Initial commit for TizenHEADsubmit/2.0alpha/20130201.014617accepted/2.0alpha/20130201.0121062.0alpha

60 files changed, 17093 insertions, 0 deletions

diff --git a/compat/basename.c b/compat/basename.c
new file mode 100644
index 0000000..d8f8a3c
--- /dev/null
+++ b/compat/basename.c
@@ -0,0 +1,15 @@
+#include "../git-compat-util.h"
+
+/* Adapted from libiberty's basename.c.  */
+char *gitbasename (char *path)
+{
+	const char *base;
+	/* Skip over the disk name in MSDOS pathnames. */
+	if (has_dos_drive_prefix(path))
+		path += 2;
+	for (base = path; *path; path++) {
+		if (is_dir_sep(*path))
+			base = path + 1;
+	}
+	return (char *)base;
+}
diff --git a/compat/bswap.h b/compat/bswap.h
new file mode 100644
index 0000000..54756db
--- /dev/null
+++ b/compat/bswap.h
@@ -0,0 +1,48 @@
+/*
+ * Let's make sure we always have a sane definition for ntohl()/htonl().
+ * Some libraries define those as a function call, just to perform byte
+ * shifting, bringing significant overhead to what should be a simple
+ * operation.
+ */
+
+/*
+ * Default version that the compiler ought to optimize properly with
+ * constant values.
+ */
+static inline uint32_t default_swab32(uint32_t val)
+{
+	return (((val & 0xff000000) >> 24) |
+		((val & 0x00ff0000) >>  8) |
+		((val & 0x0000ff00) <<  8) |
+		((val & 0x000000ff) << 24));
+}
+
+#undef bswap32
+
+#if defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
+
+#define bswap32(x) ({ \
+	uint32_t __res; \
+	if (__builtin_constant_p(x)) { \
+		__res = default_swab32(x); \
+	} else { \
+		__asm__("bswap %0" : "=r" (__res) : "0" ((uint32_t)(x))); \
+	} \
+	__res; })
+
+#elif defined(_MSC_VER) && (defined(_M_IX86) || defined(_M_X64))
+
+#include <stdlib.h>
+
+#define bswap32(x) _byteswap_ulong(x)
+
+#endif
+
+#ifdef bswap32
+
+#undef ntohl
+#undef htonl
+#define ntohl(x) bswap32(x)
+#define htonl(x) bswap32(x)
+
+#endif
diff --git a/compat/cygwin.c b/compat/cygwin.c
new file mode 100644
index 0000000..b4a51b9
--- /dev/null
+++ b/compat/cygwin.c
@@ -0,0 +1,143 @@
+#define WIN32_LEAN_AND_MEAN
+#include "../git-compat-util.h"
+#include "win32.h"
+#include "../cache.h" /* to read configuration */
+
+static inline void filetime_to_timespec(const FILETIME *ft, struct timespec *ts)
+{
+	long long winTime = ((long long)ft->dwHighDateTime << 32) +
+			ft->dwLowDateTime;
+	winTime -= 116444736000000000LL; /* Windows to Unix Epoch conversion */
+	/* convert 100-nsecond interval to seconds and nanoseconds */
+	ts->tv_sec = (time_t)(winTime/10000000);
+	ts->tv_nsec = (long)(winTime - ts->tv_sec*10000000LL) * 100;
+}
+
+#define size_to_blocks(s) (((s)+511)/512)
+
+/* do_stat is a common implementation for cygwin_lstat and cygwin_stat.
+ *
+ * To simplify its logic, in the case of cygwin symlinks, this implementation
+ * falls back to the cygwin version of stat/lstat, which is provided as the
+ * last argument.
+ */
+static int do_stat(const char *file_name, struct stat *buf, stat_fn_t cygstat)
+{
+	WIN32_FILE_ATTRIBUTE_DATA fdata;
+
+	if (file_name[0] == '/')
+		return cygstat (file_name, buf);
+
+	if (!(errno = get_file_attr(file_name, &fdata))) {
+		/*
+		 * If the system attribute is set and it is not a directory then
+		 * it could be a symbol link created in the nowinsymlinks mode.
+		 * Normally, Cygwin works in the winsymlinks mode, so this situation
+		 * is very unlikely. For the sake of simplicity of our code, let's
+		 * Cygwin to handle it.
+		 */
+		if ((fdata.dwFileAttributes & FILE_ATTRIBUTE_SYSTEM) &&
+		    !(fdata.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY))
+			return cygstat(file_name, buf);
+
+		/* fill out the stat structure */
+		buf->st_dev = buf->st_rdev = 0; /* not used by Git */
+		buf->st_ino = 0;
+		buf->st_mode = file_attr_to_st_mode(fdata.dwFileAttributes);
+		buf->st_nlink = 1;
+		buf->st_uid = buf->st_gid = 0;
+#ifdef __CYGWIN_USE_BIG_TYPES__
+		buf->st_size = ((_off64_t)fdata.nFileSizeHigh << 32) +
+			fdata.nFileSizeLow;
+#else
+		buf->st_size = (off_t)fdata.nFileSizeLow;
+#endif
+		buf->st_blocks = size_to_blocks(buf->st_size);
+		filetime_to_timespec(&fdata.ftLastAccessTime, &buf->st_atim);
+		filetime_to_timespec(&fdata.ftLastWriteTime, &buf->st_mtim);
+		filetime_to_timespec(&fdata.ftCreationTime, &buf->st_ctim);
+		return 0;
+	} else if (errno == ENOENT) {
+		/*
+		 * In the winsymlinks mode (which is the default), Cygwin
+		 * emulates symbol links using Windows shortcut files. These
+		 * files are formed by adding .lnk extension. So, if we have
+		 * not found the specified file name, it could be that it is
+		 * a symbol link. Let's Cygwin to deal with that.
+		 */
+		return cygstat(file_name, buf);
+	}
+	return -1;
+}
+
+/* We provide our own lstat/stat functions, since the provided Cygwin versions
+ * of these functions are too slow. These stat functions are tailored for Git's
+ * usage, and therefore they are not meant to be complete and correct emulation
+ * of lstat/stat functionality.
+ */
+static int cygwin_lstat(const char *path, struct stat *buf)
+{
+	return do_stat(path, buf, lstat);
+}
+
+static int cygwin_stat(const char *path, struct stat *buf)
+{
+	return do_stat(path, buf, stat);
+}
+
+
+/*
+ * At start up, we are trying to determine whether Win32 API or cygwin stat
+ * functions should be used. The choice is determined by core.ignorecygwinfstricks.
+ * Reading this option is not always possible immediately as git_dir may
+ * not be set yet. So until it is set, use cygwin lstat/stat functions.
+ * However, if core.filemode is set, we must use the Cygwin posix
+ * stat/lstat as the Windows stat functions do not determine posix filemode.
+ *
+ * Note that git_cygwin_config() does NOT call git_default_config() and this
+ * is deliberate.  Many commands read from config to establish initial
+ * values in variables and later tweak them from elsewhere (e.g. command line).
+ * init_stat() is called lazily on demand, typically much late in the program,
+ * and calling git_default_config() from here would break such variables.
+ */
+static int native_stat = 1;
+static int core_filemode;
+
+static int git_cygwin_config(const char *var, const char *value, void *cb)
+{
+	if (!strcmp(var, "core.ignorecygwinfstricks"))
+		native_stat = git_config_bool(var, value);
+	else if (!strcmp(var, "core.filemode"))
+		core_filemode = git_config_bool(var, value);
+	return 0;
+}
+
+static int init_stat(void)
+{
+	if (have_git_dir()) {
+		git_config(git_cygwin_config, NULL);
+		if (!core_filemode && native_stat) {
+			cygwin_stat_fn = cygwin_stat;
+			cygwin_lstat_fn = cygwin_lstat;
+		} else {
+			cygwin_stat_fn = stat;
+			cygwin_lstat_fn = lstat;
+		}
+		return 0;
+	}
+	return 1;
+}
+
+static int cygwin_stat_stub(const char *file_name, struct stat *buf)
+{
+	return (init_stat() ? stat : *cygwin_stat_fn)(file_name, buf);
+}
+
+static int cygwin_lstat_stub(const char *file_name, struct stat *buf)
+{
+	return (init_stat() ? lstat : *cygwin_lstat_fn)(file_name, buf);
+}
+
+stat_fn_t cygwin_stat_fn = cygwin_stat_stub;
+stat_fn_t cygwin_lstat_fn = cygwin_lstat_stub;
+
diff --git a/compat/cygwin.h b/compat/cygwin.h
new file mode 100644
index 0000000..a3229f5
--- /dev/null
+++ b/compat/cygwin.h
@@ -0,0 +1,9 @@
+#include <sys/types.h>
+#include <sys/stat.h>
+
+typedef int (*stat_fn_t)(const char*, struct stat*);
+extern stat_fn_t cygwin_stat_fn;
+extern stat_fn_t cygwin_lstat_fn;
+
+#define stat(path, buf) (*cygwin_stat_fn)(path, buf)
+#define lstat(path, buf) (*cygwin_lstat_fn)(path, buf)
diff --git a/compat/fnmatch/fnmatch.c b/compat/fnmatch/fnmatch.c
new file mode 100644
index 0000000..14feac7
--- /dev/null
+++ b/compat/fnmatch/fnmatch.c
@@ -0,0 +1,488 @@
+/* Copyright (C) 1991, 92, 93, 96, 97, 98, 99 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Library General Public License as
+   published by the Free Software Foundation; either version 2 of the
+   License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Library General Public License for more details.
+
+   You should have received a copy of the GNU Library General Public
+   License along with this library; see the file COPYING.LIB.  If not,
+   write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+   Boston, MA 02111-1307, USA.  */
+
+#if HAVE_CONFIG_H
+# include <config.h>
+#endif
+
+/* Enable GNU extensions in fnmatch.h.  */
+#ifndef _GNU_SOURCE
+# define _GNU_SOURCE	1
+#endif
+
+#include <errno.h>
+#include <fnmatch.h>
+#include <ctype.h>
+
+#if HAVE_STRING_H || defined _LIBC
+# include <string.h>
+#else
+# include <strings.h>
+#endif
+
+#if defined STDC_HEADERS || defined _LIBC
+# include <stdlib.h>
+#endif
+
+/* For platforms which support the ISO C amendment 1 functionality we
+   support user defined character classes.  */
+#if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
+/* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>.  */
+# include <wchar.h>
+# include <wctype.h>
+#endif
+
+/* Comment out all this code if we are using the GNU C Library, and are not
+   actually compiling the library itself.  This code is part of the GNU C
+   Library, but also included in many other GNU distributions.  Compiling
+   and linking in this code is a waste when using the GNU C library
+   (especially if it is a shared library).  Rather than having every GNU
+   program understand `configure --with-gnu-libc' and omit the object files,
+   it is simpler to just do this in the source for each such file.  */
+
+#if defined _LIBC || !defined __GNU_LIBRARY__
+
+
+# if defined STDC_HEADERS || !defined isascii
+#  define ISASCII(c) 1
+# else
+#  define ISASCII(c) isascii(c)
+# endif
+
+# ifdef isblank
+#  define ISBLANK(c) (ISASCII (c) && isblank (c))
+# else
+#  define ISBLANK(c) ((c) == ' ' || (c) == '\t')
+# endif
+# ifdef isgraph
+#  define ISGRAPH(c) (ISASCII (c) && isgraph (c))
+# else
+#  define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
+# endif
+
+# define ISPRINT(c) (ISASCII (c) && isprint (c))
+# define ISDIGIT(c) (ISASCII (c) && isdigit (c))
+# define ISALNUM(c) (ISASCII (c) && isalnum (c))
+# define ISALPHA(c) (ISASCII (c) && isalpha (c))
+# define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
+# define ISLOWER(c) (ISASCII (c) && islower (c))
+# define ISPUNCT(c) (ISASCII (c) && ispunct (c))
+# define ISSPACE(c) (ISASCII (c) && isspace (c))
+# define ISUPPER(c) (ISASCII (c) && isupper (c))
+# define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
+
+# define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
+
+# if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
+/* The GNU C library provides support for user-defined character classes
+   and the functions from ISO C amendment 1.  */
+#  ifdef CHARCLASS_NAME_MAX
+#   define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
+#  else
+/* This shouldn't happen but some implementation might still have this
+   problem.  Use a reasonable default value.  */
+#   define CHAR_CLASS_MAX_LENGTH 256
+#  endif
+
+#  ifdef _LIBC
+#   define IS_CHAR_CLASS(string) __wctype (string)
+#  else
+#   define IS_CHAR_CLASS(string) wctype (string)
+#  endif
+# else
+#  define CHAR_CLASS_MAX_LENGTH  6 /* Namely, `xdigit'.  */
+
+#  define IS_CHAR_CLASS(string)						      \
+   (STREQ (string, "alpha") || STREQ (string, "upper")			      \
+    || STREQ (string, "lower") || STREQ (string, "digit")		      \
+    || STREQ (string, "alnum") || STREQ (string, "xdigit")		      \
+    || STREQ (string, "space") || STREQ (string, "print")		      \
+    || STREQ (string, "punct") || STREQ (string, "graph")		      \
+    || STREQ (string, "cntrl") || STREQ (string, "blank"))
+# endif
+
+/* Avoid depending on library functions or files
+   whose names are inconsistent.  */
+
+# if !defined _LIBC && !defined getenv
+extern char *getenv ();
+# endif
+
+# ifndef errno
+extern int errno;
+# endif
+
+/* This function doesn't exist on most systems.  */
+
+# if !defined HAVE___STRCHRNUL && !defined _LIBC
+static char *
+__strchrnul (s, c)
+     const char *s;
+     int c;
+{
+  char *result = strchr (s, c);
+  if (result == NULL)
+    result = strchr (s, '\0');
+  return result;
+}
+# endif
+
+# ifndef internal_function
+/* Inside GNU libc we mark some function in a special way.  In other
+   environments simply ignore the marking.  */
+#  define internal_function
+# endif
+
+/* Match STRING against the filename pattern PATTERN, returning zero if
+   it matches, nonzero if not.  */
+static int internal_fnmatch __P ((const char *pattern, const char *string,
+				  int no_leading_period, int flags))
+     internal_function;
+static int
+internal_function
+internal_fnmatch (pattern, string, no_leading_period, flags)
+     const char *pattern;
+     const char *string;
+     int no_leading_period;
+     int flags;
+{
+  register const char *p = pattern, *n = string;
+  register unsigned char c;
+
+/* Note that this evaluates C many times.  */
+# ifdef _LIBC
+#  define FOLD(c) ((flags & FNM_CASEFOLD) ? tolower (c) : (c))
+# else
+#  define FOLD(c) ((flags & FNM_CASEFOLD) && ISUPPER (c) ? tolower (c) : (c))
+# endif
+
+  while ((c = *p++) != '\0')
+    {
+      c = FOLD (c);
+
+      switch (c)
+	{
+	case '?':
+	  if (*n == '\0')
+	    return FNM_NOMATCH;
+	  else if (*n == '/' && (flags & FNM_FILE_NAME))
+	    return FNM_NOMATCH;
+	  else if (*n == '.' && no_leading_period
+		   && (n == string
+		       || (n[-1] == '/' && (flags & FNM_FILE_NAME))))
+	    return FNM_NOMATCH;
+	  break;
+
+	case '\\':
+	  if (!(flags & FNM_NOESCAPE))
+	    {
+	      c = *p++;
+	      if (c == '\0')
+		/* Trailing \ loses.  */
+		return FNM_NOMATCH;
+	      c = FOLD (c);
+	    }
+	  if (FOLD ((unsigned char) *n) != c)
+	    return FNM_NOMATCH;
+	  break;
+
+	case '*':
+	  if (*n == '.' && no_leading_period
+	      && (n == string
+		  || (n[-1] == '/' && (flags & FNM_FILE_NAME))))
+	    return FNM_NOMATCH;
+
+	  for (c = *p++; c == '?' || c == '*'; c = *p++)
+	    {
+	      if (*n == '/' && (flags & FNM_FILE_NAME))
+		/* A slash does not match a wildcard under FNM_FILE_NAME.  */
+		return FNM_NOMATCH;
+	      else if (c == '?')
+		{
+		  /* A ? needs to match one character.  */
+		  if (*n == '\0')
+		    /* There isn't another character; no match.  */
+		    return FNM_NOMATCH;
+		  else
+		    /* One character of the string is consumed in matching
+		       this ? wildcard, so *??? won't match if there are
+		       less than three characters.  */
+		    ++n;
+		}
+	    }
+
+	  if (c == '\0')
+	    /* The wildcard(s) is/are the last element of the pattern.
+	       If the name is a file name and contains another slash
+	       this does mean it cannot match.  */
+	    return ((flags & FNM_FILE_NAME) && strchr (n, '/') != NULL
+		    ? FNM_NOMATCH : 0);
+	  else
+	    {
+	      const char *endp;
+
+	      endp = __strchrnul (n, (flags & FNM_FILE_NAME) ? '/' : '\0');
+
+	      if (c == '[')
+		{
+		  int flags2 = ((flags & FNM_FILE_NAME)
+				? flags : (flags & ~FNM_PERIOD));
+
+		  for (--p; n < endp; ++n)
+		    if (internal_fnmatch (p, n,
+					  (no_leading_period
+					   && (n == string
+					       || (n[-1] == '/'
+						   && (flags
+						       & FNM_FILE_NAME)))),
+					  flags2)
+			== 0)
+		      return 0;
+		}
+	      else if (c == '/' && (flags & FNM_FILE_NAME))
+		{
+		  while (*n != '\0' && *n != '/')
+		    ++n;
+		  if (*n == '/'
+		      && (internal_fnmatch (p, n + 1, flags & FNM_PERIOD,
+					    flags) == 0))
+		    return 0;
+		}
+	      else
+		{
+		  int flags2 = ((flags & FNM_FILE_NAME)
+				? flags : (flags & ~FNM_PERIOD));
+
+		  if (c == '\\' && !(flags & FNM_NOESCAPE))
+		    c = *p;
+		  c = FOLD (c);
+		  for (--p; n < endp; ++n)
+		    if (FOLD ((unsigned char) *n) == c
+			&& (internal_fnmatch (p, n,
+					      (no_leading_period
+					       && (n == string
+						   || (n[-1] == '/'
+						       && (flags
+							   & FNM_FILE_NAME)))),
+					      flags2) == 0))
+		      return 0;
+		}
+	    }
+
+	  /* If we come here no match is possible with the wildcard.  */
+	  return FNM_NOMATCH;
+
+	case '[':
+	  {
+	    /* Nonzero if the sense of the character class is inverted.  */
+	    static int posixly_correct;
+	    register int not;
+	    char cold;
+
+	    if (posixly_correct == 0)
+	      posixly_correct = getenv ("POSIXLY_CORRECT") != NULL ? 1 : -1;
+
+	    if (*n == '\0')
+	      return FNM_NOMATCH;
+
+	    if (*n == '.' && no_leading_period && (n == string
+						   || (n[-1] == '/'
+						       && (flags
+							   & FNM_FILE_NAME))))
+	      return FNM_NOMATCH;
+
+	    if (*n == '/' && (flags & FNM_FILE_NAME))
+	      /* `/' cannot be matched.  */
+	      return FNM_NOMATCH;
+
+	    not = (*p == '!' || (posixly_correct < 0 && *p == '^'));
+	    if (not)
+	      ++p;
+
+	    c = *p++;
+	    for (;;)
+	      {
+		unsigned char fn = FOLD ((unsigned char) *n);
+
+		if (!(flags & FNM_NOESCAPE) && c == '\\')
+		  {
+		    if (*p == '\0')
+		      return FNM_NOMATCH;
+		    c = FOLD ((unsigned char) *p);
+		    ++p;
+
+		    if (c == fn)
+		      goto matched;
+		  }
+		else if (c == '[' && *p == ':')
+		  {
+		    /* Leave room for the null.  */
+		    char str[CHAR_CLASS_MAX_LENGTH + 1];
+		    size_t c1 = 0;
+# if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
+		    wctype_t wt;
+# endif
+		    const char *startp = p;
+
+		    for (;;)
+		      {
+			if (c1 == CHAR_CLASS_MAX_LENGTH)
+			  /* The name is too long and therefore the pattern
+			     is ill-formed.  */
+			  return FNM_NOMATCH;
+
+			c = *++p;
+			if (c == ':' && p[1] == ']')
+			  {
+			    p += 2;
+			    break;
+			  }
+			if (c < 'a' || c >= 'z')
+			  {
+			    /* This cannot possibly be a character class name.
+			       Match it as a normal range.  */
+			    p = startp;
+			    c = '[';
+			    goto normal_bracket;
+			  }
+			str[c1++] = c;
+		      }
+		    str[c1] = '\0';
+
+# if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
+		    wt = IS_CHAR_CLASS (str);
+		    if (wt == 0)
+		      /* Invalid character class name.  */
+		      return FNM_NOMATCH;
+
+		    if (__iswctype (__btowc ((unsigned char) *n), wt))
+		      goto matched;
+# else
+		    if ((STREQ (str, "alnum") && ISALNUM ((unsigned char) *n))
+			|| (STREQ (str, "alpha") && ISALPHA ((unsigned char) *n))
+			|| (STREQ (str, "blank") && ISBLANK ((unsigned char) *n))
+			|| (STREQ (str, "cntrl") && ISCNTRL ((unsigned char) *n))
+			|| (STREQ (str, "digit") && ISDIGIT ((unsigned char) *n))
+			|| (STREQ (str, "graph") && ISGRAPH ((unsigned char) *n))
+			|| (STREQ (str, "lower") && ISLOWER ((unsigned char) *n))
+			|| (STREQ (str, "print") && ISPRINT ((unsigned char) *n))
+			|| (STREQ (str, "punct") && ISPUNCT ((unsigned char) *n))
+			|| (STREQ (str, "space") && ISSPACE ((unsigned char) *n))
+			|| (STREQ (str, "upper") && ISUPPER ((unsigned char) *n))
+			|| (STREQ (str, "xdigit") && ISXDIGIT ((unsigned char) *n)))
+		      goto matched;
+# endif
+		  }
+		else if (c == '\0')
+		  /* [ (unterminated) loses.  */
+		  return FNM_NOMATCH;
+		else
+		  {
+		  normal_bracket:
+		    if (FOLD (c) == fn)
+		      goto matched;
+
+		    cold = c;
+		    c = *p++;
+
+		    if (c == '-' && *p != ']')
+		      {
+			/* It is a range.  */
+			unsigned char cend = *p++;
+			if (!(flags & FNM_NOESCAPE) && cend == '\\')
+			  cend = *p++;
+			if (cend == '\0')
+			  return FNM_NOMATCH;
+
+			if (cold <= fn && fn <= FOLD (cend))
+			  goto matched;
+
+			c = *p++;
+		      }
+		  }
+
+		if (c == ']')
+		  break;
+	      }
+
+	    if (!not)
+	      return FNM_NOMATCH;
+	    break;
+
+	  matched:
+	    /* Skip the rest of the [...] that already matched.  */
+	    while (c != ']')
+	      {
+		if (c == '\0')
+		  /* [... (unterminated) loses.  */
+		  return FNM_NOMATCH;
+
+		c = *p++;
+		if (!(flags & FNM_NOESCAPE) && c == '\\')
+		  {
+		    if (*p == '\0')
+		      return FNM_NOMATCH;
+		    /* XXX 1003.2d11 is unclear if this is right.  */
+		    ++p;
+		  }
+		else if (c == '[' && *p == ':')
+		  {
+		    do
+		      if (*++p == '\0')
+			return FNM_NOMATCH;
+		    while (*p != ':' || p[1] == ']');
+		    p += 2;
+		    c = *p;
+		  }
+	      }
+	    if (not)
+	      return FNM_NOMATCH;
+	  }
+	  break;
+
+	default:
+	  if (c != FOLD ((unsigned char) *n))
+	    return FNM_NOMATCH;
+	}
+
+      ++n;
+    }
+
+  if (*n == '\0')
+    return 0;
+
+  if ((flags & FNM_LEADING_DIR) && *n == '/')
+    /* The FNM_LEADING_DIR flag says that "foo*" matches "foobar/frobozz".  */
+    return 0;
+
+  return FNM_NOMATCH;
+
+# undef FOLD
+}
+
+
+int
+fnmatch (pattern, string, flags)
+     const char *pattern;
+     const char *string;
+     int flags;
+{
+  return internal_fnmatch (pattern, string, flags & FNM_PERIOD, flags);
+}
+
+#endif	/* _LIBC or not __GNU_LIBRARY__.  */
diff --git a/compat/fnmatch/fnmatch.h b/compat/fnmatch/fnmatch.h
new file mode 100644
index 0000000..cc3ec37
--- /dev/null
+++ b/compat/fnmatch/fnmatch.h
@@ -0,0 +1,84 @@
+/* Copyright (C) 1991, 92, 93, 96, 97, 98, 99 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Library General Public License as
+   published by the Free Software Foundation; either version 2 of the
+   License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Library General Public License for more details.
+
+   You should have received a copy of the GNU Library General Public
+   License along with the GNU C Library; see the file COPYING.LIB.  If not,
+   write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+   Boston, MA 02111-1307, USA.  */
+
+#ifndef	_FNMATCH_H
+#define	_FNMATCH_H	1
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+#if defined __cplusplus || (defined __STDC__ && __STDC__) || defined WINDOWS32
+# if !defined __GLIBC__ || !defined __P
+#  undef	__P
+#  define __P(protos)	protos
+# endif
+#else /* Not C++ or ANSI C.  */
+# undef	__P
+# define __P(protos)	()
+/* We can get away without defining `const' here only because in this file
+   it is used only inside the prototype for `fnmatch', which is elided in
+   non-ANSI C where `const' is problematical.  */
+#endif /* C++ or ANSI C.  */
+
+#ifndef const
+# if (defined __STDC__ && __STDC__) || defined __cplusplus
+#  define __const	const
+# else
+#  define __const
+# endif
+#endif
+
+/* We #undef these before defining them because some losing systems
+   (HP-UX A.08.07 for example) define these in <unistd.h>.  */
+#undef	FNM_PATHNAME
+#undef	FNM_NOESCAPE
+#undef	FNM_PERIOD
+
+/* Bits set in the FLAGS argument to `fnmatch'.  */
+#define	FNM_PATHNAME	(1 << 0) /* No wildcard can ever match `/'.  */
+#define	FNM_NOESCAPE	(1 << 1) /* Backslashes don't quote special chars.  */
+#define	FNM_PERIOD	(1 << 2) /* Leading `.' is matched only explicitly.  */
+
+#if !defined _POSIX_C_SOURCE || _POSIX_C_SOURCE < 2 || defined _GNU_SOURCE
+# define FNM_FILE_NAME	 FNM_PATHNAME	/* Preferred GNU name.  */
+# define FNM_LEADING_DIR (1 << 3)	/* Ignore `/...' after a match.  */
+# define FNM_CASEFOLD	 (1 << 4)	/* Compare without regard to case.  */
+#endif
+
+/* Value returned by `fnmatch' if STRING does not match PATTERN.  */
+#define	FNM_NOMATCH	1
+
+/* This value is returned if the implementation does not support
+   `fnmatch'.  Since this is not the case here it will never be
+   returned but the conformance test suites still require the symbol
+   to be defined.  */
+#ifdef _XOPEN_SOURCE
+# define FNM_NOSYS	(-1)
+#endif
+
+/* Match NAME against the filename pattern PATTERN,
+   returning zero if it matches, FNM_NOMATCH if not.  */
+extern int fnmatch __P ((__const char *__pattern, __const char *__name,
+			 int __flags));
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif /* fnmatch.h */
diff --git a/compat/fopen.c b/compat/fopen.c
new file mode 100644
index 0000000..b5ca142
--- /dev/null
+++ b/compat/fopen.c
@@ -0,0 +1,37 @@
+/*
+ *  The order of the following two lines is important.
+ *
+ *  FREAD_READS_DIRECTORIES is undefined before including git-compat-util.h
+ *  to avoid the redefinition of fopen within git-compat-util.h. This is
+ *  necessary since fopen is a macro on some platforms which may be set
+ *  based on compiler options. For example, on AIX fopen is set to fopen64
+ *  when _LARGE_FILES is defined. The previous technique of merely undefining
+ *  fopen after including git-compat-util.h is inadequate in this case.
+ */
+#undef FREAD_READS_DIRECTORIES
+#include "../git-compat-util.h"
+
+FILE *git_fopen(const char *path, const char *mode)
+{
+	FILE *fp;
+	struct stat st;
+
+	if (mode[0] == 'w' || mode[0] == 'a')
+		return fopen(path, mode);
+
+	if (!(fp = fopen(path, mode)))
+		return NULL;
+
+	if (fstat(fileno(fp), &st)) {
+		fclose(fp);
+		return NULL;
+	}
+
+	if (S_ISDIR(st.st_mode)) {
+		fclose(fp);
+		errno = EISDIR;
+		return NULL;
+	}
+
+	return fp;
+}
diff --git a/compat/hstrerror.c b/compat/hstrerror.c
new file mode 100644
index 0000000..069c555
--- /dev/null
+++ b/compat/hstrerror.c
@@ -0,0 +1,21 @@
+#include <string.h>
+#include <stdio.h>
+#include <netdb.h>
+
+const char *githstrerror(int err)
+{
+	static char buffer[48];
+	switch (err)
+	{
+	case HOST_NOT_FOUND:
+		return "Authoritative answer: host not found";
+	case NO_DATA:
+		return "Valid name, no data record of requested type";
+	case NO_RECOVERY:
+		return "Non recoverable errors, FORMERR, REFUSED, NOTIMP";
+	case TRY_AGAIN:
+		return "Non-authoritative \"host not found\", or SERVERFAIL";
+	}
+	sprintf(buffer, "Name resolution error %d", err);
+	return buffer;
+}
diff --git a/compat/inet_ntop.c b/compat/inet_ntop.c
new file mode 100644
index 0000000..f444982
--- /dev/null
+++ b/compat/inet_ntop.c
@@ -0,0 +1,199 @@
+/*
+ * Copyright (c) 1996-1999 by Internet Software Consortium.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND INTERNET SOFTWARE CONSORTIUM DISCLAIMS
+ * ALL WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES
+ * OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL INTERNET SOFTWARE
+ * CONSORTIUM BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL
+ * DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR
+ * PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS
+ * ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS
+ * SOFTWARE.
+ */
+
+#include <errno.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <netinet/in.h>
+#include <arpa/inet.h>
+#include <stdio.h>
+#include <string.h>
+
+#ifndef NS_INADDRSZ
+#define NS_INADDRSZ	4
+#endif
+#ifndef NS_IN6ADDRSZ
+#define NS_IN6ADDRSZ	16
+#endif
+#ifndef NS_INT16SZ
+#define NS_INT16SZ	2
+#endif
+
+/*
+ * WARNING: Don't even consider trying to compile this on a system where
+ * sizeof(int) < 4.  sizeof(int) > 4 is fine; all the world's not a VAX.
+ */
+
+/* const char *
+ * inet_ntop4(src, dst, size)
+ *	format an IPv4 address
+ * return:
+ *	`dst' (as a const)
+ * notes:
+ *	(1) uses no statics
+ *	(2) takes a u_char* not an in_addr as input
+ * author:
+ *	Paul Vixie, 1996.
+ */
+static const char *
+inet_ntop4(src, dst, size)
+	const u_char *src;
+	char *dst;
+	size_t size;
+{
+	static const char fmt[] = "%u.%u.%u.%u";
+	char tmp[sizeof "255.255.255.255"];
+	int nprinted;
+
+	nprinted = snprintf(tmp, sizeof(tmp), fmt, src[0], src[1], src[2], src[3]);
+	if (nprinted < 0)
+		return (NULL);	/* we assume "errno" was set by "snprintf()" */
+	if ((size_t)nprinted > size) {
+		errno = ENOSPC;
+		return (NULL);
+	}
+	strcpy(dst, tmp);
+	return (dst);
+}
+
+#ifndef NO_IPV6
+/* const char *
+ * inet_ntop6(src, dst, size)
+ *	convert IPv6 binary address into presentation (printable) format
+ * author:
+ *	Paul Vixie, 1996.
+ */
+static const char *
+inet_ntop6(src, dst, size)
+	const u_char *src;
+	char *dst;
+	size_t size;
+{
+	/*
+	 * Note that int32_t and int16_t need only be "at least" large enough
+	 * to contain a value of the specified size.  On some systems, like
+	 * Crays, there is no such thing as an integer variable with 16 bits.
+	 * Keep this in mind if you think this function should have been coded
+	 * to use pointer overlays.  All the world's not a VAX.
+	 */
+	char tmp[sizeof "ffff:ffff:ffff:ffff:ffff:ffff:255.255.255.255"], *tp;
+	struct { int base, len; } best, cur;
+	unsigned int words[NS_IN6ADDRSZ / NS_INT16SZ];
+	int i;
+
+	/*
+	 * Preprocess:
+	 *	Copy the input (bytewise) array into a wordwise array.
+	 *	Find the longest run of 0x00's in src[] for :: shorthanding.
+	 */
+	memset(words, '\0', sizeof words);
+	for (i = 0; i < NS_IN6ADDRSZ; i++)
+		words[i / 2] |= (src[i] << ((1 - (i % 2)) << 3));
+	best.base = -1;
+	cur.base = -1;
+	for (i = 0; i < (NS_IN6ADDRSZ / NS_INT16SZ); i++) {
+		if (words[i] == 0) {
+			if (cur.base == -1)
+				cur.base = i, cur.len = 1;
+			else
+				cur.len++;
+		} else {
+			if (cur.base != -1) {
+				if (best.base == -1 || cur.len > best.len)
+					best = cur;
+				cur.base = -1;
+			}
+		}
+	}
+	if (cur.base != -1) {
+		if (best.base == -1 || cur.len > best.len)
+			best = cur;
+	}
+	if (best.base != -1 && best.len < 2)
+		best.base = -1;
+
+	/*
+	 * Format the result.
+	 */
+	tp = tmp;
+	for (i = 0; i < (NS_IN6ADDRSZ / NS_INT16SZ); i++) {
+		/* Are we inside the best run of 0x00's? */
+		if (best.base != -1 && i >= best.base &&
+		    i < (best.base + best.len)) {
+			if (i == best.base)
+				*tp++ = ':';
+			continue;
+		}
+		/* Are we following an initial run of 0x00s or any real hex? */
+		if (i != 0)
+			*tp++ = ':';
+		/* Is this address an encapsulated IPv4? */
+		if (i == 6 && best.base == 0 &&
+		    (best.len == 6 || (best.len == 5 && words[5] == 0xffff))) {
+			if (!inet_ntop4(src+12, tp, sizeof tmp - (tp - tmp)))
+				return (NULL);
+			tp += strlen(tp);
+			break;
+		}
+		tp += snprintf(tp, sizeof tmp - (tp - tmp), "%x", words[i]);
+	}
+	/* Was it a trailing run of 0x00's? */
+	if (best.base != -1 && (best.base + best.len) ==
+	    (NS_IN6ADDRSZ / NS_INT16SZ))
+		*tp++ = ':';
+	*tp++ = '\0';
+
+	/*
+	 * Check for overflow, copy, and we're done.
+	 */
+	if ((size_t)(tp - tmp) > size) {
+		errno = ENOSPC;
+		return (NULL);
+	}
+	strcpy(dst, tmp);
+	return (dst);
+}
+#endif
+
+/* char *
+ * inet_ntop(af, src, dst, size)
+ *	convert a network format address to presentation format.
+ * return:
+ *	pointer to presentation format address (`dst'), or NULL (see errno).
+ * author:
+ *	Paul Vixie, 1996.
+ */
+const char *
+inet_ntop(af, src, dst, size)
+	int af;
+	const void *src;
+	char *dst;
+	size_t size;
+{
+	switch (af) {
+	case AF_INET:
+		return (inet_ntop4(src, dst, size));
+#ifndef NO_IPV6
+	case AF_INET6:
+		return (inet_ntop6(src, dst, size));
+#endif
+	default:
+		errno = EAFNOSUPPORT;
+		return (NULL);
+	}
+	/* NOTREACHED */
+}
diff --git a/compat/inet_pton.c b/compat/inet_pton.c
new file mode 100644
index 0000000..4078fc0
--- /dev/null
+++ b/compat/inet_pton.c
@@ -0,0 +1,219 @@
+/*
+ * Copyright (C) 1996-2001  Internet Software Consortium.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND INTERNET SOFTWARE CONSORTIUM
+ * DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL
+ * INTERNET SOFTWARE CONSORTIUM BE LIABLE FOR ANY SPECIAL, DIRECT,
+ * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING
+ * FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT,
+ * NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION
+ * WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include <errno.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <netinet/in.h>
+#include <arpa/inet.h>
+#include <stdio.h>
+#include <string.h>
+
+#ifndef NS_INT16SZ
+#define NS_INT16SZ       2
+#endif
+
+#ifndef NS_INADDRSZ
+#define NS_INADDRSZ      4
+#endif
+
+#ifndef NS_IN6ADDRSZ
+#define NS_IN6ADDRSZ    16
+#endif
+
+/*
+ * WARNING: Don't even consider trying to compile this on a system where
+ * sizeof(int) < 4.  sizeof(int) > 4 is fine; all the world's not a VAX.
+ */
+
+static int inet_pton4(const char *src, unsigned char *dst);
+static int inet_pton6(const char *src, unsigned char *dst);
+
+/* int
+ * inet_pton4(src, dst)
+ *      like inet_aton() but without all the hexadecimal and shorthand.
+ * return:
+ *      1 if `src' is a valid dotted quad, else 0.
+ * notice:
+ *      does not touch `dst' unless it's returning 1.
+ * author:
+ *      Paul Vixie, 1996.
+ */
+static int
+inet_pton4(const char *src, unsigned char *dst)
+{
+        static const char digits[] = "0123456789";
+        int saw_digit, octets, ch;
+        unsigned char tmp[NS_INADDRSZ], *tp;
+
+        saw_digit = 0;
+        octets = 0;
+        *(tp = tmp) = 0;
+        while ((ch = *src++) != '\0') {
+                const char *pch;
+
+                if ((pch = strchr(digits, ch)) != NULL) {
+                        unsigned int new = *tp * 10 + (pch - digits);
+
+                        if (new > 255)
+                                return (0);
+                        *tp = new;
+                        if (! saw_digit) {
+                                if (++octets > 4)
+                                        return (0);
+                                saw_digit = 1;
+                        }
+                } else if (ch == '.' && saw_digit) {
+                        if (octets == 4)
+                                return (0);
+                        *++tp = 0;
+                        saw_digit = 0;
+                } else
+                        return (0);
+        }
+        if (octets < 4)
+                return (0);
+        memcpy(dst, tmp, NS_INADDRSZ);
+        return (1);
+}
+
+/* int
+ * inet_pton6(src, dst)
+ *      convert presentation level address to network order binary form.
+ * return:
+ *      1 if `src' is a valid [RFC1884 2.2] address, else 0.
+ * notice:
+ *      (1) does not touch `dst' unless it's returning 1.
+ *      (2) :: in a full address is silently ignored.
+ * credit:
+ *      inspired by Mark Andrews.
+ * author:
+ *      Paul Vixie, 1996.
+ */
+
+#ifndef NO_IPV6
+static int
+inet_pton6(const char *src, unsigned char *dst)
+{
+        static const char xdigits_l[] = "0123456789abcdef",
+                          xdigits_u[] = "0123456789ABCDEF";
+        unsigned char tmp[NS_IN6ADDRSZ], *tp, *endp, *colonp;
+        const char *xdigits, *curtok;
+        int ch, saw_xdigit;
+        unsigned int val;
+
+        memset((tp = tmp), '\0', NS_IN6ADDRSZ);
+        endp = tp + NS_IN6ADDRSZ;
+        colonp = NULL;
+        /* Leading :: requires some special handling. */
+        if (*src == ':')
+                if (*++src != ':')
+                        return (0);
+        curtok = src;
+        saw_xdigit = 0;
+        val = 0;
+        while ((ch = *src++) != '\0') {
+                const char *pch;
+
+                if ((pch = strchr((xdigits = xdigits_l), ch)) == NULL)
+                        pch = strchr((xdigits = xdigits_u), ch);
+                if (pch != NULL) {
+                        val <<= 4;
+                        val |= (pch - xdigits);
+                        if (val > 0xffff)
+                                return (0);
+                        saw_xdigit = 1;
+                        continue;
+                }
+                if (ch == ':') {
+                        curtok = src;
+                        if (!saw_xdigit) {
+                                if (colonp)
+                                        return (0);
+                                colonp = tp;
+                                continue;
+                        }
+                        if (tp + NS_INT16SZ > endp)
+                                return (0);
+                        *tp++ = (unsigned char) (val >> 8) & 0xff;
+                        *tp++ = (unsigned char) val & 0xff;
+                        saw_xdigit = 0;
+                        val = 0;
+                        continue;
+                }
+                if (ch == '.' && ((tp + NS_INADDRSZ) <= endp) &&
+                    inet_pton4(curtok, tp) > 0) {
+                        tp += NS_INADDRSZ;
+                        saw_xdigit = 0;
+                        break;  /* '\0' was seen by inet_pton4(). */
+                }
+                return (0);
+        }
+        if (saw_xdigit) {
+                if (tp + NS_INT16SZ > endp)
+                        return (0);
+                *tp++ = (unsigned char) (val >> 8) & 0xff;
+                *tp++ = (unsigned char) val & 0xff;
+        }
+        if (colonp != NULL) {
+                /*
+                 * Since some memmove()'s erroneously fail to handle
+                 * overlapping regions, we'll do the shift by hand.
+                 */
+                const int n = tp - colonp;
+                int i;
+
+                for (i = 1; i <= n; i++) {
+                        endp[- i] = colonp[n - i];
+                        colonp[n - i] = 0;
+                }
+                tp = endp;
+        }
+        if (tp != endp)
+                return (0);
+        memcpy(dst, tmp, NS_IN6ADDRSZ);
+        return (1);
+}
+#endif
+
+/* int
+ * isc_net_pton(af, src, dst)
+ *      convert from presentation format (which usually means ASCII printable)
+ *      to network format (which is usually some kind of binary format).
+ * return:
+ *      1 if the address was valid for the specified address family
+ *      0 if the address wasn't valid (`dst' is untouched in this case)
+ *      -1 if some other error occurred (`dst' is untouched in this case, too)
+ * author:
+ *      Paul Vixie, 1996.
+ */
+int
+inet_pton(int af, const char *src, void *dst)
+{
+        switch (af) {
+        case AF_INET:
+                return (inet_pton4(src, dst));
+#ifndef NO_IPV6
+        case AF_INET6:
+                return (inet_pton6(src, dst));
+#endif
+        default:
+                errno = EAFNOSUPPORT;
+                return (-1);
+        }
+        /* NOTREACHED */
+}
diff --git a/compat/memmem.c b/compat/memmem.c
new file mode 100644
index 0000000..56bcb42
--- /dev/null
+++ b/compat/memmem.c
@@ -0,0 +1,32 @@
+#include "../git-compat-util.h"
+
+void *gitmemmem(const void *haystack, size_t haystack_len,
+                const void *needle, size_t needle_len)
+{
+	const char *begin = haystack;
+	const char *last_possible = begin + haystack_len - needle_len;
+	const char *tail = needle;
+	char point;
+
+	/*
+	 * The first occurrence of the empty string is deemed to occur at
+	 * the beginning of the string.
+	 */
+	if (needle_len == 0)
+		return (void *)begin;
+
+	/*
+	 * Sanity check, otherwise the loop might search through the whole
+	 * memory.
+	 */
+	if (haystack_len < needle_len)
+		return NULL;
+
+	point = *tail++;
+	for (; begin <= last_possible; begin++) {
+		if (*begin == point && !memcmp(begin + 1, tail, needle_len - 1))
+			return (void *)begin;
+	}
+
+	return NULL;
+}
diff --git a/compat/mingw.c b/compat/mingw.c
new file mode 100644
index 0000000..9a8e336
--- /dev/null
+++ b/compat/mingw.c
@@ -0,0 +1,1492 @@
+#include "../git-compat-util.h"
+#include "win32.h"
+#include <conio.h>
+#include "../strbuf.h"
+
+int err_win_to_posix(DWORD winerr)
+{
+	int error = ENOSYS;
+	switch(winerr) {
+	case ERROR_ACCESS_DENIED: error = EACCES; break;
+	case ERROR_ACCOUNT_DISABLED: error = EACCES; break;
+	case ERROR_ACCOUNT_RESTRICTION: error = EACCES; break;
+	case ERROR_ALREADY_ASSIGNED: error = EBUSY; break;
+	case ERROR_ALREADY_EXISTS: error = EEXIST; break;
+	case ERROR_ARITHMETIC_OVERFLOW: error = ERANGE; break;
+	case ERROR_BAD_COMMAND: error = EIO; break;
+	case ERROR_BAD_DEVICE: error = ENODEV; break;
+	case ERROR_BAD_DRIVER_LEVEL: error = ENXIO; break;
+	case ERROR_BAD_EXE_FORMAT: error = ENOEXEC; break;
+	case ERROR_BAD_FORMAT: error = ENOEXEC; break;
+	case ERROR_BAD_LENGTH: error = EINVAL; break;
+	case ERROR_BAD_PATHNAME: error = ENOENT; break;
+	case ERROR_BAD_PIPE: error = EPIPE; break;
+	case ERROR_BAD_UNIT: error = ENODEV; break;
+	case ERROR_BAD_USERNAME: error = EINVAL; break;
+	case ERROR_BROKEN_PIPE: error = EPIPE; break;
+	case ERROR_BUFFER_OVERFLOW: error = ENAMETOOLONG; break;
+	case ERROR_BUSY: error = EBUSY; break;
+	case ERROR_BUSY_DRIVE: error = EBUSY; break;
+	case ERROR_CALL_NOT_IMPLEMENTED: error = ENOSYS; break;
+	case ERROR_CANNOT_MAKE: error = EACCES; break;
+	case ERROR_CANTOPEN: error = EIO; break;
+	case ERROR_CANTREAD: error = EIO; break;
+	case ERROR_CANTWRITE: error = EIO; break;
+	case ERROR_CRC: error = EIO; break;
+	case ERROR_CURRENT_DIRECTORY: error = EACCES; break;
+	case ERROR_DEVICE_IN_USE: error = EBUSY; break;
+	case ERROR_DEV_NOT_EXIST: error = ENODEV; break;
+	case ERROR_DIRECTORY: error = EINVAL; break;
+	case ERROR_DIR_NOT_EMPTY: error = ENOTEMPTY; break;
+	case ERROR_DISK_CHANGE: error = EIO; break;
+	case ERROR_DISK_FULL: error = ENOSPC; break;
+	case ERROR_DRIVE_LOCKED: error = EBUSY; break;
+	case ERROR_ENVVAR_NOT_FOUND: error = EINVAL; break;
+	case ERROR_EXE_MARKED_INVALID: error = ENOEXEC; break;
+	case ERROR_FILENAME_EXCED_RANGE: error = ENAMETOOLONG; break;
+	case ERROR_FILE_EXISTS: error = EEXIST; break;
+	case ERROR_FILE_INVALID: error = ENODEV; break;
+	case ERROR_FILE_NOT_FOUND: error = ENOENT; break;
+	case ERROR_GEN_FAILURE: error = EIO; break;
+	case ERROR_HANDLE_DISK_FULL: error = ENOSPC; break;
+	case ERROR_INSUFFICIENT_BUFFER: error = ENOMEM; break;
+	case ERROR_INVALID_ACCESS: error = EACCES; break;
+	case ERROR_INVALID_ADDRESS: error = EFAULT; break;
+	case ERROR_INVALID_BLOCK: error = EFAULT; break;
+	case ERROR_INVALID_DATA: error = EINVAL; break;
+	case ERROR_INVALID_DRIVE: error = ENODEV; break;
+	case ERROR_INVALID_EXE_SIGNATURE: error = ENOEXEC; break;
+	case ERROR_INVALID_FLAGS: error = EINVAL; break;
+	case ERROR_INVALID_FUNCTION: error = ENOSYS; break;
+	case ERROR_INVALID_HANDLE: error = EBADF; break;
+	case ERROR_INVALID_LOGON_HOURS: error = EACCES; break;
+	case ERROR_INVALID_NAME: error = EINVAL; break;
+	case ERROR_INVALID_OWNER: error = EINVAL; break;
+	case ERROR_INVALID_PARAMETER: error = EINVAL; break;
+	case ERROR_INVALID_PASSWORD: error = EPERM; break;
+	case ERROR_INVALID_PRIMARY_GROUP: error = EINVAL; break;
+	case ERROR_INVALID_SIGNAL_NUMBER: error = EINVAL; break;
+	case ERROR_INVALID_TARGET_HANDLE: error = EIO; break;
+	case ERROR_INVALID_WORKSTATION: error = EACCES; break;
+	case ERROR_IO_DEVICE: error = EIO; break;
+	case ERROR_IO_INCOMPLETE: error = EINTR; break;
+	case ERROR_LOCKED: error = EBUSY; break;
+	case ERROR_LOCK_VIOLATION: error = EACCES; break;
+	case ERROR_LOGON_FAILURE: error = EACCES; break;
+	case ERROR_MAPPED_ALIGNMENT: error = EINVAL; break;
+	case ERROR_META_EXPANSION_TOO_LONG: error = E2BIG; break;
+	case ERROR_MORE_DATA: error = EPIPE; break;
+	case ERROR_NEGATIVE_SEEK: error = ESPIPE; break;
+	case ERROR_NOACCESS: error = EFAULT; break;
+	case ERROR_NONE_MAPPED: error = EINVAL; break;
+	case ERROR_NOT_ENOUGH_MEMORY: error = ENOMEM; break;
+	case ERROR_NOT_READY: error = EAGAIN; break;
+	case ERROR_NOT_SAME_DEVICE: error = EXDEV; break;
+	case ERROR_NO_DATA: error = EPIPE; break;
+	case ERROR_NO_MORE_SEARCH_HANDLES: error = EIO; break;
+	case ERROR_NO_PROC_SLOTS: error = EAGAIN; break;
+	case ERROR_NO_SUCH_PRIVILEGE: error = EACCES; break;
+	case ERROR_OPEN_FAILED: error = EIO; break;
+	case ERROR_OPEN_FILES: error = EBUSY; break;
+	case ERROR_OPERATION_ABORTED: error = EINTR; break;
+	case ERROR_OUTOFMEMORY: error = ENOMEM; break;
+	case ERROR_PASSWORD_EXPIRED: error = EACCES; break;
+	case ERROR_PATH_BUSY: error = EBUSY; break;
+	case ERROR_PATH_NOT_FOUND: error = ENOENT; break;
+	case ERROR_PIPE_BUSY: error = EBUSY; break;
+	case ERROR_PIPE_CONNECTED: error = EPIPE; break;
+	case ERROR_PIPE_LISTENING: error = EPIPE; break;
+	case ERROR_PIPE_NOT_CONNECTED: error = EPIPE; break;
+	case ERROR_PRIVILEGE_NOT_HELD: error = EACCES; break;
+	case ERROR_READ_FAULT: error = EIO; break;
+	case ERROR_SEEK: error = EIO; break;
+	case ERROR_SEEK_ON_DEVICE: error = ESPIPE; break;
+	case ERROR_SHARING_BUFFER_EXCEEDED: error = ENFILE; break;
+	case ERROR_SHARING_VIOLATION: error = EACCES; break;
+	case ERROR_STACK_OVERFLOW: error = ENOMEM; break;
+	case ERROR_SWAPERROR: error = ENOENT; break;
+	case ERROR_TOO_MANY_MODULES: error = EMFILE; break;
+	case ERROR_TOO_MANY_OPEN_FILES: error = EMFILE; break;
+	case ERROR_UNRECOGNIZED_MEDIA: error = ENXIO; break;
+	case ERROR_UNRECOGNIZED_VOLUME: error = ENODEV; break;
+	case ERROR_WAIT_NO_CHILDREN: error = ECHILD; break;
+	case ERROR_WRITE_FAULT: error = EIO; break;
+	case ERROR_WRITE_PROTECT: error = EROFS; break;
+	}
+	return error;
+}
+
+#undef open
+int mingw_open (const char *filename, int oflags, ...)
+{
+	va_list args;
+	unsigned mode;
+	int fd;
+
+	va_start(args, oflags);
+	mode = va_arg(args, int);
+	va_end(args);
+
+	if (!strcmp(filename, "/dev/null"))
+		filename = "nul";
+
+	fd = open(filename, oflags, mode);
+
+	if (fd < 0 && (oflags & O_CREAT) && errno == EACCES) {
+		DWORD attrs = GetFileAttributes(filename);
+		if (attrs != INVALID_FILE_ATTRIBUTES && (attrs & FILE_ATTRIBUTE_DIRECTORY))
+			errno = EISDIR;
+	}
+	return fd;
+}
+
+#undef write
+ssize_t mingw_write(int fd, const void *buf, size_t count)
+{
+	/*
+	 * While write() calls to a file on a local disk are translated
+	 * into WriteFile() calls with a maximum size of 64KB on Windows
+	 * XP and 256KB on Vista, no such cap is placed on writes to
+	 * files over the network on Windows XP.  Unfortunately, there
+	 * seems to be a limit of 32MB-28KB on X64 and 64MB-32KB on x86;
+	 * bigger writes fail on Windows XP.
+	 * So we cap to a nice 31MB here to avoid write failures over
+	 * the net without changing the number of WriteFile() calls in
+	 * the local case.
+	 */
+	return write(fd, buf, min(count, 31 * 1024 * 1024));
+}
+
+#undef fopen
+FILE *mingw_fopen (const char *filename, const char *otype)
+{
+	if (!strcmp(filename, "/dev/null"))
+		filename = "nul";
+	return fopen(filename, otype);
+}
+
+#undef freopen
+FILE *mingw_freopen (const char *filename, const char *otype, FILE *stream)
+{
+	if (filename && !strcmp(filename, "/dev/null"))
+		filename = "nul";
+	return freopen(filename, otype, stream);
+}
+
+/*
+ * The unit of FILETIME is 100-nanoseconds since January 1, 1601, UTC.
+ * Returns the 100-nanoseconds ("hekto nanoseconds") since the epoch.
+ */
+static inline long long filetime_to_hnsec(const FILETIME *ft)
+{
+	long long winTime = ((long long)ft->dwHighDateTime << 32) + ft->dwLowDateTime;
+	/* Windows to Unix Epoch conversion */
+	return winTime - 116444736000000000LL;
+}
+
+static inline time_t filetime_to_time_t(const FILETIME *ft)
+{
+	return (time_t)(filetime_to_hnsec(ft) / 10000000);
+}
+
+/* We keep the do_lstat code in a separate function to avoid recursion.
+ * When a path ends with a slash, the stat will fail with ENOENT. In
+ * this case, we strip the trailing slashes and stat again.
+ */
+static int do_lstat(const char *file_name, struct stat *buf)
+{
+	WIN32_FILE_ATTRIBUTE_DATA fdata;
+
+	if (!(errno = get_file_attr(file_name, &fdata))) {
+		buf->st_ino = 0;
+		buf->st_gid = 0;
+		buf->st_uid = 0;
+		buf->st_nlink = 1;
+		buf->st_mode = file_attr_to_st_mode(fdata.dwFileAttributes);
+		buf->st_size = fdata.nFileSizeLow |
+			(((off_t)fdata.nFileSizeHigh)<<32);
+		buf->st_dev = buf->st_rdev = 0; /* not used by Git */
+		buf->st_atime = filetime_to_time_t(&(fdata.ftLastAccessTime));
+		buf->st_mtime = filetime_to_time_t(&(fdata.ftLastWriteTime));
+		buf->st_ctime = filetime_to_time_t(&(fdata.ftCreationTime));
+		return 0;
+	}
+	return -1;
+}
+
+/* We provide our own lstat/fstat functions, since the provided
+ * lstat/fstat functions are so slow. These stat functions are
+ * tailored for Git's usage (read: fast), and are not meant to be
+ * complete. Note that Git stat()s are redirected to mingw_lstat()
+ * too, since Windows doesn't really handle symlinks that well.
+ */
+int mingw_lstat(const char *file_name, struct stat *buf)
+{
+	int namelen;
+	static char alt_name[PATH_MAX];
+
+	if (!do_lstat(file_name, buf))
+		return 0;
+
+	/* if file_name ended in a '/', Windows returned ENOENT;
+	 * try again without trailing slashes
+	 */
+	if (errno != ENOENT)
+		return -1;
+
+	namelen = strlen(file_name);
+	if (namelen && file_name[namelen-1] != '/')
+		return -1;
+	while (namelen && file_name[namelen-1] == '/')
+		--namelen;
+	if (!namelen || namelen >= PATH_MAX)
+		return -1;
+
+	memcpy(alt_name, file_name, namelen);
+	alt_name[namelen] = 0;
+	return do_lstat(alt_name, buf);
+}
+
+#undef fstat
+int mingw_fstat(int fd, struct stat *buf)
+{
+	HANDLE fh = (HANDLE)_get_osfhandle(fd);
+	BY_HANDLE_FILE_INFORMATION fdata;
+
+	if (fh == INVALID_HANDLE_VALUE) {
+		errno = EBADF;
+		return -1;
+	}
+	/* direct non-file handles to MS's fstat() */
+	if (GetFileType(fh) != FILE_TYPE_DISK)
+		return _fstati64(fd, buf);
+
+	if (GetFileInformationByHandle(fh, &fdata)) {
+		buf->st_ino = 0;
+		buf->st_gid = 0;
+		buf->st_uid = 0;
+		buf->st_nlink = 1;
+		buf->st_mode = file_attr_to_st_mode(fdata.dwFileAttributes);
+		buf->st_size = fdata.nFileSizeLow |
+			(((off_t)fdata.nFileSizeHigh)<<32);
+		buf->st_dev = buf->st_rdev = 0; /* not used by Git */
+		buf->st_atime = filetime_to_time_t(&(fdata.ftLastAccessTime));
+		buf->st_mtime = filetime_to_time_t(&(fdata.ftLastWriteTime));
+		buf->st_ctime = filetime_to_time_t(&(fdata.ftCreationTime));
+		return 0;
+	}
+	errno = EBADF;
+	return -1;
+}
+
+static inline void time_t_to_filetime(time_t t, FILETIME *ft)
+{
+	long long winTime = t * 10000000LL + 116444736000000000LL;
+	ft->dwLowDateTime = winTime;
+	ft->dwHighDateTime = winTime >> 32;
+}
+
+int mingw_utime (const char *file_name, const struct utimbuf *times)
+{
+	FILETIME mft, aft;
+	int fh, rc;
+
+	/* must have write permission */
+	DWORD attrs = GetFileAttributes(file_name);
+	if (attrs != INVALID_FILE_ATTRIBUTES &&
+	    (attrs & FILE_ATTRIBUTE_READONLY)) {
+		/* ignore errors here; open() will report them */
+		SetFileAttributes(file_name, attrs & ~FILE_ATTRIBUTE_READONLY);
+	}
+
+	if ((fh = open(file_name, O_RDWR | O_BINARY)) < 0) {
+		rc = -1;
+		goto revert_attrs;
+	}
+
+	time_t_to_filetime(times->modtime, &mft);
+	time_t_to_filetime(times->actime, &aft);
+	if (!SetFileTime((HANDLE)_get_osfhandle(fh), NULL, &aft, &mft)) {
+		errno = EINVAL;
+		rc = -1;
+	} else
+		rc = 0;
+	close(fh);
+
+revert_attrs:
+	if (attrs != INVALID_FILE_ATTRIBUTES &&
+	    (attrs & FILE_ATTRIBUTE_READONLY)) {
+		/* ignore errors again */
+		SetFileAttributes(file_name, attrs);
+	}
+	return rc;
+}
+
+unsigned int sleep (unsigned int seconds)
+{
+	Sleep(seconds*1000);
+	return 0;
+}
+
+int mkstemp(char *template)
+{
+	char *filename = mktemp(template);
+	if (filename == NULL)
+		return -1;
+	return open(filename, O_RDWR | O_CREAT, 0600);
+}
+
+int gettimeofday(struct timeval *tv, void *tz)
+{
+	FILETIME ft;
+	long long hnsec;
+
+	GetSystemTimeAsFileTime(&ft);
+	hnsec = filetime_to_hnsec(&ft);
+	tv->tv_sec = hnsec / 10000000;
+	tv->tv_usec = (hnsec % 10000000) / 10;
+	return 0;
+}
+
+int pipe(int filedes[2])
+{
+	HANDLE h[2];
+
+	/* this creates non-inheritable handles */
+	if (!CreatePipe(&h[0], &h[1], NULL, 8192)) {
+		errno = err_win_to_posix(GetLastError());
+		return -1;
+	}
+	filedes[0] = _open_osfhandle((int)h[0], O_NOINHERIT);
+	if (filedes[0] < 0) {
+		CloseHandle(h[0]);
+		CloseHandle(h[1]);
+		return -1;
+	}
+	filedes[1] = _open_osfhandle((int)h[1], O_NOINHERIT);
+	if (filedes[0] < 0) {
+		close(filedes[0]);
+		CloseHandle(h[1]);
+		return -1;
+	}
+	return 0;
+}
+
+int poll(struct pollfd *ufds, unsigned int nfds, int timeout)
+{
+	int i, pending;
+
+	if (timeout >= 0) {
+		if (nfds == 0) {
+			Sleep(timeout);
+			return 0;
+		}
+		return errno = EINVAL, error("poll timeout not supported");
+	}
+
+	/* When there is only one fd to wait for, then we pretend that
+	 * input is available and let the actual wait happen when the
+	 * caller invokes read().
+	 */
+	if (nfds == 1) {
+		if (!(ufds[0].events & POLLIN))
+			return errno = EINVAL, error("POLLIN not set");
+		ufds[0].revents = POLLIN;
+		return 0;
+	}
+
+repeat:
+	pending = 0;
+	for (i = 0; i < nfds; i++) {
+		DWORD avail = 0;
+		HANDLE h = (HANDLE) _get_osfhandle(ufds[i].fd);
+		if (h == INVALID_HANDLE_VALUE)
+			return -1;	/* errno was set */
+
+		if (!(ufds[i].events & POLLIN))
+			return errno = EINVAL, error("POLLIN not set");
+
+		/* this emulation works only for pipes */
+		if (!PeekNamedPipe(h, NULL, 0, NULL, &avail, NULL)) {
+			int err = GetLastError();
+			if (err == ERROR_BROKEN_PIPE) {
+				ufds[i].revents = POLLHUP;
+				pending++;
+			} else {
+				errno = EINVAL;
+				return error("PeekNamedPipe failed,"
+					" GetLastError: %u", err);
+			}
+		} else if (avail) {
+			ufds[i].revents = POLLIN;
+			pending++;
+		} else
+			ufds[i].revents = 0;
+	}
+	if (!pending) {
+		/* The only times that we spin here is when the process
+		 * that is connected through the pipes is waiting for
+		 * its own input data to become available. But since
+		 * the process (pack-objects) is itself CPU intensive,
+		 * it will happily pick up the time slice that we are
+		 * relinquishing here.
+		 */
+		Sleep(0);
+		goto repeat;
+	}
+	return 0;
+}
+
+struct tm *gmtime_r(const time_t *timep, struct tm *result)
+{
+	/* gmtime() in MSVCRT.DLL is thread-safe, but not reentrant */
+	memcpy(result, gmtime(timep), sizeof(struct tm));
+	return result;
+}
+
+struct tm *localtime_r(const time_t *timep, struct tm *result)
+{
+	/* localtime() in MSVCRT.DLL is thread-safe, but not reentrant */
+	memcpy(result, localtime(timep), sizeof(struct tm));
+	return result;
+}
+
+#undef getcwd
+char *mingw_getcwd(char *pointer, int len)
+{
+	int i;
+	char *ret = getcwd(pointer, len);
+	if (!ret)
+		return ret;
+	for (i = 0; pointer[i]; i++)
+		if (pointer[i] == '\\')
+			pointer[i] = '/';
+	return ret;
+}
+
+#undef getenv
+char *mingw_getenv(const char *name)
+{
+	char *result = getenv(name);
+	if (!result && !strcmp(name, "TMPDIR")) {
+		/* on Windows it is TMP and TEMP */
+		result = getenv("TMP");
+		if (!result)
+			result = getenv("TEMP");
+	}
+	return result;
+}
+
+/*
+ * See http://msdn2.microsoft.com/en-us/library/17w5ykft(vs.71).aspx
+ * (Parsing C++ Command-Line Arguments)
+ */
+static const char *quote_arg(const char *arg)
+{
+	/* count chars to quote */
+	int len = 0, n = 0;
+	int force_quotes = 0;
+	char *q, *d;
+	const char *p = arg;
+	if (!*p) force_quotes = 1;
+	while (*p) {
+		if (isspace(*p) || *p == '*' || *p == '?' || *p == '{' || *p == '\'')
+			force_quotes = 1;
+		else if (*p == '"')
+			n++;
+		else if (*p == '\\') {
+			int count = 0;
+			while (*p == '\\') {
+				count++;
+				p++;
+				len++;
+			}
+			if (*p == '"')
+				n += count*2 + 1;
+			continue;
+		}
+		len++;
+		p++;
+	}
+	if (!force_quotes && n == 0)
+		return arg;
+
+	/* insert \ where necessary */
+	d = q = xmalloc(len+n+3);
+	*d++ = '"';
+	while (*arg) {
+		if (*arg == '"')
+			*d++ = '\\';
+		else if (*arg == '\\') {
+			int count = 0;
+			while (*arg == '\\') {
+				count++;
+				*d++ = *arg++;
+			}
+			if (*arg == '"') {
+				while (count-- > 0)
+					*d++ = '\\';
+				*d++ = '\\';
+			}
+		}
+		*d++ = *arg++;
+	}
+	*d++ = '"';
+	*d++ = 0;
+	return q;
+}
+
+static const char *parse_interpreter(const char *cmd)
+{
+	static char buf[100];
+	char *p, *opt;
+	int n, fd;
+
+	/* don't even try a .exe */
+	n = strlen(cmd);
+	if (n >= 4 && !strcasecmp(cmd+n-4, ".exe"))
+		return NULL;
+
+	fd = open(cmd, O_RDONLY);
+	if (fd < 0)
+		return NULL;
+	n = read(fd, buf, sizeof(buf)-1);
+	close(fd);
+	if (n < 4)	/* at least '#!/x' and not error */
+		return NULL;
+
+	if (buf[0] != '#' || buf[1] != '!')
+		return NULL;
+	buf[n] = '\0';
+	p = buf + strcspn(buf, "\r\n");
+	if (!*p)
+		return NULL;
+
+	*p = '\0';
+	if (!(p = strrchr(buf+2, '/')) && !(p = strrchr(buf+2, '\\')))
+		return NULL;
+	/* strip options */
+	if ((opt = strchr(p+1, ' ')))
+		*opt = '\0';
+	return p+1;
+}
+
+/*
+ * Splits the PATH into parts.
+ */
+static char **get_path_split(void)
+{
+	char *p, **path, *envpath = getenv("PATH");
+	int i, n = 0;
+
+	if (!envpath || !*envpath)
+		return NULL;
+
+	envpath = xstrdup(envpath);
+	p = envpath;
+	while (p) {
+		char *dir = p;
+		p = strchr(p, ';');
+		if (p) *p++ = '\0';
+		if (*dir) {	/* not earlier, catches series of ; */
+			++n;
+		}
+	}
+	if (!n)
+		return NULL;
+
+	path = xmalloc((n+1)*sizeof(char *));
+	p = envpath;
+	i = 0;
+	do {
+		if (*p)
+			path[i++] = xstrdup(p);
+		p = p+strlen(p)+1;
+	} while (i < n);
+	path[i] = NULL;
+
+	free(envpath);
+
+	return path;
+}
+
+static void free_path_split(char **path)
+{
+	char **p = path;
+
+	if (!path)
+		return;
+
+	while (*p)
+		free(*p++);
+	free(path);
+}
+
+/*
+ * exe_only means that we only want to detect .exe files, but not scripts
+ * (which do not have an extension)
+ */
+static char *lookup_prog(const char *dir, const char *cmd, int isexe, int exe_only)
+{
+	char path[MAX_PATH];
+	snprintf(path, sizeof(path), "%s/%s.exe", dir, cmd);
+
+	if (!isexe && access(path, F_OK) == 0)
+		return xstrdup(path);
+	path[strlen(path)-4] = '\0';
+	if ((!exe_only || isexe) && access(path, F_OK) == 0)
+		if (!(GetFileAttributes(path) & FILE_ATTRIBUTE_DIRECTORY))
+			return xstrdup(path);
+	return NULL;
+}
+
+/*
+ * Determines the absolute path of cmd using the the split path in path.
+ * If cmd contains a slash or backslash, no lookup is performed.
+ */
+static char *path_lookup(const char *cmd, char **path, int exe_only)
+{
+	char *prog = NULL;
+	int len = strlen(cmd);
+	int isexe = len >= 4 && !strcasecmp(cmd+len-4, ".exe");
+
+	if (strchr(cmd, '/') || strchr(cmd, '\\'))
+		prog = xstrdup(cmd);
+
+	while (!prog && *path)
+		prog = lookup_prog(*path++, cmd, isexe, exe_only);
+
+	return prog;
+}
+
+static int env_compare(const void *a, const void *b)
+{
+	char *const *ea = a;
+	char *const *eb = b;
+	return strcasecmp(*ea, *eb);
+}
+
+static pid_t mingw_spawnve_fd(const char *cmd, const char **argv, char **env,
+			      const char *dir,
+			      int prepend_cmd, int fhin, int fhout, int fherr)
+{
+	STARTUPINFO si;
+	PROCESS_INFORMATION pi;
+	struct strbuf envblk, args;
+	unsigned flags;
+	BOOL ret;
+
+	/* Determine whether or not we are associated to a console */
+	HANDLE cons = CreateFile("CONOUT$", GENERIC_WRITE,
+			FILE_SHARE_WRITE, NULL, OPEN_EXISTING,
+			FILE_ATTRIBUTE_NORMAL, NULL);
+	if (cons == INVALID_HANDLE_VALUE) {
+		/* There is no console associated with this process.
+		 * Since the child is a console process, Windows
+		 * would normally create a console window. But
+		 * since we'll be redirecting std streams, we do
+		 * not need the console.
+		 * It is necessary to use DETACHED_PROCESS
+		 * instead of CREATE_NO_WINDOW to make ssh
+		 * recognize that it has no console.
+		 */
+		flags = DETACHED_PROCESS;
+	} else {
+		/* There is already a console. If we specified
+		 * DETACHED_PROCESS here, too, Windows would
+		 * disassociate the child from the console.
+		 * The same is true for CREATE_NO_WINDOW.
+		 * Go figure!
+		 */
+		flags = 0;
+		CloseHandle(cons);
+	}
+	memset(&si, 0, sizeof(si));
+	si.cb = sizeof(si);
+	si.dwFlags = STARTF_USESTDHANDLES;
+	si.hStdInput = (HANDLE) _get_osfhandle(fhin);
+	si.hStdOutput = (HANDLE) _get_osfhandle(fhout);
+	si.hStdError = (HANDLE) _get_osfhandle(fherr);
+
+	/* concatenate argv, quoting args as we go */
+	strbuf_init(&args, 0);
+	if (prepend_cmd) {
+		char *quoted = (char *)quote_arg(cmd);
+		strbuf_addstr(&args, quoted);
+		if (quoted != cmd)
+			free(quoted);
+	}
+	for (; *argv; argv++) {
+		char *quoted = (char *)quote_arg(*argv);
+		if (*args.buf)
+			strbuf_addch(&args, ' ');
+		strbuf_addstr(&args, quoted);
+		if (quoted != *argv)
+			free(quoted);
+	}
+
+	if (env) {
+		int count = 0;
+		char **e, **sorted_env;
+
+		for (e = env; *e; e++)
+			count++;
+
+		/* environment must be sorted */
+		sorted_env = xmalloc(sizeof(*sorted_env) * (count + 1));
+		memcpy(sorted_env, env, sizeof(*sorted_env) * (count + 1));
+		qsort(sorted_env, count, sizeof(*sorted_env), env_compare);
+
+		strbuf_init(&envblk, 0);
+		for (e = sorted_env; *e; e++) {
+			strbuf_addstr(&envblk, *e);
+			strbuf_addch(&envblk, '\0');
+		}
+		free(sorted_env);
+	}
+
+	memset(&pi, 0, sizeof(pi));
+	ret = CreateProcess(cmd, args.buf, NULL, NULL, TRUE, flags,
+		env ? envblk.buf : NULL, dir, &si, &pi);
+
+	if (env)
+		strbuf_release(&envblk);
+	strbuf_release(&args);
+
+	if (!ret) {
+		errno = ENOENT;
+		return -1;
+	}
+	CloseHandle(pi.hThread);
+	return (pid_t)pi.hProcess;
+}
+
+static pid_t mingw_spawnve(const char *cmd, const char **argv, char **env,
+			   int prepend_cmd)
+{
+	return mingw_spawnve_fd(cmd, argv, env, NULL, prepend_cmd, 0, 1, 2);
+}
+
+pid_t mingw_spawnvpe(const char *cmd, const char **argv, char **env,
+		     const char *dir,
+		     int fhin, int fhout, int fherr)
+{
+	pid_t pid;
+	char **path = get_path_split();
+	char *prog = path_lookup(cmd, path, 0);
+
+	if (!prog) {
+		errno = ENOENT;
+		pid = -1;
+	}
+	else {
+		const char *interpr = parse_interpreter(prog);
+
+		if (interpr) {
+			const char *argv0 = argv[0];
+			char *iprog = path_lookup(interpr, path, 1);
+			argv[0] = prog;
+			if (!iprog) {
+				errno = ENOENT;
+				pid = -1;
+			}
+			else {
+				pid = mingw_spawnve_fd(iprog, argv, env, dir, 1,
+						       fhin, fhout, fherr);
+				free(iprog);
+			}
+			argv[0] = argv0;
+		}
+		else
+			pid = mingw_spawnve_fd(prog, argv, env, dir, 0,
+					       fhin, fhout, fherr);
+		free(prog);
+	}
+	free_path_split(path);
+	return pid;
+}
+
+static int try_shell_exec(const char *cmd, char *const *argv, char **env)
+{
+	const char *interpr = parse_interpreter(cmd);
+	char **path;
+	char *prog;
+	int pid = 0;
+
+	if (!interpr)
+		return 0;
+	path = get_path_split();
+	prog = path_lookup(interpr, path, 1);
+	if (prog) {
+		int argc = 0;
+		const char **argv2;
+		while (argv[argc]) argc++;
+		argv2 = xmalloc(sizeof(*argv) * (argc+1));
+		argv2[0] = (char *)cmd;	/* full path to the script file */
+		memcpy(&argv2[1], &argv[1], sizeof(*argv) * argc);
+		pid = mingw_spawnve(prog, argv2, env, 1);
+		if (pid >= 0) {
+			int status;
+			if (waitpid(pid, &status, 0) < 0)
+				status = 255;
+			exit(status);
+		}
+		pid = 1;	/* indicate that we tried but failed */
+		free(prog);
+		free(argv2);
+	}
+	free_path_split(path);
+	return pid;
+}
+
+static void mingw_execve(const char *cmd, char *const *argv, char *const *env)
+{
+	/* check if git_command is a shell script */
+	if (!try_shell_exec(cmd, argv, (char **)env)) {
+		int pid, status;
+
+		pid = mingw_spawnve(cmd, (const char **)argv, (char **)env, 0);
+		if (pid < 0)
+			return;
+		if (waitpid(pid, &status, 0) < 0)
+			status = 255;
+		exit(status);
+	}
+}
+
+void mingw_execvp(const char *cmd, char *const *argv)
+{
+	char **path = get_path_split();
+	char *prog = path_lookup(cmd, path, 0);
+
+	if (prog) {
+		mingw_execve(prog, argv, environ);
+		free(prog);
+	} else
+		errno = ENOENT;
+
+	free_path_split(path);
+}
+
+static char **copy_environ(void)
+{
+	char **env;
+	int i = 0;
+	while (environ[i])
+		i++;
+	env = xmalloc((i+1)*sizeof(*env));
+	for (i = 0; environ[i]; i++)
+		env[i] = xstrdup(environ[i]);
+	env[i] = NULL;
+	return env;
+}
+
+void free_environ(char **env)
+{
+	int i;
+	for (i = 0; env[i]; i++)
+		free(env[i]);
+	free(env);
+}
+
+static int lookup_env(char **env, const char *name, size_t nmln)
+{
+	int i;
+
+	for (i = 0; env[i]; i++) {
+		if (0 == strncmp(env[i], name, nmln)
+		    && '=' == env[i][nmln])
+			/* matches */
+			return i;
+	}
+	return -1;
+}
+
+/*
+ * If name contains '=', then sets the variable, otherwise it unsets it
+ */
+static char **env_setenv(char **env, const char *name)
+{
+	char *eq = strchrnul(name, '=');
+	int i = lookup_env(env, name, eq-name);
+
+	if (i < 0) {
+		if (*eq) {
+			for (i = 0; env[i]; i++)
+				;
+			env = xrealloc(env, (i+2)*sizeof(*env));
+			env[i] = xstrdup(name);
+			env[i+1] = NULL;
+		}
+	}
+	else {
+		free(env[i]);
+		if (*eq)
+			env[i] = xstrdup(name);
+		else
+			for (; env[i]; i++)
+				env[i] = env[i+1];
+	}
+	return env;
+}
+
+/*
+ * Copies global environ and adjusts variables as specified by vars.
+ */
+char **make_augmented_environ(const char *const *vars)
+{
+	char **env = copy_environ();
+
+	while (*vars)
+		env = env_setenv(env, *vars++);
+	return env;
+}
+
+/*
+ * Note, this isn't a complete replacement for getaddrinfo. It assumes
+ * that service contains a numerical port, or that it it is null. It
+ * does a simple search using gethostbyname, and returns one IPv4 host
+ * if one was found.
