0.9.7 (1232780) - bug fix for -MD and -MF - bug fix for gcc 4.6 plugins 0.9.6 (1158638) - fix installation issues - add a value type of ticks in /proc - and make it long long - fix kFreeBSD build - fix run-failure with glibc 2.11 - allow tgz as extension - support more local args (bnc#625621) 0.9.5 (1083362) - fix: close the file descriptor (novell bug 525799) - log PID even if no prefix is specified - allow get_msg() to fetch data from the kernel even if timeout is 0. - clean up event loop of gcc invocation - Install an "icerun" symlink that will serialize commands through icecream's local daemon 0.9.4 (961598): - fix compilation warnings - don't leak file descriptor to create-env - don't use the shell to call simple commands - assert current_kids is > 0 before decrementing (it's unsigned) - also check for EAGAIN (needed on OS X) - add ICECC_LOGFILE to get trace() output into a file - Fix compilation when using -Wformat -Werror=format-security flags - make a special exception for - in rest args (bnc#495786) - Fix way icecream changes permissions of /var/cache/icecream 0.9.3 (926474): - Fix FTBFS with GCC 4.4 - Explicitly set group ownership and permissions on the base cache dir and target dir in case root has a umask of e.g. 0077. Otherwise nobody user won't be ableto untar. - make sure that a too strict umask doesn't ruin the game fixes https://bugzilla.novell.com/show_bug.cgi?id=446329 - some create-env fixes for fedora 0.9.2 (879112): - On OS X, append the release number to Darwin (Darwin8 for Tiger, Darwin9 for Leopard) as their icecream environments are not compatible - Add accessor for the hostname of the scheduler, so that we can show it in a tooltip in icemon in the future. - Only localize job if we find -Wa,.*-a[a-z]*= in an argument - also add accessor for network name - if called as "icecc" (i.e., not through a symlink) and the first argument starts with a slash, use that as the command to run. 0.9.1 (822102): - trying to support gentoo x86_64 - -mcpu=native has to compile locally (gentoo bug 183586#c13) - don't compile C files with C++ compile if ICECC_CXX is set 0.9.0 (807014): - create a working env on OS X - don't hardcode tar path - implement load informations on OS X - make it work again on non-Linux platforms: Jobs still reported their environment platform as i386 etc, even if they came from Darwin or BSD (thus the scheduler never finds a matching compile server, or wrongly directs a non-Linux client to use a Linux compile server) - fixing the init script - patch by Tais M. Hansen to ignore directories - compile on OS X 10.5 - some preparations for rsync support - fix valgrind warning - add configure check for librsync - don't kick daemons that run as user, just don't give them any jobs. this helps the people that just don't trust the daemon to be able to at least send out jobs - remove the now pointless "runasuser" option. it will just not accept any jobs then - add log message when the local daemon hangs again - accept TERM=dumb as well - implement support for scheduling local non-compile jobs. For example: you can add a symlink named "meinproc4" pointing to icecc to your path, and it will automatically schedule the job and then invoke meinproc4 on it locally. the advantage is that you can use cmake -j50 without having to worry that there are going to be 50 meinproc jobs killing your machine. - do an asynchronous connection to the scheduler: * helps when the scheduler is down, as the daemon is otherwise locked up then and can't schedule local jobs * helps when the scheduler is flaky and blocked itself on network (like in the suse network). - fix connect if scheduler is known - send jobs to a monitor non-blocking - implement host blocking - switch monitor connections to bulk transfer - fix double deletion when control connection disconnects - add greeting banner that gives a rough statistic (for pretty monitoring graphs :) - protocol cleanup - fix error handling - adding more API for better monitor behaviour - various protections against malicious requests - add a try-restart target for the rpm package - always read the scheduler message in complete before trying to fetch a message. fixes a daemon hangup (surprising noone noticed this before) - various code cleanups - don't flood the scheduler - implement bulk sending - reduce overhead by not sending a full path - implement much shorter monitor messages - reenable cork handling on linux only. helps a lot - fix job statistics in monitor - check if the compile server sent us a message before die'ing another death - avoid killing the daemon on protocol errors - support "quit" on the normal daemon port as well - ext/hash_set is not needed and does not exist anymore - DoS protection for the scheduler - avoid that daemons hang around in environment installation state forever - fix daemons being overrun by wrongly assigned jobs while they're busy installing - add a segfault catcher - don't allow tiny environments - they are very likely broken - always relogin to the scheduler even if environment transfer failed - implement proper handling for PCH files for e.g. Qt compilation - fix compilation with glibc 2.8 0.8.0 (657903): - Find md5 on FreeBSD - apply patch to support cygwin a bit better - write a pid file. patch by Ismail Doenmez - precompiled header support, needed for compiling Qt 4.3 - add support for --no-remote. patch by Matt Gruenke - Make it work on DragonFlyBSD. - provide SuSEfirewall2 information - Make the result of make dist-gzip usable. - update node name each time we login to the scheduler. - treat -combine and -fdump* as always local - Merged make-it-cool branch: * implement job preloading * asynchronous install of new environments * uses tcp keepalive for network connection tracking - make it work on OS X * don' hardcode tar path * create environments correctly on OS X * report load correctly on mach kernels (Darwin) * prepend os to the machine type if it's not linux (resulting in e.g. Darwin_PowerMacintosh) 