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

Comparing Coro/Changes (file contents):
Revision 1.408 by root, Fri Nov 14 02:29:09 2008 UTC vs.
Revision 1.601 by root, Wed Oct 31 20:15:40 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 7
85.0 8TODO: add peerhost/peerport from IO::Socket::INET etc.
9 - ARCHITECTURAL REDESIGN, use 4.912 for the stable version. 9
10 - support magic values as timed_io_once args.
11 - recommend AnyEvent 7+ or EV 4+, also require EV
12 version 4 or newer for Coro::EV.
13
146.10 Tue Oct 9 01:14:27 CEST 2012
15 - updated ecb.h, it had a typo that caused it to not compile on many
16 big endian systems (reported by many people).
17 - disable memory fences in ecb.h to improve portability.
18
196.09 Sat Oct 6 23:25:02 CEST 2012
20 - Coro::EV I/O watchers were not interruptible by exceptions
21 (Coro::State::throw) (testcase by sten).
22 - ->throw now puts threads into the ready queue, as this seems to
23 be expected by existing code, and code that doesn't cope with spurious
24 wakeups needs fixing anyway.
25 - use fd -1 in mmap.
26 - cast I32 to int in error message printf.
27 - warn about broken so-called "hardened" kernels.
28
296.08 Fri Apr 13 12:05:47 CEST 2012
30 - be more aggressive about exiting like perl does - formerly,
31 exiting from the non-main thread would not execute END blocks.
32
336.07 Fri Nov 11 21:21:48 CET 2011
34 - work around a bug in PerlIO (setting $SIG{__WARN__} to a PVCV).
35 - update ecb.h.
36
376.06 Mon Aug 8 23:59:48 CEST 2011
38 - cygwin unfortunately patches the stack at runtime, so we use the pthreads
39 backend, which is an order of magnitude slower. unfortunately, cygwins
40 pthread implementation isn't very complete either, so allocate the stack
41 twice just to be sure.
42 (note: cygwin also enables mymalloc, which is NOT THREADSAFE ON WINDOWS,
43 in its ithreaded perl - best recompile cygwin and use the 'w'indows
44 backend for much better performance. also disable ithreads for
45 even better performance...).
46
476.05 Thu Aug 4 21:36:36 CEST 2011
48 - blush, condvar values would not be propagated from send to recv anymore
49 (reported by Chip Salzenberg).
50 - use exponential increase for the readline buffer length in
51 Coro::Handle. also reduce initial allocation to 1020 from 4096 bytes.
52
536.04 Wed Aug 3 17:15:45 CEST 2011
54 - use even more efficient and more compatible condvars for
55 compatibility to AnyEvent 6.x :)
56 - more inconsequential ecb.h updates.
57
586.03 Wed Aug 3 11:41:30 CEST 2011
59 - change how Coro patches AnyEvent condvars for compatibility to
60 AnyEvent 6.x.
61 - update ecb.h, to no longer include <pthread.h> in case WinNT.h
62 hasn't been included.
63
646.02 Wed Jul 13 04:35:19 CEST 2011
65 - "improve portability to Gentoo" - gentoo manages to put perl variables
66 in memory areas that are farther than 2gb apart, which the jit couldn't
67 handle and barfed. now it's just a bit slower on gentoo and similar
68 systems.
69
706.01 Sun Jul 3 12:31:14 CEST 2011
71 - workarounds are good, but the test for whether pthreads are used
72 was not good. thisone should be better.
73 - check differently whether gcc generates cfi instructions itself.
74
756.0 Wed Jun 29 19:43:35 CEST 2011
76 - INCOMPATIBLE CHANGE: unreferenced coro objects will now be
77 destroyed and cleaned up automatically (e.g. async { schedule }).
78 - implement a JIT compiler for part of the thread switch code,
79 which gives a 50% speed improvement on threaded perls, and
80 about 4% on non-thraeded perls (so threaded perls now finally
81 reach about half the speed of non-threaded perls).
82 - slightly modernise Coro::Intro, add section about rouse functions.
83 - avoid DEFSV and ERRSV, giving another 10% improvement
84 in thread switching.
85 - Coro::State->is_destroyed is now called is_zombie.
86 - implement a Coro->safe_cancel method that might fail, but
87 cancels in a "safer" way if it succeeds.
88 - add preliminary support for DEBUGGING perls.
89 - get rid of two hash-accesses when initialising a new Coro - this
90 speeds up coro creation by almost a factor of two.
