ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/Coro/Changes
(Generate patch)

Comparing Coro/Changes (file contents):
Revision 1.387 by root, Mon Nov 3 16:17:30 2008 UTC vs.
Revision 1.592 by root, Mon Jul 2 20:08:06 2012 UTC

1Revision history for Perl extension Coro. 1Revision history for Perl extension Coro.
2 2
3TODO: should explore PerlIO::coroaio (perl leaks like hell). 3TODO: should explore PerlIO::coroaio (perl leaks like hell).
4TODO: maybe implement a default message channel, very much like Erlang's 4TODO: channel->maxsize(newsize)?
5 actor model (which is cool in a lot of important aspects (failures!), 5TODO: http://www.microsoft.com/msj/archive/s2ce.aspx
6 but very lacking in others (higher level ipc)). 6TODO: __GCC_HAVE_DWARF2_CFI_ASM
7
8 - use fd = -1 in mmap.
9 - cast I32 to int in error message printf.
10
116.08 Fri Apr 13 12:05:47 CEST 2012
12 - be more aggressive about exiting like perl does - formerly,
13 exiting from the non-main thread would not execute END blocks.
14
156.07 Fri Nov 11 21:21:48 CET 2011
16 - work around a bug in PerlIO (setting $SIG{__WARN__} to a PVCV).
17 - update ecb.h.
18
196.06 Mon Aug 8 23:59:48 CEST 2011
20 - cygwin unfortunately patches the stack at runtime, so we use the pthreads
21 backend, which is an order of magnitude slower. unfortunately, cygwins
22 pthread implementation isn't very complete either, so allocate the stack
23 twice just to be sure.
24 (note: cygwin also enables mymalloc, which is NOT THREADSAFE ON WINDOWS,
25 in its ithreaded perl - best recompile cygwin and use the 'w'indows
26 backend for much better performance. also disable ithreads for
27 even better performance...).
28
296.05 Thu Aug 4 21:36:36 CEST 2011
30 - blush, condvar values would not be propagated from send to recv anymore
31 (reported by Chip Salzenberg).
32 - use exponential increase for the readline buffer length in
33 Coro::Handle. also reduce initial allocation to 1020 from 4096 bytes.
34
356.04 Wed Aug 3 17:15:45 CEST 2011
36 - use even more efficient and more compatible condvars for
37 compatibility to AnyEvent 6.x :)
38 - more inconsequential ecb.h updates.
39
406.03 Wed Aug 3 11:41:30 CEST 2011
41 - change how Coro patches AnyEvent condvars for compatibility to
42 AnyEvent 6.x.
43 - update ecb.h, to no longer include <pthread.h> in case WinNT.h
44 hasn't been included.
45
466.02 Wed Jul 13 04:35:19 CEST 2011
47 - "improve portability to Gentoo" - gentoo manages to put perl variables
48 in memory areas that are farther than 2gb apart, which the jit couldn't
49 handle and barfed. now it's just a bit slower on gentoo and similar
50 systems.
51
526.01 Sun Jul 3 12:31:14 CEST 2011
53 - workarounds are good, but the test for whether pthreads are used
54 was not good. thisone should be better.
55 - check differently whether gcc generates cfi instructions itself.
56
576.0 Wed Jun 29 19:43:35 CEST 2011
58 - INCOMPATIBLE CHANGE: unreferenced coro objects will now be
59 destroyed and cleaned up automatically (e.g. async { schedule }).
60 - implement a JIT compiler for part of the thread switch code,
61 which gives a 50% speed improvement on threaded perls, and
62 about 4% on non-thraeded perls (so threaded perls now finally
63 reach about half the speed of non-threaded perls).
64 - slightly modernise Coro::Intro, add section about rouse functions.
65 - avoid DEFSV and ERRSV, giving another 10% improvement
66 in thread switching.
67 - Coro::State->is_destroyed is now called is_zombie.
68 - implement a Coro->safe_cancel method that might fail, but
69 cancels in a "safer" way if it succeeds.
70 - add preliminary support for DEBUGGING perls.
71 - get rid of two hash-accesses when initialising a new Coro - this
72 speeds up coro creation by almost a factor of two.