0.7.14 (583722): - fix current_kids getting out of sync if send_scheduler failes in an unfortunate moment - fix reporting the error - add more to the internals dump - trying to make the scheduler kicks a little bit less frequent - if the compiler is supposed to keep temp files, then we have to do it locally 0.7.13 (583347): - use file -L to follow symlinks in create-env - fix an easy valgrind error - make handling of the return values consistent - if the client is in WAITCOMPILE then it's waiting for some other host to compile, not this one. So job == 0 -> crash - fix crash if the scheduler goes away while daemon transfers environment - apparently sometimes gcc can hang forever in some cases. By the time the client disconects, we know we don't have to wait any longer, because there is nobody left caring about the result. - update node name each time we login to the scheduler. Fixes tons of "linux" hosts appearing in the icecream monitor 0.7.12 (r581454): - fixing error handling when scheduler restarts - do not waitpid before we're sure we read all of g++'s output otherwise g++ waits for us to read and we're waiting for g++ to finish -> deadlock 0.7.11 (r581076): - fix a crash in the daemon when the scheduler was gone while local jobs were waiting for finishing - separate stat handling from ping handling to avoid excessive stat/ping loops (increases internal version number) - only reset scheduler ping time if the scheduler ping'ed us - even when we can't determine native environment, we can still use the daemon for inter-process locking instead of falling back to file locking. - quicker reap of dead daemons - improved load guessing - fix stupid logic bug in ping tracking 0.7.10 (r580794): - handle errors in installing environments correctly - block daemons that have full discs - add -pipe to the command line to reduce disk usage - fix cancelling of jobs awaiting a remote job (were hanging in scheduler forever) - if ICECC=no is set, don't try to be clever - adding two more flags for local compilation (profile feedback related) - flush debug files before every fork to avoid dups in log output - be stricter in what messages are required to keep the daemon<->scheduler communication intact (network drops again) 0.7.9 (r580498): - scheduler will ping the daemon periodically, daemon will disconnect if not pinged from time to time (to avoid network drops being unnoticed) - removed some debug output - make logging more consistent - increased internal protocol version (still compatible) - try even harder to avoid races in daemon - rework how clients are catched - some random cleanup - remove the "this should be an exception" output to avoid failing configure checks for nothing - make sure the assembler is dead before we cleanup object files 0.7.8 (r579795): - fd leak fix, which caused a deadlock under certain conditions - rework some FreeBSD patches that caused problems - fix race between select and SIGCHILD 0.7.7 (r574260): - clear all internal maps - patches by Frerich to support FreeBSD - avoid busy loop in bizarre conditions - found another case of endless hanging jobs - some process fixes 0.7.6 (r561764): - report all daemon error message to the client - fix support for biarch machines that have only one environment installed with another environment with the same name and different architecture being around - work-around a bug in older libstdc++ - ensure nobody uses -u root 0.7.5 (r561480): - don't block while looking for a scheduler over broadcast search. - fix logrotate - immediately kick broken daemon connections 0.7.4 (r559927): - Add -fpreprocessed argument - Fix daemons getting stuck after a while with 100% CPU - fix for kubuntu where libs are deeply nested such as /lib/tls/i686/cmov/libc.so.6 - find the compiler even though icecc wasn't in the path - fix deletion of old environments - a lot more error handling - fix assertion failures in daemon upon sigchild delivery - use -fPIC also for C files - avoid race between local and remote daemon about who has the right exit status - rename the package to icecc 0.7.3 (r552930): - more work on error handling - corrected icecc --help output - handle local icecream load correctly - reduce the number of load updates from daemon to scheduler - early launch the compiler even before the whole preprocessed file has been received - handle low-latency networks better - builtin colorgcc - let the daemon schedule local load even without scheduler (important for multi processor machines) 0.7.2 (r549574): - split out libicecream-devel for the monitor - introducing ICECC_PREFERRED_HOST (=mybigfathost) to let the scheduler ignore all rules - this is meant to be a debugging aid - fix network performance between client and local daemon - replaced create-env with icecc --build-native (respecting the user's $PATH by default) 0.7.1 (r548846): - Monitor is an extra package now - don't make non-existant cross compilers fatal, but just use the other environments in ICECC_VERSION - always handle aborted syscalls - update minilzo - never exit() or return from fork()'ed code paths, 0.7.0 (r547196): - WARNING: the protocol is not compatible with earlier versions, old clients will refuse to connect, so update the full farm - protocol reworked to no longer require a client->scheduler connection (the daemon will proxy it all) - all local jobs are now scheduled by the local daemon to make sure multiprocessor machines are handling as many jobs as they have jobs configured (# processors). They used to be "scheduled" by lock file - fixed grave bug that removed all daemon environments when the cache grew over limit - new telnet interface command: internals to output daemon states - stricter error handling to avoid suprises - SIGHUP to daemon and scheduler will reopen the log file to allow logrotate - a restart of the daemon and scheduler won't flush the log file, but append to it - create-env will create .tar.gz now - the client allows environments to be .tar, .tar.bz2 and .tar.gz