+ */
+static int WSAAPI getaddrinfo_stub(const char *node, const char *service,
+				   const struct addrinfo *hints,
+				   struct addrinfo **res)
+{
+	struct hostent *h = gethostbyname(node);
+	struct addrinfo *ai;
+	struct sockaddr_in *sin;
+
+	if (!h)
+		return WSAGetLastError();
+
+	ai = xmalloc(sizeof(struct addrinfo));
+	*res = ai;
+	ai->ai_flags = 0;
+	ai->ai_family = AF_INET;
+	ai->ai_socktype = hints->ai_socktype;
+	switch (hints->ai_socktype) {
+	case SOCK_STREAM:
+		ai->ai_protocol = IPPROTO_TCP;
+		break;
+	case SOCK_DGRAM:
+		ai->ai_protocol = IPPROTO_UDP;
+		break;
+	default:
+		ai->ai_protocol = 0;
+		break;
+	}
+	ai->ai_addrlen = sizeof(struct sockaddr_in);
+	ai->ai_canonname = strdup(h->h_name);
+
+	sin = xmalloc(ai->ai_addrlen);
+	memset(sin, 0, ai->ai_addrlen);
+	sin->sin_family = AF_INET;
+	if (service)
+		sin->sin_port = htons(atoi(service));
+	sin->sin_addr = *(struct in_addr *)h->h_addr;
+	ai->ai_addr = (struct sockaddr *)sin;
+	ai->ai_next = 0;
+	return 0;
+}
+
+static void WSAAPI freeaddrinfo_stub(struct addrinfo *res)
+{
+	free(res->ai_canonname);
+	free(res->ai_addr);
+	free(res);
+}
+
+static int WSAAPI getnameinfo_stub(const struct sockaddr *sa, socklen_t salen,
+				   char *host, DWORD hostlen,
+				   char *serv, DWORD servlen, int flags)
+{
+	const struct sockaddr_in *sin = (const struct sockaddr_in *)sa;
+	if (sa->sa_family != AF_INET)
+		return EAI_FAMILY;
+	if (!host && !serv)
+		return EAI_NONAME;
+
+	if (host && hostlen > 0) {
+		struct hostent *ent = NULL;
+		if (!(flags & NI_NUMERICHOST))
+			ent = gethostbyaddr((const char *)&sin->sin_addr,
+					    sizeof(sin->sin_addr), AF_INET);
+
+		if (ent)
+			snprintf(host, hostlen, "%s", ent->h_name);
+		else if (flags & NI_NAMEREQD)
+			return EAI_NONAME;
+		else
+			snprintf(host, hostlen, "%s", inet_ntoa(sin->sin_addr));
+	}
+
+	if (serv && servlen > 0) {
+		struct servent *ent = NULL;
+		if (!(flags & NI_NUMERICSERV))
+			ent = getservbyport(sin->sin_port,
+					    flags & NI_DGRAM ? "udp" : "tcp");
+
+		if (ent)
+			snprintf(serv, servlen, "%s", ent->s_name);
+		else
+			snprintf(serv, servlen, "%d", ntohs(sin->sin_port));
+	}
+
+	return 0;
+}
+
+static HMODULE ipv6_dll = NULL;
+static void (WSAAPI *ipv6_freeaddrinfo)(struct addrinfo *res);
+static int (WSAAPI *ipv6_getaddrinfo)(const char *node, const char *service,
+				      const struct addrinfo *hints,
+				      struct addrinfo **res);
+static int (WSAAPI *ipv6_getnameinfo)(const struct sockaddr *sa, socklen_t salen,
+				      char *host, DWORD hostlen,
+				      char *serv, DWORD servlen, int flags);
+/*
+ * gai_strerror is an inline function in the ws2tcpip.h header, so we
+ * don't need to try to load that one dynamically.
+ */
+
+static void socket_cleanup(void)
+{
+	WSACleanup();
+	if (ipv6_dll)
+		FreeLibrary(ipv6_dll);
+	ipv6_dll = NULL;
+	ipv6_freeaddrinfo = freeaddrinfo_stub;
+	ipv6_getaddrinfo = getaddrinfo_stub;
+	ipv6_getnameinfo = getnameinfo_stub;
+}
+
+static void ensure_socket_initialization(void)
+{
+	WSADATA wsa;
+	static int initialized = 0;
+	const char *libraries[] = { "ws2_32.dll", "wship6.dll", NULL };
+	const char **name;
+
+	if (initialized)
+		return;
+
+	if (WSAStartup(MAKEWORD(2,2), &wsa))
+		die("unable to initialize winsock subsystem, error %d",
+			WSAGetLastError());
+
+	for (name = libraries; *name; name++) {
+		ipv6_dll = LoadLibrary(*name);
+		if (!ipv6_dll)
+			continue;
+
+		ipv6_freeaddrinfo = (void (WSAAPI *)(struct addrinfo *))
+			GetProcAddress(ipv6_dll, "freeaddrinfo");
+		ipv6_getaddrinfo = (int (WSAAPI *)(const char *, const char *,
+						   const struct addrinfo *,
+						   struct addrinfo **))
+			GetProcAddress(ipv6_dll, "getaddrinfo");
+		ipv6_getnameinfo = (int (WSAAPI *)(const struct sockaddr *,
+						   socklen_t, char *, DWORD,
+						   char *, DWORD, int))
+			GetProcAddress(ipv6_dll, "getnameinfo");
+		if (!ipv6_freeaddrinfo || !ipv6_getaddrinfo || !ipv6_getnameinfo) {
+			FreeLibrary(ipv6_dll);
+			ipv6_dll = NULL;
+		} else
+			break;
+	}
+	if (!ipv6_freeaddrinfo || !ipv6_getaddrinfo || !ipv6_getnameinfo) {
+		ipv6_freeaddrinfo = freeaddrinfo_stub;
+		ipv6_getaddrinfo = getaddrinfo_stub;
+		ipv6_getnameinfo = getnameinfo_stub;
+	}
+
+	atexit(socket_cleanup);
+	initialized = 1;
+}
+
+#undef gethostbyname
+struct hostent *mingw_gethostbyname(const char *host)
+{
+	ensure_socket_initialization();
+	return gethostbyname(host);
+}
+
+void mingw_freeaddrinfo(struct addrinfo *res)
+{
+	ipv6_freeaddrinfo(res);
+}
+
+int mingw_getaddrinfo(const char *node, const char *service,
+		      const struct addrinfo *hints, struct addrinfo **res)
+{
+	ensure_socket_initialization();
+	return ipv6_getaddrinfo(node, service, hints, res);
+}
+
+int mingw_getnameinfo(const struct sockaddr *sa, socklen_t salen,
+		      char *host, DWORD hostlen, char *serv, DWORD servlen,
+		      int flags)
+{
+	ensure_socket_initialization();
+	return ipv6_getnameinfo(sa, salen, host, hostlen, serv, servlen, flags);
+}
+
+int mingw_socket(int domain, int type, int protocol)
+{
+	int sockfd;
+	SOCKET s = WSASocket(domain, type, protocol, NULL, 0, 0);
+	if (s == INVALID_SOCKET) {
+		/*
+		 * WSAGetLastError() values are regular BSD error codes
+		 * biased by WSABASEERR.
+		 * However, strerror() does not know about networking
+		 * specific errors, which are values beginning at 38 or so.
+		 * Therefore, we choose to leave the biased error code
+		 * in errno so that _if_ someone looks up the code somewhere,
+		 * then it is at least the number that are usually listed.
+		 */
+		errno = WSAGetLastError();
+		return -1;
+	}
+	/* convert into a file descriptor */
+	if ((sockfd = _open_osfhandle(s, O_RDWR|O_BINARY)) < 0) {
+		closesocket(s);
+		return error("unable to make a socket file descriptor: %s",
+			strerror(errno));
+	}
+	return sockfd;
+}
+
+#undef connect
+int mingw_connect(int sockfd, struct sockaddr *sa, size_t sz)
+{
+	SOCKET s = (SOCKET)_get_osfhandle(sockfd);
+	return connect(s, sa, sz);
+}
+
+#undef rename
+int mingw_rename(const char *pold, const char *pnew)
+{
+	DWORD attrs, gle;
+	int tries = 0;
+	static const int delay[] = { 0, 1, 10, 20, 40 };
+
+	/*
+	 * Try native rename() first to get errno right.
+	 * It is based on MoveFile(), which cannot overwrite existing files.
+	 */
+	if (!rename(pold, pnew))
+		return 0;
+	if (errno != EEXIST)
+		return -1;
+repeat:
+	if (MoveFileEx(pold, pnew, MOVEFILE_REPLACE_EXISTING))
+		return 0;
+	/* TODO: translate more errors */
+	gle = GetLastError();
+	if (gle == ERROR_ACCESS_DENIED &&
+	    (attrs = GetFileAttributes(pnew)) != INVALID_FILE_ATTRIBUTES) {
+		if (attrs & FILE_ATTRIBUTE_DIRECTORY) {
+			errno = EISDIR;
+			return -1;
+		}
+		if ((attrs & FILE_ATTRIBUTE_READONLY) &&
+		    SetFileAttributes(pnew, attrs & ~FILE_ATTRIBUTE_READONLY)) {
+			if (MoveFileEx(pold, pnew, MOVEFILE_REPLACE_EXISTING))
+				return 0;
+			gle = GetLastError();
+			/* revert file attributes on failure */
+			SetFileAttributes(pnew, attrs);
+		}
+	}
+	if (tries < ARRAY_SIZE(delay) && gle == ERROR_ACCESS_DENIED) {
+		/*
+		 * We assume that some other process had the source or
+		 * destination file open at the wrong moment and retry.
+		 * In order to give the other process a higher chance to
+		 * complete its operation, we give up our time slice now.
+		 * If we have to retry again, we do sleep a bit.
+		 */
+		Sleep(delay[tries]);
+		tries++;
+		goto repeat;
+	}
+	errno = EACCES;
+	return -1;
+}
+
+/*
+ * Note that this doesn't return the actual pagesize, but
+ * the allocation granularity. If future Windows specific git code
+ * needs the real getpagesize function, we need to find another solution.
+ */
+int mingw_getpagesize(void)
+{
+	SYSTEM_INFO si;
+	GetSystemInfo(&si);
+	return si.dwAllocationGranularity;
+}
+
+struct passwd *getpwuid(int uid)
+{
+	static char user_name[100];
+	static struct passwd p;
+
+	DWORD len = sizeof(user_name);
+	if (!GetUserName(user_name, &len))
+		return NULL;
+	p.pw_name = user_name;
+	p.pw_gecos = "unknown";
+	p.pw_dir = NULL;
+	return &p;
+}
+
+static HANDLE timer_event;
+static HANDLE timer_thread;
+static int timer_interval;
+static int one_shot;
+static sig_handler_t timer_fn = SIG_DFL;
+
+/* The timer works like this:
+ * The thread, ticktack(), is a trivial routine that most of the time
+ * only waits to receive the signal to terminate. The main thread tells
+ * the thread to terminate by setting the timer_event to the signalled
+ * state.
+ * But ticktack() interrupts the wait state after the timer's interval
+ * length to call the signal handler.
+ */
+
+static unsigned __stdcall ticktack(void *dummy)
+{
+	while (WaitForSingleObject(timer_event, timer_interval) == WAIT_TIMEOUT) {
+		if (timer_fn == SIG_DFL)
+			die("Alarm");
+		if (timer_fn != SIG_IGN)
+			timer_fn(SIGALRM);
+		if (one_shot)
+			break;
+	}
+	return 0;
+}
+
+static int start_timer_thread(void)
+{
+	timer_event = CreateEvent(NULL, FALSE, FALSE, NULL);
+	if (timer_event) {
+		timer_thread = (HANDLE) _beginthreadex(NULL, 0, ticktack, NULL, 0, NULL);
+		if (!timer_thread )
+			return errno = ENOMEM,
+				error("cannot start timer thread");
+	} else
+		return errno = ENOMEM,
+			error("cannot allocate resources for timer");
+	return 0;
+}
+
+static void stop_timer_thread(void)
+{
+	if (timer_event)
+		SetEvent(timer_event);	/* tell thread to terminate */
+	if (timer_thread) {
+		int rc = WaitForSingleObject(timer_thread, 1000);
+		if (rc == WAIT_TIMEOUT)
+			error("timer thread did not terminate timely");
+		else if (rc != WAIT_OBJECT_0)
+			error("waiting for timer thread failed: %lu",
+			      GetLastError());
+		CloseHandle(timer_thread);
+	}
+	if (timer_event)
+		CloseHandle(timer_event);
+	timer_event = NULL;
+	timer_thread = NULL;
+}
+
+static inline int is_timeval_eq(const struct timeval *i1, const struct timeval *i2)
+{
+	return i1->tv_sec == i2->tv_sec && i1->tv_usec == i2->tv_usec;
+}
+
+int setitimer(int type, struct itimerval *in, struct itimerval *out)
+{
+	static const struct timeval zero;
+	static int atexit_done;
+
+	if (out != NULL)
+		return errno = EINVAL,
+			error("setitimer param 3 != NULL not implemented");
+	if (!is_timeval_eq(&in->it_interval, &zero) &&
+	    !is_timeval_eq(&in->it_interval, &in->it_value))
+		return errno = EINVAL,
+			error("setitimer: it_interval must be zero or eq it_value");
+
+	if (timer_thread)
+		stop_timer_thread();
+
+	if (is_timeval_eq(&in->it_value, &zero) &&
+	    is_timeval_eq(&in->it_interval, &zero))
+		return 0;
+
+	timer_interval = in->it_value.tv_sec * 1000 + in->it_value.tv_usec / 1000;
+	one_shot = is_timeval_eq(&in->it_interval, &zero);
+	if (!atexit_done) {
+		atexit(stop_timer_thread);
+		atexit_done = 1;
+	}
+	return start_timer_thread();
+}
+
+int sigaction(int sig, struct sigaction *in, struct sigaction *out)
+{
+	if (sig != SIGALRM)
+		return errno = EINVAL,
+			error("sigaction only implemented for SIGALRM");
+	if (out != NULL)
+		return errno = EINVAL,
+			error("sigaction: param 3 != NULL not implemented");
+
+	timer_fn = in->sa_handler;
+	return 0;
+}
+
+#undef signal
+sig_handler_t mingw_signal(int sig, sig_handler_t handler)
+{
+	sig_handler_t old = timer_fn;
+	if (sig != SIGALRM)
+		return signal(sig, handler);
+	timer_fn = handler;
+	return old;
+}
+
+static const char *make_backslash_path(const char *path)
+{
+	static char buf[PATH_MAX + 1];
+	char *c;
+
+	if (strlcpy(buf, path, PATH_MAX) >= PATH_MAX)
+		die("Too long path: %.*s", 60, path);
+
+	for (c = buf; *c; c++) {
+		if (*c == '/')
+			*c = '\\';
+	}
+	return buf;
+}
+
+void mingw_open_html(const char *unixpath)
+{
+	const char *htmlpath = make_backslash_path(unixpath);
+	typedef HINSTANCE (WINAPI *T)(HWND, const char *,
+			const char *, const char *, const char *, INT);
+	T ShellExecute;
+	HMODULE shell32;
+
+	shell32 = LoadLibrary("shell32.dll");
+	if (!shell32)
+		die("cannot load shell32.dll");
+	ShellExecute = (T)GetProcAddress(shell32, "ShellExecuteA");
+	if (!ShellExecute)
+		die("cannot run browser");
+
+	printf("Launching default browser to display HTML ...\n");
+	ShellExecute(NULL, "open", htmlpath, NULL, "\\", 0);
+
+	FreeLibrary(shell32);
+}
+
+int link(const char *oldpath, const char *newpath)
+{
+	typedef BOOL (WINAPI *T)(const char*, const char*, LPSECURITY_ATTRIBUTES);
+	static T create_hard_link = NULL;
+	if (!create_hard_link) {
+		create_hard_link = (T) GetProcAddress(
+			GetModuleHandle("kernel32.dll"), "CreateHardLinkA");
+		if (!create_hard_link)
+			create_hard_link = (T)-1;
+	}
+	if (create_hard_link == (T)-1) {
+		errno = ENOSYS;
+		return -1;
+	}
+	if (!create_hard_link(newpath, oldpath, NULL)) {
+		errno = err_win_to_posix(GetLastError());
+		return -1;
+	}
+	return 0;
+}
+
+char *getpass(const char *prompt)
+{
+	struct strbuf buf = STRBUF_INIT;
+
+	fputs(prompt, stderr);
+	for (;;) {
+		char c = _getch();
+		if (c == '\r' || c == '\n')
+			break;
+		strbuf_addch(&buf, c);
+	}
+	fputs("\n", stderr);
+	return strbuf_detach(&buf, NULL);
+}
+
+#ifndef NO_MINGW_REPLACE_READDIR
+/* MinGW readdir implementation to avoid extra lstats for Git */
+struct mingw_DIR
+{
+	struct _finddata_t	dd_dta;		/* disk transfer area for this dir */
+	struct mingw_dirent	dd_dir;		/* Our own implementation, including d_type */
+	long			dd_handle;	/* _findnext handle */
+	int			dd_stat; 	/* 0 = next entry to read is first entry, -1 = off the end, positive = 0 based index of next entry */
+	char			dd_name[1]; 	/* given path for dir with search pattern (struct is extended) */
+};
+
+struct dirent *mingw_readdir(DIR *dir)
+{
+	WIN32_FIND_DATAA buf;
+	HANDLE handle;
+	struct mingw_DIR *mdir = (struct mingw_DIR*)dir;
+
+	if (!dir->dd_handle) {
+		errno = EBADF; /* No set_errno for mingw */
+		return NULL;
+	}
+
+	if (dir->dd_handle == (long)INVALID_HANDLE_VALUE && dir->dd_stat == 0)
+	{
+		DWORD lasterr;
+		handle = FindFirstFileA(dir->dd_name, &buf);
+		lasterr = GetLastError();
+		dir->dd_handle = (long)handle;
+		if (handle == INVALID_HANDLE_VALUE && (lasterr != ERROR_NO_MORE_FILES)) {
+			errno = err_win_to_posix(lasterr);
+			return NULL;
+		}
+	} else if (dir->dd_handle == (long)INVALID_HANDLE_VALUE) {
+		return NULL;
+	} else if (!FindNextFileA((HANDLE)dir->dd_handle, &buf)) {
+		DWORD lasterr = GetLastError();
+		FindClose((HANDLE)dir->dd_handle);
+		dir->dd_handle = (long)INVALID_HANDLE_VALUE;
+		/* POSIX says you shouldn't set errno when readdir can't
+		   find any more files; so, if another error we leave it set. */
+		if (lasterr != ERROR_NO_MORE_FILES)
+			errno = err_win_to_posix(lasterr);
+		return NULL;
+	}
+
+	/* We get here if `buf' contains valid data.  */
+	strcpy(dir->dd_dir.d_name, buf.cFileName);
+	++dir->dd_stat;
+
+	/* Set file type, based on WIN32_FIND_DATA */
+	mdir->dd_dir.d_type = 0;
+	if (buf.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY)
+		mdir->dd_dir.d_type |= DT_DIR;
+	else
+		mdir->dd_dir.d_type |= DT_REG;
+
+	return (struct dirent*)&dir->dd_dir;
+}
+#endif // !NO_MINGW_REPLACE_READDIR
diff --git a/compat/mingw.h b/compat/mingw.h
new file mode 100644
index 0000000..3b2477b
--- /dev/null
+++ b/compat/mingw.h
@@ -0,0 +1,327 @@
+#include <winsock2.h>
+#include <ws2tcpip.h>
+
+/*
+ * things that are not available in header files
+ */
+
+typedef int pid_t;
+#define hstrerror strerror
+
+#define S_IFLNK    0120000 /* Symbolic link */
+#define S_ISLNK(x) (((x) & S_IFMT) == S_IFLNK)
+#define S_ISSOCK(x) 0
+#define S_IRGRP 0
+#define S_IWGRP 0
+#define S_IXGRP 0
+#define S_ISGID 0
+#define S_IROTH 0
+#define S_IXOTH 0
+
+#define WIFEXITED(x) 1
+#define WIFSIGNALED(x) 0
+#define WEXITSTATUS(x) ((x) & 0xff)
+#define WTERMSIG(x) SIGTERM
+
+#define SIGHUP 1
+#define SIGQUIT 3
+#define SIGKILL 9
+#define SIGPIPE 13
+#define SIGALRM 14
+#define SIGCHLD 17
+
+#define F_GETFD 1
+#define F_SETFD 2
+#define FD_CLOEXEC 0x1
+
+struct passwd {
+	char *pw_name;
+	char *pw_gecos;
+	char *pw_dir;
+};
+
+extern char *getpass(const char *prompt);
+
+#ifndef POLLIN
+struct pollfd {
+	int fd;           /* file descriptor */
+	short events;     /* requested events */
+	short revents;    /* returned events */
+};
+#define POLLIN 1
+#define POLLHUP 2
+#endif
+
+typedef void (__cdecl *sig_handler_t)(int);
+struct sigaction {
+	sig_handler_t sa_handler;
+	unsigned sa_flags;
+};
+#define sigemptyset(x) (void)0
+#define SA_RESTART 0
+
+struct itimerval {
+	struct timeval it_value, it_interval;
+};
+#define ITIMER_REAL 0
+
+/*
+ * trivial stubs
+ */
+
+static inline int readlink(const char *path, char *buf, size_t bufsiz)
+{ errno = ENOSYS; return -1; }
+static inline int symlink(const char *oldpath, const char *newpath)
+{ errno = ENOSYS; return -1; }
+static inline int fchmod(int fildes, mode_t mode)
+{ errno = ENOSYS; return -1; }
+static inline int fork(void)
+{ errno = ENOSYS; return -1; }
+static inline unsigned int alarm(unsigned int seconds)
+{ return 0; }
+static inline int fsync(int fd)
+{ return _commit(fd); }
+static inline int getppid(void)
+{ return 1; }
+static inline void sync(void)
+{}
+static inline int getuid()
+{ return 1; }
+static inline struct passwd *getpwnam(const char *name)
+{ return NULL; }
+static inline int fcntl(int fd, int cmd, ...)
+{
+	if (cmd == F_GETFD || cmd == F_SETFD)
+		return 0;
+	errno = EINVAL;
+	return -1;
+}
+/* bash cannot reliably detect negative return codes as failure */
+#define exit(code) exit((code) & 0xff)
+
+/*
+ * simple adaptors
+ */
+
+static inline int mingw_mkdir(const char *path, int mode)
+{
+	return mkdir(path);
+}
+#define mkdir mingw_mkdir
+
+static inline int mingw_unlink(const char *pathname)
+{
+	/* read-only files cannot be removed */
+	chmod(pathname, 0666);
+	return unlink(pathname);
+}
+#define unlink mingw_unlink
+
+static inline int waitpid(pid_t pid, int *status, unsigned options)
+{
+	if (options == 0)
+		return _cwait(status, pid, 0);
+	errno = EINVAL;
+	return -1;
+}
+
+#ifndef NO_OPENSSL
+#include <openssl/ssl.h>
+static inline int mingw_SSL_set_fd(SSL *ssl, int fd)
+{
+	return SSL_set_fd(ssl, _get_osfhandle(fd));
+}
+#define SSL_set_fd mingw_SSL_set_fd
+
+static inline int mingw_SSL_set_rfd(SSL *ssl, int fd)
+{
+	return SSL_set_rfd(ssl, _get_osfhandle(fd));
+}
+#define SSL_set_rfd mingw_SSL_set_rfd
+
+static inline int mingw_SSL_set_wfd(SSL *ssl, int fd)
+{
+	return SSL_set_wfd(ssl, _get_osfhandle(fd));
+}
+#define SSL_set_wfd mingw_SSL_set_wfd
+#endif
+
+/*
+ * implementations of missing functions
+ */
+
+int pipe(int filedes[2]);
+unsigned int sleep (unsigned int seconds);
+int mkstemp(char *template);
+int gettimeofday(struct timeval *tv, void *tz);
+int poll(struct pollfd *ufds, unsigned int nfds, int timeout);
+struct tm *gmtime_r(const time_t *timep, struct tm *result);
+struct tm *localtime_r(const time_t *timep, struct tm *result);
+int getpagesize(void);	/* defined in MinGW's libgcc.a */
+struct passwd *getpwuid(int uid);
+int setitimer(int type, struct itimerval *in, struct itimerval *out);
+int sigaction(int sig, struct sigaction *in, struct sigaction *out);
+int link(const char *oldpath, const char *newpath);
+
+/*
+ * replacements of existing functions
+ */
+
+int mingw_open (const char *filename, int oflags, ...);
+#define open mingw_open
+
+ssize_t mingw_write(int fd, const void *buf, size_t count);
+#define write mingw_write
+
+FILE *mingw_fopen (const char *filename, const char *otype);
+#define fopen mingw_fopen
+
+FILE *mingw_freopen (const char *filename, const char *otype, FILE *stream);
+#define freopen mingw_freopen
+
+char *mingw_getcwd(char *pointer, int len);
+#define getcwd mingw_getcwd
+
+char *mingw_getenv(const char *name);
+#define getenv mingw_getenv
+
+struct hostent *mingw_gethostbyname(const char *host);
+#define gethostbyname mingw_gethostbyname
+
+void mingw_freeaddrinfo(struct addrinfo *res);
+#define freeaddrinfo mingw_freeaddrinfo
+
+int mingw_getaddrinfo(const char *node, const char *service,
+		      const struct addrinfo *hints, struct addrinfo **res);
+#define getaddrinfo mingw_getaddrinfo
+
+int mingw_getnameinfo(const struct sockaddr *sa, socklen_t salen,
+		      char *host, DWORD hostlen, char *serv, DWORD servlen,
+		      int flags);
+#define getnameinfo mingw_getnameinfo
+
+int mingw_socket(int domain, int type, int protocol);
+#define socket mingw_socket
+
+int mingw_connect(int sockfd, struct sockaddr *sa, size_t sz);
+#define connect mingw_connect
+
+int mingw_rename(const char*, const char*);
+#define rename mingw_rename
+
+#if defined(USE_WIN32_MMAP) || defined(_MSC_VER)
+int mingw_getpagesize(void);
+#define getpagesize mingw_getpagesize
+#endif
+
+/* Use mingw_lstat() instead of lstat()/stat() and
+ * mingw_fstat() instead of fstat() on Windows.
+ */
+#define off_t off64_t
+#define lseek _lseeki64
+#ifndef ALREADY_DECLARED_STAT_FUNCS
+#define stat _stati64
+int mingw_lstat(const char *file_name, struct stat *buf);
+int mingw_fstat(int fd, struct stat *buf);
+#define fstat mingw_fstat
+#define lstat mingw_lstat
+#define _stati64(x,y) mingw_lstat(x,y)
+#endif
+
+int mingw_utime(const char *file_name, const struct utimbuf *times);
+#define utime mingw_utime
+
+pid_t mingw_spawnvpe(const char *cmd, const char **argv, char **env,
+		     const char *dir,
+		     int fhin, int fhout, int fherr);
+void mingw_execvp(const char *cmd, char *const *argv);
+#define execvp mingw_execvp
+
+static inline unsigned int git_ntohl(unsigned int x)
+{ return (unsigned int)ntohl(x); }
+#define ntohl git_ntohl
+
+sig_handler_t mingw_signal(int sig, sig_handler_t handler);
+#define signal mingw_signal
+
+/*
+ * ANSI emulation wrappers
+ */
+
+int winansi_fputs(const char *str, FILE *stream);
+int winansi_printf(const char *format, ...) __attribute__((format (printf, 1, 2)));
+int winansi_fprintf(FILE *stream, const char *format, ...) __attribute__((format (printf, 2, 3)));
+#define fputs winansi_fputs
+#define printf(...) winansi_printf(__VA_ARGS__)
+#define fprintf(...) winansi_fprintf(__VA_ARGS__)
+
+/*
+ * git specific compatibility
+ */
+
+#define has_dos_drive_prefix(path) (isalpha(*(path)) && (path)[1] == ':')
+#define is_dir_sep(c) ((c) == '/' || (c) == '\\')
+#define PATH_SEP ';'
+#define PRIuMAX "I64u"
+
+void mingw_open_html(const char *path);
+#define open_html mingw_open_html
+
+/*
+ * helpers
+ */
+
+char **make_augmented_environ(const char *const *vars);
+void free_environ(char **env);
+
+/*
+ * A replacement of main() that ensures that argv[0] has a path
+ * and that default fmode and std(in|out|err) are in binary mode
+ */
+
+#define main(c,v) dummy_decl_mingw_main(); \
+static int mingw_main(); \
+int main(int argc, const char **argv) \
+{ \
+	_fmode = _O_BINARY; \
+	_setmode(_fileno(stdin), _O_BINARY); \
+	_setmode(_fileno(stdout), _O_BINARY); \
+	_setmode(_fileno(stderr), _O_BINARY); \
+	argv[0] = xstrdup(_pgmptr); \
+	return mingw_main(argc, argv); \
+} \
+static int mingw_main(c,v)
+
+#ifndef NO_MINGW_REPLACE_READDIR
+/*
+ * A replacement of readdir, to ensure that it reads the file type at
+ * the same time. This avoid extra unneeded lstats in git on MinGW
+ */
+#undef DT_UNKNOWN
+#undef DT_DIR
+#undef DT_REG
+#undef DT_LNK
+#define DT_UNKNOWN	0
+#define DT_DIR		1
+#define DT_REG		2
+#define DT_LNK		3
+
+struct mingw_dirent
+{
+	long		d_ino;			/* Always zero. */
+	union {
+		unsigned short	d_reclen;	/* Always zero. */
+		unsigned char   d_type;		/* Reimplementation adds this */
+	};
+	unsigned short	d_namlen;		/* Length of name in d_name. */
+	char		d_name[FILENAME_MAX];	/* File name. */
+};
+#define dirent mingw_dirent
+#define readdir(x) mingw_readdir(x)
+struct dirent *mingw_readdir(DIR *dir);
+#endif // !NO_MINGW_REPLACE_READDIR
+
+/*
+ * Used by Pthread API implementation for Windows
+ */
+extern int err_win_to_posix(DWORD winerr);
diff --git a/compat/mkdtemp.c b/compat/mkdtemp.c
new file mode 100644
index 0000000..1136119
--- /dev/null
+++ b/compat/mkdtemp.c
@@ -0,0 +1,8 @@
+#include "../git-compat-util.h"
+
+char *gitmkdtemp(char *template)
+{
+	if (!*mktemp(template) || mkdir(template, 0700))
+		return NULL;
+	return template;
+}
diff --git a/compat/mmap.c b/compat/mmap.c
new file mode 100644
index 0000000..c9d46d1
--- /dev/null
+++ b/compat/mmap.c
@@ -0,0 +1,42 @@
+#include "../git-compat-util.h"
+
+void *git_mmap(void *start, size_t length, int prot, int flags, int fd, off_t offset)
+{
+	size_t n = 0;
+
+	if (start != NULL || !(flags & MAP_PRIVATE))
+		die("Invalid usage of mmap when built with NO_MMAP");
+
+	start = xmalloc(length);
+	if (start == NULL) {
+		errno = ENOMEM;
+		return MAP_FAILED;
+	}
+
+	while (n < length) {
+		ssize_t count = pread(fd, (char *)start + n, length - n, offset + n);
+
+		if (count == 0) {
+			memset((char *)start+n, 0, length-n);
+			break;
+		}
+
+		if (count < 0) {
+			if (errno == EAGAIN || errno == EINTR)
+				continue;
+			free(start);
+			errno = EACCES;
+			return MAP_FAILED;
+		}
+
+		n += count;
+	}
+
+	return start;
+}
+
+int git_munmap(void *start, size_t length)
+{
+	free(start);
+	return 0;
+}
diff --git a/compat/msvc.c b/compat/msvc.c
new file mode 100644
index 0000000..ac04a4c
--- /dev/null
+++ b/compat/msvc.c
@@ -0,0 +1,35 @@
+#include "../git-compat-util.h"
+#include "win32.h"
+#include <conio.h>
+#include "../strbuf.h"
+
+DIR *opendir(const char *name)
+{
+	int len;
+	DIR *p;
+	p = (DIR*)malloc(sizeof(DIR));
+	memset(p, 0, sizeof(DIR));
+	strncpy(p->dd_name, name, PATH_MAX);
+	len = strlen(p->dd_name);
+	p->dd_name[len] = '/';
+	p->dd_name[len+1] = '*';
+
+	if (p == NULL)
+		return NULL;
+
+	p->dd_handle = _findfirst(p->dd_name, &p->dd_dta);
+
+	if (p->dd_handle == -1) {
+		free(p);
+		return NULL;
+	}
+	return p;
+}
+int closedir(DIR *dir)
+{
+	_findclose(dir->dd_handle);
+	free(dir);
+	return 0;
+}
+
+#include "mingw.c"
diff --git a/compat/msvc.h b/compat/msvc.h
new file mode 100644
index 0000000..023aba0
--- /dev/null
+++ b/compat/msvc.h
@@ -0,0 +1,42 @@
+#ifndef __MSVC__HEAD
+#define __MSVC__HEAD
+
+#include <direct.h>
+#include <process.h>
+#include <malloc.h>
+
+/* porting function */
+#define inline __inline
+#define __inline__ __inline
+#define __attribute__(x)
+#define va_copy(dst, src)     ((dst) = (src))
+#define strncasecmp  _strnicmp
+#define ftruncate    _chsize
+
+static __inline int strcasecmp (const char *s1, const char *s2)
+{
+	int size1 = strlen(s1);
+	int sisz2 = strlen(s2);
+	return _strnicmp(s1, s2, sisz2 > size1 ? sisz2 : size1);
+}
+
+#undef ERROR
+
+/* Use mingw_lstat() instead of lstat()/stat() and mingw_fstat() instead
+ * of fstat(). We add the declaration of these functions here, suppressing
+ * the corresponding declarations in mingw.h, so that we can use the
+ * appropriate structure type (and function) names from the msvc headers.
+ */
+#define stat _stat64
+int mingw_lstat(const char *file_name, struct stat *buf);
+int mingw_fstat(int fd, struct stat *buf);
+#define fstat mingw_fstat
+#define lstat mingw_lstat
+#define _stat64(x,y) mingw_lstat(x,y)
+#define ALREADY_DECLARED_STAT_FUNCS
+
+#include "compat/mingw.h"
+
+#undef ALREADY_DECLARED_STAT_FUNCS
+
+#endif
diff --git a/compat/nedmalloc/License.txt b/compat/nedmalloc/License.txt
new file mode 100644
index 0000000..36b7cd9
--- /dev/null
+++ b/compat/nedmalloc/License.txt
@@ -0,0 +1,23 @@
+Boost Software License - Version 1.0 - August 17th, 2003
+
+Permission is hereby granted, free of charge, to any person or organization
+obtaining a copy of the software and accompanying documentation covered by
+this license (the "Software") to use, reproduce, display, distribute,
+execute, and transmit the Software, and to prepare derivative works of the
+Software, and to permit third-parties to whom the Software is furnished to
+do so, all subject to the following:
+
+The copyright notices in the Software and this entire statement, including
+the above license grant, this restriction and the following disclaimer,
+must be included in all copies of the Software, in whole or in part, and
+all derivative works of the Software, unless such copies or derivative
+works are solely in the form of machine-executable object code generated by
+a source language processor.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE, TITLE AND NON-INFRINGEMENT. IN NO EVENT
+SHALL THE COPYRIGHT HOLDERS OR ANYONE DISTRIBUTING THE SOFTWARE BE LIABLE
+FOR ANY DAMAGES OR OTHER LIABILITY, WHETHER IN CONTRACT, TORT OR OTHERWISE,
+ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+DEALINGS IN THE SOFTWARE.
diff --git a/compat/nedmalloc/Readme.txt b/compat/nedmalloc/Readme.txt
new file mode 100644
index 0000000..8763656
--- /dev/null
+++ b/compat/nedmalloc/Readme.txt
@@ -0,0 +1,136 @@
+nedalloc v1.05 15th June 2008:
+-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
+
+by Niall Douglas (http://www.nedprod.com/programs/portable/nedmalloc/)
+
+Enclosed is nedalloc, an alternative malloc implementation for multiple
+threads without lock contention based on dlmalloc v2.8.4. It is more
+or less a newer implementation of ptmalloc2, the standard allocator in
+Linux (which is based on dlmalloc v2.7.0) but also contains a per-thread
+cache for maximum CPU scalability.
+
+It is licensed under the Boost Software License which basically means
+you can do anything you like with it. This does not apply to the malloc.c.h
+file which remains copyright to others.
+
+It has been tested on win32 (x86), win64 (x64), Linux (x64), FreeBSD (x64)
+and Apple MacOS X (x86). It works very well on all of these and is very
+significantly faster than the system allocator on all of these platforms.
+
+By literally dropping in this allocator as a replacement for your system
+allocator, you can see real world improvements of up to three times in normal
+code!
+
+To use:
+-=-=-=-
+Drop in nedmalloc.h, nedmalloc.c and malloc.c.h into your project.
+Configure using the instructions in nedmalloc.h. Run and enjoy.
+
+To test, compile test.c. It will run a comparison between your system
+allocator and nedalloc and tell you how much faster nedalloc is. It also
+serves as an example of usage.
+
+Notes:
+-=-=-=
+If you want the very latest version of this allocator, get it from the
+TnFOX SVN repository at svn://svn.berlios.de/viewcvs/tnfox/trunk/src/nedmalloc
+
+Because of how nedalloc allocates an mspace per thread, it can cause
+severe bloating of memory usage under certain allocation patterns.
+You can substantially reduce this wastage by setting MAXTHREADSINPOOL
+or the threads parameter to nedcreatepool() to a fraction of the number of
+threads which would normally be in a pool at once. This will reduce
+bloating at the cost of an increase in lock contention. If allocated size
+is less than THREADCACHEMAX, locking is avoided 90-99% of the time and
+if most of your allocations are below this value, you can safely set
+MAXTHREADSINPOOL to one.
+
+You will suffer memory leakage unless you call neddisablethreadcache()
+per pool for every thread which exits. This is because nedalloc cannot
+portably know when a thread exits and thus when its thread cache can
+be returned for use by other code. Don't forget pool zero, the system pool.
+
+For C++ type allocation patterns (where the same sizes of memory are
+regularly allocated and deallocated as objects are created and destroyed),
+the threadcache always benefits performance. If however your allocation
+patterns are different, searching the threadcache may significantly slow
+down your code - as a rule of thumb, if cache utilisation is below 80%
+(see the source for neddisablethreadcache() for how to enable debug
+printing in release mode) then you should disable the thread cache for
+that thread. You can compile out the threadcache code by setting
+THREADCACHEMAX to zero.
+
+Speed comparisons:
+-=-=-=-=-=-=-=-=-=
+See Benchmarks.xls for details.
+
+The enclosed test.c can do two things: it can be a torture test or a speed
+test. The speed test is designed to be a representative synthetic
+memory allocator test. It works by randomly mixing allocations with frees
+with half of the allocation sizes being a two power multiple less than
+512 bytes (to mimic C++ stack instantiated objects) and the other half
+being a simple random value less than 16Kb.
+
+The real world code results are from Tn's TestIO benchmark. This is a
+heavily multithreaded and memory intensive benchmark with a lot of branching
+and other stuff modern processors don't like so much. As you'll note, the
+test doesn't show the benefits of the threadcache mostly due to the saturation
+of the memory bus being the limiting factor.
+
+ChangeLog:
+-=-=-=-=-=
+v1.05 15th June 2008:
+ * { 1042 } Added error check for TLSSET() and TLSFREE() macros. Thanks to
+Markus Elfring for reporting this.
+ * { 1043 } Fixed a segfault when freeing memory allocated using
+nedindependent_comalloc(). Thanks to Pavel Vozenilek for reporting this.
+
+v1.04 14th July 2007:
+ * Fixed a bug with the new optimised implementation that failed to lock
+on a realloc under certain conditions.
+ * Fixed lack of thread synchronisation in InitPool() causing pool corruption
+ * Fixed a memory leak of thread cache contents on disabling. Thanks to Earl
+Chew for reporting this.
+ * Added a sanity check for freed blocks being valid.
+ * Reworked test.c into being a torture test.
+ * Fixed GCC assembler optimisation misspecification
+
+v1.04alpha_svn915 7th October 2006:
+ * Fixed failure to unlock thread cache list if allocating a new list failed.
+Thanks to Dmitry Chichkov for reporting this. Futher thanks to Aleksey Sanin.
+ * Fixed realloc(0, <size>) segfaulting. Thanks to Dmitry Chichkov for
+reporting this.
+ * Made config defines #ifndef so they can be overriden by the build system.
+Thanks to Aleksey Sanin for suggesting this.
+ * Fixed deadlock in nedprealloc() due to unnecessary locking of preferred
+thread mspace when mspace_realloc() always uses the original block's mspace
+anyway. Thanks to Aleksey Sanin for reporting this.
+ * Made some speed improvements by hacking mspace_malloc() to no longer lock
+its mspace, thus allowing the recursive mutex implementation to be removed
+with an associated speed increase. Thanks to Aleksey Sanin for suggesting this.
+ * Fixed a bug where allocating mspaces overran its max limit. Thanks to
+Aleksey Sanin for reporting this.
+
+v1.03 10th July 2006:
+ * Fixed memory corruption bug in threadcache code which only appeared with >4
+threads and in heavy use of the threadcache.
+
+v1.02 15th May 2006:
+ * Integrated dlmalloc v2.8.4, fixing the win32 memory release problem and
+improving performance still further. Speed is now up to twice the speed of v1.01
+(average is 67% faster).
+ * Fixed win32 critical section implementation. Thanks to Pavel Kuznetsov
+for reporting this.
+ * Wasn't locking mspace if all mspaces were locked. Thanks to Pavel Kuznetsov
+for reporting this.
+ * Added Apple Mac OS X support.
+
+v1.01 24th February 2006:
+ * Fixed multiprocessor scaling problems by removing sources of cache sloshing
+ * Earl Chew <earl_chew <at> agilent <dot> com> sent patches for the following:
+   1. size2binidx() wasn't working for default code path (non x86)
+   2. Fixed failure to release mspace lock under certain circumstances which
+      caused a deadlock
+
+v1.00 1st January 2006:
+ * First release
diff --git a/compat/nedmalloc/malloc.c.h b/compat/nedmalloc/malloc.c.h
new file mode 100644
index 0000000..74c42e3
--- /dev/null
+++ b/compat/nedmalloc/malloc.c.h
@@ -0,0 +1,5752 @@
+/*
+  This is a version (aka dlmalloc) of malloc/free/realloc written by
+  Doug Lea and released to the public domain, as explained at
+  http://creativecommons.org/licenses/publicdomain.  Send questions,
+  comments, complaints, performance data, etc to dl@cs.oswego.edu
+
+* Version pre-2.8.4 Mon Nov 27 11:22:37 2006    (dl at gee)
+
+   Note: There may be an updated version of this malloc obtainable at
+	   ftp://gee.cs.oswego.edu/pub/misc/malloc.c
+	 Check before installing!
+
+* Quickstart
+
+  This library is all in one file to simplify the most common usage:
+  ftp it, compile it (-O3), and link it into another program. All of
+  the compile-time options default to reasonable values for use on
+  most platforms.  You might later want to step through various
+  compile-time and dynamic tuning options.
+
+  For convenience, an include file for code using this malloc is at:
+     ftp://gee.cs.oswego.edu/pub/misc/malloc-2.8.4.h
+  You don't really need this .h file unless you call functions not
+  defined in your system include files.  The .h file contains only the
+  excerpts from this file needed for using this malloc on ANSI C/C++
+  systems, so long as you haven't changed compile-time options about
+  naming and tuning parameters.  If you do, then you can create your
+  own malloc.h that does include all settings by cutting at the point
+  indicated below. Note that you may already by default be using a C
+  library containing a malloc that is based on some version of this
+  malloc (for example in linux). You might still want to use the one
+  in this file to customize settings or to avoid overheads associated
+  with library versions.
+
+* Vital statistics:
+
+  Supported pointer/size_t representation:       4 or 8 bytes
+       size_t MUST be an unsigned type of the same width as
+       pointers. (If you are using an ancient system that declares
+       size_t as a signed type, or need it to be a different width
+       than pointers, you can use a previous release of this malloc
+       (e.g. 2.7.2) supporting these.)
+
+  Alignment:                                     8 bytes (default)
+       This suffices for nearly all current machines and C compilers.
+       However, you can define MALLOC_ALIGNMENT to be wider than this
+       if necessary (up to 128bytes), at the expense of using more space.
+
+  Minimum overhead per allocated chunk:   4 or  8 bytes (if 4byte sizes)
+					  8 or 16 bytes (if 8byte sizes)
+       Each malloced chunk has a hidden word of overhead holding size
+       and status information, and additional cross-check word
+       if FOOTERS is defined.
+
+  Minimum allocated size: 4-byte ptrs:  16 bytes    (including overhead)
+			  8-byte ptrs:  32 bytes    (including overhead)
+
+       Even a request for zero bytes (i.e., malloc(0)) returns a
+       pointer to something of the minimum allocatable size.
+       The maximum overhead wastage (i.e., number of extra bytes
+       allocated than were requested in malloc) is less than or equal
+       to the minimum size, except for requests >= mmap_threshold that
+       are serviced via mmap(), where the worst case wastage is about
+       32 bytes plus the remainder from a system page (the minimal
+       mmap unit); typically 4096 or 8192 bytes.
+
+  Security: static-safe; optionally more or less
+       The "security" of malloc refers to the ability of malicious
+       code to accentuate the effects of errors (for example, freeing
+       space that is not currently malloc'ed or overwriting past the
+       ends of chunks) in code that calls malloc.  This malloc
+       guarantees not to modify any memory locations below the base of
+       heap, i.e., static variables, even in the presence of usage
+       errors.  The routines additionally detect most improper frees
+       and reallocs.  All this holds as long as the static bookkeeping
+       for malloc itself is not corrupted by some other means.  This
+       is only one aspect of security -- these checks do not, and
+       cannot, detect all possible programming errors.
+
+       If FOOTERS is defined nonzero, then each allocated chunk
+       carries an additional check word to verify that it was malloced
+       from its space.  These check words are the same within each
+       execution of a program using malloc, but differ across
+       executions, so externally crafted fake chunks cannot be
+       freed. This improves security by rejecting frees/reallocs that
+       could corrupt heap memory, in addition to the checks preventing
+       writes to statics that are always on.  This may further improve
+       security at the expense of time and space overhead.  (Note that
+       FOOTERS may also be worth using with MSPACES.)
+
+       By default detected errors cause the program to abort (calling
+       "abort()"). You can override this to instead proceed past
+       errors by defining PROCEED_ON_ERROR.  In this case, a bad free
+       has no effect, and a malloc that encounters a bad address
+       caused by user overwrites will ignore the bad address by
+       dropping pointers and indices to all known memory. This may
+       be appropriate for programs that should continue if at all
+       possible in the face of programming errors, although they may
+       run out of memory because dropped memory is never reclaimed.
+
+       If you don't like either of these options, you can define
+       CORRUPTION_ERROR_ACTION and USAGE_ERROR_ACTION to do anything
+       else. And if if you are sure that your program using malloc has
+       no errors or vulnerabilities, you can define INSECURE to 1,
+       which might (or might not) provide a small performance improvement.
+
+  Thread-safety: NOT thread-safe unless USE_LOCKS defined
+       When USE_LOCKS is defined, each public call to malloc, free,
+       etc is surrounded with either a pthread mutex or a win32
+       spinlock (depending on WIN32). This is not especially fast, and
+       can be a major bottleneck.  It is designed only to provide
+       minimal protection in concurrent environments, and to provide a
+       basis for extensions.  If you are using malloc in a concurrent
+       program, consider instead using nedmalloc
+       (http://www.nedprod.com/programs/portable/nedmalloc/) or
+       ptmalloc (See http://www.malloc.de), which are derived
+       from versions of this malloc.
+
+  System requirements: Any combination of MORECORE and/or MMAP/MUNMAP
+       This malloc can use unix sbrk or any emulation (invoked using
+       the CALL_MORECORE macro) and/or mmap/munmap or any emulation
+       (invoked using CALL_MMAP/CALL_MUNMAP) to get and release system
+       memory.  On most unix systems, it tends to work best if both
+       MORECORE and MMAP are enabled.  On Win32, it uses emulations
+       based on VirtualAlloc. It also uses common C library functions
+       like memset.
+
+  Compliance: I believe it is compliant with the Single Unix Specification
+       (See http://www.unix.org). Also SVID/XPG, ANSI C, and probably
+       others as well.
+
+* Overview of algorithms
+
+  This is not the fastest, most space-conserving, most portable, or
+  most tunable malloc ever written. However it is among the fastest
+  while also being among the most space-conserving, portable and
+  tunable.  Consistent balance across these factors results in a good
+  general-purpose allocator for malloc-intensive programs.
+
+  In most ways, this malloc is a best-fit allocator. Generally, it
+  chooses the best-fitting existing chunk for a request, with ties
+  broken in approximately least-recently-used order. (This strategy
+  normally maintains low fragmentation.) However, for requests less
+  than 256bytes, it deviates from best-fit when there is not an
+  exactly fitting available chunk by preferring to use space adjacent
+  to that used for the previous small request, as well as by breaking
+  ties in approximately most-recently-used order. (These enhance
+  locality of series of small allocations.)  And for very large requests
+  (>= 256Kb by default), it relies on system memory mapping
+  facilities, if supported.  (This helps avoid carrying around and
+  possibly fragmenting memory used only for large chunks.)
+
+  All operations (except malloc_stats and mallinfo) have execution
+  times that are bounded by a constant factor of the number of bits in
+  a size_t, not counting any clearing in calloc or copying in realloc,
+  or actions surrounding MORECORE and MMAP that have times
+  proportional to the number of non-contiguous regions returned by
+  system allocation routines, which is often just 1. In real-time
+  applications, you can optionally suppress segment traversals using
+  NO_SEGMENT_TRAVERSAL, which assures bounded execution even when
+  system allocators return non-contiguous spaces, at the typical
+  expense of carrying around more memory and increased fragmentation.
+
+  The implementation is not very modular and seriously overuses
+  macros. Perhaps someday all C compilers will do as good a job
+  inlining modular code as can now be done by brute-force expansion,
+  but now, enough of them seem not to.
+
+  Some compilers issue a lot of warnings about code that is
+  dead/unreachable only on some platforms, and also about intentional
+  uses of negation on unsigned types. All known cases of each can be
+  ignored.
+
+  For a longer but out of date high-level description, see
+     http://gee.cs.oswego.edu/dl/html/malloc.html
+
+* MSPACES
+  If MSPACES is defined, then in addition to malloc, free, etc.,
+  this file also defines mspace_malloc, mspace_free, etc. These
+  are versions of malloc routines that take an "mspace" argument
+  obtained using create_mspace, to control all internal bookkeeping.
+  If ONLY_MSPACES is defined, only these versions are compiled.
+  So if you would like to use this allocator for only some allocations,
+  and your system malloc for others, you can compile with
+  ONLY_MSPACES and then do something like...
+    static mspace mymspace = create_mspace(0,0); // for example
+    #define mymalloc(bytes)  mspace_malloc(mymspace, bytes)
+
+  (Note: If you only need one instance of an mspace, you can instead
+  use "USE_DL_PREFIX" to relabel the global malloc.)
+
+  You can similarly create thread-local allocators by storing
+  mspaces as thread-locals. For example:
+    static __thread mspace tlms = 0;
+    void*  tlmalloc(size_t bytes) {
+      if (tlms == 0) tlms = create_mspace(0, 0);
+      return mspace_malloc(tlms, bytes);
+    }
+    void  tlfree(void* mem) { mspace_free(tlms, mem); }
+
+  Unless FOOTERS is defined, each mspace is completely independent.
+  You cannot allocate from one and free to another (although
+  conformance is only weakly checked, so usage errors are not always
+  caught). If FOOTERS is defined, then each chunk carries around a tag
+  indicating its originating mspace, and frees are directed to their
+  originating spaces.
+
+ -------------------------  Compile-time options ---------------------------
+
+Be careful in setting #define values for numerical constants of type
+size_t. On some systems, literal values are not automatically extended
+to size_t precision unless they are explicitly casted. You can also
+use the symbolic values MAX_SIZE_T, SIZE_T_ONE, etc below.
+
+WIN32                    default: defined if _WIN32 defined
+  Defining WIN32 sets up defaults for MS environment and compilers.
+  Otherwise defaults are for unix. Beware that there seem to be some
+  cases where this malloc might not be a pure drop-in replacement for
+  Win32 malloc: Random-looking failures from Win32 GDI API's (eg;
+  SetDIBits()) may be due to bugs in some video driver implementations
+  when pixel buffers are malloc()ed, and the region spans more than
+  one VirtualAlloc()ed region. Because dlmalloc uses a small (64Kb)
+  default granularity, pixel buffers may straddle virtual allocation
+  regions more often than when using the Microsoft allocator.  You can
+  avoid this by using VirtualAlloc() and VirtualFree() for all pixel
+  buffers rather than using malloc().  If this is not possible,
+  recompile this malloc with a larger DEFAULT_GRANULARITY.
+
+MALLOC_ALIGNMENT         default: (size_t)8
+  Controls the minimum alignment for malloc'ed chunks.  It must be a
+  power of two and at least 8, even on machines for which smaller
+  alignments would suffice. It may be defined as larger than this
+  though. Note however that code and data structures are optimized for
+  the case of 8-byte alignment.
+
+MSPACES                  default: 0 (false)
+  If true, compile in support for independent allocation spaces.
+  This is only supported if HAVE_MMAP is true.
+
+ONLY_MSPACES             default: 0 (false)
+  If true, only compile in mspace versions, not regular versions.
+
+USE_LOCKS                default: 0 (false)
+  Causes each call to each public routine to be surrounded with
+  pthread or WIN32 mutex lock/unlock. (If set true, this can be
+  overridden on a per-mspace basis for mspace versions.) If set to a
+  non-zero value other than 1, locks are used, but their
+  implementation is left out, so lock functions must be supplied manually.
+
+USE_SPIN_LOCKS           default: 1 iff USE_LOCKS and on x86 using gcc or MSC
+  If true, uses custom spin locks for locking. This is currently
+  supported only for x86 platforms using gcc or recent MS compilers.
+  Otherwise, posix locks or win32 critical sections are used.
+
+FOOTERS                  default: 0
+  If true, provide extra checking and dispatching by placing
+  information in the footers of allocated chunks. This adds
+  space and time overhead.
+
+INSECURE                 default: 0
+  If true, omit checks for usage errors and heap space overwrites.
+
+USE_DL_PREFIX            default: NOT defined
+  Causes compiler to prefix all public routines with the string 'dl'.
+  This can be useful when you only want to use this malloc in one part
+  of a program, using your regular system malloc elsewhere.
+
+ABORT                    default: defined as abort()
+  Defines how to abort on failed checks.  On most systems, a failed
+  check cannot die with an "assert" or even print an informative
+  message, because the underlying print routines in turn call malloc,
+  which will fail again.  Generally, the best policy is to simply call
+  abort(). It's not very useful to do more than this because many
+  errors due to overwriting will show up as address faults (null, odd
+  addresses etc) rather than malloc-triggered checks, so will also
+  abort.  Also, most compilers know that abort() does not return, so
+  can better optimize code conditionally calling it.
+
+PROCEED_ON_ERROR           default: defined as 0 (false)
+  Controls whether detected bad addresses cause them to bypassed
+  rather than aborting. If set, detected bad arguments to free and
+  realloc are ignored. And all bookkeeping information is zeroed out
+  upon a detected overwrite of freed heap space, thus losing the
+  ability to ever return it from malloc again, but enabling the
+  application to proceed. If PROCEED_ON_ERROR is defined, the
+  static variable malloc_corruption_error_count is compiled in
+  and can be examined to see if errors have occurred. This option
+  generates slower code than the default abort policy.
+
+DEBUG                    default: NOT defined
+  The DEBUG setting is mainly intended for people trying to modify
+  this code or diagnose problems when porting to new platforms.
+  However, it may also be able to better isolate user errors than just
+  using runtime checks.  The assertions in the check routines spell
+  out in more detail the assumptions and invariants underlying the
+  algorithms.  The checking is fairly extensive, and will slow down
+  execution noticeably. Calling malloc_stats or mallinfo with DEBUG
+  set will attempt to check every non-mmapped allocated and free chunk
+  in the course of computing the summaries.
+
+ABORT_ON_ASSERT_FAILURE   default: defined as 1 (true)
+  Debugging assertion failures can be nearly impossible if your
+  version of the assert macro causes malloc to be called, which will
+  lead to a cascade of further failures, blowing the runtime stack.
+  ABORT_ON_ASSERT_FAILURE cause assertions failures to call abort(),
+  which will usually make debugging easier.
+
+MALLOC_FAILURE_ACTION     default: sets errno to ENOMEM, or no-op on win32
+  The action to take before "return 0" when malloc fails to be able to
+  return memory because there is none available.
+
+HAVE_MORECORE             default: 1 (true) unless win32 or ONLY_MSPACES
+  True if this system supports sbrk or an emulation of it.
+
+MORECORE                  default: sbrk
+  The name of the sbrk-style system routine to call to obtain more
+  memory.  See below for guidance on writing custom MORECORE
+  functions. The type of the argument to sbrk/MORECORE varies across
+  systems.  It cannot be size_t, because it supports negative
+  arguments, so it is normally the signed type of the same width as
+  size_t (sometimes declared as "intptr_t").  It doesn't much matter
+  though. Internally, we only call it with arguments less than half
+  the max value of a size_t, which should work across all reasonable
+  possibilities, although sometimes generating compiler warnings.
+
+MORECORE_CONTIGUOUS       default: 1 (true) if HAVE_MORECORE
+  If true, take advantage of fact that consecutive calls to MORECORE
+  with positive arguments always return contiguous increasing
+  addresses.  This is true of unix sbrk. It does not hurt too much to
+  set it true anyway, since malloc copes with non-contiguities.
+  Setting it false when definitely non-contiguous saves time
+  and possibly wasted space it would take to discover this though.
+
+MORECORE_CANNOT_TRIM      default: NOT defined
+  True if MORECORE cannot release space back to the system when given
+  negative arguments. This is generally necessary only if you are
+  using a hand-crafted MORECORE function that cannot handle negative
+  arguments.
+
+NO_SEGMENT_TRAVERSAL       default: 0
+  If non-zero, suppresses traversals of memory segments
+  returned by either MORECORE or CALL_MMAP. This disables
+  merging of segments that are contiguous, and selectively
+  releasing them to the OS if unused, but bounds execution times.
+
+HAVE_MMAP                 default: 1 (true)
+  True if this system supports mmap or an emulation of it.  If so, and
+  HAVE_MORECORE is not true, MMAP is used for all system
+  allocation. If set and HAVE_MORECORE is true as well, MMAP is
+  primarily used to directly allocate very large blocks. It is also
+  used as a backup strategy in cases where MORECORE fails to provide
+  space from system. Note: A single call to MUNMAP is assumed to be
+  able to unmap memory that may have be allocated using multiple calls
+  to MMAP, so long as they are adjacent.
+
+HAVE_MREMAP               default: 1 on linux, else 0
+  If true realloc() uses mremap() to re-allocate large blocks and
+  extend or shrink allocation spaces.
+
+MMAP_CLEARS               default: 1 except on WINCE.
+  True if mmap clears memory so calloc doesn't need to. This is true
+  for standard unix mmap using /dev/zero and on WIN32 except for WINCE.
+
+USE_BUILTIN_FFS            default: 0 (i.e., not used)
+  Causes malloc to use the builtin ffs() function to compute indices.
+  Some compilers may recognize and intrinsify ffs to be faster than the
+  supplied C version. Also, the case of x86 using gcc is special-cased
+  to an asm instruction, so is already as fast as it can be, and so
+  this setting has no effect. Similarly for Win32 under recent MS compilers.
+  (On most x86s, the asm version is only slightly faster than the C version.)
+
+malloc_getpagesize         default: derive from system includes, or 4096.
+  The system page size. To the extent possible, this malloc manages
+  memory from the system in page-size units.  This may be (and
+  usually is) a function rather than a constant. This is ignored
+  if WIN32, where page size is determined using getSystemInfo during
+  initialization.
+
+USE_DEV_RANDOM             default: 0 (i.e., not used)
+  Causes malloc to use /dev/random to initialize secure magic seed for
+  stamping footers. Otherwise, the current time is used.
+
+NO_MALLINFO                default: 0
+  If defined, don't compile "mallinfo". This can be a simple way
+  of dealing with mismatches between system declarations and
+  those in this file.
+
+MALLINFO_FIELD_TYPE        default: size_t
+  The type of the fields in the mallinfo struct. This was originally
+  defined as "int" in SVID etc, but is more usefully defined as
+  size_t. The value is used only if  HAVE_USR_INCLUDE_MALLOC_H is not set
+
+REALLOC_ZERO_BYTES_FREES    default: not defined
+  This should be set if a call to realloc with zero bytes should
+  be the same as a call to free. Some people think it should. Otherwise,
+  since this malloc returns a unique pointer for malloc(0), so does
+  realloc(p, 0).
+
+LACKS_UNISTD_H, LACKS_FCNTL_H, LACKS_SYS_PARAM_H, LACKS_SYS_MMAN_H
+LACKS_STRINGS_H, LACKS_STRING_H, LACKS_SYS_TYPES_H,  LACKS_ERRNO_H
+LACKS_STDLIB_H                default: NOT defined unless on WIN32
+  Define these if your system does not have these header files.
+  You might need to manually insert some of the declarations they provide.
+
+DEFAULT_GRANULARITY        default: page size if MORECORE_CONTIGUOUS,
+				system_info.dwAllocationGranularity in WIN32,
+				otherwise 64K.
+      Also settable using mallopt(M_GRANULARITY, x)
+  The unit for allocating and deallocating memory from the system.  On
+  most systems with contiguous MORECORE, there is no reason to
+  make this more than a page. However, systems with MMAP tend to
+  either require or encourage larger granularities.  You can increase
+  this value to prevent system allocation functions to be called so
+  often, especially if they are slow.  The value must be at least one
+  page and must be a power of two.  Setting to 0 causes initialization
+  to either page size or win32 region size.  (Note: In previous
+  versions of malloc, the equivalent of this option was called
+  "TOP_PAD")
+
+DEFAULT_TRIM_THRESHOLD    default: 2MB
+      Also settable using mallopt(M_TRIM_THRESHOLD, x)
+  The maximum amount of unused top-most memory to keep before
+  releasing via malloc_trim in free().  Automatic trimming is mainly
+  useful in long-lived programs using contiguous MORECORE.  Because
+  trimming via sbrk can be slow on some systems, and can sometimes be
+  wasteful (in cases where programs immediately afterward allocate
+  more large chunks) the value should be high enough so that your
+  overall system performance would improve by releasing this much
+  memory.  As a rough guide, you might set to a value close to the
+  average size of a process (program) running on your system.
+  Releasing this much memory would allow such a process to run in
+  memory.  Generally, it is worth tuning trim thresholds when a
+  program undergoes phases where several large chunks are allocated
+  and released in ways that can reuse each other's storage, perhaps
+  mixed with phases where there are no such chunks at all. The trim
+  value must be greater than page size to have any useful effect.  To
+  disable trimming completely, you can set to MAX_SIZE_T. Note that the trick
+  some people use of mallocing a huge space and then freeing it at
+  program startup, in an attempt to reserve system memory, doesn't
+  have the intended effect under automatic trimming, since that memory
+  will immediately be returned to the system.
+
+DEFAULT_MMAP_THRESHOLD       default: 256K
+      Also settable using mallopt(M_MMAP_THRESHOLD, x)
+  The request size threshold for using MMAP to directly service a
+  request. Requests of at least this size that cannot be allocated
+  using already-existing space will be serviced via mmap.  (If enough
+  normal freed space already exists it is used instead.)  Using mmap
+  segregates relatively large chunks of memory so that they can be
+  individually obtained and released from the host system. A request
+  serviced through mmap is never reused by any other request (at least
+  not directly; the system may just so happen to remap successive
+  requests to the same locations).  Segregating space in this way has
+  the benefits that: Mmapped space can always be individually released
+  back to the system, which helps keep the system level memory demands
+  of a long-lived program low.  Also, mapped memory doesn't become
+  `locked' between other chunks, as can happen with normally allocated
+  chunks, which means that even trimming via malloc_trim would not
+  release them.  However, it has the disadvantage that the space
+  cannot be reclaimed, consolidated, and then used to service later
+  requests, as happens with normal chunks.  The advantages of mmap
+  nearly always outweigh disadvantages for "large" chunks, but the
+  value of "large" may vary across systems.  The default is an
+  empirically derived value that works well in most systems. You can
+  disable mmap by setting to MAX_SIZE_T.
+
+MAX_RELEASE_CHECK_RATE   default: 4095 unless not HAVE_MMAP
+  The number of consolidated frees between checks to release
+  unused segments when freeing. When using non-contiguous segments,
+  especially with multiple mspaces, checking only for topmost space
+  doesn't always suffice to trigger trimming. To compensate for this,
+  free() will, with a period of MAX_RELEASE_CHECK_RATE (or the
+  current number of segments, if greater) try to release unused
+  segments to the OS when freeing chunks that result in
+  consolidation. The best value for this parameter is a compromise
+  between slowing down frees with relatively costly checks that
+  rarely trigger versus holding on to unused memory. To effectively
+  disable, set to MAX_SIZE_T. This may lead to a very slight speed
+  improvement at the expense of carrying around more memory.
+*/
+
+/* Version identifier to allow people to support multiple versions */
+#ifndef DLMALLOC_VERSION
+#define DLMALLOC_VERSION 20804
+#endif /* DLMALLOC_VERSION */
+
+#ifndef WIN32
+#ifdef _WIN32
+#define WIN32 1
+#endif  /* _WIN32 */
+#ifdef _WIN32_WCE
+#define LACKS_FCNTL_H
+#define WIN32 1
+#endif /* _WIN32_WCE */
+#endif  /* WIN32 */
+#ifdef WIN32
+#define WIN32_LEAN_AND_MEAN
+#define _WIN32_WINNT 0x403
+#include <windows.h>
+#define HAVE_MMAP 1
+#define HAVE_MORECORE 0
+#define LACKS_UNISTD_H
+#define LACKS_SYS_PARAM_H
+#define LACKS_SYS_MMAN_H
+#define LACKS_STRING_H
+#define LACKS_STRINGS_H
+#define LACKS_SYS_TYPES_H
+#define LACKS_ERRNO_H
+#ifndef MALLOC_FAILURE_ACTION
+#define MALLOC_FAILURE_ACTION
+#endif /* MALLOC_FAILURE_ACTION */
+#ifdef _WIN32_WCE /* WINCE reportedly does not clear */
+#define MMAP_CLEARS 0
+#else
+#define MMAP_CLEARS 1
+#endif /* _WIN32_WCE */
+#endif  /* WIN32 */
+
+#if defined(DARWIN) || defined(_DARWIN)
+/* Mac OSX docs advise not to use sbrk; it seems better to use mmap */
+#ifndef HAVE_MORECORE
+#define HAVE_MORECORE 0
+#define HAVE_MMAP 1
+/* OSX allocators provide 16 byte alignment */
+#ifndef MALLOC_ALIGNMENT
+#define MALLOC_ALIGNMENT ((size_t)16U)
+#endif
+#endif  /* HAVE_MORECORE */
+#endif  /* DARWIN */
+
+#ifndef LACKS_SYS_TYPES_H
+#include <sys/types.h>  /* For size_t */
+#endif  /* LACKS_SYS_TYPES_H */
+
+/* The maximum possible size_t value has all bits set */
+#define MAX_SIZE_T           (~(size_t)0)
+
+#ifndef ONLY_MSPACES
+#define ONLY_MSPACES 0     /* define to a value */
+#else
+#define ONLY_MSPACES 1
+#endif  /* ONLY_MSPACES */
+#ifndef MSPACES
+#if ONLY_MSPACES
+#define MSPACES 1
+#else   /* ONLY_MSPACES */
+#define MSPACES 0
+#endif  /* ONLY_MSPACES */
+#endif  /* MSPACES */
+#ifndef MALLOC_ALIGNMENT
+#define MALLOC_ALIGNMENT ((size_t)8U)
+#endif  /* MALLOC_ALIGNMENT */
+#ifndef FOOTERS
+#define FOOTERS 0
+#endif  /* FOOTERS */
+#ifndef ABORT
+#define ABORT  abort()
+#endif  /* ABORT */
+#ifndef ABORT_ON_ASSERT_FAILURE
+#define ABORT_ON_ASSERT_FAILURE 1
+#endif  /* ABORT_ON_ASSERT_FAILURE */
+#ifndef PROCEED_ON_ERROR
+#define PROCEED_ON_ERROR 0
+#endif  /* PROCEED_ON_ERROR */
+#ifndef USE_LOCKS
+#define USE_LOCKS 0
+#endif  /* USE_LOCKS */
+#ifndef USE_SPIN_LOCKS
+#if USE_LOCKS && (defined(__GNUC__) && ((defined(__i386__) || defined(__x86_64__)))) || (defined(_MSC_VER) && _MSC_VER>=1310)
+#define USE_SPIN_LOCKS 1
+#else
+#define USE_SPIN_LOCKS 0
+#endif /* USE_LOCKS && ... */
+#endif /* USE_SPIN_LOCKS */
+#ifndef INSECURE
+#define INSECURE 0
+#endif  /* INSECURE */
+#ifndef HAVE_MMAP
+#define HAVE_MMAP 1
+#endif  /* HAVE_MMAP */
+#ifndef MMAP_CLEARS
+#define MMAP_CLEARS 1
+#endif  /* MMAP_CLEARS */
+#ifndef HAVE_MREMAP
+#ifdef linux
+#define HAVE_MREMAP 1
+#else   /* linux */
+#define HAVE_MREMAP 0
+#endif  /* linux */
+#endif  /* HAVE_MREMAP */
+#ifndef MALLOC_FAILURE_ACTION
+#define MALLOC_FAILURE_ACTION  errno = ENOMEM;
+#endif  /* MALLOC_FAILURE_ACTION */
+#ifndef HAVE_MORECORE
+#if ONLY_MSPACES
+#define HAVE_MORECORE 0
+#else   /* ONLY_MSPACES */
+#define HAVE_MORECORE 1
+#endif  /* ONLY_MSPACES */
+#endif  /* HAVE_MORECORE */
+#if !HAVE_MORECORE
+#define MORECORE_CONTIGUOUS 0
+#else   /* !HAVE_MORECORE */
+#define MORECORE_DEFAULT sbrk
+#ifndef MORECORE_CONTIGUOUS
+#define MORECORE_CONTIGUOUS 1
+#endif  /* MORECORE_CONTIGUOUS */
+#endif  /* HAVE_MORECORE */
+#ifndef DEFAULT_GRANULARITY
+#if (MORECORE_CONTIGUOUS || defined(WIN32))
+#define DEFAULT_GRANULARITY (0)  /* 0 means to compute in init_mparams */
+#else   /* MORECORE_CONTIGUOUS */
+#define DEFAULT_GRANULARITY ((size_t)64U * (size_t)1024U)
+#endif  /* MORECORE_CONTIGUOUS */
+#endif  /* DEFAULT_GRANULARITY */
+#ifndef DEFAULT_TRIM_THRESHOLD
+#ifndef MORECORE_CANNOT_TRIM
+#define DEFAULT_TRIM_THRESHOLD ((size_t)2U * (size_t)1024U * (size_t)1024U)
+#else   /* MORECORE_CANNOT_TRIM */
+#define DEFAULT_TRIM_THRESHOLD MAX_SIZE_T
+#endif  /* MORECORE_CANNOT_TRIM */
+#endif  /* DEFAULT_TRIM_THRESHOLD */
+#ifndef DEFAULT_MMAP_THRESHOLD
+#if HAVE_MMAP
+#define DEFAULT_MMAP_THRESHOLD ((size_t)256U * (size_t)1024U)
+#else   /* HAVE_MMAP */
+#define DEFAULT_MMAP_THRESHOLD MAX_SIZE_T
+#endif  /* HAVE_MMAP */
+#endif  /* DEFAULT_MMAP_THRESHOLD */
+#ifndef MAX_RELEASE_CHECK_RATE
+#if HAVE_MMAP
+#define MAX_RELEASE_CHECK_RATE 4095
+#else
+#define MAX_RELEASE_CHECK_RATE MAX_SIZE_T
+#endif /* HAVE_MMAP */
+#endif /* MAX_RELEASE_CHECK_RATE */
+#ifndef USE_BUILTIN_FFS
+#define USE_BUILTIN_FFS 0
+#endif  /* USE_BUILTIN_FFS */
+#ifndef USE_DEV_RANDOM
+#define USE_DEV_RANDOM 0
+#endif  /* USE_DEV_RANDOM */
+#ifndef NO_MALLINFO
+#define NO_MALLINFO 0
+#endif  /* NO_MALLINFO */
+#ifndef MALLINFO_FIELD_TYPE
+#define MALLINFO_FIELD_TYPE size_t
+#endif  /* MALLINFO_FIELD_TYPE */
+#ifndef NO_SEGMENT_TRAVERSAL
+#define NO_SEGMENT_TRAVERSAL 0
+#endif /* NO_SEGMENT_TRAVERSAL */
+
+/*
+  mallopt tuning options.  SVID/XPG defines four standard parameter
+  numbers for mallopt, normally defined in malloc.h.  None of these
+  are used in this malloc, so setting them has no effect. But this
+  malloc does support the following options.
+*/
+
+#define M_TRIM_THRESHOLD     (-1)
+#define M_GRANULARITY        (-2)
+#define M_MMAP_THRESHOLD     (-3)
+
+/* ------------------------ Mallinfo declarations ------------------------ */
+
+#if !NO_MALLINFO
+/*
+  This version of malloc supports the standard SVID/XPG mallinfo
+  routine that returns a struct containing usage properties and
+  statistics. It should work on any system that has a
+  /usr/include/malloc.h defining struct mallinfo.  The main
+  declaration needed is the mallinfo struct that is returned (by-copy)
+  by mallinfo().  The malloinfo struct contains a bunch of fields that
+  are not even meaningful in this version of malloc.  These fields are
+  are instead filled by mallinfo() with other numbers that might be of
+  interest.
+
+  HAVE_USR_INCLUDE_MALLOC_H should be set if you have a
+  /usr/include/malloc.h file that includes a declaration of struct
+  mallinfo.  If so, it is included; else a compliant version is
+  declared below.  These must be precisely the same for mallinfo() to
+  work.  The original SVID version of this struct, defined on most
+  systems with mallinfo, declares all fields as ints. But some others
+  define as unsigned long. If your system defines the fields using a
+  type of different width than listed here, you MUST #include your
+  system version and #define HAVE_USR_INCLUDE_MALLOC_H.
+*/
+
+/* #define HAVE_USR_INCLUDE_MALLOC_H */
+
+#ifdef HAVE_USR_INCLUDE_MALLOC_H
+#include "/usr/include/malloc.h"
+#else /* HAVE_USR_INCLUDE_MALLOC_H */
+#ifndef STRUCT_MALLINFO_DECLARED
+#define STRUCT_MALLINFO_DECLARED 1
+struct mallinfo {
+  MALLINFO_FIELD_TYPE arena;    /* non-mmapped space allocated from system */
+  MALLINFO_FIELD_TYPE ordblks;  /* number of free chunks */
+  MALLINFO_FIELD_TYPE smblks;   /* always 0 */
+  MALLINFO_FIELD_TYPE hblks;    /* always 0 */
+  MALLINFO_FIELD_TYPE hblkhd;   /* space in mmapped regions */
+  MALLINFO_FIELD_TYPE usmblks;  /* maximum total allocated space */
+  MALLINFO_FIELD_TYPE fsmblks;  /* always 0 */
+  MALLINFO_FIELD_TYPE uordblks; /* total allocated space */
+  MALLINFO_FIELD_TYPE fordblks; /* total free space */
+  MALLINFO_FIELD_TYPE keepcost; /* releasable (via malloc_trim) space */
+};
+#endif /* STRUCT_MALLINFO_DECLARED */
+#endif /* HAVE_USR_INCLUDE_MALLOC_H */
+#endif /* NO_MALLINFO */
+
+/*
+  Try to persuade compilers to inline. The most critical functions for
+  inlining are defined as macros, so these aren't used for them.
+*/
+
+#ifndef FORCEINLINE
+  #if defined(__GNUC__)
+#define FORCEINLINE __inline __attribute__ ((always_inline))
+  #elif defined(_MSC_VER)
+    #define FORCEINLINE __forceinline
+  #endif
+#endif
+#ifndef NOINLINE
+  #if defined(__GNUC__)
+    #define NOINLINE __attribute__ ((noinline))
+  #elif defined(_MSC_VER)
+    #define NOINLINE __declspec(noinline)
+  #else
+    #define NOINLINE
+  #endif
+#endif
+
+#ifdef __cplusplus
+extern "C" {
+#ifndef FORCEINLINE
+ #define FORCEINLINE inline
+#endif
+#endif /* __cplusplus */
+#ifndef FORCEINLINE
+ #define FORCEINLINE
+#endif
+
+#if !ONLY_MSPACES
+
+/* ------------------- Declarations of public routines ------------------- */
+
+#ifndef USE_DL_PREFIX
+#define dlcalloc               calloc
+#define dlfree                 free
+#define dlmalloc               malloc
+#define dlmemalign             memalign
+#define dlrealloc              realloc
+#define dlvalloc               valloc
+#define dlpvalloc              pvalloc
+#define dlmallinfo             mallinfo
+#define dlmallopt              mallopt
+#define dlmalloc_trim          malloc_trim
+#define dlmalloc_stats         malloc_stats
+#define dlmalloc_usable_size   malloc_usable_size
+#define dlmalloc_footprint     malloc_footprint
+#define dlmalloc_max_footprint malloc_max_footprint
+#define dlindependent_calloc   independent_calloc
+#define dlindependent_comalloc independent_comalloc
+#endif /* USE_DL_PREFIX */
+
+
+/*
+  malloc(size_t n)
+  Returns a pointer to a newly allocated chunk of at least n bytes, or
+  null if no space is available, in which case errno is set to ENOMEM
+  on ANSI C systems.