73 - croak when a coro that is being cancelled tries to block
74 (e.g. while executing a guard block), instead of crashing or
75 deadlocking.
76 - use a more robust and also faster method to identify Coro::State
77 objects - speeds up everything a bit.
78 - implement Coro->cancel in XS for a 20% speed improvement, and to
79 be able to implement mutual cancellation.
80 - speed up context switches by a percent or two by more efficiently
81 allocating context stack entries.
82 - implement Coro->join and Coro->on_destroy in XS for a speedup and
83 a reduction in memory use.
84 - cancelling a coro while it itself is cancelling another coro is
85 now supported and working, instead of triggering an assertion.
86 - be a bit more crash-resistant when calling (buggy) on_destroy
87 callbacks (best effort).
88 - move on_destroy into the slf_frame, to allow extension slf
89 functions to have destructors.
90 - get rid if coro refcounting - simply crash in other interpreter
91 threads by nulling the pointers on clone.
92 - simplify warn/die hook handling when loading Coro - the convoluted
93 logic seems to be no longer neccessary.
94 - use libecb instead of our own home-grown gcc hacks.
95 - document alternatives to Coro::LWP. Please use them :)
96 - work around another mindless idiotic NEEDLESS bug in openbsd/mirbsds
97 sigaltstack. Really. wine suffers from it, erlang suffers from it,
98 and it's known since at least 2006.
99
1005.372 Wed Feb 23 06:14:30 CET 2011
101 - apparently mingw doesn't provide a working gettimeofday, try to
102 work around that by relying on Time::HiRes (indirectly brought to
103 my attention by Max Maischein).
104 - fix some portability issues when Time::HiRes was used.
105
1065.371 Mon Feb 21 14:36:08 CET 2011
107 - backport to windows process emulation code again.
108
1095.37 Sat Feb 19 07:49:44 CET 2011
110 - add a big "Coro thread life cycle" section to "man Coro".
111 - try a tentative workaround against the breakage that 5.13 has
112 introduced without depreciation period. sigh.
113 - no longer use Time::HiRes if gettimeofday is available, which
114 saves quite a lot of memory.
115
1165.36 Sun Feb 13 05:33:41 CET 2011
117 - automatically load Coro::Channel, Coro::RWLock, Coro::Semaphore,
118 Coro::SemaphoreSet, Coro::Signal and Coro::Specific on first "new"
119 method call.
120 - undocument Coro::Timer::sleep and obsolete whole module.
121 - optimise Coro::Timer::timeout memory and cpu usage.
122 - slightly updated Coro::Intro for recent changes.
123 - do not initialise PL_dirty anymore.
124
1255.25 Thu Nov 11 01:08:39 CET 2010
126 - try a different approach on netbsd - netbsd 5 finally has marginally
127 working pthreads, but still broken ucontext/sigaltstack.
128 - openbsd 4.8 finally got their act together, Coro works out of the box
129 with asm, setjmp and pthreads (no change, just informational).
130
1315.24 Sat Oct 23 11:27:12 CEST 2010
132 - port to the EV 4.0 API.
133 - work around bugs in mingw32, making strawberry perl work
134 out of the box.
135 - correctly modify Coro::AIO function prototypes
136 so that they reflect the "no optional parameters" rule.
137 - "ported" libcoro to C++.
138
1395.23 Mon May 17 18:50:42 CEST 2010
140 - be more resistant to ordering changes when initialising
141 Coro::AnyEvent, Coro::EV and Coro::Event (reported by Matthias
142 Waldorf).
143 - document that perl 5.12 deliberately removed support for cloning.
144
1455.22 Wed Apr 14 03:55:35 CEST 2010
146 - correctly return udnef on errors in Coro::Handle::read/write
147 (testcase by Marc Mims).
148 - convert Coro::Util into a "perl compatibility wrapper" - the functions
149 are less useful now, but are drop-in replacements for existing
150 functions, listing better alternatives in the documentation. This also
151 fixes a bug in Coro::LWP which naively substituted Socket::inet_aton
152 with Coro::Util::inet_aton.