91 - croak when a coro that is being cancelled tries to block
92 (e.g. while executing a guard block), instead of crashing or
93 deadlocking.
94 - use a more robust and also faster method to identify Coro::State
95 objects - speeds up everything a bit.
96 - implement Coro->cancel in XS for a 20% speed improvement, and to
97 be able to implement mutual cancellation.
98 - speed up context switches by a percent or two by more efficiently
99 allocating context stack entries.
100 - implement Coro->join and Coro->on_destroy in XS for a speedup and
101 a reduction in memory use.
102 - cancelling a coro while it itself is cancelling another coro is
103 now supported and working, instead of triggering an assertion.
104 - be a bit more crash-resistant when calling (buggy) on_destroy
105 callbacks (best effort).
106 - move on_destroy into the slf_frame, to allow extension slf
107 functions to have destructors.
108 - get rid if coro refcounting - simply crash in other interpreter
109 threads by nulling the pointers on clone.
110 - simplify warn/die hook handling when loading Coro - the convoluted
111 logic seems to be no longer neccessary.
112 - use libecb instead of our own home-grown gcc hacks.
113 - document alternatives to Coro::LWP. Please use them :)
114 - work around another mindless idiotic NEEDLESS bug in openbsd/mirbsds
115 sigaltstack. Really. wine suffers from it, erlang suffers from it,
116 and it's known since at least 2006.
117
1185.372 Wed Feb 23 06:14:30 CET 2011
119 - apparently mingw doesn't provide a working gettimeofday, try to
120 work around that by relying on Time::HiRes (indirectly brought to
121 my attention by Max Maischein).
122 - fix some portability issues when Time::HiRes was used.
123
1245.371 Mon Feb 21 14:36:08 CET 2011
125 - backport to windows process emulation code again.
126
1275.37 Sat Feb 19 07:49:44 CET 2011
128 - add a big "Coro thread life cycle" section to "man Coro".
129 - try a tentative workaround against the breakage that 5.13 has
130 introduced without depreciation period. sigh.
131 - no longer use Time::HiRes if gettimeofday is available, which
132 saves quite a lot of memory.
133
1345.36 Sun Feb 13 05:33:41 CET 2011
135 - automatically load Coro::Channel, Coro::RWLock, Coro::Semaphore,
136 Coro::SemaphoreSet, Coro::Signal and Coro::Specific on first "new"
137 method call.
138 - undocument Coro::Timer::sleep and obsolete whole module.
139 - optimise Coro::Timer::timeout memory and cpu usage.
140 - slightly updated Coro::Intro for recent changes.
141 - do not initialise PL_dirty anymore.
142
1435.25 Thu Nov 11 01:08:39 CET 2010
144 - try a different approach on netbsd - netbsd 5 finally has marginally
145 working pthreads, but still broken ucontext/sigaltstack.
146 - openbsd 4.8 finally got their act together, Coro works out of the box
147 with asm, setjmp and pthreads (no change, just informational).
148
1495.24 Sat Oct 23 11:27:12 CEST 2010
150 - port to the EV 4.0 API.
151 - work around bugs in mingw32, making strawberry perl work
152 out of the box.
153 - correctly modify Coro::AIO function prototypes
154 so that they reflect the "no optional parameters" rule.
155 - "ported" libcoro to C++.
156
1575.23 Mon May 17 18:50:42 CEST 2010
158 - be more resistant to ordering changes when initialising
159 Coro::AnyEvent, Coro::EV and Coro::Event (reported by Matthias
160 Waldorf).
161 - document that perl 5.12 deliberately removed support for cloning.
162
1635.22 Wed Apr 14 03:55:35 CEST 2010
164 - correctly return udnef on errors in Coro::Handle::read/write
165 (testcase by Marc Mims).
166 - convert Coro::Util into a "perl compatibility wrapper" - the functions
167 are less useful now, but are drop-in replacements for existing
168 functions, listing better alternatives in the documentation. This also
169 fixes a bug in Coro::LWP which naively substituted Socket::inet_aton
170 with Coro::Util::inet_aton.
171 - do not override $Coro::idle unconditionally in Coro.pm, as other
172 modules could have provided their own idle coro already
173 (for exmaple, Coro::AnyEvent).
174 - fix Coro::Util::gethost* functions.
175 - Coro::Timer corretcly exports it's symbols (reported by Hideki Yamamura).
176
1775.21 Wed Dec 16 07:19:51 CET 2009
178 - automatically load Coro::AnyEvent when AnyEvent and Coro are used
179 together.