+
+  If n is zero, malloc returns a minimum-sized chunk. (The minimum
+  size is 16 bytes on most 32bit systems, and 32 bytes on 64bit
+  systems.)  Note that size_t is an unsigned type, so calls with
+  arguments that would be negative if signed are interpreted as
+  requests for huge amounts of space, which will often fail. The
+  maximum supported value of n differs across systems, but is in all
+  cases less than the maximum representable value of a size_t.
+*/
+void* dlmalloc(size_t);
+
+/*
+  free(void* p)
+  Releases the chunk of memory pointed to by p, that had been previously
+  allocated using malloc or a related routine such as realloc.
+  It has no effect if p is null. If p was not malloced or already
+  freed, free(p) will by default cause the current program to abort.
+*/
+void  dlfree(void*);
+
+/*
+  calloc(size_t n_elements, size_t element_size);
+  Returns a pointer to n_elements * element_size bytes, with all locations
+  set to zero.
+*/
+void* dlcalloc(size_t, size_t);
+
+/*
+  realloc(void* p, size_t n)
+  Returns a pointer to a chunk of size n that contains the same data
+  as does chunk p up to the minimum of (n, p's size) bytes, or null
+  if no space is available.
+
+  The returned pointer may or may not be the same as p. The algorithm
+  prefers extending p in most cases when possible, otherwise it
+  employs the equivalent of a malloc-copy-free sequence.
+
+  If p is null, realloc is equivalent to malloc.
+
+  If space is not available, realloc returns null, errno is set (if on
+  ANSI) and p is NOT freed.
+
+  if n is for fewer bytes than already held by p, the newly unused
+  space is lopped off and freed if possible.  realloc with a size
+  argument of zero (re)allocates a minimum-sized chunk.
+
+  The old unix realloc convention of allowing the last-free'd chunk
+  to be used as an argument to realloc is not supported.
+*/
+
+void* dlrealloc(void*, size_t);
+
+/*
+  memalign(size_t alignment, size_t n);
+  Returns a pointer to a newly allocated chunk of n bytes, aligned
+  in accord with the alignment argument.
+
+  The alignment argument should be a power of two. If the argument is
+  not a power of two, the nearest greater power is used.
+  8-byte alignment is guaranteed by normal malloc calls, so don't
+  bother calling memalign with an argument of 8 or less.
+
+  Overreliance on memalign is a sure way to fragment space.
+*/
+void* dlmemalign(size_t, size_t);
+
+/*
+  valloc(size_t n);
+  Equivalent to memalign(pagesize, n), where pagesize is the page
+  size of the system. If the pagesize is unknown, 4096 is used.
+*/
+void* dlvalloc(size_t);
+
+/*
+  mallopt(int parameter_number, int parameter_value)
+  Sets tunable parameters The format is to provide a
+  (parameter-number, parameter-value) pair.  mallopt then sets the
+  corresponding parameter to the argument value if it can (i.e., so
+  long as the value is meaningful), and returns 1 if successful else
+  0.  To workaround the fact that mallopt is specified to use int,
+  not size_t parameters, the value -1 is specially treated as the
+  maximum unsigned size_t value.
+
+  SVID/XPG/ANSI defines four standard param numbers for mallopt,
+  normally defined in malloc.h.  None of these are use in this malloc,
+  so setting them has no effect. But this malloc also supports other
+  options in mallopt. See below for details.  Briefly, supported
+  parameters are as follows (listed defaults are for "typical"
+  configurations).
+
+  Symbol            param #  default    allowed param values
+  M_TRIM_THRESHOLD     -1   2*1024*1024   any   (-1 disables)
+  M_GRANULARITY        -2     page size   any power of 2 >= page size
+  M_MMAP_THRESHOLD     -3      256*1024   any   (or 0 if no MMAP support)
+*/
+int dlmallopt(int, int);
+
+/*
+  malloc_footprint();
+  Returns the number of bytes obtained from the system.  The total
+  number of bytes allocated by malloc, realloc etc., is less than this
+  value. Unlike mallinfo, this function returns only a precomputed
+  result, so can be called frequently to monitor memory consumption.
+  Even if locks are otherwise defined, this function does not use them,
+  so results might not be up to date.
+*/
+size_t dlmalloc_footprint(void);
+
+/*
+  malloc_max_footprint();
+  Returns the maximum number of bytes obtained from the system. This
+  value will be greater than current footprint if deallocated space
+  has been reclaimed by the system. The peak number of bytes allocated
+  by malloc, realloc etc., is less than this value. Unlike mallinfo,
+  this function returns only a precomputed result, so can be called
+  frequently to monitor memory consumption.  Even if locks are
+  otherwise defined, this function does not use them, so results might
+  not be up to date.
+*/
+size_t dlmalloc_max_footprint(void);
+
+#if !NO_MALLINFO
+/*
+  mallinfo()
+  Returns (by copy) a struct containing various summary statistics:
+
+  arena:     current total non-mmapped bytes allocated from system
+  ordblks:   the number of free chunks
+  smblks:    always zero.
+  hblks:     current number of mmapped regions
+  hblkhd:    total bytes held in mmapped regions
+  usmblks:   the maximum total allocated space. This will be greater
+		than current total if trimming has occurred.
+  fsmblks:   always zero
+  uordblks:  current total allocated space (normal or mmapped)
+  fordblks:  total free space
+  keepcost:  the maximum number of bytes that could ideally be released
+	       back to system via malloc_trim. ("ideally" means that
+	       it ignores page restrictions etc.)
+
+  Because these fields are ints, but internal bookkeeping may
+  be kept as longs, the reported values may wrap around zero and
+  thus be inaccurate.
+*/
+struct mallinfo dlmallinfo(void);
+#endif /* NO_MALLINFO */
+
+/*
+  independent_calloc(size_t n_elements, size_t element_size, void* chunks[]);
+
+  independent_calloc is similar to calloc, but instead of returning a
+  single cleared space, it returns an array of pointers to n_elements
+  independent elements that can hold contents of size elem_size, each
+  of which starts out cleared, and can be independently freed,
+  realloc'ed etc. The elements are guaranteed to be adjacently
+  allocated (this is not guaranteed to occur with multiple callocs or
+  mallocs), which may also improve cache locality in some
+  applications.
+
+  The "chunks" argument is optional (i.e., may be null, which is
+  probably the most typical usage). If it is null, the returned array
+  is itself dynamically allocated and should also be freed when it is
+  no longer needed. Otherwise, the chunks array must be of at least
+  n_elements in length. It is filled in with the pointers to the
+  chunks.
+
+  In either case, independent_calloc returns this pointer array, or
+  null if the allocation failed.  If n_elements is zero and "chunks"
+  is null, it returns a chunk representing an array with zero elements
+  (which should be freed if not wanted).
+
+  Each element must be individually freed when it is no longer
+  needed. If you'd like to instead be able to free all at once, you
+  should instead use regular calloc and assign pointers into this
+  space to represent elements.  (In this case though, you cannot
+  independently free elements.)
+
+  independent_calloc simplifies and speeds up implementations of many
+  kinds of pools.  It may also be useful when constructing large data
+  structures that initially have a fixed number of fixed-sized nodes,
+  but the number is not known at compile time, and some of the nodes
+  may later need to be freed. For example:
+
+  struct Node { int item; struct Node* next; };
+
+  struct Node* build_list() {
+    struct Node** pool;
+    int n = read_number_of_nodes_needed();
+    if (n <= 0) return 0;
+    pool = (struct Node**)(independent_calloc(n, sizeof(struct Node), 0);
+    if (pool == 0) die();
+    // organize into a linked list...
+    struct Node* first = pool[0];
+    for (i = 0; i < n-1; ++i)
+      pool[i]->next = pool[i+1];
+    free(pool);     // Can now free the array (or not, if it is needed later)
+    return first;
+  }
+*/
+void** dlindependent_calloc(size_t, size_t, void**);
+
+/*
+  independent_comalloc(size_t n_elements, size_t sizes[], void* chunks[]);
+
+  independent_comalloc allocates, all at once, a set of n_elements
+  chunks with sizes indicated in the "sizes" array.    It returns
+  an array of pointers to these elements, each of which can be
+  independently freed, realloc'ed etc. The elements are guaranteed to
+  be adjacently allocated (this is not guaranteed to occur with
+  multiple callocs or mallocs), which may also improve cache locality
+  in some applications.
+
+  The "chunks" argument is optional (i.e., may be null). If it is null
+  the returned array is itself dynamically allocated and should also
+  be freed when it is no longer needed. Otherwise, the chunks array
+  must be of at least n_elements in length. It is filled in with the
+  pointers to the chunks.
+
+  In either case, independent_comalloc returns this pointer array, or
+  null if the allocation failed.  If n_elements is zero and chunks is
+  null, it returns a chunk representing an array with zero elements
+  (which should be freed if not wanted).
+
+  Each element must be individually freed when it is no longer
+  needed. If you'd like to instead be able to free all at once, you
+  should instead use a single regular malloc, and assign pointers at
+  particular offsets in the aggregate space. (In this case though, you
+  cannot independently free elements.)
+
+  independent_comallac differs from independent_calloc in that each
+  element may have a different size, and also that it does not
+  automatically clear elements.
+
+  independent_comalloc can be used to speed up allocation in cases
+  where several structs or objects must always be allocated at the
+  same time.  For example:
+
+  struct Head { ... }
+  struct Foot { ... }
+
+  void send_message(char* msg) {
+    int msglen = strlen(msg);
+    size_t sizes[3] = { sizeof(struct Head), msglen, sizeof(struct Foot) };
+    void* chunks[3];
+    if (independent_comalloc(3, sizes, chunks) == 0)
+      die();
+    struct Head* head = (struct Head*)(chunks[0]);
+    char*        body = (char*)(chunks[1]);
+    struct Foot* foot = (struct Foot*)(chunks[2]);
+    // ...
+  }
+
+  In general though, independent_comalloc is worth using only for
+  larger values of n_elements. For small values, you probably won't
+  detect enough difference from series of malloc calls to bother.
+
+  Overuse of independent_comalloc can increase overall memory usage,
+  since it cannot reuse existing noncontiguous small chunks that
+  might be available for some of the elements.
+*/
+void** dlindependent_comalloc(size_t, size_t*, void**);
+
+
+/*
+  pvalloc(size_t n);
+  Equivalent to valloc(minimum-page-that-holds(n)), that is,
+  round up n to nearest pagesize.
+ */
+void*  dlpvalloc(size_t);
+
+/*
+  malloc_trim(size_t pad);
+
+  If possible, gives memory back to the system (via negative arguments
+  to sbrk) if there is unused memory at the `high' end of the malloc
+  pool or in unused MMAP segments. You can call this after freeing
+  large blocks of memory to potentially reduce the system-level memory
+  requirements of a program. However, it cannot guarantee to reduce
+  memory. Under some allocation patterns, some large free blocks of
+  memory will be locked between two used chunks, so they cannot be
+  given back to the system.
+
+  The `pad' argument to malloc_trim represents the amount of free
+  trailing space to leave untrimmed. If this argument is zero, only
+  the minimum amount of memory to maintain internal data structures
+  will be left. Non-zero arguments can be supplied to maintain enough
+  trailing space to service future expected allocations without having
+  to re-obtain memory from the system.
+
+  Malloc_trim returns 1 if it actually released any memory, else 0.
+*/
+int  dlmalloc_trim(size_t);
+
+/*
+  malloc_stats();
+  Prints on stderr the amount of space obtained from the system (both
+  via sbrk and mmap), the maximum amount (which may be more than
+  current if malloc_trim and/or munmap got called), and the current
+  number of bytes allocated via malloc (or realloc, etc) but not yet
+  freed. Note that this is the number of bytes allocated, not the
+  number requested. It will be larger than the number requested
+  because of alignment and bookkeeping overhead. Because it includes
+  alignment wastage as being in use, this figure may be greater than
+  zero even when no user-level chunks are allocated.
+
+  The reported current and maximum system memory can be inaccurate if
+  a program makes other calls to system memory allocation functions
+  (normally sbrk) outside of malloc.
+
+  malloc_stats prints only the most commonly interesting statistics.
+  More information can be obtained by calling mallinfo.
+*/
+void  dlmalloc_stats(void);
+
+#endif /* ONLY_MSPACES */
+
+/*
+  malloc_usable_size(void* p);
+
+  Returns the number of bytes you can actually use in
+  an allocated chunk, which may be more than you requested (although
+  often not) due to alignment and minimum size constraints.
+  You can use this many bytes without worrying about
+  overwriting other allocated objects. This is not a particularly great
+  programming practice. malloc_usable_size can be more useful in
+  debugging and assertions, for example:
+
+  p = malloc(n);
+  assert(malloc_usable_size(p) >= 256);
+*/
+size_t dlmalloc_usable_size(void*);
+
+
+#if MSPACES
+
+/*
+  mspace is an opaque type representing an independent
+  region of space that supports mspace_malloc, etc.
+*/
+typedef void* mspace;
+
+/*
+  create_mspace creates and returns a new independent space with the
+  given initial capacity, or, if 0, the default granularity size.  It
+  returns null if there is no system memory available to create the
+  space.  If argument locked is non-zero, the space uses a separate
+  lock to control access. The capacity of the space will grow
+  dynamically as needed to service mspace_malloc requests.  You can
+  control the sizes of incremental increases of this space by
+  compiling with a different DEFAULT_GRANULARITY or dynamically
+  setting with mallopt(M_GRANULARITY, value).
+*/
+mspace create_mspace(size_t capacity, int locked);
+
+/*
+  destroy_mspace destroys the given space, and attempts to return all
+  of its memory back to the system, returning the total number of
+  bytes freed. After destruction, the results of access to all memory
+  used by the space become undefined.
+*/
+size_t destroy_mspace(mspace msp);
+
+/*
+  create_mspace_with_base uses the memory supplied as the initial base
+  of a new mspace. Part (less than 128*sizeof(size_t) bytes) of this
+  space is used for bookkeeping, so the capacity must be at least this
+  large. (Otherwise 0 is returned.) When this initial space is
+  exhausted, additional memory will be obtained from the system.
+  Destroying this space will deallocate all additionally allocated
+  space (if possible) but not the initial base.
+*/
+mspace create_mspace_with_base(void* base, size_t capacity, int locked);
+
+/*
+  mspace_mmap_large_chunks controls whether requests for large chunks
+  are allocated in their own mmapped regions, separate from others in
+  this mspace. By default this is enabled, which reduces
+  fragmentation. However, such chunks are not necessarily released to
+  the system upon destroy_mspace.  Disabling by setting to false may
+  increase fragmentation, but avoids leakage when relying on
+  destroy_mspace to release all memory allocated using this space.
+*/
+int mspace_mmap_large_chunks(mspace msp, int enable);
+
+
+/*
+  mspace_malloc behaves as malloc, but operates within
+  the given space.
+*/
+void* mspace_malloc(mspace msp, size_t bytes);
+
+/*
+  mspace_free behaves as free, but operates within
+  the given space.
+
+  If compiled with FOOTERS==1, mspace_free is not actually needed.
+  free may be called instead of mspace_free because freed chunks from
+  any space are handled by their originating spaces.
+*/
+void mspace_free(mspace msp, void* mem);
+
+/*
+  mspace_realloc behaves as realloc, but operates within
+  the given space.
+
+  If compiled with FOOTERS==1, mspace_realloc is not actually
+  needed.  realloc may be called instead of mspace_realloc because
+  realloced chunks from any space are handled by their originating
+  spaces.
+*/
+void* mspace_realloc(mspace msp, void* mem, size_t newsize);
+
+/*
+  mspace_calloc behaves as calloc, but operates within
+  the given space.
+*/
+void* mspace_calloc(mspace msp, size_t n_elements, size_t elem_size);
+
+/*
+  mspace_memalign behaves as memalign, but operates within
+  the given space.
+*/
+void* mspace_memalign(mspace msp, size_t alignment, size_t bytes);
+
+/*
+  mspace_independent_calloc behaves as independent_calloc, but
+  operates within the given space.
+*/
+void** mspace_independent_calloc(mspace msp, size_t n_elements,
+				 size_t elem_size, void* chunks[]);
+
+/*
+  mspace_independent_comalloc behaves as independent_comalloc, but
+  operates within the given space.
+*/
+void** mspace_independent_comalloc(mspace msp, size_t n_elements,
+				   size_t sizes[], void* chunks[]);
+
+/*
+  mspace_footprint() returns the number of bytes obtained from the
+  system for this space.
+*/
+size_t mspace_footprint(mspace msp);
+
+/*
+  mspace_max_footprint() returns the peak number of bytes obtained from the
+  system for this space.
+*/
+size_t mspace_max_footprint(mspace msp);
+
+
+#if !NO_MALLINFO
+/*
+  mspace_mallinfo behaves as mallinfo, but reports properties of
+  the given space.
+*/
+struct mallinfo mspace_mallinfo(mspace msp);
+#endif /* NO_MALLINFO */
+
+/*
+  malloc_usable_size(void* p) behaves the same as malloc_usable_size;
+*/
+  size_t mspace_usable_size(void* mem);
+
+/*
+  mspace_malloc_stats behaves as malloc_stats, but reports
+  properties of the given space.
+*/
+void mspace_malloc_stats(mspace msp);
+
+/*
+  mspace_trim behaves as malloc_trim, but
+  operates within the given space.
+*/
+int mspace_trim(mspace msp, size_t pad);
+
+/*
+  An alias for mallopt.
+*/
+int mspace_mallopt(int, int);
+
+#endif /* MSPACES */
+
+#ifdef __cplusplus
+};  /* end of extern "C" */
+#endif /* __cplusplus */
+
+/*
+  ========================================================================
+  To make a fully customizable malloc.h header file, cut everything
+  above this line, put into file malloc.h, edit to suit, and #include it
+  on the next line, as well as in programs that use this malloc.
+  ========================================================================
+*/
+
+/* #include "malloc.h" */
+
+/*------------------------------ internal #includes ---------------------- */
+
+#ifdef WIN32
+#ifndef __GNUC__
+#pragma warning( disable : 4146 ) /* no "unsigned" warnings */
+#endif
+#endif /* WIN32 */
+
+#include <stdio.h>       /* for printing in malloc_stats */
+
+#ifndef LACKS_ERRNO_H
+#include <errno.h>       /* for MALLOC_FAILURE_ACTION */
+#endif /* LACKS_ERRNO_H */
+#if FOOTERS
+#include <time.h>        /* for magic initialization */
+#endif /* FOOTERS */
+#ifndef LACKS_STDLIB_H
+#include <stdlib.h>      /* for abort() */
+#endif /* LACKS_STDLIB_H */
+#ifdef DEBUG
+#if ABORT_ON_ASSERT_FAILURE
+#define assert(x) if(!(x)) ABORT
+#else /* ABORT_ON_ASSERT_FAILURE */
+#include <assert.h>
+#endif /* ABORT_ON_ASSERT_FAILURE */
+#else  /* DEBUG */
+#ifndef assert
+#define assert(x)
+#endif
+#define DEBUG 0
+#endif /* DEBUG */
+#ifndef LACKS_STRING_H
+#include <string.h>      /* for memset etc */
+#endif  /* LACKS_STRING_H */
+#if USE_BUILTIN_FFS
+#ifndef LACKS_STRINGS_H
+#include <strings.h>     /* for ffs */
+#endif /* LACKS_STRINGS_H */
+#endif /* USE_BUILTIN_FFS */
+#if HAVE_MMAP
+#ifndef LACKS_SYS_MMAN_H
+#include <sys/mman.h>    /* for mmap */
+#endif /* LACKS_SYS_MMAN_H */
+#ifndef LACKS_FCNTL_H
+#include <fcntl.h>
+#endif /* LACKS_FCNTL_H */
+#endif /* HAVE_MMAP */
+#ifndef LACKS_UNISTD_H
+#include <unistd.h>     /* for sbrk, sysconf */
+#else /* LACKS_UNISTD_H */
+#if !defined(__FreeBSD__) && !defined(__OpenBSD__) && !defined(__NetBSD__)
+extern void*     sbrk(ptrdiff_t);
+#endif /* FreeBSD etc */
+#endif /* LACKS_UNISTD_H */
+
+/* Declarations for locking */
+#if USE_LOCKS
+#ifndef WIN32
+#include <pthread.h>
+#if defined (__SVR4) && defined (__sun)  /* solaris */
+#include <thread.h>
+#endif /* solaris */
+#else
+#ifndef _M_AMD64
+/* These are already defined on AMD64 builds */
+#ifdef __cplusplus
+extern "C" {
+#endif /* __cplusplus */
+#ifndef __MINGW32__
+LONG __cdecl _InterlockedCompareExchange(LONG volatile *Dest, LONG Exchange, LONG Comp);
+LONG __cdecl _InterlockedExchange(LONG volatile *Target, LONG Value);
+#endif
+#ifdef __cplusplus
+}
+#endif /* __cplusplus */
+#endif /* _M_AMD64 */
+#ifndef __MINGW32__
+#pragma intrinsic (_InterlockedCompareExchange)
+#pragma intrinsic (_InterlockedExchange)
+#else
+  /* --[ start GCC compatibility ]----------------------------------------------
+   * Compatibility <intrin_x86.h> header for GCC -- GCC equivalents of intrinsic
+   * Microsoft Visual C++ functions. Originally developed for the ReactOS
+   * (<http://www.reactos.org/>) and TinyKrnl (<http://www.tinykrnl.org/>)
+   * projects.
+   *
+   * Copyright (c) 2006 KJK::Hyperion <hackbunny@reactos.com>
+   *
+   * Permission is hereby granted, free of charge, to any person obtaining a
+   * copy of this software and associated documentation files (the "Software"),
+   * to deal in the Software without restriction, including without limitation
+   * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   * and/or sell copies of the Software, and to permit persons to whom the
+   * Software is furnished to do so, subject to the following conditions:
+   *
+   * The above copyright notice and this permission notice shall be included in
+   * all copies or substantial portions of the Software.
+   *
+   * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   * DEALINGS IN THE SOFTWARE.
+   */
+
+  /*** Atomic operations ***/
+  #if (__GNUC__ * 10000 + __GNUC_MINOR__ * 100 + __GNUC_PATCHLEVEL__) > 40100
+    #define _ReadWriteBarrier() __sync_synchronize()
+  #else
+    static __inline__ __attribute__((always_inline)) long __sync_lock_test_and_set(volatile long * const Target, const long Value)
+    {
+      long res;
+      __asm__ __volatile__("xchg%z0 %2, %0" : "=g" (*(Target)), "=r" (res) : "1" (Value));
+      return res;
+    }
+    static void __inline__ __attribute__((always_inline)) _MemoryBarrier(void)
+    {
+      __asm__ __volatile__("" : : : "memory");
+    }
+    #define _ReadWriteBarrier() _MemoryBarrier()
+  #endif
+  /* BUGBUG: GCC only supports full barriers */
+  static __inline__ __attribute__((always_inline)) long _InterlockedExchange(volatile long * const Target, const long Value)
+  {
+    /* NOTE: __sync_lock_test_and_set would be an acquire barrier, so we force a full barrier */
+    _ReadWriteBarrier();
+    return __sync_lock_test_and_set(Target, Value);
+  }
+  /* --[ end GCC compatibility ]---------------------------------------------- */
+#endif
+#define interlockedcompareexchange _InterlockedCompareExchange
+#define interlockedexchange _InterlockedExchange
+#endif /* Win32 */
+#endif /* USE_LOCKS */
+
+/* Declarations for bit scanning on win32 */
+#if defined(_MSC_VER) && _MSC_VER>=1300
+#ifndef BitScanForward	/* Try to avoid pulling in WinNT.h */
+#ifdef __cplusplus
+extern "C" {
+#endif /* __cplusplus */
+unsigned char _BitScanForward(unsigned long *index, unsigned long mask);
+unsigned char _BitScanReverse(unsigned long *index, unsigned long mask);
+#ifdef __cplusplus
+}
+#endif /* __cplusplus */
+
+#define BitScanForward _BitScanForward
+#define BitScanReverse _BitScanReverse
+#pragma intrinsic(_BitScanForward)
+#pragma intrinsic(_BitScanReverse)
+#endif /* BitScanForward */
+#endif /* defined(_MSC_VER) && _MSC_VER>=1300 */
+
+#ifndef WIN32
+#ifndef malloc_getpagesize
+#  ifdef _SC_PAGESIZE         /* some SVR4 systems omit an underscore */
+#    ifndef _SC_PAGE_SIZE
+#      define _SC_PAGE_SIZE _SC_PAGESIZE
+#    endif
+#  endif
+#  ifdef _SC_PAGE_SIZE
+#    define malloc_getpagesize sysconf(_SC_PAGE_SIZE)
+#  else
+#    if defined(BSD) || defined(DGUX) || defined(HAVE_GETPAGESIZE)
+       extern size_t getpagesize();
+#      define malloc_getpagesize getpagesize()
+#    else
+#      ifdef WIN32 /* use supplied emulation of getpagesize */
+#        define malloc_getpagesize getpagesize()
+#      else
+#        ifndef LACKS_SYS_PARAM_H
+#          include <sys/param.h>
+#        endif
+#        ifdef EXEC_PAGESIZE
+#          define malloc_getpagesize EXEC_PAGESIZE
+#        else
+#          ifdef NBPG
+#            ifndef CLSIZE
+#              define malloc_getpagesize NBPG
+#            else
+#              define malloc_getpagesize (NBPG * CLSIZE)
+#            endif
+#          else
+#            ifdef NBPC
+#              define malloc_getpagesize NBPC
+#            else
+#              ifdef PAGESIZE
+#                define malloc_getpagesize PAGESIZE
+#              else /* just guess */
+#                define malloc_getpagesize ((size_t)4096U)
+#              endif
+#            endif
+#          endif
+#        endif
+#      endif
+#    endif
+#  endif
+#endif
+#endif
+
+
+
+/* ------------------- size_t and alignment properties -------------------- */
+
+/* The byte and bit size of a size_t */
+#define SIZE_T_SIZE         (sizeof(size_t))
+#define SIZE_T_BITSIZE      (sizeof(size_t) << 3)
+
+/* Some constants coerced to size_t */
+/* Annoying but necessary to avoid errors on some platforms */
+#define SIZE_T_ZERO         ((size_t)0)
+#define SIZE_T_ONE          ((size_t)1)
+#define SIZE_T_TWO          ((size_t)2)
+#define SIZE_T_FOUR         ((size_t)4)
+#define TWO_SIZE_T_SIZES    (SIZE_T_SIZE<<1)
+#define FOUR_SIZE_T_SIZES   (SIZE_T_SIZE<<2)
+#define SIX_SIZE_T_SIZES    (FOUR_SIZE_T_SIZES+TWO_SIZE_T_SIZES)
+#define HALF_MAX_SIZE_T     (MAX_SIZE_T / 2U)
+
+/* The bit mask value corresponding to MALLOC_ALIGNMENT */
+#define CHUNK_ALIGN_MASK    (MALLOC_ALIGNMENT - SIZE_T_ONE)
+
+/* True if address a has acceptable alignment */
+#define is_aligned(A)       (((size_t)((A)) & (CHUNK_ALIGN_MASK)) == 0)
+
+/* the number of bytes to offset an address to align it */
+#define align_offset(A)\
+ ((((size_t)(A) & CHUNK_ALIGN_MASK) == 0)? 0 :\
+  ((MALLOC_ALIGNMENT - ((size_t)(A) & CHUNK_ALIGN_MASK)) & CHUNK_ALIGN_MASK))
+
+/* -------------------------- MMAP preliminaries ------------------------- */
+
+/*
+   If HAVE_MORECORE or HAVE_MMAP are false, we just define calls and
+   checks to fail so compiler optimizer can delete code rather than
+   using so many "#if"s.
+*/
+
+
+/* MORECORE and MMAP must return MFAIL on failure */
+#define MFAIL                ((void*)(MAX_SIZE_T))
+#define CMFAIL               ((char*)(MFAIL)) /* defined for convenience */
+
+#if HAVE_MMAP
+
+#ifndef WIN32
+#define MUNMAP_DEFAULT(a, s)  munmap((a), (s))
+#define MMAP_PROT            (PROT_READ|PROT_WRITE)
+#if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
+#define MAP_ANONYMOUS        MAP_ANON
+#endif /* MAP_ANON */
+#ifdef MAP_ANONYMOUS
+#define MMAP_FLAGS           (MAP_PRIVATE|MAP_ANONYMOUS)
+#define MMAP_DEFAULT(s)       mmap(0, (s), MMAP_PROT, MMAP_FLAGS, -1, 0)
+#else /* MAP_ANONYMOUS */
+/*
+   Nearly all versions of mmap support MAP_ANONYMOUS, so the following
+   is unlikely to be needed, but is supplied just in case.
+*/
+#define MMAP_FLAGS           (MAP_PRIVATE)
+static int dev_zero_fd = -1; /* Cached file descriptor for /dev/zero. */
+#define MMAP_DEFAULT(s) ((dev_zero_fd < 0) ? \
+	   (dev_zero_fd = open("/dev/zero", O_RDWR), \
+	    mmap(0, (s), MMAP_PROT, MMAP_FLAGS, dev_zero_fd, 0)) : \
+	    mmap(0, (s), MMAP_PROT, MMAP_FLAGS, dev_zero_fd, 0))
+#endif /* MAP_ANONYMOUS */
+
+#define DIRECT_MMAP_DEFAULT(s) MMAP_DEFAULT(s)
+
+#else /* WIN32 */
+
+/* Win32 MMAP via VirtualAlloc */
+static FORCEINLINE void* win32mmap(size_t size) {
+  void* ptr = VirtualAlloc(0, size, MEM_RESERVE|MEM_COMMIT, PAGE_READWRITE);
+  return (ptr != 0)? ptr: MFAIL;
+}
+
+/* For direct MMAP, use MEM_TOP_DOWN to minimize interference */
+static FORCEINLINE void* win32direct_mmap(size_t size) {
+  void* ptr = VirtualAlloc(0, size, MEM_RESERVE|MEM_COMMIT|MEM_TOP_DOWN,
+			   PAGE_READWRITE);
+  return (ptr != 0)? ptr: MFAIL;
+}
+
+/* This function supports releasing coalesed segments */
+static FORCEINLINE int win32munmap(void* ptr, size_t size) {
+  MEMORY_BASIC_INFORMATION minfo;
+  char* cptr = (char*)ptr;
+  while (size) {
+    if (VirtualQuery(cptr, &minfo, sizeof(minfo)) == 0)
+      return -1;
+    if (minfo.BaseAddress != cptr || minfo.AllocationBase != cptr ||
+	minfo.State != MEM_COMMIT || minfo.RegionSize > size)
+      return -1;
+    if (VirtualFree(cptr, 0, MEM_RELEASE) == 0)
+      return -1;
+    cptr += minfo.RegionSize;
+    size -= minfo.RegionSize;
+  }
+  return 0;
+}
+
+#define MMAP_DEFAULT(s)             win32mmap(s)
+#define MUNMAP_DEFAULT(a, s)        win32munmap((a), (s))
+#define DIRECT_MMAP_DEFAULT(s)      win32direct_mmap(s)
+#endif /* WIN32 */
+#endif /* HAVE_MMAP */
+
+#if HAVE_MREMAP
+#ifndef WIN32
+#define MREMAP_DEFAULT(addr, osz, nsz, mv) mremap((addr), (osz), (nsz), (mv))
+#endif /* WIN32 */
+#endif /* HAVE_MREMAP */
+
+
+/**
+ * Define CALL_MORECORE
+ */
+#if HAVE_MORECORE
+    #ifdef MORECORE
+	#define CALL_MORECORE(S)    MORECORE(S)
+    #else  /* MORECORE */
+	#define CALL_MORECORE(S)    MORECORE_DEFAULT(S)
+    #endif /* MORECORE */
+#else  /* HAVE_MORECORE */
+    #define CALL_MORECORE(S)        MFAIL
+#endif /* HAVE_MORECORE */
+
+/**
+ * Define CALL_MMAP/CALL_MUNMAP/CALL_DIRECT_MMAP
+ */
+#if HAVE_MMAP
+    #define IS_MMAPPED_BIT          (SIZE_T_ONE)
+    #define USE_MMAP_BIT            (SIZE_T_ONE)
+
+    #ifdef MMAP
+	#define CALL_MMAP(s)        MMAP(s)
+    #else /* MMAP */
+	#define CALL_MMAP(s)        MMAP_DEFAULT(s)
+    #endif /* MMAP */
+    #ifdef MUNMAP
+	#define CALL_MUNMAP(a, s)   MUNMAP((a), (s))
+    #else /* MUNMAP */
+	#define CALL_MUNMAP(a, s)   MUNMAP_DEFAULT((a), (s))
+    #endif /* MUNMAP */
+    #ifdef DIRECT_MMAP
+	#define CALL_DIRECT_MMAP(s) DIRECT_MMAP(s)
+    #else /* DIRECT_MMAP */
+	#define CALL_DIRECT_MMAP(s) DIRECT_MMAP_DEFAULT(s)
+    #endif /* DIRECT_MMAP */
+#else  /* HAVE_MMAP */
+    #define IS_MMAPPED_BIT          (SIZE_T_ZERO)
+    #define USE_MMAP_BIT            (SIZE_T_ZERO)
+
+    #define MMAP(s)                 MFAIL
+    #define MUNMAP(a, s)            (-1)
+    #define DIRECT_MMAP(s)          MFAIL
+    #define CALL_DIRECT_MMAP(s)     DIRECT_MMAP(s)
+    #define CALL_MMAP(s)            MMAP(s)
+    #define CALL_MUNMAP(a, s)       MUNMAP((a), (s))
+#endif /* HAVE_MMAP */
+
+/**
+ * Define CALL_MREMAP
+ */
+#if HAVE_MMAP && HAVE_MREMAP
+    #ifdef MREMAP
+	#define CALL_MREMAP(addr, osz, nsz, mv) MREMAP((addr), (osz), (nsz), (mv))
+    #else /* MREMAP */
+	#define CALL_MREMAP(addr, osz, nsz, mv) MREMAP_DEFAULT((addr), (osz), (nsz), (mv))
+    #endif /* MREMAP */
+#else  /* HAVE_MMAP && HAVE_MREMAP */
+    #define CALL_MREMAP(addr, osz, nsz, mv)     MFAIL
+#endif /* HAVE_MMAP && HAVE_MREMAP */
+
+/* mstate bit set if continguous morecore disabled or failed */
+#define USE_NONCONTIGUOUS_BIT (4U)
+
+/* segment bit set in create_mspace_with_base */
+#define EXTERN_BIT            (8U)
+
+
+/* --------------------------- Lock preliminaries ------------------------ */
+
+/*
+  When locks are defined, there is one global lock, plus
+  one per-mspace lock.
+
+  The global lock_ensures that mparams.magic and other unique
+  mparams values are initialized only once. It also protects
+  sequences of calls to MORECORE.  In many cases sys_alloc requires
+  two calls, that should not be interleaved with calls by other
+  threads.  This does not protect against direct calls to MORECORE
+  by other threads not using this lock, so there is still code to
+  cope the best we can on interference.
+
+  Per-mspace locks surround calls to malloc, free, etc.  To enable use
+  in layered extensions, per-mspace locks are reentrant.
+
+  Because lock-protected regions generally have bounded times, it is
+  OK to use the supplied simple spinlocks in the custom versions for
+  x86.
+
+  If USE_LOCKS is > 1, the definitions of lock routines here are
+  bypassed, in which case you will need to define at least
+  INITIAL_LOCK, ACQUIRE_LOCK, RELEASE_LOCK and possibly TRY_LOCK
+  (which is not used in this malloc, but commonly needed in
+  extensions.)
+*/
+
+#if USE_LOCKS == 1
+
+#if USE_SPIN_LOCKS
+#ifndef WIN32
+
+/* Custom pthread-style spin locks on x86 and x64 for gcc */
+struct pthread_mlock_t {
+  volatile unsigned int l;
+  volatile unsigned int c;
+  volatile pthread_t threadid;
+};
+#define MLOCK_T struct        pthread_mlock_t
+#define CURRENT_THREAD        pthread_self()
+#define INITIAL_LOCK(sl)      (memset(sl, 0, sizeof(MLOCK_T)), 0)
+#define ACQUIRE_LOCK(sl)      pthread_acquire_lock(sl)
+#define RELEASE_LOCK(sl)      pthread_release_lock(sl)
+#define TRY_LOCK(sl)          pthread_try_lock(sl)
+#define SPINS_PER_YIELD       63
+
+static MLOCK_T malloc_global_mutex = { 0, 0, 0};
+
+static FORCEINLINE int pthread_acquire_lock (MLOCK_T *sl) {
+  int spins = 0;
+  volatile unsigned int* lp = &sl->l;
+  for (;;) {
+    if (*lp != 0) {
+      if (sl->threadid == CURRENT_THREAD) {
+	++sl->c;
+	return 0;
+      }
+    }
+    else {
+      /* place args to cmpxchgl in locals to evade oddities in some gccs */
+      int cmp = 0;
+      int val = 1;
+      int ret;
+      __asm__ __volatile__  ("lock; cmpxchgl %1, %2"
+			     : "=a" (ret)
+			     : "r" (val), "m" (*(lp)), "0"(cmp)
+			     : "memory", "cc");
+      if (!ret) {
+	assert(!sl->threadid);
+	sl->c = 1;
+	sl->threadid = CURRENT_THREAD;
+	return 0;
+      }
+      if ((++spins & SPINS_PER_YIELD) == 0) {
+#if defined (__SVR4) && defined (__sun) /* solaris */
+	thr_yield();
+#else
+#if defined(__linux__) || defined(__FreeBSD__) || defined(__APPLE__)
+	sched_yield();
+#else  /* no-op yield on unknown systems */
+	;
+#endif /* __linux__ || __FreeBSD__ || __APPLE__ */
+#endif /* solaris */
+      }
+    }
+  }
+}
+
+static FORCEINLINE void pthread_release_lock (MLOCK_T *sl) {
+  assert(sl->l != 0);
+  assert(sl->threadid == CURRENT_THREAD);
+  if (--sl->c == 0) {
+    sl->threadid = 0;
+    volatile unsigned int* lp = &sl->l;
+    int prev = 0;
+    int ret;
+    __asm__ __volatile__ ("lock; xchgl %0, %1"
+			  : "=r" (ret)
+			  : "m" (*(lp)), "0"(prev)
+			  : "memory");
+  }
+}
+
+static FORCEINLINE int pthread_try_lock (MLOCK_T *sl) {
+  volatile unsigned int* lp = &sl->l;
+  if (*lp != 0) {
+      if (sl->threadid == CURRENT_THREAD) {
+	++sl->c;
+	return 1;
+      }
+  }
+  else {
+    int cmp = 0;
+    int val = 1;
+    int ret;
+    __asm__ __volatile__  ("lock; cmpxchgl %1, %2"
+			   : "=a" (ret)
+			   : "r" (val), "m" (*(lp)), "0"(cmp)
+			   : "memory", "cc");
+    if (!ret) {
+      assert(!sl->threadid);
+      sl->c = 1;
+      sl->threadid = CURRENT_THREAD;
+      return 1;
+    }
+  }
+  return 0;
+}
+
+
+#else /* WIN32 */
+/* Custom win32-style spin locks on x86 and x64 for MSC */
+struct win32_mlock_t
+{
+  volatile long l;
+  volatile unsigned int c;
+  volatile long threadid;
+};
+
+#define MLOCK_T               struct win32_mlock_t
+#define CURRENT_THREAD        win32_getcurrentthreadid()
+#define INITIAL_LOCK(sl)      (memset(sl, 0, sizeof(MLOCK_T)), 0)
+#define ACQUIRE_LOCK(sl)      win32_acquire_lock(sl)
+#define RELEASE_LOCK(sl)      win32_release_lock(sl)
+#define TRY_LOCK(sl)          win32_try_lock(sl)
+#define SPINS_PER_YIELD       63
+
+static MLOCK_T malloc_global_mutex = { 0, 0, 0};
+
+static FORCEINLINE long win32_getcurrentthreadid() {
+#ifdef _MSC_VER
+#if defined(_M_IX86)
+  long *threadstruct=(long *)__readfsdword(0x18);
+  long threadid=threadstruct[0x24/sizeof(long)];
+  return threadid;
+#elif defined(_M_X64)
+  /* todo */
+  return GetCurrentThreadId();
+#else
+  return GetCurrentThreadId();
+#endif
+#else
+  return GetCurrentThreadId();
+#endif
+}
+
+static FORCEINLINE int win32_acquire_lock (MLOCK_T *sl) {
+  int spins = 0;
+  for (;;) {
+    if (sl->l != 0) {
+      if (sl->threadid == CURRENT_THREAD) {
+	++sl->c;
+	return 0;
+      }
+    }
+    else {
+      if (!interlockedexchange(&sl->l, 1)) {
+	assert(!sl->threadid);
+		sl->c=CURRENT_THREAD;
+	sl->threadid = CURRENT_THREAD;
+	sl->c = 1;
+	return 0;
+      }
+    }
+    if ((++spins & SPINS_PER_YIELD) == 0)
+      SleepEx(0, FALSE);
+  }
+}
+
+static FORCEINLINE void win32_release_lock (MLOCK_T *sl) {
+  assert(sl->threadid == CURRENT_THREAD);
+  assert(sl->l != 0);
+  if (--sl->c == 0) {
+    sl->threadid = 0;
+    interlockedexchange (&sl->l, 0);
+  }
+}
+
+static FORCEINLINE int win32_try_lock (MLOCK_T *sl) {
+  if(sl->l != 0) {
+      if (sl->threadid == CURRENT_THREAD) {
+	++sl->c;
+	return 1;
+      }
+  }
+  else {
+    if (!interlockedexchange(&sl->l, 1)){
+      assert(!sl->threadid);
+      sl->threadid = CURRENT_THREAD;
+      sl->c = 1;
+      return 1;
+    }
+  }
+  return 0;
+}
+
+#endif /* WIN32 */
+#else /* USE_SPIN_LOCKS */
+
+#ifndef WIN32
+/* pthreads-based locks */
+
+#define MLOCK_T               pthread_mutex_t
+#define CURRENT_THREAD        pthread_self()
+#define INITIAL_LOCK(sl)      pthread_init_lock(sl)
+#define ACQUIRE_LOCK(sl)      pthread_mutex_lock(sl)
+#define RELEASE_LOCK(sl)      pthread_mutex_unlock(sl)
+#define TRY_LOCK(sl)          (!pthread_mutex_trylock(sl))
+
+static MLOCK_T malloc_global_mutex = PTHREAD_MUTEX_INITIALIZER;
+
+/* Cope with old-style linux recursive lock initialization by adding */
+/* skipped internal declaration from pthread.h */
+#ifdef linux
+#ifndef PTHREAD_MUTEX_RECURSIVE
+extern int pthread_mutexattr_setkind_np __P ((pthread_mutexattr_t *__attr,
+					   int __kind));
+#define PTHREAD_MUTEX_RECURSIVE PTHREAD_MUTEX_RECURSIVE_NP
+#define pthread_mutexattr_settype(x,y) pthread_mutexattr_setkind_np(x,y)
+#endif
+#endif
+
+static int pthread_init_lock (MLOCK_T *sl) {
+  pthread_mutexattr_t attr;
+  if (pthread_mutexattr_init(&attr)) return 1;
+  if (pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE)) return 1;
+  if (pthread_mutex_init(sl, &attr)) return 1;
+  if (pthread_mutexattr_destroy(&attr)) return 1;
+  return 0;
+}
+
+#else /* WIN32 */
+/* Win32 critical sections */
+#define MLOCK_T               CRITICAL_SECTION
+#define CURRENT_THREAD        GetCurrentThreadId()
+#define INITIAL_LOCK(s)       (!InitializeCriticalSectionAndSpinCount((s), 0x80000000|4000))
+#define ACQUIRE_LOCK(s)       (EnterCriticalSection(s), 0)
+#define RELEASE_LOCK(s)       LeaveCriticalSection(s)
+#define TRY_LOCK(s)           TryEnterCriticalSection(s)
+#define NEED_GLOBAL_LOCK_INIT
+
+static MLOCK_T malloc_global_mutex;
+static volatile long malloc_global_mutex_status;
+
+/* Use spin loop to initialize global lock */
+static void init_malloc_global_mutex() {
+  for (;;) {
+    long stat = malloc_global_mutex_status;
+    if (stat > 0)
+      return;
+    /* transition to < 0 while initializing, then to > 0) */
+    if (stat == 0 &&
+	interlockedcompareexchange(&malloc_global_mutex_status, -1, 0) == 0) {
+      InitializeCriticalSection(&malloc_global_mutex);
+      interlockedexchange(&malloc_global_mutex_status,1);
+      return;
+    }
+    SleepEx(0, FALSE);
+  }
+}
+
+#endif /* WIN32 */
+#endif /* USE_SPIN_LOCKS */
+#endif /* USE_LOCKS == 1 */
+
+/* -----------------------  User-defined locks ------------------------ */
+
+#if USE_LOCKS > 1
+/* Define your own lock implementation here */
+/* #define INITIAL_LOCK(sl)  ... */
+/* #define ACQUIRE_LOCK(sl)  ... */
+/* #define RELEASE_LOCK(sl)  ... */
+/* #define TRY_LOCK(sl) ... */
+/* static MLOCK_T malloc_global_mutex = ... */
+#endif /* USE_LOCKS > 1 */
+
+/* -----------------------  Lock-based state ------------------------ */
+
+#if USE_LOCKS
+#define USE_LOCK_BIT               (2U)
+#else  /* USE_LOCKS */
+#define USE_LOCK_BIT               (0U)
+#define INITIAL_LOCK(l)
+#endif /* USE_LOCKS */
+
+#if USE_LOCKS
+#define ACQUIRE_MALLOC_GLOBAL_LOCK()  ACQUIRE_LOCK(&malloc_global_mutex);
+#define RELEASE_MALLOC_GLOBAL_LOCK()  RELEASE_LOCK(&malloc_global_mutex);
+#else  /* USE_LOCKS */
+#define ACQUIRE_MALLOC_GLOBAL_LOCK()
+#define RELEASE_MALLOC_GLOBAL_LOCK()
+#endif /* USE_LOCKS */
+
+
+/* -----------------------  Chunk representations ------------------------ */
+
+/*
+  (The following includes lightly edited explanations by Colin Plumb.)
+
+  The malloc_chunk declaration below is misleading (but accurate and
+  necessary).  It declares a "view" into memory allowing access to
+  necessary fields at known offsets from a given base.
+
+  Chunks of memory are maintained using a `boundary tag' method as
+  originally described by Knuth.  (See the paper by Paul Wilson
+  ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a survey of such
+  techniques.)  Sizes of free chunks are stored both in the front of
+  each chunk and at the end.  This makes consolidating fragmented
+  chunks into bigger chunks fast.  The head fields also hold bits
+  representing whether chunks are free or in use.
+
+  Here are some pictures to make it clearer.  They are "exploded" to
+  show that the state of a chunk can be thought of as extending from
+  the high 31 bits of the head field of its header through the
+  prev_foot and PINUSE_BIT bit of the following chunk header.
+
+  A chunk that's in use looks like:
+
+   chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	   | Size of previous chunk (if P = 0)                             |
+	   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |P|
+	 | Size of this chunk                                         1| +-+
+   mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	 |                                                               |
+	 +-                                                             -+
+	 |                                                               |
+	 +-                                                             -+
+	 |                                                               :
+	 +-      size - sizeof(size_t) available payload bytes          -+
+	 :                                                               |
+ chunk-> +-                                                             -+
+	 |                                                               |
+	 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |1|
+       | Size of next chunk (may or may not be in use)               | +-+
+ mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+
+    And if it's free, it looks like this:
+
+   chunk-> +-                                                             -+
+	   | User payload (must be in use, or we would have merged!)       |
+	   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |P|
+	 | Size of this chunk                                         0| +-+
+   mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	 | Next pointer                                                  |
+	 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	 | Prev pointer                                                  |
+	 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	 |                                                               :
+	 +-      size - sizeof(struct chunk) unused bytes               -+
+	 :                                                               |
+ chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	 | Size of this chunk                                            |
+	 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |0|
+       | Size of next chunk (must be in use, or we would have merged)| +-+
+ mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+       |                                                               :
+       +- User payload                                                -+
+       :                                                               |
+       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+								     |0|
+								     +-+
+  Note that since we always merge adjacent free chunks, the chunks
+  adjacent to a free chunk must be in use.
+
+  Given a pointer to a chunk (which can be derived trivially from the
+  payload pointer) we can, in O(1) time, find out whether the adjacent
+  chunks are free, and if so, unlink them from the lists that they
+  are on and merge them with the current chunk.
+
+  Chunks always begin on even word boundaries, so the mem portion
+  (which is returned to the user) is also on an even word boundary, and
+  thus at least double-word aligned.
+
+  The P (PINUSE_BIT) bit, stored in the unused low-order bit of the
+  chunk size (which is always a multiple of two words), is an in-use
+  bit for the *previous* chunk.  If that bit is *clear*, then the
+  word before the current chunk size contains the previous chunk
+  size, and can be used to find the front of the previous chunk.
+  The very first chunk allocated always has this bit set, preventing
+  access to non-existent (or non-owned) memory. If pinuse is set for
+  any given chunk, then you CANNOT determine the size of the
+  previous chunk, and might even get a memory addressing fault when
+  trying to do so.
+
+  The C (CINUSE_BIT) bit, stored in the unused second-lowest bit of
+  the chunk size redundantly records whether the current chunk is
+  inuse. This redundancy enables usage checks within free and realloc,
+  and reduces indirection when freeing and consolidating chunks.
+
+  Each freshly allocated chunk must have both cinuse and pinuse set.
+  That is, each allocated chunk borders either a previously allocated
+  and still in-use chunk, or the base of its memory arena. This is
+  ensured by making all allocations from the the `lowest' part of any
+  found chunk.  Further, no free chunk physically borders another one,
+  so each free chunk is known to be preceded and followed by either
+  inuse chunks or the ends of memory.
+
+  Note that the `foot' of the current chunk is actually represented
+  as the prev_foot of the NEXT chunk. This makes it easier to
+  deal with alignments etc but can be very confusing when trying
+  to extend or adapt this code.
+
+  The exceptions to all this are
+
+     1. The special chunk `top' is the top-most available chunk (i.e.,
+	the one bordering the end of available memory). It is treated
+	specially.  Top is never included in any bin, is used only if
+	no other chunk is available, and is released back to the
+	system if it is very large (see M_TRIM_THRESHOLD).  In effect,
+	the top chunk is treated as larger (and thus less well
+	fitting) than any other available chunk.  The top chunk
+	doesn't update its trailing size field since there is no next
+	contiguous chunk that would have to index off it. However,
+	space is still allocated for it (TOP_FOOT_SIZE) to enable
+	separation or merging when space is extended.
+
+     3. Chunks allocated via mmap, which have the lowest-order bit
+	(IS_MMAPPED_BIT) set in their prev_foot fields, and do not set
+	PINUSE_BIT in their head fields.  Because they are allocated
+	one-by-one, each must carry its own prev_foot field, which is
+	also used to hold the offset this chunk has within its mmapped
+	region, which is needed to preserve alignment. Each mmapped
+	chunk is trailed by the first two fields of a fake next-chunk
+	for sake of usage checks.
+
+*/
+
+struct malloc_chunk {
+  size_t               prev_foot;  /* Size of previous chunk (if free).  */
+  size_t               head;       /* Size and inuse bits. */
+  struct malloc_chunk* fd;         /* double links -- used only if free. */
+  struct malloc_chunk* bk;
+};
+
+typedef struct malloc_chunk  mchunk;
+typedef struct malloc_chunk* mchunkptr;
+typedef struct malloc_chunk* sbinptr;  /* The type of bins of chunks */
+typedef unsigned int bindex_t;         /* Described below */
+typedef unsigned int binmap_t;         /* Described below */
+typedef unsigned int flag_t;           /* The type of various bit flag sets */
+
+/* ------------------- Chunks sizes and alignments ----------------------- */
+
+#define MCHUNK_SIZE         (sizeof(mchunk))
+
+#if FOOTERS
+#define CHUNK_OVERHEAD      (TWO_SIZE_T_SIZES)
+#else /* FOOTERS */
+#define CHUNK_OVERHEAD      (SIZE_T_SIZE)
+#endif /* FOOTERS */
+
+/* MMapped chunks need a second word of overhead ... */
+#define MMAP_CHUNK_OVERHEAD (TWO_SIZE_T_SIZES)
+/* ... and additional padding for fake next-chunk at foot */
+#define MMAP_FOOT_PAD       (FOUR_SIZE_T_SIZES)
+
+/* The smallest size we can malloc is an aligned minimal chunk */
+#define MIN_CHUNK_SIZE\
+  ((MCHUNK_SIZE + CHUNK_ALIGN_MASK) & ~CHUNK_ALIGN_MASK)
+
+/* conversion from malloc headers to user pointers, and back */
+#define chunk2mem(p)        ((void*)((char*)(p)       + TWO_SIZE_T_SIZES))
+#define mem2chunk(mem)      ((mchunkptr)((char*)(mem) - TWO_SIZE_T_SIZES))
+/* chunk associated with aligned address A */
+#define align_as_chunk(A)   (mchunkptr)((A) + align_offset(chunk2mem(A)))
+
+/* Bounds on request (not chunk) sizes. */
+#define MAX_REQUEST         ((-MIN_CHUNK_SIZE) << 2)
+#define MIN_REQUEST         (MIN_CHUNK_SIZE - CHUNK_OVERHEAD - SIZE_T_ONE)
+
+/* pad request bytes into a usable size */
+#define pad_request(req) \
+   (((req) + CHUNK_OVERHEAD + CHUNK_ALIGN_MASK) & ~CHUNK_ALIGN_MASK)
+
+/* pad request, checking for minimum (but not maximum) */
+#define request2size(req) \
+  (((req) < MIN_REQUEST)? MIN_CHUNK_SIZE : pad_request(req))
+
+
+/* ------------------ Operations on head and foot fields ----------------- */
+
+/*
+  The head field of a chunk is or'ed with PINUSE_BIT when previous
+  adjacent chunk in use, and or'ed with CINUSE_BIT if this chunk is in
+  use. If the chunk was obtained with mmap, the prev_foot field has
+  IS_MMAPPED_BIT set, otherwise holding the offset of the base of the
+  mmapped region to the base of the chunk.
+
+  FLAG4_BIT is not used by this malloc, but might be useful in extensions.
+*/
+
+#define PINUSE_BIT          (SIZE_T_ONE)
+#define CINUSE_BIT          (SIZE_T_TWO)
+#define FLAG4_BIT           (SIZE_T_FOUR)
+#define INUSE_BITS          (PINUSE_BIT|CINUSE_BIT)
+#define FLAG_BITS           (PINUSE_BIT|CINUSE_BIT|FLAG4_BIT)
+
+/* Head value for fenceposts */
+#define FENCEPOST_HEAD      (INUSE_BITS|SIZE_T_SIZE)
+
+/* extraction of fields from head words */
+#define cinuse(p)           ((p)->head & CINUSE_BIT)
+#define pinuse(p)           ((p)->head & PINUSE_BIT)
+#define chunksize(p)        ((p)->head & ~(FLAG_BITS))
+
+#define clear_pinuse(p)     ((p)->head &= ~PINUSE_BIT)
+#define clear_cinuse(p)     ((p)->head &= ~CINUSE_BIT)
+
+/* Treat space at ptr +/- offset as a chunk */
+#define chunk_plus_offset(p, s)  ((mchunkptr)(((char*)(p)) + (s)))
+#define chunk_minus_offset(p, s) ((mchunkptr)(((char*)(p)) - (s)))
+
+/* Ptr to next or previous physical malloc_chunk. */
+#define next_chunk(p) ((mchunkptr)( ((char*)(p)) + ((p)->head & ~FLAG_BITS)))
+#define prev_chunk(p) ((mchunkptr)( ((char*)(p)) - ((p)->prev_foot) ))
+
+/* extract next chunk's pinuse bit */
+#define next_pinuse(p)  ((next_chunk(p)->head) & PINUSE_BIT)
+
+/* Get/set size at footer */
+#define get_foot(p, s)  (((mchunkptr)((char*)(p) + (s)))->prev_foot)
+#define set_foot(p, s)  (((mchunkptr)((char*)(p) + (s)))->prev_foot = (s))
+
+/* Set size, pinuse bit, and foot */
+#define set_size_and_pinuse_of_free_chunk(p, s)\
+  ((p)->head = (s|PINUSE_BIT), set_foot(p, s))
+
+/* Set size, pinuse bit, foot, and clear next pinuse */
+#define set_free_with_pinuse(p, s, n)\
+  (clear_pinuse(n), set_size_and_pinuse_of_free_chunk(p, s))
+
+#define is_mmapped(p)\
+  (!((p)->head & PINUSE_BIT) && ((p)->prev_foot & IS_MMAPPED_BIT))
+
+/* Get the internal overhead associated with chunk p */
+#define overhead_for(p)\
+ (is_mmapped(p)? MMAP_CHUNK_OVERHEAD : CHUNK_OVERHEAD)
+
+/* Return true if malloced space is not necessarily cleared */
+#if MMAP_CLEARS
+#define calloc_must_clear(p) (!is_mmapped(p))
+#else /* MMAP_CLEARS */
+#define calloc_must_clear(p) (1)
+#endif /* MMAP_CLEARS */
+
+/* ---------------------- Overlaid data structures ----------------------- */
+
+/*
+  When chunks are not in use, they are treated as nodes of either
+  lists or trees.
+
+  "Small"  chunks are stored in circular doubly-linked lists, and look
+  like this:
+
+    chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Size of previous chunk                            |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    `head:' |             Size of chunk, in bytes                         |P|
+      mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Forward pointer to next chunk in list             |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Back pointer to previous chunk in list            |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Unused space (may be 0 bytes long)                .
+	    .                                                               .
+	    .                                                               |
+nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    `foot:' |             Size of chunk, in bytes                           |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+
+  Larger chunks are kept in a form of bitwise digital trees (aka
+  tries) keyed on chunksizes.  Because malloc_tree_chunks are only for
+  free chunks greater than 256 bytes, their size doesn't impose any
+  constraints on user chunk sizes.  Each node looks like:
+
+    chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Size of previous chunk                            |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    `head:' |             Size of chunk, in bytes                         |P|
+      mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Forward pointer to next chunk of same size        |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Back pointer to previous chunk of same size       |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Pointer to left child (child[0])                  |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Pointer to right child (child[1])                 |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Pointer to parent                                 |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             bin index of this chunk                           |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Unused space                                      .
+	    .                                                               |
+nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    `foot:' |             Size of chunk, in bytes                           |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+
+  Each tree holding treenodes is a tree of unique chunk sizes.  Chunks
+  of the same size are arranged in a circularly-linked list, with only
+  the oldest chunk (the next to be used, in our FIFO ordering)
+  actually in the tree.  (Tree members are distinguished by a non-null
+  parent pointer.)  If a chunk with the same size an an existing node
+  is inserted, it is linked off the existing node using pointers that
+  work in the same way as fd/bk pointers of small chunks.
+
+  Each tree contains a power of 2 sized range of chunk sizes (the
+  smallest is 0x100 <= x < 0x180), which is is divided in half at each
+  tree level, with the chunks in the smaller half of the range (0x100
+  <= x < 0x140 for the top nose) in the left subtree and the larger
+  half (0x140 <= x < 0x180) in the right subtree.  This is, of course,
+  done by inspecting individual bits.
+
+  Using these rules, each node's left subtree contains all smaller
+  sizes than its right subtree.  However, the node at the root of each
+  subtree has no particular ordering relationship to either.  (The
+  dividing line between the subtree sizes is based on trie relation.)
+  If we remove the last chunk of a given size from the interior of the
+  tree, we need to replace it with a leaf node.  The tree ordering
+  rules permit a node to be replaced by any leaf below it.
+
+  The smallest chunk in a tree (a common operation in a best-fit
+  allocator) can be found by walking a path to the leftmost leaf in
+  the tree.  Unlike a usual binary tree, where we follow left child
+  pointers until we reach a null, here we follow the right child
+  pointer any time the left one is null, until we reach a leaf with
+  both child pointers null. The smallest chunk in the tree will be
+  somewhere along that path.
+
+  The worst case number of steps to add, find, or remove a node is
+  bounded by the number of bits differentiating chunks within
+  bins. Under current bin calculations, this ranges from 6 up to 21
+  (for 32 bit sizes) or up to 53 (for 64 bit sizes). The typical case
+  is of course much better.
+*/
+
+struct malloc_tree_chunk {
+  /* The first four fields must be compatible with malloc_chunk */
+  size_t                    prev_foot;
+  size_t                    head;
+  struct malloc_tree_chunk* fd;
+  struct malloc_tree_chunk* bk;
+
+  struct malloc_tree_chunk* child[2];
+  struct malloc_tree_chunk* parent;
+  bindex_t                  index;
+};
+
+typedef struct malloc_tree_chunk  tchunk;
+typedef struct malloc_tree_chunk* tchunkptr;
+typedef struct malloc_tree_chunk* tbinptr; /* The type of bins of trees */
+
+/* A little helper macro for trees */
+#define leftmost_child(t) ((t)->child[0] != 0? (t)->child[0] : (t)->child[1])
+
+/* ----------------------------- Segments -------------------------------- */
+
+/*
+  Each malloc space may include non-contiguous segments, held in a
+  list headed by an embedded malloc_segment record representing the
+  top-most space. Segments also include flags holding properties of
+  the space. Large chunks that are directly allocated by mmap are not
+  included in this list. They are instead independently created and
+  destroyed without otherwise keeping track of them.
+
+  Segment management mainly comes into play for spaces allocated by
+  MMAP.  Any call to MMAP might or might not return memory that is
+  adjacent to an existing segment.  MORECORE normally contiguously
+  extends the current space, so this space is almost always adjacent,
+  which is simpler and faster to deal with. (This is why MORECORE is
+  used preferentially to MMAP when both are available -- see
+  sys_alloc.)  When allocating using MMAP, we don't use any of the
+  hinting mechanisms (inconsistently) supported in various
+  implementations of unix mmap, or distinguish reserving from
+  committing memory. Instead, we just ask for space, and exploit
+  contiguity when we get it.  It is probably possible to do
+  better than this on some systems, but no general scheme seems
+  to be significantly better.
+
+  Management entails a simpler variant of the consolidation scheme
+  used for chunks to reduce fragmentation -- new adjacent memory is
+  normally prepended or appended to an existing segment. However,
+  there are limitations compared to chunk consolidation that mostly
+  reflect the fact that segment processing is relatively infrequent
+  (occurring only when getting memory from system) and that we
+  don't expect to have huge numbers of segments:
+
+  * Segments are not indexed, so traversal requires linear scans.  (It
+    would be possible to index these, but is not worth the extra
+    overhead and complexity for most programs on most platforms.)
+  * New segments are only appended to old ones when holding top-most
+    memory; if they cannot be prepended to others, they are held in
+    different segments.
+
+  Except for the top-most segment of an mstate, each segment record
+  is kept at the tail of its segment. Segments are added by pushing
+  segment records onto the list headed by &mstate.seg for the
+  containing mstate.
+
+  Segment flags control allocation/merge/deallocation policies:
+  * If EXTERN_BIT set, then we did not allocate this segment,
+    and so should not try to deallocate or merge with others.
+    (This currently holds only for the initial segment passed
+    into create_mspace_with_base.)
+  * If IS_MMAPPED_BIT set, the segment may be merged with
+    other surrounding mmapped segments and trimmed/de-allocated
+    using munmap.
+  * If neither bit is set, then the segment was obtained using
+    MORECORE so can be merged with surrounding MORECORE'd segments
+    and deallocated/trimmed using MORECORE with negative arguments.
+*/
+
+struct malloc_segment {
+  char*        base;             /* base address */
+  size_t       size;             /* allocated size */
+  struct malloc_segment* next;   /* ptr to next segment */
+  flag_t       sflags;           /* mmap and extern flag */
+};
+
+#define is_mmapped_segment(S)  ((S)->sflags & IS_MMAPPED_BIT)
+#define is_extern_segment(S)   ((S)->sflags & EXTERN_BIT)
+
+typedef struct malloc_segment  msegment;
+typedef struct malloc_segment* msegmentptr;
+
+/* ---------------------------- malloc_state ----------------------------- */
+
+/*
+   A malloc_state holds all of the bookkeeping for a space.
+   The main fields are:
+
+  Top
+    The topmost chunk of the currently active segment. Its size is
+    cached in topsize.  The actual size of topmost space is
+    topsize+TOP_FOOT_SIZE, which includes space reserved for adding
+    fenceposts and segment records if necessary when getting more
+    space from the system.  The size at which to autotrim top is
+    cached from mparams in trim_check, except that it is disabled if
+    an autotrim fails.
+
+  Designated victim (dv)
+    This is the preferred chunk for servicing small requests that
+    don't have exact fits.  It is normally the chunk split off most
+    recently to service another small request.  Its size is cached in
+    dvsize. The link fields of this chunk are not maintained since it
+    is not kept in a bin.
+
+  SmallBins
+    An array of bin headers for free chunks.  These bins hold chunks
+    with sizes less than MIN_LARGE_SIZE bytes. Each bin contains
+    chunks of all the same size, spaced 8 bytes apart.  To simplify
+    use in double-linked lists, each bin header acts as a malloc_chunk
+    pointing to the real first node, if it exists (else pointing to
+    itself).  This avoids special-casing for headers.  But to avoid
+    waste, we allocate only the fd/bk pointers of bins, and then use
+    repositioning tricks to treat these as the fields of a chunk.
+
+  TreeBins
+    Treebins are pointers to the roots of trees holding a range of
+    sizes. There are 2 equally spaced treebins for each power of two
+    from TREE_SHIFT to TREE_SHIFT+16. The last bin holds anything
+    larger.
+
+  Bin maps
+    There is one bit map for small bins ("smallmap") and one for
+    treebins ("treemap).  Each bin sets its bit when non-empty, and
+    clears the bit when empty.  Bit operations are then used to avoid
+    bin-by-bin searching -- nearly all "search" is done without ever
+    looking at bins that won't be selected.  The bit maps
+    conservatively use 32 bits per map word, even if on 64bit system.
+    For a good description of some of the bit-based techniques used
+    here, see Henry S. Warren Jr's book "Hacker's Delight" (and
+    supplement at http://hackersdelight.org/). Many of these are
+    intended to reduce the branchiness of paths through malloc etc, as
+    well as to reduce the number of memory locations read or written.
+
+  Segments
+    A list of segments headed by an embedded malloc_segment record
+    representing the initial space.
+
+  Address check support
+    The least_addr field is the least address ever obtained from
+    MORECORE or MMAP. Attempted frees and reallocs of any address less
+    than this are trapped (unless INSECURE is defined).
+
+  Magic tag
+    A cross-check field that should always hold same value as mparams.magic.
+
+  Flags
+    Bits recording whether to use MMAP, locks, or contiguous MORECORE
+
+  Statistics
+    Each space keeps track of current and maximum system memory
+    obtained via MORECORE or MMAP.
+
+  Trim support
+    Fields holding the amount of unused topmost memory that should trigger
+    timming, and a counter to force periodic scanning to release unused
+    non-topmost segments.
+
+  Locking
+    If USE_LOCKS is defined, the "mutex" lock is acquired and released
+    around every public call using this mspace.
+
+  Extension support
+    A void* pointer and a size_t field that can be used to help implement
+    extensions to this malloc.
+*/
+
+/* Bin types, widths and sizes */
+#define NSMALLBINS        (32U)
+#define NTREEBINS         (32U)
+#define SMALLBIN_SHIFT    (3U)
+#define SMALLBIN_WIDTH    (SIZE_T_ONE << SMALLBIN_SHIFT)
+#define TREEBIN_SHIFT     (8U)
+#define MIN_LARGE_SIZE    (SIZE_T_ONE << TREEBIN_SHIFT)
+#define MAX_SMALL_SIZE    (MIN_LARGE_SIZE - SIZE_T_ONE)
+#define MAX_SMALL_REQUEST (MAX_SMALL_SIZE - CHUNK_ALIGN_MASK - CHUNK_OVERHEAD)
+
+struct malloc_state {
+  binmap_t   smallmap;
+  binmap_t   treemap;
+  size_t     dvsize;
+  size_t     topsize;
+  char*      least_addr;
+  mchunkptr  dv;
+  mchunkptr  top;
+  size_t     trim_check;
+  size_t     release_checks;
+  size_t     magic;
+  mchunkptr  smallbins[(NSMALLBINS+1)*2];
+  tbinptr    treebins[NTREEBINS];
+  size_t     footprint;
+  size_t     max_footprint;
+  flag_t     mflags;
+#if USE_LOCKS
+  MLOCK_T    mutex;     /* locate lock among fields that rarely change */
+#endif /* USE_LOCKS */
+  msegment   seg;
+  void*      extp;      /* Unused but available for extensions */
+  size_t     exts;
+};
+
+typedef struct malloc_state*    mstate;
+
+/* ------------- Global malloc_state and malloc_params ------------------- */
+
+/*
+  malloc_params holds global properties, including those that can be
+  dynamically set using mallopt. There is a single instance, mparams,
+  initialized in init_mparams. Note that the non-zeroness of "magic"
+  also serves as an initialization flag.
+*/
+
+struct malloc_params {
+  volatile size_t magic;
+  size_t page_size;
+  size_t granularity;
+  size_t mmap_threshold;
+  size_t trim_threshold;
+  flag_t default_mflags;
+};
+
+static struct malloc_params mparams;
+
+/* Ensure mparams initialized */
+#define ensure_initialization() ((void)(mparams.magic != 0 || init_mparams()))
+
+#if !ONLY_MSPACES
+
+/* The global malloc_state used for all non-"mspace" calls */
+static struct malloc_state _gm_;
+#define gm                 (&_gm_)
+#define is_global(M)       ((M) == &_gm_)
+
+#endif /* !ONLY_MSPACES */
+
+#define is_initialized(M)  ((M)->top != 0)
+
+/* -------------------------- system alloc setup ------------------------- */
+
+/* Operations on mflags */
+
+#define use_lock(M)           ((M)->mflags &   USE_LOCK_BIT)
+#define enable_lock(M)        ((M)->mflags |=  USE_LOCK_BIT)
+#define disable_lock(M)       ((M)->mflags &= ~USE_LOCK_BIT)
+
+#define use_mmap(M)           ((M)->mflags &   USE_MMAP_BIT)
+#define enable_mmap(M)        ((M)->mflags |=  USE_MMAP_BIT)
+#define disable_mmap(M)       ((M)->mflags &= ~USE_MMAP_BIT)
+
+#define use_noncontiguous(M)  ((M)->mflags &   USE_NONCONTIGUOUS_BIT)
+#define disable_contiguous(M) ((M)->mflags |=  USE_NONCONTIGUOUS_BIT)
+
+#define set_lock(M,L)\
+ ((M)->mflags = (L)?\
+  ((M)->mflags | USE_LOCK_BIT) :\
+  ((M)->mflags & ~USE_LOCK_BIT))
+
+/* page-align a size */
+#define page_align(S)\
+ (((S) + (mparams.page_size - SIZE_T_ONE)) & ~(mparams.page_size - SIZE_T_ONE))
+
+/* granularity-align a size */
+#define granularity_align(S)\
+  (((S) + (mparams.granularity - SIZE_T_ONE))\
+   & ~(mparams.granularity - SIZE_T_ONE))
+
+
+/* For mmap, use granularity alignment on windows, else page-align */
+#ifdef WIN32
+#define mmap_align(S) granularity_align(S)
+#else
+#define mmap_align(S) page_align(S)
+#endif
+
+/* For sys_alloc, enough padding to ensure can malloc request on success */
+#define SYS_ALLOC_PADDING (TOP_FOOT_SIZE + MALLOC_ALIGNMENT)
+
+#define is_page_aligned(S)\
+   (((size_t)(S) & (mparams.page_size - SIZE_T_ONE)) == 0)
+#define is_granularity_aligned(S)\
+   (((size_t)(S) & (mparams.granularity - SIZE_T_ONE)) == 0)
+
+/*  True if segment S holds address A */
+#define segment_holds(S, A)\
+  ((char*)(A) >= S->base && (char*)(A) < S->base + S->size)
+
+/* Return segment holding given address */
+static msegmentptr segment_holding(mstate m, char* addr) {
+  msegmentptr sp = &m->seg;
+  for (;;) {
+    if (addr >= sp->base && addr < sp->base + sp->size)
+      return sp;
+    if ((sp = sp->next) == 0)
+      return 0;
+  }
+}
+
+/* Return true if segment contains a segment link */
+static int has_segment_link(mstate m, msegmentptr ss) {
+  msegmentptr sp = &m->seg;
+  for (;;) {
+    if ((char*)sp >= ss->base && (char*)sp < ss->base + ss->size)
+      return 1;
+    if ((sp = sp->next) == 0)
+      return 0;
+  }
+}
+
+#ifndef MORECORE_CANNOT_TRIM
+#define should_trim(M,s)  ((s) > (M)->trim_check)
+#else  /* MORECORE_CANNOT_TRIM */
+#define should_trim(M,s)  (0)
+#endif /* MORECORE_CANNOT_TRIM */
+
+/*
+  TOP_FOOT_SIZE is padding at the end of a segment, including space
+  that may be needed to place segment records and fenceposts when new
+  noncontiguous segments are added.
+*/
+#define TOP_FOOT_SIZE\
+  (align_offset(chunk2mem(0))+pad_request(sizeof(struct malloc_segment))+MIN_CHUNK_SIZE)
+
+
+/* -------------------------------  Hooks -------------------------------- */
+
+/*
+  PREACTION should be defined to return 0 on success, and nonzero on
+  failure. If you are not using locking, you can redefine these to do
+  anything you like.
+*/
+
+#if USE_LOCKS
+
+#define PREACTION(M)  ((use_lock(M))? ACQUIRE_LOCK(&(M)->mutex) : 0)
+#define POSTACTION(M) { if (use_lock(M)) RELEASE_LOCK(&(M)->mutex); }
+#else /* USE_LOCKS */
+
+#ifndef PREACTION
+#define PREACTION(M) (0)
+#endif  /* PREACTION */
+
+#ifndef POSTACTION
+#define POSTACTION(M)
+#endif  /* POSTACTION */
+
+#endif /* USE_LOCKS */
+
+/*
+  CORRUPTION_ERROR_ACTION is triggered upon detected bad addresses.
+  USAGE_ERROR_ACTION is triggered on detected bad frees and
+  reallocs. The argument p is an address that might have triggered the
+  fault. It is ignored by the two predefined actions, but might be
+  useful in custom actions that try to help diagnose errors.