153 - do not override $Coro::idle unconditionally in Coro.pm, as other
154 modules could have provided their own idle coro already
155 (for exmaple, Coro::AnyEvent).
156 - fix Coro::Util::gethost* functions.
157 - Coro::Timer corretcly exports it's symbols (reported by Hideki Yamamura).
158
1595.21 Wed Dec 16 07:19:51 CET 2009
160 - automatically load Coro::AnyEvent when AnyEvent and Coro are used
161 together.
162 - add some examples on how to combine other event loops with Coro in
163 Coro::AnyEvent, and how to run it (and not to block). Seems to be
164 the most common source of confusion.
165 - try to catch people naively blocking in an event callback.
166 - work around the perl filehandle bug issue in conjunction with
167 older common::sense (as indirectly pointed out by ZSystem).
168 - clarify the "not from signal handlers" section.
169
1705.2 Sun Oct 4 14:54:24 CEST 2009
171 - Coro::Storable destroyed the prototypes of the functions it wrapped.
172 - export rouse_cb and rouse_wait by default now.
173 - fix various prototype mismatches in Coro::AnyEvent and Coro::Handle.
174 - new method $state->swap_sv.
175 - added section on "windows process emulation" to the manpage,
176 after a not-so-fruitful (nor-friendly) "discussion" with chip
177 salzenberg (discussion implies arguments, but his only arguments
178 were ad-hominems, one wonders why he started it in the first
179 place). I hope this explains it well enough for him to understand,
180 and maybe well enough for others to understand.
181 - use common::sense everywhere now.
182 - idle callbacks are no longer supported, use idle coros instead.
183 - print a thread listing when a deadlock is detected.
184
1855.17 Sat Aug 22 23:09:31 CEST 2009
186 - work around a bug in the perl debugger causing crashes
187 when running under the debugger by marking _pool_handler as nodebug.
188 - speed up Coro::async considerably.
189 - try some hacks to get netbsd to work "more often" - their broken
190 setjmp/longjmp, ucontext *and* phtreads are really hard on Coro.
191 - convert Coro to AE 5.0 API.
192
1935.162 Tue Jul 28 04:04:03 CEST 2009
194 - perl 5.8.2 is now minimum requirement.
195 - skip t/19_handle.t on broken windows perls.
196
1975.161 Wed Jul 22 04:47:38 CEST 2009
198 - Coro::AnyEvent::poll could have a different prototype when EV was
199 used as backend (analyzed by Tatsuhiko Miyagawa).
200 - Coro::AnyEvent errornously initialised the event loop when loaded,
201 not on demand.
202 - try to workaround rare */t/01_unblock.t failures.
203
2045.16 Tue Jul 21 01:44:37 CEST 2009
205 - Coro::AnyEvent failed to hook into the event loop
206 when no threads had been readied between detecting
207 the event loop and actually running it.
208 - considerably speed up Coro::Select by taking avdantage
209 of AnyEvent > 4.8 and some other optimisations.
210 - implement paragraph readline mode in Coro::Handle
211 (based on patches by Zsbán Ambrus).
212 - replace WSAEINPROGRESS by WSAEWOULDBLOCK (reported
213 and analyzed by Yasuhiro MATSUMOTO).
214 - clarified libcoro license and copyright.
215 - someone stole my EXTRA_META!!!
216 - implement Coro::Select::patch_pp_sselect and it's brother,
217 for hardcode select overriding.
218
2195.151 Mon Jul 6 05:41:57 CEST 2009
220 - backport to windows process emulation code again (patch by
221 Yasuhiro MATSUMOTO).
222 - slightly update Coro::MakeMaker.
223
2245.15 Tue Jun 30 10:28:06 CEST 2009
225 - deprecate Coro::Socket, document how to get ipv6 support via
226 AnyEvent::Socket instead.
227 - implement signal->wait ($cb) interface, similar to semaphores.
228 - work around SvOK not supporting getmagic, so we have to getmagic
229 to test for undef :/ (reported by Matthias Waldorf).
230 - load Coro::AnyEvent in all modules using AnyEvent.
231 - work around perl corrupting our internal data structures,
232 reported by Tokuhiro Matsuno.
