--- Coro/Changes 2008/11/19 05:52:42 1.437 +++ Coro/Changes 2013/05/08 00:44:44 1.615 @@ -1,22 +1,393 @@ Revision history for Perl extension Coro. TODO: should explore PerlIO::coroaio (perl leaks like hell). -TODO: maybe implement a default message channel, very much like Erlang's - actor model (which is cool in a lot of important aspects (failures!), - but very lacking in others (higher level ipc)). - -TODO: testcancel -TODO: guard as SLF - to avoid deadlocks. - - avoid deadlock in Coro::Channel if maxsize == 1 (Richard Hundt). +TODO: channel->maxsize(newsize)? +TODO: __GCC_HAVE_DWARF2_CFI_ASM -5.0 +6.29 + - when an on_destroy handler destructs the coro currently being + destructed a perl scalar could be accessed after being freed, + likely causing a crash. + +6.28 Wed Mar 6 06:58:02 CET 2013 + - clean remnants of existing __DIE__ and __WARN__ handlers so + they lose their magic and will not cause segfaults later + (testcase by Andrey Sagulin). + - improved Coro::State documentation a bit. + - Coro::Debug::command now flushes the output. + - add hack detection code for x32 abi, because the braindead slugs + who designed that made it look exactly like x86_64 without + providing proper compile time symbols to test for it. as a result, + this detection cannot work reliably. + - valgrind stack registering was broken. + - do not rely on Time::HiRes anymore in Coro::Debug. + +6.23 Fri Dec 7 23:36:37 CET 2012 + - use experimental fiber implementation on native windows + perls. + - use sizeof (void *) as multiplication factor for stack sizes, + to accomodate the totally braindamaged microsoft 64 bit "os". + - changed verifier host from win2k-ap510-32 to win7-sp516-32/64. + activeperl 5.16 crashes when PerlIO_define_layer is called due + to some bug in the perl dll, strawberry perl at least passes + the testsuite. + - implement Coro::Handle->peeraddr/host/port, for slightly + improved compatibility with LWP. + - implement 5.17 compatibility by almost blindly applying a + good-looking patch by Father Chrysostomos. + - move stack management functions into libcoro 3. + - libcoro version 3 "released". + - support magic values as timed_io_once args. + - recommend AnyEvent 7+ or EV 4+, also require EV + version 4 or newer for Coro::EV. + +6.10 Tue Oct 9 01:14:27 CEST 2012 + - updated ecb.h, it had a typo that caused it to not compile on many + big endian systems (reported by many people). + - disable memory fences in ecb.h to improve portability. + +6.09 Sat Oct 6 23:25:02 CEST 2012 + - Coro::EV I/O watchers were not interruptible by exceptions + (Coro::State::throw) (testcase by sten). + - ->throw now puts threads into the ready queue, as this seems to + be expected by existing code, and code that doesn't cope with spurious + wakeups needs fixing anyway. + - use fd -1 in mmap. + - cast I32 to int in error message printf. + - warn about broken so-called "hardened" kernels. + +6.08 Fri Apr 13 12:05:47 CEST 2012 + - be more aggressive about exiting like perl does - formerly, + exiting from the non-main thread would not execute END blocks. + +6.07 Fri Nov 11 21:21:48 CET 2011 + - work around a bug in PerlIO (setting $SIG{__WARN__} to a PVCV). + - update ecb.h. + +6.06 Mon Aug 8 23:59:48 CEST 2011 + - cygwin unfortunately patches the stack at runtime, so we use the pthreads + backend, which is an order of magnitude slower. unfortunately, cygwins + pthread implementation isn't very complete either, so allocate the stack + twice just to be sure. + (note: cygwin also enables mymalloc, which is NOT THREADSAFE ON WINDOWS, + in its ithreaded perl - best recompile cygwin and use the 'w'indows + backend for much better performance. also disable ithreads for + even better performance...). + +6.05 Thu Aug 4 21:36:36 CEST 2011 + - blush, condvar values would not be propagated from send to recv anymore + (reported by Chip Salzenberg). + - use exponential increase for the readline buffer length in + Coro::Handle. also reduce initial allocation to 1020 from 4096 bytes. + +6.04 Wed Aug 3 17:15:45 CEST 2011 + - use even more efficient and more compatible condvars for + compatibility to AnyEvent 6.x :) + - more inconsequential ecb.h updates. + +6.03 Wed Aug 3 11:41:30 CEST 2011 + - change how Coro patches AnyEvent condvars for compatibility to + AnyEvent 6.x. + - update ecb.h, to no longer include in case WinNT.h + hasn't been included. + +6.02 Wed Jul 13 04:35:19 CEST 2011 + - "improve portability to Gentoo" - gentoo manages to put perl variables + in memory areas that are farther than 2gb apart, which the jit couldn't + handle and barfed. now it's just a bit slower on gentoo and similar + systems. + +6.01 Sun Jul 3 12:31:14 CEST 2011 + - workarounds are good, but the test for whether pthreads are used + was not good. thisone should be better. + - check differently whether gcc generates cfi instructions itself. + +6.0 Wed Jun 29 19:43:35 CEST 2011 + - INCOMPATIBLE CHANGE: unreferenced coro objects will now be + destroyed and cleaned up automatically (e.g. async { schedule }). + - implement a JIT compiler for part of the thread switch code, + which gives a 50% speed improvement on threaded perls, and + about 4% on non-thraeded perls (so threaded perls now finally + reach about half the speed of non-threaded perls). + - slightly modernise Coro::Intro, add section about rouse functions. + - avoid DEFSV and ERRSV, giving another 10% improvement + in thread switching. + - Coro::State->is_destroyed is now called is_zombie. + - implement a Coro->safe_cancel method that might fail, but + cancels in a "safer" way if it succeeds. + - add preliminary support for DEBUGGING perls. + - get rid of two hash-accesses when initialising a new Coro - this + speeds up coro creation by almost a factor of two. + - croak when a coro that is being cancelled tries to block + (e.g. while executing a guard block), instead of crashing or + deadlocking. + - use a more robust and also faster method to identify Coro::State + objects - speeds up everything a bit. + - implement Coro->cancel in XS for a 20% speed improvement, and to + be able to implement mutual cancellation. + - speed up context switches by a percent or two by more efficiently + allocating context stack entries. + - implement Coro->join and Coro->on_destroy in XS for a speedup and + a reduction in memory use. + - cancelling a coro while it itself is cancelling another coro is + now supported and working, instead of triggering an assertion. + - be a bit more crash-resistant when calling (buggy) on_destroy + callbacks (best effort). + - move on_destroy into the slf_frame, to allow extension slf + functions to have destructors. + - get rid if coro refcounting - simply crash in other interpreter + threads by nulling the pointers on clone. + - simplify warn/die hook handling when loading Coro - the convoluted + logic seems to be no longer neccessary. + - use libecb instead of our own home-grown gcc hacks. + - document alternatives to Coro::LWP. Please use them :) + - work around another mindless idiotic NEEDLESS bug in openbsd/mirbsds + sigaltstack. Really. wine suffers from it, erlang suffers from it, + and it's known since at least 2006. + +5.372 Wed Feb 23 06:14:30 CET 2011 + - apparently mingw doesn't provide a working gettimeofday, try to + work around that by relying on Time::HiRes (indirectly brought to + my attention by Max Maischein). + - fix some portability issues when Time::HiRes was used. + +5.371 Mon Feb 21 14:36:08 CET 2011 + - backport to windows process emulation code again. + +5.37 Sat Feb 19 07:49:44 CET 2011 + - add a big "Coro thread life cycle" section to "man Coro". + - try a tentative workaround against the breakage that 5.13 has + introduced without depreciation period. sigh. + - no longer use Time::HiRes if gettimeofday is available, which + saves quite a lot of memory. + +5.36 Sun Feb 13 05:33:41 CET 2011 + - automatically load Coro::Channel, Coro::RWLock, Coro::Semaphore, + Coro::SemaphoreSet, Coro::Signal and Coro::Specific on first "new" + method call. + - undocument Coro::Timer::sleep and obsolete whole module. + - optimise Coro::Timer::timeout memory and cpu usage. + - slightly updated Coro::Intro for recent changes. + - do not initialise PL_dirty anymore. + +5.25 Thu Nov 11 01:08:39 CET 2010 + - try a different approach on netbsd - netbsd 5 finally has marginally + working pthreads, but still broken ucontext/sigaltstack. + - openbsd 4.8 finally got their act together, Coro works out of the box + with asm, setjmp and pthreads (no change, just informational). + +5.24 Sat Oct 23 11:27:12 CEST 2010 + - port to the EV 4.0 API. + - work around bugs in mingw32, making strawberry perl work + out of the box. + - correctly modify Coro::AIO function prototypes + so that they reflect the "no optional parameters" rule. + - "ported" libcoro to C++. + +5.23 Mon May 17 18:50:42 CEST 2010 + - be more resistant to ordering changes when initialising + Coro::AnyEvent, Coro::EV and Coro::Event (reported by Matthias + Waldorf). + - document that perl 5.12 deliberately removed support for cloning. + +5.22 Wed Apr 14 03:55:35 CEST 2010 + - correctly return udnef on errors in Coro::Handle::read/write + (testcase by Marc Mims). + - convert Coro::Util into a "perl compatibility wrapper" - the functions + are less useful now, but are drop-in replacements for existing + functions, listing better alternatives in the documentation. This also + fixes a bug in Coro::LWP which naively substituted Socket::inet_aton + with Coro::Util::inet_aton. + - do not override $Coro::idle unconditionally in Coro.pm, as other + modules could have provided their own idle coro already + (for exmaple, Coro::AnyEvent). + - fix Coro::Util::gethost* functions. + - Coro::Timer corretcly exports it's symbols (reported by Hideki Yamamura). + +5.21 Wed Dec 16 07:19:51 CET 2009 + - automatically load Coro::AnyEvent when AnyEvent and Coro are used + together. + - add some examples on how to combine other event loops with Coro in + Coro::AnyEvent, and how to run it (and not to block). Seems to be + the most common source of confusion. + - try to catch people naively blocking in an event callback. + - work around the perl filehandle bug issue in conjunction with + older common::sense (as indirectly pointed out by ZSystem). + - clarify the "not from signal handlers" section. + +5.2 Sun Oct 4 14:54:24 CEST 2009 + - Coro::Storable destroyed the prototypes of the functions it wrapped. + - export rouse_cb and rouse_wait by default now. + - fix various prototype mismatches in Coro::AnyEvent and Coro::Handle. + - new method $state->swap_sv. + - added section on "windows process emulation" to the manpage, + after a not-so-fruitful (nor-friendly) "discussion" with chip + salzenberg (discussion implies arguments, but his only arguments + were ad-hominems, one wonders why he started it in the first + place). I hope this explains it well enough for him to understand, + and maybe well enough for others to understand. + - use common::sense everywhere now. + - idle callbacks are no longer supported, use idle coros instead. + - print a thread listing when a deadlock is detected. + +5.17 Sat Aug 22 23:09:31 CEST 2009 + - work around a bug in the perl debugger causing crashes + when running under the debugger by marking _pool_handler as nodebug. + - speed up Coro::async considerably. + - try some hacks to get netbsd to work "more often" - their broken + setjmp/longjmp, ucontext *and* phtreads are really hard on Coro. + - convert Coro to AE 5.0 API. + +5.162 Tue Jul 28 04:04:03 CEST 2009 + - perl 5.8.2 is now minimum requirement. + - skip t/19_handle.t on broken windows perls. + +5.161 Wed Jul 22 04:47:38 CEST 2009 + - Coro::AnyEvent::poll could have a different prototype when EV was + used as backend (analyzed by Tatsuhiko Miyagawa). + - Coro::AnyEvent errornously initialised the event loop when loaded, + not on demand. + - try to workaround rare */t/01_unblock.t failures. + +5.16 Tue Jul 21 01:44:37 CEST 2009 + - Coro::AnyEvent failed to hook into the event loop + when no threads had been readied between detecting + the event loop and actually running it. + - considerably speed up Coro::Select by taking avdantage + of AnyEvent > 4.8 and some other optimisations. + - implement paragraph readline mode in Coro::Handle + (based on patches by Zsbán Ambrus). + - replace WSAEINPROGRESS by WSAEWOULDBLOCK (reported + and analyzed by Yasuhiro MATSUMOTO). + - clarified libcoro license and copyright. + - someone stole my EXTRA_META!!! + - implement Coro::Select::patch_pp_sselect and it's brother, + for hardcode select overriding. + +5.151 Mon Jul 6 05:41:57 CEST 2009 + - backport to windows process emulation code again (patch by + Yasuhiro MATSUMOTO). + - slightly update Coro::MakeMaker. + +5.15 Tue Jun 30 10:28:06 CEST 2009 + - deprecate Coro::Socket, document how to get ipv6 support via + AnyEvent::Socket instead. + - implement signal->wait ($cb) interface, similar to semaphores. + - work around SvOK not supporting getmagic, so we have to getmagic + to test for undef :/ (reported by Matthias Waldorf). + - load Coro::AnyEvent in all modules using AnyEvent. + - work around perl corrupting our internal data structures, + reported by Tokuhiro Matsuno. + - enable per-coroutine real and cpu time gathering + (Coro::State::enable_times). + +5.14 Wed Jun 24 01:37:48 CEST 2009 + - provide explicit functions to "cede" to the event loop in Coro::AnyEvent, + as this seems to have been a difficult concept (poll, sleep, idle, + idle_upto). + - add Coro::AnyEvent::readable/writable functions. + - clarify Coro::EV/Event/AnyEvent manpages. + - free per-thread global scalars in the thread calling ->cancel, to + avoid crashes when $_, $@ etc., are magical but some of those + had already been freed. + - "unexperimentalise" the callback interface for Coro::Semaphore. + - speed up ready queue management/context switching by using a linked + list instead of an array (~5-10%). + - implement "watch" command in Coro::Debug shells. + - for fun, implement time-slicing as an example in the manpage. + - if AnyEvent detects EV or Event, but we don't have Coro::EV or + Coro::Event, use the normal AnyEvent handling instead of dieing + (the same is true for Coro::Handle). + - properly document Coro::EV::timed_io_once. + - avoid unneccessary ->cancel calls in Coro::Handle. + - maybe make it work on mingw32 with win32 backend + (based on patch by Yasuhiro Matsumoto). + +5.132 Fri May 29 09:00:39 CEST 2009 + - do not keep a reference to the argument itself in + Coro::Semaphore::guard, as it could change later. + - support SO_RCVBUF/SO_SNDBUF nonstandard Coro::Socket options, + should support a prepare callback. + +5.131 Mon Mar 16 23:20:37 CET 2009 + - implement and document Coro->suspend, Coro->resume. + - fix Coro::Select implementation to not (often) close + the passed file descriptors (testcase provided by pippijn). + +5.13 Mon Dec 15 21:51:42 CET 2008 + - EXPERIMENTAL: implement dynamic winds (on_enter/on_leave). + - don't set diehook to C but instead to NULL, to avoid + spurious warnings. + - fix a lot of bugs in Coro::SemaphoreSet. + - Coro::SemaphoreSet will less often create a semaphore needlessly. + - add Coro::SemaphoreSet::count and wait methods. + - take advantage of the new Guard module. + - deprecate Coro::guard. + - try to fix the dreaded 01_unblock tests once more. I hate it when + testsuites need more fixing than the code they are supposed to test. + - croak in more cases when a required callback isn't resolvable. + - fix some minor issues in Coro::State->call/eval. + - use current coroutine context instead of a temporary one + when temporarily switching to another coroutine. + - do not call C-level on_destroy handlers during global destruction, + to avoid needless segfaults. + +5.12 Sun Dec 7 13:30:38 CET 2008 + - add default config for MirOS, which seems to be a bug-to-bug + compatible fork of openbsd ("world domination by releasing + openbsd cvs before the openbsd folks do it" or so :). + - free_padlist did destroy the names pad, not good, but didn't + seem to bother perl - this could fix issues such as eval "" + inside a function called from multiple coroutines. + - use a different method to detect destruction time. + - be more careful when freeing padlists just before global + destruction. + - fixed and expanded the call/cc example. + - renamed _terminate to _coro_run. + - new method Coro::Channel->shutdown. + - try