180 - add some examples on how to combine other event loops with Coro in
181 Coro::AnyEvent, and how to run it (and not to block). Seems to be
182 the most common source of confusion.
183 - try to catch people naively blocking in an event callback.
184 - work around the perl filehandle bug issue in conjunction with
185 older common::sense (as indirectly pointed out by ZSystem).
186 - clarify the "not from signal handlers" section.
187
1885.2 Sun Oct 4 14:54:24 CEST 2009
189 - Coro::Storable destroyed the prototypes of the functions it wrapped.
190 - export rouse_cb and rouse_wait by default now.
191 - fix various prototype mismatches in Coro::AnyEvent and Coro::Handle.
192 - new method $state->swap_sv.
193 - added section on "windows process emulation" to the manpage,
194 after a not-so-fruitful (nor-friendly) "discussion" with chip
195 salzenberg (discussion implies arguments, but his only arguments
196 were ad-hominems, one wonders why he started it in the first
197 place). I hope this explains it well enough for him to understand,
198 and maybe well enough for others to understand.
199 - use common::sense everywhere now.
200 - idle callbacks are no longer supported, use idle coros instead.
201 - print a thread listing when a deadlock is detected.
202
2035.17 Sat Aug 22 23:09:31 CEST 2009
204 - work around a bug in the perl debugger causing crashes
205 when running under the debugger by marking _pool_handler as nodebug.
206 - speed up Coro::async considerably.
207 - try some hacks to get netbsd to work "more often" - their broken
208 setjmp/longjmp, ucontext *and* phtreads are really hard on Coro.
209 - convert Coro to AE 5.0 API.
210
2115.162 Tue Jul 28 04:04:03 CEST 2009
212 - perl 5.8.2 is now minimum requirement.
213 - skip t/19_handle.t on broken windows perls.
214
2155.161 Wed Jul 22 04:47:38 CEST 2009
216 - Coro::AnyEvent::poll could have a different prototype when EV was
217 used as backend (analyzed by Tatsuhiko Miyagawa).
218 - Coro::AnyEvent errornously initialised the event loop when loaded,
219 not on demand.
220 - try to workaround rare */t/01_unblock.t failures.
221
2225.16 Tue Jul 21 01:44:37 CEST 2009
223 - Coro::AnyEvent failed to hook into the event loop
224 when no threads had been readied between detecting
225 the event loop and actually running it.
226 - considerably speed up Coro::Select by taking avdantage
227 of AnyEvent > 4.8 and some other optimisations.
228 - implement paragraph readline mode in Coro::Handle
229 (based on patches by Zsbán Ambrus).
230 - replace WSAEINPROGRESS by WSAEWOULDBLOCK (reported
231 and analyzed by Yasuhiro MATSUMOTO).
232 - clarified libcoro license and copyright.
233 - someone stole my EXTRA_META!!!
234 - implement Coro::Select::patch_pp_sselect and it's brother,
235 for hardcode select overriding.
236
2375.151 Mon Jul 6 05:41:57 CEST 2009
238 - backport to windows process emulation code again (patch by
239 Yasuhiro MATSUMOTO).
240 - slightly update Coro::MakeMaker.
241
2425.15 Tue Jun 30 10:28:06 CEST 2009
243 - deprecate Coro::Socket, document how to get ipv6 support via
244 AnyEvent::Socket instead.
245 - implement signal->wait ($cb) interface, similar to semaphores.
246 - work around SvOK not supporting getmagic, so we have to getmagic
247 to test for undef :/ (reported by Matthias Waldorf).
248 - load Coro::AnyEvent in all modules using AnyEvent.
249 - work around perl corrupting our internal data structures,
250 reported by Tokuhiro Matsuno.
251 - enable per-coroutine real and cpu time gathering
252 (Coro::State::enable_times).
253
2545.14 Wed Jun 24 01:37:48 CEST 2009
255 - provide explicit functions to "cede" to the event loop in Coro::AnyEvent,
256 as this seems to have been a difficult concept (poll, sleep, idle,
257 idle_upto).
258 - add Coro::AnyEvent::readable/writable functions.
259 - clarify Coro::EV/Event/AnyEvent manpages.
260 - free per-thread global scalars in the thread calling ->cancel, to
261 avoid crashes when $_, $@ etc., are magical but some of those
262 had already been freed.
263 - "unexperimentalise" the callback interface for Coro::Semaphore.
264 - speed up ready queue management/context switching by using a linked
265 list instead of an array (~5-10%).
266 - implement "watch" command in Coro::Debug shells.