233 - enable per-coroutine real and cpu time gathering
234 (Coro::State::enable_times).
235
2365.14 Wed Jun 24 01:37:48 CEST 2009
237 - provide explicit functions to "cede" to the event loop in Coro::AnyEvent,
238 as this seems to have been a difficult concept (poll, sleep, idle,
239 idle_upto).
240 - add Coro::AnyEvent::readable/writable functions.
241 - clarify Coro::EV/Event/AnyEvent manpages.
242 - free per-thread global scalars in the thread calling ->cancel, to
243 avoid crashes when $_, $@ etc., are magical but some of those
244 had already been freed.
245 - "unexperimentalise" the callback interface for Coro::Semaphore.
246 - speed up ready queue management/context switching by using a linked
247 list instead of an array (~5-10%).
248 - implement "watch" command in Coro::Debug shells.
249 - for fun, implement time-slicing as an example in the manpage.
250 - if AnyEvent detects EV or Event, but we don't have Coro::EV or
251 Coro::Event, use the normal AnyEvent handling instead of dieing
252 (the same is true for Coro::Handle).
253 - properly document Coro::EV::timed_io_once.
254 - avoid unneccessary ->cancel calls in Coro::Handle.
255 - maybe make it work on mingw32 with win32 backend
256 (based on patch by Yasuhiro Matsumoto).
257
2585.132 Fri May 29 09:00:39 CEST 2009
259 - do not keep a reference to the argument itself in
260 Coro::Semaphore::guard, as it could change later.
261 - support SO_RCVBUF/SO_SNDBUF nonstandard Coro::Socket options,
262 should support a prepare callback.
263
2645.131 Mon Mar 16 23:20:37 CET 2009
265 - implement and document Coro->suspend, Coro->resume.
266 - fix Coro::Select implementation to not (often) close
267 the passed file descriptors (testcase provided by pippijn).
268
2695.13 Mon Dec 15 21:51:42 CET 2008
270 - EXPERIMENTAL: implement dynamic winds (on_enter/on_leave).
271 - don't set diehook to C<undef> but instead to NULL, to avoid
272 spurious warnings.
273 - fix a lot of bugs in Coro::SemaphoreSet.
274 - Coro::SemaphoreSet will less often create a semaphore needlessly.
275 - add Coro::SemaphoreSet::count and wait methods.
276 - take advantage of the new Guard module.
277 - deprecate Coro::guard.
278 - try to fix the dreaded 01_unblock tests once more. I hate it when
279 testsuites need more fixing than the code they are supposed to test.
280 - croak in more cases when a required callback isn't resolvable.
281 - fix some minor issues in Coro::State->call/eval.
282 - use current coroutine context instead of a temporary one
283 when temporarily switching to another coroutine.
284 - do not call C-level on_destroy handlers during global destruction,
285 to avoid needless segfaults.
286
2875.12 Sun Dec 7 13:30:38 CET 2008
288 - add default config for MirOS, which seems to be a bug-to-bug
289 compatible fork of openbsd ("world domination by releasing
290 openbsd cvs before the openbsd folks do it" or so :).
291 - free_padlist did destroy the names pad, not good, but didn't
292 seem to bother perl - this could fix issues such as eval ""
293 inside a function called from multiple coroutines.
294 - use a different method to detect destruction time.
295 - be more careful when freeing padlists just before global
296 destruction.
297 - fixed and expanded the call/cc example.
298 - renamed _terminate to _coro_run.
299 - new method Coro::Channel->shutdown.
300 - try pthreads on openbsd <4.4 (broken sigaltstack, will
301 pthreads fare better?).
302 - be less picky about destroying "a" running coroutine.
303
3045.11 Tue Nov 25 21:49:05 CET 2008
305 - DEBUGGING in 5.10.0 is a mess: it constantly flags perfectly
306 working code with failed assertions, introducing more bugs than
307 it fixes, requiring elaborate workarounds :(
308
3095.1 Mon Nov 24 08:54:59 CET 2008
310 - wrote a small introductory tutorial - Coro::Intro.
311 - convert Coro::Timer, Coro::Select and Coro::Util to rouse API.