+*/
+
+#if PROCEED_ON_ERROR
+
+/* A count of the number of corruption errors causing resets */
+int malloc_corruption_error_count;
+
+/* default corruption action */
+static void reset_on_error(mstate m);
+
+#define CORRUPTION_ERROR_ACTION(m)  reset_on_error(m)
+#define USAGE_ERROR_ACTION(m, p)
+
+#else /* PROCEED_ON_ERROR */
+
+#ifndef CORRUPTION_ERROR_ACTION
+#define CORRUPTION_ERROR_ACTION(m) ABORT
+#endif /* CORRUPTION_ERROR_ACTION */
+
+#ifndef USAGE_ERROR_ACTION
+#define USAGE_ERROR_ACTION(m,p) ABORT
+#endif /* USAGE_ERROR_ACTION */
+
+#endif /* PROCEED_ON_ERROR */
+
+/* -------------------------- Debugging setup ---------------------------- */
+
+#if ! DEBUG
+
+#define check_free_chunk(M,P)
+#define check_inuse_chunk(M,P)
+#define check_malloced_chunk(M,P,N)
+#define check_mmapped_chunk(M,P)
+#define check_malloc_state(M)
+#define check_top_chunk(M,P)
+
+#else /* DEBUG */
+#define check_free_chunk(M,P)       do_check_free_chunk(M,P)
+#define check_inuse_chunk(M,P)      do_check_inuse_chunk(M,P)
+#define check_top_chunk(M,P)        do_check_top_chunk(M,P)
+#define check_malloced_chunk(M,P,N) do_check_malloced_chunk(M,P,N)
+#define check_mmapped_chunk(M,P)    do_check_mmapped_chunk(M,P)
+#define check_malloc_state(M)       do_check_malloc_state(M)
+
+static void   do_check_any_chunk(mstate m, mchunkptr p);
+static void   do_check_top_chunk(mstate m, mchunkptr p);
+static void   do_check_mmapped_chunk(mstate m, mchunkptr p);
+static void   do_check_inuse_chunk(mstate m, mchunkptr p);
+static void   do_check_free_chunk(mstate m, mchunkptr p);
+static void   do_check_malloced_chunk(mstate m, void* mem, size_t s);
+static void   do_check_tree(mstate m, tchunkptr t);
+static void   do_check_treebin(mstate m, bindex_t i);
+static void   do_check_smallbin(mstate m, bindex_t i);
+static void   do_check_malloc_state(mstate m);
+static int    bin_find(mstate m, mchunkptr x);
+static size_t traverse_and_check(mstate m);
+#endif /* DEBUG */
+
+/* ---------------------------- Indexing Bins ---------------------------- */
+
+#define is_small(s)         (((s) >> SMALLBIN_SHIFT) < NSMALLBINS)
+#define small_index(s)      ((s)  >> SMALLBIN_SHIFT)
+#define small_index2size(i) ((i)  << SMALLBIN_SHIFT)
+#define MIN_SMALL_INDEX     (small_index(MIN_CHUNK_SIZE))
+
+/* addressing by index. See above about smallbin repositioning */
+#define smallbin_at(M, i)   ((sbinptr)((char*)&((M)->smallbins[(i)<<1])))
+#define treebin_at(M,i)     (&((M)->treebins[i]))
+
+/* assign tree index for size S to variable I. Use x86 asm if possible  */
+#if defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
+#define compute_tree_index(S, I)\
+{\
+  unsigned int X = S >> TREEBIN_SHIFT;\
+  if (X == 0)\
+    I = 0;\
+  else if (X > 0xFFFF)\
+    I = NTREEBINS-1;\
+  else {\
+    unsigned int K;\
+    __asm__("bsrl\t%1, %0\n\t" : "=r" (K) : "rm"  (X));\
+    I =  (bindex_t)((K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1)));\
+  }\
+}
+
+#elif defined (__INTEL_COMPILER)
+#define compute_tree_index(S, I)\
+{\
+  size_t X = S >> TREEBIN_SHIFT;\
+  if (X == 0)\
+    I = 0;\
+  else if (X > 0xFFFF)\
+    I = NTREEBINS-1;\
+  else {\
+    unsigned int K = _bit_scan_reverse (X); \
+    I =  (bindex_t)((K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1)));\
+  }\
+}
+
+#elif defined(_MSC_VER) && _MSC_VER>=1300
+#define compute_tree_index(S, I)\
+{\
+  size_t X = S >> TREEBIN_SHIFT;\
+  if (X == 0)\
+    I = 0;\
+  else if (X > 0xFFFF)\
+    I = NTREEBINS-1;\
+  else {\
+    unsigned int K;\
+    _BitScanReverse((DWORD *) &K, X);\
+    I =  (bindex_t)((K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1)));\
+  }\
+}
+
+#else /* GNUC */
+#define compute_tree_index(S, I)\
+{\
+  size_t X = S >> TREEBIN_SHIFT;\
+  if (X == 0)\
+    I = 0;\
+  else if (X > 0xFFFF)\
+    I = NTREEBINS-1;\
+  else {\
+    unsigned int Y = (unsigned int)X;\
+    unsigned int N = ((Y - 0x100) >> 16) & 8;\
+    unsigned int K = (((Y <<= N) - 0x1000) >> 16) & 4;\
+    N += K;\
+    N += K = (((Y <<= K) - 0x4000) >> 16) & 2;\
+    K = 14 - N + ((Y <<= K) >> 15);\
+    I = (K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1));\
+  }\
+}
+#endif /* GNUC */
+
+/* Bit representing maximum resolved size in a treebin at i */
+#define bit_for_tree_index(i) \
+   (i == NTREEBINS-1)? (SIZE_T_BITSIZE-1) : (((i) >> 1) + TREEBIN_SHIFT - 2)
+
+/* Shift placing maximum resolved bit in a treebin at i as sign bit */
+#define leftshift_for_tree_index(i) \
+   ((i == NTREEBINS-1)? 0 : \
+    ((SIZE_T_BITSIZE-SIZE_T_ONE) - (((i) >> 1) + TREEBIN_SHIFT - 2)))
+
+/* The size of the smallest chunk held in bin with index i */
+#define minsize_for_tree_index(i) \
+   ((SIZE_T_ONE << (((i) >> 1) + TREEBIN_SHIFT)) |  \
+   (((size_t)((i) & SIZE_T_ONE)) << (((i) >> 1) + TREEBIN_SHIFT - 1)))
+
+
+/* ------------------------ Operations on bin maps ----------------------- */
+
+/* bit corresponding to given index */
+#define idx2bit(i)              ((binmap_t)(1) << (i))
+
+/* Mark/Clear bits with given index */
+#define mark_smallmap(M,i)      ((M)->smallmap |=  idx2bit(i))
+#define clear_smallmap(M,i)     ((M)->smallmap &= ~idx2bit(i))
+#define smallmap_is_marked(M,i) ((M)->smallmap &   idx2bit(i))
+
+#define mark_treemap(M,i)       ((M)->treemap  |=  idx2bit(i))
+#define clear_treemap(M,i)      ((M)->treemap  &= ~idx2bit(i))
+#define treemap_is_marked(M,i)  ((M)->treemap  &   idx2bit(i))
+
+/* isolate the least set bit of a bitmap */
+#define least_bit(x)         ((x) & -(x))
+
+/* mask with all bits to left of least bit of x on */
+#define left_bits(x)         ((x<<1) | -(x<<1))
+
+/* mask with all bits to left of or equal to least bit of x on */
+#define same_or_left_bits(x) ((x) | -(x))
+
+/* index corresponding to given bit. Use x86 asm if possible */
+
+#if defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
+#define compute_bit2idx(X, I)\
+{\
+  unsigned int J;\
+  __asm__("bsfl\t%1, %0\n\t" : "=r" (J) : "rm" (X));\
+  I = (bindex_t)J;\
+}
+
+#elif defined (__INTEL_COMPILER)
+#define compute_bit2idx(X, I)\
+{\
+  unsigned int J;\
+  J = _bit_scan_forward (X); \
+  I = (bindex_t)J;\
+}
+
+#elif defined(_MSC_VER) && _MSC_VER>=1300
+#define compute_bit2idx(X, I)\
+{\
+  unsigned int J;\
+  _BitScanForward((DWORD *) &J, X);\
+  I = (bindex_t)J;\
+}
+
+#elif USE_BUILTIN_FFS
+#define compute_bit2idx(X, I) I = ffs(X)-1
+
+#else
+#define compute_bit2idx(X, I)\
+{\
+  unsigned int Y = X - 1;\
+  unsigned int K = Y >> (16-4) & 16;\
+  unsigned int N = K;        Y >>= K;\
+  N += K = Y >> (8-3) &  8;  Y >>= K;\
+  N += K = Y >> (4-2) &  4;  Y >>= K;\
+  N += K = Y >> (2-1) &  2;  Y >>= K;\
+  N += K = Y >> (1-0) &  1;  Y >>= K;\
+  I = (bindex_t)(N + Y);\
+}
+#endif /* GNUC */
+
+
+/* ----------------------- Runtime Check Support ------------------------- */
+
+/*
+  For security, the main invariant is that malloc/free/etc never
+  writes to a static address other than malloc_state, unless static
+  malloc_state itself has been corrupted, which cannot occur via
+  malloc (because of these checks). In essence this means that we
+  believe all pointers, sizes, maps etc held in malloc_state, but
+  check all of those linked or offsetted from other embedded data
+  structures.  These checks are interspersed with main code in a way
+  that tends to minimize their run-time cost.
+
+  When FOOTERS is defined, in addition to range checking, we also
+  verify footer fields of inuse chunks, which can be used guarantee
+  that the mstate controlling malloc/free is intact.  This is a
+  streamlined version of the approach described by William Robertson
+  et al in "Run-time Detection of Heap-based Overflows" LISA'03
+  http://www.usenix.org/events/lisa03/tech/robertson.html The footer
+  of an inuse chunk holds the xor of its mstate and a random seed,
+  that is checked upon calls to free() and realloc().  This is
+  (probablistically) unguessable from outside the program, but can be
+  computed by any code successfully malloc'ing any chunk, so does not
+  itself provide protection against code that has already broken
+  security through some other means.  Unlike Robertson et al, we
+  always dynamically check addresses of all offset chunks (previous,
+  next, etc). This turns out to be cheaper than relying on hashes.
+*/
+
+#if !INSECURE
+/* Check if address a is at least as high as any from MORECORE or MMAP */
+#define ok_address(M, a) ((char*)(a) >= (M)->least_addr)
+/* Check if address of next chunk n is higher than base chunk p */
+#define ok_next(p, n)    ((char*)(p) < (char*)(n))
+/* Check if p has its cinuse bit on */
+#define ok_cinuse(p)     cinuse(p)
+/* Check if p has its pinuse bit on */
+#define ok_pinuse(p)     pinuse(p)
+
+#else /* !INSECURE */
+#define ok_address(M, a) (1)
+#define ok_next(b, n)    (1)
+#define ok_cinuse(p)     (1)
+#define ok_pinuse(p)     (1)
+#endif /* !INSECURE */
+
+#if (FOOTERS && !INSECURE)
+/* Check if (alleged) mstate m has expected magic field */
+#define ok_magic(M)      ((M)->magic == mparams.magic)
+#else  /* (FOOTERS && !INSECURE) */
+#define ok_magic(M)      (1)
+#endif /* (FOOTERS && !INSECURE) */
+
+
+/* In gcc, use __builtin_expect to minimize impact of checks */
+#if !INSECURE
+#if defined(__GNUC__) && __GNUC__ >= 3
+#define RTCHECK(e)  __builtin_expect(e, 1)
+#else /* GNUC */
+#define RTCHECK(e)  (e)
+#endif /* GNUC */
+#else /* !INSECURE */
+#define RTCHECK(e)  (1)
+#endif /* !INSECURE */
+
+/* macros to set up inuse chunks with or without footers */
+
+#if !FOOTERS
+
+#define mark_inuse_foot(M,p,s)
+
+/* Set cinuse bit and pinuse bit of next chunk */
+#define set_inuse(M,p,s)\
+  ((p)->head = (((p)->head & PINUSE_BIT)|s|CINUSE_BIT),\
+  ((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT)
+
+/* Set cinuse and pinuse of this chunk and pinuse of next chunk */
+#define set_inuse_and_pinuse(M,p,s)\
+  ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
+  ((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT)
+
+/* Set size, cinuse and pinuse bit of this chunk */
+#define set_size_and_pinuse_of_inuse_chunk(M, p, s)\
+  ((p)->head = (s|PINUSE_BIT|CINUSE_BIT))
+
+#else /* FOOTERS */
+
+/* Set foot of inuse chunk to be xor of mstate and seed */
+#define mark_inuse_foot(M,p,s)\
+  (((mchunkptr)((char*)(p) + (s)))->prev_foot = ((size_t)(M) ^ mparams.magic))
+
+#define get_mstate_for(p)\
+  ((mstate)(((mchunkptr)((char*)(p) +\
+    (chunksize(p))))->prev_foot ^ mparams.magic))
+
+#define set_inuse(M,p,s)\
+  ((p)->head = (((p)->head & PINUSE_BIT)|s|CINUSE_BIT),\
+  (((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT), \
+  mark_inuse_foot(M,p,s))
+
+#define set_inuse_and_pinuse(M,p,s)\
+  ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
+  (((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT),\
+ mark_inuse_foot(M,p,s))
+
+#define set_size_and_pinuse_of_inuse_chunk(M, p, s)\
+  ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
+  mark_inuse_foot(M, p, s))
+
+#endif /* !FOOTERS */
+
+/* ---------------------------- setting mparams -------------------------- */
+
+/* Initialize mparams */
+static int init_mparams(void) {
+#ifdef NEED_GLOBAL_LOCK_INIT
+  if (malloc_global_mutex_status <= 0)
+    init_malloc_global_mutex();
+#endif
+
+  ACQUIRE_MALLOC_GLOBAL_LOCK();
+  if (mparams.magic == 0) {
+    size_t magic;
+    size_t psize;
+    size_t gsize;
+
+#ifndef WIN32
+    psize = malloc_getpagesize;
+    gsize = ((DEFAULT_GRANULARITY != 0)? DEFAULT_GRANULARITY : psize);
+#else /* WIN32 */
+    {
+      SYSTEM_INFO system_info;
+      GetSystemInfo(&system_info);
+      psize = system_info.dwPageSize;
+      gsize = ((DEFAULT_GRANULARITY != 0)?
+	       DEFAULT_GRANULARITY : system_info.dwAllocationGranularity);
+    }
+#endif /* WIN32 */
+
+    /* Sanity-check configuration:
+       size_t must be unsigned and as wide as pointer type.
+       ints must be at least 4 bytes.
+       alignment must be at least 8.
+       Alignment, min chunk size, and page size must all be powers of 2.
+    */
+    if ((sizeof(size_t) != sizeof(char*)) ||
+	(MAX_SIZE_T < MIN_CHUNK_SIZE)  ||
+	(sizeof(int) < 4)  ||
+	(MALLOC_ALIGNMENT < (size_t)8U) ||
+	((MALLOC_ALIGNMENT & (MALLOC_ALIGNMENT-SIZE_T_ONE)) != 0) ||
+	((MCHUNK_SIZE      & (MCHUNK_SIZE-SIZE_T_ONE))      != 0) ||
+	((gsize            & (gsize-SIZE_T_ONE))            != 0) ||
+	((psize            & (psize-SIZE_T_ONE))            != 0))
+      ABORT;
+
+    mparams.granularity = gsize;
+    mparams.page_size = psize;
+    mparams.mmap_threshold = DEFAULT_MMAP_THRESHOLD;
+    mparams.trim_threshold = DEFAULT_TRIM_THRESHOLD;
+#if MORECORE_CONTIGUOUS
+    mparams.default_mflags = USE_LOCK_BIT|USE_MMAP_BIT;
+#else  /* MORECORE_CONTIGUOUS */
+    mparams.default_mflags = USE_LOCK_BIT|USE_MMAP_BIT|USE_NONCONTIGUOUS_BIT;
+#endif /* MORECORE_CONTIGUOUS */
+
+#if !ONLY_MSPACES
+    /* Set up lock for main malloc area */
+    gm->mflags = mparams.default_mflags;
+    INITIAL_LOCK(&gm->mutex);
+#endif
+
+#if (FOOTERS && !INSECURE)
+    {
+#if USE_DEV_RANDOM
+      int fd;
+      unsigned char buf[sizeof(size_t)];
+      /* Try to use /dev/urandom, else fall back on using time */
+      if ((fd = open("/dev/urandom", O_RDONLY)) >= 0 &&
+	  read(fd, buf, sizeof(buf)) == sizeof(buf)) {
+	magic = *((size_t *) buf);
+	close(fd);
+      }
+      else
+#endif /* USE_DEV_RANDOM */
+#ifdef WIN32
+	magic = (size_t)(GetTickCount() ^ (size_t)0x55555555U);
+#else
+      magic = (size_t)(time(0) ^ (size_t)0x55555555U);
+#endif
+      magic |= (size_t)8U;    /* ensure nonzero */
+      magic &= ~(size_t)7U;   /* improve chances of fault for bad values */
+    }
+#else /* (FOOTERS && !INSECURE) */
+    magic = (size_t)0x58585858U;
+#endif /* (FOOTERS && !INSECURE) */
+
+    mparams.magic = magic;
+  }
+
+  RELEASE_MALLOC_GLOBAL_LOCK();
+  return 1;
+}
+
+/* support for mallopt */
+static int change_mparam(int param_number, int value) {
+  size_t val = (value == -1)? MAX_SIZE_T : (size_t)value;
+  ensure_initialization();
+  switch(param_number) {
+  case M_TRIM_THRESHOLD:
+    mparams.trim_threshold = val;
+    return 1;
+  case M_GRANULARITY:
+    if (val >= mparams.page_size && ((val & (val-1)) == 0)) {
+      mparams.granularity = val;
+      return 1;
+    }
+    else
+      return 0;
+  case M_MMAP_THRESHOLD:
+    mparams.mmap_threshold = val;
+    return 1;
+  default:
+    return 0;
+  }
+}
+
+#if DEBUG
+/* ------------------------- Debugging Support --------------------------- */
+
+/* Check properties of any chunk, whether free, inuse, mmapped etc  */
+static void do_check_any_chunk(mstate m, mchunkptr p) {
+  assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
+  assert(ok_address(m, p));
+}
+
+/* Check properties of top chunk */
+static void do_check_top_chunk(mstate m, mchunkptr p) {
+  msegmentptr sp = segment_holding(m, (char*)p);
+  size_t  sz = p->head & ~INUSE_BITS; /* third-lowest bit can be set! */
+  assert(sp != 0);
+  assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
+  assert(ok_address(m, p));
+  assert(sz == m->topsize);
+  assert(sz > 0);
+  assert(sz == ((sp->base + sp->size) - (char*)p) - TOP_FOOT_SIZE);
+  assert(pinuse(p));
+  assert(!pinuse(chunk_plus_offset(p, sz)));
+}
+
+/* Check properties of (inuse) mmapped chunks */
+static void do_check_mmapped_chunk(mstate m, mchunkptr p) {
+  size_t  sz = chunksize(p);
+  size_t len = (sz + (p->prev_foot & ~IS_MMAPPED_BIT) + MMAP_FOOT_PAD);
+  assert(is_mmapped(p));
+  assert(use_mmap(m));
+  assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
+  assert(ok_address(m, p));
+  assert(!is_small(sz));
+  assert((len & (mparams.page_size-SIZE_T_ONE)) == 0);
+  assert(chunk_plus_offset(p, sz)->head == FENCEPOST_HEAD);
+  assert(chunk_plus_offset(p, sz+SIZE_T_SIZE)->head == 0);
+}
+
+/* Check properties of inuse chunks */
+static void do_check_inuse_chunk(mstate m, mchunkptr p) {
+  do_check_any_chunk(m, p);
+  assert(cinuse(p));
+  assert(next_pinuse(p));
+  /* If not pinuse and not mmapped, previous chunk has OK offset */
+  assert(is_mmapped(p) || pinuse(p) || next_chunk(prev_chunk(p)) == p);
+  if (is_mmapped(p))
+    do_check_mmapped_chunk(m, p);
+}
+
+/* Check properties of free chunks */
+static void do_check_free_chunk(mstate m, mchunkptr p) {
+  size_t sz = chunksize(p);
+  mchunkptr next = chunk_plus_offset(p, sz);
+  do_check_any_chunk(m, p);
+  assert(!cinuse(p));
+  assert(!next_pinuse(p));
+  assert (!is_mmapped(p));
+  if (p != m->dv && p != m->top) {
+    if (sz >= MIN_CHUNK_SIZE) {
+      assert((sz & CHUNK_ALIGN_MASK) == 0);
+      assert(is_aligned(chunk2mem(p)));
+      assert(next->prev_foot == sz);
+      assert(pinuse(p));
+      assert (next == m->top || cinuse(next));
+      assert(p->fd->bk == p);
+      assert(p->bk->fd == p);
+    }
+    else  /* markers are always of size SIZE_T_SIZE */
+      assert(sz == SIZE_T_SIZE);
+  }
+}
+
+/* Check properties of malloced chunks at the point they are malloced */
+static void do_check_malloced_chunk(mstate m, void* mem, size_t s) {
+  if (mem != 0) {
+    mchunkptr p = mem2chunk(mem);
+    size_t sz = p->head & ~(PINUSE_BIT|CINUSE_BIT);
+    do_check_inuse_chunk(m, p);
+    assert((sz & CHUNK_ALIGN_MASK) == 0);
+    assert(sz >= MIN_CHUNK_SIZE);
+    assert(sz >= s);
+    /* unless mmapped, size is less than MIN_CHUNK_SIZE more than request */
+    assert(is_mmapped(p) || sz < (s + MIN_CHUNK_SIZE));
+  }
+}
+
+/* Check a tree and its subtrees.  */
+static void do_check_tree(mstate m, tchunkptr t) {
+  tchunkptr head = 0;
+  tchunkptr u = t;
+  bindex_t tindex = t->index;
+  size_t tsize = chunksize(t);
+  bindex_t idx;
+  compute_tree_index(tsize, idx);
+  assert(tindex == idx);
+  assert(tsize >= MIN_LARGE_SIZE);
+  assert(tsize >= minsize_for_tree_index(idx));
+  assert((idx == NTREEBINS-1) || (tsize < minsize_for_tree_index((idx+1))));
+
+  do { /* traverse through chain of same-sized nodes */
+    do_check_any_chunk(m, ((mchunkptr)u));
+    assert(u->index == tindex);
+    assert(chunksize(u) == tsize);
+    assert(!cinuse(u));
+    assert(!next_pinuse(u));
+    assert(u->fd->bk == u);
+    assert(u->bk->fd == u);
+    if (u->parent == 0) {
+      assert(u->child[0] == 0);
+      assert(u->child[1] == 0);
+    }
+    else {
+      assert(head == 0); /* only one node on chain has parent */
+      head = u;
+      assert(u->parent != u);
+      assert (u->parent->child[0] == u ||
+	      u->parent->child[1] == u ||
+	      *((tbinptr*)(u->parent)) == u);
+      if (u->child[0] != 0) {
+	assert(u->child[0]->parent == u);
+	assert(u->child[0] != u);
+	do_check_tree(m, u->child[0]);
+      }
+      if (u->child[1] != 0) {
+	assert(u->child[1]->parent == u);
+	assert(u->child[1] != u);
+	do_check_tree(m, u->child[1]);
+      }
+      if (u->child[0] != 0 && u->child[1] != 0) {
+	assert(chunksize(u->child[0]) < chunksize(u->child[1]));
+      }
+    }
+    u = u->fd;
+  } while (u != t);
+  assert(head != 0);
+}
+
+/*  Check all the chunks in a treebin.  */
+static void do_check_treebin(mstate m, bindex_t i) {
+  tbinptr* tb = treebin_at(m, i);
+  tchunkptr t = *tb;
+  int empty = (m->treemap & (1U << i)) == 0;
+  if (t == 0)
+    assert(empty);
+  if (!empty)
+    do_check_tree(m, t);
+}
+
+/*  Check all the chunks in a smallbin.  */
+static void do_check_smallbin(mstate m, bindex_t i) {
+  sbinptr b = smallbin_at(m, i);
+  mchunkptr p = b->bk;
+  unsigned int empty = (m->smallmap & (1U << i)) == 0;
+  if (p == b)
+    assert(empty);
+  if (!empty) {
+    for (; p != b; p = p->bk) {
+      size_t size = chunksize(p);
+      mchunkptr q;
+      /* each chunk claims to be free */
+      do_check_free_chunk(m, p);
+      /* chunk belongs in bin */
+      assert(small_index(size) == i);
+      assert(p->bk == b || chunksize(p->bk) == chunksize(p));
+      /* chunk is followed by an inuse chunk */
+      q = next_chunk(p);
+      if (q->head != FENCEPOST_HEAD)
+	do_check_inuse_chunk(m, q);
+    }
+  }
+}
+
+/* Find x in a bin. Used in other check functions. */
+static int bin_find(mstate m, mchunkptr x) {
+  size_t size = chunksize(x);
+  if (is_small(size)) {
+    bindex_t sidx = small_index(size);
+    sbinptr b = smallbin_at(m, sidx);
+    if (smallmap_is_marked(m, sidx)) {
+      mchunkptr p = b;
+      do {
+	if (p == x)
+	  return 1;
+      } while ((p = p->fd) != b);
+    }
+  }
+  else {
+    bindex_t tidx;
+    compute_tree_index(size, tidx);
+    if (treemap_is_marked(m, tidx)) {
+      tchunkptr t = *treebin_at(m, tidx);
+      size_t sizebits = size << leftshift_for_tree_index(tidx);
+      while (t != 0 && chunksize(t) != size) {
+	t = t->child[(sizebits >> (SIZE_T_BITSIZE-SIZE_T_ONE)) & 1];
+	sizebits <<= 1;
+      }
+      if (t != 0) {
+	tchunkptr u = t;
+	do {
+	  if (u == (tchunkptr)x)
+	    return 1;
+	} while ((u = u->fd) != t);
+      }
+    }
+  }
+  return 0;
+}
+
+/* Traverse each chunk and check it; return total */
+static size_t traverse_and_check(mstate m) {
+  size_t sum = 0;
+  if (is_initialized(m)) {
+    msegmentptr s = &m->seg;
+    sum += m->topsize + TOP_FOOT_SIZE;
+    while (s != 0) {
+      mchunkptr q = align_as_chunk(s->base);
+      mchunkptr lastq = 0;
+      assert(pinuse(q));
+      while (segment_holds(s, q) &&
+	     q != m->top && q->head != FENCEPOST_HEAD) {
+	sum += chunksize(q);
+	if (cinuse(q)) {
+	  assert(!bin_find(m, q));
+	  do_check_inuse_chunk(m, q);
+	}
+	else {
+	  assert(q == m->dv || bin_find(m, q));
+	  assert(lastq == 0 || cinuse(lastq)); /* Not 2 consecutive free */
+	  do_check_free_chunk(m, q);
+	}
+	lastq = q;
+	q = next_chunk(q);
+      }
+      s = s->next;
+    }
+  }
+  return sum;
+}
+
+/* Check all properties of malloc_state. */
+static void do_check_malloc_state(mstate m) {
+  bindex_t i;
+  size_t total;
+  /* check bins */
+  for (i = 0; i < NSMALLBINS; ++i)
+    do_check_smallbin(m, i);
+  for (i = 0; i < NTREEBINS; ++i)
+    do_check_treebin(m, i);
+
+  if (m->dvsize != 0) { /* check dv chunk */
+    do_check_any_chunk(m, m->dv);
+    assert(m->dvsize == chunksize(m->dv));
+    assert(m->dvsize >= MIN_CHUNK_SIZE);
+    assert(bin_find(m, m->dv) == 0);
+  }
+
+  if (m->top != 0) {   /* check top chunk */
+    do_check_top_chunk(m, m->top);
+    /*assert(m->topsize == chunksize(m->top)); redundant */
+    assert(m->topsize > 0);
+    assert(bin_find(m, m->top) == 0);
+  }
+
+  total = traverse_and_check(m);
+  assert(total <= m->footprint);
+  assert(m->footprint <= m->max_footprint);
+}
+#endif /* DEBUG */
+
+/* ----------------------------- statistics ------------------------------ */
+
+#if !NO_MALLINFO
+static struct mallinfo internal_mallinfo(mstate m) {
+  struct mallinfo nm = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+  ensure_initialization();
+  if (!PREACTION(m)) {
+    check_malloc_state(m);
+    if (is_initialized(m)) {
+      size_t nfree = SIZE_T_ONE; /* top always free */
+      size_t mfree = m->topsize + TOP_FOOT_SIZE;
+      size_t sum = mfree;
+      msegmentptr s = &m->seg;
+      while (s != 0) {
+	mchunkptr q = align_as_chunk(s->base);
+	while (segment_holds(s, q) &&
+	       q != m->top && q->head != FENCEPOST_HEAD) {
+	  size_t sz = chunksize(q);
+	  sum += sz;
+	  if (!cinuse(q)) {
+	    mfree += sz;
+	    ++nfree;
+	  }
+	  q = next_chunk(q);
+	}
+	s = s->next;
+      }
+
+      nm.arena    = sum;
+      nm.ordblks  = nfree;
+      nm.hblkhd   = m->footprint - sum;
+      nm.usmblks  = m->max_footprint;
+      nm.uordblks = m->footprint - mfree;
+      nm.fordblks = mfree;
+      nm.keepcost = m->topsize;
+    }
+
+    POSTACTION(m);
+  }
+  return nm;
+}
+#endif /* !NO_MALLINFO */
+
+static void internal_malloc_stats(mstate m) {
+  ensure_initialization();
+  if (!PREACTION(m)) {
+    size_t maxfp = 0;
+    size_t fp = 0;
+    size_t used = 0;
+    check_malloc_state(m);
+    if (is_initialized(m)) {
+      msegmentptr s = &m->seg;
+      maxfp = m->max_footprint;
+      fp = m->footprint;
+      used = fp - (m->topsize + TOP_FOOT_SIZE);
+
+      while (s != 0) {
+	mchunkptr q = align_as_chunk(s->base);
+	while (segment_holds(s, q) &&
+	       q != m->top && q->head != FENCEPOST_HEAD) {
+	  if (!cinuse(q))
+	    used -= chunksize(q);
+	  q = next_chunk(q);
+	}
+	s = s->next;
+      }
+    }
+
+    fprintf(stderr, "max system bytes = %10lu\n", (unsigned long)(maxfp));
+    fprintf(stderr, "system bytes     = %10lu\n", (unsigned long)(fp));
+    fprintf(stderr, "in use bytes     = %10lu\n", (unsigned long)(used));
+
+    POSTACTION(m);
+  }
+}
+
+/* ----------------------- Operations on smallbins ----------------------- */
+
+/*
+  Various forms of linking and unlinking are defined as macros.  Even
+  the ones for trees, which are very long but have very short typical
+  paths.  This is ugly but reduces reliance on inlining support of
+  compilers.
+*/
+
+/* Link a free chunk into a smallbin  */
+#define insert_small_chunk(M, P, S) {\
+  bindex_t I  = small_index(S);\
+  mchunkptr B = smallbin_at(M, I);\
+  mchunkptr F = B;\
+  assert(S >= MIN_CHUNK_SIZE);\
+  if (!smallmap_is_marked(M, I))\
+    mark_smallmap(M, I);\
+  else if (RTCHECK(ok_address(M, B->fd)))\
+    F = B->fd;\
+  else {\
+    CORRUPTION_ERROR_ACTION(M);\
+  }\
+  B->fd = P;\
+  F->bk = P;\
+  P->fd = F;\
+  P->bk = B;\
+}
+
+/* Unlink a chunk from a smallbin  */
+#define unlink_small_chunk(M, P, S) {\
+  mchunkptr F = P->fd;\
+  mchunkptr B = P->bk;\
+  bindex_t I = small_index(S);\
+  assert(P != B);\
+  assert(P != F);\
+  assert(chunksize(P) == small_index2size(I));\
+  if (F == B)\
+    clear_smallmap(M, I);\
+  else if (RTCHECK((F == smallbin_at(M,I) || ok_address(M, F)) &&\
+		   (B == smallbin_at(M,I) || ok_address(M, B)))) {\
+    F->bk = B;\
+    B->fd = F;\
+  }\
+  else {\
+    CORRUPTION_ERROR_ACTION(M);\
+  }\
+}
+
+/* Unlink the first chunk from a smallbin */
+#define unlink_first_small_chunk(M, B, P, I) {\
+  mchunkptr F = P->fd;\
+  assert(P != B);\
+  assert(P != F);\
+  assert(chunksize(P) == small_index2size(I));\
+  if (B == F)\
+    clear_smallmap(M, I);\
+  else if (RTCHECK(ok_address(M, F))) {\
+    B->fd = F;\
+    F->bk = B;\
+  }\
+  else {\
+    CORRUPTION_ERROR_ACTION(M);\
+  }\
+}
+
+
+
+/* Replace dv node, binning the old one */
+/* Used only when dvsize known to be small */
+#define replace_dv(M, P, S) {\
+  size_t DVS = M->dvsize;\
+  if (DVS != 0) {\
+    mchunkptr DV = M->dv;\
+    assert(is_small(DVS));\
+    insert_small_chunk(M, DV, DVS);\
+  }\
+  M->dvsize = S;\
+  M->dv = P;\
+}
+
+/* ------------------------- Operations on trees ------------------------- */
+
+/* Insert chunk into tree */
+#define insert_large_chunk(M, X, S) {\
+  tbinptr* H;\
+  bindex_t I;\
+  compute_tree_index(S, I);\
+  H = treebin_at(M, I);\
+  X->index = I;\
+  X->child[0] = X->child[1] = 0;\
+  if (!treemap_is_marked(M, I)) {\
+    mark_treemap(M, I);\
+    *H = X;\
+    X->parent = (tchunkptr)H;\
+    X->fd = X->bk = X;\
+  }\
+  else {\
+    tchunkptr T = *H;\
+    size_t K = S << leftshift_for_tree_index(I);\
+    for (;;) {\
+      if (chunksize(T) != S) {\
+	tchunkptr* C = &(T->child[(K >> (SIZE_T_BITSIZE-SIZE_T_ONE)) & 1]);\
+	K <<= 1;\
+	if (*C != 0)\
+	  T = *C;\
+	else if (RTCHECK(ok_address(M, C))) {\
+	  *C = X;\
+	  X->parent = T;\
+	  X->fd = X->bk = X;\
+	  break;\
+	}\
+	else {\
+	  CORRUPTION_ERROR_ACTION(M);\
+	  break;\
+	}\
+      }\
+      else {\
+	tchunkptr F = T->fd;\
+	if (RTCHECK(ok_address(M, T) && ok_address(M, F))) {\
+	  T->fd = F->bk = X;\
+	  X->fd = F;\
+	  X->bk = T;\
+	  X->parent = 0;\
+	  break;\
+	}\
+	else {\
+	  CORRUPTION_ERROR_ACTION(M);\
+	  break;\
+	}\
+      }\
+    }\
+  }\
+}
+
+/*
+  Unlink steps:
+
+  1. If x is a chained node, unlink it from its same-sized fd/bk links
+     and choose its bk node as its replacement.
+  2. If x was the last node of its size, but not a leaf node, it must
+     be replaced with a leaf node (not merely one with an open left or
+     right), to make sure that lefts and rights of descendents
+     correspond properly to bit masks.  We use the rightmost descendent
+     of x.  We could use any other leaf, but this is easy to locate and
+     tends to counteract removal of leftmosts elsewhere, and so keeps
+     paths shorter than minimally guaranteed.  This doesn't loop much
+     because on average a node in a tree is near the bottom.
+  3. If x is the base of a chain (i.e., has parent links) relink
+     x's parent and children to x's replacement (or null if none).
+*/
+
+#define unlink_large_chunk(M, X) {\
+  tchunkptr XP = X->parent;\
+  tchunkptr R;\
+  if (X->bk != X) {\
+    tchunkptr F = X->fd;\
+    R = X->bk;\
+    if (RTCHECK(ok_address(M, F))) {\
+      F->bk = R;\
+      R->fd = F;\
+    }\
+    else {\
+      CORRUPTION_ERROR_ACTION(M);\
+    }\
+  }\
+  else {\
+    tchunkptr* RP;\
+    if (((R = *(RP = &(X->child[1]))) != 0) ||\
+	((R = *(RP = &(X->child[0]))) != 0)) {\
+      tchunkptr* CP;\
+      while ((*(CP = &(R->child[1])) != 0) ||\
+	     (*(CP = &(R->child[0])) != 0)) {\
+	R = *(RP = CP);\
+      }\
+      if (RTCHECK(ok_address(M, RP)))\
+	*RP = 0;\
+      else {\
+	CORRUPTION_ERROR_ACTION(M);\
+      }\
+    }\
+  }\
+  if (XP != 0) {\
+    tbinptr* H = treebin_at(M, X->index);\
+    if (X == *H) {\
+      if ((*H = R) == 0) \
+	clear_treemap(M, X->index);\
+    }\
+    else if (RTCHECK(ok_address(M, XP))) {\
+      if (XP->child[0] == X) \
+	XP->child[0] = R;\
+      else \
+	XP->child[1] = R;\
+    }\
+    else\
+      CORRUPTION_ERROR_ACTION(M);\
+    if (R != 0) {\
+      if (RTCHECK(ok_address(M, R))) {\
+	tchunkptr C0, C1;\
+	R->parent = XP;\
+	if ((C0 = X->child[0]) != 0) {\
+	  if (RTCHECK(ok_address(M, C0))) {\
+	    R->child[0] = C0;\
+	    C0->parent = R;\
+	  }\
+	  else\
+	    CORRUPTION_ERROR_ACTION(M);\
+	}\
+	if ((C1 = X->child[1]) != 0) {\
+	  if (RTCHECK(ok_address(M, C1))) {\
+	    R->child[1] = C1;\
+	    C1->parent = R;\
+	  }\
+	  else\
+	    CORRUPTION_ERROR_ACTION(M);\
+	}\
+      }\
+      else\
+	CORRUPTION_ERROR_ACTION(M);\
+    }\
+  }\
+}
+
+/* Relays to large vs small bin operations */
+
+#define insert_chunk(M, P, S)\
+  if (is_small(S)) insert_small_chunk(M, P, S)\
+  else { tchunkptr TP = (tchunkptr)(P); insert_large_chunk(M, TP, S); }
+
+#define unlink_chunk(M, P, S)\
+  if (is_small(S)) unlink_small_chunk(M, P, S)\
+  else { tchunkptr TP = (tchunkptr)(P); unlink_large_chunk(M, TP); }
+
+
+/* Relays to internal calls to malloc/free from realloc, memalign etc */
+
+#if ONLY_MSPACES
+#define internal_malloc(m, b) mspace_malloc(m, b)
+#define internal_free(m, mem) mspace_free(m,mem);
+#else /* ONLY_MSPACES */
+#if MSPACES
+#define internal_malloc(m, b)\
+   (m == gm)? dlmalloc(b) : mspace_malloc(m, b)
+#define internal_free(m, mem)\
+   if (m == gm) dlfree(mem); else mspace_free(m,mem);
+#else /* MSPACES */
+#define internal_malloc(m, b) dlmalloc(b)
+#define internal_free(m, mem) dlfree(mem)
+#endif /* MSPACES */
+#endif /* ONLY_MSPACES */
+
+/* -----------------------  Direct-mmapping chunks ----------------------- */
+
+/*
+  Directly mmapped chunks are set up with an offset to the start of
+  the mmapped region stored in the prev_foot field of the chunk. This
+  allows reconstruction of the required argument to MUNMAP when freed,
+  and also allows adjustment of the returned chunk to meet alignment
+  requirements (especially in memalign).  There is also enough space
+  allocated to hold a fake next chunk of size SIZE_T_SIZE to maintain
+  the PINUSE bit so frees can be checked.
+*/
+
+/* Malloc using mmap */
+static void* mmap_alloc(mstate m, size_t nb) {
+  size_t mmsize = mmap_align(nb + SIX_SIZE_T_SIZES + CHUNK_ALIGN_MASK);
+  if (mmsize > nb) {     /* Check for wrap around 0 */
+    char* mm = (char*)(CALL_DIRECT_MMAP(mmsize));
+    if (mm != CMFAIL) {
+      size_t offset = align_offset(chunk2mem(mm));
+      size_t psize = mmsize - offset - MMAP_FOOT_PAD;
+      mchunkptr p = (mchunkptr)(mm + offset);
+      p->prev_foot = offset | IS_MMAPPED_BIT;
+      (p)->head = (psize|CINUSE_BIT);
+      mark_inuse_foot(m, p, psize);
+      chunk_plus_offset(p, psize)->head = FENCEPOST_HEAD;
+      chunk_plus_offset(p, psize+SIZE_T_SIZE)->head = 0;
+
+      if (mm < m->least_addr)
+	m->least_addr = mm;
+      if ((m->footprint += mmsize) > m->max_footprint)
+	m->max_footprint = m->footprint;
+      assert(is_aligned(chunk2mem(p)));
+      check_mmapped_chunk(m, p);
+      return chunk2mem(p);
+    }
+  }
+  return 0;
+}
+
+/* Realloc using mmap */
+static mchunkptr mmap_resize(mstate m, mchunkptr oldp, size_t nb) {
+  size_t oldsize = chunksize(oldp);
+  if (is_small(nb)) /* Can't shrink mmap regions below small size */
+    return 0;
+  /* Keep old chunk if big enough but not too big */
+  if (oldsize >= nb + SIZE_T_SIZE &&
+      (oldsize - nb) <= (mparams.granularity << 1))
+    return oldp;
+  else {
+    size_t offset = oldp->prev_foot & ~IS_MMAPPED_BIT;
+    size_t oldmmsize = oldsize + offset + MMAP_FOOT_PAD;
+    size_t newmmsize = mmap_align(nb + SIX_SIZE_T_SIZES + CHUNK_ALIGN_MASK);
+    char* cp = (char*)CALL_MREMAP((char*)oldp - offset,
+				  oldmmsize, newmmsize, 1);
+    if (cp != CMFAIL) {
+      mchunkptr newp = (mchunkptr)(cp + offset);
+      size_t psize = newmmsize - offset - MMAP_FOOT_PAD;
+      newp->head = (psize|CINUSE_BIT);
+      mark_inuse_foot(m, newp, psize);
+      chunk_plus_offset(newp, psize)->head = FENCEPOST_HEAD;
+      chunk_plus_offset(newp, psize+SIZE_T_SIZE)->head = 0;
+
+      if (cp < m->least_addr)
+	m->least_addr = cp;
+      if ((m->footprint += newmmsize - oldmmsize) > m->max_footprint)
+	m->max_footprint = m->footprint;
+      check_mmapped_chunk(m, newp);
+      return newp;
+    }
+  }
+  return 0;
+}
+
+/* -------------------------- mspace management -------------------------- */
+
+/* Initialize top chunk and its size */
+static void init_top(mstate m, mchunkptr p, size_t psize) {
+  /* Ensure alignment */
+  size_t offset = align_offset(chunk2mem(p));
+  p = (mchunkptr)((char*)p + offset);
+  psize -= offset;
+
+  m->top = p;
+  m->topsize = psize;
+  p->head = psize | PINUSE_BIT;
+  /* set size of fake trailing chunk holding overhead space only once */
+  chunk_plus_offset(p, psize)->head = TOP_FOOT_SIZE;
+  m->trim_check = mparams.trim_threshold; /* reset on each update */
+}
+
+/* Initialize bins for a new mstate that is otherwise zeroed out */
+static void init_bins(mstate m) {
+  /* Establish circular links for smallbins */
+  bindex_t i;
+  for (i = 0; i < NSMALLBINS; ++i) {
+    sbinptr bin = smallbin_at(m,i);
+    bin->fd = bin->bk = bin;
+  }
+}
+
+#if PROCEED_ON_ERROR
+
+/* default corruption action */
+static void reset_on_error(mstate m) {
+  int i;
+  ++malloc_corruption_error_count;
+  /* Reinitialize fields to forget about all memory */
+  m->smallbins = m->treebins = 0;
+  m->dvsize = m->topsize = 0;
+  m->seg.base = 0;
+  m->seg.size = 0;
+  m->seg.next = 0;
+  m->top = m->dv = 0;
+  for (i = 0; i < NTREEBINS; ++i)
+    *treebin_at(m, i) = 0;
+  init_bins(m);
+}
+#endif /* PROCEED_ON_ERROR */
+
+/* Allocate chunk and prepend remainder with chunk in successor base. */
+static void* prepend_alloc(mstate m, char* newbase, char* oldbase,
+			   size_t nb) {
+  mchunkptr p = align_as_chunk(newbase);
+  mchunkptr oldfirst = align_as_chunk(oldbase);
+  size_t psize = (char*)oldfirst - (char*)p;
+  mchunkptr q = chunk_plus_offset(p, nb);
+  size_t qsize = psize - nb;
+  set_size_and_pinuse_of_inuse_chunk(m, p, nb);
+
+  assert((char*)oldfirst > (char*)q);
+  assert(pinuse(oldfirst));
+  assert(qsize >= MIN_CHUNK_SIZE);
+
+  /* consolidate remainder with first chunk of old base */
+  if (oldfirst == m->top) {
+    size_t tsize = m->topsize += qsize;
+    m->top = q;
+    q->head = tsize | PINUSE_BIT;
+    check_top_chunk(m, q);
+  }
+  else if (oldfirst == m->dv) {
+    size_t dsize = m->dvsize += qsize;
+    m->dv = q;
+    set_size_and_pinuse_of_free_chunk(q, dsize);
+  }
+  else {
+    if (!cinuse(oldfirst)) {
+      size_t nsize = chunksize(oldfirst);
+      unlink_chunk(m, oldfirst, nsize);
+      oldfirst = chunk_plus_offset(oldfirst, nsize);
+      qsize += nsize;
+    }
+    set_free_with_pinuse(q, qsize, oldfirst);
+    insert_chunk(m, q, qsize);
+    check_free_chunk(m, q);
+  }
+
+  check_malloced_chunk(m, chunk2mem(p), nb);
+  return chunk2mem(p);
+}
+
+/* Add a segment to hold a new noncontiguous region */
+static void add_segment(mstate m, char* tbase, size_t tsize, flag_t mmapped) {
+  /* Determine locations and sizes of segment, fenceposts, old top */
+  char* old_top = (char*)m->top;
+  msegmentptr oldsp = segment_holding(m, old_top);
+  char* old_end = oldsp->base + oldsp->size;
+  size_t ssize = pad_request(sizeof(struct malloc_segment));
+  char* rawsp = old_end - (ssize + FOUR_SIZE_T_SIZES + CHUNK_ALIGN_MASK);
+  size_t offset = align_offset(chunk2mem(rawsp));
+  char* asp = rawsp + offset;
+  char* csp = (asp < (old_top + MIN_CHUNK_SIZE))? old_top : asp;
+  mchunkptr sp = (mchunkptr)csp;
+  msegmentptr ss = (msegmentptr)(chunk2mem(sp));
+  mchunkptr tnext = chunk_plus_offset(sp, ssize);
+  mchunkptr p = tnext;
+  int nfences = 0;
+
+  /* reset top to new space */
+  init_top(m, (mchunkptr)tbase, tsize - TOP_FOOT_SIZE);
+
+  /* Set up segment record */
+  assert(is_aligned(ss));
+  set_size_and_pinuse_of_inuse_chunk(m, sp, ssize);
+  *ss = m->seg; /* Push current record */
+  m->seg.base = tbase;
+  m->seg.size = tsize;
+  m->seg.sflags = mmapped;
+  m->seg.next = ss;
+
+  /* Insert trailing fenceposts */
+  for (;;) {
+    mchunkptr nextp = chunk_plus_offset(p, SIZE_T_SIZE);
+    p->head = FENCEPOST_HEAD;
+    ++nfences;
+    if ((char*)(&(nextp->head)) < old_end)
+      p = nextp;
+    else
+      break;
+  }
+  assert(nfences >= 2);
+
+  /* Insert the rest of old top into a bin as an ordinary free chunk */
+  if (csp != old_top) {
+    mchunkptr q = (mchunkptr)old_top;
+    size_t psize = csp - old_top;
+    mchunkptr tn = chunk_plus_offset(q, psize);
+    set_free_with_pinuse(q, psize, tn);
+    insert_chunk(m, q, psize);
+  }
+
+  check_top_chunk(m, m->top);
+}
+
+/* -------------------------- System allocation -------------------------- */
+
+/* Get memory from system using MORECORE or MMAP */
+static void* sys_alloc(mstate m, size_t nb) {
+  char* tbase = CMFAIL;
+  size_t tsize = 0;
+  flag_t mmap_flag = 0;
+
+  ensure_initialization();
+
+  /* Directly map large chunks */
+  if (use_mmap(m) && nb >= mparams.mmap_threshold) {
+    void* mem = mmap_alloc(m, nb);
+    if (mem != 0)
+      return mem;
+  }
+
+  /*
+    Try getting memory in any of three ways (in most-preferred to
+    least-preferred order):
+    1. A call to MORECORE that can normally contiguously extend memory.
+       (disabled if not MORECORE_CONTIGUOUS or not HAVE_MORECORE or
+       or main space is mmapped or a previous contiguous call failed)
+    2. A call to MMAP new space (disabled if not HAVE_MMAP).
+       Note that under the default settings, if MORECORE is unable to
+       fulfill a request, and HAVE_MMAP is true, then mmap is
+       used as a noncontiguous system allocator. This is a useful backup
+       strategy for systems with holes in address spaces -- in this case
+       sbrk cannot contiguously expand the heap, but mmap may be able to
+       find space.
+    3. A call to MORECORE that cannot usually contiguously extend memory.
+       (disabled if not HAVE_MORECORE)
+
+   In all cases, we need to request enough bytes from system to ensure
+   we can malloc nb bytes upon success, so pad with enough space for
+   top_foot, plus alignment-pad to make sure we don't lose bytes if
+   not on boundary, and round this up to a granularity unit.
+  */
+
+  if (MORECORE_CONTIGUOUS && !use_noncontiguous(m)) {
+    char* br = CMFAIL;
+    msegmentptr ss = (m->top == 0)? 0 : segment_holding(m, (char*)m->top);
+    size_t asize = 0;
+    ACQUIRE_MALLOC_GLOBAL_LOCK();
+
+    if (ss == 0) {  /* First time through or recovery */
+      char* base = (char*)CALL_MORECORE(0);
+      if (base != CMFAIL) {
+	asize = granularity_align(nb + SYS_ALLOC_PADDING);
+	/* Adjust to end on a page boundary */
+	if (!is_page_aligned(base))
+	  asize += (page_align((size_t)base) - (size_t)base);
+	/* Can't call MORECORE if size is negative when treated as signed */
+	if (asize < HALF_MAX_SIZE_T &&
+	    (br = (char*)(CALL_MORECORE(asize))) == base) {
+	  tbase = base;
+	  tsize = asize;
+	}
+      }
+    }
+    else {
+      /* Subtract out existing available top space from MORECORE request. */
+      asize = granularity_align(nb - m->topsize + SYS_ALLOC_PADDING);
+      /* Use mem here only if it did continuously extend old space */
+      if (asize < HALF_MAX_SIZE_T &&
+	  (br = (char*)(CALL_MORECORE(asize))) == ss->base+ss->size) {
+	tbase = br;
+	tsize = asize;
+      }
+    }
+
+    if (tbase == CMFAIL) {    /* Cope with partial failure */
+      if (br != CMFAIL) {    /* Try to use/extend the space we did get */
+	if (asize < HALF_MAX_SIZE_T &&
+	    asize < nb + SYS_ALLOC_PADDING) {
+	  size_t esize = granularity_align(nb + SYS_ALLOC_PADDING - asize);
+	  if (esize < HALF_MAX_SIZE_T) {
+	    char* end = (char*)CALL_MORECORE(esize);
+	    if (end != CMFAIL)
+	      asize += esize;
+	    else {            /* Can't use; try to release */
+	      (void) CALL_MORECORE(-asize);
+	      br = CMFAIL;
+	    }
+	  }
+	}
+      }
+      if (br != CMFAIL) {    /* Use the space we did get */
+	tbase = br;
+	tsize = asize;
+      }
+      else
+	disable_contiguous(m); /* Don't try contiguous path in the future */
+    }
+
+    RELEASE_MALLOC_GLOBAL_LOCK();
+  }
+
+  if (HAVE_MMAP && tbase == CMFAIL) {  /* Try MMAP */
+    size_t rsize = granularity_align(nb + SYS_ALLOC_PADDING);
+    if (rsize > nb) { /* Fail if wraps around zero */
+      char* mp = (char*)(CALL_MMAP(rsize));
+      if (mp != CMFAIL) {
+	tbase = mp;
+	tsize = rsize;
+	mmap_flag = IS_MMAPPED_BIT;
+      }
+    }
+  }
+
+  if (HAVE_MORECORE && tbase == CMFAIL) { /* Try noncontiguous MORECORE */
+    size_t asize = granularity_align(nb + SYS_ALLOC_PADDING);
+    if (asize < HALF_MAX_SIZE_T) {
+      char* br = CMFAIL;
+      char* end = CMFAIL;
+      ACQUIRE_MALLOC_GLOBAL_LOCK();
+      br = (char*)(CALL_MORECORE(asize));
+      end = (char*)(CALL_MORECORE(0));
+      RELEASE_MALLOC_GLOBAL_LOCK();
+      if (br != CMFAIL && end != CMFAIL && br < end) {
+	size_t ssize = end - br;
+	if (ssize > nb + TOP_FOOT_SIZE) {
+	  tbase = br;
+	  tsize = ssize;
+	}
+      }
+    }
+  }
+
+  if (tbase != CMFAIL) {
+
+    if ((m->footprint += tsize) > m->max_footprint)
+      m->max_footprint = m->footprint;
+
+    if (!is_initialized(m)) { /* first-time initialization */
+      m->seg.base = m->least_addr = tbase;
+      m->seg.size = tsize;
+      m->seg.sflags = mmap_flag;
+      m->magic = mparams.magic;
+      m->release_checks = MAX_RELEASE_CHECK_RATE;
+      init_bins(m);
+#if !ONLY_MSPACES
+      if (is_global(m))
+	init_top(m, (mchunkptr)tbase, tsize - TOP_FOOT_SIZE);
+      else
+#endif
+      {
+	/* Offset top by embedded malloc_state */
+	mchunkptr mn = next_chunk(mem2chunk(m));
+	init_top(m, mn, (size_t)((tbase + tsize) - (char*)mn) -TOP_FOOT_SIZE);
+      }
+    }
+
+    else {
+      /* Try to merge with an existing segment */
+      msegmentptr sp = &m->seg;
+      /* Only consider most recent segment if traversal suppressed */
+      while (sp != 0 && tbase != sp->base + sp->size)
+	sp = (NO_SEGMENT_TRAVERSAL) ? 0 : sp->next;
+      if (sp != 0 &&
+	  !is_extern_segment(sp) &&
+	  (sp->sflags & IS_MMAPPED_BIT) == mmap_flag &&
+	  segment_holds(sp, m->top)) { /* append */
+	sp->size += tsize;
+	init_top(m, m->top, m->topsize + tsize);
+      }
+      else {
+	if (tbase < m->least_addr)
+	  m->least_addr = tbase;
+	sp = &m->seg;
+	while (sp != 0 && sp->base != tbase + tsize)
+	  sp = (NO_SEGMENT_TRAVERSAL) ? 0 : sp->next;
+	if (sp != 0 &&
+	    !is_extern_segment(sp) &&
+	    (sp->sflags & IS_MMAPPED_BIT) == mmap_flag) {
+	  char* oldbase = sp->base;
+	  sp->base = tbase;
+	  sp->size += tsize;
+	  return prepend_alloc(m, tbase, oldbase, nb);
+	}
+	else
+	  add_segment(m, tbase, tsize, mmap_flag);
+      }
+    }
+
+    if (nb < m->topsize) { /* Allocate from new or extended top space */
+      size_t rsize = m->topsize -= nb;
+      mchunkptr p = m->top;
+      mchunkptr r = m->top = chunk_plus_offset(p, nb);
+      r->head = rsize | PINUSE_BIT;
+      set_size_and_pinuse_of_inuse_chunk(m, p, nb);
+      check_top_chunk(m, m->top);
+      check_malloced_chunk(m, chunk2mem(p), nb);
+      return chunk2mem(p);
+    }
+  }
+
+  MALLOC_FAILURE_ACTION;
+  return 0;
+}
+
+/* -----------------------  system deallocation -------------------------- */
+
+/* Unmap and unlink any mmapped segments that don't contain used chunks */
+static size_t release_unused_segments(mstate m) {
+  size_t released = 0;
+  int nsegs = 0;
+  msegmentptr pred = &m->seg;
+  msegmentptr sp = pred->next;
+  while (sp != 0) {
+    char* base = sp->base;
+    size_t size = sp->size;
+    msegmentptr next = sp->next;
+    ++nsegs;
+    if (is_mmapped_segment(sp) && !is_extern_segment(sp)) {
+      mchunkptr p = align_as_chunk(base);
+      size_t psize = chunksize(p);
+      /* Can unmap if first chunk holds entire segment and not pinned */
+      if (!cinuse(p) && (char*)p + psize >= base + size - TOP_FOOT_SIZE) {
+	tchunkptr tp = (tchunkptr)p;
+	assert(segment_holds(sp, (char*)sp));
+	if (p == m->dv) {
+	  m->dv = 0;
+	  m->dvsize = 0;
+	}
+	else {
+	  unlink_large_chunk(m, tp);
+	}
+	if (CALL_MUNMAP(base, size) == 0) {
+	  released += size;
+	  m->footprint -= size;
+	  /* unlink obsoleted record */
+	  sp = pred;
+	  sp->next = next;
+	}
+	else { /* back out if cannot unmap */
+	  insert_large_chunk(m, tp, psize);
+	}
+      }
+    }
+    if (NO_SEGMENT_TRAVERSAL) /* scan only first segment */
+      break;
+    pred = sp;
+    sp = next;
+  }
+  /* Reset check counter */
+  m->release_checks = ((nsegs > MAX_RELEASE_CHECK_RATE)?
+		       nsegs : MAX_RELEASE_CHECK_RATE);
+  return released;
+}
+
+static int sys_trim(mstate m, size_t pad) {
+  size_t released = 0;
+  ensure_initialization();
+  if (pad < MAX_REQUEST && is_initialized(m)) {
+    pad += TOP_FOOT_SIZE; /* ensure enough room for segment overhead */
+
+    if (m->topsize > pad) {
+      /* Shrink top space in granularity-size units, keeping at least one */
+      size_t unit = mparams.granularity;
+      size_t extra = ((m->topsize - pad + (unit - SIZE_T_ONE)) / unit -
+		      SIZE_T_ONE) * unit;
+      msegmentptr sp = segment_holding(m, (char*)m->top);
+
+      if (!is_extern_segment(sp)) {
+	if (is_mmapped_segment(sp)) {
+	  if (HAVE_MMAP &&
+	      sp->size >= extra &&
+	      !has_segment_link(m, sp)) { /* can't shrink if pinned */
+	    size_t newsize = sp->size - extra;
+	    /* Prefer mremap, fall back to munmap */
+	    if ((CALL_MREMAP(sp->base, sp->size, newsize, 0) != MFAIL) ||
+		(CALL_MUNMAP(sp->base + newsize, extra) == 0)) {
+	      released = extra;
+	    }
+	  }
+	}
+	else if (HAVE_MORECORE) {
+	  if (extra >= HALF_MAX_SIZE_T) /* Avoid wrapping negative */
+	    extra = (HALF_MAX_SIZE_T) + SIZE_T_ONE - unit;
+	  ACQUIRE_MALLOC_GLOBAL_LOCK();
+	  {
+	    /* Make sure end of memory is where we last set it. */
+	    char* old_br = (char*)(CALL_MORECORE(0));
+	    if (old_br == sp->base + sp->size) {
+	      char* rel_br = (char*)(CALL_MORECORE(-extra));
+	      char* new_br = (char*)(CALL_MORECORE(0));
+	      if (rel_br != CMFAIL && new_br < old_br)
+		released = old_br - new_br;
+	    }
+	  }
+	  RELEASE_MALLOC_GLOBAL_LOCK();
+	}
+      }
+
+      if (released != 0) {
+	sp->size -= released;
+	m->footprint -= released;
+	init_top(m, m->top, m->topsize - released);
+	check_top_chunk(m, m->top);
+      }
+    }
+
+    /* Unmap any unused mmapped segments */
+    if (HAVE_MMAP)
+      released += release_unused_segments(m);
+
+    /* On failure, disable autotrim to avoid repeated failed future calls */
+    if (released == 0 && m->topsize > m->trim_check)
+      m->trim_check = MAX_SIZE_T;
+  }
+
+  return (released != 0)? 1 : 0;
+}
+
+
+/* ---------------------------- malloc support --------------------------- */
+
+/* allocate a large request from the best fitting chunk in a treebin */
+static void* tmalloc_large(mstate m, size_t nb) {
+  tchunkptr v = 0;
+  size_t rsize = -nb; /* Unsigned negation */
+  tchunkptr t;
+  bindex_t idx;
+  compute_tree_index(nb, idx);
+  if ((t = *treebin_at(m, idx)) != 0) {
+    /* Traverse tree for this bin looking for node with size == nb */
+    size_t sizebits = nb << leftshift_for_tree_index(idx);
+    tchunkptr rst = 0;  /* The deepest untaken right subtree */
+    for (;;) {
+      tchunkptr rt;
+      size_t trem = chunksize(t) - nb;
+      if (trem < rsize) {
+	v = t;
+	if ((rsize = trem) == 0)
+	  break;
+      }
+      rt = t->child[1];
+      t = t->child[(sizebits >> (SIZE_T_BITSIZE-SIZE_T_ONE)) & 1];
+      if (rt != 0 && rt != t)
+	rst = rt;
+      if (t == 0) {
+	t = rst; /* set t to least subtree holding sizes > nb */
+	break;
+      }
+      sizebits <<= 1;
+    }
+  }
+  if (t == 0 && v == 0) { /* set t to root of next non-empty treebin */
+    binmap_t leftbits = left_bits(idx2bit(idx)) & m->treemap;
+    if (leftbits != 0) {
+      bindex_t i;
+      binmap_t leastbit = least_bit(leftbits);
+      compute_bit2idx(leastbit, i);
+      t = *treebin_at(m, i);
+    }
+  }
+
+  while (t != 0) { /* find smallest of tree or subtree */
+    size_t trem = chunksize(t) - nb;
+    if (trem < rsize) {
+      rsize = trem;
+      v = t;
+    }
+    t = leftmost_child(t);
+  }
+
+  /*  If dv is a better fit, return 0 so malloc will use it */
+  if (v != 0 && rsize < (size_t)(m->dvsize - nb)) {
+    if (RTCHECK(ok_address(m, v))) { /* split */
+      mchunkptr r = chunk_plus_offset(v, nb);
+      assert(chunksize(v) == rsize + nb);
+      if (RTCHECK(ok_next(v, r))) {
+	unlink_large_chunk(m, v);
+	if (rsize < MIN_CHUNK_SIZE)
+	  set_inuse_and_pinuse(m, v, (rsize + nb));
+	else {
+	  set_size_and_pinuse_of_inuse_chunk(m, v, nb);
+	  set_size_and_pinuse_of_free_chunk(r, rsize);
+	  insert_chunk(m, r, rsize);
+	}
+	return chunk2mem(v);
+      }
+    }
+    CORRUPTION_ERROR_ACTION(m);
+  }
+  return 0;
+}
+
+/* allocate a small request from the best fitting chunk in a treebin */
+static void* tmalloc_small(mstate m, size_t nb) {
+  tchunkptr t, v;
+  size_t rsize;
+  bindex_t i;
+  binmap_t leastbit = least_bit(m->treemap);
+  compute_bit2idx(leastbit, i);
+  v = t = *treebin_at(m, i);
+  rsize = chunksize(t) - nb;
+
+  while ((t = leftmost_child(t)) != 0) {
+    size_t trem = chunksize(t) - nb;
+    if (trem < rsize) {
+      rsize = trem;
+      v = t;
+    }
+  }
+
+  if (RTCHECK(ok_address(m, v))) {
+    mchunkptr r = chunk_plus_offset(v, nb);
+    assert(chunksize(v) == rsize + nb);
+    if (RTCHECK(ok_next(v, r))) {
+      unlink_large_chunk(m, v);
+      if (rsize < MIN_CHUNK_SIZE)
+	set_inuse_and_pinuse(m, v, (rsize + nb));
+      else {
+	set_size_and_pinuse_of_inuse_chunk(m, v, nb);
+	set_size_and_pinuse_of_free_chunk(r, rsize);
+	replace_dv(m, r, rsize);
+      }
+      return chunk2mem(v);
+    }
+  }
+
+  CORRUPTION_ERROR_ACTION(m);
+  return 0;
+}
+
+/* --------------------------- realloc support --------------------------- */
+
+static void* internal_realloc(mstate m, void* oldmem, size_t bytes) {
+  if (bytes >= MAX_REQUEST) {
+    MALLOC_FAILURE_ACTION;
+    return 0;
+  }
+  if (!PREACTION(m)) {
+    mchunkptr oldp = mem2chunk(oldmem);
+    size_t oldsize = chunksize(oldp);
+    mchunkptr next = chunk_plus_offset(oldp, oldsize);
+    mchunkptr newp = 0;
+    void* extra = 0;
+
+    /* Try to either shrink or extend into top. Else malloc-copy-free */
+
+    if (RTCHECK(ok_address(m, oldp) && ok_cinuse(oldp) &&
+		ok_next(oldp, next) && ok_pinuse(next))) {
+      size_t nb = request2size(bytes);
+      if (is_mmapped(oldp))
+	newp = mmap_resize(m, oldp, nb);
+      else if (oldsize >= nb) { /* already big enough */
+	size_t rsize = oldsize - nb;
+	newp = oldp;
+	if (rsize >= MIN_CHUNK_SIZE) {
+	  mchunkptr remainder = chunk_plus_offset(newp, nb);
+	  set_inuse(m, newp, nb);
+	  set_inuse(m, remainder, rsize);
+	  extra = chunk2mem(remainder);
+	}
+      }
+      else if (next == m->top && oldsize + m->topsize > nb) {
+	/* Expand into top */
+	size_t newsize = oldsize + m->topsize;
+	size_t newtopsize = newsize - nb;
+	mchunkptr newtop = chunk_plus_offset(oldp, nb);
+	set_inuse(m, oldp, nb);
+	newtop->head = newtopsize |PINUSE_BIT;
+	m->top = newtop;
+	m->topsize = newtopsize;
+	newp = oldp;
+      }
+    }
+    else {
+      USAGE_ERROR_ACTION(m, oldmem);
+      POSTACTION(m);
+      return 0;
+    }
+
+    POSTACTION(m);
+
+    if (newp != 0) {
+      if (extra != 0) {
+	internal_free(m, extra);
+      }
+      check_inuse_chunk(m, newp);
+      return chunk2mem(newp);
+    }
+    else {
+      void* newmem = internal_malloc(m, bytes);
+      if (newmem != 0) {
+	size_t oc = oldsize - overhead_for(oldp);
+	memcpy(newmem, oldmem, (oc < bytes)? oc : bytes);
+	internal_free(m, oldmem);
+      }
+      return newmem;
+    }
+  }
+  return 0;
+}
+
+/* --------------------------- memalign support -------------------------- */
+
+static void* internal_memalign(mstate m, size_t alignment, size_t bytes) {
+  if (alignment <= MALLOC_ALIGNMENT)    /* Can just use malloc */
+    return internal_malloc(m, bytes);
+  if (alignment <  MIN_CHUNK_SIZE) /* must be at least a minimum chunk size */
+    alignment = MIN_CHUNK_SIZE;
+  if ((alignment & (alignment-SIZE_T_ONE)) != 0) {/* Ensure a power of 2 */
+    size_t a = MALLOC_ALIGNMENT << 1;
+    while (a < alignment) a <<= 1;
+    alignment = a;
+  }
+
+  if (bytes >= MAX_REQUEST - alignment) {
+    if (m != 0)  { /* Test isn't needed but avoids compiler warning */
+      MALLOC_FAILURE_ACTION;
+    }
+  }
+  else {
+    size_t nb = request2size(bytes);
+    size_t req = nb + alignment + MIN_CHUNK_SIZE - CHUNK_OVERHEAD;
+    char* mem = (char*)internal_malloc(m, req);
+    if (mem != 0) {
+      void* leader = 0;
+      void* trailer = 0;
+      mchunkptr p = mem2chunk(mem);
+
+      if (PREACTION(m)) return 0;
+      if ((((size_t)(mem)) % alignment) != 0) { /* misaligned */
+	/*
+	  Find an aligned spot inside chunk.  Since we need to give
+	  back leading space in a chunk of at least MIN_CHUNK_SIZE, if
+	  the first calculation places us at a spot with less than
+	  MIN_CHUNK_SIZE leader, we can move to the next aligned spot.
+	  We've allocated enough total room so that this is always
+	  possible.
+	*/
+	char* br = (char*)mem2chunk((size_t)(((size_t)(mem +
+						       alignment -
+						       SIZE_T_ONE)) &
+					     -alignment));
+	char* pos = ((size_t)(br - (char*)(p)) >= MIN_CHUNK_SIZE)?
+	  br : br+alignment;
+	mchunkptr newp = (mchunkptr)pos;
+	size_t leadsize = pos - (char*)(p);
+	size_t newsize = chunksize(p) - leadsize;
+
+	if (is_mmapped(p)) { /* For mmapped chunks, just adjust offset */
+	  newp->prev_foot = p->prev_foot + leadsize;
+	  newp->head = (newsize|CINUSE_BIT);
+	}
+	else { /* Otherwise, give back leader, use the rest */
+	  set_inuse(m, newp, newsize);
+	  set_inuse(m, p, leadsize);
+	  leader = chunk2mem(p);
+	}
+	p = newp;
+      }
+
+      /* Give back spare room at the end */
+      if (!is_mmapped(p)) {
+	size_t size = chunksize(p);
+	if (size > nb + MIN_CHUNK_SIZE) {
+	  size_t remainder_size = size - nb;
+	  mchunkptr remainder = chunk_plus_offset(p, nb);
+	  set_inuse(m, p, nb);
+	  set_inuse(m, remainder, remainder_size);
+	  trailer = chunk2mem(remainder);
+	}
+      }
+
+      assert (chunksize(p) >= nb);
+      assert((((size_t)(chunk2mem(p))) % alignment) == 0);
+      check_inuse_chunk(m, p);
+      POSTACTION(m);
+      if (leader != 0) {
+	internal_free(m, leader);
+      }
+      if (trailer != 0) {
+	internal_free(m, trailer);
+      }
+      return chunk2mem(p);
+    }
+  }
+  return 0;
+}
+
+/* ------------------------ comalloc/coalloc support --------------------- */
+
+static void** ialloc(mstate m,
+		     size_t n_elements,
+		     size_t* sizes,
+		     int opts,
+		     void* chunks[]) {
+  /*
+    This provides common support for independent_X routines, handling
+    all of the combinations that can result.
+
+    The opts arg has:
+    bit 0 set if all elements are same size (using sizes[0])
+    bit 1 set if elements should be zeroed
+  */
+
+  size_t    element_size;   /* chunksize of each element, if all same */
+  size_t    contents_size;  /* total size of elements */
+  size_t    array_size;     /* request size of pointer array */
+  void*     mem;            /* malloced aggregate space */
+  mchunkptr p;              /* corresponding chunk */
+  size_t    remainder_size; /* remaining bytes while splitting */
+  void**    marray;         /* either "chunks" or malloced ptr array */
+  mchunkptr array_chunk;    /* chunk for malloced ptr array */
+  flag_t    was_enabled;    /* to disable mmap */
+  size_t    size;
+  size_t    i;
+
+  ensure_initialization();
+  /* compute array length, if needed */
+  if (chunks != 0) {
+    if (n_elements == 0)
+      return chunks; /* nothing to do */
+    marray = chunks;
+    array_size = 0;
+  }
+  else {
+    /* if empty req, must still return chunk representing empty array */
+    if (n_elements == 0)
+      return (void**)internal_malloc(m, 0);
+    marray = 0;
+    array_size = request2size(n_elements * (sizeof(void*)));
+  }
+
+  /* compute total element size */
+  if (opts & 0x1) { /* all-same-size */
+    element_size = request2size(*sizes);
+    contents_size = n_elements * element_size;
+  }
+  else { /* add up all the sizes */
+    element_size = 0;
+    contents_size = 0;
+    for (i = 0; i != n_elements; ++i)
+      contents_size += request2size(sizes[i]);
+  }
+
+  size = contents_size + array_size;
+
+  /*
+     Allocate the aggregate chunk.  First disable direct-mmapping so
+     malloc won't use it, since we would not be able to later
+     free/realloc space internal to a segregated mmap region.
+  */
+  was_enabled = use_mmap(m);
+  disable_mmap(m);
+  mem = internal_malloc(m, size - CHUNK_OVERHEAD);
+  if (was_enabled)
+    enable_mmap(m);
+  if (mem == 0)
+    return 0;
+
+  if (PREACTION(m)) return 0;
+  p = mem2chunk(mem);
+  remainder_size = chunksize(p);
+
+  assert(!is_mmapped(p));
+
+  if (opts & 0x2) {       /* optionally clear the elements */
+    memset((size_t*)mem, 0, remainder_size - SIZE_T_SIZE - array_size);
+  }
+
+  /* If not provided, allocate the pointer array as final part of chunk */
+  if (marray == 0) {
+    size_t  array_chunk_size;
+    array_chunk = chunk_plus_offset(p, contents_size);
+    array_chunk_size = remainder_size - contents_size;
+    marray = (void**) (chunk2mem(array_chunk));
+    set_size_and_pinuse_of_inuse_chunk(m, array_chunk, array_chunk_size);
+    remainder_size = contents_size;
+  }
+
+  /* split out elements */
+  for (i = 0; ; ++i) {
+    marray[i] = chunk2mem(p);
+    if (i != n_elements-1) {
+      if (element_size != 0)
+	size = element_size;
+      else
+	size = request2size(sizes[i]);
+      remainder_size -= size;
+      set_size_and_pinuse_of_inuse_chunk(m, p, size);
+      p = chunk_plus_offset(p, size);
+    }
+    else { /* the final element absorbs any overallocation slop */
+      set_size_and_pinuse_of_inuse_chunk(m, p, remainder_size);
+      break;
+    }
+  }
+
+#if DEBUG
+  if (marray != chunks) {
+    /* final element must have exactly exhausted chunk */
+    if (element_size != 0) {
+      assert(remainder_size == element_size);
+    }
+    else {
+      assert(remainder_size == request2size(sizes[i]));
+    }
+    check_inuse_chunk(m, mem2chunk(marray));
+  }
+  for (i = 0; i != n_elements; ++i)
+    check_inuse_chunk(m, mem2chunk(marray[i]));
+
+#endif /* DEBUG */
+
+  POSTACTION(m);
+  return marray;
+}
+
+
+/* -------------------------- public routines ---------------------------- */
+
+#if !ONLY_MSPACES
+
+void* dlmalloc(size_t bytes) {
+  /*
+     Basic algorithm:
+     If a small request (< 256 bytes minus per-chunk overhead):
+       1. If one exists, use a remainderless chunk in associated smallbin.
+	  (Remainderless means that there are too few excess bytes to
+	  represent as a chunk.)
+       2. If it is big enough, use the dv chunk, which is normally the
+	  chunk adjacent to the one used for the most recent small request.
+       3. If one exists, split the smallest available chunk in a bin,
+	  saving remainder in dv.
+       4. If it is big enough, use the top chunk.
+       5. If available, get memory from system and use it
+     Otherwise, for a large request:
+       1. Find the smallest available binned chunk that fits, and use it
+	  if it is better fitting than dv chunk, splitting if necessary.
+       2. If better fitting than any binned chunk, use the dv chunk.
+       3. If it is big enough, use the top chunk.
+       4. If request size >= mmap threshold, try to directly mmap this chunk.
+       5. If available, get memory from system and use it
+
+     The ugly goto's here ensure that postaction occurs along all paths.
+  */
+
+#if USE_LOCKS
+  ensure_initialization(); /* initialize in sys_alloc if not using locks */
+#endif
+
+  if (!PREACTION(gm)) {
+    void* mem;
+    size_t nb;
+    if (bytes <= MAX_SMALL_REQUEST) {
+      bindex_t idx;
+      binmap_t smallbits;
+      nb = (bytes < MIN_REQUEST)? MIN_CHUNK_SIZE : pad_request(bytes);
+      idx = small_index(nb);
+      smallbits = gm->smallmap >> idx;
+
+      if ((smallbits & 0x3U) != 0) { /* Remainderless fit to a smallbin. */
+	mchunkptr b, p;
+	idx += ~smallbits & 1;       /* Uses next bin if idx empty */
+	b = smallbin_at(gm, idx);
+	p = b->fd;
+	assert(chunksize(p) == small_index2size(idx));
+	unlink_first_small_chunk(gm, b, p, idx);
+	set_inuse_and_pinuse(gm, p, small_index2size(idx));
+	mem = chunk2mem(p);
+	check_malloced_chunk(gm, mem, nb);
+	goto postaction;
+      }
+
+      else if (nb > gm->dvsize) {
+	if (smallbits != 0) { /* Use chunk in next nonempty smallbin */
+	  mchunkptr b, p, r;
+	  size_t rsize;
+	  bindex_t i;
+	  binmap_t leftbits = (smallbits << idx) & left_bits(idx2bit(idx));
+	  binmap_t leastbit = least_bit(leftbits);
+	  compute_bit2idx(leastbit, i);
+	  b = smallbin_at(gm, i);
+	  p = b->fd;
+	  assert(chunksize(p) == small_index2size(i));
+	  unlink_first_small_chunk(gm, b, p, i);
+	  rsize = small_index2size(i) - nb;
+	  /* Fit here cannot be remainderless if 4byte sizes */
+	  if (SIZE_T_SIZE != 4 && rsize < MIN_CHUNK_SIZE)
+	    set_inuse_and_pinuse(gm, p, small_index2size(i));
+	  else {
+	    set_size_and_pinuse_of_inuse_chunk(gm, p, nb);
+	    r = chunk_plus_offset(p, nb);
+	    set_size_and_pinuse_of_free_chunk(r, rsize);
+	    replace_dv(gm, r, rsize);
+	  }
+	  mem = chunk2mem(p);
+	  check_malloced_chunk(gm, mem, nb);
+	  goto postaction;
+	}
+
+	else if (gm->treemap != 0 && (mem = tmalloc_small(gm, nb)) != 0) {
+	  check_malloced_chunk(gm, mem, nb);
+	  goto postaction;
+	}
+      }
+    }
+    else if (bytes >= MAX_REQUEST)
+      nb = MAX_SIZE_T; /* Too big to allocate. Force failure (in sys alloc) */
+    else {
+      nb = pad_request(bytes);
+      if (gm->treemap != 0 && (mem = tmalloc_large(gm, nb)) != 0) {
+	check_malloced_chunk(gm, mem, nb);
+	goto postaction;
+      }
+    }
+
+    if (nb <= gm->dvsize) {
+      size_t rsize = gm->dvsize - nb;
+      mchunkptr p = gm->dv;
+      if (rsize >= MIN_CHUNK_SIZE) { /* split dv */
+	mchunkptr r = gm->dv = chunk_plus_offset(p, nb);
+	gm->dvsize = rsize;
+	set_size_and_pinuse_of_free_chunk(r, rsize);
+	set_size_and_pinuse_of_inuse_chunk(gm, p, nb);
+      }
+      else { /* exhaust dv */
+	size_t dvs = gm->dvsize;
+	gm->dvsize = 0;
+	gm->dv = 0;
+	set_inuse_and_pinuse(gm, p, dvs);
+      }
+      mem = chunk2mem(p);
+      check_malloced_chunk(gm, mem, nb);
+      goto postaction;
+    }
+
+    else if (nb < gm->topsize) { /* Split top */
+      size_t rsize = gm->topsize -= nb;
+      mchunkptr p = gm->top;
+      mchunkptr r = gm->top = chunk_plus_offset(p, nb);
+      r->head = rsize | PINUSE_BIT;
+      set_size_and_pinuse_of_inuse_chunk(gm, p, nb);
+      mem = chunk2mem(p);
+      check_top_chunk(gm, gm->top);
+      check_malloced_chunk(gm, mem, nb);
+      goto postaction;
+    }
+
+    mem = sys_alloc(gm, nb);
+
+  postaction:
+    POSTACTION(gm);
+    return mem;
+  }
+
+  return 0;
+}
+
+void dlfree(void* mem) {
+  /*
+     Consolidate freed chunks with preceeding or succeeding bordering
+     free chunks, if they exist, and then place in a bin.  Intermixed
+     with special cases for top, dv, mmapped chunks, and usage errors.