267 - for fun, implement time-slicing as an example in the manpage.
268 - if AnyEvent detects EV or Event, but we don't have Coro::EV or
269 Coro::Event, use the normal AnyEvent handling instead of dieing
270 (the same is true for Coro::Handle).
271 - properly document Coro::EV::timed_io_once.
272 - avoid unneccessary ->cancel calls in Coro::Handle.
273 - maybe make it work on mingw32 with win32 backend
274 (based on patch by Yasuhiro Matsumoto).
275
2765.132 Fri May 29 09:00:39 CEST 2009
277 - do not keep a reference to the argument itself in
278 Coro::Semaphore::guard, as it could change later.
279 - support SO_RCVBUF/SO_SNDBUF nonstandard Coro::Socket options,
280 should support a prepare callback.
281
2825.131 Mon Mar 16 23:20:37 CET 2009
283 - implement and document Coro->suspend, Coro->resume.
284 - fix Coro::Select implementation to not (often) close
285 the passed file descriptors (testcase provided by pippijn).
286
2875.13 Mon Dec 15 21:51:42 CET 2008
288 - EXPERIMENTAL: implement dynamic winds (on_enter/on_leave).
289 - don't set diehook to C<undef> but instead to NULL, to avoid
290 spurious warnings.
291 - fix a lot of bugs in Coro::SemaphoreSet.
292 - Coro::SemaphoreSet will less often create a semaphore needlessly.
293 - add Coro::SemaphoreSet::count and wait methods.
294 - take advantage of the new Guard module.
295 - deprecate Coro::guard.
296 - try to fix the dreaded 01_unblock tests once more. I hate it when
297 testsuites need more fixing than the code they are supposed to test.
298 - croak in more cases when a required callback isn't resolvable.
299 - fix some minor issues in Coro::State->call/eval.
300 - use current coroutine context instead of a temporary one
301 when temporarily switching to another coroutine.
302 - do not call C-level on_destroy handlers during global destruction,
303 to avoid needless segfaults.
304
3055.12 Sun Dec 7 13:30:38 CET 2008
306 - add default config for MirOS, which seems to be a bug-to-bug
307 compatible fork of openbsd ("world domination by releasing
308 openbsd cvs before the openbsd folks do it" or so :).
309 - free_padlist did destroy the names pad, not good, but didn't
310 seem to bother perl - this could fix issues such as eval ""
311 inside a function called from multiple coroutines.
312 - use a different method to detect destruction time.
313 - be more careful when freeing padlists just before global
314 destruction.
315 - fixed and expanded the call/cc example.
316 - renamed _terminate to _coro_run.
317 - new method Coro::Channel->shutdown.
318 - try pthreads on openbsd <4.4 (broken sigaltstack, will
319 pthreads fare better?).
320 - be less picky about destroying "a" running coroutine.
321
3225.11 Tue Nov 25 21:49:05 CET 2008
323 - DEBUGGING in 5.10.0 is a mess: it constantly flags perfectly
324 working code with failed assertions, introducing more bugs than
325 it fixes, requiring elaborate workarounds :(
326
3275.1 Mon Nov 24 08:54:59 CET 2008
328 - wrote a small introductory tutorial - Coro::Intro.
329 - convert Coro::Timer, Coro::Select and Coro::Util to rouse API.
330 - Coro::Select did errornously dup the file descriptors
331 and didn't work with all AnyEvent backends.
332 - Coro::Select wasn't imported correctly form Coro::LWP, causing blocking
333 LWP data transfers.
334 - disassociate c contexts from coro objects - this is agruably more
335 correct, but mostly allows sharing of cctxs between coro and state
336 objects, for added memory savings and speed increases.
337 - bumped $Coro::POOL_RSS up to 32kb by default.
338 - no longer set the optype to OP_CUSTOM, as B::* understandably
339 doesn't like this very much (and we *are* a type of entersub).
340 - implement state cloning, just to prove that call/cc can be done.
341 - automatically load Coro::AnyEvent in Coro::Handle.
342 - wrap ->cancel calls in eval inside Coro::Handle as EV watchers
343 do not have this method (and don't need it either).
344 - speed up generic anyevent methods in Coro::Handle by using rouse
345 callbacks.
346 - allow coroutines in $Coro::IDLE, speeding up Coro::AnyEvent and
347 others. It also makes the debugger happier, as you can trace
348 through the idle threads now.
349 - add comppad_name* and hints ($^H, %^H) to per-thread variables.
350 - eg/event was pretty broken.
351 - better 5.8.6 compatibility.
352
3535.0 Thu Nov 20 10:35:05 CET 2008
354 - NEW ARCHITECTURE: use the latest 4.x version if you experience
355 stability issues.
356 - bump API version to 7 - all dependents must be recompiled.
357 - removed timed_* functions - they were not being used anyways
358 and should be replaced by a more generic mechanism -
359 and were annoying to support anyways :)
360 - removed SemaphoreSet's waiter method - use sem method instead.
361 - Coro::Semaphore->adjust didn't correctly wake up enough waiters.
362 - async_pool did free a scalar value twice
363 ("Attempt to unreference...").
364 - fix a longstanding bug where calling terminate on a coro that
365 was waiting for a semaphore that was just becoming available
366 would cause a deadlock (semaphore would get into a state where
367 it was available but waiters were still blocked).
368 - calling throw on a coroutine that is waiting for a semaphore will
369 no longer make it acquire the semaphore (and thus leak a count).
370 - perl's process emulation is now not even theoretically supported
371 anymore.
372 - new functions Coro::rouse_cb and Coro::rouse_wait for easier
373 conversion of callback-style to blocking-style.
374 - new methods $coro->schedule_to and ->cede_to, to specifically
375 schedule or cede to a specific coroutine.
376 - new function Coro::Semaphore::wait.
377 - use named constants in Coro::Channel (Richard Hundt).
10 - directly patch the entersub opcode calling ssl functions (cede, 378 - directly patch the entersub opcode calling SLF functions (cede,
11 transfer and so on). this does speed up context switching by 379 transfer and so on). this does speed up context switching, but
12 up to 10%. More importanly, it frees us from the hardcoded 380 more importanly, it frees us from the hardcoded behaviour of
13 behaviour of entersub, so we might actually be able to return 381 entersub, so we might actually be able to return something from
14 something from those functions. 382 those functions and atcually create new ones.
383 - take advantage of __builtin_frame_address on gcc.
384 - expose THX in coroapi (not sure whether this was a wise decision,
385 as "threaded" perls are running at half speed anyways).
386 - implement execute_slf (schedule-like-function) interface that makes
387 it possible to implement schedule-like-functions in XS.
388 - use new SLF interface to massively speed up Coro::EV by roughly a
389 factor of two.
390 - used new SLF interface to massively speed up Coro::Semaphore by a
391 factor of three.
392 - used new SLF interface to speed up Coro::AIO by roughly a factor of
393 four and reduce its memory usage considerably.
394 - implement Coro::SemaphoreSet purely in terms of Coro::Semaphore,
395 for a nice speedup and vastly more correct behaviour. Also implement
396 a new method "sem" to get at the underlying semaphore object.
397 - implement Coro::Channel in terms of Coro::Semaphore, for a moderate
398 (in comparison) 20-40% speedup.
399 - used new SLF interface to reimplement Coro::Signal gaining
400 some unknown (because I was too lazy), but certain, speedup, and also
401 making signals reliable for the first time.
402 - used new SLF interface and other optimisations to speed up async_pool
403 by a factor of two. It also doesn't rely on perl's exception mechanism
404 to exit anymore. The overhead for terminating an async_pool, coro over
405 a normal async is now very small.
406 - sped up coroutine creation/destruction by 40%.
407 - forgot to include Coro/libcoro/README in the dist for all these years.
408 - work around a freebsd pthreads bug (manual testcancel is required as
409 pthread_cond_wait isn't a cancellation point on freebsd).
410 - use new rouse functions to speed up and simplify Coro::BDB.
411 - make "prefer perl native functions" work with threaded perls.
412 - condense Coro::Debug ps output, hint at v and w flags.
413 - (libcoro) lots of minor cleanups and portability improvements.
414
4154.914 Wed Nov 19 12:54:18 CET 2008
416 - fix a disastrous bug in the readline optimisation
417 introduced in 4.801.
418
4194.913 Sat Nov 15 07:58:28 CET 2008
420 - async_pool did free a scalar value twice
421 ("Attempt to unreference...").
15 422
164.912 Thu Nov 13 18:31:23 CET 2008 4234.912 Thu Nov 13 18:31:23 CET 2008
17 - minor cleanups. 424 - minor cleanups.
18 - use much larger stacks on linux and perl < 5.8.8. 425 - use much larger stacks on linux and perl < 5.8.8.
19 - Coro::Debug::new_unix_server did not unlink the socket 426 - Coro::Debug::new_unix_server did not unlink the socket

Diff Legend

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