312 - Coro::Select did errornously dup the file descriptors
313 and didn't work with all AnyEvent backends.
314 - Coro::Select wasn't imported correctly form Coro::LWP, causing blocking
315 LWP data transfers.
316 - disassociate c contexts from coro objects - this is agruably more
317 correct, but mostly allows sharing of cctxs between coro and state
318 objects, for added memory savings and speed increases.
319 - bumped $Coro::POOL_RSS up to 32kb by default.
320 - no longer set the optype to OP_CUSTOM, as B::* understandably
321 doesn't like this very much (and we *are* a type of entersub).
322 - implement state cloning, just to prove that call/cc can be done.
323 - automatically load Coro::AnyEvent in Coro::Handle.
324 - wrap ->cancel calls in eval inside Coro::Handle as EV watchers
325 do not have this method (and don't need it either).
326 - speed up generic anyevent methods in Coro::Handle by using rouse
327 callbacks.
328 - allow coroutines in $Coro::IDLE, speeding up Coro::AnyEvent and
329 others. It also makes the debugger happier, as you can trace
330 through the idle threads now.
331 - add comppad_name* and hints ($^H, %^H) to per-thread variables.
332 - eg/event was pretty broken.
333 - better 5.8.6 compatibility.
334
3355.0 Thu Nov 20 10:35:05 CET 2008
336 - NEW ARCHITECTURE: use the latest 4.x version if you experience
337 stability issues.
338 - bump API version to 7 - all dependents must be recompiled.
339 - removed timed_* functions - they were not being used anyways
340 and should be replaced by a more generic mechanism -
341 and were annoying to support anyways :)
342 - removed SemaphoreSet's waiter method - use sem method instead.
343 - Coro::Semaphore->adjust didn't correctly wake up enough waiters.
344 - async_pool did free a scalar value twice
345 ("Attempt to unreference...").
346 - fix a longstanding bug where calling terminate on a coro that
347 was waiting for a semaphore that was just becoming available
348 would cause a deadlock (semaphore would get into a state where
349 it was available but waiters were still blocked).
350 - calling throw on a coroutine that is waiting for a semaphore will
351 no longer make it acquire the semaphore (and thus leak a count).
352 - perl's process emulation is now not even theoretically supported
353 anymore.
354 - new functions Coro::rouse_cb and Coro::rouse_wait for easier
355 conversion of callback-style to blocking-style.
356 - new methods $coro->schedule_to and ->cede_to, to specifically
357 schedule or cede to a specific coroutine.
358 - new function Coro::Semaphore::wait.
359 - use named constants in Coro::Channel (Richard Hundt).
360 - directly patch the entersub opcode calling SLF functions (cede,
361 transfer and so on). this does speed up context switching, but
362 more importanly, it frees us from the hardcoded behaviour of
363 entersub, so we might actually be able to return something from
364 those functions and atcually create new ones.
365 - take advantage of __builtin_frame_address on gcc.
366 - expose THX in coroapi (not sure whether this was a wise decision,
367 as "threaded" perls are running at half speed anyways).
368 - implement execute_slf (schedule-like-function) interface that makes
369 it possible to implement schedule-like-functions in XS.
370 - use new SLF interface to massively speed up Coro::EV by roughly a
371 factor of two.
372 - used new SLF interface to massively speed up Coro::Semaphore by a
373 factor of three.
374 - used new SLF interface to speed up Coro::AIO by roughly a factor of
375 four and reduce its memory usage considerably.
376 - implement Coro::SemaphoreSet purely in terms of Coro::Semaphore,
377 for a nice speedup and vastly more correct behaviour. Also implement
378 a new method "sem" to get at the underlying semaphore object.
379 - implement Coro::Channel in terms of Coro::Semaphore, for a moderate
380 (in comparison) 20-40% speedup.
381 - used new SLF interface to reimplement Coro::Signal gaining
382 some unknown (because I was too lazy), but certain, speedup, and also
383 making signals reliable for the first time.
384 - used new SLF interface and other optimisations to speed up async_pool
385 by a factor of two. It also doesn't rely on perl's exception mechanism
386 to exit anymore. The overhead for terminating an async_pool, coro over
387 a normal async is now very small.