+  */
+
+  if (mem != 0) {
+    mchunkptr p  = mem2chunk(mem);
+#if FOOTERS
+    mstate fm = get_mstate_for(p);
+    if (!ok_magic(fm)) {
+      USAGE_ERROR_ACTION(fm, p);
+      return;
+    }
+#else /* FOOTERS */
+#define fm gm
+#endif /* FOOTERS */
+    if (!PREACTION(fm)) {
+      check_inuse_chunk(fm, p);
+      if (RTCHECK(ok_address(fm, p) && ok_cinuse(p))) {
+	size_t psize = chunksize(p);
+	mchunkptr next = chunk_plus_offset(p, psize);
+	if (!pinuse(p)) {
+	  size_t prevsize = p->prev_foot;
+	  if ((prevsize & IS_MMAPPED_BIT) != 0) {
+	    prevsize &= ~IS_MMAPPED_BIT;
+	    psize += prevsize + MMAP_FOOT_PAD;
+	    if (CALL_MUNMAP((char*)p - prevsize, psize) == 0)
+	      fm->footprint -= psize;
+	    goto postaction;
+	  }
+	  else {
+	    mchunkptr prev = chunk_minus_offset(p, prevsize);
+	    psize += prevsize;
+	    p = prev;
+	    if (RTCHECK(ok_address(fm, prev))) { /* consolidate backward */
+	      if (p != fm->dv) {
+		unlink_chunk(fm, p, prevsize);
+	      }
+	      else if ((next->head & INUSE_BITS) == INUSE_BITS) {
+		fm->dvsize = psize;
+		set_free_with_pinuse(p, psize, next);
+		goto postaction;
+	      }
+	    }
+	    else
+	      goto erroraction;
+	  }
+	}
+
+	if (RTCHECK(ok_next(p, next) && ok_pinuse(next))) {
+	  if (!cinuse(next)) {  /* consolidate forward */
+	    if (next == fm->top) {
+	      size_t tsize = fm->topsize += psize;
+	      fm->top = p;
+	      p->head = tsize | PINUSE_BIT;
+	      if (p == fm->dv) {
+		fm->dv = 0;
+		fm->dvsize = 0;
+	      }
+	      if (should_trim(fm, tsize))
+		sys_trim(fm, 0);
+	      goto postaction;
+	    }
+	    else if (next == fm->dv) {
+	      size_t dsize = fm->dvsize += psize;
+	      fm->dv = p;
+	      set_size_and_pinuse_of_free_chunk(p, dsize);
+	      goto postaction;
+	    }
+	    else {
+	      size_t nsize = chunksize(next);
+	      psize += nsize;
+	      unlink_chunk(fm, next, nsize);
+	      set_size_and_pinuse_of_free_chunk(p, psize);
+	      if (p == fm->dv) {
+		fm->dvsize = psize;
+		goto postaction;
+	      }
+	    }
+	  }
+	  else
+	    set_free_with_pinuse(p, psize, next);
+
+	  if (is_small(psize)) {
+	    insert_small_chunk(fm, p, psize);
+	    check_free_chunk(fm, p);
+	  }
+	  else {
+	    tchunkptr tp = (tchunkptr)p;
+	    insert_large_chunk(fm, tp, psize);
+	    check_free_chunk(fm, p);
+	    if (--fm->release_checks == 0)
+	      release_unused_segments(fm);
+	  }
+	  goto postaction;
+	}
+      }
+    erroraction:
+      USAGE_ERROR_ACTION(fm, p);
+    postaction:
+      POSTACTION(fm);
+    }
+  }
+#if !FOOTERS
+#undef fm
+#endif /* FOOTERS */
+}
+
+void* dlcalloc(size_t n_elements, size_t elem_size) {
+  void* mem;
+  size_t req = 0;
+  if (n_elements != 0) {
+    req = n_elements * elem_size;
+    if (((n_elements | elem_size) & ~(size_t)0xffff) &&
+	(req / n_elements != elem_size))
+      req = MAX_SIZE_T; /* force downstream failure on overflow */
+  }
+  mem = dlmalloc(req);
+  if (mem != 0 && calloc_must_clear(mem2chunk(mem)))
+    memset(mem, 0, req);
+  return mem;
+}
+
+void* dlrealloc(void* oldmem, size_t bytes) {
+  if (oldmem == 0)
+    return dlmalloc(bytes);
+#ifdef REALLOC_ZERO_BYTES_FREES
+  if (bytes == 0) {
+    dlfree(oldmem);
+    return 0;
+  }
+#endif /* REALLOC_ZERO_BYTES_FREES */
+  else {
+#if ! FOOTERS
+    mstate m = gm;
+#else /* FOOTERS */
+    mstate m = get_mstate_for(mem2chunk(oldmem));
+    if (!ok_magic(m)) {
+      USAGE_ERROR_ACTION(m, oldmem);
+      return 0;
+    }
+#endif /* FOOTERS */
+    return internal_realloc(m, oldmem, bytes);
+  }
+}
+
+void* dlmemalign(size_t alignment, size_t bytes) {
+  return internal_memalign(gm, alignment, bytes);
+}
+
+void** dlindependent_calloc(size_t n_elements, size_t elem_size,
+				 void* chunks[]) {
+  size_t sz = elem_size; /* serves as 1-element array */
+  return ialloc(gm, n_elements, &sz, 3, chunks);
+}
+
+void** dlindependent_comalloc(size_t n_elements, size_t sizes[],
+				   void* chunks[]) {
+  return ialloc(gm, n_elements, sizes, 0, chunks);
+}
+
+void* dlvalloc(size_t bytes) {
+  size_t pagesz;
+  ensure_initialization();
+  pagesz = mparams.page_size;
+  return dlmemalign(pagesz, bytes);
+}
+
+void* dlpvalloc(size_t bytes) {
+  size_t pagesz;
+  ensure_initialization();
+  pagesz = mparams.page_size;
+  return dlmemalign(pagesz, (bytes + pagesz - SIZE_T_ONE) & ~(pagesz - SIZE_T_ONE));
+}
+
+int dlmalloc_trim(size_t pad) {
+  ensure_initialization();
+  int result = 0;
+  if (!PREACTION(gm)) {
+    result = sys_trim(gm, pad);
+    POSTACTION(gm);
+  }
+  return result;
+}
+
+size_t dlmalloc_footprint(void) {
+  return gm->footprint;
+}
+
+size_t dlmalloc_max_footprint(void) {
+  return gm->max_footprint;
+}
+
+#if !NO_MALLINFO
+struct mallinfo dlmallinfo(void) {
+  return internal_mallinfo(gm);
+}
+#endif /* NO_MALLINFO */
+
+void dlmalloc_stats() {
+  internal_malloc_stats(gm);
+}
+
+int dlmallopt(int param_number, int value) {
+  return change_mparam(param_number, value);
+}
+
+#endif /* !ONLY_MSPACES */
+
+size_t dlmalloc_usable_size(void* mem) {
+  if (mem != 0) {
+    mchunkptr p = mem2chunk(mem);
+    if (cinuse(p))
+      return chunksize(p) - overhead_for(p);
+  }
+  return 0;
+}
+
+/* ----------------------------- user mspaces ---------------------------- */
+
+#if MSPACES
+
+static mstate init_user_mstate(char* tbase, size_t tsize) {
+  size_t msize = pad_request(sizeof(struct malloc_state));
+  mchunkptr mn;
+  mchunkptr msp = align_as_chunk(tbase);
+  mstate m = (mstate)(chunk2mem(msp));
+  memset(m, 0, msize);
+  INITIAL_LOCK(&m->mutex);
+  msp->head = (msize|PINUSE_BIT|CINUSE_BIT);
+  m->seg.base = m->least_addr = tbase;
+  m->seg.size = m->footprint = m->max_footprint = tsize;
+  m->magic = mparams.magic;
+  m->release_checks = MAX_RELEASE_CHECK_RATE;
+  m->mflags = mparams.default_mflags;
+  m->extp = 0;
+  m->exts = 0;
+  disable_contiguous(m);
+  init_bins(m);
+  mn = next_chunk(mem2chunk(m));
+  init_top(m, mn, (size_t)((tbase + tsize) - (char*)mn) - TOP_FOOT_SIZE);
+  check_top_chunk(m, m->top);
+  return m;
+}
+
+mspace create_mspace(size_t capacity, int locked) {
+  mstate m = 0;
+  size_t msize;
+  ensure_initialization();
+  msize = pad_request(sizeof(struct malloc_state));
+  if (capacity < (size_t) -(msize + TOP_FOOT_SIZE + mparams.page_size)) {
+    size_t rs = ((capacity == 0)? mparams.granularity :
+		 (capacity + TOP_FOOT_SIZE + msize));
+    size_t tsize = granularity_align(rs);
+    char* tbase = (char*)(CALL_MMAP(tsize));
+    if (tbase != CMFAIL) {
+      m = init_user_mstate(tbase, tsize);
+      m->seg.sflags = IS_MMAPPED_BIT;
+      set_lock(m, locked);
+    }
+  }
+  return (mspace)m;
+}
+
+mspace create_mspace_with_base(void* base, size_t capacity, int locked) {
+  mstate m = 0;
+  size_t msize;
+  ensure_initialization();
+  msize = pad_request(sizeof(struct malloc_state));
+  if (capacity > msize + TOP_FOOT_SIZE &&
+      capacity < (size_t) -(msize + TOP_FOOT_SIZE + mparams.page_size)) {
+    m = init_user_mstate((char*)base, capacity);
+    m->seg.sflags = EXTERN_BIT;
+    set_lock(m, locked);
+  }
+  return (mspace)m;
+}
+
+int mspace_mmap_large_chunks(mspace msp, int enable) {
+  int ret = 0;
+  mstate ms = (mstate)msp;
+  if (!PREACTION(ms)) {
+    if (use_mmap(ms))
+      ret = 1;
+    if (enable)
+      enable_mmap(ms);
+    else
+      disable_mmap(ms);
+    POSTACTION(ms);
+  }
+  return ret;
+}
+
+size_t destroy_mspace(mspace msp) {
+  size_t freed = 0;
+  mstate ms = (mstate)msp;
+  if (ok_magic(ms)) {
+    msegmentptr sp = &ms->seg;
+    while (sp != 0) {
+      char* base = sp->base;
+      size_t size = sp->size;
+      flag_t flag = sp->sflags;
+      sp = sp->next;
+      if ((flag & IS_MMAPPED_BIT) && !(flag & EXTERN_BIT) &&
+	  CALL_MUNMAP(base, size) == 0)
+	freed += size;
+    }
+  }
+  else {
+    USAGE_ERROR_ACTION(ms,ms);
+  }
+  return freed;
+}
+
+/*
+  mspace versions of routines are near-clones of the global
+  versions. This is not so nice but better than the alternatives.
+*/
+
+
+void* mspace_malloc(mspace msp, size_t bytes) {
+  mstate ms = (mstate)msp;
+  if (!ok_magic(ms)) {
+    USAGE_ERROR_ACTION(ms,ms);
+    return 0;
+  }
+  if (!PREACTION(ms)) {
+    void* mem;
+    size_t nb;
+    if (bytes <= MAX_SMALL_REQUEST) {
+      bindex_t idx;
+      binmap_t smallbits;
+      nb = (bytes < MIN_REQUEST)? MIN_CHUNK_SIZE : pad_request(bytes);
+      idx = small_index(nb);
+      smallbits = ms->smallmap >> idx;
+
+      if ((smallbits & 0x3U) != 0) { /* Remainderless fit to a smallbin. */
+	mchunkptr b, p;
+	idx += ~smallbits & 1;       /* Uses next bin if idx empty */
+	b = smallbin_at(ms, idx);
+	p = b->fd;
+	assert(chunksize(p) == small_index2size(idx));
+	unlink_first_small_chunk(ms, b, p, idx);
+	set_inuse_and_pinuse(ms, p, small_index2size(idx));
+	mem = chunk2mem(p);
+	check_malloced_chunk(ms, mem, nb);
+	goto postaction;
+      }
+
+      else if (nb > ms->dvsize) {
+	if (smallbits != 0) { /* Use chunk in next nonempty smallbin */
+	  mchunkptr b, p, r;
+	  size_t rsize;
+	  bindex_t i;
+	  binmap_t leftbits = (smallbits << idx) & left_bits(idx2bit(idx));
+	  binmap_t leastbit = least_bit(leftbits);
+	  compute_bit2idx(leastbit, i);
+	  b = smallbin_at(ms, i);
+	  p = b->fd;
+	  assert(chunksize(p) == small_index2size(i));
+	  unlink_first_small_chunk(ms, b, p, i);
+	  rsize = small_index2size(i) - nb;
+	  /* Fit here cannot be remainderless if 4byte sizes */
+	  if (SIZE_T_SIZE != 4 && rsize < MIN_CHUNK_SIZE)
+	    set_inuse_and_pinuse(ms, p, small_index2size(i));
+	  else {
+	    set_size_and_pinuse_of_inuse_chunk(ms, p, nb);
+	    r = chunk_plus_offset(p, nb);
+	    set_size_and_pinuse_of_free_chunk(r, rsize);
+	    replace_dv(ms, r, rsize);
+	  }
+	  mem = chunk2mem(p);
+	  check_malloced_chunk(ms, mem, nb);
+	  goto postaction;
+	}
+
+	else if (ms->treemap != 0 && (mem = tmalloc_small(ms, nb)) != 0) {
+	  check_malloced_chunk(ms, mem, nb);
+	  goto postaction;
+	}
+      }
+    }
+    else if (bytes >= MAX_REQUEST)
+      nb = MAX_SIZE_T; /* Too big to allocate. Force failure (in sys alloc) */
+    else {
+      nb = pad_request(bytes);
+      if (ms->treemap != 0 && (mem = tmalloc_large(ms, nb)) != 0) {
+	check_malloced_chunk(ms, mem, nb);
+	goto postaction;
+      }
+    }
+
+    if (nb <= ms->dvsize) {
+      size_t rsize = ms->dvsize - nb;
+      mchunkptr p = ms->dv;
+      if (rsize >= MIN_CHUNK_SIZE) { /* split dv */
+	mchunkptr r = ms->dv = chunk_plus_offset(p, nb);
+	ms->dvsize = rsize;
+	set_size_and_pinuse_of_free_chunk(r, rsize);
+	set_size_and_pinuse_of_inuse_chunk(ms, p, nb);
+      }
+      else { /* exhaust dv */
+	size_t dvs = ms->dvsize;
+	ms->dvsize = 0;
+	ms->dv = 0;
+	set_inuse_and_pinuse(ms, p, dvs);
+      }
+      mem = chunk2mem(p);
+      check_malloced_chunk(ms, mem, nb);
+      goto postaction;
+    }
+
+    else if (nb < ms->topsize) { /* Split top */
+      size_t rsize = ms->topsize -= nb;
+      mchunkptr p = ms->top;
+      mchunkptr r = ms->top = chunk_plus_offset(p, nb);
+      r->head = rsize | PINUSE_BIT;
+      set_size_and_pinuse_of_inuse_chunk(ms, p, nb);
+      mem = chunk2mem(p);
+      check_top_chunk(ms, ms->top);
+      check_malloced_chunk(ms, mem, nb);
+      goto postaction;
+    }
+
+    mem = sys_alloc(ms, nb);
+
+  postaction:
+    POSTACTION(ms);
+    return mem;
+  }
+
+  return 0;
+}
+
+void mspace_free(mspace msp, void* mem) {
+  if (mem != 0) {
+    mchunkptr p  = mem2chunk(mem);
+#if FOOTERS
+    mstate fm = get_mstate_for(p);
+#else /* FOOTERS */
+    mstate fm = (mstate)msp;
+#endif /* FOOTERS */
+    if (!ok_magic(fm)) {
+      USAGE_ERROR_ACTION(fm, p);
+      return;
+    }
+    if (!PREACTION(fm)) {
+      check_inuse_chunk(fm, p);
+      if (RTCHECK(ok_address(fm, p) && ok_cinuse(p))) {
+	size_t psize = chunksize(p);
+	mchunkptr next = chunk_plus_offset(p, psize);
+	if (!pinuse(p)) {
+	  size_t prevsize = p->prev_foot;
+	  if ((prevsize & IS_MMAPPED_BIT) != 0) {
+	    prevsize &= ~IS_MMAPPED_BIT;
+	    psize += prevsize + MMAP_FOOT_PAD;
+	    if (CALL_MUNMAP((char*)p - prevsize, psize) == 0)
+	      fm->footprint -= psize;
+	    goto postaction;
+	  }
+	  else {
+	    mchunkptr prev = chunk_minus_offset(p, prevsize);
+	    psize += prevsize;
+	    p = prev;
+	    if (RTCHECK(ok_address(fm, prev))) { /* consolidate backward */
+	      if (p != fm->dv) {
+		unlink_chunk(fm, p, prevsize);
+	      }
+	      else if ((next->head & INUSE_BITS) == INUSE_BITS) {
+		fm->dvsize = psize;
+		set_free_with_pinuse(p, psize, next);
+		goto postaction;
+	      }
+	    }
+	    else
+	      goto erroraction;
+	  }
+	}
+
+	if (RTCHECK(ok_next(p, next) && ok_pinuse(next))) {
+	  if (!cinuse(next)) {  /* consolidate forward */
+	    if (next == fm->top) {
+	      size_t tsize = fm->topsize += psize;
+	      fm->top = p;
+	      p->head = tsize | PINUSE_BIT;
+	      if (p == fm->dv) {
+		fm->dv = 0;
+		fm->dvsize = 0;
+	      }
+	      if (should_trim(fm, tsize))
+		sys_trim(fm, 0);
+	      goto postaction;
+	    }
+	    else if (next == fm->dv) {
+	      size_t dsize = fm->dvsize += psize;
+	      fm->dv = p;
+	      set_size_and_pinuse_of_free_chunk(p, dsize);
+	      goto postaction;
+	    }
+	    else {
+	      size_t nsize = chunksize(next);
+	      psize += nsize;
+	      unlink_chunk(fm, next, nsize);
+	      set_size_and_pinuse_of_free_chunk(p, psize);
+	      if (p == fm->dv) {
+		fm->dvsize = psize;
+		goto postaction;
+	      }
+	    }
+	  }
+	  else
+	    set_free_with_pinuse(p, psize, next);
+
+	  if (is_small(psize)) {
+	    insert_small_chunk(fm, p, psize);
+	    check_free_chunk(fm, p);
+	  }
+	  else {
+	    tchunkptr tp = (tchunkptr)p;
+	    insert_large_chunk(fm, tp, psize);
+	    check_free_chunk(fm, p);
+	    if (--fm->release_checks == 0)
+	      release_unused_segments(fm);
+	  }
+	  goto postaction;
+	}
+      }
+    erroraction:
+      USAGE_ERROR_ACTION(fm, p);
+    postaction:
+      POSTACTION(fm);
+    }
+  }
+}
+
+void* mspace_calloc(mspace msp, size_t n_elements, size_t elem_size) {
+  void* mem;
+  size_t req = 0;
+  mstate ms = (mstate)msp;
+  if (!ok_magic(ms)) {
+    USAGE_ERROR_ACTION(ms,ms);
+    return 0;
+  }
+  if (n_elements != 0) {
+    req = n_elements * elem_size;
+    if (((n_elements | elem_size) & ~(size_t)0xffff) &&
+	(req / n_elements != elem_size))
+      req = MAX_SIZE_T; /* force downstream failure on overflow */
+  }
+  mem = internal_malloc(ms, req);
+  if (mem != 0 && calloc_must_clear(mem2chunk(mem)))
+    memset(mem, 0, req);
+  return mem;
+}
+
+void* mspace_realloc(mspace msp, void* oldmem, size_t bytes) {
+  if (oldmem == 0)
+    return mspace_malloc(msp, bytes);
+#ifdef REALLOC_ZERO_BYTES_FREES
+  if (bytes == 0) {
+    mspace_free(msp, oldmem);
+    return 0;
+  }
+#endif /* REALLOC_ZERO_BYTES_FREES */
+  else {
+#if FOOTERS
+    mchunkptr p  = mem2chunk(oldmem);
+    mstate ms = get_mstate_for(p);
+#else /* FOOTERS */
+    mstate ms = (mstate)msp;
+#endif /* FOOTERS */
+    if (!ok_magic(ms)) {
+      USAGE_ERROR_ACTION(ms,ms);
+      return 0;
+    }
+    return internal_realloc(ms, oldmem, bytes);
+  }
+}
+
+void* mspace_memalign(mspace msp, size_t alignment, size_t bytes) {
+  mstate ms = (mstate)msp;
+  if (!ok_magic(ms)) {
+    USAGE_ERROR_ACTION(ms,ms);
+    return 0;
+  }
+  return internal_memalign(ms, alignment, bytes);
+}
+
+void** mspace_independent_calloc(mspace msp, size_t n_elements,
+				 size_t elem_size, void* chunks[]) {
+  size_t sz = elem_size; /* serves as 1-element array */
+  mstate ms = (mstate)msp;
+  if (!ok_magic(ms)) {
+    USAGE_ERROR_ACTION(ms,ms);
+    return 0;
+  }
+  return ialloc(ms, n_elements, &sz, 3, chunks);
+}
+
+void** mspace_independent_comalloc(mspace msp, size_t n_elements,
+				   size_t sizes[], void* chunks[]) {
+  mstate ms = (mstate)msp;
+  if (!ok_magic(ms)) {
+    USAGE_ERROR_ACTION(ms,ms);
+    return 0;
+  }
+  return ialloc(ms, n_elements, sizes, 0, chunks);
+}
+
+int mspace_trim(mspace msp, size_t pad) {
+  int result = 0;
+  mstate ms = (mstate)msp;
+  if (ok_magic(ms)) {
+    if (!PREACTION(ms)) {
+      result = sys_trim(ms, pad);
+      POSTACTION(ms);
+    }
+  }
+  else {
+    USAGE_ERROR_ACTION(ms,ms);
+  }
+  return result;
+}
+
+void mspace_malloc_stats(mspace msp) {
+  mstate ms = (mstate)msp;
+  if (ok_magic(ms)) {
+    internal_malloc_stats(ms);
+  }
+  else {
+    USAGE_ERROR_ACTION(ms,ms);
+  }
+}
+
+size_t mspace_footprint(mspace msp) {
+  size_t result = 0;
+  mstate ms = (mstate)msp;
+  if (ok_magic(ms)) {
+    result = ms->footprint;
+  }
+  else {
+    USAGE_ERROR_ACTION(ms,ms);
+  }
+  return result;
+}
+
+
+size_t mspace_max_footprint(mspace msp) {
+  size_t result = 0;
+  mstate ms = (mstate)msp;
+  if (ok_magic(ms)) {
+    result = ms->max_footprint;
+  }
+  else {
+    USAGE_ERROR_ACTION(ms,ms);
+  }
+  return result;
+}
+
+
+#if !NO_MALLINFO
+struct mallinfo mspace_mallinfo(mspace msp) {
+  mstate ms = (mstate)msp;
+  if (!ok_magic(ms)) {
+    USAGE_ERROR_ACTION(ms,ms);
+  }
+  return internal_mallinfo(ms);
+}
+#endif /* NO_MALLINFO */
+
+size_t mspace_usable_size(void* mem) {
+  if (mem != 0) {
+    mchunkptr p = mem2chunk(mem);
+    if (cinuse(p))
+      return chunksize(p) - overhead_for(p);
+  }
+  return 0;
+}
+
+int mspace_mallopt(int param_number, int value) {
+  return change_mparam(param_number, value);
+}
+
+#endif /* MSPACES */
+
+/* -------------------- Alternative MORECORE functions ------------------- */
+
+/*
+  Guidelines for creating a custom version of MORECORE:
+
+  * For best performance, MORECORE should allocate in multiples of pagesize.
+  * MORECORE may allocate more memory than requested. (Or even less,
+      but this will usually result in a malloc failure.)
+  * MORECORE must not allocate memory when given argument zero, but
+      instead return one past the end address of memory from previous
+      nonzero call.
+  * For best performance, consecutive calls to MORECORE with positive
+      arguments should return increasing addresses, indicating that
+      space has been contiguously extended.
+  * Even though consecutive calls to MORECORE need not return contiguous
+      addresses, it must be OK for malloc'ed chunks to span multiple
+      regions in those cases where they do happen to be contiguous.
+  * MORECORE need not handle negative arguments -- it may instead
+      just return MFAIL when given negative arguments.
+      Negative arguments are always multiples of pagesize. MORECORE
+      must not misinterpret negative args as large positive unsigned
+      args. You can suppress all such calls from even occurring by defining
+      MORECORE_CANNOT_TRIM,
+
+  As an example alternative MORECORE, here is a custom allocator
+  kindly contributed for pre-OSX macOS.  It uses virtually but not
+  necessarily physically contiguous non-paged memory (locked in,
+  present and won't get swapped out).  You can use it by uncommenting
+  this section, adding some #includes, and setting up the appropriate
+  defines above:
+
+      #define MORECORE osMoreCore
+
+  There is also a shutdown routine that should somehow be called for
+  cleanup upon program exit.
+
+  #define MAX_POOL_ENTRIES 100
+  #define MINIMUM_MORECORE_SIZE  (64 * 1024U)
+  static int next_os_pool;
+  void *our_os_pools[MAX_POOL_ENTRIES];
+
+  void *osMoreCore(int size)
+  {
+    void *ptr = 0;
+    static void *sbrk_top = 0;
+
+    if (size > 0)
+    {
+      if (size < MINIMUM_MORECORE_SIZE)
+	 size = MINIMUM_MORECORE_SIZE;
+      if (CurrentExecutionLevel() == kTaskLevel)
+	 ptr = PoolAllocateResident(size + RM_PAGE_SIZE, 0);
+      if (ptr == 0)
+      {
+	return (void *) MFAIL;
+      }
+      // save ptrs so they can be freed during cleanup
+      our_os_pools[next_os_pool] = ptr;
+      next_os_pool++;
+      ptr = (void *) ((((size_t) ptr) + RM_PAGE_MASK) & ~RM_PAGE_MASK);
+      sbrk_top = (char *) ptr + size;
+      return ptr;
+    }
+    else if (size < 0)
+    {
+      // we don't currently support shrink behavior
+      return (void *) MFAIL;
+    }
+    else
+    {
+      return sbrk_top;
+    }
+  }
+
+  // cleanup any allocated memory pools
+  // called as last thing before shutting down driver
+
+  void osCleanupMem(void)
+  {
+    void **ptr;
+
+    for (ptr = our_os_pools; ptr < &our_os_pools[MAX_POOL_ENTRIES]; ptr++)
+      if (*ptr)
+      {
+	 PoolDeallocate(*ptr);
+	 *ptr = 0;
+      }
+  }
+
+*/
+
+
+/* -----------------------------------------------------------------------
+History:
+    V2.8.4 (not yet released)
+      * Add mspace_mmap_large_chunks; thanks to Jean Brouwers
+      * Fix insufficient sys_alloc padding when using 16byte alignment
+      * Fix bad error check in mspace_footprint
+      * Adaptations for ptmalloc, courtesy of Wolfram Gloger.
+      * Reentrant spin locks, courtesy of Earl Chew and others
+      * Win32 improvements, courtesy of Niall Douglas and Earl Chew
+      * Add NO_SEGMENT_TRAVERSAL and MAX_RELEASE_CHECK_RATE options
+      * Extension hook in malloc_state
+      * Various small adjustments to reduce warnings on some compilers
+      * Various configuration extensions/changes for more platforms. Thanks
+	 to all who contributed these.
+
+    V2.8.3 Thu Sep 22 11:16:32 2005  Doug Lea  (dl at gee)
+      * Add max_footprint functions
+      * Ensure all appropriate literals are size_t
+      * Fix conditional compilation problem for some #define settings
+      * Avoid concatenating segments with the one provided
+	in create_mspace_with_base
+      * Rename some variables to avoid compiler shadowing warnings
+      * Use explicit lock initialization.
+      * Better handling of sbrk interference.
+      * Simplify and fix segment insertion, trimming and mspace_destroy
+      * Reinstate REALLOC_ZERO_BYTES_FREES option from 2.7.x
+      * Thanks especially to Dennis Flanagan for help on these.
+
+    V2.8.2 Sun Jun 12 16:01:10 2005  Doug Lea  (dl at gee)
+      * Fix memalign brace error.
+
+    V2.8.1 Wed Jun  8 16:11:46 2005  Doug Lea  (dl at gee)
+      * Fix improper #endif nesting in C++
+      * Add explicit casts needed for C++
+
+    V2.8.0 Mon May 30 14:09:02 2005  Doug Lea  (dl at gee)
+      * Use trees for large bins
+      * Support mspaces
+      * Use segments to unify sbrk-based and mmap-based system allocation,
+	removing need for emulation on most platforms without sbrk.
+      * Default safety checks
+      * Optional footer checks. Thanks to William Robertson for the idea.
+      * Internal code refactoring
+      * Incorporate suggestions and platform-specific changes.
+	Thanks to Dennis Flanagan, Colin Plumb, Niall Douglas,
+	Aaron Bachmann,  Emery Berger, and others.
+      * Speed up non-fastbin processing enough to remove fastbins.
+      * Remove useless cfree() to avoid conflicts with other apps.
+      * Remove internal memcpy, memset. Compilers handle builtins better.
+      * Remove some options that no one ever used and rename others.
+
+    V2.7.2 Sat Aug 17 09:07:30 2002  Doug Lea  (dl at gee)
+      * Fix malloc_state bitmap array misdeclaration
+
+    V2.7.1 Thu Jul 25 10:58:03 2002  Doug Lea  (dl at gee)
+      * Allow tuning of FIRST_SORTED_BIN_SIZE
+      * Use PTR_UINT as type for all ptr->int casts. Thanks to John Belmonte.
+      * Better detection and support for non-contiguousness of MORECORE.
+	Thanks to Andreas Mueller, Conal Walsh, and Wolfram Gloger
+      * Bypass most of malloc if no frees. Thanks To Emery Berger.
+      * Fix freeing of old top non-contiguous chunk im sysmalloc.
+      * Raised default trim and map thresholds to 256K.
+      * Fix mmap-related #defines. Thanks to Lubos Lunak.
+      * Fix copy macros; added LACKS_FCNTL_H. Thanks to Neal Walfield.
+      * Branch-free bin calculation
+      * Default trim and mmap thresholds now 256K.
+
+    V2.7.0 Sun Mar 11 14:14:06 2001  Doug Lea  (dl at gee)
+      * Introduce independent_comalloc and independent_calloc.
+	Thanks to Michael Pachos for motivation and help.
+      * Make optional .h file available
+      * Allow > 2GB requests on 32bit systems.
+      * new WIN32 sbrk, mmap, munmap, lock code from <Walter@GeNeSys-e.de>.
+	Thanks also to Andreas Mueller <a.mueller at paradatec.de>,
+	and Anonymous.
+      * Allow override of MALLOC_ALIGNMENT (Thanks to Ruud Waij for
+	helping test this.)
+      * memalign: check alignment arg
+      * realloc: don't try to shift chunks backwards, since this
+	leads to  more fragmentation in some programs and doesn't
+	seem to help in any others.
+      * Collect all cases in malloc requiring system memory into sysmalloc
+      * Use mmap as backup to sbrk
+      * Place all internal state in malloc_state
+      * Introduce fastbins (although similar to 2.5.1)
+      * Many minor tunings and cosmetic improvements
+      * Introduce USE_PUBLIC_MALLOC_WRAPPERS, USE_MALLOC_LOCK
+      * Introduce MALLOC_FAILURE_ACTION, MORECORE_CONTIGUOUS
+	Thanks to Tony E. Bennett <tbennett@nvidia.com> and others.
+      * Include errno.h to support default failure action.
+
+    V2.6.6 Sun Dec  5 07:42:19 1999  Doug Lea  (dl at gee)
+      * return null for negative arguments
+      * Added Several WIN32 cleanups from Martin C. Fong <mcfong at yahoo.com>
+	 * Add 'LACKS_SYS_PARAM_H' for those systems without 'sys/param.h'
+	  (e.g. WIN32 platforms)
+	 * Cleanup header file inclusion for WIN32 platforms
+	 * Cleanup code to avoid Microsoft Visual C++ compiler complaints
+	 * Add 'USE_DL_PREFIX' to quickly allow co-existence with existing
+	   memory allocation routines
+	 * Set 'malloc_getpagesize' for WIN32 platforms (needs more work)
+	 * Use 'assert' rather than 'ASSERT' in WIN32 code to conform to
+	   usage of 'assert' in non-WIN32 code
+	 * Improve WIN32 'sbrk()' emulation's 'findRegion()' routine to
+	   avoid infinite loop
+      * Always call 'fREe()' rather than 'free()'
+
+    V2.6.5 Wed Jun 17 15:57:31 1998  Doug Lea  (dl at gee)
+      * Fixed ordering problem with boundary-stamping
+
+    V2.6.3 Sun May 19 08:17:58 1996  Doug Lea  (dl at gee)
+      * Added pvalloc, as recommended by H.J. Liu
+      * Added 64bit pointer support mainly from Wolfram Gloger
+      * Added anonymously donated WIN32 sbrk emulation
+      * Malloc, calloc, getpagesize: add optimizations from Raymond Nijssen
+      * malloc_extend_top: fix mask error that caused wastage after
+	foreign sbrks
+      * Add linux mremap support code from HJ Liu
+
+    V2.6.2 Tue Dec  5 06:52:55 1995  Doug Lea  (dl at gee)
+      * Integrated most documentation with the code.
+      * Add support for mmap, with help from
+	Wolfram Gloger (Gloger@lrz.uni-muenchen.de).
+      * Use last_remainder in more cases.
+      * Pack bins using idea from  colin@nyx10.cs.du.edu
+      * Use ordered bins instead of best-fit threshhold
+      * Eliminate block-local decls to simplify tracing and debugging.
+      * Support another case of realloc via move into top
+      * Fix error occuring when initial sbrk_base not word-aligned.
+      * Rely on page size for units instead of SBRK_UNIT to
+	avoid surprises about sbrk alignment conventions.
+      * Add mallinfo, mallopt. Thanks to Raymond Nijssen
+	(raymond@es.ele.tue.nl) for the suggestion.
+      * Add `pad' argument to malloc_trim and top_pad mallopt parameter.
+      * More precautions for cases where other routines call sbrk,
+	courtesy of Wolfram Gloger (Gloger@lrz.uni-muenchen.de).
+      * Added macros etc., allowing use in linux libc from
+	H.J. Lu (hjl@gnu.ai.mit.edu)
+      * Inverted this history list
+
+    V2.6.1 Sat Dec  2 14:10:57 1995  Doug Lea  (dl at gee)
+      * Re-tuned and fixed to behave more nicely with V2.6.0 changes.
+      * Removed all preallocation code since under current scheme
+	the work required to undo bad preallocations exceeds
+	the work saved in good cases for most test programs.
+      * No longer use return list or unconsolidated bins since
+	no scheme using them consistently outperforms those that don't
+	given above changes.
+      * Use best fit for very large chunks to prevent some worst-cases.
+      * Added some support for debugging
+
+    V2.6.0 Sat Nov  4 07:05:23 1995  Doug Lea  (dl at gee)
+      * Removed footers when chunks are in use. Thanks to
+	Paul Wilson (wilson@cs.texas.edu) for the suggestion.
+
+    V2.5.4 Wed Nov  1 07:54:51 1995  Doug Lea  (dl at gee)
+      * Added malloc_trim, with help from Wolfram Gloger
+	(wmglo@Dent.MED.Uni-Muenchen.DE).
+
+    V2.5.3 Tue Apr 26 10:16:01 1994  Doug Lea  (dl at g)
+
+    V2.5.2 Tue Apr  5 16:20:40 1994  Doug Lea  (dl at g)
+      * realloc: try to expand in both directions
+      * malloc: swap order of clean-bin strategy;
+      * realloc: only conditionally expand backwards
+      * Try not to scavenge used bins
+      * Use bin counts as a guide to preallocation
+      * Occasionally bin return list chunks in first scan
+      * Add a few optimizations from colin@nyx10.cs.du.edu
+
+    V2.5.1 Sat Aug 14 15:40:43 1993  Doug Lea  (dl at g)
+      * faster bin computation & slightly different binning
+      * merged all consolidations to one part of malloc proper
+	 (eliminating old malloc_find_space & malloc_clean_bin)
+      * Scan 2 returns chunks (not just 1)
+      * Propagate failure in realloc if malloc returns 0
+      * Add stuff to allow compilation on non-ANSI compilers
+	  from kpv@research.att.com
+
+    V2.5 Sat Aug  7 07:41:59 1993  Doug Lea  (dl at g.oswego.edu)
+      * removed potential for odd address access in prev_chunk
+      * removed dependency on getpagesize.h
+      * misc cosmetics and a bit more internal documentation
+      * anticosmetics: mangled names in macros to evade debugger strangeness
+      * tested on sparc, hp-700, dec-mips, rs6000
+	  with gcc & native cc (hp, dec only) allowing
+	  Detlefs & Zorn comparison study (in SIGPLAN Notices.)
+
+    Trial version Fri Aug 28 13:14:29 1992  Doug Lea  (dl at g.oswego.edu)
+      * Based loosely on libg++-1.2X malloc. (It retains some of the overall
+	 structure of old version,  but most details differ.)
+
+*/
+
+
diff --git a/compat/nedmalloc/nedmalloc.c b/compat/nedmalloc/nedmalloc.c
new file mode 100644
index 0000000..d9a17a8
--- /dev/null
+++ b/compat/nedmalloc/nedmalloc.c
@@ -0,0 +1,966 @@
+/* Alternative malloc implementation for multiple threads without
+lock contention based on dlmalloc. (C) 2005-2006 Niall Douglas
+
+Boost Software License - Version 1.0 - August 17th, 2003
+
+Permission is hereby granted, free of charge, to any person or organization
+obtaining a copy of the software and accompanying documentation covered by
+this license (the "Software") to use, reproduce, display, distribute,
+execute, and transmit the Software, and to prepare derivative works of the
+Software, and to permit third-parties to whom the Software is furnished to
+do so, all subject to the following:
+
+The copyright notices in the Software and this entire statement, including
+the above license grant, this restriction and the following disclaimer,
+must be included in all copies of the Software, in whole or in part, and
+all derivative works of the Software, unless such copies or derivative
+works are solely in the form of machine-executable object code generated by
+a source language processor.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE, TITLE AND NON-INFRINGEMENT. IN NO EVENT
+SHALL THE COPYRIGHT HOLDERS OR ANYONE DISTRIBUTING THE SOFTWARE BE LIABLE
+FOR ANY DAMAGES OR OTHER LIABILITY, WHETHER IN CONTRACT, TORT OR OTHERWISE,
+ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+DEALINGS IN THE SOFTWARE.
+*/
+
+#ifdef _MSC_VER
+/* Enable full aliasing on MSVC */
+/*#pragma optimize("a", on)*/
+#endif
+
+/*#define FULLSANITYCHECKS*/
+
+#include "nedmalloc.h"
+#if defined(WIN32)
+ #include <malloc.h>
+#endif
+#define MSPACES 1
+#define ONLY_MSPACES 1
+#ifndef USE_LOCKS
+ #define USE_LOCKS 1
+#endif
+#define FOOTERS 1           /* Need to enable footers so frees lock the right mspace */
+#undef DEBUG				/* dlmalloc wants DEBUG either 0 or 1 */
+#ifdef _DEBUG
+ #define DEBUG 1
+#else
+ #define DEBUG 0
+#endif
+#ifdef NDEBUG               /* Disable assert checking on release builds */
+ #undef DEBUG
+#endif
+/* The default of 64Kb means we spend too much time kernel-side */
+#ifndef DEFAULT_GRANULARITY
+#define DEFAULT_GRANULARITY (1*1024*1024)
+#endif
+/*#define USE_SPIN_LOCKS 0*/
+
+
+/*#define FORCEINLINE*/
+#include "malloc.c.h"
+#ifdef NDEBUG               /* Disable assert checking on release builds */
+ #undef DEBUG
+#endif
+
+/* The maximum concurrent threads in a pool possible */
+#ifndef MAXTHREADSINPOOL
+#define MAXTHREADSINPOOL 16
+#endif
+/* The maximum number of threadcaches which can be allocated */
+#ifndef THREADCACHEMAXCACHES
+#define THREADCACHEMAXCACHES 256
+#endif
+/* The maximum size to be allocated from the thread cache */
+#ifndef THREADCACHEMAX
+#define THREADCACHEMAX 8192
+#endif
+#if 0
+/* The number of cache entries for finer grained bins. This is (topbitpos(THREADCACHEMAX)-4)*2 */
+#define THREADCACHEMAXBINS ((13-4)*2)
+#else
+/* The number of cache entries. This is (topbitpos(THREADCACHEMAX)-4) */
+#define THREADCACHEMAXBINS (13-4)
+#endif
+/* Point at which the free space in a thread cache is garbage collected */
+#ifndef THREADCACHEMAXFREESPACE
+#define THREADCACHEMAXFREESPACE (512*1024)
+#endif
+
+
+#ifdef WIN32
+ #define TLSVAR			DWORD
+ #define TLSALLOC(k)	(*(k)=TlsAlloc(), TLS_OUT_OF_INDEXES==*(k))
+ #define TLSFREE(k)		(!TlsFree(k))
+ #define TLSGET(k)		TlsGetValue(k)
+ #define TLSSET(k, a)	(!TlsSetValue(k, a))
+ #ifdef DEBUG
+static LPVOID ChkedTlsGetValue(DWORD idx)
+{
+	LPVOID ret=TlsGetValue(idx);
+	assert(S_OK==GetLastError());
+	return ret;
+}
+  #undef TLSGET
+  #define TLSGET(k) ChkedTlsGetValue(k)
+ #endif
+#else
+ #define TLSVAR			pthread_key_t
+ #define TLSALLOC(k)	pthread_key_create(k, 0)
+ #define TLSFREE(k)		pthread_key_delete(k)
+ #define TLSGET(k)		pthread_getspecific(k)
+ #define TLSSET(k, a)	pthread_setspecific(k, a)
+#endif
+
+#if 0
+/* Only enable if testing with valgrind. Causes misoperation */
+#define mspace_malloc(p, s) malloc(s)
+#define mspace_realloc(p, m, s) realloc(m, s)
+#define mspace_calloc(p, n, s) calloc(n, s)
+#define mspace_free(p, m) free(m)
+#endif
+
+
+#if defined(__cplusplus)
+#if !defined(NO_NED_NAMESPACE)
+namespace nedalloc {
+#else
+extern "C" {
+#endif
+#endif
+
+size_t nedblksize(void *mem) THROWSPEC
+{
+#if 0
+	/* Only enable if testing with valgrind. Causes misoperation */
+	return THREADCACHEMAX;
+#else
+	if(mem)
+	{
+		mchunkptr p=mem2chunk(mem);
+		assert(cinuse(p));	/* If this fails, someone tried to free a block twice */
+		if(cinuse(p))
+			return chunksize(p)-overhead_for(p);
+	}
+	return 0;
+#endif
+}
+
+void nedsetvalue(void *v) THROWSPEC					{ nedpsetvalue(0, v); }
+void * nedmalloc(size_t size) THROWSPEC				{ return nedpmalloc(0, size); }
+void * nedcalloc(size_t no, size_t size) THROWSPEC	{ return nedpcalloc(0, no, size); }
+void * nedrealloc(void *mem, size_t size) THROWSPEC	{ return nedprealloc(0, mem, size); }
+void   nedfree(void *mem) THROWSPEC					{ nedpfree(0, mem); }
+void * nedmemalign(size_t alignment, size_t bytes) THROWSPEC { return nedpmemalign(0, alignment, bytes); }
+#if !NO_MALLINFO
+struct mallinfo nedmallinfo(void) THROWSPEC			{ return nedpmallinfo(0); }
+#endif
+int    nedmallopt(int parno, int value) THROWSPEC	{ return nedpmallopt(0, parno, value); }
+int    nedmalloc_trim(size_t pad) THROWSPEC			{ return nedpmalloc_trim(0, pad); }
+void   nedmalloc_stats() THROWSPEC					{ nedpmalloc_stats(0); }
+size_t nedmalloc_footprint() THROWSPEC				{ return nedpmalloc_footprint(0); }
+void **nedindependent_calloc(size_t elemsno, size_t elemsize, void **chunks) THROWSPEC	{ return nedpindependent_calloc(0, elemsno, elemsize, chunks); }
+void **nedindependent_comalloc(size_t elems, size_t *sizes, void **chunks) THROWSPEC	{ return nedpindependent_comalloc(0, elems, sizes, chunks); }
+
+struct threadcacheblk_t;
+typedef struct threadcacheblk_t threadcacheblk;
+struct threadcacheblk_t
+{	/* Keep less than 16 bytes on 32 bit systems and 32 bytes on 64 bit systems */
+#ifdef FULLSANITYCHECKS
+	unsigned int magic;
+#endif
+	unsigned int lastUsed, size;
+	threadcacheblk *next, *prev;
+};
+typedef struct threadcache_t
+{
+#ifdef FULLSANITYCHECKS
+	unsigned int magic1;
+#endif
+	int mymspace;						/* Last mspace entry this thread used */
+	long threadid;
+	unsigned int mallocs, frees, successes;
+	size_t freeInCache;					/* How much free space is stored in this cache */
+	threadcacheblk *bins[(THREADCACHEMAXBINS+1)*2];
+#ifdef FULLSANITYCHECKS
+	unsigned int magic2;
+#endif
+} threadcache;
+struct nedpool_t
+{
+	MLOCK_T mutex;
+	void *uservalue;
+	int threads;						/* Max entries in m to use */
+	threadcache *caches[THREADCACHEMAXCACHES];
+	TLSVAR mycache;						/* Thread cache for this thread. 0 for unset, negative for use mspace-1 directly, otherwise is cache-1 */
+	mstate m[MAXTHREADSINPOOL+1];		/* mspace entries for this pool */
+};
+static nedpool syspool;
+
+static FORCEINLINE unsigned int size2binidx(size_t _size) THROWSPEC
+{	/* 8=1000	16=10000	20=10100	24=11000	32=100000	48=110000	4096=1000000000000 */
+	unsigned int topbit, size=(unsigned int)(_size>>4);
+	/* 16=1		20=1	24=1	32=10	48=11	64=100	96=110	128=1000	4096=100000000 */
+
+#if defined(__GNUC__)
+	topbit = sizeof(size)*__CHAR_BIT__ - 1 - __builtin_clz(size);
+#elif defined(_MSC_VER) && _MSC_VER>=1300
+	{
+	    unsigned long bsrTopBit;
+
+	    _BitScanReverse(&bsrTopBit, size);
+
+	    topbit = bsrTopBit;
+	}
+#else
+#if 0
+	union {
+		unsigned asInt[2];
+		double asDouble;
+	};
+	int n;
+
+	asDouble = (double)size + 0.5;
+	topbit = (asInt[!FOX_BIGENDIAN] >> 20) - 1023;
+#else
+	{
+		unsigned int x=size;
+		x = x | (x >> 1);
+		x = x | (x >> 2);
+		x = x | (x >> 4);
+		x = x | (x >> 8);
+		x = x | (x >>16);
+		x = ~x;
+		x = x - ((x >> 1) & 0x55555555);
+		x = (x & 0x33333333) + ((x >> 2) & 0x33333333);
+		x = (x + (x >> 4)) & 0x0F0F0F0F;
+		x = x + (x << 8);
+		x = x + (x << 16);
+		topbit=31 - (x >> 24);
+	}
+#endif
+#endif
+	return topbit;
+}
+
+
+#ifdef FULLSANITYCHECKS
+static void tcsanitycheck(threadcacheblk **ptr) THROWSPEC
+{
+	assert((ptr[0] && ptr[1]) || (!ptr[0] && !ptr[1]));
+	if(ptr[0] && ptr[1])
+	{
+		assert(nedblksize(ptr[0])>=sizeof(threadcacheblk));
+		assert(nedblksize(ptr[1])>=sizeof(threadcacheblk));
+		assert(*(unsigned int *) "NEDN"==ptr[0]->magic);
+		assert(*(unsigned int *) "NEDN"==ptr[1]->magic);
+		assert(!ptr[0]->prev);
+		assert(!ptr[1]->next);
+		if(ptr[0]==ptr[1])
+		{
+			assert(!ptr[0]->next);
+			assert(!ptr[1]->prev);
+		}
+	}
+}
+static void tcfullsanitycheck(threadcache *tc) THROWSPEC
+{
+	threadcacheblk **tcbptr=tc->bins;
+	int n;
+	for(n=0; n<=THREADCACHEMAXBINS; n++, tcbptr+=2)
+	{
+		threadcacheblk *b, *ob=0;
+		tcsanitycheck(tcbptr);
+		for(b=tcbptr[0]; b; ob=b, b=b->next)
+		{
+			assert(*(unsigned int *) "NEDN"==b->magic);
+			assert(!ob || ob->next==b);
+			assert(!ob || b->prev==ob);
+		}
+	}
+}
+#endif
+
+static NOINLINE void RemoveCacheEntries(nedpool *p, threadcache *tc, unsigned int age) THROWSPEC
+{
+#ifdef FULLSANITYCHECKS
+	tcfullsanitycheck(tc);
+#endif
+	if(tc->freeInCache)
+	{
+		threadcacheblk **tcbptr=tc->bins;
+		int n;
+		for(n=0; n<=THREADCACHEMAXBINS; n++, tcbptr+=2)
+		{
+			threadcacheblk **tcb=tcbptr+1;		/* come from oldest end of list */
+			/*tcsanitycheck(tcbptr);*/
+			for(; *tcb && tc->frees-(*tcb)->lastUsed>=age; )
+			{
+				threadcacheblk *f=*tcb;
+				size_t blksize=f->size; /*nedblksize(f);*/
+				assert(blksize<=nedblksize(f));
+				assert(blksize);
+#ifdef FULLSANITYCHECKS
+				assert(*(unsigned int *) "NEDN"==(*tcb)->magic);
+#endif
+				*tcb=(*tcb)->prev;
+				if(*tcb)
+					(*tcb)->next=0;
+				else
+					*tcbptr=0;
+				tc->freeInCache-=blksize;
+				assert((long) tc->freeInCache>=0);
+				mspace_free(0, f);
+				/*tcsanitycheck(tcbptr);*/
+			}
+		}
+	}
+#ifdef FULLSANITYCHECKS
+	tcfullsanitycheck(tc);
+#endif
+}
+static void DestroyCaches(nedpool *p) THROWSPEC
+{
+	if(p->caches)
+	{
+		threadcache *tc;
+		int n;
+		for(n=0; n<THREADCACHEMAXCACHES; n++)
+		{
+			if((tc=p->caches[n]))
+			{
+				tc->frees++;
+				RemoveCacheEntries(p, tc, 0);
+				assert(!tc->freeInCache);
+				tc->mymspace=-1;
+				tc->threadid=0;
+				mspace_free(0, tc);
+				p->caches[n]=0;
+			}
+		}
+	}
+}
+
+static NOINLINE threadcache *AllocCache(nedpool *p) THROWSPEC
+{
+	threadcache *tc=0;
+	int n, end;
+	ACQUIRE_LOCK(&p->mutex);
+	for(n=0; n<THREADCACHEMAXCACHES && p->caches[n]; n++);
+	if(THREADCACHEMAXCACHES==n)
+	{	/* List exhausted, so disable for this thread */
+		RELEASE_LOCK(&p->mutex);
+		return 0;
+	}
+	tc=p->caches[n]=(threadcache *) mspace_calloc(p->m[0], 1, sizeof(threadcache));
+	if(!tc)
+	{
+		RELEASE_LOCK(&p->mutex);
+		return 0;
+	}
+#ifdef FULLSANITYCHECKS
+	tc->magic1=*(unsigned int *)"NEDMALC1";
+	tc->magic2=*(unsigned int *)"NEDMALC2";
+#endif
+	tc->threadid=(long)(size_t)CURRENT_THREAD;
+	for(end=0; p->m[end]; end++);
+	tc->mymspace=tc->threadid % end;
+	RELEASE_LOCK(&p->mutex);
+	if(TLSSET(p->mycache, (void *)(size_t)(n+1))) abort();
+	return tc;
+}
+
+static void *threadcache_malloc(nedpool *p, threadcache *tc, size_t *size) THROWSPEC
+{
+	void *ret=0;
+	unsigned int bestsize;
+	unsigned int idx=size2binidx(*size);
+	size_t blksize=0;
+	threadcacheblk *blk, **binsptr;
+#ifdef FULLSANITYCHECKS
+	tcfullsanitycheck(tc);
+#endif
+	/* Calculate best fit bin size */
+	bestsize=1<<(idx+4);
+#if 0
+	/* Finer grained bin fit */
+	idx<<=1;
+	if(*size>bestsize)
+	{
+		idx++;
+		bestsize+=bestsize>>1;
+	}
+	if(*size>bestsize)
+	{
+		idx++;
+		bestsize=1<<(4+(idx>>1));
+	}
+#else
+	if(*size>bestsize)
+	{
+		idx++;
+		bestsize<<=1;
+	}
+#endif
+	assert(bestsize>=*size);
+	if(*size<bestsize) *size=bestsize;
+	assert(*size<=THREADCACHEMAX);
+	assert(idx<=THREADCACHEMAXBINS);
+	binsptr=&tc->bins[idx*2];
+	/* Try to match close, but move up a bin if necessary */
+	blk=*binsptr;
+	if(!blk || blk->size<*size)
+	{	/* Bump it up a bin */
+		if(idx<THREADCACHEMAXBINS)
+		{
+			idx++;
+			binsptr+=2;
+			blk=*binsptr;
+		}
+	}
+	if(blk)
+	{
+		blksize=blk->size; /*nedblksize(blk);*/
+		assert(nedblksize(blk)>=blksize);
+		assert(blksize>=*size);
+		if(blk->next)
+			blk->next->prev=0;
+		*binsptr=blk->next;
+		if(!*binsptr)
+			binsptr[1]=0;
+#ifdef FULLSANITYCHECKS
+		blk->magic=0;
+#endif
+		assert(binsptr[0]!=blk && binsptr[1]!=blk);
+		assert(nedblksize(blk)>=sizeof(threadcacheblk) && nedblksize(blk)<=THREADCACHEMAX+CHUNK_OVERHEAD);
+		/*printf("malloc: %p, %p, %p, %lu\n", p, tc, blk, (long) size);*/
+		ret=(void *) blk;
+	}
+	++tc->mallocs;
+	if(ret)
+	{
+		assert(blksize>=*size);
+		++tc->successes;
+		tc->freeInCache-=blksize;
+		assert((long) tc->freeInCache>=0);
+	}
+#if defined(DEBUG) && 0
+	if(!(tc->mallocs & 0xfff))
+	{
+		printf("*** threadcache=%u, mallocs=%u (%f), free=%u (%f), freeInCache=%u\n", (unsigned int) tc->threadid, tc->mallocs,
+			(float) tc->successes/tc->mallocs, tc->frees, (float) tc->successes/tc->frees, (unsigned int) tc->freeInCache);
+	}
+#endif
+#ifdef FULLSANITYCHECKS
+	tcfullsanitycheck(tc);
+#endif
+	return ret;
+}
+static NOINLINE void ReleaseFreeInCache(nedpool *p, threadcache *tc, int mymspace) THROWSPEC
+{
+	unsigned int age=THREADCACHEMAXFREESPACE/8192;
+	/*ACQUIRE_LOCK(&p->m[mymspace]->mutex);*/
+	while(age && tc->freeInCache>=THREADCACHEMAXFREESPACE)
+	{
+		RemoveCacheEntries(p, tc, age);
+		/*printf("*** Removing cache entries older than %u (%u)\n", age, (unsigned int) tc->freeInCache);*/
+		age>>=1;
+	}
+	/*RELEASE_LOCK(&p->m[mymspace]->mutex);*/
+}
+static void threadcache_free(nedpool *p, threadcache *tc, int mymspace, void *mem, size_t size) THROWSPEC
+{
+	unsigned int bestsize;
+	unsigned int idx=size2binidx(size);
+	threadcacheblk **binsptr, *tck=(threadcacheblk *) mem;
+	assert(size>=sizeof(threadcacheblk) && size<=THREADCACHEMAX+CHUNK_OVERHEAD);
+#ifdef DEBUG
+	{	/* Make sure this is a valid memory block */
+	    mchunkptr p  = mem2chunk(mem);
+	    mstate fm = get_mstate_for(p);
+	    if (!ok_magic(fm)) {
+	      USAGE_ERROR_ACTION(fm, p);
+	      return;
+	    }
+	}
+#endif
+#ifdef FULLSANITYCHECKS
+	tcfullsanitycheck(tc);
+#endif
+	/* Calculate best fit bin size */
+	bestsize=1<<(idx+4);
+#if 0
+	/* Finer grained bin fit */
+	idx<<=1;
+	if(size>bestsize)
+	{
+		unsigned int biggerbestsize=bestsize+bestsize<<1;
+		if(size>=biggerbestsize)
+		{
+			idx++;
+			bestsize=biggerbestsize;
+		}
+	}
+#endif
+	if(bestsize!=size)	/* dlmalloc can round up, so we round down to preserve indexing */
+		size=bestsize;
+	binsptr=&tc->bins[idx*2];
+	assert(idx<=THREADCACHEMAXBINS);
+	if(tck==*binsptr)
+	{
+		fprintf(stderr, "Attempt to free already freed memory block %p - aborting!\n", tck);
+		abort();
+	}
+#ifdef FULLSANITYCHECKS
+	tck->magic=*(unsigned int *) "NEDN";
+#endif
+	tck->lastUsed=++tc->frees;
+	tck->size=(unsigned int) size;
+	tck->next=*binsptr;
+	tck->prev=0;
+	if(tck->next)
+		tck->next->prev=tck;
+	else
+		binsptr[1]=tck;
+	assert(!*binsptr || (*binsptr)->size==tck->size);
+	*binsptr=tck;
+	assert(tck==tc->bins[idx*2]);
+	assert(tc->bins[idx*2+1]==tck || binsptr[0]->next->prev==tck);
+	/*printf("free: %p, %p, %p, %lu\n", p, tc, mem, (long) size);*/
+	tc->freeInCache+=size;
+#ifdef FULLSANITYCHECKS
+	tcfullsanitycheck(tc);
+#endif
+#if 1
+	if(tc->freeInCache>=THREADCACHEMAXFREESPACE)
+		ReleaseFreeInCache(p, tc, mymspace);
+#endif
+}
+
+
+
+
+static NOINLINE int InitPool(nedpool *p, size_t capacity, int threads) THROWSPEC
+{	/* threads is -1 for system pool */
+	ensure_initialization();
+	ACQUIRE_MALLOC_GLOBAL_LOCK();
+	if(p->threads) goto done;
+	if(INITIAL_LOCK(&p->mutex)) goto err;
+	if(TLSALLOC(&p->mycache)) goto err;
+	if(!(p->m[0]=(mstate) create_mspace(capacity, 1))) goto err;
+	p->m[0]->extp=p;
+	p->threads=(threads<1 || threads>MAXTHREADSINPOOL) ? MAXTHREADSINPOOL : threads;
+done:
+	RELEASE_MALLOC_GLOBAL_LOCK();
+	return 1;
+err:
+	if(threads<0)
+		abort();			/* If you can't allocate for system pool, we're screwed */
+	DestroyCaches(p);
+	if(p->m[0])
+	{
+		destroy_mspace(p->m[0]);
+		p->m[0]=0;
+	}
+	if(p->mycache)
+	{
+		if(TLSFREE(p->mycache)) abort();
+		p->mycache=0;
+	}
+	RELEASE_MALLOC_GLOBAL_LOCK();
+	return 0;
+}
+static NOINLINE mstate FindMSpace(nedpool *p, threadcache *tc, int *lastUsed, size_t size) THROWSPEC
+{	/* Gets called when thread's last used mspace is in use. The strategy
+	is to run through the list of all available mspaces looking for an
+	unlocked one and if we fail, we create a new one so long as we don't
+	exceed p->threads */
+	int n, end;
+	for(n=end=*lastUsed+1; p->m[n]; end=++n)
+	{
+		if(TRY_LOCK(&p->m[n]->mutex)) goto found;
+	}
+	for(n=0; n<*lastUsed && p->m[n]; n++)
+	{
+		if(TRY_LOCK(&p->m[n]->mutex)) goto found;
+	}
+	if(end<p->threads)
+	{
+		mstate temp;
+		if(!(temp=(mstate) create_mspace(size, 1)))
+			goto badexit;
+		/* Now we're ready to modify the lists, we lock */
+		ACQUIRE_LOCK(&p->mutex);
+		while(p->m[end] && end<p->threads)
+			end++;
+		if(end>=p->threads)
+		{	/* Drat, must destroy it now */
+			RELEASE_LOCK(&p->mutex);
+			destroy_mspace((mspace) temp);
+			goto badexit;
+		}
+		/* We really want to make sure this goes into memory now but we
+		have to be careful of breaking aliasing rules, so write it twice */
+		*((volatile struct malloc_state **) &p->m[end])=p->m[end]=temp;
+		ACQUIRE_LOCK(&p->m[end]->mutex);
+		/*printf("Created mspace idx %d\n", end);*/
+		RELEASE_LOCK(&p->mutex);
+		n=end;
+		goto found;
+	}
+	/* Let it lock on the last one it used */
+badexit:
+	ACQUIRE_LOCK(&p->m[*lastUsed]->mutex);
+	return p->m[*lastUsed];
+found:
+	*lastUsed=n;
+	if(tc)
+		tc->mymspace=n;
+	else
+	{
+		if(TLSSET(p->mycache, (void *)(size_t)(-(n+1)))) abort();
+	}
+	return p->m[n];
+}
+
+nedpool *nedcreatepool(size_t capacity, int threads) THROWSPEC
+{
+	nedpool *ret;
+	if(!(ret=(nedpool *) nedpcalloc(0, 1, sizeof(nedpool)))) return 0;
+	if(!InitPool(ret, capacity, threads))
+	{
+		nedpfree(0, ret);
+		return 0;
+	}
+	return ret;
+}
+void neddestroypool(nedpool *p) THROWSPEC
+{
+	int n;
+	ACQUIRE_LOCK(&p->mutex);
+	DestroyCaches(p);
+	for(n=0; p->m[n]; n++)
+	{
+		destroy_mspace(p->m[n]);
+		p->m[n]=0;
+	}
+	RELEASE_LOCK(&p->mutex);
+	if(TLSFREE(p->mycache)) abort();
+	nedpfree(0, p);
+}
+
+void nedpsetvalue(nedpool *p, void *v) THROWSPEC
+{
+	if(!p) { p=&syspool; if(!syspool.threads) InitPool(&syspool, 0, -1); }
+	p->uservalue=v;
+}
+void *nedgetvalue(nedpool **p, void *mem) THROWSPEC
+{
+	nedpool *np=0;
+	mchunkptr mcp=mem2chunk(mem);
+	mstate fm;
+	if(!(is_aligned(chunk2mem(mcp))) && mcp->head != FENCEPOST_HEAD) return 0;
+	if(!cinuse(mcp)) return 0;
+	if(!next_pinuse(mcp)) return 0;
+	if(!is_mmapped(mcp) && !pinuse(mcp))
+	{
+		if(next_chunk(prev_chunk(mcp))!=mcp) return 0;
+	}
+	fm=get_mstate_for(mcp);
+	if(!ok_magic(fm)) return 0;
+	if(!ok_address(fm, mcp)) return 0;
+	if(!fm->extp) return 0;
+	np=(nedpool *) fm->extp;
+	if(p) *p=np;
+	return np->uservalue;
+}
+
+void neddisablethreadcache(nedpool *p) THROWSPEC
+{
+	int mycache;
+	if(!p)
+	{
+		p=&syspool;
+		if(!syspool.threads) InitPool(&syspool, 0, -1);
+	}
+	mycache=(int)(size_t) TLSGET(p->mycache);
+	if(!mycache)
+	{	/* Set to mspace 0 */
+		if(TLSSET(p->mycache, (void *)-1)) abort();
+	}
+	else if(mycache>0)
+	{	/* Set to last used mspace */
+		threadcache *tc=p->caches[mycache-1];
+#if defined(DEBUG)
+		printf("Threadcache utilisation: %lf%% in cache with %lf%% lost to other threads\n",
+			100.0*tc->successes/tc->mallocs, 100.0*((double) tc->mallocs-tc->frees)/tc->mallocs);
+#endif
+		if(TLSSET(p->mycache, (void *)(size_t)(-tc->mymspace))) abort();
+		tc->frees++;
+		RemoveCacheEntries(p, tc, 0);
+		assert(!tc->freeInCache);
+		tc->mymspace=-1;
+		tc->threadid=0;
+		mspace_free(0, p->caches[mycache-1]);
+		p->caches[mycache-1]=0;
+	}
+}
+
+#define GETMSPACE(m,p,tc,ms,s,action)           \
+  do                                            \
+  {                                             \
+    mstate m = GetMSpace((p),(tc),(ms),(s));    \
+    action;                                     \
+    RELEASE_LOCK(&m->mutex);                    \
+  } while (0)
+
+static FORCEINLINE mstate GetMSpace(nedpool *p, threadcache *tc, int mymspace, size_t size) THROWSPEC
+{	/* Returns a locked and ready for use mspace */
+	mstate m=p->m[mymspace];
+	assert(m);
+	if(!TRY_LOCK(&p->m[mymspace]->mutex)) m=FindMSpace(p, tc, &mymspace, size);\
+	/*assert(IS_LOCKED(&p->m[mymspace]->mutex));*/
+	return m;
+}
+static FORCEINLINE void GetThreadCache(nedpool **p, threadcache **tc, int *mymspace, size_t *size) THROWSPEC
+{
+	int mycache;
+	if(size && *size<sizeof(threadcacheblk)) *size=sizeof(threadcacheblk);
+	if(!*p)
+	{
+		*p=&syspool;
+		if(!syspool.threads) InitPool(&syspool, 0, -1);
+	}
+	mycache=(int)(size_t) TLSGET((*p)->mycache);
+	if(mycache>0)
+	{
+		*tc=(*p)->caches[mycache-1];
+		*mymspace=(*tc)->mymspace;
+	}
+	else if(!mycache)
+	{
+		*tc=AllocCache(*p);
+		if(!*tc)
+		{	/* Disable */
+			if(TLSSET((*p)->mycache, (void *)-1)) abort();
+			*mymspace=0;
+		}
+		else
+			*mymspace=(*tc)->mymspace;
+	}
+	else
+	{
+		*tc=0;
+		*mymspace=-mycache-1;
+	}
+	assert(*mymspace>=0);
+	assert((long)(size_t)CURRENT_THREAD==(*tc)->threadid);
+#ifdef FULLSANITYCHECKS
+	if(*tc)
+	{
+		if(*(unsigned int *)"NEDMALC1"!=(*tc)->magic1 || *(unsigned int *)"NEDMALC2"!=(*tc)->magic2)
+		{
+			abort();
+		}
+	}
+#endif
+}
+
+void * nedpmalloc(nedpool *p, size_t size) THROWSPEC
+{
+	void *ret=0;
+	threadcache *tc;
+	int mymspace;
+	GetThreadCache(&p, &tc, &mymspace, &size);
+#if THREADCACHEMAX
+	if(tc && size<=THREADCACHEMAX)
+	{	/* Use the thread cache */
+		ret=threadcache_malloc(p, tc, &size);
+	}
+#endif
+	if(!ret)
+	{	/* Use this thread's mspace */
+	GETMSPACE(m, p, tc, mymspace, size,
+		  ret=mspace_malloc(m, size));
+	}
+	return ret;
+}
+void * nedpcalloc(nedpool *p, size_t no, size_t size) THROWSPEC
+{
+	size_t rsize=size*no;
+	void *ret=0;
+	threadcache *tc;
+	int mymspace;
+	GetThreadCache(&p, &tc, &mymspace, &rsize);
+#if THREADCACHEMAX
+	if(tc && rsize<=THREADCACHEMAX)
+	{	/* Use the thread cache */
+		if((ret=threadcache_malloc(p, tc, &rsize)))
+			memset(ret, 0, rsize);
+	}
+#endif
+	if(!ret)
+	{	/* Use this thread's mspace */
+	GETMSPACE(m, p, tc, mymspace, rsize,
+		  ret=mspace_calloc(m, 1, rsize));
+	}
+	return ret;
+}
+void * nedprealloc(nedpool *p, void *mem, size_t size) THROWSPEC
+{
+	void *ret=0;
+	threadcache *tc;
+	int mymspace;
+	if(!mem) return nedpmalloc(p, size);
+	GetThreadCache(&p, &tc, &mymspace, &size);
+#if THREADCACHEMAX
+	if(tc && size && size<=THREADCACHEMAX)
+	{	/* Use the thread cache */
+		size_t memsize=nedblksize(mem);
+		assert(memsize);
+		if((ret=threadcache_malloc(p, tc, &size)))
+		{
+			memcpy(ret, mem, memsize<size ? memsize : size);
+			if(memsize<=THREADCACHEMAX)
+				threadcache_free(p, tc, mymspace, mem, memsize);
+			else
+				mspace_free(0, mem);
+		}
+	}
+#endif
+	if(!ret)
+	{	/* Reallocs always happen in the mspace they happened in, so skip
+		locking the preferred mspace for this thread */
+		ret=mspace_realloc(0, mem, size);
+	}
+	return ret;
+}
+void   nedpfree(nedpool *p, void *mem) THROWSPEC
+{	/* Frees always happen in the mspace they happened in, so skip
+	locking the preferred mspace for this thread */
+	threadcache *tc;
+	int mymspace;
+	size_t memsize;
+	assert(mem);
+	GetThreadCache(&p, &tc, &mymspace, 0);
+#if THREADCACHEMAX
+	memsize=nedblksize(mem);
+	assert(memsize);
+	if(mem && tc && memsize<=(THREADCACHEMAX+CHUNK_OVERHEAD))
+		threadcache_free(p, tc, mymspace, mem, memsize);
+	else
+#endif
+		mspace_free(0, mem);
+}
+void * nedpmemalign(nedpool *p, size_t alignment, size_t bytes) THROWSPEC
+{
+	void *ret;
+	threadcache *tc;
+	int mymspace;
+	GetThreadCache(&p, &tc, &mymspace, &bytes);
+	{	/* Use this thread's mspace */
+	GETMSPACE(m, p, tc, mymspace, bytes,
+		  ret=mspace_memalign(m, alignment, bytes));
+	}
+	return ret;
+}
+#if !NO_MALLINFO
+struct mallinfo nedpmallinfo(nedpool *p) THROWSPEC
+{
+	int n;
+	struct mallinfo ret={0};
+	if(!p) { p=&syspool; if(!syspool.threads) InitPool(&syspool, 0, -1); }
+	for(n=0; p->m[n]; n++)
+	{
+		struct mallinfo t=mspace_mallinfo(p->m[n]);
+		ret.arena+=t.arena;
+		ret.ordblks+=t.ordblks;
+		ret.hblkhd+=t.hblkhd;
+		ret.usmblks+=t.usmblks;
+		ret.uordblks+=t.uordblks;
+		ret.fordblks+=t.fordblks;
+		ret.keepcost+=t.keepcost;
+	}
+	return ret;
+}
+#endif
+int    nedpmallopt(nedpool *p, int parno, int value) THROWSPEC
+{
+	return mspace_mallopt(parno, value);
+}
+int    nedpmalloc_trim(nedpool *p, size_t pad) THROWSPEC
+{
+	int n, ret=0;
+	if(!p) { p=&syspool; if(!syspool.threads) InitPool(&syspool, 0, -1); }
+	for(n=0; p->m[n]; n++)
+	{
+		ret+=mspace_trim(p->m[n], pad);
+	}
+	return ret;
+}
+void   nedpmalloc_stats(nedpool *p) THROWSPEC
+{
+	int n;
+	if(!p) { p=&syspool; if(!syspool.threads) InitPool(&syspool, 0, -1); }
+	for(n=0; p->m[n]; n++)
+	{
+		mspace_malloc_stats(p->m[n]);
+	}
+}
+size_t nedpmalloc_footprint(nedpool *p) THROWSPEC
+{
+	size_t ret=0;
+	int n;
+	if(!p) { p=&syspool; if(!syspool.threads) InitPool(&syspool, 0, -1); }
+	for(n=0; p->m[n]; n++)
+	{
+		ret+=mspace_footprint(p->m[n]);
+	}
+	return ret;
+}
+void **nedpindependent_calloc(nedpool *p, size_t elemsno, size_t elemsize, void **chunks) THROWSPEC
+{
+	void **ret;
+	threadcache *tc;
+	int mymspace;
+	GetThreadCache(&p, &tc, &mymspace, &elemsize);
+    GETMSPACE(m, p, tc, mymspace, elemsno*elemsize,
+	      ret=mspace_independent_calloc(m, elemsno, elemsize, chunks));
+	return ret;
+}
+void **nedpindependent_comalloc(nedpool *p, size_t elems, size_t *sizes, void **chunks) THROWSPEC
+{
+	void **ret;
+	threadcache *tc;
+	int mymspace;
+    size_t i, *adjustedsizes=(size_t *) alloca(elems*sizeof(size_t));
+    if(!adjustedsizes) return 0;
+    for(i=0; i<elems; i++)
+	adjustedsizes[i]=sizes[i]<sizeof(threadcacheblk) ? sizeof(threadcacheblk) : sizes[i];
+	GetThreadCache(&p, &tc, &mymspace, 0);
+	GETMSPACE(m, p, tc, mymspace, 0,
+	      ret=mspace_independent_comalloc(m, elems, adjustedsizes, chunks));
+	return ret;
+}
+
+#ifdef OVERRIDE_STRDUP
+/*
+ * This implementation is purely there to override the libc version, to
+ * avoid a crash due to allocation and free on different 'heaps'.
+ */
+char *strdup(const char *s1)
+{
+	char *s2 = 0;
+	if (s1) {
+		s2 = malloc(strlen(s1) + 1);
+		strcpy(s2, s1);
+	}
+	return s2;
+}
+#endif
+
+#if defined(__cplusplus)
+}
+#endif
diff --git a/compat/nedmalloc/nedmalloc.h b/compat/nedmalloc/nedmalloc.h
new file mode 100644
index 0000000..f960e66
--- /dev/null
+++ b/compat/nedmalloc/nedmalloc.h
@@ -0,0 +1,180 @@
+/* nedalloc, an alternative malloc implementation for multiple threads without
+lock contention based on dlmalloc v2.8.3. (C) 2005 Niall Douglas
+
+Boost Software License - Version 1.0 - August 17th, 2003
+
+Permission is hereby granted, free of charge, to any person or organization
+obtaining a copy of the software and accompanying documentation covered by
+this license (the "Software") to use, reproduce, display, distribute,
+execute, and transmit the Software, and to prepare derivative works of the
+Software, and to permit third-parties to whom the Software is furnished to
+do so, all subject to the following:
+
+The copyright notices in the Software and this entire statement, including
+the above license grant, this restriction and the following disclaimer,
+must be included in all copies of the Software, in whole or in part, and
+all derivative works of the Software, unless such copies or derivative
+works are solely in the form of machine-executable object code generated by
+a source language processor.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE, TITLE AND NON-INFRINGEMENT. IN NO EVENT
+SHALL THE COPYRIGHT HOLDERS OR ANYONE DISTRIBUTING THE SOFTWARE BE LIABLE
+FOR ANY DAMAGES OR OTHER LIABILITY, WHETHER IN CONTRACT, TORT OR OTHERWISE,
+ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef NEDMALLOC_H
+#define NEDMALLOC_H
+
+
+/* See malloc.c.h for what each function does.
+
+REPLACE_SYSTEM_ALLOCATOR causes nedalloc's functions to be called malloc,
+free etc. instead of nedmalloc, nedfree etc. You may or may not want this.
+
+NO_NED_NAMESPACE prevents the functions from being defined in the nedalloc
+namespace when in C++ (uses the global namespace instead).
+
+EXTSPEC can be defined to be __declspec(dllexport) or
+__attribute__ ((visibility("default"))) or whatever you like. It defaults
+to extern.
+
+USE_LOCKS can be 2 if you want to define your own MLOCK_T, INITIAL_LOCK,
+ACQUIRE_LOCK, RELEASE_LOCK, TRY_LOCK, IS_LOCKED and NULL_LOCK_INITIALIZER.
+
+*/
+
+#include <stddef.h>   /* for size_t */
+
+#ifndef EXTSPEC
+ #define EXTSPEC extern
+#endif
+
+#if defined(_MSC_VER) && _MSC_VER>=1400
+ #define MALLOCATTR __declspec(restrict)
+#endif
+#ifdef __GNUC__
+ #define MALLOCATTR __attribute__ ((malloc))
+#endif
+#ifndef MALLOCATTR
+ #define MALLOCATTR
+#endif
+
+#ifdef REPLACE_SYSTEM_ALLOCATOR
+ #define nedmalloc               malloc
+ #define nedcalloc               calloc
+ #define nedrealloc              realloc
+ #define nedfree                 free
+ #define nedmemalign             memalign
+ #define nedmallinfo             mallinfo
+ #define nedmallopt              mallopt
+ #define nedmalloc_trim          malloc_trim
+ #define nedmalloc_stats         malloc_stats
+ #define nedmalloc_footprint     malloc_footprint
+ #define nedindependent_calloc   independent_calloc
+ #define nedindependent_comalloc independent_comalloc
+ #ifdef _MSC_VER
+  #define nedblksize              _msize
+ #endif
+#endif
+
+#ifndef NO_MALLINFO
+#define NO_MALLINFO 0
+#endif
+
+#if !NO_MALLINFO
+struct mallinfo;
+#endif
+
+#if defined(__cplusplus)
+ #if !defined(NO_NED_NAMESPACE)
+namespace nedalloc {
+ #else
+extern "C" {
+ #endif
+ #define THROWSPEC throw()
+#else
+ #define THROWSPEC
+#endif
+
+/* These are the global functions */
+
+/* Gets the usable size of an allocated block. Note this will always be bigger than what was
+asked for due to rounding etc.
+*/
+EXTSPEC size_t nedblksize(void *mem) THROWSPEC;
+
+EXTSPEC void nedsetvalue(void *v) THROWSPEC;
+
+EXTSPEC MALLOCATTR void * nedmalloc(size_t size) THROWSPEC;
+EXTSPEC MALLOCATTR void * nedcalloc(size_t no, size_t size) THROWSPEC;
+EXTSPEC MALLOCATTR void * nedrealloc(void *mem, size_t size) THROWSPEC;
+EXTSPEC void   nedfree(void *mem) THROWSPEC;
+EXTSPEC MALLOCATTR void * nedmemalign(size_t alignment, size_t bytes) THROWSPEC;
+#if !NO_MALLINFO
+EXTSPEC struct mallinfo nedmallinfo(void) THROWSPEC;
+#endif
+EXTSPEC int    nedmallopt(int parno, int value) THROWSPEC;
+EXTSPEC int    nedmalloc_trim(size_t pad) THROWSPEC;
+EXTSPEC void   nedmalloc_stats(void) THROWSPEC;
+EXTSPEC size_t nedmalloc_footprint(void) THROWSPEC;
+EXTSPEC MALLOCATTR void **nedindependent_calloc(size_t elemsno, size_t elemsize, void **chunks) THROWSPEC;
+EXTSPEC MALLOCATTR void **nedindependent_comalloc(size_t elems, size_t *sizes, void **chunks) THROWSPEC;
+
+/* These are the pool functions */
+struct nedpool_t;
+typedef struct nedpool_t nedpool;
+
+/* Creates a memory pool for use with the nedp* functions below.
+Capacity is how much to allocate immediately (if you know you'll be allocating a lot
+of memory very soon) which you can leave at zero. Threads specifies how many threads
+will *normally* be accessing the pool concurrently. Setting this to zero means it
+extends on demand, but be careful of this as it can rapidly consume system resources
+where bursts of concurrent threads use a pool at once.
+*/
+EXTSPEC MALLOCATTR nedpool *nedcreatepool(size_t capacity, int threads) THROWSPEC;
+
+/* Destroys a memory pool previously created by nedcreatepool().
+*/
+EXTSPEC void neddestroypool(nedpool *p) THROWSPEC;
+
+/* Sets a value to be associated with a pool. You can retrieve this value by passing
+any memory block allocated from that pool.
+*/
+EXTSPEC void nedpsetvalue(nedpool *p, void *v) THROWSPEC;
+/* Gets a previously set value using nedpsetvalue() or zero if memory is unknown.
+Optionally can also retrieve pool.
+*/
+EXTSPEC void *nedgetvalue(nedpool **p, void *mem) THROWSPEC;
+
+/* Disables the thread cache for the calling thread, returning any existing cache
+data to the central pool.
+*/
+EXTSPEC void neddisablethreadcache(nedpool *p) THROWSPEC;
+
+EXTSPEC MALLOCATTR void * nedpmalloc(nedpool *p, size_t size) THROWSPEC;
+EXTSPEC MALLOCATTR void * nedpcalloc(nedpool *p, size_t no, size_t size) THROWSPEC;
+EXTSPEC MALLOCATTR void * nedprealloc(nedpool *p, void *mem, size_t size) THROWSPEC;
+EXTSPEC void   nedpfree(nedpool *p, void *mem) THROWSPEC;
+EXTSPEC MALLOCATTR void * nedpmemalign(nedpool *p, size_t alignment, size_t bytes) THROWSPEC;
+#if !NO_MALLINFO
+EXTSPEC struct mallinfo nedpmallinfo(nedpool *p) THROWSPEC;
+#endif
+EXTSPEC int    nedpmallopt(nedpool *p, int parno, int value) THROWSPEC;
+EXTSPEC int    nedpmalloc_trim(nedpool *p, size_t pad) THROWSPEC;
+EXTSPEC void   nedpmalloc_stats(nedpool *p) THROWSPEC;
+EXTSPEC size_t nedpmalloc_footprint(nedpool *p) THROWSPEC;
+EXTSPEC MALLOCATTR void **nedpindependent_calloc(nedpool *p, size_t elemsno, size_t elemsize, void **chunks) THROWSPEC;
+EXTSPEC MALLOCATTR void **nedpindependent_comalloc(nedpool *p, size_t elems, size_t *sizes, void **chunks) THROWSPEC;
+
+#if defined(__cplusplus)
+}
+#endif
+
+#undef MALLOCATTR
+#undef EXTSPEC
+
+#endif
diff --git a/compat/pread.c b/compat/pread.c
new file mode 100644
index 0000000..978cac4
--- /dev/null
+++ b/compat/pread.c
@@ -0,0 +1,18 @@
+#include "../git-compat-util.h"
+
+ssize_t git_pread(int fd, void *buf, size_t count, off_t offset)
+{
+        off_t current_offset;
+        ssize_t rc;
+
+        current_offset = lseek(fd, 0, SEEK_CUR);
+
+        if (lseek(fd, offset, SEEK_SET) < 0)
+                return -1;
+
+        rc = read_in_full(fd, buf, count);
+
+        if (current_offset != lseek(fd, current_offset, SEEK_SET))
+                return -1;
+        return rc;
+}
diff --git a/compat/qsort.c b/compat/qsort.c
new file mode 100644
index 0000000..d93dce2
--- /dev/null
+++ b/compat/qsort.c
@@ -0,0 +1,62 @@
+#include "../git-compat-util.h"
+
+/*
+ * A merge sort implementation, simplified from the qsort implementation
+ * by Mike Haertel, which is a part of the GNU C Library.
+ */
+
+static void msort_with_tmp(void *b, size_t n, size_t s,
+			   int (*cmp)(const void *, const void *),
+			   char *t)
+{
+	char *tmp;
+	char *b1, *b2;
+	size_t n1, n2;
+
+	if (n <= 1)
+		return;
+
+	n1 = n / 2;
+	n2 = n - n1;
+	b1 = b;
+	b2 = (char *)b + (n1 * s);
+
+	msort_with_tmp(b1, n1, s, cmp, t);
+	msort_with_tmp(b2, n2, s, cmp, t);
+
+	tmp = t;
+
+	while (n1 > 0 && n2 > 0) {
+		if (cmp(b1, b2) <= 0) {
+			memcpy(tmp, b1, s);
+			tmp += s;
+			b1 += s;
+			--n1;
+		} else {
+			memcpy(tmp, b2, s);
+			tmp += s;
+			b2 += s;
+			--n2;
+		}
+	}
+	if (n1 > 0)
+		memcpy(tmp, b1, n1 * s);
+	memcpy(b, t, (n - n2) * s);
+}
+
+void git_qsort(void *b, size_t n, size_t s,
+	       int (*cmp)(const void *, const void *))
+{
+	const size_t size = n * s;
+	char buf[1024];
+
+	if (size < sizeof(buf)) {
+		/* The temporary array fits on the small on-stack buffer. */
+		msort_with_tmp(b, n, s, cmp, buf);
+	} else {
+		/* It's somewhat large, so malloc it.  */
+		char *tmp = malloc(size);
+		msort_with_tmp(b, n, s, cmp, tmp);
+		free(tmp);
+	}
+}
diff --git a/compat/regex/regex.c b/compat/regex/regex.c
new file mode 100644
index 0000000..556d8ab
--- /dev/null
+++ b/compat/regex/regex.c
@@ -0,0 +1,4924 @@
+/* Extended regular expression matching and search library,
+   version 0.12.
+   (Implements POSIX draft P10003.2/D11.2, except for
+   internationalization features.)
+
+   Copyright (C) 1993 Free Software Foundation, Inc.
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 2, or (at your option)
+   any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program; if not, write to the Free Software
+   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.  */
+
+/* AIX requires this to be the first thing in the file. */
+#if defined (_AIX) && !defined (REGEX_MALLOC)
+  #pragma alloca
+#endif
+
+#define _GNU_SOURCE
+
+/* We need this for `regex.h', and perhaps for the Emacs include files.  */
+#include <sys/types.h>
+
+/* We used to test for `BSTRING' here, but only GCC and Emacs define
+   `BSTRING', as far as I know, and neither of them use this code.  */
+#include <string.h>
+#ifndef bcmp
+#define bcmp(s1, s2, n)	memcmp ((s1), (s2), (n))
+#endif
+#ifndef bcopy
+#define bcopy(s, d, n)	memcpy ((d), (s), (n))
+#endif
+#ifndef bzero
+#define bzero(s, n)	memset ((s), 0, (n))
+#endif
+
+#include <stdlib.h>
+
+
+/* Define the syntax stuff for \<, \>, etc.  */
+
+/* This must be nonzero for the wordchar and notwordchar pattern
+   commands in re_match_2.  */
+#ifndef Sword
+#define Sword 1
+#endif
+
+#ifdef SYNTAX_TABLE
+
+extern char *re_syntax_table;
+
+#else /* not SYNTAX_TABLE */
+
+/* How many characters in the character set.  */
+#define CHAR_SET_SIZE 256
+
+static char re_syntax_table[CHAR_SET_SIZE];
+
+static void
+init_syntax_once ()
+{
+   register int c;
+   static int done = 0;
+
+   if (done)
+     return;
+
+   bzero (re_syntax_table, sizeof re_syntax_table);
+
+   for (c = 'a'; c <= 'z'; c++)
+     re_syntax_table[c] = Sword;
+
+   for (c = 'A'; c <= 'Z'; c++)
+     re_syntax_table[c] = Sword;
+
+   for (c = '0'; c <= '9'; c++)
+     re_syntax_table[c] = Sword;
+
+   re_syntax_table['_'] = Sword;
+
+   done = 1;
+}
+
+#endif /* not SYNTAX_TABLE */
+
+#define SYNTAX(c) re_syntax_table[c]
+
+
+/* Get the interface, including the syntax bits.  */
+#include "regex.h"
+
+/* isalpha etc. are used for the character classes.  */
+#include <ctype.h>
+
+#ifndef isascii
+#define isascii(c) 1
+#endif
+
+#ifdef isblank
+#define ISBLANK(c) (isascii (c) && isblank (c))
+#else
+#define ISBLANK(c) ((c) == ' ' || (c) == '\t')
+#endif
+#ifdef isgraph
+#define ISGRAPH(c) (isascii (c) && isgraph (c))
+#else
+#define ISGRAPH(c) (isascii (c) && isprint (c) && !isspace (c))
+#endif
+
+#define ISPRINT(c) (isascii (c) && isprint (c))
+#define ISDIGIT(c) (isascii (c) && isdigit (c))
+#define ISALNUM(c) (isascii (c) && isalnum (c))
+#define ISALPHA(c) (isascii (c) && isalpha (c))
+#define ISCNTRL(c) (isascii (c) && iscntrl (c))
+#define ISLOWER(c) (isascii (c) && islower (c))
+#define ISPUNCT(c) (isascii (c) && ispunct (c))
+#define ISSPACE(c) (isascii (c) && isspace (c))
+#define ISUPPER(c) (isascii (c) && isupper (c))
+#define ISXDIGIT(c) (isascii (c) && isxdigit (c))
+
+#ifndef NULL
+#define NULL 0
+#endif
+
+/* We remove any previous definition of `SIGN_EXTEND_CHAR',
+   since ours (we hope) works properly with all combinations of
+   machines, compilers, `char' and `unsigned char' argument types.
+   (Per Bothner suggested the basic approach.)  */
+#undef SIGN_EXTEND_CHAR
+#if __STDC__
+#define SIGN_EXTEND_CHAR(c) ((signed char) (c))
+#else  /* not __STDC__ */
+/* As in Harbison and Steele.  */
+#define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
+#endif
+
+/* Should we use malloc or alloca?  If REGEX_MALLOC is not defined, we
+   use `alloca' instead of `malloc'.  This is because using malloc in
+   re_search* or re_match* could cause memory leaks when C-g is used in
+   Emacs; also, malloc is slower and causes storage fragmentation.  On
+   the other hand, malloc is more portable, and easier to debug.
+
+   Because we sometimes use alloca, some routines have to be macros,
+   not functions -- `alloca'-allocated space disappears at the end of the
+   function it is called in.  */
+
+#ifdef REGEX_MALLOC
+
+#define REGEX_ALLOCATE malloc
+#define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
+
+#else /* not REGEX_MALLOC  */
+
+/* Emacs already defines alloca, sometimes.  */
+#ifndef alloca
+
+/* Make alloca work the best possible way.  */
+#ifdef __GNUC__
+#define alloca __builtin_alloca
+#else /* not __GNUC__ */
+#if HAVE_ALLOCA_H
+#include <alloca.h>
+#else /* not __GNUC__ or HAVE_ALLOCA_H */
+#ifndef _AIX /* Already did AIX, up at the top.  */
+char *alloca ();
+#endif /* not _AIX */
+#endif /* not HAVE_ALLOCA_H */
+#endif /* not __GNUC__ */
+
+#endif /* not alloca */
+
+#define REGEX_ALLOCATE alloca
+
+/* Assumes a `char *destination' variable.  */
+#define REGEX_REALLOCATE(source, osize, nsize)				\
+  (destination = (char *) alloca (nsize),				\
+   bcopy (source, destination, osize),					\
+   destination)
+
+#endif /* not REGEX_MALLOC */
+
+
+/* True if `size1' is non-NULL and PTR is pointing anywhere inside
+   `string1' or just past its end.  This works if PTR is NULL, which is
+   a good thing.  */
+#define FIRST_STRING_P(ptr) 					\
+  (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
+
+/* (Re)Allocate N items of type T using malloc, or fail.  */
+#define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
+#define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
+#define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
+
+#define BYTEWIDTH 8 /* In bits.  */
+
+#define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
+
+#define MAX(a, b) ((a) > (b) ? (a) : (b))
+#define MIN(a, b) ((a) < (b) ? (a) : (b))
+
+typedef char boolean;
+#define false 0
+#define true 1
+
+/* These are the command codes that appear in compiled regular
+   expressions.  Some opcodes are followed by argument bytes.  A
+   command code can specify any interpretation whatsoever for its
+   arguments.  Zero bytes may appear in the compiled regular expression.
+
+   The value of `exactn' is needed in search.c (search_buffer) in Emacs.
+   So regex.h defines a symbol `RE_EXACTN_VALUE' to be 1; the value of
+   `exactn' we use here must also be 1.  */
+
+typedef enum
+{
+  no_op = 0,
+
+	/* Followed by one byte giving n, then by n literal bytes.  */
+  exactn = 1,
+
+	/* Matches any (more or less) character.  */
+  anychar,
+
+	/* Matches any one char belonging to specified set.  First
+	   following byte is number of bitmap bytes.  Then come bytes
+	   for a bitmap saying which chars are in.  Bits in each byte
+	   are ordered low-bit-first.  A character is in the set if its
+	   bit is 1.  A character too large to have a bit in the map is
+	   automatically not in the set.  */
+  charset,
+
+	/* Same parameters as charset, but match any character that is
+	   not one of those specified.  */
+  charset_not,
+
+	/* Start remembering the text that is matched, for storing in a
+	   register.  Followed by one byte with the register number, in
+	   the range 0 to one less than the pattern buffer's re_nsub
+	   field.  Then followed by one byte with the number of groups
+	   inner to this one.  (This last has to be part of the
+	   start_memory only because we need it in the on_failure_jump
+	   of re_match_2.)  */
+  start_memory,
+
+	/* Stop remembering the text that is matched and store it in a
+	   memory register.  Followed by one byte with the register
+	   number, in the range 0 to one less than `re_nsub' in the
+	   pattern buffer, and one byte with the number of inner groups,
+	   just like `start_memory'.  (We need the number of inner
+	   groups here because we don't have any easy way of finding the
+	   corresponding start_memory when we're at a stop_memory.)  */
+  stop_memory,
+
+	/* Match a duplicate of something remembered. Followed by one
+	   byte containing the register number.  */
+  duplicate,
+
+	/* Fail unless at beginning of line.  */
+  begline,
+
+	/* Fail unless at end of line.  */
+  endline,
+
+	/* Succeeds if at beginning of buffer (if emacs) or at beginning
+	   of string to be matched (if not).  */
+  begbuf,
+
+	/* Analogously, for end of buffer/string.  */
+  endbuf,
+
+	/* Followed by two byte relative address to which to jump.  */
+  jump,
+
+	/* Same as jump, but marks the end of an alternative.  */
+  jump_past_alt,
+
+	/* Followed by two-byte relative address of place to resume at
+	   in case of failure.  */
+  on_failure_jump,
+
+	/* Like on_failure_jump, but pushes a placeholder instead of the
+	   current string position when executed.  */
+  on_failure_keep_string_jump,
+
+	/* Throw away latest failure point and then jump to following
+	   two-byte relative address.  */
+  pop_failure_jump,
+
+	/* Change to pop_failure_jump if know won't have to backtrack to
+	   match; otherwise change to jump.  This is used to jump
+	   back to the beginning of a repeat.  If what follows this jump
+	   clearly won't match what the repeat does, such that we can be
+	   sure that there is no use backtracking out of repetitions
+	   already matched, then we change it to a pop_failure_jump.
+	   Followed by two-byte address.  */
+  maybe_pop_jump,
+
+	/* Jump to following two-byte address, and push a dummy failure
+	   point. This failure point will be thrown away if an attempt
+	   is made to use it for a failure.  A `+' construct makes this
+	   before the first repeat.  Also used as an intermediary kind
+	   of jump when compiling an alternative.  */
+  dummy_failure_jump,
+
+	/* Push a dummy failure point and continue.  Used at the end of
+	   alternatives.  */
+  push_dummy_failure,
+
+	/* Followed by two-byte relative address and two-byte number n.
+	   After matching N times, jump to the address upon failure.  */
+  succeed_n,
+
+	/* Followed by two-byte relative address, and two-byte number n.
+	   Jump to the address N times, then fail.  */
+  jump_n,
+
+	/* Set the following two-byte relative address to the
+	   subsequent two-byte number.  The address *includes* the two
+	   bytes of number.  */
+  set_number_at,
+
+  wordchar,	/* Matches any word-constituent character.  */
+  notwordchar,	/* Matches any char that is not a word-constituent.  */
+
+  wordbeg,	/* Succeeds if at word beginning.  */
+  wordend,	/* Succeeds if at word end.  */
+
+  wordbound,	/* Succeeds if at a word boundary.  */
+  notwordbound	/* Succeeds if not at a word boundary.  */
+
+#ifdef emacs
+  ,before_dot,	/* Succeeds if before point.  */
+  at_dot,	/* Succeeds if at point.  */
+  after_dot,	/* Succeeds if after point.  */
+
+	/* Matches any character whose syntax is specified.  Followed by
+	   a byte which contains a syntax code, e.g., Sword.  */
+  syntaxspec,
+
+	/* Matches any character whose syntax is not that specified.  */
+  notsyntaxspec
+#endif /* emacs */
+} re_opcode_t;
+
+/* Common operations on the compiled pattern.  */
+
+/* Store NUMBER in two contiguous bytes starting at DESTINATION.  */
+
+#define STORE_NUMBER(destination, number)				\
+  do {									\
+    (destination)[0] = (number) & 0377;					\
+    (destination)[1] = (number) >> 8;					\
+  } while (0)
+
+/* Same as STORE_NUMBER, except increment DESTINATION to
+   the byte after where the number is stored.  Therefore, DESTINATION
+   must be an lvalue.  */
+
+#define STORE_NUMBER_AND_INCR(destination, number)			\
+  do {									\
+    STORE_NUMBER (destination, number);					\
+    (destination) += 2;							\
+  } while (0)
+
+/* Put into DESTINATION a number stored in two contiguous bytes starting
+   at SOURCE.  */
+
+#define EXTRACT_NUMBER(destination, source)				\
+  do {									\
+    (destination) = *(source) & 0377;					\
+    (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8;		\
+  } while (0)
+
+#ifdef DEBUG
+static void
+extract_number (dest, source)
+    int *dest;
+    unsigned char *source;
+{
+  int temp = SIGN_EXTEND_CHAR (*(source + 1));
+  *dest = *source & 0377;
+  *dest += temp << 8;
+}
+
+#ifndef EXTRACT_MACROS /* To debug the macros.  */
+#undef EXTRACT_NUMBER
+#define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
+#endif /* not EXTRACT_MACROS */
+
+#endif /* DEBUG */
+
+/* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
+   SOURCE must be an lvalue.  */
+
+#define EXTRACT_NUMBER_AND_INCR(destination, source)			\
+  do {									\
+    EXTRACT_NUMBER (destination, source);				\
+    (source) += 2; 							\
+  } while (0)
+
+#ifdef DEBUG
+static void
+extract_number_and_incr (destination, source)
+    int *destination;
+    unsigned char **source;
+{
+  extract_number (destination, *source);
+  *source += 2;
+}
+
+#ifndef EXTRACT_MACROS
+#undef EXTRACT_NUMBER_AND_INCR
+#define EXTRACT_NUMBER_AND_INCR(dest, src) \
+  extract_number_and_incr (&dest, &src)
+#endif /* not EXTRACT_MACROS */
+
+#endif /* DEBUG */
+
+/* If DEBUG is defined, Regex prints many voluminous messages about what
+   it is doing (if the variable `debug' is nonzero).  If linked with the
+   main program in `iregex.c', you can enter patterns and strings
+   interactively.  And if linked with the main program in `main.c' and
+   the other test files, you can run the already-written tests.  */
+
+#ifdef DEBUG
+
+/* We use standard I/O for debugging.  */
+#include <stdio.h>
+
+/* It is useful to test things that ``must'' be true when debugging.  */
+#include <assert.h>
+
+static int debug = 0;
+
+#define DEBUG_STATEMENT(e) e
+#define DEBUG_PRINT1(x) if (debug) printf (x)
+#define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
+#define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
+#define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
+#define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) 				\
+  if (debug) print_partial_compiled_pattern (s, e)
+#define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)			\
+  if (debug) print_double_string (w, s1, sz1, s2, sz2)
+
+
+extern void printchar ();
+
+/* Print the fastmap in human-readable form.  */
+
+void
+print_fastmap (fastmap)
+    char *fastmap;
+{
+  unsigned was_a_range = 0;
+  unsigned i = 0;
+
+  while (i < (1 << BYTEWIDTH))
+    {
+      if (fastmap[i++])
+	{
+	  was_a_range = 0;
+	  printchar (i - 1);
+	  while (i < (1 << BYTEWIDTH)  &&  fastmap[i])
+	    {
+	      was_a_range = 1;
+	      i++;
+	    }
+	  if (was_a_range)
+	    {
+	      printf ("-");
+	      printchar (i - 1);
+	    }
+	}
+    }
+  putchar ('\n');
+}
+
+
+/* Print a compiled pattern string in human-readable form, starting at
+   the START pointer into it and ending just before the pointer END.  */
+
+void
+print_partial_compiled_pattern (start, end)
+    unsigned char *start;
+    unsigned char *end;
+{
+  int mcnt, mcnt2;
+  unsigned char *p = start;
+  unsigned char *pend = end;
+
+  if (start == NULL)
+    {
+      printf ("(null)\n");
+      return;
+    }
+
+  /* Loop over pattern commands.  */
+  while (p < pend)
+    {
+      switch ((re_opcode_t) *p++)
+	{
+	case no_op:
+	  printf ("/no_op");
+	  break;
+
+	case exactn:
+	  mcnt = *p++;
+	  printf ("/exactn/%d", mcnt);
+	  do
+	    {
+	      putchar ('/');
+	      printchar (*p++);
+	    }
+	  while (--mcnt);
+	  break;
+
+	case start_memory:
+	  mcnt = *p++;
+	  printf ("/start_memory/%d/%d", mcnt, *p++);
+	  break;
+
+	case stop_memory:
+	  mcnt = *p++;
+	  printf ("/stop_memory/%d/%d", mcnt, *p++);
+	  break;
+
+	case duplicate:
+	  printf ("/duplicate/%d", *p++);
+	  break;
+
+	case anychar:
+	  printf ("/anychar");
+	  break;
+
+	case charset:
+	case charset_not:
+	  {
+	    register int c;
+
+	    printf ("/charset%s",
+		    (re_opcode_t) *(p - 1) == charset_not ? "_not" : "");
+
+	    assert (p + *p < pend);
+
+	    for (c = 0; c < *p; c++)
+	      {
+		unsigned bit;
+		unsigned char map_byte = p[1 + c];
+
+		putchar ('/');
+
+		for (bit = 0; bit < BYTEWIDTH; bit++)
+		  if (map_byte & (1 << bit))
+		    printchar (c * BYTEWIDTH + bit);
+	      }
+	    p += 1 + *p;
+	    break;
+	  }
+
+	case begline:
+	  printf ("/begline");
+	  break;
+
+	case endline:
+	  printf ("/endline");
+	  break;
+
+	case on_failure_jump:
+	  extract_number_and_incr (&mcnt, &p);
+	  printf ("/on_failure_jump/0/%d", mcnt);
+	  break;
+
+	case on_failure_keep_string_jump:
+	  extract_number_and_incr (&mcnt, &p);
+	  printf ("/on_failure_keep_string_jump/0/%d", mcnt);
+	  break;
+
+	case dummy_failure_jump:
+	  extract_number_and_incr (&mcnt, &p);
+	  printf ("/dummy_failure_jump/0/%d", mcnt);
+	  break;
+
+	case push_dummy_failure:
+	  printf ("/push_dummy_failure");
+	  break;
+
+	case maybe_pop_jump:
+	  extract_number_and_incr (&mcnt, &p);
+	  printf ("/maybe_pop_jump/0/%d", mcnt);
+	  break;
+
+	case pop_failure_jump:
+	  extract_number_and_incr (&mcnt, &p);
+	  printf ("/pop_failure_jump/0/%d", mcnt);
+	  break;
+
+	case jump_past_alt:
+	  extract_number_and_incr (&mcnt, &p);
+	  printf ("/jump_past_alt/0/%d", mcnt);
+	  break;
+
+	case jump:
+	  extract_number_and_incr (&mcnt, &p);
+	  printf ("/jump/0/%d", mcnt);
+	  break;
+
+	case succeed_n:
+	  extract_number_and_incr (&mcnt, &p);
+	  extract_number_and_incr (&mcnt2, &p);
+	  printf ("/succeed_n/0/%d/0/%d", mcnt, mcnt2);
+	  break;
+
+	case jump_n:
+	  extract_number_and_incr (&mcnt, &p);
+	  extract_number_and_incr (&mcnt2, &p);
+	  printf ("/jump_n/0/%d/0/%d", mcnt, mcnt2);
+	  break;
+
+	case set_number_at:
+	  extract_number_and_incr (&mcnt, &p);
+	  extract_number_and_incr (&mcnt2, &p);
+	  printf ("/set_number_at/0/%d/0/%d", mcnt, mcnt2);
+	  break;
+
+	case wordbound:
+	  printf ("/wordbound");
+	  break;
+
+	case notwordbound:
+	  printf ("/notwordbound");
+	  break;
+
+	case wordbeg:
+	  printf ("/wordbeg");
+	  break;
+
+	case wordend:
+	  printf ("/wordend");
+
+#ifdef emacs
+	case before_dot:
+	  printf ("/before_dot");
+	  break;
+
+	case at_dot:
+	  printf ("/at_dot");
+	  break;
+
+	case after_dot:
+	  printf ("/after_dot");
+	  break;
+
+	case syntaxspec:
+	  printf ("/syntaxspec");
+	  mcnt = *p++;
+	  printf ("/%d", mcnt);
+	  break;
+
+	case notsyntaxspec:
+	  printf ("/notsyntaxspec");
+	  mcnt = *p++;
+	  printf ("/%d", mcnt);
+	  break;
+#endif /* emacs */
+
+	case wordchar:
+	  printf ("/wordchar");
+	  break;
+
+	case notwordchar:
+	  printf ("/notwordchar");
+	  break;
+
+	case begbuf:
+	  printf ("/begbuf");
+	  break;
+
+	case endbuf:
+	  printf ("/endbuf");
+	  break;
+
+	default:
+	  printf ("?%d", *(p-1));
+	}
+    }
+  printf ("/\n");
+}
+
+
+void
+print_compiled_pattern (bufp)
+    struct re_pattern_buffer *bufp;
+{
+  unsigned char *buffer = bufp->buffer;
+
+  print_partial_compiled_pattern (buffer, buffer + bufp->used);
+  printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
+
+  if (bufp->fastmap_accurate && bufp->fastmap)
+    {
+      printf ("fastmap: ");
+      print_fastmap (bufp->fastmap);
+    }
+
+  printf ("re_nsub: %d\t", bufp->re_nsub);
+  printf ("regs_alloc: %d\t", bufp->regs_allocated);
+  printf ("can_be_null: %d\t", bufp->can_be_null);
+  printf ("newline_anchor: %d\n", bufp->newline_anchor);
+  printf ("no_sub: %d\t", bufp->no_sub);
+  printf ("not_bol: %d\t", bufp->not_bol);
+  printf ("not_eol: %d\t", bufp->not_eol);
+  printf ("syntax: %d\n", bufp->syntax);
+  /* Perhaps we should print the translate table?  */
+}
+
+
+void
+print_double_string (where, string1, size1, string2, size2)
+    const char *where;
+    const char *string1;
+    const char *string2;
+    int size1;
+    int size2;
+{
+  unsigned this_char;
+
+  if (where == NULL)
+    printf ("(null)");
+  else
+    {
+      if (FIRST_STRING_P (where))
+	{
+	  for (this_char = where - string1; this_char < size1; this_char++)
+	    printchar (string1[this_char]);
+
+	  where = string2;
+	}
+
+      for (this_char = where - string2; this_char < size2; this_char++)
+	printchar (string2[this_char]);
+    }
+}
+
+#else /* not DEBUG */
+
+#undef assert
+#define assert(e)
+
+#define DEBUG_STATEMENT(e)
+#define DEBUG_PRINT1(x)
+#define DEBUG_PRINT2(x1, x2)
+#define DEBUG_PRINT3(x1, x2, x3)
+#define DEBUG_PRINT4(x1, x2, x3, x4)
+#define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
+#define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
+
+#endif /* not DEBUG */
+
+/* Set by `re_set_syntax' to the current regexp syntax to recognize.  Can
+   also be assigned to arbitrarily: each pattern buffer stores its own
+   syntax, so it can be changed between regex compilations.  */
+reg_syntax_t re_syntax_options = RE_SYNTAX_EMACS;
+
+
+/* Specify the precise syntax of regexps for compilation.  This provides
+   for compatibility for various utilities which historically have
+   different, incompatible syntaxes.
+
+   The argument SYNTAX is a bit mask comprised of the various bits
+   defined in regex.h.  We return the old syntax.  */
+
+reg_syntax_t
+re_set_syntax (syntax)
+    reg_syntax_t syntax;
+{
+  reg_syntax_t ret = re_syntax_options;
+
+  re_syntax_options = syntax;
+  return ret;
+}
+
+/* This table gives an error message for each of the error codes listed
+   in regex.h.  Obviously the order here has to be same as there.  */
+
+static const char *re_error_msg[] =
+  { NULL,					/* REG_NOERROR */
+    "No match",					/* REG_NOMATCH */
+    "Invalid regular expression",		/* REG_BADPAT */
+    "Invalid collation character",		/* REG_ECOLLATE */
+    "Invalid character class name",		/* REG_ECTYPE */
+    "Trailing backslash",			/* REG_EESCAPE */
+    "Invalid back reference",			/* REG_ESUBREG */
+    "Unmatched [ or [^",			/* REG_EBRACK */
+    "Unmatched ( or \\(",			/* REG_EPAREN */
+    "Unmatched \\{",				/* REG_EBRACE */
+    "Invalid content of \\{\\}",		/* REG_BADBR */
+    "Invalid range end",			/* REG_ERANGE */
+    "Memory exhausted",				/* REG_ESPACE */
+    "Invalid preceding regular expression",	/* REG_BADRPT */
+    "Premature end of regular expression",	/* REG_EEND */
+    "Regular expression too big",		/* REG_ESIZE */
+    "Unmatched ) or \\)",			/* REG_ERPAREN */
+  };
+
+/* Subroutine declarations and macros for regex_compile.  */
+
+static void store_op1 (), store_op2 ();
+static void insert_op1 (), insert_op2 ();
+static boolean at_begline_loc_p (), at_endline_loc_p ();
+static boolean group_in_compile_stack ();
+static reg_errcode_t compile_range ();
+
+/* Fetch the next character in the uncompiled pattern---translating it
+   if necessary.  Also cast from a signed character in the constant
+   string passed to us by the user to an unsigned char that we can use
+   as an array index (in, e.g., `translate').  */
+#define PATFETCH(c)							\
+  do {if (p == pend) return REG_EEND;					\
+    c = (unsigned char) *p++;						\
+    if (translate) c = translate[c]; 					\
+  } while (0)
+
+/* Fetch the next character in the uncompiled pattern, with no
+   translation.  */
+#define PATFETCH_RAW(c)							\
+  do {if (p == pend) return REG_EEND;					\
+    c = (unsigned char) *p++; 						\
+  } while (0)
+
+/* Go backwards one character in the pattern.  */
+#define PATUNFETCH p--
+
+
+/* If `translate' is non-null, return translate[D], else just D.  We
+   cast the subscript to translate because some data is declared as
+   `char *', to avoid warnings when a string constant is passed.  But
+   when we use a character as a subscript we must make it unsigned.  */
+#define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
+
+
+/* Macros for outputting the compiled pattern into `buffer'.  */
+
+/* If the buffer isn't allocated when it comes in, use this.  */
+#define INIT_BUF_SIZE  32
+
+/* Make sure we have at least N more bytes of space in buffer.  */
+#define GET_BUFFER_SPACE(n)						\
+    while (b - bufp->buffer + (n) > bufp->allocated)			\
+      EXTEND_BUFFER ()
+
+/* Make sure we have one more byte of buffer space and then add C to it.  */
+#define BUF_PUSH(c)							\
+  do {									\
+    GET_BUFFER_SPACE (1);						\
+    *b++ = (unsigned char) (c);						\
+  } while (0)
+
+
+/* Ensure we have two more bytes of buffer space and then append C1 and C2.  */
+#define BUF_PUSH_2(c1, c2)						\
+  do {									\
+    GET_BUFFER_SPACE (2);						\
+    *b++ = (unsigned char) (c1);					\
+    *b++ = (unsigned char) (c2);					\
+  } while (0)
+
+
+/* As with BUF_PUSH_2, except for three bytes.  */
+#define BUF_PUSH_3(c1, c2, c3)						\
+  do {									\
+    GET_BUFFER_SPACE (3);						\
+    *b++ = (unsigned char) (c1);					\
+    *b++ = (unsigned char) (c2);					\
+    *b++ = (unsigned char) (c3);					\
+  } while (0)
+
+
+/* Store a jump with opcode OP at LOC to location TO.  We store a
+   relative address offset by the three bytes the jump itself occupies.  */
+#define STORE_JUMP(op, loc, to) \
+  store_op1 (op, loc, (to) - (loc) - 3)
+
+/* Likewise, for a two-argument jump.  */
+#define STORE_JUMP2(op, loc, to, arg) \
+  store_op2 (op, loc, (to) - (loc) - 3, arg)
+
+/* Like `STORE_JUMP', but for inserting.  Assume `b' is the buffer end.  */
+#define INSERT_JUMP(op, loc, to) \
+  insert_op1 (op, loc, (to) - (loc) - 3, b)
+
+/* Like `STORE_JUMP2', but for inserting.  Assume `b' is the buffer end.  */
+#define INSERT_JUMP2(op, loc, to, arg) \
+  insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
+
+
+/* This is not an arbitrary limit: the arguments which represent offsets
+   into the pattern are two bytes long.  So if 2^16 bytes turns out to
+   be too small, many things would have to change.  */
+#define MAX_BUF_SIZE (1L << 16)
+
+
+/* Extend the buffer by twice its current size via realloc and
+   reset the pointers that pointed into the old block to point to the
+   correct places in the new one.  If extending the buffer results in it
+   being larger than MAX_BUF_SIZE, then flag memory exhausted.  */
+#define EXTEND_BUFFER()							\
+  do { 									\
+    unsigned char *old_buffer = bufp->buffer;				\
+    if (bufp->allocated == MAX_BUF_SIZE) 				\
+      return REG_ESIZE;							\
+    bufp->allocated <<= 1;						\
+    if (bufp->allocated > MAX_BUF_SIZE)					\
+      bufp->allocated = MAX_BUF_SIZE; 					\
+    bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
+    if (bufp->buffer == NULL)						\
+      return REG_ESPACE;						\
+    /* If the buffer moved, move all the pointers into it.  */		\
+    if (old_buffer != bufp->buffer)					\
+      {									\
+	b = (b - old_buffer) + bufp->buffer;				\
+	begalt = (begalt - old_buffer) + bufp->buffer;			\
+	if (fixup_alt_jump)						\
+	  fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
+	if (laststart)							\
+	  laststart = (laststart - old_buffer) + bufp->buffer;		\
+	if (pending_exact)						\
+	  pending_exact = (pending_exact - old_buffer) + bufp->buffer;	\
+      }									\
+  } while (0)
+
+
+/* Since we have one byte reserved for the register number argument to
+   {start,stop}_memory, the maximum number of groups we can report
+   things about is what fits in that byte.  */
+#define MAX_REGNUM 255
+
+/* But patterns can have more than `MAX_REGNUM' registers.  We just
+   ignore the excess.  */
+typedef unsigned regnum_t;
+
+
+/* Macros for the compile stack.  */
+
+/* Since offsets can go either forwards or backwards, this type needs to
+   be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1.  */
+typedef int pattern_offset_t;
+
+typedef struct
+{
+  pattern_offset_t begalt_offset;
+  pattern_offset_t fixup_alt_jump;
+  pattern_offset_t inner_group_offset;
+  pattern_offset_t laststart_offset;
+  regnum_t regnum;
+} compile_stack_elt_t;
+
+
+typedef struct
+{
+  compile_stack_elt_t *stack;
+  unsigned size;
+  unsigned avail;			/* Offset of next open position.  */
+} compile_stack_type;
+
+
+#define INIT_COMPILE_STACK_SIZE 32
+
+#define COMPILE_STACK_EMPTY  (compile_stack.avail == 0)
+#define COMPILE_STACK_FULL  (compile_stack.avail == compile_stack.size)
+
+/* The next available element.  */
+#define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
+
+
+/* Set the bit for character C in a list.  */
+#define SET_LIST_BIT(c)                               \
+  (b[((unsigned char) (c)) / BYTEWIDTH]               \
+   |= 1 << (((unsigned char) c) % BYTEWIDTH))
+
+
+/* Get the next unsigned number in the uncompiled pattern.  */
+#define GET_UNSIGNED_NUMBER(num) 					\
+  { if (p != pend)							\
+     {									\
+       PATFETCH (c); 							\
+       while (ISDIGIT (c)) 						\
+	 { 								\
+	   if (num < 0)							\
+	      num = 0;							\
+	   num = num * 10 + c - '0'; 					\
+	   if (p == pend) 						\
+	      break; 							\
+	   PATFETCH (c);						\
+	 } 								\
+       } 								\
+    }
+
+#define CHAR_CLASS_MAX_LENGTH  6 /* Namely, `xdigit'.  */
+
+#define IS_CHAR_CLASS(string)						\
+   (STREQ (string, "alpha") || STREQ (string, "upper")			\
+    || STREQ (string, "lower") || STREQ (string, "digit")		\
+    || STREQ (string, "alnum") || STREQ (string, "xdigit")		\
+    || STREQ (string, "space") || STREQ (string, "print")		\
+    || STREQ (string, "punct") || STREQ (string, "graph")		\
+    || STREQ (string, "cntrl") || STREQ (string, "blank"))
+
+/* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
+   Returns one of error codes defined in `regex.h', or zero for success.
+
+   Assumes the `allocated' (and perhaps `buffer') and `translate'
+   fields are set in BUFP on entry.
+
+   If it succeeds, results are put in BUFP (if it returns an error, the
+   contents of BUFP are undefined):
+     `buffer' is the compiled pattern;
+     `syntax' is set to SYNTAX;
+     `used' is set to the length of the compiled pattern;
+     `fastmap_accurate' is zero;
+     `re_nsub' is the number of subexpressions in PATTERN;
+     `not_bol' and `not_eol' are zero;
+
+   The `fastmap' and `newline_anchor' fields are neither
+   examined nor set.  */
+
+static reg_errcode_t
+regex_compile (pattern, size, syntax, bufp)
+     const char *pattern;
+     int size;
+     reg_syntax_t syntax;
+     struct re_pattern_buffer *bufp;
+{
+  /* We fetch characters from PATTERN here.  Even though PATTERN is
+     `char *' (i.e., signed), we declare these variables as unsigned, so
+     they can be reliably used as array indices.  */
+  register unsigned char c, c1;
+
+  /* A random temporary spot in PATTERN.  */
+  const char *p1;
+
+  /* Points to the end of the buffer, where we should append.  */
+  register unsigned char *b;
+
+  /* Keeps track of unclosed groups.  */
+  compile_stack_type compile_stack;
+
+  /* Points to the current (ending) position in the pattern.  */
+  const char *p = pattern;
+  const char *pend = pattern + size;
+
+  /* How to translate the characters in the pattern.  */
+  char *translate = bufp->translate;
+
+  /* Address of the count-byte of the most recently inserted `exactn'
+     command.  This makes it possible to tell if a new exact-match
+     character can be added to that command or if the character requires
+     a new `exactn' command.  */
+  unsigned char *pending_exact = 0;
+
+  /* Address of start of the most recently finished expression.
+     This tells, e.g., postfix * where to find the start of its
+     operand.  Reset at the beginning of groups and alternatives.  */
+  unsigned char *laststart = 0;
+
+  /* Address of beginning of regexp, or inside of last group.  */
+  unsigned char *begalt;
+
+  /* Place in the uncompiled pattern (i.e., the {) to
+     which to go back if the interval is invalid.  */
+  const char *beg_interval;
+
+  /* Address of the place where a forward jump should go to the end of
+     the containing expression.  Each alternative of an `or' -- except the
+     last -- ends with a forward jump of this sort.  */
+  unsigned char *fixup_alt_jump = 0;
+
+  /* Counts open-groups as they are encountered.  Remembered for the
+     matching close-group on the compile stack, so the same register
+     number is put in the stop_memory as the start_memory.  */
+  regnum_t regnum = 0;
+
+#ifdef DEBUG
+  DEBUG_PRINT1 ("\nCompiling pattern: ");
+  if (debug)
+    {
+      unsigned debug_count;
+
+      for (debug_count = 0; debug_count < size; debug_count++)
+	printchar (pattern[debug_count]);
+      putchar ('\n');
+    }
+#endif /* DEBUG */
+
+  /* Initialize the compile stack.  */
+  compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
+  if (compile_stack.stack == NULL)
+    return REG_ESPACE;
+
+  compile_stack.size = INIT_COMPILE_STACK_SIZE;
+  compile_stack.avail = 0;
+
+  /* Initialize the pattern buffer.  */
+  bufp->syntax = syntax;
+  bufp->fastmap_accurate = 0;
+  bufp->not_bol = bufp->not_eol = 0;
+
+  /* Set `used' to zero, so that if we return an error, the pattern
+     printer (for debugging) will think there's no pattern.  We reset it
+     at the end.  */
+  bufp->used = 0;
+
+  /* Always count groups, whether or not bufp->no_sub is set.  */
+  bufp->re_nsub = 0;
+
+#if !defined (emacs) && !defined (SYNTAX_TABLE)
+  /* Initialize the syntax table.  */
+   init_syntax_once ();
+#endif
+
+  if (bufp->allocated == 0)
+    {
+      if (bufp->buffer)
+	{ /* If zero allocated, but buffer is non-null, try to realloc
+	     enough space.  This loses if buffer's address is bogus, but
+	     that is the user's responsibility.  */
+	  RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
+	}
+      else
+	{ /* Caller did not allocate a buffer.  Do it for them.  */
+	  bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
+	}
+      if (!bufp->buffer) return REG_ESPACE;
+
+      bufp->allocated = INIT_BUF_SIZE;
+    }
+
+  begalt = b = bufp->buffer;
+
+  /* Loop through the uncompiled pattern until we're at the end.  */
+  while (p != pend)
+    {
+      PATFETCH (c);
+
+      switch (c)
+	{
+	case '^':
+	  {
+	    if (   /* If at start of pattern, it's an operator.  */
+		   p == pattern + 1
+		   /* If context independent, it's an operator.  */
+		|| syntax & RE_CONTEXT_INDEP_ANCHORS
+		   /* Otherwise, depends on what's come before.  */
+		|| at_begline_loc_p (pattern, p, syntax))
+	      BUF_PUSH (begline);
+	    else
+	      goto normal_char;
+	  }
+	  break;
+
+
+	case '$':
+	  {
+	    if (   /* If at end of pattern, it's an operator.  */
+		   p == pend
+		   /* If context independent, it's an operator.  */
+		|| syntax & RE_CONTEXT_INDEP_ANCHORS
+		   /* Otherwise, depends on what's next.  */
+		|| at_endline_loc_p (p, pend, syntax))
+	       BUF_PUSH (endline);
+	     else
+	       goto normal_char;
+	   }
+	   break;
+
+
+	case '+':
+	case '?':
+	  if ((syntax & RE_BK_PLUS_QM)
+	      || (syntax & RE_LIMITED_OPS))
+	    goto normal_char;
+	handle_plus:
+	case '*':
+	  /* If there is no previous pattern... */
+	  if (!laststart)
+	    {
+	      if (syntax & RE_CONTEXT_INVALID_OPS)
+		return REG_BADRPT;
+	      else if (!(syntax & RE_CONTEXT_INDEP_OPS))
+		goto normal_char;
+	    }
+
+	  {
+	    /* Are we optimizing this jump?  */
+	    boolean keep_string_p = false;
+
+	    /* 1 means zero (many) matches is allowed.  */
+	    char zero_times_ok = 0, many_times_ok = 0;
+
+	    /* If there is a sequence of repetition chars, collapse it
+	       down to just one (the right one).  We can't combine
+	       interval operators with these because of, e.g., `a{2}*',
+	       which should only match an even number of `a's.  */
+
+	    for (;;)
+	      {
+		zero_times_ok |= c != '+';
+		many_times_ok |= c != '?';
+
+		if (p == pend)
+		  break;
+
+		PATFETCH (c);
+
+		if (c == '*'
+		    || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
+		  ;
+
+		else if (syntax & RE_BK_PLUS_QM  &&  c == '\\')
+		  {
+		    if (p == pend) return REG_EESCAPE;
+
+		    PATFETCH (c1);
+		    if (!(c1 == '+' || c1 == '?'))
+		      {
+			PATUNFETCH;
+			PATUNFETCH;
+			break;
+		      }
+
+		    c = c1;
+		  }
+		else
+		  {
+		    PATUNFETCH;
+		    break;
+		  }
+
+		/* If we get here, we found another repeat character.  */
+	       }
+
+	    /* Star, etc. applied to an empty pattern is equivalent
+	       to an empty pattern.  */
+	    if (!laststart)
+	      break;
+
+	    /* Now we know whether or not zero matches is allowed
+	       and also whether or not two or more matches is allowed.  */
+	    if (many_times_ok)
+	      { /* More than one repetition is allowed, so put in at the
+		   end a backward relative jump from `b' to before the next
+		   jump we're going to put in below (which jumps from
+		   laststart to after this jump).
+
+		   But if we are at the `*' in the exact sequence `.*\n',
+		   insert an unconditional jump backwards to the .,
+		   instead of the beginning of the loop.  This way we only
+		   push a failure point once, instead of every time
+		   through the loop.  */
+		assert (p - 1 > pattern);
+
+		/* Allocate the space for the jump.  */
+		GET_BUFFER_SPACE (3);
+
+		/* We know we are not at the first character of the pattern,
+		   because laststart was nonzero.  And we've already
+		   incremented `p', by the way, to be the character after
+		   the `*'.  Do we have to do something analogous here
+		   for null bytes, because of RE_DOT_NOT_NULL?  */
+		if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
+		    && zero_times_ok
+		    && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
+		    && !(syntax & RE_DOT_NEWLINE))
+		  { /* We have .*\n.  */
+		    STORE_JUMP (jump, b, laststart);
+		    keep_string_p = true;
+		  }
+		else
+		  /* Anything else.  */
+		  STORE_JUMP (maybe_pop_jump, b, laststart - 3);
+
+		/* We've added more stuff to the buffer.  */
+		b += 3;
+	      }
+
+	    /* On failure, jump from laststart to b + 3, which will be the
+	       end of the buffer after this jump is inserted.  */
+	    GET_BUFFER_SPACE (3);
+	    INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
+				       : on_failure_jump,
+			 laststart, b + 3);
+	    pending_exact = 0;
+	    b += 3;
+
+	    if (!zero_times_ok)
+	      {
+		/* At least one repetition is required, so insert a
+		   `dummy_failure_jump' before the initial
+		   `on_failure_jump' instruction of the loop. This
+		   effects a skip over that instruction the first time
+		   we hit that loop.  */
+		GET_BUFFER_SPACE (3);
+		INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
+		b += 3;
+	      }
+	    }
+	  break;
+
+
+	case '.':
+	  laststart = b;
+	  BUF_PUSH (anychar);
+	  break;
+
+
+	case '[':
+	  {
+	    boolean had_char_class = false;
+
+	    if (p == pend) return REG_EBRACK;
+
+	    /* Ensure that we have enough space to push a charset: the
+	       opcode, the length count, and the bitset; 34 bytes in all.  */
+	    GET_BUFFER_SPACE (34);
+
+	    laststart = b;
+
+	    /* We test `*p == '^' twice, instead of using an if
+	       statement, so we only need one BUF_PUSH.  */
+	    BUF_PUSH (*p == '^' ? charset_not : charset);
+	    if (*p == '^')
+	      p++;
+
+	    /* Remember the first position in the bracket expression.  */
+	    p1 = p;
+
+	    /* Push the number of bytes in the bitmap.  */
+	    BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
+
+	    /* Clear the whole map.  */
+	    bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
+
+	    /* charset_not matches newline according to a syntax bit.  */
+	    if ((re_opcode_t) b[-2] == charset_not
+		&& (syntax & RE_HAT_LISTS_NOT_NEWLINE))
+	      SET_LIST_BIT ('\n');
+
+	    /* Read in characters and ranges, setting map bits.  */
+	    for (;;)
+	      {
+		if (p == pend) return REG_EBRACK;
+
+		PATFETCH (c);
+
+		/* \ might escape characters inside [...] and [^...].  */
+		if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
+		  {
+		    if (p == pend) return REG_EESCAPE;
+
+		    PATFETCH (c1);
+		    SET_LIST_BIT (c1);
+		    continue;
+		  }
+
+		/* Could be the end of the bracket expression.  If it's
+		   not (i.e., when the bracket expression is `[]' so
+		   far), the ']' character bit gets set way below.  */
+		if (c == ']' && p != p1 + 1)
+		  break;
+
+		/* Look ahead to see if it's a range when the last thing
+		   was a character class.  */
+		if (had_char_class && c == '-' && *p != ']')
+		  return REG_ERANGE;
+
+		/* Look ahead to see if it's a range when the last thing
+		   was a character: if this is a hyphen not at the
+		   beginning or the end of a list, then it's the range
+		   operator.  */
+		if (c == '-'
+		    && !(p - 2 >= pattern && p[-2] == '[')
+		    && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
+		    && *p != ']')
+		  {
+		    reg_errcode_t ret
+		      = compile_range (&p, pend, translate, syntax, b);
+		    if (ret != REG_NOERROR) return ret;
+		  }
+
+		else if (p[0] == '-' && p[1] != ']')
+		  { /* This handles ranges made up of characters only.  */
+		    reg_errcode_t ret;
+
+		    /* Move past the `-'.  */
+		    PATFETCH (c1);
+
+		    ret = compile_range (&p, pend, translate, syntax, b);
+		    if (ret != REG_NOERROR) return ret;
+		  }
+
+		/* See if we're at the beginning of a possible character
+		   class.  */
+
+		else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
+		  { /* Leave room for the null.  */
+		    char str[CHAR_CLASS_MAX_LENGTH + 1];
+
+		    PATFETCH (c);
+		    c1 = 0;
+
+		    /* If pattern is `[[:'.  */
+		    if (p == pend) return REG_EBRACK;
+
+		    for (;;)
+		      {
+			PATFETCH (c);
+			if (c == ':' || c == ']' || p == pend
+			    || c1 == CHAR_CLASS_MAX_LENGTH)
+			  break;
+			str[c1++] = c;
+		      }
+		    str[c1] = '\0';
+
+		    /* If isn't a word bracketed by `[:' and:`]':
+		       undo the ending character, the letters, and leave
+		       the leading `:' and `[' (but set bits for them).  */
+		    if (c == ':' && *p == ']')
+		      {
+			int ch;
+			boolean is_alnum = STREQ (str, "alnum");
+			boolean is_alpha = STREQ (str, "alpha");
+			boolean is_blank = STREQ (str, "blank");
+			boolean is_cntrl = STREQ (str, "cntrl");
+			boolean is_digit = STREQ (str, "digit");
+			boolean is_graph = STREQ (str, "graph");
+			boolean is_lower = STREQ (str, "lower");
+			boolean is_print = STREQ (str, "print");
+			boolean is_punct = STREQ (str, "punct");
+			boolean is_space = STREQ (str, "space");
+			boolean is_upper = STREQ (str, "upper");
+			boolean is_xdigit = STREQ (str, "xdigit");
+
+			if (!IS_CHAR_CLASS (str)) return REG_ECTYPE;
+
+			/* Throw away the ] at the end of the character
+			   class.  */
+			PATFETCH (c);
+
+			if (p == pend) return REG_EBRACK;
+
+			for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
+			  {
+			    if (   (is_alnum  && ISALNUM (ch))
+				|| (is_alpha  && ISALPHA (ch))
+				|| (is_blank  && ISBLANK (ch))
+				|| (is_cntrl  && ISCNTRL (ch))
+				|| (is_digit  && ISDIGIT (ch))
+				|| (is_graph  && ISGRAPH (ch))
+				|| (is_lower  && ISLOWER (ch))
+				|| (is_print  && ISPRINT (ch))
+				|| (is_punct  && ISPUNCT (ch))
+				|| (is_space  && ISSPACE (ch))
+				|| (is_upper  && ISUPPER (ch))
+				|| (is_xdigit && ISXDIGIT (ch)))
+			    SET_LIST_BIT (ch);
+			  }
+			had_char_class = true;
+		      }
+		    else
+		      {
+			c1++;
+			while (c1--)
+			  PATUNFETCH;
+			SET_LIST_BIT ('[');
+			SET_LIST_BIT (':');
+			had_char_class = false;
+		      }
+		  }
+		else
+		  {
+		    had_char_class = false;
+		    SET_LIST_BIT (c);
+		  }
+	      }
+
+	    /* Discard any (non)matching list bytes that are all 0 at the
+	       end of the map.  Decrease the map-length byte too.  */
+	    while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
+	      b[-1]--;
+	    b += b[-1];
+	  }
+	  break;
+
+
+	case '(':
+	  if (syntax & RE_NO_BK_PARENS)
+	    goto handle_open;
+	  else
+	    goto normal_char;
+
+
+	case ')':
+	  if (syntax & RE_NO_BK_PARENS)
+	    goto handle_close;
+	  else
+	    goto normal_char;
+
+
+	case '\n':
+	  if (syntax & RE_NEWLINE_ALT)
+	    goto handle_alt;
+	  else
+	    goto normal_char;
+
+
+	case '|':
+	  if (syntax & RE_NO_BK_VBAR)
+	    goto handle_alt;
+	  else
+	    goto normal_char;
+
+
+	case '{':
+	   if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
+	     goto handle_interval;
+	   else
+	     goto normal_char;
+
+
+	case '\\':
+	  if (p == pend) return REG_EESCAPE;
+
+	  /* Do not translate the character after the \, so that we can
+	     distinguish, e.g., \B from \b, even if we normally would
+	     translate, e.g., B to b.  */
+	  PATFETCH_RAW (c);
+
+	  switch (c)
+	    {
+	    case '(':
+	      if (syntax & RE_NO_BK_PARENS)
+		goto normal_backslash;
+
+	    handle_open:
+	      bufp->re_nsub++;
+	      regnum++;
+
+	      if (COMPILE_STACK_FULL)
+		{
+		  RETALLOC (compile_stack.stack, compile_stack.size << 1,
+			    compile_stack_elt_t);
+		  if (compile_stack.stack == NULL) return REG_ESPACE;
+
+		  compile_stack.size <<= 1;
+		}
+
+	      /* These are the values to restore when we hit end of this
+		 group.  They are all relative offsets, so that if the
+		 whole pattern moves because of realloc, they will still
+		 be valid.  */
+	      COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
+	      COMPILE_STACK_TOP.fixup_alt_jump
+		= fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
+	      COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
+	      COMPILE_STACK_TOP.regnum = regnum;
+
+	      /* We will eventually replace the 0 with the number of
+		 groups inner to this one.  But do not push a
+		 start_memory for groups beyond the last one we can
+		 represent in the compiled pattern.  */
+	      if (regnum <= MAX_REGNUM)
+		{
+		  COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
+		  BUF_PUSH_3 (start_memory, regnum, 0);
+		}
+
+	      compile_stack.avail++;
+
+	      fixup_alt_jump = 0;
+	      laststart = 0;
+	      begalt = b;
+	      /* If we've reached MAX_REGNUM groups, then this open
+		 won't actually generate any code, so we'll have to
+		 clear pending_exact explicitly.  */
+	      pending_exact = 0;
+	      break;
+
+
+	    case ')':
+	      if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
+
+	      if (COMPILE_STACK_EMPTY)
+	      {
+		if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
+		  goto normal_backslash;
+		else
+		  return REG_ERPAREN;
+	      }
+
+	    handle_close:
+	      if (fixup_alt_jump)
+		{ /* Push a dummy failure point at the end of the
+		     alternative for a possible future
+		     `pop_failure_jump' to pop.  See comments at
+		     `push_dummy_failure' in `re_match_2'.  */
+		  BUF_PUSH (push_dummy_failure);
+
+		  /* We allocated space for this jump when we assigned
+		     to `fixup_alt_jump', in the `handle_alt' case below.  */
+		  STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
+		}
+
+	      /* See similar code for backslashed left paren above.  */
+	      if (COMPILE_STACK_EMPTY)
+	      {
+		if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
+		  goto normal_char;
+		else
+		  return REG_ERPAREN;
+	      }
+
+	      /* Since we just checked for an empty stack above, this
+		 ``can't happen''.  */
+	      assert (compile_stack.avail != 0);
+	      {
+		/* We don't just want to restore into `regnum', because
+		   later groups should continue to be numbered higher,
+		   as in `(ab)c(de)' -- the second group is #2.  */
+		regnum_t this_group_regnum;
+
+		compile_stack.avail--;
+		begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
+		fixup_alt_jump
+		  = COMPILE_STACK_TOP.fixup_alt_jump
+		    ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
+		    : 0;
+		laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
+		this_group_regnum = COMPILE_STACK_TOP.regnum;
+		/* If we've reached MAX_REGNUM groups, then this open
+		   won't actually generate any code, so we'll have to
+		   clear pending_exact explicitly.  */
+		pending_exact = 0;
+
+		/* We're at the end of the group, so now we know how many
+		   groups were inside this one.  */
+		if (this_group_regnum <= MAX_REGNUM)
+		  {
+		    unsigned char *inner_group_loc
+		      = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
+
+		    *inner_group_loc = regnum - this_group_regnum;
+		    BUF_PUSH_3 (stop_memory, this_group_regnum,
+				regnum - this_group_regnum);
+		  }
+	      }
+	      break;
+
+
+	    case '|':					/* `\|'.  */
+	      if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
+		goto normal_backslash;
+	    handle_alt:
+	      if (syntax & RE_LIMITED_OPS)
+		goto normal_char;
+
+	      /* Insert before the previous alternative a jump which
+		 jumps to this alternative if the former fails.  */
+	      GET_BUFFER_SPACE (3);
+	      INSERT_JUMP (on_failure_jump, begalt, b + 6);
+	      pending_exact = 0;
+	      b += 3;
+
+	      /* The alternative before this one has a jump after it
+		 which gets executed if it gets matched.  Adjust that
+		 jump so it will jump to this alternative's analogous
+		 jump (put in below, which in turn will jump to the next
+		 (if any) alternative's such jump, etc.).  The last such
+		 jump jumps to the correct final destination.  A picture:
+			  _____ _____
+			  |   | |   |
+			  |   v |   v
+			 a | b   | c
+
+		 If we are at `b', then fixup_alt_jump right now points to a
+		 three-byte space after `a'.  We'll put in the jump, set
+		 fixup_alt_jump to right after `b', and leave behind three
+		 bytes which we'll fill in when we get to after `c'.  */
+
+	      if (fixup_alt_jump)
+		STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
+
+	      /* Mark and leave space for a jump after this alternative,
+		 to be filled in later either by next alternative or
+		 when know we're at the end of a series of alternatives.  */
+	      fixup_alt_jump = b;
+	      GET_BUFFER_SPACE (3);
+	      b += 3;
+
+	      laststart = 0;
+	      begalt = b;
+	      break;
+
+
+	    case '{':
+	      /* If \{ is a literal.  */
+	      if (!(syntax & RE_INTERVALS)
+		     /* If we're at `\{' and it's not the open-interval
+			operator.  */
+		  || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
+		  || (p - 2 == pattern  &&  p == pend))
+		goto normal_backslash;
+
+	    handle_interval:
+	      {
+		/* If got here, then the syntax allows intervals.  */
+
+		/* At least (most) this many matches must be made.  */
+		int lower_bound = -1, upper_bound = -1;
+
+		beg_interval = p - 1;
+
+		if (p == pend)
+		  {
+		    if (syntax & RE_NO_BK_BRACES)
+		      goto unfetch_interval;
+		    else
+		      return REG_EBRACE;
+		  }
+
+		GET_UNSIGNED_NUMBER (lower_bound);
+
+		if (c == ',')
+		  {
+		    GET_UNSIGNED_NUMBER (upper_bound);
+		    if (upper_bound < 0) upper_bound = RE_DUP_MAX;
+		  }
+		else
+		  /* Interval such as `{1}' => match exactly once. */
+		  upper_bound = lower_bound;
+
+		if (lower_bound < 0 || upper_bound > RE_DUP_MAX
+		    || lower_bound > upper_bound)
+		  {
+		    if (syntax & RE_NO_BK_BRACES)
+		      goto unfetch_interval;
+		    else
+		      return REG_BADBR;
+		  }
+
+		if (!(syntax & RE_NO_BK_BRACES))
+		  {
+		    if (c != '\\') return REG_EBRACE;
+
+		    PATFETCH (c);
+		  }
+
+		if (c != '}')
+		  {
+		    if (syntax & RE_NO_BK_BRACES)
+		      goto unfetch_interval;
+		    else
+		      return REG_BADBR;
+		  }
+
+		/* We just parsed a valid interval.  */
+
+		/* If it's invalid to have no preceding re.  */
+		if (!laststart)
+		  {
+		    if (syntax & RE_CONTEXT_INVALID_OPS)
+		      return REG_BADRPT;
+		    else if (syntax & RE_CONTEXT_INDEP_OPS)
+		      laststart = b;
+		    else
+		      goto unfetch_interval;
+		  }
+
+		/* If the upper bound is zero, don't want to succeed at
+		   all; jump from `laststart' to `b + 3', which will be
+		   the end of the buffer after we insert the jump.  */
+		 if (upper_bound == 0)
+		   {
+		     GET_BUFFER_SPACE (3);
+		     INSERT_JUMP (jump, laststart, b + 3);
+		     b += 3;
+		   }
+
+		 /* Otherwise, we have a nontrivial interval.  When
+		    we're all done, the pattern will look like:
+		      set_number_at <jump count> <upper bound>
+		      set_number_at <succeed_n count> <lower bound>
+		      succeed_n <after jump addr> <succeed_n count>
+		      <body of loop>
+		      jump_n <succeed_n addr> <jump count>
+		    (The upper bound and `jump_n' are omitted if
+		    `upper_bound' is 1, though.)  */
+		 else
+		   { /* If the upper bound is > 1, we need to insert
+			more at the end of the loop.  */
+		     unsigned nbytes = 10 + (upper_bound > 1) * 10;
+
+		     GET_BUFFER_SPACE (nbytes);
+
+		     /* Initialize lower bound of the `succeed_n', even
+			though it will be set during matching by its
+			attendant `set_number_at' (inserted next),
+			because `re_compile_fastmap' needs to know.
+			Jump to the `jump_n' we might insert below.  */
+		     INSERT_JUMP2 (succeed_n, laststart,
+				   b + 5 + (upper_bound > 1) * 5,
+				   lower_bound);
+		     b += 5;
+
+		     /* Code to initialize the lower bound.  Insert
+			before the `succeed_n'.  The `5' is the last two
+			bytes of this `set_number_at', plus 3 bytes of
+			the following `succeed_n'.  */
+		     insert_op2 (set_number_at, laststart, 5, lower_bound, b);
+		     b += 5;
+
+		     if (upper_bound > 1)
+		       { /* More than one repetition is allowed, so
+			    append a backward jump to the `succeed_n'
+			    that starts this interval.
+
+			    When we've reached this during matching,
+			    we'll have matched the interval once, so
+			    jump back only `upper_bound - 1' times.  */
+			 STORE_JUMP2 (jump_n, b, laststart + 5,
+				      upper_bound - 1);
+			 b += 5;
+
+			 /* The location we want to set is the second
+			    parameter of the `jump_n'; that is `b-2' as
+			    an absolute address.  `laststart' will be
+			    the `set_number_at' we're about to insert;
+			    `laststart+3' the number to set, the source
+			    for the relative address.  But we are
+			    inserting into the middle of the pattern --
+			    so everything is getting moved up by 5.
+			    Conclusion: (b - 2) - (laststart + 3) + 5,
+			    i.e., b - laststart.
+
+			    We insert this at the beginning of the loop
+			    so that if we fail during matching, we'll
+			    reinitialize the bounds.  */
+			 insert_op2 (set_number_at, laststart, b - laststart,
+				     upper_bound - 1, b);
+			 b += 5;
+		       }
+		   }
+		pending_exact = 0;
+		beg_interval = NULL;
+	      }
+	      break;
+
+	    unfetch_interval:
+	      /* If an invalid interval, match the characters as literals.  */
+	       assert (beg_interval);
+	       p = beg_interval;
+	       beg_interval = NULL;
+
+	       /* normal_char and normal_backslash need `c'.  */
+	       PATFETCH (c);
+
+	       if (!(syntax & RE_NO_BK_BRACES))
+		 {
+		   if (p > pattern  &&  p[-1] == '\\')
+		     goto normal_backslash;
+		 }
+	       goto normal_char;
+
+#ifdef emacs
+	    /* There is no way to specify the before_dot and after_dot
+	       operators.  rms says this is ok.  --karl  */
+	    case '=':
+	      BUF_PUSH (at_dot);
+	      break;
+
+	    case 's':
+	      laststart = b;
+	      PATFETCH (c);
+	      BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
+	      break;
+
+	    case 'S':
+	      laststart = b;
+	      PATFETCH (c);
+	      BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
+	      break;
+#endif /* emacs */
+
+
+	    case 'w':
+	      laststart = b;
+	      BUF_PUSH (wordchar);
+	      break;
+
+
+	    case 'W':
+	      laststart = b;
+	      BUF_PUSH (notwordchar);
+	      break;
+
+
+	    case '<':
+	      BUF_PUSH (wordbeg);
+	      break;
+
+	    case '>':
+	      BUF_PUSH (wordend);
+	      break;
+
+	    case 'b':
+	      BUF_PUSH (wordbound);
+	      break;
+
+	    case 'B':
+	      BUF_PUSH (notwordbound);
+	      break;
+
+	    case '`':
+	      BUF_PUSH (begbuf);
+	      break;
+
+	    case '\'':
+	      BUF_PUSH (endbuf);
+	      break;
+
+	    case '1': case '2': case '3': case '4': case '5':
+	    case '6': case '7': case '8': case '9':
+	      if (syntax & RE_NO_BK_REFS)
+		goto normal_char;
+
+	      c1 = c - '0';
+
+	      if (c1 > regnum)
+		return REG_ESUBREG;
+
+	      /* Can't back reference to a subexpression if inside of it.  */
+	      if (group_in_compile_stack (compile_stack, c1))
+		goto normal_char;
+
+	      laststart = b;
+	      BUF_PUSH_2 (duplicate, c1);
+	      break;
+
+
+	    case '+':
+	    case '?':
+	      if (syntax & RE_BK_PLUS_QM)
+		goto handle_plus;
+	      else
+		goto normal_backslash;
+
+	    default:
+	    normal_backslash:
+	      /* You might think it would be useful for \ to mean
+		 not to translate; but if we don't translate it
+		 it will never match anything.  */
+	      c = TRANSLATE (c);
+	      goto normal_char;
+	    }
+	  break;
+
+
+	default:
+	/* Expects the character in `c'.  */
+	normal_char:
+	      /* If no exactn currently being built.  */
+	  if (!pending_exact
+
+	      /* If last exactn not at current position.  */
+	      || pending_exact + *pending_exact + 1 != b
+
+	      /* We have only one byte following the exactn for the count.  */
+	      || *pending_exact == (1 << BYTEWIDTH) - 1
+
+	      /* If followed by a repetition operator.  */
+	      || *p == '*' || *p == '^'
+	      || ((syntax & RE_BK_PLUS_QM)
+		  ? *p == '\\' && (p[1] == '+' || p[1] == '?')
+		  : (*p == '+' || *p == '?'))
+	      || ((syntax & RE_INTERVALS)
+		  && ((syntax & RE_NO_BK_BRACES)
+		      ? *p == '{'
+		      : (p[0] == '\\' && p[1] == '{'))))
+	    {
+	      /* Start building a new exactn.  */
+
+	      laststart = b;
+
+	      BUF_PUSH_2 (exactn, 0);
+	      pending_exact = b - 1;
+	    }
+
+	  BUF_PUSH (c);
+	  (*pending_exact)++;
+	  break;
+	} /* switch (c) */
+    } /* while p != pend */
+
+
+  /* Through the pattern now.  */
+
+  if (fixup_alt_jump)
+    STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
+
+  if (!COMPILE_STACK_EMPTY)
+    return REG_EPAREN;
+
+  free (compile_stack.stack);
+
+  /* We have succeeded; set the length of the buffer.  */
+  bufp->used = b - bufp->buffer;
+
+#ifdef DEBUG
+  if (debug)
+    {
+      DEBUG_PRINT1 ("\nCompiled pattern: ");
+      print_compiled_pattern (bufp);
+    }
+#endif /* DEBUG */
+
+  return REG_NOERROR;
+} /* regex_compile */
+
+/* Subroutines for `regex_compile'.  */
+
+/* Store OP at LOC followed by two-byte integer parameter ARG.  */
+
+static void
+store_op1 (op, loc, arg)
+    re_opcode_t op;
+    unsigned char *loc;
+    int arg;
+{
+  *loc = (unsigned char) op;
+  STORE_NUMBER (loc + 1, arg);
+}
+
+
+/* Like `store_op1', but for two two-byte parameters ARG1 and ARG2.  */
+
+static void
+store_op2 (op, loc, arg1, arg2)
+    re_opcode_t op;
+    unsigned char *loc;
+    int arg1, arg2;
+{
+  *loc = (unsigned char) op;
+  STORE_NUMBER (loc + 1, arg1);
+  STORE_NUMBER (loc + 3, arg2);
+}
+
+
+/* Copy the bytes from LOC to END to open up three bytes of space at LOC
+   for OP followed by two-byte integer parameter ARG.  */
+
+static void
+insert_op1 (op, loc, arg, end)
+    re_opcode_t op;
+    unsigned char *loc;
+    int arg;
+    unsigned char *end;
+{
+  register unsigned char *pfrom = end;
+  register unsigned char *pto = end + 3;
+
+  while (pfrom != loc)
+    *--pto = *--pfrom;
+
+  store_op1 (op, loc, arg);
+}
+
+
+/* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2.  */
+
+static void
+insert_op2 (op, loc, arg1, arg2, end)
+    re_opcode_t op;
+    unsigned char *loc;
+    int arg1, arg2;
+    unsigned char *end;
+{
+  register unsigned char *pfrom = end;
+  register unsigned char *pto = end + 5;
+
+  while (pfrom != loc)
+    *--pto = *--pfrom;
+
+  store_op2 (op, loc, arg1, arg2);
+}
+
+
+/* P points to just after a ^ in PATTERN.  Return true if that ^ comes
+   after an alternative or a begin-subexpression.  We assume there is at
+   least one character before the ^.  */
+
+static boolean
+at_begline_loc_p (pattern, p, syntax)
+    const char *pattern, *p;
+    reg_syntax_t syntax;
+{
+  const char *prev = p - 2;
+  boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
+
+  return
+       /* After a subexpression?  */
+       (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
+       /* After an alternative?  */
+    || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
+}
+
+
+/* The dual of at_begline_loc_p.  This one is for $.  We assume there is
+   at least one character after the $, i.e., `P < PEND'.  */
+
+static boolean
+at_endline_loc_p (p, pend, syntax)
+    const char *p, *pend;
+    int syntax;
+{
+  const char *next = p;
+  boolean next_backslash = *next == '\\';
+  const char *next_next = p + 1 < pend ? p + 1 : NULL;
+
+  return
+       /* Before a subexpression?  */
+       (syntax & RE_NO_BK_PARENS ? *next == ')'
+	: next_backslash && next_next && *next_next == ')')
+       /* Before an alternative?  */
+    || (syntax & RE_NO_BK_VBAR ? *next == '|'
+	: next_backslash && next_next && *next_next == '|');
+}
+
+
+/* Returns true if REGNUM is in one of COMPILE_STACK's elements and
+   false if it's not.  */
+
+static boolean
+group_in_compile_stack (compile_stack, regnum)
+    compile_stack_type compile_stack;
+    regnum_t regnum;
+{
+  int this_element;
+
+  for (this_element = compile_stack.avail - 1;
+       this_element >= 0;
+       this_element--)
+    if (compile_stack.stack[this_element].regnum == regnum)
+      return true;
+
+  return false;
+}
+
+
+/* Read the ending character of a range (in a bracket expression) from the
+   uncompiled pattern *P_PTR (which ends at PEND).  We assume the
+   starting character is in `P[-2]'.  (`P[-1]' is the character `-'.)
+   Then we set the translation of all bits between the starting and
+   ending characters (inclusive) in the compiled pattern B.
+
+   Return an error code.
+
+   We use these short variable names so we can use the same macros as
+   `regex_compile' itself.  */
+
+static reg_errcode_t
+compile_range (p_ptr, pend, translate, syntax, b)
+    const char **p_ptr, *pend;
+    char *translate;
+    reg_syntax_t syntax;
+    unsigned char *b;
+{
+  unsigned this_char;
+
+  const char *p = *p_ptr;
+  int range_start, range_end;
+
+  if (p == pend)
+    return REG_ERANGE;
+
+  /* Even though the pattern is a signed `char *', we need to fetch
+     with unsigned char *'s; if the high bit of the pattern character
+     is set, the range endpoints will be negative if we fetch using a
+     signed char *.
+
+     We also want to fetch the endpoints without translating them; the
+     appropriate translation is done in the bit-setting loop below.  */
+  range_start = ((unsigned char *) p)[-2];
+  range_end   = ((unsigned char *) p)[0];
+
+  /* Have to increment the pointer into the pattern string, so the
+     caller isn't still at the ending character.  */
+  (*p_ptr)++;
+
+  /* If the start is after the end, the range is empty.  */
+  if (range_start > range_end)
+    return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
+
+  /* Here we see why `this_char' has to be larger than an `unsigned
+     char' -- the range is inclusive, so if `range_end' == 0xff
+     (assuming 8-bit characters), we would otherwise go into an infinite
+     loop, since all characters <= 0xff.  */
+  for (this_char = range_start; this_char <= range_end; this_char++)
+    {
+      SET_LIST_BIT (TRANSLATE (this_char));
+    }
+
+  return REG_NOERROR;
+}
+
+/* Failure stack declarations and macros; both re_compile_fastmap and
+   re_match_2 use a failure stack.  These have to be macros because of
+   REGEX_ALLOCATE.  */
+
+
+/* Number of failure points for which to initially allocate space
+   when matching.  If this number is exceeded, we allocate more
+   space, so it is not a hard limit.  */
+#ifndef INIT_FAILURE_ALLOC
+#define INIT_FAILURE_ALLOC 5
+#endif
+
+/* Roughly the maximum number of failure points on the stack.  Would be
+   exactly that if always used MAX_FAILURE_SPACE each time we failed.
+   This is a variable only so users of regex can assign to it; we never
+   change it ourselves.  */
+int re_max_failures = 2000;
+
+typedef const unsigned char *fail_stack_elt_t;
+
+typedef struct
+{
+  fail_stack_elt_t *stack;
+  unsigned size;
+  unsigned avail;			/* Offset of next open position.  */
+} fail_stack_type;
+
+#define FAIL_STACK_EMPTY()     (fail_stack.avail == 0)
+#define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
+#define FAIL_STACK_FULL()      (fail_stack.avail == fail_stack.size)
+#define FAIL_STACK_TOP()       (fail_stack.stack[fail_stack.avail])
+
+
+/* Initialize `fail_stack'.  Do `return -2' if the alloc fails.  */
+
+#define INIT_FAIL_STACK()						\
+  do {									\
+    fail_stack.stack = (fail_stack_elt_t *)				\
+      REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t));	\
+									\
+    if (fail_stack.stack == NULL)					\
+      return -2;							\
+									\
+    fail_stack.size = INIT_FAILURE_ALLOC;				\
+    fail_stack.avail = 0;						\
+  } while (0)
+
+
+/* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
+
+   Return 1 if succeeds, and 0 if either ran out of memory
+   allocating space for it or it was already too large.
+
+   REGEX_REALLOCATE requires `destination' be declared.   */
+
+#define DOUBLE_FAIL_STACK(fail_stack)					\
+  ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS		\
+   ? 0									\
+   : ((fail_stack).stack = (fail_stack_elt_t *)				\
+	REGEX_REALLOCATE ((fail_stack).stack, 				\
+	  (fail_stack).size * sizeof (fail_stack_elt_t),		\
+	  ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)),	\
+									\
+      (fail_stack).stack == NULL					\
+      ? 0								\
+      : ((fail_stack).size <<= 1, 					\
+	 1)))
+
+
+/* Push PATTERN_OP on FAIL_STACK.
+
+   Return 1 if was able to do so and 0 if ran out of memory allocating
+   space to do so.  */
+#define PUSH_PATTERN_OP(pattern_op, fail_stack)				\
+  ((FAIL_STACK_FULL ()							\
+    && !DOUBLE_FAIL_STACK (fail_stack))					\
+    ? 0									\
+    : ((fail_stack).stack[(fail_stack).avail++] = pattern_op,		\
+       1))
+
+/* This pushes an item onto the failure stack.  Must be a four-byte
+   value.  Assumes the variable `fail_stack'.  Probably should only
+   be called from within `PUSH_FAILURE_POINT'.  */
+#define PUSH_FAILURE_ITEM(item)						\
+  fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item
+
+/* The complement operation.  Assumes `fail_stack' is nonempty.  */
+#define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail]
+
+/* Used to omit pushing failure point id's when we're not debugging.  */
+#ifdef DEBUG
+#define DEBUG_PUSH PUSH_FAILURE_ITEM
+#define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM ()
+#else
+#define DEBUG_PUSH(item)
+#define DEBUG_POP(item_addr)
+#endif
+
+
+/* Push the information about the state we will need
+   if we ever fail back to it.
+
+   Requires variables fail_stack, regstart, regend, reg_info, and
+   num_regs be declared.  DOUBLE_FAIL_STACK requires `destination' be
+   declared.
+
+   Does `return FAILURE_CODE' if runs out of memory.  */
+
+#define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code)	\
+  do {									\
+    char *destination;							\
+    /* Must be int, so when we don't save any registers, the arithmetic	\
+       of 0 + -1 isn't done as unsigned.  */				\
+    int this_reg;							\
+									\
+    DEBUG_STATEMENT (failure_id++);					\
+    DEBUG_STATEMENT (nfailure_points_pushed++);				\
+    DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id);		\
+    DEBUG_PRINT2 ("  Before push, next avail: %d\n", (fail_stack).avail);\
+    DEBUG_PRINT2 ("                     size: %d\n", (fail_stack).size);\
+									\
+    DEBUG_PRINT2 ("  slots needed: %d\n", NUM_FAILURE_ITEMS);		\
+    DEBUG_PRINT2 ("     available: %d\n", REMAINING_AVAIL_SLOTS);	\
+									\
+    /* Ensure we have enough space allocated for what we will push.  */	\
+    while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS)			\
+      {									\
+	if (!DOUBLE_FAIL_STACK (fail_stack))			\
+	  return failure_code;						\
+									\
+	DEBUG_PRINT2 ("\n  Doubled stack; size now: %d\n",		\
+		       (fail_stack).size);				\
+	DEBUG_PRINT2 ("  slots available: %d\n", REMAINING_AVAIL_SLOTS);\
+      }									\
+									\
+    /* Push the info, starting with the registers.  */			\
+    DEBUG_PRINT1 ("\n");						\
+									\
+    for (this_reg = lowest_active_reg; this_reg <= highest_active_reg;	\
+	 this_reg++)							\
+      {									\
+	DEBUG_PRINT2 ("  Pushing reg: %d\n", this_reg);			\
+	DEBUG_STATEMENT (num_regs_pushed++);				\
+									\
+	DEBUG_PRINT2 ("    start: 0x%x\n", regstart[this_reg]);		\
+	PUSH_FAILURE_ITEM (regstart[this_reg]);				\
+									\
+	DEBUG_PRINT2 ("    end: 0x%x\n", regend[this_reg]);		\
+	PUSH_FAILURE_ITEM (regend[this_reg]);				\
+									\
+	DEBUG_PRINT2 ("    info: 0x%x\n      ", reg_info[this_reg]);	\
+	DEBUG_PRINT2 (" match_null=%d",					\
+		      REG_MATCH_NULL_STRING_P (reg_info[this_reg]));	\
+	DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg]));	\
+	DEBUG_PRINT2 (" matched_something=%d",				\
+		      MATCHED_SOMETHING (reg_info[this_reg]));		\
+	DEBUG_PRINT2 (" ever_matched=%d",				\
+		      EVER_MATCHED_SOMETHING (reg_info[this_reg]));	\
+	DEBUG_PRINT1 ("\n");						\
+	PUSH_FAILURE_ITEM (reg_info[this_reg].word);			\
+      }									\
+									\
+    DEBUG_PRINT2 ("  Pushing  low active reg: %d\n", lowest_active_reg);\
+    PUSH_FAILURE_ITEM (lowest_active_reg);				\
+									\
+    DEBUG_PRINT2 ("  Pushing high active reg: %d\n", highest_active_reg);\
+    PUSH_FAILURE_ITEM (highest_active_reg);				\
+									\
+    DEBUG_PRINT2 ("  Pushing pattern 0x%x: ", pattern_place);		\
+    DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend);		\
+    PUSH_FAILURE_ITEM (pattern_place);					\
+									\
+    DEBUG_PRINT2 ("  Pushing string 0x%x: `", string_place);		\
+    DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2,   \
+				 size2);				\
+    DEBUG_PRINT1 ("'\n");						\
+    PUSH_FAILURE_ITEM (string_place);					\
+									\
+    DEBUG_PRINT2 ("  Pushing failure id: %u\n", failure_id);		\
+    DEBUG_PUSH (failure_id);						\
+  } while (0)
+
+/* This is the number of items that are pushed and popped on the stack
+   for each register.  */
+#define NUM_REG_ITEMS  3
+
+/* Individual items aside from the registers.  */
+#ifdef DEBUG
+#define NUM_NONREG_ITEMS 5 /* Includes failure point id.  */
+#else
+#define NUM_NONREG_ITEMS 4
+#endif
+
+/* We push at most this many items on the stack.  */
+#define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
+
+/* We actually push this many items.  */
+#define NUM_FAILURE_ITEMS						\
+  ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS 	\
+    + NUM_NONREG_ITEMS)
+
+/* How many items can still be added to the stack without overflowing it.  */
+#define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
+
+
+/* Pops what PUSH_FAIL_STACK pushes.
+
+   We restore into the parameters, all of which should be lvalues:
+     STR -- the saved data position.
+     PAT -- the saved pattern position.
+     LOW_REG, HIGH_REG -- the highest and lowest active registers.
+     REGSTART, REGEND -- arrays of string positions.
+     REG_INFO -- array of information about each subexpression.
+
+   Also assumes the variables `fail_stack' and (if debugging), `bufp',
+   `pend', `string1', `size1', `string2', and `size2'.  */
+
+#define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
+{									\
+  DEBUG_STATEMENT (fail_stack_elt_t failure_id;)			\
+  int this_reg;								\
+  const unsigned char *string_temp;					\
+									\
+  assert (!FAIL_STACK_EMPTY ());					\
+									\
+  /* Remove failure points and point to how many regs pushed.  */	\
+  DEBUG_PRINT1 ("POP_FAILURE_POINT:\n");				\
+  DEBUG_PRINT2 ("  Before pop, next avail: %d\n", fail_stack.avail);	\
+  DEBUG_PRINT2 ("                    size: %d\n", fail_stack.size);	\
+									\
+  assert (fail_stack.avail >= NUM_NONREG_ITEMS);			\
+									\
+  DEBUG_POP (&failure_id);						\
+  DEBUG_PRINT2 ("  Popping failure id: %u\n", failure_id);		\
+									\
+  /* If the saved string location is NULL, it came from an		\
+     on_failure_keep_string_jump opcode, and we want to throw away the	\
+     saved NULL, thus retaining our current position in the string.  */	\
+  string_temp = POP_FAILURE_ITEM ();					\
+  if (string_temp != NULL)						\
+    str = (const char *) string_temp;					\
+									\
+  DEBUG_PRINT2 ("  Popping string 0x%x: `", str);			\
+  DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2);	\
+  DEBUG_PRINT1 ("'\n");							\
+									\
+  pat = (unsigned char *) POP_FAILURE_ITEM ();				\
+  DEBUG_PRINT2 ("  Popping pattern 0x%x: ", pat);			\
+  DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend);			\
+									\
+  /* Restore register info.  */						\
+  high_reg = (unsigned) POP_FAILURE_ITEM ();				\
+  DEBUG_PRINT2 ("  Popping high active reg: %d\n", high_reg);		\
+									\
+  low_reg = (unsigned) POP_FAILURE_ITEM ();				\
+  DEBUG_PRINT2 ("  Popping  low active reg: %d\n", low_reg);		\
+									\
+  for (this_reg = high_reg; this_reg >= low_reg; this_reg--)		\
+    {									\
+      DEBUG_PRINT2 ("    Popping reg: %d\n", this_reg);			\
+									\
+      reg_info[this_reg].word = POP_FAILURE_ITEM ();			\
+      DEBUG_PRINT2 ("      info: 0x%x\n", reg_info[this_reg]);		\
+									\
+      regend[this_reg] = (const char *) POP_FAILURE_ITEM ();		\
+      DEBUG_PRINT2 ("      end: 0x%x\n", regend[this_reg]);		\
+									\
+      regstart[this_reg] = (const char *) POP_FAILURE_ITEM ();		\
+      DEBUG_PRINT2 ("      start: 0x%x\n", regstart[this_reg]);		\
+    }									\
+									\
+  DEBUG_STATEMENT (nfailure_points_popped++);				\
+} /* POP_FAILURE_POINT */
+
+/* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
+   BUFP.  A fastmap records which of the (1 << BYTEWIDTH) possible
+   characters can start a string that matches the pattern.  This fastmap
+   is used by re_search to skip quickly over impossible starting points.
+
+   The caller must supply the address of a (1 << BYTEWIDTH)-byte data
+   area as BUFP->fastmap.
+
+   We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
+   the pattern buffer.
+
+   Returns 0 if we succeed, -2 if an internal error.   */
+
+int
+re_compile_fastmap (bufp)
+     struct re_pattern_buffer *bufp;
+{
+  int j, k;
+  fail_stack_type fail_stack;
+#ifndef REGEX_MALLOC
+  char *destination;
+#endif
+  /* We don't push any register information onto the failure stack.  */
+  unsigned num_regs = 0;
+
+  register char *fastmap = bufp->fastmap;
+  unsigned char *pattern = bufp->buffer;
+  unsigned long size = bufp->used;
+  const unsigned char *p = pattern;
+  register unsigned char *pend = pattern + size;
+
+  /* Assume that each path through the pattern can be null until
+     proven otherwise.  We set this false at the bottom of switch
+     statement, to which we get only if a particular path doesn't
+     match the empty string.  */
+  boolean path_can_be_null = true;
+
+  /* We aren't doing a `succeed_n' to begin with.  */
+  boolean succeed_n_p = false;
+
+  assert (fastmap != NULL && p != NULL);
+
+  INIT_FAIL_STACK ();
+  bzero (fastmap, 1 << BYTEWIDTH);  /* Assume nothing's valid.  */
+  bufp->fastmap_accurate = 1;	    /* It will be when we're done.  */
+  bufp->can_be_null = 0;
+
+  while (p != pend || !FAIL_STACK_EMPTY ())
+    {
+      if (p == pend)
+	{
+	  bufp->can_be_null |= path_can_be_null;
+
+	  /* Reset for next path.  */
+	  path_can_be_null = true;
+
+	  p = fail_stack.stack[--fail_stack.avail];
+	}
+
+      /* We should never be about to go beyond the end of the pattern.  */
+      assert (p < pend);
+
+#ifdef SWITCH_ENUM_BUG
+      switch ((int) ((re_opcode_t) *p++))
+#else
+      switch ((re_opcode_t) *p++)
+#endif
+	{
+
+	/* I guess the idea here is to simply not bother with a fastmap
+	   if a backreference is used, since it's too hard to figure out
+	   the fastmap for the corresponding group.  Setting
+	   `can_be_null' stops `re_search_2' from using the fastmap, so
+	   that is all we do.  */
+	case duplicate:
+	  bufp->can_be_null = 1;
+	  return 0;
+
+
+      /* Following are the cases which match a character.  These end
+	 with `break'.  */
+
+	case exactn:
+	  fastmap[p[1]] = 1;
+	  break;
+
+
+	case charset:
+	  for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
+	    if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
+	      fastmap[j] = 1;
+	  break;
+
+
+	case charset_not:
+	  /* Chars beyond end of map must be allowed.  */
+	  for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
+	    fastmap[j] = 1;
+
+	  for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
+	    if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
+	      fastmap[j] = 1;
+	  break;
+
+
+	case wordchar:
+	  for (j = 0; j < (1 << BYTEWIDTH); j++)
+	    if (SYNTAX (j) == Sword)
+	      fastmap[j] = 1;
+	  break;
+
+
+	case notwordchar:
+	  for (j = 0; j < (1 << BYTEWIDTH); j++)
+	    if (SYNTAX (j) != Sword)
+	      fastmap[j] = 1;
+	  break;
+
+
+	case anychar:
+	  /* `.' matches anything ...  */
+	  for (j = 0; j < (1 << BYTEWIDTH); j++)
+	    fastmap[j] = 1;
+
+	  /* ... except perhaps newline.  */
+	  if (!(bufp->syntax & RE_DOT_NEWLINE))
+	    fastmap['\n'] = 0;
+
+	  /* Return if we have already set `can_be_null'; if we have,
+	     then the fastmap is irrelevant.  Something's wrong here.  */
+	  else if (bufp->can_be_null)
+	    return 0;
+
+	  /* Otherwise, have to check alternative paths.  */
+	  break;
+
+
+#ifdef emacs
+	case syntaxspec:
+	  k = *p++;
+	  for (j = 0; j < (1 << BYTEWIDTH); j++)
+	    if (SYNTAX (j) == (enum syntaxcode) k)
+	      fastmap[j] = 1;
+	  break;
+
+
+	case notsyntaxspec:
+	  k = *p++;
+	  for (j = 0; j < (1 << BYTEWIDTH); j++)
+	    if (SYNTAX (j) != (enum syntaxcode) k)
+	      fastmap[j] = 1;
+	  break;
+
+
+      /* All cases after this match the empty string.  These end with
+	 `continue'.  */
+
+
+	case before_dot:
+	case at_dot:
+	case after_dot:
+	  continue;
+#endif /* not emacs */
+
+
+	case no_op:
+	case begline:
+	case endline:
+	case begbuf:
+	case endbuf:
+	case wordbound:
+	case notwordbound:
+	case wordbeg:
+	case wordend:
+	case push_dummy_failure:
+	  continue;
+
+
+	case jump_n:
+	case pop_failure_jump:
+	case maybe_pop_jump:
+	case jump:
+	case jump_past_alt:
+	case dummy_failure_jump:
+	  EXTRACT_NUMBER_AND_INCR (j, p);
+	  p += j;
+	  if (j > 0)
+	    continue;
+
+	  /* Jump backward implies we just went through the body of a
+	     loop and matched nothing.  Opcode jumped to should be
+	     `on_failure_jump' or `succeed_n'.  Just treat it like an
+	     ordinary jump.  For a * loop, it has pushed its failure
+	     point already; if so, discard that as redundant.  */
+	  if ((re_opcode_t) *p != on_failure_jump
+	      && (re_opcode_t) *p != succeed_n)
+	    continue;
+
+	  p++;
+	  EXTRACT_NUMBER_AND_INCR (j, p);
+	  p += j;
+
+	  /* If what's on the stack is where we are now, pop it.  */
+	  if (!FAIL_STACK_EMPTY ()
+	      && fail_stack.stack[fail_stack.avail - 1] == p)
+	    fail_stack.avail--;
+
+	  continue;
+
+
+	case on_failure_jump:
+	case on_failure_keep_string_jump:
+	handle_on_failure_jump:
+	  EXTRACT_NUMBER_AND_INCR (j, p);
+
+	  /* For some patterns, e.g., `(a?)?', `p+j' here points to the
+	     end of the pattern.  We don't want to push such a point,
+	     since when we restore it above, entering the switch will
+	     increment `p' past the end of the pattern.  We don't need
+	     to push such a point since we obviously won't find any more
+	     fastmap entries beyond `pend'.  Such a pattern can match
+	     the null string, though.  */
+	  if (p + j < pend)
+	    {
+	      if (!PUSH_PATTERN_OP (p + j, fail_stack))
+		return -2;
+	    }
+	  else
+	    bufp->can_be_null = 1;
+
+	  if (succeed_n_p)
+	    {
+	      EXTRACT_NUMBER_AND_INCR (k, p);	/* Skip the n.  */
+	      succeed_n_p = false;
+	    }
+
+	  continue;
+
+
+	case succeed_n:
+	  /* Get to the number of times to succeed.  */
+	  p += 2;
+
+	  /* Increment p past the n for when k != 0.  */
+	  EXTRACT_NUMBER_AND_INCR (k, p);
+	  if (k == 0)
+	    {
+	      p -= 4;
+	      succeed_n_p = true;  /* Spaghetti code alert.  */
+	      goto handle_on_failure_jump;
+	    }
+	  continue;
+
+
+	case set_number_at:
+	  p += 4;
+	  continue;
+
+
+	case start_memory:
+	case stop_memory:
+	  p += 2;
+	  continue;
+
+
+	default:
+	  abort (); /* We have listed all the cases.  */
+	} /* switch *p++ */
+
+      /* Getting here means we have found the possible starting
+	 characters for one path of the pattern -- and that the empty
+	 string does not match.  We need not follow this path further.
+	 Instead, look at the next alternative (remembered on the
+	 stack), or quit if no more.  The test at the top of the loop
+	 does these things.  */
+      path_can_be_null = false;
+      p = pend;
+    } /* while p */
+
+  /* Set `can_be_null' for the last path (also the first path, if the
+     pattern is empty).  */
+  bufp->can_be_null |= path_can_be_null;
+  return 0;
+} /* re_compile_fastmap */
+
+/* Set REGS to hold NUM_REGS registers, storing them in STARTS and
+   ENDS.  Subsequent matches using PATTERN_BUFFER and REGS will use
+   this memory for recording register information.  STARTS and ENDS
+   must be allocated using the malloc library routine, and must each
+   be at least NUM_REGS * sizeof (regoff_t) bytes long.
+
+   If NUM_REGS == 0, then subsequent matches should allocate their own
+   register data.
+
+   Unless this function is called, the first search or match using
+   PATTERN_BUFFER will allocate its own register data, without
+   freeing the old data.  */
+
+void
+re_set_registers (bufp, regs, num_regs, starts, ends)
+    struct re_pattern_buffer *bufp;
+    struct re_registers *regs;
+    unsigned num_regs;
+    regoff_t *starts, *ends;
+{
+  if (num_regs)
+    {
+      bufp->regs_allocated = REGS_REALLOCATE;
+      regs->num_regs = num_regs;
+      regs->start = starts;
+      regs->end = ends;
+    }
+  else
+    {
+      bufp->regs_allocated = REGS_UNALLOCATED;
+      regs->num_regs = 0;
+      regs->start = regs->end = (regoff_t *) 0;
+    }
+}
+
+/* Searching routines.  */
+
+/* Like re_search_2, below, but only one string is specified, and
+   doesn't let you say where to stop matching. */
+
+int
+re_search (bufp, string, size, startpos, range, regs)
+     struct re_pattern_buffer *bufp;
+     const char *string;
+     int size, startpos, range;
+     struct re_registers *regs;
+{
+  return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
+		      regs, size);
+}
+
+
+/* Using the compiled pattern in BUFP->buffer, first tries to match the
+   virtual concatenation of STRING1 and STRING2, starting first at index
+   STARTPOS, then at STARTPOS + 1, and so on.
+
+   STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
+
+   RANGE is how far to scan while trying to match.  RANGE = 0 means try
+   only at STARTPOS; in general, the last start tried is STARTPOS +
+   RANGE.
+
+   In REGS, return the indices of the virtual concatenation of STRING1
+   and STRING2 that matched the entire BUFP->buffer and its contained
+   subexpressions.
+
+   Do not consider matching one past the index STOP in the virtual
+   concatenation of STRING1 and STRING2.
+
+   We return either the position in the strings at which the match was
+   found, -1 if no match, or -2 if error (such as failure
+   stack overflow).  */
+
+int
+re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
+     struct re_pattern_buffer *bufp;
+     const char *string1, *string2;
+     int size1, size2;
+     int startpos;
+     int range;
+     struct re_registers *regs;
+     int stop;
+{
+  int val;
+  register char *fastmap = bufp->fastmap;
+  register char *translate = bufp->translate;
+  int total_size = size1 + size2;
+  int endpos = startpos + range;
+
+  /* Check for out-of-range STARTPOS.  */
+  if (startpos < 0 || startpos > total_size)
+    return -1;
+
+  /* Fix up RANGE if it might eventually take us outside
+     the virtual concatenation of STRING1 and STRING2.  */
+  if (endpos < -1)
+    range = -1 - startpos;
+  else if (endpos > total_size)
+    range = total_size - startpos;
+
+  /* If the search isn't to be a backwards one, don't waste time in a
+     search for a pattern that must be anchored.  */
+  if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
+    {
+      if (startpos > 0)
+	return -1;
+      else
+	range = 1;
+    }
+
+  /* Update the fastmap now if not correct already.  */
+  if (fastmap && !bufp->fastmap_accurate)
+    if (re_compile_fastmap (bufp) == -2)
+      return -2;
+
+  /* Loop through the string, looking for a place to start matching.  */
+  for (;;)
+    {
+      /* If a fastmap is supplied, skip quickly over characters that
+	 cannot be the start of a match.  If the pattern can match the
+	 null string, however, we don't need to skip characters; we want
+	 the first null string.  */
+      if (fastmap && startpos < total_size && !bufp->can_be_null)
+	{
+	  if (range > 0)	/* Searching forwards.  */
+	    {
+	      register const char *d;
+	      register int lim = 0;
+	      int irange = range;
+
+	      if (startpos < size1 && startpos + range >= size1)
+		lim = range - (size1 - startpos);
+
+	      d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
+
+	      /* Written out as an if-else to avoid testing `translate'
+		 inside the loop.  */
+	      if (translate)
+		while (range > lim
+		       && !fastmap[(unsigned char)
+				   translate[(unsigned char) *d++]])
+		  range--;
+	      else
+		while (range > lim && !fastmap[(unsigned char) *d++])
+		  range--;
+
+	      startpos += irange - range;
+	    }
+	  else				/* Searching backwards.  */
+	    {
+	      register char c = (size1 == 0 || startpos >= size1
+				 ? string2[startpos - size1]
+				 : string1[startpos]);
+
+	      if (!fastmap[(unsigned char) TRANSLATE (c)])
+		goto advance;
+	    }
+	}
+
+      /* If can't match the null string, and that's all we have left, fail.  */
+      if (range >= 0 && startpos == total_size && fastmap
+	  && !bufp->can_be_null)
+	return -1;
+
+      val = re_match_2 (bufp, string1, size1, string2, size2,
+			startpos, regs, stop);
+      if (val >= 0)
+	return startpos;
+
+      if (val == -2)
+	return -2;
+
+    advance:
+      if (!range)
+	break;
+      else if (range > 0)
+	{
+	  range--;
+	  startpos++;
+	}
+      else
+	{
+	  range++;
+	  startpos--;
+	}
+    }
+  return -1;
+} /* re_search_2 */
+
+/* Declarations and macros for re_match_2.  */
+
+static int bcmp_translate ();
+static boolean alt_match_null_string_p (),
+	       common_op_match_null_string_p (),
+	       group_match_null_string_p ();
+
+/* Structure for per-register (a.k.a. per-group) information.
+   This must not be longer than one word, because we push this value
+   onto the failure stack.  Other register information, such as the
+   starting and ending positions (which are addresses), and the list of
+   inner groups (which is a bits list) are maintained in separate
+   variables.
+
+   We are making a (strictly speaking) nonportable assumption here: that
+   the compiler will pack our bit fields into something that fits into
+   the type of `word', i.e., is something that fits into one item on the
+   failure stack.  */
+typedef union
+{
+  fail_stack_elt_t word;
+  struct
+  {
+      /* This field is one if this group can match the empty string,
+	 zero if not.  If not yet determined,  `MATCH_NULL_UNSET_VALUE'.  */
+#define MATCH_NULL_UNSET_VALUE 3
+    unsigned match_null_string_p : 2;
+    unsigned is_active : 1;
+    unsigned matched_something : 1;
+    unsigned ever_matched_something : 1;
+  } bits;
+} register_info_type;
+
+#define REG_MATCH_NULL_STRING_P(R)  ((R).bits.match_null_string_p)
+#define IS_ACTIVE(R)  ((R).bits.is_active)
+#define MATCHED_SOMETHING(R)  ((R).bits.matched_something)
+#define EVER_MATCHED_SOMETHING(R)  ((R).bits.ever_matched_something)
+
+
+/* Call this when have matched a real character; it sets `matched' flags
+   for the subexpressions which we are currently inside.  Also records
+   that those subexprs have matched.  */
+#define SET_REGS_MATCHED()						\
+  do									\
+    {									\
+      unsigned r;							\
+      for (r = lowest_active_reg; r <= highest_active_reg; r++)		\
+	{								\
+	  MATCHED_SOMETHING (reg_info[r])				\
+	    = EVER_MATCHED_SOMETHING (reg_info[r])			\
+	    = 1;							\
+	}								\
+    }									\
+  while (0)
+
+
+/* This converts PTR, a pointer into one of the search strings `string1'
+   and `string2' into an offset from the beginning of that string.  */
+#define POINTER_TO_OFFSET(ptr)						\
+  (FIRST_STRING_P (ptr) ? (ptr) - string1 : (ptr) - string2 + size1)
+
+/* Registers are set to a sentinel when they haven't yet matched.  */
+#define REG_UNSET_VALUE ((char *) -1)
+#define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
+
+
+/* Macros for dealing with the split strings in re_match_2.  */
+
+#define MATCHING_IN_FIRST_STRING  (dend == end_match_1)
+
+/* Call before fetching a character with *d.  This switches over to
+   string2 if necessary.  */
+#define PREFETCH()							\
+  while (d == dend)						    	\
+    {									\
+      /* End of string2 => fail.  */					\
+      if (dend == end_match_2) 						\
+	goto fail;							\
+      /* End of string1 => advance to string2.  */ 			\
+      d = string2;						        \
+      dend = end_match_2;						\
+    }
+
+
+/* Test if at very beginning or at very end of the virtual concatenation
+   of `string1' and `string2'.  If only one string, it's `string2'.  */
+#define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
+#define AT_STRINGS_END(d) ((d) == end2)
+
+
+/* Test if D points to a character which is word-constituent.  We have
+   two special cases to check for: if past the end of string1, look at
+   the first character in string2; and if before the beginning of
+   string2, look at the last character in string1.  */
+#define WORDCHAR_P(d)							\
+  (SYNTAX ((d) == end1 ? *string2					\
+	   : (d) == string2 - 1 ? *(end1 - 1) : *(d))			\
+   == Sword)
+
+/* Test if the character before D and the one at D differ with respect
+   to being word-constituent.  */
+#define AT_WORD_BOUNDARY(d)						\
+  (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)				\
+   || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
+
+
+/* Free everything we malloc.  */
+#ifdef REGEX_MALLOC
+#define FREE_VAR(var) if (var) free (var); var = NULL
+#define FREE_VARIABLES()						\
+  do {									\
+    FREE_VAR (fail_stack.stack);					\
+    FREE_VAR (regstart);						\
+    FREE_VAR (regend);							\
+    FREE_VAR (old_regstart);						\
+    FREE_VAR (old_regend);						\
+    FREE_VAR (best_regstart);						\
+    FREE_VAR (best_regend);						\
+    FREE_VAR (reg_info);						\
+    FREE_VAR (reg_dummy);						\
+    FREE_VAR (reg_info_dummy);						\
+  } while (0)
+#else /* not REGEX_MALLOC */
+/* Some MIPS systems (at least) want this to free alloca'd storage.  */
+#define FREE_VARIABLES() alloca (0)
+#endif /* not REGEX_MALLOC */
+
+
+/* These values must meet several constraints.  They must not be valid
+   register values; since we have a limit of 255 registers (because
+   we use only one byte in the pattern for the register number), we can
+   use numbers larger than 255.  They must differ by 1, because of
+   NUM_FAILURE_ITEMS above.  And the value for the lowest register must
+   be larger than the value for the highest register, so we do not try
+   to actually save any registers when none are active.  */
+#define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
+#define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
+
+/* Matching routines.  */
+
+#ifndef emacs   /* Emacs never uses this.  */
+/* re_match is like re_match_2 except it takes only a single string.  */
+
+int
+re_match (bufp, string, size, pos, regs)
+     struct re_pattern_buffer *bufp;
+     const char *string;
+     int size, pos;
+     struct re_registers *regs;
+ {
+  return re_match_2 (bufp, NULL, 0, string, size, pos, regs, size);
+}
+#endif /* not emacs */
+
+
+/* re_match_2 matches the compiled pattern in BUFP against the
+   the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
+   and SIZE2, respectively).  We start matching at POS, and stop
+   matching at STOP.
+
+   If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
+   store offsets for the substring each group matched in REGS.  See the
+   documentation for exactly how many groups we fill.
+
+   We return -1 if no match, -2 if an internal error (such as the
+   failure stack overflowing).  Otherwise, we return the length of the
+   matched substring.  */
+
+int
+re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
+     struct re_pattern_buffer *bufp;
+     const char *string1, *string2;
+     int size1, size2;
+     int pos;
+     struct re_registers *regs;
+     int stop;
+{
+  /* General temporaries.  */
+  int mcnt;
+  unsigned char *p1;
+
+  /* Just past the end of the corresponding string.  */
+  const char *end1, *end2;
+
+  /* Pointers into string1 and string2, just past the last characters in
+     each to consider matching.  */
+  const char *end_match_1, *end_match_2;
+
+  /* Where we are in the data, and the end of the current string.  */
+  const char *d, *dend;
+
+  /* Where we are in the pattern, and the end of the pattern.  */
+  unsigned char *p = bufp->buffer;
+  register unsigned char *pend = p + bufp->used;
+
+  /* We use this to map every character in the string.  */
+  char *translate = bufp->translate;
+
+  /* Failure point stack.  Each place that can handle a failure further
+     down the line pushes a failure point on this stack.  It consists of
+     restart, regend, and reg_info for all registers corresponding to
+     the subexpressions we're currently inside, plus the number of such
+     registers, and, finally, two char *'s.  The first char * is where
+     to resume scanning the pattern; the second one is where to resume
+     scanning the strings.  If the latter is zero, the failure point is
+     a ``dummy''; if a failure happens and the failure point is a dummy,
+     it gets discarded and the next next one is tried.  */
+  fail_stack_type fail_stack;
+#ifdef DEBUG
+  static unsigned failure_id = 0;
+  unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
+#endif
+
+  /* We fill all the registers internally, independent of what we
+     return, for use in backreferences.  The number here includes
+     an element for register zero.  */
+  unsigned num_regs = bufp->re_nsub + 1;
+
+  /* The currently active registers.  */
+  unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
+  unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
+
+  /* Information on the contents of registers. These are pointers into
+     the input strings; they record just what was matched (on this
+     attempt) by a subexpression part of the pattern, that is, the
+     regnum-th regstart pointer points to where in the pattern we began
+     matching and the regnum-th regend points to right after where we
+     stopped matching the regnum-th subexpression.  (The zeroth register
+     keeps track of what the whole pattern matches.)  */
+  const char **regstart = NULL, **regend = NULL;
+
+  /* If a group that's operated upon by a repetition operator fails to
+     match anything, then the register for its start will need to be
+     restored because it will have been set to wherever in the string we
+     are when we last see its open-group operator.  Similarly for a
+     register's end.  */
+  const char **old_regstart = NULL, **old_regend = NULL;
+
+  /* The is_active field of reg_info helps us keep track of which (possibly
+     nested) subexpressions we are currently in. The matched_something
+     field of reg_info[reg_num] helps us tell whether or not we have
+     matched any of the pattern so far this time through the reg_num-th
+     subexpression.  These two fields get reset each time through any
+     loop their register is in.  */
+  register_info_type *reg_info = NULL;
+
+  /* The following record the register info as found in the above
+     variables when we find a match better than any we've seen before.
+     This happens as we backtrack through the failure points, which in
+     turn happens only if we have not yet matched the entire string. */
+  unsigned best_regs_set = false;
+  const char **best_regstart = NULL, **best_regend = NULL;
+
+  /* Logically, this is `best_regend[0]'.  But we don't want to have to
+     allocate space for that if we're not allocating space for anything
+     else (see below).  Also, we never need info about register 0 for
+     any of the other register vectors, and it seems rather a kludge to
+     treat `best_regend' differently than the rest.  So we keep track of
+     the end of the best match so far in a separate variable.  We
+     initialize this to NULL so that when we backtrack the first time
+     and need to test it, it's not garbage.  */
+  const char *match_end = NULL;
+
+  /* Used when we pop values we don't care about.  */
+  const char **reg_dummy = NULL;
+  register_info_type *reg_info_dummy = NULL;
+
+#ifdef DEBUG
+  /* Counts the total number of registers pushed.  */
+  unsigned num_regs_pushed = 0;
+#endif
+
+  DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
+
+  INIT_FAIL_STACK ();
+
+  /* Do not bother to initialize all the register variables if there are
+     no groups in the pattern, as it takes a fair amount of time.  If
+     there are groups, we include space for register 0 (the whole
+     pattern), even though we never use it, since it simplifies the
+     array indexing.  We should fix this.  */
+  if (bufp->re_nsub)
+    {
+      regstart = REGEX_TALLOC (num_regs, const char *);
+      regend = REGEX_TALLOC (num_regs, const char *);
+      old_regstart = REGEX_TALLOC (num_regs, const char *);
+      old_regend = REGEX_TALLOC (num_regs, const char *);
+      best_regstart = REGEX_TALLOC (num_regs, const char *);
+      best_regend = REGEX_TALLOC (num_regs, const char *);
+      reg_info = REGEX_TALLOC (num_regs, register_info_type);
+      reg_dummy = REGEX_TALLOC (num_regs, const char *);
+      reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
+
+      if (!(regstart && regend && old_regstart && old_regend && reg_info
+	    && best_regstart && best_regend && reg_dummy && reg_info_dummy))
+	{
+	  FREE_VARIABLES ();
+	  return -2;
+	}
+    }
+#ifdef REGEX_MALLOC
+  else
+    {
+      /* We must initialize all our variables to NULL, so that
+	 `FREE_VARIABLES' doesn't try to free them.  */
+      regstart = regend = old_regstart = old_regend = best_regstart
+	= best_regend = reg_dummy = NULL;
+      reg_info = reg_info_dummy = (register_info_type *) NULL;
+    }
+#endif /* REGEX_MALLOC */
+
+  /* The starting position is bogus.  */
+  if (pos < 0 || pos > size1 + size2)
+    {
+      FREE_VARIABLES ();
+      return -1;
+    }
+
+  /* Initialize subexpression text positions to -1 to mark ones that no
+     start_memory/stop_memory has been seen for. Also initialize the
+     register information struct.  */
+  for (mcnt = 1; mcnt < num_regs; mcnt++)
+    {
+      regstart[mcnt] = regend[mcnt]
+	= old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
+
+      REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
+      IS_ACTIVE (reg_info[mcnt]) = 0;
+      MATCHED_SOMETHING (reg_info[mcnt]) = 0;
+      EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
+    }
+
+  /* We move `string1' into `string2' if the latter's empty -- but not if
+     `string1' is null.  */
+  if (size2 == 0 && string1 != NULL)
+    {
+      string2 = string1;
+      size2 = size1;
+      string1 = 0;
+      size1 = 0;
+    }
+  end1 = string1 + size1;
+  end2 = string2 + size2;
+
+  /* Compute where to stop matching, within the two strings.  */
+  if (stop <= size1)
+    {
+      end_match_1 = string1 + stop;
+      end_match_2 = string2;
+    }
+  else
+    {
+      end_match_1 = end1;
+      end_match_2 = string2 + stop - size1;
+    }
+
+  /* `p' scans through the pattern as `d' scans through the data.
+     `dend' is the end of the input string that `d' points within.  `d'
+     is advanced into the following input string whenever necessary, but
+     this happens before fetching; therefore, at the beginning of the
+     loop, `d' can be pointing at the end of a string, but it cannot
+     equal `string2'.  */
+  if (size1 > 0 && pos <= size1)
+    {
+      d = string1 + pos;
+      dend = end_match_1;
+    }
+  else
+    {
+      d = string2 + pos - size1;
+      dend = end_match_2;
+    }
+
+  DEBUG_PRINT1 ("The compiled pattern is: ");
+  DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
+  DEBUG_PRINT1 ("The string to match is: `");
+  DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
+  DEBUG_PRINT1 ("'\n");
+
+  /* This loops over pattern commands.  It exits by returning from the
+     function if the match is complete, or it drops through if the match
+     fails at this starting point in the input data.  */
+  for (;;)
+    {
+      DEBUG_PRINT2 ("\n0x%x: ", p);
+
+      if (p == pend)
+	{ /* End of pattern means we might have succeeded.  */
+	  DEBUG_PRINT1 ("end of pattern ... ");
+
+	  /* If we haven't matched the entire string, and we want the
+	     longest match, try backtracking.  */
+	  if (d != end_match_2)
+	    {
+	      DEBUG_PRINT1 ("backtracking.\n");
+
+	      if (!FAIL_STACK_EMPTY ())
+		{ /* More failure points to try.  */
+		  boolean same_str_p = (FIRST_STRING_P (match_end)
+					== MATCHING_IN_FIRST_STRING);
+
+		  /* If exceeds best match so far, save it.  */
+		  if (!best_regs_set
+		      || (same_str_p && d > match_end)
+		      || (!same_str_p && !MATCHING_IN_FIRST_STRING))
+		    {
+		      best_regs_set = true;
+		      match_end = d;
+
+		      DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
+
+		      for (mcnt = 1; mcnt < num_regs; mcnt++)
+			{
+			  best_regstart[mcnt] = regstart[mcnt];
+			  best_regend[mcnt] = regend[mcnt];
+			}
+		    }
+		  goto fail;
+		}
+
+	      /* If no failure points, don't restore garbage.  */
+	      else if (best_regs_set)
+		{
+		restore_best_regs:
+		  /* Restore best match.  It may happen that `dend ==
+		     end_match_1' while the restored d is in string2.
+		     For example, the pattern `x.*y.*z' against the
+		     strings `x-' and `y-z-', if the two strings are
+		     not consecutive in memory.  */
+		  DEBUG_PRINT1 ("Restoring best registers.\n");
+
+		  d = match_end;
+		  dend = ((d >= string1 && d <= end1)
+			   ? end_match_1 : end_match_2);
+
+		  for (mcnt = 1; mcnt < num_regs; mcnt++)
+		    {
+		      regstart[mcnt] = best_regstart[mcnt];
+		      regend[mcnt] = best_regend[mcnt];
+		    }
+		}
+	    } /* d != end_match_2 */
+
+	  DEBUG_PRINT1 ("Accepting match.\n");
+
+	  /* If caller wants register contents data back, do it.  */
+	  if (regs && !bufp->no_sub)
+	    {
+	      /* Have the register data arrays been allocated?  */
+	      if (bufp->regs_allocated == REGS_UNALLOCATED)
+		{ /* No.  So allocate them with malloc.  We need one
+		     extra element beyond `num_regs' for the `-1' marker
+		     GNU code uses.  */
+		  regs->num_regs = MAX (RE_NREGS, num_regs + 1);
+		  regs->start = TALLOC (regs->num_regs, regoff_t);
+		  regs->end = TALLOC (regs->num_regs, regoff_t);
+		  if (regs->start == NULL || regs->end == NULL)
+		    return -2;
+		  bufp->regs_allocated = REGS_REALLOCATE;
+		}
+	      else if (bufp->regs_allocated == REGS_REALLOCATE)
+		{ /* Yes.  If we need more elements than were already
+		     allocated, reallocate them.  If we need fewer, just
+		     leave it alone.  */
+		  if (regs->num_regs < num_regs + 1)
+		    {
+		      regs->num_regs = num_regs + 1;
+		      RETALLOC (regs->start, regs->num_regs, regoff_t);
+		      RETALLOC (regs->end, regs->num_regs, regoff_t);
+		      if (regs->start == NULL || regs->end == NULL)
+			return -2;
+		    }
+		}
+	      else
+		assert (bufp->regs_allocated == REGS_FIXED);
+
+	      /* Convert the pointer data in `regstart' and `regend' to
+		 indices.  Register zero has to be set differently,
+		 since we haven't kept track of any info for it.  */
+	      if (regs->num_regs > 0)
+		{
+		  regs->start[0] = pos;
+		  regs->end[0] = (MATCHING_IN_FIRST_STRING ? d - string1
+				  : d - string2 + size1);
+		}
+
+	      /* Go through the first `min (num_regs, regs->num_regs)'
+		 registers, since that is all we initialized.  */
+	      for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
+		{
+		  if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
+		    regs->start[mcnt] = regs->end[mcnt] = -1;
+		  else
+		    {
+		      regs->start[mcnt] = POINTER_TO_OFFSET (regstart[mcnt]);
+		      regs->end[mcnt] = POINTER_TO_OFFSET (regend[mcnt]);
+		    }
+		}
+
+	      /* If the regs structure we return has more elements than
+		 were in the pattern, set the extra elements to -1.  If
+		 we (re)allocated the registers, this is the case,
+		 because we always allocate enough to have at least one
+		 -1 at the end.  */
+	      for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
+		regs->start[mcnt] = regs->end[mcnt] = -1;
+	    } /* regs && !bufp->no_sub */
+
+	  FREE_VARIABLES ();
+	  DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
+			nfailure_points_pushed, nfailure_points_popped,
+			nfailure_points_pushed - nfailure_points_popped);
+	  DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
+
+	  mcnt = d - pos - (MATCHING_IN_FIRST_STRING
+			    ? string1
+			    : string2 - size1);
+
+	  DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
+
+	  return mcnt;
+	}
+
+      /* Otherwise match next pattern command.  */
+#ifdef SWITCH_ENUM_BUG
+      switch ((int) ((re_opcode_t) *p++))
+#else
+      switch ((re_opcode_t) *p++)
+#endif
+	{
+	/* Ignore these.  Used to ignore the n of succeed_n's which
+	   currently have n == 0.  */
+	case no_op:
+	  DEBUG_PRINT1 ("EXECUTING no_op.\n");
+	  break;
+
+
+	/* Match the next n pattern characters exactly.  The following
+	   byte in the pattern defines n, and the n bytes after that
+	   are the characters to match.  */
+	case exactn:
+	  mcnt = *p++;
+	  DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
+
+	  /* This is written out as an if-else so we don't waste time
+	     testing `translate' inside the loop.  */
+	  if (translate)
+	    {
+	      do
+		{
+		  PREFETCH ();
+		  if (translate[(unsigned char) *d++] != (char) *p++)
+		    goto fail;
+		}
+	      while (--mcnt);
+	    }
+	  else
+	    {
+	      do
+		{
+		  PREFETCH ();
+		  if (*d++ != (char) *p++) goto fail;
+		}
+	      while (--mcnt);
+	    }
+	  SET_REGS_MATCHED ();
+	  break;
+
+
+	/* Match any character except possibly a newline or a null.  */
+	case anychar:
+	  DEBUG_PRINT1 ("EXECUTING anychar.\n");
+
+	  PREFETCH ();
+
+	  if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
+	      || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
+	    goto fail;
+
+	  SET_REGS_MATCHED ();
+	  DEBUG_PRINT2 ("  Matched `%d'.\n", *d);
+	  d++;
+	  break;
+
+
+	case charset:
+	case charset_not:
+	  {
+	    register unsigned char c;
+	    boolean not = (re_opcode_t) *(p - 1) == charset_not;
+
+	    DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
+
+	    PREFETCH ();
+	    c = TRANSLATE (*d); /* The character to match.  */
+
+	    /* Cast to `unsigned' instead of `unsigned char' in case the
+	       bit list is a full 32 bytes long.  */
+	    if (c < (unsigned) (*p * BYTEWIDTH)
+		&& p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
+	      not = !not;
+
+	    p += 1 + *p;
+
+	    if (!not) goto fail;
+
+	    SET_REGS_MATCHED ();
+	    d++;
+	    break;
+	  }
+
+
+	/* The beginning of a group is represented by start_memory.
+	   The arguments are the register number in the next byte, and the
+	   number of groups inner to this one in the next.  The text
+	   matched within the group is recorded (in the internal
+	   registers data structure) under the register number.  */
+	case start_memory:
+	  DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
+
+	  /* Find out if this group can match the empty string.  */
+	  p1 = p;		/* To send to group_match_null_string_p.  */
+
+	  if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
+	    REG_MATCH_NULL_STRING_P (reg_info[*p])
+	      = group_match_null_string_p (&p1, pend, reg_info);
+
+	  /* Save the position in the string where we were the last time
+	     we were at this open-group operator in case the group is
+	     operated upon by a repetition operator, e.g., with `(a*)*b'
+	     against `ab'; then we want to ignore where we are now in
+	     the string in case this attempt to match fails.  */
+	  old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
+			     ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
+			     : regstart[*p];
+	  DEBUG_PRINT2 ("  old_regstart: %d\n",
+			 POINTER_TO_OFFSET (old_regstart[*p]));
+
+	  regstart[*p] = d;
+	  DEBUG_PRINT2 ("  regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
+
+	  IS_ACTIVE (reg_info[*p]) = 1;
+	  MATCHED_SOMETHING (reg_info[*p]) = 0;
+
+	  /* This is the new highest active register.  */
+	  highest_active_reg = *p;
+
+	  /* If nothing was active before, this is the new lowest active
+	     register.  */
+	  if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
+	    lowest_active_reg = *p;
+
+	  /* Move past the register number and inner group count.  */
+	  p += 2;
+	  break;
+
+
+	/* The stop_memory opcode represents the end of a group.  Its
+	   arguments are the same as start_memory's: the register
+	   number, and the number of inner groups.  */
+	case stop_memory:
+	  DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
+
+	  /* We need to save the string position the last time we were at
+	     this close-group operator in case the group is operated
+	     upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
+	     against `aba'; then we want to ignore where we are now in
+	     the string in case this attempt to match fails.  */
+	  old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
+			   ? REG_UNSET (regend[*p]) ? d : regend[*p]
+			   : regend[*p];
+	  DEBUG_PRINT2 ("      old_regend: %d\n",
+			 POINTER_TO_OFFSET (old_regend[*p]));
+
+	  regend[*p] = d;
+	  DEBUG_PRINT2 ("      regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
+
+	  /* This register isn't active anymore.  */
+	  IS_ACTIVE (reg_info[*p]) = 0;
+
+	  /* If this was the only register active, nothing is active
+	     anymore.  */
+	  if (lowest_active_reg == highest_active_reg)
+	    {
+	      lowest_active_reg = NO_LOWEST_ACTIVE_REG;
+	      highest_active_reg = NO_HIGHEST_ACTIVE_REG;
+	    }
+	  else
+	    { /* We must scan for the new highest active register, since
+		 it isn't necessarily one less than now: consider
+		 (a(b)c(d(e)f)g).  When group 3 ends, after the f), the
+		 new highest active register is 1.  */
+	      unsigned char r = *p - 1;
+	      while (r > 0 && !IS_ACTIVE (reg_info[r]))
+		r--;
+
+	      /* If we end up at register zero, that means that we saved
+		 the registers as the result of an `on_failure_jump', not
+		 a `start_memory', and we jumped to past the innermost
+		 `stop_memory'.  For example, in ((.)*) we save
+		 registers 1 and 2 as a result of the *, but when we pop
+		 back to the second ), we are at the stop_memory 1.
+		 Thus, nothing is active.  */
+	      if (r == 0)
+		{
+		  lowest_active_reg = NO_LOWEST_ACTIVE_REG;
+		  highest_active_reg = NO_HIGHEST_ACTIVE_REG;
+		}
+	      else
+		highest_active_reg = r;
+	    }
+
+	  /* If just failed to match something this time around with a
+	     group that's operated on by a repetition operator, try to
+	     force exit from the ``loop'', and restore the register
+	     information for this group that we had before trying this
+	     last match.  */
+	  if ((!MATCHED_SOMETHING (reg_info[*p])
+	       || (re_opcode_t) p[-3] == start_memory)
+	      && (p + 2) < pend)
+	    {
+	      boolean is_a_jump_n = false;
+
+	      p1 = p + 2;
+	      mcnt = 0;
+	      switch ((re_opcode_t) *p1++)
+		{
+		  case jump_n:
+		    is_a_jump_n = true;
+		  case pop_failure_jump:
+		  case maybe_pop_jump:
+		  case jump:
+		  case dummy_failure_jump:
+		    EXTRACT_NUMBER_AND_INCR (mcnt, p1);
+		    if (is_a_jump_n)
+		      p1 += 2;
+		    break;
+
+		  default:
+		    /* do nothing */ ;
+		}
+	      p1 += mcnt;
+
+	      /* If the next operation is a jump backwards in the pattern
+		 to an on_failure_jump right before the start_memory
+		 corresponding to this stop_memory, exit from the loop
+		 by forcing a failure after pushing on the stack the
+		 on_failure_jump's jump in the pattern, and d.  */
+	      if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
+		  && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
+		{
+		  /* If this group ever matched anything, then restore
+		     what its registers were before trying this last
+		     failed match, e.g., with `(a*)*b' against `ab' for
+		     regstart[1], and, e.g., with `((a*)*(b*)*)*'
+		     against `aba' for regend[3].
+
+		     Also restore the registers for inner groups for,
+		     e.g., `((a*)(b*))*' against `aba' (register 3 would
+		     otherwise get trashed).  */
+
+		  if (EVER_MATCHED_SOMETHING (reg_info[*p]))
+		    {
+		      unsigned r;
+
+		      EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
+
+		      /* Restore this and inner groups' (if any) registers.  */
+		      for (r = *p; r < *p + *(p + 1); r++)
+			{
+			  regstart[r] = old_regstart[r];
+
+			  /* xx why this test?  */
+			  if ((int) old_regend[r] >= (int) regstart[r])
+			    regend[r] = old_regend[r];
+			}
+		    }
+		  p1++;
+		  EXTRACT_NUMBER_AND_INCR (mcnt, p1);
+		  PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
+
+		  goto fail;
+		}
+	    }
+
+	  /* Move past the register number and the inner group count.  */
+	  p += 2;
+	  break;
+
+
+	/* \<digit> has been turned into a `duplicate' command which is
+	   followed by the numeric value of <digit> as the register number.  */
+	case duplicate:
+	  {
+	    register const char *d2, *dend2;
+	    int regno = *p++;   /* Get which register to match against.  */
+	    DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
+
+	    /* Can't back reference a group which we've never matched.  */
+	    if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
+	      goto fail;
+
+	    /* Where in input to try to start matching.  */
+	    d2 = regstart[regno];
+
+	    /* Where to stop matching; if both the place to start and
+	       the place to stop matching are in the same string, then
+	       set to the place to stop, otherwise, for now have to use
+	       the end of the first string.  */
+
+	    dend2 = ((FIRST_STRING_P (regstart[regno])
+		      == FIRST_STRING_P (regend[regno]))
+		     ? regend[regno] : end_match_1);
+	    for (;;)
+	      {
+		/* If necessary, advance to next segment in register
+		   contents.  */
+		while (d2 == dend2)
+		  {
+		    if (dend2 == end_match_2) break;
+		    if (dend2 == regend[regno]) break;
+
+		    /* End of string1 => advance to string2. */
+		    d2 = string2;
+		    dend2 = regend[regno];
+		  }
+		/* At end of register contents => success */
+		if (d2 == dend2) break;
+
+		/* If necessary, advance to next segment in data.  */
+		PREFETCH ();
+
+		/* How many characters left in this segment to match.  */
+		mcnt = dend - d;
+
+		/* Want how many consecutive characters we can match in
+		   one shot, so, if necessary, adjust the count.  */
+		if (mcnt > dend2 - d2)
+		  mcnt = dend2 - d2;
+
+		/* Compare that many; failure if mismatch, else move
+		   past them.  */
+		if (translate
+		    ? bcmp_translate (d, d2, mcnt, translate)
+		    : bcmp (d, d2, mcnt))
+		  goto fail;
+		d += mcnt, d2 += mcnt;
+	      }
+	  }
+	  break;
+
+
+	/* begline matches the empty string at the beginning of the string
+	   (unless `not_bol' is set in `bufp'), and, if
+	   `newline_anchor' is set, after newlines.  */
+	case begline:
+	  DEBUG_PRINT1 ("EXECUTING begline.\n");
+
+	  if (AT_STRINGS_BEG (d))
+	    {
+	      if (!bufp->not_bol) break;
+	    }
+	  else if (d[-1] == '\n' && bufp->newline_anchor)
+	    {
+	      break;
+	    }
+	  /* In all other cases, we fail.  */
+	  goto fail;
+
+
+	/* endline is the dual of begline.  */
+	case endline:
+	  DEBUG_PRINT1 ("EXECUTING endline.\n");
+
+	  if (AT_STRINGS_END (d))
+	    {
+	      if (!bufp->not_eol) break;
+	    }
+
+	  /* We have to ``prefetch'' the next character.  */
+	  else if ((d == end1 ? *string2 : *d) == '\n'
+		   && bufp->newline_anchor)
+	    {
+	      break;
+	    }
+	  goto fail;
+
+
+	/* Match at the very beginning of the data.  */
+	case begbuf:
+	  DEBUG_PRINT1 ("EXECUTING begbuf.\n");
+	  if (AT_STRINGS_BEG (d))
+	    break;
+	  goto fail;
+
+
+	/* Match at the very end of the data.  */
+	case endbuf:
+	  DEBUG_PRINT1 ("EXECUTING endbuf.\n");
+	  if (AT_STRINGS_END (d))
+	    break;
+	  goto fail;
+
+
+	/* on_failure_keep_string_jump is used to optimize `.*\n'.  It
+	   pushes NULL as the value for the string on the stack.  Then
+	   `pop_failure_point' will keep the current value for the
+	   string, instead of restoring it.  To see why, consider
+	   matching `foo\nbar' against `.*\n'.  The .* matches the foo;
+	   then the . fails against the \n.  But the next thing we want
+	   to do is match the \n against the \n; if we restored the
+	   string value, we would be back at the foo.
+
+	   Because this is used only in specific cases, we don't need to
+	   check all the things that `on_failure_jump' does, to make
+	   sure the right things get saved on the stack.  Hence we don't
+	   share its code.  The only reason to push anything on the
+	   stack at all is that otherwise we would have to change
+	   `anychar's code to do something besides goto fail in this
+	   case; that seems worse than this.  */
+	case on_failure_keep_string_jump:
+	  DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
+
+	  EXTRACT_NUMBER_AND_INCR (mcnt, p);
+	  DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
+
+	  PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
+	  break;
+
+
+	/* Uses of on_failure_jump:
+
+	   Each alternative starts with an on_failure_jump that points
+	   to the beginning of the next alternative.  Each alternative
+	   except the last ends with a jump that in effect jumps past
+	   the rest of the alternatives.  (They really jump to the
+	   ending jump of the following alternative, because tensioning
+	   these jumps is a hassle.)
+
+	   Repeats start with an on_failure_jump that points past both
+	   the repetition text and either the following jump or
+	   pop_failure_jump back to this on_failure_jump.  */
+	case on_failure_jump:
+	on_failure:
+	  DEBUG_PRINT1 ("EXECUTING on_failure_jump");
+
+	  EXTRACT_NUMBER_AND_INCR (mcnt, p);
+	  DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
+
+	  /* If this on_failure_jump comes right before a group (i.e.,
+	     the original * applied to a group), save the information
+	     for that group and all inner ones, so that if we fail back
+	     to this point, the group's information will be correct.
+	     For example, in \(a*\)*\1, we need the preceding group,
+	     and in \(\(a*\)b*\)\2, we need the inner group.  */
+
+	  /* We can't use `p' to check ahead because we push
+	     a failure point to `p + mcnt' after we do this.  */
+	  p1 = p;
+
+	  /* We need to skip no_op's before we look for the
+	     start_memory in case this on_failure_jump is happening as
+	     the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
+	     against aba.  */
+	  while (p1 < pend && (re_opcode_t) *p1 == no_op)
+	    p1++;
+
+	  if (p1 < pend && (re_opcode_t) *p1 == start_memory)
+	    {
+	      /* We have a new highest active register now.  This will
+		 get reset at the start_memory we are about to get to,
+		 but we will have saved all the registers relevant to
+		 this repetition op, as described above.  */
+	      highest_active_reg = *(p1 + 1) + *(p1 + 2);
+	      if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
+		lowest_active_reg = *(p1 + 1);
+	    }
+
+	  DEBUG_PRINT1 (":\n");
+	  PUSH_FAILURE_POINT (p + mcnt, d, -2);
+	  break;
+
+
+	/* A smart repeat ends with `maybe_pop_jump'.
+	   We change it to either `pop_failure_jump' or `jump'.  */
+	case maybe_pop_jump:
+	  EXTRACT_NUMBER_AND_INCR (mcnt, p);
+	  DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
+	  {
+	    register unsigned char *p2 = p;
+
+	    /* Compare the beginning of the repeat with what in the
+	       pattern follows its end. If we can establish that there
+	       is nothing that they would both match, i.e., that we
+	       would have to backtrack because of (as in, e.g., `a*a')
+	       then we can change to pop_failure_jump, because we'll
+	       never have to backtrack.
+
+	       This is not true in the case of alternatives: in
+	       `(a|ab)*' we do need to backtrack to the `ab' alternative
+	       (e.g., if the string was `ab').  But instead of trying to
+	       detect that here, the alternative has put on a dummy
+	       failure point which is what we will end up popping.  */
+
+	    /* Skip over open/close-group commands.  */
+	    while (p2 + 2 < pend
+		   && ((re_opcode_t) *p2 == stop_memory
+		       || (re_opcode_t) *p2 == start_memory))
+	      p2 += 3;			/* Skip over args, too.  */
+
+	    /* If we're at the end of the pattern, we can change.  */
+	    if (p2 == pend)
+	      {
+		/* Consider what happens when matching ":\(.*\)"
+		   against ":/".  I don't really understand this code
+		   yet.  */
+		p[-3] = (unsigned char) pop_failure_jump;
+		DEBUG_PRINT1
+		  ("  End of pattern: change to `pop_failure_jump'.\n");
+	      }
+
+	    else if ((re_opcode_t) *p2 == exactn
+		     || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
+	      {
+		register unsigned char c
+		  = *p2 == (unsigned char) endline ? '\n' : p2[2];
+		p1 = p + mcnt;
+
+		/* p1[0] ... p1[2] are the `on_failure_jump' corresponding
+		   to the `maybe_finalize_jump' of this case.  Examine what
+		   follows.  */
+		if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
+		  {
+		    p[-3] = (unsigned char) pop_failure_jump;
+		    DEBUG_PRINT3 ("  %c != %c => pop_failure_jump.\n",
+				  c, p1[5]);
+		  }
+
+		else if ((re_opcode_t) p1[3] == charset
+			 || (re_opcode_t) p1[3] == charset_not)
+		  {
+		    int not = (re_opcode_t) p1[3] == charset_not;
+
+		    if (c < (unsigned char) (p1[4] * BYTEWIDTH)
+			&& p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
+		      not = !not;
+
+		    /* `not' is equal to 1 if c would match, which means
+			that we can't change to pop_failure_jump.  */
+		    if (!not)
+		      {
+			p[-3] = (unsigned char) pop_failure_jump;
+			DEBUG_PRINT1 ("  No match => pop_failure_jump.\n");
+		      }
+		  }
+	      }
+	  }
+	  p -= 2;		/* Point at relative address again.  */
+	  if ((re_opcode_t) p[-1] != pop_failure_jump)
+	    {
+	      p[-1] = (unsigned char) jump;
+	      DEBUG_PRINT1 ("  Match => jump.\n");
+	      goto unconditional_jump;
+	    }
+	/* Note fall through.  */
+
+
+	/* The end of a simple repeat has a pop_failure_jump back to
+	   its matching on_failure_jump, where the latter will push a
+	   failure point.  The pop_failure_jump takes off failure
+	   points put on by this pop_failure_jump's matching
+	   on_failure_jump; we got through the pattern to here from the
+	   matching on_failure_jump, so didn't fail.  */
+	case pop_failure_jump:
+	  {
+	    /* We need to pass separate storage for the lowest and
+	       highest registers, even though we don't care about the
+	       actual values.  Otherwise, we will restore only one
+	       register from the stack, since lowest will == highest in
+	       `pop_failure_point'.  */
+	    unsigned dummy_low_reg, dummy_high_reg;
+	    unsigned char *pdummy;
+	    const char *sdummy;
+
+	    DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
+	    POP_FAILURE_POINT (sdummy, pdummy,
+			       dummy_low_reg, dummy_high_reg,
+			       reg_dummy, reg_dummy, reg_info_dummy);
+	  }
+	  /* Note fall through.  */
+
+
+	/* Unconditionally jump (without popping any failure points).  */
+	case jump:
+	unconditional_jump:
+	  EXTRACT_NUMBER_AND_INCR (mcnt, p);	/* Get the amount to jump.  */
+	  DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
+	  p += mcnt;				/* Do the jump.  */
+	  DEBUG_PRINT2 ("(to 0x%x).\n", p);
+	  break;
+
+
+	/* We need this opcode so we can detect where alternatives end
+	   in `group_match_null_string_p' et al.  */
+	case jump_past_alt:
+	  DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
+	  goto unconditional_jump;
+
+
+	/* Normally, the on_failure_jump pushes a failure point, which
+	   then gets popped at pop_failure_jump.  We will end up at
+	   pop_failure_jump, also, and with a pattern of, say, `a+', we
+	   are skipping over the on_failure_jump, so we have to push
+	   something meaningless for pop_failure_jump to pop.  */
+	case dummy_failure_jump:
+	  DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
+	  /* It doesn't matter what we push for the string here.  What
+	     the code at `fail' tests is the value for the pattern.  */
+	  PUSH_FAILURE_POINT (0, 0, -2);
+	  goto unconditional_jump;
+
+
+	/* At the end of an alternative, we need to push a dummy failure
+	   point in case we are followed by a `pop_failure_jump', because
+	   we don't want the failure point for the alternative to be
+	   popped.  For example, matching `(a|ab)*' against `aab'
+	   requires that we match the `ab' alternative.  */
+	case push_dummy_failure:
+	  DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
+	  /* See comments just above at `dummy_failure_jump' about the
+	     two zeroes.  */
+	  PUSH_FAILURE_POINT (0, 0, -2);
+	  break;
+
+	/* Have to succeed matching what follows at least n times.
+	   After that, handle like `on_failure_jump'.  */
+	case succeed_n:
+	  EXTRACT_NUMBER (mcnt, p + 2);
+	  DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
+
+	  assert (mcnt >= 0);
+	  /* Originally, this is how many times we HAVE to succeed.  */
+	  if (mcnt > 0)
+	    {
+	       mcnt--;
+	       p += 2;
+	       STORE_NUMBER_AND_INCR (p, mcnt);
+	       DEBUG_PRINT3 ("  Setting 0x%x to %d.\n", p, mcnt);
+	    }
+	  else if (mcnt == 0)
+	    {
+	      DEBUG_PRINT2 ("  Setting two bytes from 0x%x to no_op.\n", p+2);
+	      p[2] = (unsigned char) no_op;
+	      p[3] = (unsigned char) no_op;
+	      goto on_failure;
+	    }
+	  break;
+
+	case jump_n:
+	  EXTRACT_NUMBER (mcnt, p + 2);
+	  DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
+
+	  /* Originally, this is how many times we CAN jump.  */
+	  if (mcnt)
+	    {
+	       mcnt--;
+	       STORE_NUMBER (p + 2, mcnt);
+	       goto unconditional_jump;
+	    }
+	  /* If don't have to jump any more, skip over the rest of command.  */
+	  else
+	    p += 4;
+	  break;
+
+	case set_number_at:
+	  {
+	    DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
+
+	    EXTRACT_NUMBER_AND_INCR (mcnt, p);
+	    p1 = p + mcnt;
+	    EXTRACT_NUMBER_AND_INCR (mcnt, p);
+	    DEBUG_PRINT3 ("  Setting 0x%x to %d.\n", p1, mcnt);
+	    STORE_NUMBER (p1, mcnt);
+	    break;
+	  }
+
+	case wordbound:
+	  DEBUG_PRINT1 ("EXECUTING wordbound.\n");
+	  if (AT_WORD_BOUNDARY (d))
+	    break;
+	  goto fail;
+
+	case notwordbound:
+	  DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
+	  if (AT_WORD_BOUNDARY (d))
+	    goto fail;
+	  break;
+
+	case wordbeg:
+	  DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
+	  if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
+	    break;
+	  goto fail;
+
+	case wordend:
+	  DEBUG_PRINT1 ("EXECUTING wordend.\n");
+	  if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
+	      && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
+	    break;
+	  goto fail;
+
+#ifdef emacs
+#ifdef emacs19
+	case before_dot:
+	  DEBUG_PRINT1 ("EXECUTING before_dot.\n");
+	  if (PTR_CHAR_POS ((unsigned char *) d) >= point)
+	    goto fail;
+	  break;
+
+	case at_dot:
+	  DEBUG_PRINT1 ("EXECUTING at_dot.\n");
+	  if (PTR_CHAR_POS ((unsigned char *) d) != point)
+	    goto fail;
+	  break;
+
+	case after_dot:
+	  DEBUG_PRINT1 ("EXECUTING after_dot.\n");
+	  if (PTR_CHAR_POS ((unsigned char *) d) <= point)
+	    goto fail;
+	  break;
+#else /* not emacs19 */
+	case at_dot:
+	  DEBUG_PRINT1 ("EXECUTING at_dot.\n");
+	  if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point)
+	    goto fail;
+	  break;
+#endif /* not emacs19 */
+
+	case syntaxspec:
+	  DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
+	  mcnt = *p++;
+	  goto matchsyntax;
+
+	case wordchar:
+	  DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
+	  mcnt = (int) Sword;
+	matchsyntax:
+	  PREFETCH ();
+	  if (SYNTAX (*d++) != (enum syntaxcode) mcnt)
+	    goto fail;
+	  SET_REGS_MATCHED ();
+	  break;
+
+	case notsyntaxspec:
+	  DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
+	  mcnt = *p++;
+	  goto matchnotsyntax;
+
+	case notwordchar:
+	  DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
+	  mcnt = (int) Sword;
+	matchnotsyntax:
+	  PREFETCH ();
+	  if (SYNTAX (*d++) == (enum syntaxcode) mcnt)
+	    goto fail;
+	  SET_REGS_MATCHED ();
+	  break;
+
+#else /* not emacs */
+	case wordchar:
+	  DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
+	  PREFETCH ();
+	  if (!WORDCHAR_P (d))
+	    goto fail;
+	  SET_REGS_MATCHED ();
+	  d++;
+	  break;
+
+	case notwordchar:
+	  DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
+	  PREFETCH ();
+	  if (WORDCHAR_P (d))
+	    goto fail;
+	  SET_REGS_MATCHED ();
+	  d++;
+	  break;
+#endif /* not emacs */
+
+	default:
+	  abort ();
+	}
+      continue;  /* Successfully executed one pattern command; keep going.  */
+
+
+    /* We goto here if a matching operation fails. */
+    fail:
+      if (!FAIL_STACK_EMPTY ())
+	{ /* A restart point is known.  Restore to that state.  */
+	  DEBUG_PRINT1 ("\nFAIL:\n");
+	  POP_FAILURE_POINT (d, p,
+			     lowest_active_reg, highest_active_reg,
+			     regstart, regend, reg_info);
+
+	  /* If this failure point is a dummy, try the next one.  */
+	  if (!p)
+	    goto fail;
+
+	  /* If we failed to the end of the pattern, don't examine *p.  */
+	  assert (p <= pend);
+	  if (p < pend)
+	    {
+	      boolean is_a_jump_n = false;
+
+	      /* If failed to a backwards jump that's part of a repetition
+		 loop, need to pop this failure point and use the next one.  */
+	      switch ((re_opcode_t) *p)
+		{
+		case jump_n:
+		  is_a_jump_n = true;
+		case maybe_pop_jump:
+		case pop_failure_jump:
+		case jump:
+		  p1 = p + 1;
+		  EXTRACT_NUMBER_AND_INCR (mcnt, p1);
+		  p1 += mcnt;
+
+		  if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
+		      || (!is_a_jump_n
+			  && (re_opcode_t) *p1 == on_failure_jump))
+		    goto fail;
+		  break;
+		default:
+		  /* do nothing */ ;
+		}
+	    }
+
+	  if (d >= string1 && d <= end1)
+	    dend = end_match_1;
+	}
+      else
+	break;   /* Matching at this starting point really fails.  */
+    } /* for (;;) */
+
+  if (best_regs_set)
+    goto restore_best_regs;
+
+  FREE_VARIABLES ();
+
+  return -1;         			/* Failure to match.  */
+} /* re_match_2 */
+
+/* Subroutine definitions for re_match_2.  */
+
+
+/* We are passed P pointing to a register number after a start_memory.
+
+   Return true if the pattern up to the corresponding stop_memory can
+   match the empty string, and false otherwise.
+
+   If we find the matching stop_memory, sets P to point to one past its number.
+   Otherwise, sets P to an undefined byte less than or equal to END.
+
+   We don't handle duplicates properly (yet).  */
+
+static boolean
+group_match_null_string_p (p, end, reg_info)
+    unsigned char **p, *end;
+    register_info_type *reg_info;
+{
+  int mcnt;
+  /* Point to after the args to the start_memory.  */
+  unsigned char *p1 = *p + 2;
+
+  while (p1 < end)
+    {
+      /* Skip over opcodes that can match nothing, and return true or
+	 false, as appropriate, when we get to one that can't, or to the
+	 matching stop_memory.  */
+
+      switch ((re_opcode_t) *p1)
+	{
+	/* Could be either a loop or a series of alternatives.  */
+	case on_failure_jump:
+	  p1++;
+	  EXTRACT_NUMBER_AND_INCR (mcnt, p1);
+
+	  /* If the next operation is not a jump backwards in the
+	     pattern.  */
+
+	  if (mcnt >= 0)
+	    {
+	      /* Go through the on_failure_jumps of the alternatives,
+		 seeing if any of the alternatives cannot match nothing.
+		 The last alternative starts with only a jump,
+		 whereas the rest start with on_failure_jump and end
+		 with a jump, e.g., here is the pattern for `a|b|c':
+
+		 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
+		 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
+		 /exactn/1/c
+
+		 So, we have to first go through the first (n-1)
+		 alternatives and then deal with the last one separately.  */
+
+
+	      /* Deal with the first (n-1) alternatives, which start
+		 with an on_failure_jump (see above) that jumps to right
+		 past a jump_past_alt.  */
+
+	      while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
+		{
+		  /* `mcnt' holds how many bytes long the alternative
+		     is, including the ending `jump_past_alt' and
+		     its number.  */
+
+		  if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
+						      reg_info))
+		    return false;
+
+		  /* Move to right after this alternative, including the
+		     jump_past_alt.  */
+		  p1 += mcnt;
+
+		  /* Break if it's the beginning of an n-th alternative
+		     that doesn't begin with an on_failure_jump.  */
+		  if ((re_opcode_t) *p1 != on_failure_jump)
+		    break;
+
+		  /* Still have to check that it's not an n-th
+		     alternative that starts with an on_failure_jump.  */
+		  p1++;
+		  EXTRACT_NUMBER_AND_INCR (mcnt, p1);
+		  if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
+		    {
+		      /* Get to the beginning of the n-th alternative.  */
+		      p1 -= 3;
+		      break;
+		    }
+		}
+
+	      /* Deal with the last alternative: go back and get number
+		 of the `jump_past_alt' just before it.  `mcnt' contains
+		 the length of the alternative.  */
+	      EXTRACT_NUMBER (mcnt, p1 - 2);
+
+	      if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
+		return false;
+
+	      p1 += mcnt;	/* Get past the n-th alternative.  */
+	    } /* if mcnt > 0 */
+	  break;
+
+
+	case stop_memory:
+	  assert (p1[1] == **p);
+	  *p = p1 + 2;
+	  return true;
+
+
+	default:
+	  if (!common_op_match_null_string_p (&p1, end, reg_info))
+	    return false;
+	}
+    } /* while p1 < end */
+
+  return false;
+} /* group_match_null_string_p */
+
+
+/* Similar to group_match_null_string_p, but doesn't deal with alternatives:
+   It expects P to be the first byte of a single alternative and END one
+   byte past the last. The alternative can contain groups.  */
+
+static boolean
+alt_match_null_string_p (p, end, reg_info)
+    unsigned char *p, *end;
+    register_info_type *reg_info;
+{
+  int mcnt;
+  unsigned char *p1 = p;
+
+  while (p1 < end)
+    {
+      /* Skip over opcodes that can match nothing, and break when we get
+	 to one that can't.  */
+
+      switch ((re_opcode_t) *p1)
+	{
+	/* It's a loop.  */
+	case on_failure_jump:
+	  p1++;
+	  EXTRACT_NUMBER_AND_INCR (mcnt, p1);
+	  p1 += mcnt;
+	  break;
+
+	default:
+	  if (!common_op_match_null_string_p (&p1, end, reg_info))
+	    return false;
+	}
+    }  /* while p1 < end */
+
+  return true;
+} /* alt_match_null_string_p */
+
+
+/* Deals with the ops common to group_match_null_string_p and
+   alt_match_null_string_p.
+
+   Sets P to one after the op and its arguments, if any.  */
+
+static boolean
+common_op_match_null_string_p (p, end, reg_info)
+    unsigned char **p, *end;
+    register_info_type *reg_info;
+{
+  int mcnt;
+  boolean ret;
+  int reg_no;
+  unsigned char *p1 = *p;
+
+  switch ((re_opcode_t) *p1++)
+    {
+    case no_op:
+    case begline:
+    case endline:
+    case begbuf:
+    case endbuf:
+    case wordbeg:
+    case wordend:
+    case wordbound:
+    case notwordbound:
+#ifdef emacs
+    case before_dot:
+    case at_dot:
+    case after_dot:
+#endif
+      break;
+
+    case start_memory:
+      reg_no = *p1;
+      assert (reg_no > 0 && reg_no <= MAX_REGNUM);
+      ret = group_match_null_string_p (&p1, end, reg_info);
+
+      /* Have to set this here in case we're checking a group which
+	 contains a group and a back reference to it.  */
+
+      if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
+	REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
+
+      if (!ret)
+	return false;
+      break;
+
+    /* If this is an optimized succeed_n for zero times, make the jump.  */
+    case jump:
+      EXTRACT_NUMBER_AND_INCR (mcnt, p1);
+      if (mcnt >= 0)
+	p1 += mcnt;
+      else
+	return false;
+      break;
+
+    case succeed_n:
+      /* Get to the number of times to succeed.  */
+      p1 += 2;
+      EXTRACT_NUMBER_AND_INCR (mcnt, p1);
+
+      if (mcnt == 0)
+	{
+	  p1 -= 4;
+	  EXTRACT_NUMBER_AND_INCR (mcnt, p1);
+	  p1 += mcnt;
+	}
+      else
+	return false;
+      break;
+
+    case duplicate:
+      if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
+	return false;
+      break;
+
+    case set_number_at:
+      p1 += 4;
+
+    default:
+      /* All other opcodes mean we cannot match the empty string.  */
+      return false;
+  }
+
+  *p = p1;
+  return true;
+} /* common_op_match_null_string_p */
+
+
+/* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
+   bytes; nonzero otherwise.  */
+
+static int
+bcmp_translate(
+     unsigned char *s1,
+     unsigned char *s2,
+     int len,
+     char *translate
+)
+{
+  register unsigned char *p1 = s1, *p2 = s2;
+  while (len)
+    {
+      if (translate[*p1++] != translate[*p2++]) return 1;
+      len--;
+    }
+  return 0;
+}
+
+/* Entry points for GNU code.  */
+
+/* re_compile_pattern is the GNU regular expression compiler: it
+   compiles PATTERN (of length SIZE) and puts the result in BUFP.
+   Returns 0 if the pattern was valid, otherwise an error string.
+
+   Assumes the `allocated' (and perhaps `buffer') and `translate' fields
+   are set in BUFP on entry.
+
+   We call regex_compile to do the actual compilation.  */
+
+const char *
+re_compile_pattern (pattern, length, bufp)
+     const char *pattern;
+     int length;
+     struct re_pattern_buffer *bufp;
+{
+  reg_errcode_t ret;
+
+  /* GNU code is written to assume at least RE_NREGS registers will be set
+     (and at least one extra will be -1).  */
+  bufp->regs_allocated = REGS_UNALLOCATED;
+
+  /* And GNU code determines whether or not to get register information
+     by passing null for the REGS argument to re_match, etc., not by
+     setting no_sub.  */
+  bufp->no_sub = 0;
+
+  /* Match anchors at newline.  */
+  bufp->newline_anchor = 1;
+
+  ret = regex_compile (pattern, length, re_syntax_options, bufp);
+
+  return re_error_msg[(int) ret];
+}
+
+/* Entry points compatible with 4.2 BSD regex library.  We don't define
+   them if this is an Emacs or POSIX compilation.  */
+
+#if !defined (emacs) && !defined (_POSIX_SOURCE)
+
+/* BSD has one and only one pattern buffer.  */
+static struct re_pattern_buffer re_comp_buf;
+
+char *
+re_comp (s)
+    const char *s;
+{
+  reg_errcode_t ret;
+
+  if (!s)
+    {
+      if (!re_comp_buf.buffer)
+	return "No previous regular expression";
+      return 0;
+    }
+
+  if (!re_comp_buf.buffer)
+    {
+      re_comp_buf.buffer = (unsigned char *) malloc (200);
+      if (re_comp_buf.buffer == NULL)
+	return "Memory exhausted";
+      re_comp_buf.allocated = 200;
+
+      re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
+      if (re_comp_buf.fastmap == NULL)
+	return "Memory exhausted";
+    }
+
+  /* Since `re_exec' always passes NULL for the `regs' argument, we
+     don't need to initialize the pattern buffer fields which affect it.  */
+
+  /* Match anchors at newlines.  */
+  re_comp_buf.newline_anchor = 1;
+
+  ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
+
+  /* Yes, we're discarding `const' here.  */
+  return (char *) re_error_msg[(int) ret];
+}
+
+
+int
+re_exec (s)
+    const char *s;
+{
+  const int len = strlen (s);
+  return
+    0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
+}
+#endif /* not emacs and not _POSIX_SOURCE */
+
+/* POSIX.2 functions.  Don't define these for Emacs.  */
+
+#ifndef emacs
+
+/* regcomp takes a regular expression as a string and compiles it.
+
+   PREG is a regex_t *.  We do not expect any fields to be initialized,
+   since POSIX says we shouldn't.  Thus, we set
+
+     `buffer' to the compiled pattern;
+     `used' to the length of the compiled pattern;
+     `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
+       REG_EXTENDED bit in CFLAGS is set; otherwise, to
+       RE_SYNTAX_POSIX_BASIC;
+     `newline_anchor' to REG_NEWLINE being set in CFLAGS;
+     `fastmap' and `fastmap_accurate' to zero;
+     `re_nsub' to the number of subexpressions in PATTERN.
+
+   PATTERN is the address of the pattern string.
+
+   CFLAGS is a series of bits which affect compilation.
+
+     If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
+     use POSIX basic syntax.
+
+     If REG_NEWLINE is set, then . and [^...] don't match newline.
+     Also, regexec will try a match beginning after every newline.
+
+     If REG_ICASE is set, then we considers upper- and lowercase
+     versions of letters to be equivalent when matching.
+
+     If REG_NOSUB is set, then when PREG is passed to regexec, that
+     routine will report only success or failure, and nothing about the
+     registers.
+
+   It returns 0 if it succeeds, nonzero if it doesn't.  (See regex.h for
+   the return codes and their meanings.)  */
+
+int
+regcomp (preg, pattern, cflags)
+    regex_t *preg;
+    const char *pattern;
+    int cflags;
+{
+  reg_errcode_t ret;
+  unsigned syntax
+    = (cflags & REG_EXTENDED) ?
+      RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
+
+  /* regex_compile will allocate the space for the compiled pattern.  */
+  preg->buffer = 0;
+  preg->allocated = 0;
+
+  /* Don't bother to use a fastmap when searching.  This simplifies the
+     REG_NEWLINE case: if we used a fastmap, we'd have to put all the
+     characters after newlines into the fastmap.  This way, we just try
+     every character.  */
+  preg->fastmap = 0;
+
+  if (cflags & REG_ICASE)
+    {
+      unsigned i;
+
+      preg->translate = (char *) malloc (CHAR_SET_SIZE);
+      if (preg->translate == NULL)
+	return (int) REG_ESPACE;
+
+      /* Map uppercase characters to corresponding lowercase ones.  */
+      for (i = 0; i < CHAR_SET_SIZE; i++)
+	preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
+    }
+  else
+    preg->translate = NULL;
+
+  /* If REG_NEWLINE is set, newlines are treated differently.  */
+  if (cflags & REG_NEWLINE)
+    { /* REG_NEWLINE implies neither . nor [^...] match newline.  */
+      syntax &= ~RE_DOT_NEWLINE;
+      syntax |= RE_HAT_LISTS_NOT_NEWLINE;
+      /* It also changes the matching behavior.  */
+      preg->newline_anchor = 1;
+    }
+  else
+    preg->newline_anchor = 0;
+
+  preg->no_sub = !!(cflags & REG_NOSUB);
+
+  /* POSIX says a null character in the pattern terminates it, so we
+     can use strlen here in compiling the pattern.  */
+  ret = regex_compile (pattern, strlen (pattern), syntax, preg);
+
+  /* POSIX doesn't distinguish between an unmatched open-group and an
+     unmatched close-group: both are REG_EPAREN.  */
+  if (ret == REG_ERPAREN) ret = REG_EPAREN;
+
+  return (int) ret;
+}
+
+
+/* regexec searches for a given pattern, specified by PREG, in the
+   string STRING.
+
+   If NMATCH is zero or REG_NOSUB was set in the cflags argument to
+   `regcomp', we ignore PMATCH.  Otherwise, we assume PMATCH has at
+   least NMATCH elements, and we set them to the offsets of the
+   corresponding matched substrings.
+
+   EFLAGS specifies `execution flags' which affect matching: if
+   REG_NOTBOL is set, then ^ does not match at the beginning of the
+   string; if REG_NOTEOL is set, then $ does not match at the end.
+
+   We return 0 if we find a match and REG_NOMATCH if not.  */
+
+int
+regexec (preg, string, nmatch, pmatch, eflags)
+    const regex_t *preg;
+    const char *string;
+    size_t nmatch;
+    regmatch_t pmatch[];
+    int eflags;
+{
+  int ret;
+  struct re_registers regs;
+  regex_t private_preg;
+  int len = strlen (string);
+  boolean want_reg_info = !preg->no_sub && nmatch > 0;
+
+  private_preg = *preg;
+
+  private_preg.not_bol = !!(eflags & REG_NOTBOL);
+  private_preg.not_eol = !!(eflags & REG_NOTEOL);
+
+  /* The user has told us exactly how many registers to return
+     information about, via `nmatch'.  We have to pass that on to the
+     matching routines.  */
+  private_preg.regs_allocated = REGS_FIXED;
+
+  if (want_reg_info)
+    {
+      regs.num_regs = nmatch;
+      regs.start = TALLOC (nmatch, regoff_t);
+      regs.end = TALLOC (nmatch, regoff_t);
+      if (regs.start == NULL || regs.end == NULL)
+	return (int) REG_NOMATCH;
+    }
+
+  /* Perform the searching operation.  */
+  ret = re_search (&private_preg, string, len,
+		   /* start: */ 0, /* range: */ len,
+		   want_reg_info ? &regs : (struct re_registers *) 0);
+
+  /* Copy the register information to the POSIX structure.  */
+  if (want_reg_info)
+    {
+      if (ret >= 0)
+	{
+	  unsigned r;
+
+	  for (r = 0; r < nmatch; r++)
+	    {
+	      pmatch[r].rm_so = regs.start[r];
+	      pmatch[r].rm_eo = regs.end[r];
+	    }
+	}
+
+      /* If we needed the temporary register info, free the space now.  */
+      free (regs.start);
+      free (regs.end);
+    }
+
+  /* We want zero return to mean success, unlike `re_search'.  */
+  return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
+}
+
+
+/* Returns a message corresponding to an error code, ERRCODE, returned
+   from either regcomp or regexec.   We don't use PREG here.  */
+
+size_t
+regerror(int errcode, const regex_t *preg,
+	 char *errbuf, size_t errbuf_size)
+{
+  const char *msg;
+  size_t msg_size;
+
+  if (errcode < 0
+      || errcode >= (sizeof (re_error_msg) / sizeof (re_error_msg[0])))
+    /* Only error codes returned by the rest of the code should be passed
+       to this routine.  If we are given anything else, or if other regex
+       code generates an invalid error code, then the program has a bug.
+       Dump core so we can fix it.  */
+    abort ();
+
+  msg = re_error_msg[errcode];
+
+  /* POSIX doesn't require that we do anything in this case, but why
+     not be nice.  */
+  if (! msg)
+    msg = "Success";
+
+  msg_size = strlen (msg) + 1; /* Includes the null.  */
+
+  if (errbuf_size != 0)
+    {
+      if (msg_size > errbuf_size)
+	{
+	  strncpy (errbuf, msg, errbuf_size - 1);
+	  errbuf[errbuf_size - 1] = 0;
+	}
+      else
+	strcpy (errbuf, msg);
+    }
+
+  return msg_size;
+}
+
+
+/* Free dynamically allocated space used by PREG.  */
+
+void
+regfree (preg)
+    regex_t *preg;
+{
+  if (preg->buffer != NULL)
+    free (preg->buffer);
+  preg->buffer = NULL;
+
+  preg->allocated = 0;
+  preg->used = 0;
+
+  if (preg->fastmap != NULL)
+    free (preg->fastmap);
+  preg->fastmap = NULL;
+  preg->fastmap_accurate = 0;
+
+  if (preg->translate != NULL)
+    free (preg->translate);
+  preg->translate = NULL;
+}
+
+#endif /* not emacs  */
+
+/*
+Local variables:
+make-backup-files: t
+version-control: t
+trim-versions-without-asking: nil
+End:
+*/
diff --git a/compat/regex/regex.h b/compat/regex/regex.h
new file mode 100644
index 0000000..6eb64f1
--- /dev/null
+++ b/compat/regex/regex.h
@@ -0,0 +1,490 @@
+/* Definitions for data structures and routines for the regular
+   expression library, version 0.12.
+
+   Copyright (C) 1985, 1989, 1990, 1991, 1992, 1993 Free Software Foundation, Inc.
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 2, or (at your option)
+   any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program; if not, write to the Free Software
+   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.  */
+
+#ifndef __REGEXP_LIBRARY_H__
+#define __REGEXP_LIBRARY_H__
+
+/* POSIX says that <sys/types.h> must be included (by the caller) before
+   <regex.h>.  */
+
+#ifdef VMS
+/* VMS doesn't have `size_t' in <sys/types.h>, even though POSIX says it
+   should be there.  */
+#include <stddef.h>
+#endif
+
+
+/* The following bits are used to determine the regexp syntax we
+   recognize.  The set/not-set meanings are chosen so that Emacs syntax
+   remains the value 0.  The bits are given in alphabetical order, and
+   the definitions shifted by one from the previous bit; thus, when we
+   add or remove a bit, only one other definition need change.  */
+typedef unsigned reg_syntax_t;
+
+/* If this bit is not set, then \ inside a bracket expression is literal.
+   If set, then such a \ quotes the following character.  */
+#define RE_BACKSLASH_ESCAPE_IN_LISTS (1)
+
+/* If this bit is not set, then + and ? are operators, and \+ and \? are
+     literals.
+   If set, then \+ and \? are operators and + and ? are literals.  */
+#define RE_BK_PLUS_QM (RE_BACKSLASH_ESCAPE_IN_LISTS << 1)
+
+/* If this bit is set, then character classes are supported.  They are:
+     [:alpha:], [:upper:], [:lower:],  [:digit:], [:alnum:], [:xdigit:],
+     [:space:], [:print:], [:punct:], [:graph:], and [:cntrl:].
+   If not set, then character classes are not supported.  */
+#define RE_CHAR_CLASSES (RE_BK_PLUS_QM << 1)
+
+/* If this bit is set, then ^ and $ are always anchors (outside bracket
+     expressions, of course).
+   If this bit is not set, then it depends:
+	^  is an anchor if it is at the beginning of a regular
+	   expression or after an open-group or an alternation operator;
+	$  is an anchor if it is at the end of a regular expression, or
+	   before a close-group or an alternation operator.
+
+   This bit could be (re)combined with RE_CONTEXT_INDEP_OPS, because
+   POSIX draft 11.2 says that * etc. in leading positions is undefined.
+   We already implemented a previous draft which made those constructs
+   invalid, though, so we haven't changed the code back.  */
+#define RE_CONTEXT_INDEP_ANCHORS (RE_CHAR_CLASSES << 1)
+
+/* If this bit is set, then special characters are always special
+     regardless of where they are in the pattern.
+   If this bit is not set, then special characters are special only in
+     some contexts; otherwise they are ordinary.  Specifically,
+     * + ? and intervals are only special when not after the beginning,
+     open-group, or alternation operator.  */
+#define RE_CONTEXT_INDEP_OPS (RE_CONTEXT_INDEP_ANCHORS << 1)
+
+/* If this bit is set, then *, +, ?, and { cannot be first in an re or
+     immediately after an alternation or begin-group operator.  */
+#define RE_CONTEXT_INVALID_OPS (RE_CONTEXT_INDEP_OPS << 1)
+
+/* If this bit is set, then . matches newline.
+   If not set, then it doesn't.  */
+#define RE_DOT_NEWLINE (RE_CONTEXT_INVALID_OPS << 1)
+
+/* If this bit is set, then . doesn't match NUL.
+   If not set, then it does.  */
+#define RE_DOT_NOT_NULL (RE_DOT_NEWLINE << 1)
+
+/* If this bit is set, nonmatching lists [^...] do not match newline.
+   If not set, they do.  */
+#define RE_HAT_LISTS_NOT_NEWLINE (RE_DOT_NOT_NULL << 1)
+
+/* If this bit is set, either \{...\} or {...} defines an
+     interval, depending on RE_NO_BK_BRACES.
+   If not set, \{, \}, {, and } are literals.  */
+#define RE_INTERVALS (RE_HAT_LISTS_NOT_NEWLINE << 1)
+
+/* If this bit is set, +, ? and | aren't recognized as operators.
+   If not set, they are.  */
+#define RE_LIMITED_OPS (RE_INTERVALS << 1)
+
+/* If this bit is set, newline is an alternation operator.
+   If not set, newline is literal.  */
+#define RE_NEWLINE_ALT (RE_LIMITED_OPS << 1)
+
+/* If this bit is set, then `{...}' defines an interval, and \{ and \}
+     are literals.
+  If not set, then `\{...\}' defines an interval.  */
+#define RE_NO_BK_BRACES (RE_NEWLINE_ALT << 1)
+
+/* If this bit is set, (...) defines a group, and \( and \) are literals.
+   If not set, \(...\) defines a group, and ( and ) are literals.  */
+#define RE_NO_BK_PARENS (RE_NO_BK_BRACES << 1)
+
+/* If this bit is set, then \<digit> matches <digit>.
+   If not set, then \<digit> is a back-reference.  */
+#define RE_NO_BK_REFS (RE_NO_BK_PARENS << 1)
+
+/* If this bit is set, then | is an alternation operator, and \| is literal.
+   If not set, then \| is an alternation operator, and | is literal.  */
+#define RE_NO_BK_VBAR (RE_NO_BK_REFS << 1)
+
+/* If this bit is set, then an ending range point collating higher
+     than the starting range point, as in [z-a], is invalid.
+   If not set, then when ending range point collates higher than the
+     starting range point, the range is ignored.  */
+#define RE_NO_EMPTY_RANGES (RE_NO_BK_VBAR << 1)
+
+/* If this bit is set, then an unmatched ) is ordinary.
+   If not set, then an unmatched ) is invalid.  */
+#define RE_UNMATCHED_RIGHT_PAREN_ORD (RE_NO_EMPTY_RANGES << 1)
+
+/* This global variable defines the particular regexp syntax to use (for
+   some interfaces).  When a regexp is compiled, the syntax used is
+   stored in the pattern buffer, so changing this does not affect
+   already-compiled regexps.  */
+extern reg_syntax_t re_syntax_options;
+
+/* Define combinations of the above bits for the standard possibilities.
+   (The [[[ comments delimit what gets put into the Texinfo file, so
+   don't delete them!)  */
+/* [[[begin syntaxes]]] */
+#define RE_SYNTAX_EMACS 0
+
+#define RE_SYNTAX_AWK							\
+  (RE_BACKSLASH_ESCAPE_IN_LISTS | RE_DOT_NOT_NULL			\
+   | RE_NO_BK_PARENS            | RE_NO_BK_REFS				\
+   | RE_NO_BK_VBAR               | RE_NO_EMPTY_RANGES			\
+   | RE_UNMATCHED_RIGHT_PAREN_ORD)
+
+#define RE_SYNTAX_POSIX_AWK 						\
+  (RE_SYNTAX_POSIX_EXTENDED | RE_BACKSLASH_ESCAPE_IN_LISTS)
+
+#define RE_SYNTAX_GREP							\
+  (RE_BK_PLUS_QM              | RE_CHAR_CLASSES				\
+   | RE_HAT_LISTS_NOT_NEWLINE | RE_INTERVALS				\
+   | RE_NEWLINE_ALT)
+
+#define RE_SYNTAX_EGREP							\
+  (RE_CHAR_CLASSES        | RE_CONTEXT_INDEP_ANCHORS			\
+   | RE_CONTEXT_INDEP_OPS | RE_HAT_LISTS_NOT_NEWLINE			\
+   | RE_NEWLINE_ALT       | RE_NO_BK_PARENS				\
+   | RE_NO_BK_VBAR)
+
+#define RE_SYNTAX_POSIX_EGREP						\
+  (RE_SYNTAX_EGREP | RE_INTERVALS | RE_NO_BK_BRACES)
+
+/* P1003.2/D11.2, section 4.20.7.1, lines 5078ff.  */
+#define RE_SYNTAX_ED RE_SYNTAX_POSIX_BASIC
+
+#define RE_SYNTAX_SED RE_SYNTAX_POSIX_BASIC
+
+/* Syntax bits common to both basic and extended POSIX regex syntax.  */
+#define _RE_SYNTAX_POSIX_COMMON						\
+  (RE_CHAR_CLASSES | RE_DOT_NEWLINE      | RE_DOT_NOT_NULL		\
+   | RE_INTERVALS  | RE_NO_EMPTY_RANGES)
+
+#define RE_SYNTAX_POSIX_BASIC						\
+  (_RE_SYNTAX_POSIX_COMMON | RE_BK_PLUS_QM)
+
+/* Differs from ..._POSIX_BASIC only in that RE_BK_PLUS_QM becomes
+   RE_LIMITED_OPS, i.e., \? \+ \| are not recognized.  Actually, this
+   isn't minimal, since other operators, such as \`, aren't disabled.  */
+#define RE_SYNTAX_POSIX_MINIMAL_BASIC					\
+  (_RE_SYNTAX_POSIX_COMMON | RE_LIMITED_OPS)
+
+#define RE_SYNTAX_POSIX_EXTENDED					\
+  (_RE_SYNTAX_POSIX_COMMON | RE_CONTEXT_INDEP_ANCHORS			\
+   | RE_CONTEXT_INDEP_OPS  | RE_NO_BK_BRACES				\
+   | RE_NO_BK_PARENS       | RE_NO_BK_VBAR				\
+   | RE_UNMATCHED_RIGHT_PAREN_ORD)
+
+/* Differs from ..._POSIX_EXTENDED in that RE_CONTEXT_INVALID_OPS
+   replaces RE_CONTEXT_INDEP_OPS and RE_NO_BK_REFS is added.  */
+#define RE_SYNTAX_POSIX_MINIMAL_EXTENDED				\
+  (_RE_SYNTAX_POSIX_COMMON  | RE_CONTEXT_INDEP_ANCHORS			\
+   | RE_CONTEXT_INVALID_OPS | RE_NO_BK_BRACES				\
+   | RE_NO_BK_PARENS        | RE_NO_BK_REFS				\
+   | RE_NO_BK_VBAR	    | RE_UNMATCHED_RIGHT_PAREN_ORD)
+/* [[[end syntaxes]]] */
+
+/* Maximum number of duplicates an interval can allow.  Some systems
+   (erroneously) define this in other header files, but we want our
+   value, so remove any previous define.  */
+#ifdef RE_DUP_MAX
+#undef RE_DUP_MAX
+#endif
+#define RE_DUP_MAX ((1 << 15) - 1)
+
+
+/* POSIX `cflags' bits (i.e., information for `regcomp').  */
+
+/* If this bit is set, then use extended regular expression syntax.
+   If not set, then use basic regular expression syntax.  */
+#define REG_EXTENDED 1
+
+/* If this bit is set, then ignore case when matching.
+   If not set, then case is significant.  */
+#define REG_ICASE (REG_EXTENDED << 1)
+
+/* If this bit is set, then anchors do not match at newline
+     characters in the string.
+   If not set, then anchors do match at newlines.  */
+#define REG_NEWLINE (REG_ICASE << 1)
+
+/* If this bit is set, then report only success or fail in regexec.
+   If not set, then returns differ between not matching and errors.  */
+#define REG_NOSUB (REG_NEWLINE << 1)
+
+
+/* POSIX `eflags' bits (i.e., information for regexec).  */
+
+/* If this bit is set, then the beginning-of-line operator doesn't match
+     the beginning of the string (presumably because it's not the
+     beginning of a line).
+   If not set, then the beginning-of-line operator does match the
+     beginning of the string.  */
+#define REG_NOTBOL 1
+
+/* Like REG_NOTBOL, except for the end-of-line.  */
+#define REG_NOTEOL (1 << 1)
+
+
+/* If any error codes are removed, changed, or added, update the
+   `re_error_msg' table in regex.c.  */
+typedef enum
+{
+  REG_NOERROR = 0,	/* Success.  */
+  REG_NOMATCH,		/* Didn't find a match (for regexec).  */
+
+  /* POSIX regcomp return error codes.  (In the order listed in the
+     standard.)  */
+  REG_BADPAT,		/* Invalid pattern.  */
+  REG_ECOLLATE,		/* Not implemented.  */
+  REG_ECTYPE,		/* Invalid character class name.  */
+  REG_EESCAPE,		/* Trailing backslash.  */
+  REG_ESUBREG,		/* Invalid back reference.  */
+  REG_EBRACK,		/* Unmatched left bracket.  */
+  REG_EPAREN,		/* Parenthesis imbalance.  */
+  REG_EBRACE,		/* Unmatched \{.  */
+  REG_BADBR,		/* Invalid contents of \{\}.  */
+  REG_ERANGE,		/* Invalid range end.  */
+  REG_ESPACE,		/* Ran out of memory.  */
+  REG_BADRPT,		/* No preceding re for repetition op.  */
+
+  /* Error codes we've added.  */
+  REG_EEND,		/* Premature end.  */
+  REG_ESIZE,		/* Compiled pattern bigger than 2^16 bytes.  */
+  REG_ERPAREN		/* Unmatched ) or \); not returned from regcomp.  */
+} reg_errcode_t;
+
+/* This data structure represents a compiled pattern.  Before calling
+   the pattern compiler, the fields `buffer', `allocated', `fastmap',
+   `translate', and `no_sub' can be set.  After the pattern has been
+   compiled, the `re_nsub' field is available.  All other fields are
+   private to the regex routines.  */
+
+struct re_pattern_buffer
+{
+/* [[[begin pattern_buffer]]] */
+	/* Space that holds the compiled pattern.  It is declared as
+	  `unsigned char *' because its elements are
+	   sometimes used as array indexes.  */
+  unsigned char *buffer;
+
+	/* Number of bytes to which `buffer' points.  */
+  unsigned long allocated;
+
+	/* Number of bytes actually used in `buffer'.  */
+  unsigned long used;
+
+	/* Syntax setting with which the pattern was compiled.  */
+  reg_syntax_t syntax;
+
+	/* Pointer to a fastmap, if any, otherwise zero.  re_search uses
+	   the fastmap, if there is one, to skip over impossible
+	   starting points for matches.  */
+  char *fastmap;
+
+	/* Either a translate table to apply to all characters before
+	   comparing them, or zero for no translation.  The translation
+	   is applied to a pattern when it is compiled and to a string
+	   when it is matched.  */
+  char *translate;
+
+	/* Number of subexpressions found by the compiler.  */
+  size_t re_nsub;
+
+	/* Zero if this pattern cannot match the empty string, one else.
+	   Well, in truth it's used only in `re_search_2', to see
+	   whether or not we should use the fastmap, so we don't set
+	   this absolutely perfectly; see `re_compile_fastmap' (the
+	   `duplicate' case).  */
+  unsigned can_be_null : 1;
+
+	/* If REGS_UNALLOCATED, allocate space in the `regs' structure
+	     for `max (RE_NREGS, re_nsub + 1)' groups.
+	   If REGS_REALLOCATE, reallocate space if necessary.
+	   If REGS_FIXED, use what's there.  */
+#define REGS_UNALLOCATED 0
+#define REGS_REALLOCATE 1
+#define REGS_FIXED 2
+  unsigned regs_allocated : 2;
+
+	/* Set to zero when `regex_compile' compiles a pattern; set to one
+	   by `re_compile_fastmap' if it updates the fastmap.  */
+  unsigned fastmap_accurate : 1;
+
+	/* If set, `re_match_2' does not return information about
+	   subexpressions.  */
+  unsigned no_sub : 1;
+
+	/* If set, a beginning-of-line anchor doesn't match at the
+	   beginning of the string.  */
+  unsigned not_bol : 1;
+
+	/* Similarly for an end-of-line anchor.  */
+  unsigned not_eol : 1;
+
+	/* If true, an anchor at a newline matches.  */
+  unsigned newline_anchor : 1;
+
+/* [[[end pattern_buffer]]] */
+};
+
+typedef struct re_pattern_buffer regex_t;
+
+
+/* search.c (search_buffer) in Emacs needs this one opcode value.  It is
+   defined both in `regex.c' and here.  */
+#define RE_EXACTN_VALUE 1
+
+/* Type for byte offsets within the string.  POSIX mandates this.  */
+typedef int regoff_t;
+
+
+/* This is the structure we store register match data in.  See
+   regex.texinfo for a full description of what registers match.  */
+struct re_registers
+{
+  unsigned num_regs;
+  regoff_t *start;
+  regoff_t *end;
+};
+
+
+/* If `regs_allocated' is REGS_UNALLOCATED in the pattern buffer,
+   `re_match_2' returns information about at least this many registers
+   the first time a `regs' structure is passed.  */
+#ifndef RE_NREGS
+#define RE_NREGS 30
+#endif
+
+
+/* POSIX specification for registers.  Aside from the different names than
+   `re_registers', POSIX uses an array of structures, instead of a
+   structure of arrays.  */
+typedef struct
+{
+  regoff_t rm_so;  /* Byte offset from string's start to substring's start.  */
+  regoff_t rm_eo;  /* Byte offset from string's start to substring's end.  */
+} regmatch_t;
+
+/* Declarations for routines.  */
+
+/* To avoid duplicating every routine declaration -- once with a
+   prototype (if we are ANSI), and once without (if we aren't) -- we
+   use the following macro to declare argument types.  This
+   unfortunately clutters up the declarations a bit, but I think it's
+   worth it.  */
+
+#if __STDC__
+
+#define _RE_ARGS(args) args
+
+#else /* not __STDC__ */
+
+#define _RE_ARGS(args) ()
+
+#endif /* not __STDC__ */
+
+/* Sets the current default syntax to SYNTAX, and return the old syntax.
+   You can also simply assign to the `re_syntax_options' variable.  */
+extern reg_syntax_t re_set_syntax _RE_ARGS ((reg_syntax_t syntax));
+
+/* Compile the regular expression PATTERN, with length LENGTH
+   and syntax given by the global `re_syntax_options', into the buffer
+   BUFFER.  Return NULL if successful, and an error string if not.  */
+extern const char *re_compile_pattern
+  _RE_ARGS ((const char *pattern, int length,
+	     struct re_pattern_buffer *buffer));
+
+
+/* Compile a fastmap for the compiled pattern in BUFFER; used to
+   accelerate searches.  Return 0 if successful and -2 if was an
+   internal error.  */
+extern int re_compile_fastmap _RE_ARGS ((struct re_pattern_buffer *buffer));
+
+
+/* Search in the string STRING (with length LENGTH) for the pattern
+   compiled into BUFFER.  Start searching at position START, for RANGE
+   characters.  Return the starting position of the match, -1 for no
+   match, or -2 for an internal error.  Also return register
+   information in REGS (if REGS and BUFFER->no_sub are nonzero).  */
+extern int re_search
+  _RE_ARGS ((struct re_pattern_buffer *buffer, const char *string,
+	    int length, int start, int range, struct re_registers *regs));
+
+
+/* Like `re_search', but search in the concatenation of STRING1 and
+   STRING2.  Also, stop searching at index START + STOP.  */
+extern int re_search_2
+  _RE_ARGS ((struct re_pattern_buffer *buffer, const char *string1,
+	     int length1, const char *string2, int length2,
+	     int start, int range, struct re_registers *regs, int stop));
+
+
+/* Like `re_search', but return how many characters in STRING the regexp
+   in BUFFER matched, starting at position START.  */
+extern int re_match
+  _RE_ARGS ((struct re_pattern_buffer *buffer, const char *string,
+	     int length, int start, struct re_registers *regs));
+
+
+/* Relates to `re_match' as `re_search_2' relates to `re_search'.  */
+extern int re_match_2
+  _RE_ARGS ((struct re_pattern_buffer *buffer, const char *string1,
+	     int length1, const char *string2, int length2,
+	     int start, struct re_registers *regs, int stop));
+
+
+/* Set REGS to hold NUM_REGS registers, storing them in STARTS and
+   ENDS.  Subsequent matches using BUFFER and REGS will use this memory
+   for recording register information.  STARTS and ENDS must be
+   allocated with malloc, and must each be at least `NUM_REGS * sizeof
+   (regoff_t)' bytes long.
+
+   If NUM_REGS == 0, then subsequent matches should allocate their own
+   register data.
+
+   Unless this function is called, the first search or match using
+   PATTERN_BUFFER will allocate its own register data, without
+   freeing the old data.  */
+extern void re_set_registers
+  _RE_ARGS ((struct re_pattern_buffer *buffer, struct re_registers *regs,
+	     unsigned num_regs, regoff_t *starts, regoff_t *ends));
+
+/* 4.2 bsd compatibility.  */
+extern char *re_comp _RE_ARGS ((const char *));
+extern int re_exec _RE_ARGS ((const char *));
+
+/* POSIX compatibility.  */
+extern int regcomp _RE_ARGS ((regex_t *preg, const char *pattern, int cflags));
+extern int regexec
+  _RE_ARGS ((const regex_t *preg, const char *string, size_t nmatch,
+	     regmatch_t pmatch[], int eflags));
+extern size_t regerror
+  _RE_ARGS ((int errcode, const regex_t *preg, char *errbuf,
+	     size_t errbuf_size));
+extern void regfree _RE_ARGS ((regex_t *preg));
+
+#endif /* not __REGEXP_LIBRARY_H__ */
+
+/*
+Local variables:
+make-backup-files: t
+version-control: t
+trim-versions-without-asking: nil
+End:
+*/
diff --git a/compat/setenv.c b/compat/setenv.c
new file mode 100644
index 0000000..3a22ea7
--- /dev/null
+++ b/compat/setenv.c
@@ -0,0 +1,34 @@
+#include "../git-compat-util.h"
+
+int gitsetenv(const char *name, const char *value, int replace)
+{
+	int out;
+	size_t namelen, valuelen;
+	char *envstr;
+
+	if (!name || !value) return -1;
+	if (!replace) {
+		char *oldval = NULL;
+		oldval = getenv(name);
+		if (oldval) return 0;
+	}
+
+	namelen = strlen(name);
+	valuelen = strlen(value);
+	envstr = malloc((namelen + valuelen + 2));
+	if (!envstr) return -1;
+
+	memcpy(envstr, name, namelen);
+	envstr[namelen] = '=';
+	memcpy(envstr + namelen + 1, value, valuelen);
+	envstr[namelen + valuelen + 1] = 0;
+
+	out = putenv(envstr);
+	/* putenv(3) makes the argument string part of the environment,
+	 * and changing that string modifies the environment --- which
+	 * means we do not own that storage anymore.  Do not free
+	 * envstr.
+	 */
+
+	return out;
+}
diff --git a/compat/snprintf.c b/compat/snprintf.c
new file mode 100644
index 0000000..e1e0e75
--- /dev/null
+++ b/compat/snprintf.c
@@ -0,0 +1,64 @@
+#include "../git-compat-util.h"
+
+/*
+ * The size parameter specifies the available space, i.e. includes
+ * the trailing NUL byte; but Windows's vsnprintf uses the entire
+ * buffer and avoids the trailing NUL, should the buffer be exactly
+ * big enough for the result. Defining SNPRINTF_SIZE_CORR to 1 will
+ * therefore remove 1 byte from the reported buffer size, so we
+ * always have room for a trailing NUL byte.
+ */
+#ifndef SNPRINTF_SIZE_CORR
+#if defined(WIN32) && (!defined(__GNUC__) || __GNUC__ < 4)
+#define SNPRINTF_SIZE_CORR 1
+#else
+#define SNPRINTF_SIZE_CORR 0
+#endif
+#endif
+
+#undef vsnprintf
+int git_vsnprintf(char *str, size_t maxsize, const char *format, va_list ap)
+{
+	char *s;
+	int ret = -1;
+
+	if (maxsize > 0) {
+		ret = vsnprintf(str, maxsize-SNPRINTF_SIZE_CORR, format, ap);
+		if (ret == maxsize-1)
+			ret = -1;
+		/* Windows does not NUL-terminate if result fills buffer */
+		str[maxsize-1] = 0;
+	}
+	if (ret != -1)
+		return ret;
+
+	s = NULL;
+	if (maxsize < 128)
+		maxsize = 128;
+
+	while (ret == -1) {
+		maxsize *= 4;
+		str = realloc(s, maxsize);
+		if (! str)
+			break;
+		s = str;
+		ret = vsnprintf(str, maxsize-SNPRINTF_SIZE_CORR, format, ap);
+		if (ret == maxsize-1)
+			ret = -1;
+	}
+	free(s);
+	return ret;
+}
+
+int git_snprintf(char *str, size_t maxsize, const char *format, ...)
+{
+	va_list ap;
+	int ret;
+
+	va_start(ap, format);
+	ret = git_vsnprintf(str, maxsize, format, ap);
+	va_end(ap);
+
+	return ret;
+}
+
diff --git a/compat/strcasestr.c b/compat/strcasestr.c
new file mode 100644
index 0000000..26896de
--- /dev/null
+++ b/compat/strcasestr.c
@@ -0,0 +1,22 @@
+#include "../git-compat-util.h"
+
+char *gitstrcasestr(const char *haystack, const char *needle)
+{
+	int nlen = strlen(needle);
+	int hlen = strlen(haystack) - nlen + 1;
+	int i;
+
+	for (i = 0; i < hlen; i++) {
+		int j;
+		for (j = 0; j < nlen; j++) {
+			unsigned char c1 = haystack[i+j];
+			unsigned char c2 = needle[j];
+			if (toupper(c1) != toupper(c2))
+				goto next;
+		}
+		return (char *) haystack + i;
+	next:
+		;
+	}
+	return NULL;
+}
diff --git a/compat/strlcpy.c b/compat/strlcpy.c
new file mode 100644
index 0000000..4024c36
--- /dev/null
+++ b/compat/strlcpy.c
@@ -0,0 +1,13 @@
+#include "../git-compat-util.h"
+
+size_t gitstrlcpy(char *dest, const char *src, size_t size)
+{
+	size_t ret = strlen(src);
+
+	if (size) {
+		size_t len = (ret >= size) ? size - 1 : ret;
+		memcpy(dest, src, len);
+		dest[len] = '\0';
+	}
+	return ret;
+}
diff --git a/compat/strtoumax.c b/compat/strtoumax.c
new file mode 100644
index 0000000..5541353
--- /dev/null
+++ b/compat/strtoumax.c
@@ -0,0 +1,10 @@
+#include "../git-compat-util.h"
+
+uintmax_t gitstrtoumax (const char *nptr, char **endptr, int base)
+{
+#if defined(NO_STRTOULL)
+	return strtoul(nptr, endptr, base);
+#else
+	return strtoull(nptr, endptr, base);
+#endif
+}
diff --git a/compat/unsetenv.c b/compat/unsetenv.c
new file mode 100644
index 0000000..eb29f5e
--- /dev/null
+++ b/compat/unsetenv.c
@@ -0,0 +1,25 @@
+#include "../git-compat-util.h"
+
+void gitunsetenv (const char *name)
+{
+     extern char **environ;
+     int src, dst;
+     size_t nmln;
+
+     nmln = strlen(name);
+
+     for (src = dst = 0; environ[src]; ++src) {
+	  size_t enln;
+	  enln = strlen(environ[src]);
+	  if (enln > nmln) {
+               /* might match, and can test for '=' safely */
+	       if (0 == strncmp (environ[src], name, nmln)
+		   && '=' == environ[src][nmln])
+		    /* matches, so skip */
+		    continue;
+	  }
+	  environ[dst] = environ[src];
+	  ++dst;
+     }
+     environ[dst] = NULL;
+}
diff --git a/compat/vcbuild/README b/compat/vcbuild/README
new file mode 100644
index 0000000..df8a657
--- /dev/null
+++ b/compat/vcbuild/README
@@ -0,0 +1,50 @@
+The Steps of Build Git with VS2008
+
+1. You need the build environment, which contains the Git dependencies
+   to be able to compile, link and run Git with MSVC.
+
+   You can either use the binary repository:
+
+       WWW: http://repo.or.cz/w/msvcgit.git
+       Git: git clone git://repo.or.cz/msvcgit.git
+       Zip: http://repo.or.cz/w/msvcgit.git?a=snapshot;h=master;sf=zip
+
+   and call the setup_32bit_env.cmd batch script before compiling Git,
+   (see repo/package README for details), or the source repository:
+
+       WWW: http://repo.or.cz/w/gitbuild.git
+       Git: git clone git://repo.or.cz/gitbuild.git
+       Zip: (None, as it's a project with submodules)
+
+   and build the support libs as instructed in that repo/package.
+
+2. Ensure you have the msysgit environment in your path, so you have
+   GNU Make, bash and perl available.
+
+       WWW: http://repo.or.cz/w/msysgit.git
+       Git: git clone git://repo.or.cz/msysgit.git
+       Zip: http://repo.or.cz/w/msysgit.git?a=snapshot;h=master;sf=zip
+
+   This environment is also needed when you use the resulting
+   executables, since Git might need to run scripts which are part of
+   the git operations.
+
+3. Inside Git's directory run the command:
+       make common-cmds.h
+   to generate the common-cmds.h file needed to compile git.
+
+4. Then either build Git with the GNU Make Makefile in the Git projects
+   root
+       make MSVC=1
+   or generate Visual Studio solution/projects (.sln/.vcproj) with the
+   command
+       perl contrib/buildsystems/generate -g Vcproj
+   and open and build the solution with the IDE
+       devenv git.sln /useenv
+   or build with the IDE build engine directly from the command line
+       devenv git.sln /useenv /build "Release|Win32"
+   The /useenv option is required, so Visual Studio picks up the
+   environment variables for the support libraries required to build
+   Git, which you set up in step 1.
+
+Done!
diff --git a/compat/vcbuild/include/alloca.h b/compat/vcbuild/include/alloca.h
new file mode 100644
index 0000000..c0d7985
--- /dev/null
+++ b/compat/vcbuild/include/alloca.h
@@ -0,0 +1 @@
+#include <malloc.h>
diff --git a/compat/vcbuild/include/arpa/inet.h b/compat/vcbuild/include/arpa/inet.h
new file mode 100644
index 0000000..0d8552a
--- /dev/null
+++ b/compat/vcbuild/include/arpa/inet.h
@@ -0,0 +1 @@
+/* Intentionally empty file to support building git with MSVC */
diff --git a/compat/vcbuild/include/dirent.h b/compat/vcbuild/include/dirent.h
new file mode 100644
index 0000000..440618d
--- /dev/null
+++ b/compat/vcbuild/include/dirent.h
@@ -0,0 +1,128 @@
+/*
+ * DIRENT.H (formerly DIRLIB.H)
+ * This file has no copyright assigned and is placed in the Public Domain.
+ * This file is a part of the mingw-runtime package.
+ *
+ * The mingw-runtime package and its code is distributed in the hope that it
+ * will be useful but WITHOUT ANY WARRANTY.  ALL WARRANTIES, EXPRESSED OR
+ * IMPLIED ARE HEREBY DISCLAIMED.  This includes but is not limited to
+ * warranties of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You are free to use this package and its code without limitation.
+ */
+#ifndef _DIRENT_H_
+#define _DIRENT_H_
+#include <io.h>
+
+#define PATH_MAX 512
+
+#define __MINGW_NOTHROW
+
+#ifndef RC_INVOKED
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+struct dirent
+{
+	long		d_ino;		/* Always zero. */
+	unsigned short	d_reclen;	/* Always zero. */
+	unsigned short	d_namlen;	/* Length of name in d_name. */
+	char		d_name[FILENAME_MAX]; /* File name. */
+};
+
+/*
+ * This is an internal data structure. Good programmers will not use it
+ * except as an argument to one of the functions below.
+ * dd_stat field is now int (was short in older versions).
+ */
+typedef struct
+{
+	/* disk transfer area for this dir */
+	struct _finddata_t	dd_dta;
+
+	/* dirent struct to return from dir (NOTE: this makes this thread
+	 * safe as long as only one thread uses a particular DIR struct at
+	 * a time) */
+	struct dirent		dd_dir;
+
+	/* _findnext handle */
+	long			dd_handle;
+
+	/*
+	 * Status of search:
+	 *   0 = not started yet (next entry to read is first entry)
+	 *  -1 = off the end
+	 *   positive = 0 based index of next entry
+	 */
+	int			dd_stat;
+
+	/* given path for dir with search pattern (struct is extended) */
+	char			dd_name[PATH_MAX+3];
+} DIR;
+
+DIR* __cdecl __MINGW_NOTHROW opendir (const char*);
+struct dirent* __cdecl __MINGW_NOTHROW readdir (DIR*);
+int __cdecl __MINGW_NOTHROW closedir (DIR*);
+void __cdecl __MINGW_NOTHROW rewinddir (DIR*);
+long __cdecl __MINGW_NOTHROW telldir (DIR*);
+void __cdecl __MINGW_NOTHROW seekdir (DIR*, long);
+
+
+/* wide char versions */
+
+struct _wdirent
+{
+	long		d_ino;		/* Always zero. */
+	unsigned short	d_reclen;	/* Always zero. */
+	unsigned short	d_namlen;	/* Length of name in d_name. */
+	wchar_t		d_name[FILENAME_MAX]; /* File name. */
+};
+
+/*
+ * This is an internal data structure. Good programmers will not use it
+ * except as an argument to one of the functions below.
+ */
+typedef struct
+{
+	/* disk transfer area for this dir */
+	//struct _wfinddata_t	dd_dta;
+
+	/* dirent struct to return from dir (NOTE: this makes this thread
+	 * safe as long as only one thread uses a particular DIR struct at
+	 * a time) */
+	struct _wdirent		dd_dir;
+
+	/* _findnext handle */
+	long			dd_handle;
+
+	/*
+	 * Status of search:
+	 *   0 = not started yet (next entry to read is first entry)
+	 *  -1 = off the end
+	 *   positive = 0 based index of next entry
+	 */
+	int			dd_stat;
+
+	/* given path for dir with search pattern (struct is extended) */
+	wchar_t			dd_name[1];
+} _WDIR;
+
+
+
+_WDIR* __cdecl __MINGW_NOTHROW _wopendir (const wchar_t*);
+struct _wdirent*  __cdecl __MINGW_NOTHROW _wreaddir (_WDIR*);
+int __cdecl __MINGW_NOTHROW _wclosedir (_WDIR*);
+void __cdecl __MINGW_NOTHROW _wrewinddir (_WDIR*);
+long __cdecl __MINGW_NOTHROW _wtelldir (_WDIR*);
+void __cdecl __MINGW_NOTHROW _wseekdir (_WDIR*, long);
+
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* Not RC_INVOKED */
+
+#endif	/* Not _DIRENT_H_ */
diff --git a/compat/vcbuild/include/grp.h b/compat/vcbuild/include/grp.h
new file mode 100644
index 0000000..0d8552a
--- /dev/null
+++ b/compat/vcbuild/include/grp.h
@@ -0,0 +1 @@
+/* Intentionally empty file to support building git with MSVC */
diff --git a/compat/vcbuild/include/inttypes.h b/compat/vcbuild/include/inttypes.h
new file mode 100644
index 0000000..0d8552a
--- /dev/null
+++ b/compat/vcbuild/include/inttypes.h
@@ -0,0 +1 @@
+/* Intentionally empty file to support building git with MSVC */
diff --git a/compat/vcbuild/include/netdb.h b/compat/vcbuild/include/netdb.h
new file mode 100644
index 0000000..0d8552a
--- /dev/null
+++ b/compat/vcbuild/include/netdb.h
@@ -0,0 +1 @@
+/* Intentionally empty file to support building git with MSVC */
diff --git a/compat/vcbuild/include/netinet/in.h b/compat/vcbuild/include/netinet/in.h
new file mode 100644
index 0000000..0d8552a
--- /dev/null
+++ b/compat/vcbuild/include/netinet/in.h
@@ -0,0 +1 @@
+/* Intentionally empty file to support building git with MSVC */
diff --git a/compat/vcbuild/include/netinet/tcp.h b/compat/vcbuild/include/netinet/tcp.h
new file mode 100644
index 0000000..0d8552a
--- /dev/null
+++ b/compat/vcbuild/include/netinet/tcp.h
@@ -0,0 +1 @@
+/* Intentionally empty file to support building git with MSVC */
diff --git a/compat/vcbuild/include/pwd.h b/compat/vcbuild/include/pwd.h
new file mode 100644
index 0000000..0d8552a
--- /dev/null
+++ b/compat/vcbuild/include/pwd.h
@@ -0,0 +1 @@
+/* Intentionally empty file to support building git with MSVC */
diff --git a/compat/vcbuild/include/sys/ioctl.h b/compat/vcbuild/include/sys/ioctl.h
new file mode 100644
index 0000000..0d8552a
--- /dev/null
+++ b/compat/vcbuild/include/sys/ioctl.h
@@ -0,0 +1 @@
+/* Intentionally empty file to support building git with MSVC */
diff --git a/compat/vcbuild/include/sys/param.h b/compat/vcbuild/include/sys/param.h
new file mode 100644
index 0000000..0d8552a
--- /dev/null
+++ b/compat/vcbuild/include/sys/param.h
@@ -0,0 +1 @@
+/* Intentionally empty file to support building git with MSVC */
diff --git a/compat/vcbuild/include/sys/poll.h b/compat/vcbuild/include/sys/poll.h
new file mode 100644
index 0000000..0d8552a
--- /dev/null
+++ b/compat/vcbuild/include/sys/poll.h
@@ -0,0 +1 @@
+/* Intentionally empty file to support building git with MSVC */
diff --git a/compat/vcbuild/include/sys/select.h b/compat/vcbuild/include/sys/select.h
new file mode 100644
index 0000000..0d8552a
--- /dev/null
+++ b/compat/vcbuild/include/sys/select.h
@@ -0,0 +1 @@
+/* Intentionally empty file to support building git with MSVC */
diff --git a/compat/vcbuild/include/sys/socket.h b/compat/vcbuild/include/sys/socket.h
new file mode 100644
index 0000000..0d8552a
--- /dev/null
+++ b/compat/vcbuild/include/sys/socket.h
@@ -0,0 +1 @@
+/* Intentionally empty file to support building git with MSVC */
diff --git a/compat/vcbuild/include/sys/time.h b/compat/vcbuild/include/sys/time.h
new file mode 100644
index 0000000..0d8552a
--- /dev/null
+++ b/compat/vcbuild/include/sys/time.h
@@ -0,0 +1 @@
+/* Intentionally empty file to support building git with MSVC */
diff --git a/compat/vcbuild/include/sys/utime.h b/compat/vcbuild/include/sys/utime.h
new file mode 100644
index 0000000..582589c
--- /dev/null
+++ b/compat/vcbuild/include/sys/utime.h
@@ -0,0 +1,34 @@
+#ifndef	_UTIME_H_
+#define	_UTIME_H_
+/*
+ * UTIME.H
+ * This file has no copyright assigned and is placed in the Public Domain.
+ * This file is a part of the mingw-runtime package.
+ *
+ * The mingw-runtime package and its code is distributed in the hope that it
+ * will be useful but WITHOUT ANY WARRANTY.  ALL WARRANTIES, EXPRESSED OR
+ * IMPLIED ARE HEREBY DISCLAIMED.  This includes but is not limited to
+ * warranties of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You are free to use this package and its code without limitation.
+ */
+
+/*
+ * Structure used by _utime function.
+ */
+struct _utimbuf
+{
+	time_t	actime;		/* Access time */
+	time_t	modtime;	/* Modification time */
+};
+
+#ifndef	_NO_OLDNAMES
+/* NOTE: Must be the same as _utimbuf above. */
+struct utimbuf
+{
+	time_t	actime;
+	time_t	modtime;
+};
+#endif	/* Not _NO_OLDNAMES */
+
+#endif
diff --git a/compat/vcbuild/include/sys/wait.h b/compat/vcbuild/include/sys/wait.h
new file mode 100644
index 0000000..0d8552a
--- /dev/null
+++ b/compat/vcbuild/include/sys/wait.h
@@ -0,0 +1 @@
+/* Intentionally empty file to support building git with MSVC */
diff --git a/compat/vcbuild/include/termios.h b/compat/vcbuild/include/termios.h
new file mode 100644
index 0000000..0d8552a
--- /dev/null
+++ b/compat/vcbuild/include/termios.h
@@ -0,0 +1 @@
+/* Intentionally empty file to support building git with MSVC */
diff --git a/compat/vcbuild/include/unistd.h b/compat/vcbuild/include/unistd.h
new file mode 100644
index 0000000..2a4f276
--- /dev/null
+++ b/compat/vcbuild/include/unistd.h
@@ -0,0 +1,92 @@
+#ifndef _UNISTD_
+#define _UNISTD_
+
+/* Win32 define for porting git*/
+
+#ifndef _MODE_T_
+#define	_MODE_T_
+typedef unsigned short _mode_t;
+
+#ifndef	_NO_OLDNAMES
+typedef _mode_t	mode_t;
+#endif
+#endif	/* Not _MODE_T_ */
+
+#ifndef _SSIZE_T_
+#define _SSIZE_T_
+typedef long _ssize_t;
+
+#ifndef	_OFF_T_
+#define	_OFF_T_
+typedef long _off_t;
+
+#ifndef	_NO_OLDNAMES
+typedef _off_t	off_t;
+#endif
+#endif	/* Not _OFF_T_ */
+
+
+#ifndef	_NO_OLDNAMES
+typedef _ssize_t ssize_t;
+#endif
+#endif /* Not _SSIZE_T_ */
+
+typedef signed char int8_t;
+typedef unsigned char   uint8_t;
+typedef short  int16_t;
+typedef unsigned short  uint16_t;
+typedef int  int32_t;
+typedef unsigned   uint32_t;
+typedef long long  int64_t;
+typedef unsigned long long   uint64_t;
+
+typedef long long  intmax_t;
+typedef unsigned long long uintmax_t;
+
+typedef int64_t off64_t;
+
+#define STDOUT_FILENO 1
+#define STDERR_FILENO 2
+
+/* Some defines for _access nAccessMode (MS doesn't define them, but
+ * it doesn't seem to hurt to add them). */
+#define	F_OK	0	/* Check for file existence */
+/* Well maybe it does hurt.  On newer versions of MSVCRT, an access mode
+   of 1 causes invalid parameter error. */
+#define	X_OK	0	/* MS access() doesn't check for execute permission. */
+#define	W_OK	2	/* Check for write permission */
+#define	R_OK	4	/* Check for read permission */
+
+#define	_S_IFIFO	0x1000	/* FIFO */
+#define	_S_IFCHR	0x2000	/* Character */
+#define	_S_IFBLK	0x3000	/* Block: Is this ever set under w32? */
+#define	_S_IFDIR	0x4000	/* Directory */
+#define	_S_IFREG	0x8000	/* Regular */
+
+#define	_S_IFMT		0xF000	/* File type mask */
+
+#define	_S_IXUSR	_S_IEXEC
+#define	_S_IWUSR	_S_IWRITE
+#define	_S_IRUSR	_S_IREAD
+#define	_S_ISDIR(m)	(((m) & _S_IFMT) == _S_IFDIR)
+
+#define	S_IFIFO		_S_IFIFO
+#define	S_IFCHR		_S_IFCHR
+#define	S_IFBLK		_S_IFBLK
+#define	S_IFDIR		_S_IFDIR
+#define	S_IFREG		_S_IFREG
+#define	S_IFMT		_S_IFMT
+#define	S_IEXEC		_S_IEXEC
+#define	S_IWRITE	_S_IWRITE
+#define	S_IREAD		_S_IREAD
+#define	S_IRWXU		_S_IRWXU
+#define	S_IXUSR		_S_IXUSR
+#define	S_IWUSR		_S_IWUSR
+#define	S_IRUSR		_S_IRUSR
+
+
+#define	S_ISDIR(m)	(((m) & S_IFMT) == S_IFDIR)
+#define	S_ISREG(m)	(((m) & S_IFMT) == S_IFREG)
+#define	S_ISFIFO(m)	(((m) & S_IFMT) == S_IFIFO)
+
+#endif
diff --git a/compat/vcbuild/include/utime.h b/compat/vcbuild/include/utime.h
new file mode 100644
index 0000000..8285f38
--- /dev/null
+++ b/compat/vcbuild/include/utime.h
@@ -0,0 +1 @@
+#include <sys/utime.h>
diff --git a/compat/vcbuild/scripts/clink.pl b/compat/vcbuild/scripts/clink.pl
new file mode 100644
index 0000000..4374771
--- /dev/null
+++ b/compat/vcbuild/scripts/clink.pl
@@ -0,0 +1,52 @@
+#!/usr/bin/perl -w
+######################################################################
+# Compiles or links files
+#
+# This is a wrapper to facilitate the compilation of Git with MSVC
+# using GNU Make as the build system. So, instead of manipulating the
+# Makefile into something nasty, just to support non-space arguments
+# etc, we use this wrapper to fix the command line options
+#
+# Copyright (C) 2009 Marius Storm-Olsen <mstormo@gmail.com>
+######################################################################
+use strict;
+my @args = ();
+my @cflags = ();
+my $is_linking = 0;
+while (@ARGV) {
+	my $arg = shift @ARGV;
+	if ("$arg" =~ /^-[DIMGO]/) {
+		push(@cflags, $arg);
+	} elsif ("$arg" eq "-o") {
+		my $file_out = shift @ARGV;
+		if ("$file_out" =~ /exe$/) {
+			$is_linking = 1;
+			push(@args, "-OUT:$file_out");
+		} else {
+			push(@args, "-Fo$file_out");
+		}
+	} elsif ("$arg" eq "-lz") {
+		push(@args, "zlib.lib");
+	} elsif ("$arg" eq "-liconv") {
+		push(@args, "iconv.lib");
+	} elsif ("$arg" eq "-lcrypto") {
+		push(@args, "libeay32.lib");
+	} elsif ("$arg" eq "-lssl") {
+		push(@args, "ssleay32.lib");
+	} elsif ("$arg" =~ /^-L/ && "$arg" ne "-LTCG") {
+		$arg =~ s/^-L/-LIBPATH:/;
+		push(@args, $arg);
+	} elsif ("$arg" =~ /^-R/) {
+		# eat
+	} else {
+		push(@args, $arg);
+	}
+}
+if ($is_linking) {
+	unshift(@args, "link.exe");
+} else {
+	unshift(@args, "cl.exe");
+	push(@args, @cflags);
+}
+#printf("**** @args\n");
+exit (system(@args) != 0);
diff --git a/compat/vcbuild/scripts/lib.pl b/compat/vcbuild/scripts/lib.pl
new file mode 100644
index 0000000..d8054e4
--- /dev/null
+++ b/compat/vcbuild/scripts/lib.pl
@@ -0,0 +1,26 @@
+#!/usr/bin/perl -w
+######################################################################
+# Libifies files on Windows
+#
+# This is a wrapper to facilitate the compilation of Git with MSVC
+# using GNU Make as the build system. So, instead of manipulating the
+# Makefile into something nasty, just to support non-space arguments
+# etc, we use this wrapper to fix the command line options
+#
+# Copyright (C) 2009 Marius Storm-Olsen <mstormo@gmail.com>
+######################################################################
+use strict;
+my @args = ();
+while (@ARGV) {
+	my $arg = shift @ARGV;
+	if ("$arg" eq "rcs") {
+		# Consume the rcs option
+	} elsif ("$arg" =~ /\.a$/) {
+		push(@args, "-OUT:$arg");
+	} else {
+		push(@args, $arg);
+	}
+}
+unshift(@args, "lib.exe");
+# printf("**** @args\n");
+exit (system(@args) != 0);
diff --git a/compat/win32.h b/compat/win32.h
new file mode 100644
index 0000000..8ce9104
--- /dev/null
+++ b/compat/win32.h
@@ -0,0 +1,41 @@
+#ifndef WIN32_H
+#define WIN32_H
+
+/* common Win32 functions for MinGW and Cygwin */
+#ifndef WIN32         /* Not defined by Cygwin */
+#include <windows.h>
+#endif
+
+static inline int file_attr_to_st_mode (DWORD attr)
+{
+	int fMode = S_IREAD;
+	if (attr & FILE_ATTRIBUTE_DIRECTORY)
+		fMode |= S_IFDIR;
+	else
+		fMode |= S_IFREG;
+	if (!(attr & FILE_ATTRIBUTE_READONLY))
+		fMode |= S_IWRITE;
+	return fMode;
+}
+
+static inline int get_file_attr(const char *fname, WIN32_FILE_ATTRIBUTE_DATA *fdata)
+{
+	if (GetFileAttributesExA(fname, GetFileExInfoStandard, fdata))
+		return 0;
+
+	switch (GetLastError()) {
+	case ERROR_ACCESS_DENIED:
+	case ERROR_SHARING_VIOLATION:
+	case ERROR_LOCK_VIOLATION:
+	case ERROR_SHARING_BUFFER_EXCEEDED:
+		return EACCES;
+	case ERROR_BUFFER_OVERFLOW:
+		return ENAMETOOLONG;
+	case ERROR_NOT_ENOUGH_MEMORY:
+		return ENOMEM;
+	default:
+		return ENOENT;
+	}
+}
+
+#endif
diff --git a/compat/win32/pthread.c b/compat/win32/pthread.c
new file mode 100644
index 0000000..010e875
--- /dev/null
+++ b/compat/win32/pthread.c
@@ -0,0 +1,196 @@
+/*
+ * Copyright (C) 2009 Andrzej K. Haczewski <ahaczewski@gmail.com>
+ *
+ * DISCLAIMER: The implementation is Git-specific, it is subset of original
+ * Pthreads API, without lots of other features that Git doesn't use.
+ * Git also makes sure that the passed arguments are valid, so there's
+ * no need for double-checking.
+ */
+
+#include "../../git-compat-util.h"
+#include "pthread.h"
+
+#include <errno.h>
+#include <limits.h>
+
+static unsigned __stdcall win32_start_routine(void *arg)
+{
+	pthread_t *thread = arg;
+	thread->tid = GetCurrentThreadId();
+	thread->arg = thread->start_routine(thread->arg);
+	return 0;
+}
+
+int pthread_create(pthread_t *thread, const void *unused,
+		   void *(*start_routine)(void*), void *arg)
+{
+	thread->arg = arg;
+	thread->start_routine = start_routine;
+	thread->handle = (HANDLE)
+		_beginthreadex(NULL, 0, win32_start_routine, thread, 0, NULL);
+
+	if (!thread->handle)
+		return errno;
+	else
+		return 0;
+}
+
+int win32_pthread_join(pthread_t *thread, void **value_ptr)
+{
+	DWORD result = WaitForSingleObject(thread->handle, INFINITE);
+	switch (result) {
+		case WAIT_OBJECT_0:
+			if (value_ptr)
+				*value_ptr = thread->arg;
+			return 0;
+		case WAIT_ABANDONED:
+			return EINVAL;
+		default:
+			return err_win_to_posix(GetLastError());
+	}
+}
+
+pthread_t pthread_self(void)
+{
+	pthread_t t = { 0 };
+	t.tid = GetCurrentThreadId();
+	return t;
+}
+
+int pthread_cond_init(pthread_cond_t *cond, const void *unused)
+{
+	cond->waiters = 0;
+	cond->was_broadcast = 0;
+	InitializeCriticalSection(&cond->waiters_lock);
+
+	cond->sema = CreateSemaphore(NULL, 0, LONG_MAX, NULL);
+	if (!cond->sema)
+		die("CreateSemaphore() failed");
+
+	cond->continue_broadcast = CreateEvent(NULL,	/* security */
+				FALSE,			/* auto-reset */
+				FALSE,			/* not signaled */
+				NULL);			/* name */
+	if (!cond->continue_broadcast)
+		die("CreateEvent() failed");
+
+	return 0;
+}
+
+int pthread_cond_destroy(pthread_cond_t *cond)
+{
+	CloseHandle(cond->sema);
+	CloseHandle(cond->continue_broadcast);
+	DeleteCriticalSection(&cond->waiters_lock);
+	return 0;
+}
+
+int pthread_cond_wait(pthread_cond_t *cond, CRITICAL_SECTION *mutex)
+{
+	int last_waiter;
+
+	EnterCriticalSection(&cond->waiters_lock);
+	cond->waiters++;
+	LeaveCriticalSection(&cond->waiters_lock);
+
+	/*
+	 * Unlock external mutex and wait for signal.
+	 * NOTE: we've held mutex locked long enough to increment
+	 * waiters count above, so there's no problem with
+	 * leaving mutex unlocked before we wait on semaphore.
+	 */
+	LeaveCriticalSection(mutex);
+
+	/* let's wait - ignore return value */
+	WaitForSingleObject(cond->sema, INFINITE);
+
+	/*
+	 * Decrease waiters count. If we are the last waiter, then we must
+	 * notify the broadcasting thread that it can continue.
+	 * But if we continued due to cond_signal, we do not have to do that
+	 * because the signaling thread knows that only one waiter continued.
+	 */
+	EnterCriticalSection(&cond->waiters_lock);
+	cond->waiters--;
+	last_waiter = cond->was_broadcast && cond->waiters == 0;
+	LeaveCriticalSection(&cond->waiters_lock);
+
+	if (last_waiter) {
+		/*
+		 * cond_broadcast was issued while mutex was held. This means
+		 * that all other waiters have continued, but are contending
+		 * for the mutex at the end of this function because the
+		 * broadcasting thread did not leave cond_broadcast, yet.
+		 * (This is so that it can be sure that each waiter has
+		 * consumed exactly one slice of the semaphor.)
+		 * The last waiter must tell the broadcasting thread that it
+		 * can go on.
+		 */
+		SetEvent(cond->continue_broadcast);
+		/*
+		 * Now we go on to contend with all other waiters for
+		 * the mutex. Auf in den Kampf!
+		 */
+	}
+	/* lock external mutex again */
+	EnterCriticalSection(mutex);
+
+	return 0;
+}
+
+/*
+ * IMPORTANT: This implementation requires that pthread_cond_signal
+ * is called while the mutex is held that is used in the corresponding
+ * pthread_cond_wait calls!
+ */
+int pthread_cond_signal(pthread_cond_t *cond)
+{
+	int have_waiters;
+
+	EnterCriticalSection(&cond->waiters_lock);
+	have_waiters = cond->waiters > 0;
+	LeaveCriticalSection(&cond->waiters_lock);
+
+	/*
+	 * Signal only when there are waiters
+	 */
+	if (have_waiters)
+		return ReleaseSemaphore(cond->sema, 1, NULL) ?
+			0 : err_win_to_posix(GetLastError());
+	else
+		return 0;
+}
+
+/*
+ * DOUBLY IMPORTANT: This implementation requires that pthread_cond_broadcast
+ * is called while the mutex is held that is used in the corresponding
+ * pthread_cond_wait calls!
+ */
+int pthread_cond_broadcast(pthread_cond_t *cond)
+{
+	EnterCriticalSection(&cond->waiters_lock);
+
+	if ((cond->was_broadcast = cond->waiters > 0)) {
+		/* wake up all waiters */
+		ReleaseSemaphore(cond->sema, cond->waiters, NULL);
+		LeaveCriticalSection(&cond->waiters_lock);
+		/*
+		 * At this point all waiters continue. Each one takes its
+		 * slice of the semaphor. Now it's our turn to wait: Since
+		 * the external mutex is held, no thread can leave cond_wait,
+		 * yet. For this reason, we can be sure that no thread gets
+		 * a chance to eat *more* than one slice. OTOH, it means
+		 * that the last waiter must send us a wake-up.
+		 */
+		WaitForSingleObject(cond->continue_broadcast, INFINITE);
+		/*
+		 * Since the external mutex is held, no thread can enter
+		 * cond_wait, and, hence, it is safe to reset this flag
+		 * without cond->waiters_lock held.
+		 */
+		cond->was_broadcast = 0;
+	} else {
+		LeaveCriticalSection(&cond->waiters_lock);
+	}
+	return 0;
+}
diff --git a/compat/win32/pthread.h b/compat/win32/pthread.h
new file mode 100644
index 0000000..2e20548
--- /dev/null
+++ b/compat/win32/pthread.h
@@ -0,0 +1,99 @@
+/*
+ * Header used to adapt pthread-based POSIX code to Windows API threads.
+ *
+ * Copyright (C) 2009 Andrzej K. Haczewski <ahaczewski@gmail.com>
+ */
+
+#ifndef PTHREAD_H
+#define PTHREAD_H
+
+#ifndef WIN32_LEAN_AND_MEAN
+#define WIN32_LEAN_AND_MEAN
+#endif
+
+#include <windows.h>
+
+/*
+ * Defines that adapt Windows API threads to pthreads API
+ */
+#define pthread_mutex_t CRITICAL_SECTION
+
+#define pthread_mutex_init(a,b) (InitializeCriticalSection((a)), 0)
+#define pthread_mutex_destroy(a) DeleteCriticalSection((a))
+#define pthread_mutex_lock EnterCriticalSection
+#define pthread_mutex_unlock LeaveCriticalSection
+
+typedef int pthread_mutexattr_t;
+#define pthread_mutexattr_init(a) (*(a) = 0)
+#define pthread_mutexattr_destroy(a) do {} while (0)
+#define pthread_mutexattr_settype(a, t) 0
+#define PTHREAD_MUTEX_RECURSIVE 0
+
+/*
+ * Implement simple condition variable for Windows threads, based on ACE
+ * implementation.
+ *
+ * See original implementation: http://bit.ly/1vkDjo
+ * ACE homepage: http://www.cse.wustl.edu/~schmidt/ACE.html
+ * See also: http://www.cse.wustl.edu/~schmidt/win32-cv-1.html
+ */
+typedef struct {
+	LONG waiters;
+	int was_broadcast;
+	CRITICAL_SECTION waiters_lock;
+	HANDLE sema;
+	HANDLE continue_broadcast;
+} pthread_cond_t;
+
+extern int pthread_cond_init(pthread_cond_t *cond, const void *unused);
+extern int pthread_cond_destroy(pthread_cond_t *cond);
+extern int pthread_cond_wait(pthread_cond_t *cond, CRITICAL_SECTION *mutex);
+extern int pthread_cond_signal(pthread_cond_t *cond);
+extern int pthread_cond_broadcast(pthread_cond_t *cond);
+
+/*
+ * Simple thread creation implementation using pthread API
+ */
+typedef struct {
+	HANDLE handle;
+	void *(*start_routine)(void*);
+	void *arg;
+	DWORD tid;
+} pthread_t;
+
+extern int pthread_create(pthread_t *thread, const void *unused,
+			  void *(*start_routine)(void*), void *arg);
+
+/*
+ * To avoid the need of copying a struct, we use small macro wrapper to pass
+ * pointer to win32_pthread_join instead.
+ */
+#define pthread_join(a, b) win32_pthread_join(&(a), (b))
+
+extern int win32_pthread_join(pthread_t *thread, void **value_ptr);
+
+#define pthread_equal(t1, t2) ((t1).tid == (t2).tid)
+extern pthread_t pthread_self(void);
+
+static inline int pthread_exit(void *ret)
+{
+	ExitThread((DWORD)ret);
+}
+
+typedef DWORD pthread_key_t;
+static inline int pthread_key_create(pthread_key_t *keyp, void (*destructor)(void *value))
+{
+	return (*keyp = TlsAlloc()) == TLS_OUT_OF_INDEXES ? EAGAIN : 0;
+}
+
+static inline int pthread_setspecific(pthread_key_t key, const void *value)
+{
+	return TlsSetValue(key, (void *)value) ? 0 : EINVAL;
+}
+
+static inline void *pthread_getspecific(pthread_key_t key)
+{
+	return TlsGetValue(key);
+}
+
+#endif /* PTHREAD_H */
diff --git a/compat/win32mmap.c b/compat/win32mmap.c
new file mode 100644
index 0000000..b58aa69
--- /dev/null
+++ b/compat/win32mmap.c
@@ -0,0 +1,41 @@
+#include "../git-compat-util.h"
+
+void *git_mmap(void *start, size_t length, int prot, int flags, int fd, off_t offset)
+{
+	HANDLE hmap;
+	void *temp;
+	off_t len;
+	struct stat st;
+	uint64_t o = offset;
+	uint32_t l = o & 0xFFFFFFFF;
+	uint32_t h = (o >> 32) & 0xFFFFFFFF;
+
+	if (!fstat(fd, &st))
+		len = st.st_size;
+	else
+		die("mmap: could not determine filesize");
+
+	if ((length + offset) > len)
+		length = xsize_t(len - offset);
+
+	if (!(flags & MAP_PRIVATE))
+		die("Invalid usage of mmap when built with USE_WIN32_MMAP");
+
+	hmap = CreateFileMapping((HANDLE)_get_osfhandle(fd), 0, PAGE_WRITECOPY,
+		0, 0, 0);
+
+	if (!hmap)
+		return MAP_FAILED;
+
+	temp = MapViewOfFileEx(hmap, FILE_MAP_COPY, h, l, length, start);
+
+	if (!CloseHandle(hmap))
+		warning("unable to close file mapping handle\n");
+
+	return temp ? temp : MAP_FAILED;
+}
+
+int git_munmap(void *start, size_t length)
+{
+	return !UnmapViewOfFile(start);
+}
diff --git a/compat/winansi.c b/compat/winansi.c
new file mode 100644
index 0000000..dedce21
--- /dev/null
+++ b/compat/winansi.c
@@ -0,0 +1,357 @@
+/*
+ * Copyright 2008 Peter Harris <git@peter.is-a-geek.org>
+ */
+
+#include "../git-compat-util.h"
+
+/*
+ Functions to be wrapped:
+*/
+#undef printf
+#undef fprintf
+#undef fputs
+/* TODO: write */
+
+/*
+ ANSI codes used by git: m, K
+
+ This file is git-specific. Therefore, this file does not attempt
+ to implement any codes that are not used by git.
+*/
+
+static HANDLE console;
+static WORD plain_attr;
+static WORD attr;
+static int negative;
+
+static void init(void)
+{
+	CONSOLE_SCREEN_BUFFER_INFO sbi;
+
+	static int initialized = 0;
+	if (initialized)
+		return;
+
+	console = GetStdHandle(STD_OUTPUT_HANDLE);
+	if (console == INVALID_HANDLE_VALUE)
+		console = NULL;
+
+	if (!console)
+		return;
+
+	GetConsoleScreenBufferInfo(console, &sbi);
+	attr = plain_attr = sbi.wAttributes;
+	negative = 0;
+
+	initialized = 1;
+}
+
+
+#define FOREGROUND_ALL (FOREGROUND_RED | FOREGROUND_GREEN | FOREGROUND_BLUE)
+#define BACKGROUND_ALL (BACKGROUND_RED | BACKGROUND_GREEN | BACKGROUND_BLUE)
+
+static void set_console_attr(void)
+{
+	WORD attributes = attr;
+	if (negative) {
+		attributes &= ~FOREGROUND_ALL;
+		attributes &= ~BACKGROUND_ALL;
+
+		/* This could probably use a bitmask
+		   instead of a series of ifs */
+		if (attr & FOREGROUND_RED)
+			attributes |= BACKGROUND_RED;
+		if (attr & FOREGROUND_GREEN)
+			attributes |= BACKGROUND_GREEN;
+		if (attr & FOREGROUND_BLUE)
+			attributes |= BACKGROUND_BLUE;
+
+		if (attr & BACKGROUND_RED)
+			attributes |= FOREGROUND_RED;
+		if (attr & BACKGROUND_GREEN)
+			attributes |= FOREGROUND_GREEN;
+		if (attr & BACKGROUND_BLUE)
+			attributes |= FOREGROUND_BLUE;
+	}
+	SetConsoleTextAttribute(console, attributes);
+}
+
+static void erase_in_line(void)
+{
+	CONSOLE_SCREEN_BUFFER_INFO sbi;
+	DWORD dummy; /* Needed for Windows 7 (or Vista) regression */
+
+	if (!console)
+		return;
+
+	GetConsoleScreenBufferInfo(console, &sbi);
+	FillConsoleOutputCharacterA(console, ' ',
+		sbi.dwSize.X - sbi.dwCursorPosition.X, sbi.dwCursorPosition,
+		&dummy);
+}
+
+
+static const char *set_attr(const char *str)
+{
+	const char *func;
+	size_t len = strspn(str, "0123456789;");
+	func = str + len;
+
+	switch (*func) {
+	case 'm':
+		do {
+			long val = strtol(str, (char **)&str, 10);
+			switch (val) {
+			case 0: /* reset */
+				attr = plain_attr;
+				negative = 0;
+				break;
+			case 1: /* bold */
+				attr |= FOREGROUND_INTENSITY;
+				break;
+			case 2:  /* faint */
+			case 22: /* normal */
+				attr &= ~FOREGROUND_INTENSITY;
+				break;
+			case 3:  /* italic */
+				/* Unsupported */
+				break;
+			case 4:  /* underline */
+			case 21: /* double underline */
+				/* Wikipedia says this flag does nothing */
+				/* Furthermore, mingw doesn't define this flag
+				attr |= COMMON_LVB_UNDERSCORE; */
+				break;
+			case 24: /* no underline */
+				/* attr &= ~COMMON_LVB_UNDERSCORE; */
+				break;
+			case 5:  /* slow blink */
+			case 6:  /* fast blink */
+				/* We don't have blink, but we do have
+				   background intensity */
+				attr |= BACKGROUND_INTENSITY;
+				break;
+			case 25: /* no blink */
+				attr &= ~BACKGROUND_INTENSITY;
+				break;
+			case 7:  /* negative */
+				negative = 1;
+				break;
+			case 27: /* positive */
+				negative = 0;
+				break;
+			case 8:  /* conceal */
+			case 28: /* reveal */
+				/* Unsupported */
+				break;
+			case 30: /* Black */
+				attr &= ~FOREGROUND_ALL;
+				break;
+			case 31: /* Red */
+				attr &= ~FOREGROUND_ALL;
+				attr |= FOREGROUND_RED;
+				break;
+			case 32: /* Green */
+				attr &= ~FOREGROUND_ALL;
+				attr |= FOREGROUND_GREEN;
+				break;
+			case 33: /* Yellow */
+				attr &= ~FOREGROUND_ALL;
+				attr |= FOREGROUND_RED | FOREGROUND_GREEN;
+				break;
+			case 34: /* Blue */
+				attr &= ~FOREGROUND_ALL;
+				attr |= FOREGROUND_BLUE;
+				break;
+			case 35: /* Magenta */
+				attr &= ~FOREGROUND_ALL;
+				attr |= FOREGROUND_RED | FOREGROUND_BLUE;
+				break;
+			case 36: /* Cyan */
+				attr &= ~FOREGROUND_ALL;
+				attr |= FOREGROUND_GREEN | FOREGROUND_BLUE;
+				break;
+			case 37: /* White */
+				attr |= FOREGROUND_RED |
+					FOREGROUND_GREEN |
+					FOREGROUND_BLUE;
+				break;
+			case 38: /* Unknown */
+				break;
+			case 39: /* reset */
+				attr &= ~FOREGROUND_ALL;
+				attr |= (plain_attr & FOREGROUND_ALL);
+				break;
+			case 40: /* Black */
+				attr &= ~BACKGROUND_ALL;
+				break;
+			case 41: /* Red */
+				attr &= ~BACKGROUND_ALL;
+				attr |= BACKGROUND_RED;
+				break;
+			case 42: /* Green */
+				attr &= ~BACKGROUND_ALL;
+				attr |= BACKGROUND_GREEN;
+				break;
+			case 43: /* Yellow */
+				attr &= ~BACKGROUND_ALL;
+				attr |= BACKGROUND_RED | BACKGROUND_GREEN;
+				break;
+			case 44: /* Blue */
+				attr &= ~BACKGROUND_ALL;
+				attr |= BACKGROUND_BLUE;
+				break;
+			case 45: /* Magenta */
+				attr &= ~BACKGROUND_ALL;
+				attr |= BACKGROUND_RED | BACKGROUND_BLUE;
+				break;
+			case 46: /* Cyan */
+				attr &= ~BACKGROUND_ALL;
+				attr |= BACKGROUND_GREEN | BACKGROUND_BLUE;
+				break;
+			case 47: /* White */
+				attr |= BACKGROUND_RED |
+					BACKGROUND_GREEN |
+					BACKGROUND_BLUE;
+				break;
+			case 48: /* Unknown */
+				break;
+			case 49: /* reset */
+				attr &= ~BACKGROUND_ALL;
+				attr |= (plain_attr & BACKGROUND_ALL);
+				break;
+			default:
+				/* Unsupported code */
+				break;
+			}
+			str++;
+		} while (*(str-1) == ';');
+
+		set_console_attr();
+		break;
+	case 'K':
+		erase_in_line();
+		break;
+	default:
+		/* Unsupported code */
+		break;
+	}
+
+	return func + 1;
+}
+
+static int ansi_emulate(const char *str, FILE *stream)
+{
+	int rv = 0;
+	const char *pos = str;
+
+	while (*pos) {
+		pos = strstr(str, "\033[");
+		if (pos) {
+			size_t len = pos - str;
+
+			if (len) {
+				size_t out_len = fwrite(str, 1, len, stream);
+				rv += out_len;
+				if (out_len < len)
+					return rv;
+			}
+
+			str = pos + 2;
+			rv += 2;
+
+			fflush(stream);
+
+			pos = set_attr(str);
+			rv += pos - str;
+			str = pos;
+		} else {
+			rv += strlen(str);
+			fputs(str, stream);
+			return rv;
+		}
+	}
+	return rv;
+}
+
+int winansi_fputs(const char *str, FILE *stream)
+{
+	int rv;
+
+	if (!isatty(fileno(stream)))
+		return fputs(str, stream);
+
+	init();
+
+	if (!console)
+		return fputs(str, stream);
+
+	rv = ansi_emulate(str, stream);
+
+	if (rv >= 0)
+		return 0;
+	else
+		return EOF;
+}
+
+static int winansi_vfprintf(FILE *stream, const char *format, va_list list)
+{
+	int len, rv;
+	char small_buf[256];
+	char *buf = small_buf;
+	va_list cp;
+
+	if (!isatty(fileno(stream)))
+		goto abort;
+
+	init();
+
+	if (!console)
+		goto abort;
+
+	va_copy(cp, list);
+	len = vsnprintf(small_buf, sizeof(small_buf), format, cp);
+	va_end(cp);
+
+	if (len > sizeof(small_buf) - 1) {
+		buf = malloc(len + 1);
+		if (!buf)
+			goto abort;
+
+		len = vsnprintf(buf, len + 1, format, list);
+	}
+
+	rv = ansi_emulate(buf, stream);
+
+	if (buf != small_buf)
+		free(buf);
+	return rv;
+
+abort:
+	rv = vfprintf(stream, format, list);
+	return rv;
+}
+
+int winansi_fprintf(FILE *stream, const char *format, ...)
+{
+	va_list list;
+	int rv;
+
+	va_start(list, format);
+	rv = winansi_vfprintf(stream, format, list);
+	va_end(list);
+
+	return rv;
+}
+
+int winansi_printf(const char *format, ...)
+{
+	va_list list;
+	int rv;
+
+	va_start(list, format);
+	rv = winansi_vfprintf(stdout, format, list);
+	va_end(list);
+
+	return rv;
+}