pthreads on openbsd <4.4 (broken sigaltstack, will + pthreads fare better?). + - be less picky about destroying "a" running coroutine. + +5.11 Tue Nov 25 21:49:05 CET 2008 + - DEBUGGING in 5.10.0 is a mess: it constantly flags perfectly + working code with failed assertions, introducing more bugs than + it fixes, requiring elaborate workarounds :( + +5.1 Mon Nov 24 08:54:59 CET 2008 + - wrote a small introductory tutorial - Coro::Intro. + - convert Coro::Timer, Coro::Select and Coro::Util to rouse API. + - Coro::Select did errornously dup the file descriptors + and didn't work with all AnyEvent backends. + - Coro::Select wasn't imported correctly form Coro::LWP, causing blocking + LWP data transfers. + - disassociate c contexts from coro objects - this is agruably more + correct, but mostly allows sharing of cctxs between coro and state + objects, for added memory savings and speed increases. + - bumped $Coro::POOL_RSS up to 32kb by default. + - no longer set the optype to OP_CUSTOM, as B::* understandably + doesn't like this very much (and we *are* a type of entersub). + - implement state cloning, just to prove that call/cc can be done. + - automatically load Coro::AnyEvent in Coro::Handle. + - wrap ->cancel calls in eval inside Coro::Handle as EV watchers + do not have this method (and don't need it either). + - speed up generic anyevent methods in Coro::Handle by using rouse + callbacks. + - allow coroutines in $Coro::IDLE, speeding up Coro::AnyEvent and + others. It also makes the debugger happier, as you can trace + through the idle threads now. + - add comppad_name* and hints ($^H, %^H) to per-thread variables. + - eg/event was pretty broken. + - better 5.8.6 compatibility. + +5.0 Thu Nov 20 10:35:05 CET 2008 - NEW ARCHITECTURE: use the latest 4.x version if you experience stability issues. - bump API version to 7 - all dependents must be recompiled. - removed timed_* functions - they were not being used anyways and should be replaced by a more generic mechanism - and were annoying to support anyways :) - - removed SemaphoreSet's waiter method - use sem instead. + - removed SemaphoreSet's waiter method - use sem method instead. - Coro::Semaphore->adjust didn't correctly wake up enough waiters. - async_pool did free a scalar value twice ("Attempt to unreference..."). @@ -28,7 +399,10 @@ no longer make it acquire the semaphore (and thus leak a count). - perl's process emulation is now not even theoretically supported anymore. - - new functions Coro::rouse_cb and Coro::rouse_wait. + - new functions Coro::rouse_cb and Coro::rouse_wait for easier + conversion of callback-style to blocking-style. + - new methods $coro->schedule_to and ->cede_to, to specifically + schedule or cede to a specific coroutine. - new function Coro::Semaphore::wait. - use named constants in Coro::Channel (Richard Hundt). - directly patch the entersub opcode calling SLF functions (cede, @@ -55,11 +429,23 @@ - used new SLF interface to reimplement Coro::Signal gaining some unknown (because I was too lazy), but certain, speedup, and also making signals reliable for the first time. + - used new SLF interface and other optimisations to speed up async_pool + by a factor of two. It also doesn't rely on perl's exception mechanism + to exit anymore. The overhead for terminating an async_pool, coro over + a normal async is now very small. + - sped up coroutine creation/destruction by 40%. - forgot to include Coro/libcoro/README in the dist for all these years. - work around a freebsd pthreads bug (manual testcancel is required as pthread_cond_wait isn't a cancellation point on freebsd). + - use new rouse functions to speed up and simplify Coro::BDB. + - make "prefer perl native functions" work with threaded perls. + - condense Coro::Debug ps output, hint at v and w flags. - (libcoro) lots of minor cleanups and portability improvements. +4.914 Wed Nov 19 12:54:18 CET 2008 + - fix a disastrous bug in the readline optimisation + introduced in 4.801. + 4.913 Sat Nov 15 07:58:28 CET 2008 - async_pool did free a scalar value twice ("Attempt to unreference...").