388 - sped up coroutine creation/destruction by 40%.
389 - forgot to include Coro/libcoro/README in the dist for all these years.
390 - work around a freebsd pthreads bug (manual testcancel is required as
391 pthread_cond_wait isn't a cancellation point on freebsd).
392 - use new rouse functions to speed up and simplify Coro::BDB.
393 - make "prefer perl native functions" work with threaded perls.
394 - condense Coro::Debug ps output, hint at v and w flags.
395 - (libcoro) lots of minor cleanups and portability improvements.
396
3974.914 Wed Nov 19 12:54:18 CET 2008
398 - fix a disastrous bug in the readline optimisation
399 introduced in 4.801.
400
4014.913 Sat Nov 15 07:58:28 CET 2008
402 - async_pool did free a scalar value twice
403 ("Attempt to unreference...").
404
4054.912 Thu Nov 13 18:31:23 CET 2008
406 - minor cleanups.
407 - use much larger stacks on linux and perl < 5.8.8.
408 - Coro::Debug::new_unix_server did not unlink the socket
409 when destroyed.
410
4114.911 Tue Nov 11 04:26:17 CET 2008
412 - "port" to threaded perls.
413
4144.91 Mon Nov 10 05:36:38 CET 2008
415 - the ->throw exception object no longer leaks.
416 - creating a new cctx leaked a scopestack entry (memleak).
417 - new coroutines didn't get created with a zero flags field
418 (unknown impact).
419 - calling ->throw on a not-yet-started coroutine should now work
420 instead of being ignored.
421 - ->throw is now supported on Coro::State objects.
422 - clean up cctx creation code a bit.
423 - entersub is actually a UNOP, not a LOGOP (not a bugfix).
424
4254.9 Sat Nov 8 17:45:27 CET 2008
426 - (libcoro) did not preserve rbp with CORO_ASM (we are getting there).
427 - (libcoro) no longer leak threads in the experimental pthread backend,
428 also speed it up considerably.
429 - (libcoro) do not rely on makecontext passing void *'s unscathed.
430 - fix compiletime dependencies on libcoro in the Makefile.
431 - cctx_count wasn't always updated properly.
432 - Coro::State::cctx_stacksize wasn't applied correctly.
433 - new function Coro::State::cctx_max_idle.
434 - the default max number of idle C contexts is now 4.
435 - (libcoro) try harder to get _setjmp/_longjmp.
436 - (libcoro) cleanup and extend the libcoro API to officially
437 allow the creation of empty source contexts.
438 - very experimental workaround for the totally broken netbsd platform.
439 - tried to hack around openbsd bugs.
440
4414.804 Wed Nov 5 16:36:00 CET 2008
442 - Coro::Debug::new_unix_server would not create a non-blocking listening
443 socket, sometimes causing freezes.
444 - (libcoro) fix misaligned stack points for setjmp and assembly
445 methods, which can cause crashes on x86/x86_64 with a sufficiently
446 aggressive compiler.
447 - new function: Coro::Debug::new_tcp_server.
448 - move ->throw into the Coro class because it only works on coro objects.
7 449
84.803 Mon Nov 3 17:16:12 CET 2008 4504.803 Mon Nov 3 17:16:12 CET 2008
9 - use a global asm statement to become independent of gcc otpimisations 451 - (libcoro) use a global asm statement to become independent of gcc
10 for CORO_ASM (thanks to pippijn for the idea). 452 otpimisations for CORO_ASM (thanks to pippijn for the idea).
11 - try to workaround yet another broken bsd, this time dragonfly. 453 - try to workaround yet another broken bsd, this time dragonfly.
12 454
134.802 Thu Oct 30 10:56:12 CET 2008 4554.802 Thu Oct 30 10:56:12 CET 2008
14 - support -fno-omit-frame-pointer on x86 with the assembly method. 456 - support -fno-omit-frame-pointer on x86 with the assembly method.
15 - tune 01_unblock.t tests a bit. 457 - tune 01_unblock.t tests a bit.

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines