/////////////////////////////////////////////////////////////////////////////// /// \file cpp_regex_traits.hpp /// Contains the definition of the cpp_regex_traits\<\> template, which is a /// wrapper for std::locale that can be used to customize the behavior of /// static and dynamic regexes. // // Copyright 2008 Eric Niebler. Distributed under the Boost // Software License, Version 1.0. (See accompanying file // LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt) #ifndef BOOST_XPRESSIVE_TRAITS_CPP_REGEX_TRAITS_HPP_EAN_10_04_2005 #define BOOST_XPRESSIVE_TRAITS_CPP_REGEX_TRAITS_HPP_EAN_10_04_2005 // MS compatible compilers support #pragma once #if defined(_MSC_VER) && (_MSC_VER >= 1020) # pragma once #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include // From John Maddock: // Fix for gcc prior to 3.4: std::ctype doesn't allow masks to be combined, for example: // std::use_facet >(locale()).is(std::ctype_base::lower|std::ctype_base::upper, L'a'); // incorrectly returns false. // NOTE: later version of the gcc define __GLIBCXX__, not __GLIBCPP__ #if BOOST_WORKAROUND(__GLIBCPP__, != 0) # define BOOST_XPRESSIVE_BUGGY_CTYPE_FACET #endif namespace boost { namespace xpressive { namespace detail { // define an unsigned integral typedef of the same size as std::ctype_base::mask typedef boost::uint_t::least umask_t; BOOST_MPL_ASSERT_RELATION(sizeof(std::ctype_base::mask), ==, sizeof(umask_t)); // Calculate what the size of the umaskex_t type should be to fix the 3 extra bitmasks // 11 char categories in ctype_base // + 3 extra categories for xpressive // = 14 total bits needed int const umaskex_bits = (14 > (sizeof(umask_t) * CHAR_BIT)) ? 14 : sizeof(umask_t) * CHAR_BIT; // define an unsigned integral type with at least umaskex_bits typedef boost::uint_t::fast umaskex_t; BOOST_MPL_ASSERT_RELATION(sizeof(umask_t), <=, sizeof(umaskex_t)); // cast a ctype mask to a umaskex_t template struct mask_cast { BOOST_STATIC_CONSTANT(umaskex_t, value = static_cast(Mask)); }; #ifdef __CYGWIN__ // Work around a gcc warning on cygwin template<> struct mask_cast { BOOST_MPL_ASSERT_RELATION('\227', ==, std::ctype_base::print); BOOST_STATIC_CONSTANT(umaskex_t, value = 0227); }; #endif #ifndef BOOST_NO_INCLASS_MEMBER_INITIALIZATION template umaskex_t const mask_cast::value; #endif #ifndef BOOST_XPRESSIVE_BUGGY_CTYPE_FACET // an unsigned integer with the highest bit set umaskex_t const highest_bit = static_cast(1) << (sizeof(umaskex_t) * CHAR_BIT - 1); /////////////////////////////////////////////////////////////////////////////// // unused_mask // find a bit in an int that isn't set template struct unused_mask { BOOST_STATIC_ASSERT(1 != Out); BOOST_STATIC_CONSTANT(umaskex_t, value = (unused_mask> 1)>::value)); }; template struct unused_mask { BOOST_STATIC_CONSTANT(umaskex_t, value = Out); }; #ifndef BOOST_NO_INCLASS_MEMBER_INITIALIZATION template umaskex_t const unused_mask::value; #endif umaskex_t const std_ctype_alnum = mask_cast::value; umaskex_t const std_ctype_alpha = mask_cast::value; umaskex_t const std_ctype_cntrl = mask_cast::value; umaskex_t const std_ctype_digit = mask_cast::value; umaskex_t const std_ctype_graph = mask_cast::value; umaskex_t const std_ctype_lower = mask_cast::value; umaskex_t const std_ctype_print = mask_cast::value; umaskex_t const std_ctype_punct = mask_cast::value; umaskex_t const std_ctype_space = mask_cast::value; umaskex_t const std_ctype_upper = mask_cast::value; umaskex_t const std_ctype_xdigit = mask_cast::value; // Reserve some bits for the implementation #if defined(__GLIBCXX__) umaskex_t const std_ctype_reserved = 0x8000; #elif defined(_CPPLIB_VER) && defined(BOOST_WINDOWS) umaskex_t const std_ctype_reserved = 0x8200; #elif defined(_LIBCPP_VERSION) umaskex_t const std_ctype_reserved = 0x8000; #else umaskex_t const std_ctype_reserved = 0; #endif // Bitwise-or all the ctype masks together umaskex_t const all_ctype_masks = std_ctype_reserved | std_ctype_alnum | std_ctype_alpha | std_ctype_cntrl | std_ctype_digit | std_ctype_graph | std_ctype_lower | std_ctype_print | std_ctype_punct | std_ctype_space | std_ctype_upper | std_ctype_xdigit; // define a new mask for "underscore" ("word" == alnum | underscore) umaskex_t const non_std_ctype_underscore = unused_mask::value; // define a new mask for "blank" umaskex_t const non_std_ctype_blank = unused_mask::value; // define a new mask for "newline" umaskex_t const non_std_ctype_newline = unused_mask::value; #else /////////////////////////////////////////////////////////////////////////////// // Ugly work-around for buggy ctype facets. umaskex_t const std_ctype_alnum = 1 << 0; umaskex_t const std_ctype_alpha = 1 << 1; umaskex_t const std_ctype_cntrl = 1 << 2; umaskex_t const std_ctype_digit = 1 << 3; umaskex_t const std_ctype_graph = 1 << 4; umaskex_t const std_ctype_lower = 1 << 5; umaskex_t const std_ctype_print = 1 << 6; umaskex_t const std_ctype_punct = 1 << 7; umaskex_t const std_ctype_space = 1 << 8; umaskex_t const std_ctype_upper = 1 << 9; umaskex_t const std_ctype_xdigit = 1 << 10; umaskex_t const non_std_ctype_underscore = 1 << 11; umaskex_t const non_std_ctype_blank = 1 << 12; umaskex_t const non_std_ctype_newline = 1 << 13; static umaskex_t const std_masks[] = { mask_cast::value , mask_cast::value , mask_cast::value , mask_cast::value , mask_cast::value , mask_cast::value , mask_cast::value , mask_cast::value , mask_cast::value , mask_cast::value , mask_cast::value }; inline int mylog2(umaskex_t i) { return "\0\0\1\0\2\0\0\0\3"[i & 0xf] + "\0\4\5\0\6\0\0\0\7"[(i & 0xf0) >> 04] + "\0\10\11\0\12\0\0\0\13"[(i & 0xf00) >> 010]; } #endif // convenient constant for the extra masks umaskex_t const non_std_ctype_masks = non_std_ctype_underscore | non_std_ctype_blank | non_std_ctype_newline; /////////////////////////////////////////////////////////////////////////////// // cpp_regex_traits_base // BUGBUG this should be replaced with a regex facet that lets you query for // an array of underscore characters and an array of line separator characters. template struct cpp_regex_traits_base { protected: void imbue(std::locale const &) { } static bool is(std::ctype const &ct, Char ch, umaskex_t mask) { #ifndef BOOST_XPRESSIVE_BUGGY_CTYPE_FACET if(ct.is((std::ctype_base::mask)(umask_t)mask, ch)) { return true; } // HACKHACK Cygwin and mingw have buggy ctype facets for wchar_t #if defined(__CYGWIN__) || defined(__MINGW32_VERSION) if (std::ctype_base::xdigit == ((std::ctype_base::mask)(umask_t)mask & std::ctype_base::xdigit)) { typename std::char_traits::int_type i = std::char_traits::to_int_type(ch); if(UCHAR_MAX >= i && std::isxdigit(static_cast(i))) return true; } #endif #else umaskex_t tmp = mask & ~non_std_ctype_masks; for(umaskex_t i; 0 != (i = (tmp & (~tmp+1))); tmp &= ~i) { std::ctype_base::mask m = (std::ctype_base::mask)(umask_t)std_masks[mylog2(i)]; if(ct.is(m, ch)) { return true; } } #endif return ((mask & non_std_ctype_blank) && cpp_regex_traits_base::is_blank(ch)) || ((mask & non_std_ctype_underscore) && cpp_regex_traits_base::is_underscore(ch)) || ((mask & non_std_ctype_newline) && cpp_regex_traits_base::is_newline(ch)); } private: static bool is_blank(Char ch) { BOOST_MPL_ASSERT_RELATION('\t', ==, L'\t'); BOOST_MPL_ASSERT_RELATION(' ', ==, L' '); return L' ' == ch || L'\t' == ch; } static bool is_underscore(Char ch) { BOOST_MPL_ASSERT_RELATION('_', ==, L'_'); return L'_' == ch; } static bool is_newline(Char ch) { BOOST_MPL_ASSERT_RELATION('\r', ==, L'\r'); BOOST_MPL_ASSERT_RELATION('\n', ==, L'\n'); BOOST_MPL_ASSERT_RELATION('\f', ==, L'\f'); return L'\r' == ch || L'\n' == ch || L'\f' == ch || (1 < SizeOfChar && (0x2028u == ch || 0x2029u == ch || 0x85u == ch)); } }; #ifndef BOOST_XPRESSIVE_BUGGY_CTYPE_FACET template struct cpp_regex_traits_base { protected: void imbue(std::locale const &loc) { int i = 0; Char allchars[UCHAR_MAX + 1]; for(i = 0; i <= static_cast(UCHAR_MAX); ++i) { allchars[i] = static_cast(i); } std::ctype const &ct = BOOST_USE_FACET(std::ctype, loc); std::ctype_base::mask tmp[UCHAR_MAX + 1]; ct.is(allchars, allchars + UCHAR_MAX + 1, tmp); for(i = 0; i <= static_cast(UCHAR_MAX); ++i) { this->masks_[i] = static_cast(tmp[i]); BOOST_ASSERT(0 == (this->masks_[i] & non_std_ctype_masks)); } this->masks_[static_cast('_')] |= non_std_ctype_underscore; this->masks_[static_cast(' ')] |= non_std_ctype_blank; this->masks_[static_cast('\t')] |= non_std_ctype_blank; this->masks_[static_cast('\n')] |= non_std_ctype_newline; this->masks_[static_cast('\r')] |= non_std_ctype_newline; this->masks_[static_cast('\f')] |= non_std_ctype_newline; } bool is(std::ctype const &, Char ch, umaskex_t mask) const { return 0 != (this->masks_[static_cast(ch)] & mask); } private: umaskex_t masks_[UCHAR_MAX + 1]; }; #endif } // namespace detail /////////////////////////////////////////////////////////////////////////////// // cpp_regex_traits // /// \brief Encapsaulates a std::locale for use by the /// basic_regex\<\> class template. template struct cpp_regex_traits : detail::cpp_regex_traits_base { typedef Char char_type; typedef std::basic_string string_type; typedef std::locale locale_type; typedef detail::umaskex_t char_class_type; typedef regex_traits_version_2_tag version_tag; typedef detail::cpp_regex_traits_base base_type; /// Initialize a cpp_regex_traits object to use the specified std::locale, /// or the global std::locale if none is specified. /// cpp_regex_traits(locale_type const &loc = locale_type()) : base_type() , loc_() { this->imbue(loc); } /// Checks two cpp_regex_traits objects for equality /// /// \return this->getloc() == that.getloc(). bool operator ==(cpp_regex_traits const &that) const { return this->loc_ == that.loc_; } /// Checks two cpp_regex_traits objects for inequality /// /// \return this->getloc() != that.getloc(). bool operator !=(cpp_regex_traits const &that) const { return this->loc_ != that.loc_; } /// Convert a char to a Char /// /// \param ch The source character. /// \return std::use_facet\ \>(this->getloc()).widen(ch). char_type widen(char ch) const { return this->ctype_->widen(ch); } /// Returns a hash value for a Char in the range [0, UCHAR_MAX] /// /// \param ch The source character. /// \return a value between 0 and UCHAR_MAX, inclusive. static unsigned char hash(char_type ch) { return static_cast(std::char_traits::to_int_type(ch)); } /// No-op /// /// \param ch The source character. /// \return ch static char_type translate(char_type ch) { return ch; } /// Converts a character to lower-case using the internally-stored std::locale. /// /// \param ch The source character. /// \return std::tolower(ch, this->getloc()). char_type translate_nocase(char_type ch) const { return this->ctype_->tolower(ch); } /// Converts a character to lower-case using the internally-stored std::locale. /// /// \param ch The source character. /// \return std::tolower(ch, this->getloc()). char_type tolower(char_type ch) const { return this->ctype_->tolower(ch); } /// Converts a character to upper-case using the internally-stored std::locale. /// /// \param ch The source character. /// \return std::toupper(ch, this->getloc()). char_type toupper(char_type ch) const { return this->ctype_->toupper(ch); } /// Returns a string_type containing all the characters that compare equal /// disregrarding case to the one passed in. This function can only be called /// if has_fold_case\ \>::value is true. /// /// \param ch The source character. /// \return string_type containing all chars which are equal to ch when disregarding /// case string_type fold_case(char_type ch) const { BOOST_MPL_ASSERT((is_same)); char_type ntcs[] = { this->ctype_->tolower(ch) , this->ctype_->toupper(ch) , 0 }; if(ntcs[1] == ntcs[0]) ntcs[1] = 0; return string_type(ntcs); } /// Checks to see if a character is within a character range. /// /// \param first The bottom of the range, inclusive. /// \param last The top of the range, inclusive. /// \param ch The source character. /// \return first <= ch && ch <= last. static bool in_range(char_type first, char_type last, char_type ch) { return first <= ch && ch <= last; } /// Checks to see if a character is within a character range, irregardless of case. /// /// \param first The bottom of the range, inclusive. /// \param last The top of the range, inclusive. /// \param ch The source character. /// \return in_range(first, last, ch) || in_range(first, last, tolower(ch, this->getloc())) || /// in_range(first, last, toupper(ch, this->getloc())) /// \attention The default implementation doesn't do proper Unicode /// case folding, but this is the best we can do with the standard /// ctype facet. bool in_range_nocase(char_type first, char_type last, char_type ch) const { // NOTE: this default implementation doesn't do proper Unicode // case folding, but this is the best we can do with the standard // std::ctype facet. return this->in_range(first, last, ch) || this->in_range(first, last, this->ctype_->toupper(ch)) || this->in_range(first, last, this->ctype_->tolower(ch)); } /// INTERNAL ONLY //string_type transform(char_type const *begin, char_type const *end) const //{ // return this->collate_->transform(begin, end); //} /// Returns a sort key for the character sequence designated by the iterator range [F1, F2) /// such that if the character sequence [G1, G2) sorts before the character sequence [H1, H2) /// then v.transform(G1, G2) \< v.transform(H1, H2). /// /// \attention Not currently used template string_type transform(FwdIter begin, FwdIter end) const { //string_type str(begin, end); //return this->transform(str.data(), str.data() + str.size()); BOOST_ASSERT(false); return string_type(); } /// Returns a sort key for the character sequence designated by the iterator range [F1, F2) /// such that if the character sequence [G1, G2) sorts before the character sequence [H1, H2) /// when character case is not considered then /// v.transform_primary(G1, G2) \< v.transform_primary(H1, H2). /// /// \attention Not currently used template string_type transform_primary(FwdIter begin, FwdIter end) const { BOOST_ASSERT(false); // TODO implement me return string_type(); } /// Returns a sequence of characters that represents the collating element /// consisting of the character sequence designated by the iterator range [F1, F2). /// Returns an empty string if the character sequence is not a valid collating element. /// /// \attention Not currently used template string_type lookup_collatename(FwdIter begin, FwdIter end) const { BOOST_ASSERT(false); // TODO implement me return string_type(); } /// For the character class name represented by the specified character sequence, /// return the corresponding bitmask representation. /// /// \param begin A forward iterator to the start of the character sequence representing /// the name of the character class. /// \param end The end of the character sequence. /// \param icase Specifies whether the returned bitmask should represent the case-insensitive /// version of the character class. /// \return A bitmask representing the character class. template char_class_type lookup_classname(FwdIter begin, FwdIter end, bool icase) const { static detail::umaskex_t const icase_masks = detail::std_ctype_lower | detail::std_ctype_upper; BOOST_ASSERT(begin != end); char_class_type char_class = this->lookup_classname_impl_(begin, end); if(0 == char_class) { // convert the string to lowercase string_type classname(begin, end); for(typename string_type::size_type i = 0, len = classname.size(); i < len; ++i) { classname[i] = this->translate_nocase(classname[i]); } char_class = this->lookup_classname_impl_(classname.begin(), classname.end()); } // erase case-sensitivity if icase==true if(icase && 0 != (char_class & icase_masks)) { char_class |= icase_masks; } return char_class; } /// Tests a character against a character class bitmask. /// /// \param ch The character to test. /// \param mask The character class bitmask against which to test. /// \pre mask is a bitmask returned by lookup_classname, or is several such masks bit-or'ed /// together. /// \return true if the character is a member of any of the specified character classes, false /// otherwise. bool isctype(char_type ch, char_class_type mask) const { return this->base_type::is(*this->ctype_, ch, mask); } /// Convert a digit character into the integer it represents. /// /// \param ch The digit character. /// \param radix The radix to use for the conversion. /// \pre radix is one of 8, 10, or 16. /// \return -1 if ch is not a digit character, the integer value of the character otherwise. /// The conversion is performed by imbueing a std::stringstream with this-\>getloc(); /// setting the radix to one of oct, hex or dec; inserting ch into the stream; and /// extracting an int. int value(char_type ch, int radix) const { BOOST_ASSERT(8 == radix || 10 == radix || 16 == radix); int val = -1; std::basic_stringstream str; str.imbue(this->getloc()); str << (8 == radix ? std::oct : (16 == radix ? std::hex : std::dec)); str.put(ch); str >> val; return str.fail() ? -1 : val; } /// Imbues *this with loc /// /// \param loc A std::locale. /// \return the previous std::locale used by *this. locale_type imbue(locale_type loc) { locale_type old_loc = this->loc_; this->loc_ = loc; this->ctype_ = &BOOST_USE_FACET(std::ctype, this->loc_); //this->collate_ = &BOOST_USE_FACET(std::collate, this->loc_); this->base_type::imbue(this->loc_); return old_loc; } /// Returns the current std::locale used by *this. /// locale_type getloc() const { return this->loc_; } private: /////////////////////////////////////////////////////////////////////////////// // char_class_pair /// INTERNAL ONLY struct char_class_pair { char_type const *class_name_; char_class_type class_type_; }; /////////////////////////////////////////////////////////////////////////////// // char_class /// INTERNAL ONLY static char_class_pair const &char_class(std::size_t j) { static char_class_pair const s_char_class_map[] = { { BOOST_XPR_CSTR_(char_type, "alnum"), detail::std_ctype_alnum } , { BOOST_XPR_CSTR_(char_type, "alpha"), detail::std_ctype_alpha } , { BOOST_XPR_CSTR_(char_type, "blank"), detail::non_std_ctype_blank } , { BOOST_XPR_CSTR_(char_type, "cntrl"), detail::std_ctype_cntrl } , { BOOST_XPR_CSTR_(char_type, "d"), detail::std_ctype_digit } , { BOOST_XPR_CSTR_(char_type, "digit"), detail::std_ctype_digit } , { BOOST_XPR_CSTR_(char_type, "graph"), detail::std_ctype_graph } , { BOOST_XPR_CSTR_(char_type, "lower"), detail::std_ctype_lower } , { BOOST_XPR_CSTR_(char_type, "newline"),detail::non_std_ctype_newline } , { BOOST_XPR_CSTR_(char_type, "print"), detail::std_ctype_print } , { BOOST_XPR_CSTR_(char_type, "punct"), detail::std_ctype_punct } , { BOOST_XPR_CSTR_(char_type, "s"), detail::std_ctype_space } , { BOOST_XPR_CSTR_(char_type, "space"), detail::std_ctype_space } , { BOOST_XPR_CSTR_(char_type, "upper"), detail::std_ctype_upper } , { BOOST_XPR_CSTR_(char_type, "w"), detail::std_ctype_alnum | detail::non_std_ctype_underscore } , { BOOST_XPR_CSTR_(char_type, "xdigit"), detail::std_ctype_xdigit } , { 0, 0 } }; return s_char_class_map[j]; } /////////////////////////////////////////////////////////////////////////////// // lookup_classname_impl /// INTERNAL ONLY template static char_class_type lookup_classname_impl_(FwdIter begin, FwdIter end) { // find the classname typedef cpp_regex_traits this_t; for(std::size_t j = 0; 0 != this_t::char_class(j).class_name_; ++j) { if(this_t::compare_(this_t::char_class(j).class_name_, begin, end)) { return this_t::char_class(j).class_type_; } } return 0; } /// INTERNAL ONLY template static bool compare_(char_type const *name, FwdIter begin, FwdIter end) { for(; *name && begin != end; ++name, ++begin) { if(*name != *begin) { return false; } } return !*name && begin == end; } locale_type loc_; std::ctype const *ctype_; //std::collate const *collate_; }; /////////////////////////////////////////////////////////////////////////////// // cpp_regex_traits<>::hash specializations template<> inline unsigned char cpp_regex_traits::hash(unsigned char ch) { return ch; } template<> inline unsigned char cpp_regex_traits::hash(char ch) { return static_cast(ch); } template<> inline unsigned char cpp_regex_traits::hash(signed char ch) { return static_cast(ch); } #ifndef BOOST_XPRESSIVE_NO_WREGEX template<> inline unsigned char cpp_regex_traits::hash(wchar_t ch) { return static_cast(ch); } #endif // Narrow C++ traits has fold_case() member function. template<> struct has_fold_case > : mpl::true_ { }; }} #endif