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

Comparing Coro/Changes (file contents):
Revision 1.416 by root, Sat Nov 15 07:01:47 2008 UTC vs.
Revision 1.606 by root, Fri Dec 7 21:42:42 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 8 - use experimental fiber implementation on native windows
9 - NEW ARCHITECTURE: use 4.912 for the stable version. 9 perls.
10 - use sizeof (void *) as multiplication factor for stack sizes,
11 to accomodate the totally braindamaged microsoft 64 bit "os".
12 - changed verifier host from win2k-ap510-32 to win7-sp516-32/64.
13 activeperl 5.16 crashes when PerlIO_define_layer is called due
14 to some bug in the perl dll, strawberry perl at least passes
15 the testsuite.
16 - implement Coro::Handle->peeraddr/host/port, for slightly
17 improved compatibility with LWP.
18 - implement 5.17 compatibility by almost blindly applying a
19 good-looking patch by Father Chrysostomos.
20 - move stack management functions into libcoro 3.
21 - libcoro version 3 "released".
22 - support magic values as timed_io_once args.
23 - recommend AnyEvent 7+ or EV 4+, also require EV
24 version 4 or newer for Coro::EV.
25
266.10 Tue Oct 9 01:14:27 CEST 2012
27 - updated ecb.h, it had a typo that caused it to not compile on many
28 big endian systems (reported by many people).
29 - disable memory fences in ecb.h to improve portability.
30
316.09 Sat Oct 6 23:25:02 CEST 2012
32 - Coro::EV I/O watchers were not interruptible by exceptions
33 (Coro::State::throw) (testcase by sten).
34 - ->throw now puts threads into the ready queue, as this seems to
35 be expected by existing code, and code that doesn't cope with spurious
36 wakeups needs fixing anyway.
37 - use fd -1 in mmap.
38 - cast I32 to int in error message printf.
39 - warn about broken so-called "hardened" kernels.
40
416.08 Fri Apr 13 12:05:47 CEST 2012
42 - be more aggressive about exiting like perl does - formerly,
43 exiting from the non-main thread would not execute END blocks.
44
456.07 Fri Nov 11 21:21:48 CET 2011
46 - work around a bug in PerlIO (setting $SIG{__WARN__} to a PVCV).
47 - update ecb.h.
48
496.06 Mon Aug 8 23:59:48 CEST 2011
50 - cygwin unfortunately patches the stack at runtime, so we use the pthreads
51 backend, which is an order of magnitude slower. unfortunately, cygwins
52 pthread implementation isn't very complete either, so allocate the stack
53 twice just to be sure.
54 (note: cygwin also enables mymalloc, which is NOT THREADSAFE ON WINDOWS,
55 in its ithreaded perl - best recompile cygwin and use the 'w'indows
56 backend for much better performance. also disable ithreads for
57 even better performance...).
58
596.05 Thu Aug 4 21:36:36 CEST 2011
60 - blush, condvar values would not be propagated from send to recv anymore
61 (reported by Chip Salzenberg).
62 - use exponential increase for the readline buffer length in
63 Coro::Handle. also reduce initial allocation to 1020 from 4096 bytes.
64
656.04 Wed Aug 3 17:15:45 CEST 2011
66 - use even more efficient and more compatible condvars for
67 compatibility to AnyEvent 6.x :)
68 - more inconsequential ecb.h updates.
69
706.03 Wed Aug 3 11:41:30 CEST 2011
71 - change how Coro patches AnyEvent condvars for compatibility to
72 AnyEvent 6.x.
73 - update ecb.h, to no longer include <pthread.h> in case WinNT.h
74 hasn't been included.
75
766.02 Wed Jul 13 04:35:19 CEST 2011
77 - "improve portability to Gentoo" - gentoo manages to put perl variables
78 in memory areas that are farther than 2gb apart, which the jit couldn't
79 handle and barfed. now it's just a bit slower on gentoo and similar
80 systems.
81
826.01 Sun Jul 3 12:31:14 CEST 2011
83 - workarounds are good, but the test for whether pthreads are used
84 was not good. thisone should be better.
85 - check differently whether gcc generates cfi instructions itself.
86
876.0 Wed Jun 29 19:43:35 CEST 2011
88 - INCOMPATIBLE CHANGE: unreferenced coro objects will now be
89 destroyed and cleaned up automatically (e.g. async { schedule }).
90 - implement a JIT compiler for part of the thread switch code,
91 which gives a 50% speed improvement on threaded perls, and
92 about 4% on non-thraeded perls (so threaded perls now finally
93 reach about half the speed of non-threaded perls).
94 - slightly modernise Coro::Intro, add section about rouse functions.
95 - avoid DEFSV and ERRSV, giving another 10% improvement
96 in thread switching.
97 - Coro::State->is_destroyed is now called is_zombie.
98 - implement a Coro->safe_cancel method that might fail, but
99 cancels in a "safer" way if it succeeds.
100 - add preliminary support for DEBUGGING perls.
101 - get rid of two hash-accesses when initialising a new Coro - this
102 speeds up coro creation by almost a factor of two.
103 - croak when a coro that is being cancelled tries to block
104 (e.g. while executing a guard block), instead of crashing or
105 deadlocking.
106 - use a more robust and also faster method to identify Coro::State
107 objects - speeds up everything a bit.
108 - implement Coro->cancel in XS for a 20% speed improvement, and to
109 be able to implement mutual cancellation.
110 - speed up context switches by a percent or two by more efficiently
111 allocating context stack entries.
112 - implement Coro->join and Coro->on_destroy in XS for a speedup and
113 a reduction in memory use.
114 - cancelling a coro while it itself is cancelling another coro is
115 now supported and working, instead of triggering an assertion.
116 - be a bit more crash-resistant when calling (buggy) on_destroy
117 callbacks (best effort).
118 - move on_destroy into the slf_frame, to allow extension slf
119 functions to have destructors.
120 - get rid if coro refcounting - simply crash in other interpreter
121 threads by nulling the pointers on clone.
122 - simplify warn/die hook handling when loading Coro - the convoluted
123 logic seems to be no longer neccessary.
124 - use libecb instead of our own home-grown gcc hacks.
125 - document alternatives to Coro::LWP. Please use them :)
126 - work around another mindless idiotic NEEDLESS bug in openbsd/mirbsds
127 sigaltstack. Really. wine suffers from it, erlang suffers from it,
128 and it's known since at least 2006.
129
1305.372 Wed Feb 23 06:14:30 CET 2011
131 - apparently mingw doesn't provide a working gettimeofday, try to
132 work around that by relying on Time::HiRes (indirectly brought to
133 my attention by Max Maischein).
134 - fix some portability issues when Time::HiRes was used.
135
1365.371 Mon Feb 21 14:36:08 CET 2011
137 - backport to windows process emulation code again.
138
1395.37 Sat Feb 19 07:49:44 CET 2011
140 - add a big "Coro thread life cycle" section to "man Coro".
141 - try a tentative workaround against the breakage that 5.13 has
142 introduced without depreciation period. sigh.
143 - no longer use Time::HiRes if gettimeofday is available, which
144 saves quite a lot of memory.
145
1465.36 Sun Feb 13 05:33:41 CET 2011
147 - automatically load Coro::Channel, Coro::RWLock, Coro::Semaphore,
148 Coro::SemaphoreSet, Coro::Signal and Coro::Specific on first "new"
149 method call.
150 - undocument Coro::Timer::sleep and obsolete whole module.
151 - optimise Coro::Timer::timeout memory and cpu usage.
152 - slightly updated Coro::Intro for recent changes.
153 - do not initialise PL_dirty anymore.
154
1555.25 Thu Nov 11 01:08:39 CET 2010
156 - try a different approach on netbsd - netbsd 5 finally has marginally
157 working pthreads, but still broken ucontext/sigaltstack.
158 - openbsd 4.8 finally got their act together, Coro works out of the box
159 with asm, setjmp and pthreads (no change, just informational).
160
1615.24 Sat Oct 23 11:27:12 CEST 2010
162 - port to the EV 4.0 API.
163 - work around bugs in mingw32, making strawberry perl work
164 out of the box.
165 - correctly modify Coro::AIO function prototypes
166 so that they reflect the "no optional parameters" rule.
167 - "ported" libcoro to C++.
168
1695.23 Mon May 17 18:50:42 CEST 2010
170 - be more resistant to ordering changes when initialising
171 Coro::AnyEvent, Coro::EV and Coro::Event (reported by Matthias
172 Waldorf).
173 - document that perl 5.12 deliberately removed support for cloning.
174
1755.22 Wed Apr 14 03:55:35 CEST 2010
176 - correctly return udnef on errors in Coro::Handle::read/write
177 (testcase by Marc Mims).
178 - convert Coro::Util into a "perl compatibility wrapper" - the functions
179 are less useful now, but are drop-in replacements for existing
180 functions, listing better alternatives in the documentation. This also
181 fixes a bug in Coro::LWP which naively substituted Socket::inet_aton
182 with Coro::Util::inet_aton.
183 - do not override $Coro::idle unconditionally in Coro.pm, as other
184 modules could have provided their own idle coro already
185 (for exmaple, Coro::AnyEvent).
186 - fix Coro::Util::gethost* functions.
187 - Coro::Timer corretcly exports it's symbols (reported by Hideki Yamamura).
188
1895.21 Wed Dec 16 07:19:51 CET 2009
190 - automatically load Coro::AnyEvent when AnyEvent and Coro are used
191 together.
192 - add some examples on how to combine other event loops with Coro in
193 Coro::AnyEvent, and how to run it (and not to block). Seems to be
194 the most common source of confusion.
195 - try to catch people naively blocking in an event callback.
196 - work around the perl filehandle bug issue in conjunction with
197 older common::sense (as indirectly pointed out by ZSystem).
198 - clarify the "not from signal handlers" section.
199
2005.2 Sun Oct 4 14:54:24 CEST 2009
201 - Coro::Storable destroyed the prototypes of the functions it wrapped.
202 - export rouse_cb and rouse_wait by default now.
203 - fix various prototype mismatches in Coro::AnyEvent and Coro::Handle.
204 - new method $state->swap_sv.
205 - added section on "windows process emulation" to the manpage,
206 after a not-so-fruitful (nor-friendly) "discussion" with chip
207 salzenberg (discussion implies arguments, but his only arguments
208 were ad-hominems, one wonders why he started it in the first
209 place). I hope this explains it well enough for him to understand,
210 and maybe well enough for others to understand.
211 - use common::sense everywhere now.
212 - idle callbacks are no longer supported, use idle coros instead.
213 - print a thread listing when a deadlock is detected.
214
2155.17 Sat Aug 22 23:09:31 CEST 2009
216 - work around a bug in the perl debugger causing crashes
217 when running under the debugger by marking _pool_handler as nodebug.
218 - speed up Coro::async considerably.
219 - try some hacks to get netbsd to work "more often" - their broken
220 setjmp/longjmp, ucontext *and* phtreads are really hard on Coro.
221 - convert Coro to AE 5.0 API.
222
2235.162 Tue Jul 28 04:04:03 CEST 2009
224 - perl 5.8.2 is now minimum requirement.
225 - skip t/19_handle.t on broken windows perls.
226
2275.161 Wed Jul 22 04:47:38 CEST 2009
228 - Coro::AnyEvent::poll could have a different prototype when EV was
229 used as backend (analyzed by Tatsuhiko Miyagawa).
230 - Coro::AnyEvent errornously initialised the event loop when loaded,
231 not on demand.
232 - try to workaround rare */t/01_unblock.t failures.
233
2345.16 Tue Jul 21 01:44:37 CEST 2009
235 - Coro::AnyEvent failed to hook into the event loop
236 when no threads had been readied between detecting
237 the event loop and actually running it.
238 - considerably speed up Coro::Select by taking avdantage
239 of AnyEvent > 4.8 and some other optimisations.
240 - implement paragraph readline mode in Coro::Handle
241 (based on patches by Zsbán Ambrus).
242 - replace WSAEINPROGRESS by WSAEWOULDBLOCK (reported
243 and analyzed by Yasuhiro MATSUMOTO).
244 - clarified libcoro license and copyright.
245 - someone stole my EXTRA_META!!!
246 - implement Coro::Select::patch_pp_sselect and it's brother,
247 for hardcode select overriding.
248
2495.151 Mon Jul 6 05:41:57 CEST 2009
250 - backport to windows process emulation code again (patch by
251 Yasuhiro MATSUMOTO).
252 - slightly update Coro::MakeMaker.
253
2545.15 Tue Jun 30 10:28:06 CEST 2009
255 - deprecate Coro::Socket, document how to get ipv6 support via
256 AnyEvent::Socket instead.
257 - implement signal->wait ($cb) interface, similar to semaphores.
258 - work around SvOK not supporting getmagic, so we have to getmagic
259 to test for undef :/ (reported by Matthias Waldorf).
260 - load Coro::AnyEvent in all modules using AnyEvent.
261 - work around perl corrupting our internal data structures,
262 reported by Tokuhiro Matsuno.
263 - enable per-coroutine real and cpu time gathering
264 (Coro::State::enable_times).
265
2665.14 Wed Jun 24 01:37:48 CEST 2009
267 - provide explicit functions to "cede" to the event loop in Coro::AnyEvent,
268 as this seems to have been a difficult concept (poll, sleep, idle,
269 idle_upto).
270 - add Coro::AnyEvent::readable/writable functions.
271 - clarify Coro::EV/Event/AnyEvent manpages.
272 - free per-thread global scalars in the thread calling ->cancel, to
273 avoid crashes when $_, $@ etc., are magical but some of those
274 had already been freed.
275 - "unexperimentalise" the callback interface for Coro::Semaphore.
276 - speed up ready queue management/context switching by using a linked
277 list instead of an array (~5-10%).
278 - implement "watch" command in Coro::Debug shells.
279 - for fun, implement time-slicing as an example in the manpage.
280 - if AnyEvent detects EV or Event, but we don't have Coro::EV or
281 Coro::Event, use the normal AnyEvent handling instead of dieing
282 (the same is true for Coro::Handle).
283 - properly document Coro::EV::timed_io_once.
284 - avoid unneccessary ->cancel calls in Coro::Handle.
285 - maybe make it work on mingw32 with win32 backend
286 (based on patch by Yasuhiro Matsumoto).
287
2885.132 Fri May 29 09:00:39 CEST 2009
289 - do not keep a reference to the argument itself in
290 Coro::Semaphore::guard, as it could change later.
291 - support SO_RCVBUF/SO_SNDBUF nonstandard Coro::Socket options,
292 should support a prepare callback.
293
2945.131 Mon Mar 16 23:20:37 CET 2009
295 - implement and document Coro->suspend, Coro->resume.
296 - fix Coro::Select implementation to not (often) close
297 the passed file descriptors (testcase provided by pippijn).
298
2995.13 Mon Dec 15 21:51:42 CET 2008
300 - EXPERIMENTAL: implement dynamic winds (on_enter/on_leave).
301 - don't set diehook to C<undef> but instead to NULL, to avoid
302 spurious warnings.
303 - fix a lot of bugs in Coro::SemaphoreSet.
304 - Coro::SemaphoreSet will less often create a semaphore needlessly.
305 - add Coro::SemaphoreSet::count and wait methods.
306 - take advantage of the new Guard module.
307 - deprecate Coro::guard.
308 - try to fix the dreaded 01_unblock tests once more. I hate it when
309 testsuites need more fixing than the code they are supposed to test.
310 - croak in more cases when a required callback isn't resolvable.
311 - fix some minor issues in Coro::State->call/eval.
312 - use current coroutine context instead of a temporary one
313 when temporarily switching to another coroutine.
314 - do not call C-level on_destroy handlers during global destruction,
315 to avoid needless segfaults.
316
3175.12 Sun Dec 7 13:30:38 CET 2008
318 - add default config for MirOS, which seems to be a bug-to-bug
319 compatible fork of openbsd ("world domination by releasing
320 openbsd cvs before the openbsd folks do it" or so :).
321 - free_padlist did destroy the names pad, not good, but didn't
322 seem to bother perl - this could fix issues such as eval ""
323 inside a function called from multiple coroutines.
324 - use a different method to detect destruction time.
325 - be more careful when freeing padlists just before global
326 destruction.
327 - fixed and expanded the call/cc example.
328 - renamed _terminate to _coro_run.
329 - new method Coro::Channel->shutdown.
330 - try pthreads on openbsd <4.4 (broken sigaltstack, will
331 pthreads fare better?).
332 - be less picky about destroying "a" running coroutine.
333
3345.11 Tue Nov 25 21:49:05 CET 2008
335 - DEBUGGING in 5.10.0 is a mess: it constantly flags perfectly
336 working code with failed assertions, introducing more bugs than
337 it fixes, requiring elaborate workarounds :(
338
3395.1 Mon Nov 24 08:54:59 CET 2008
340 - wrote a small introductory tutorial - Coro::Intro.
341 - convert Coro::Timer, Coro::Select and Coro::Util to rouse API.
342 - Coro::Select did errornously dup the file descriptors
343 and didn't work with all AnyEvent backends.
344 - Coro::Select wasn't imported correctly form Coro::LWP, causing blocking
345 LWP data transfers.
346 - disassociate c contexts from coro objects - this is agruably more
347 correct, but mostly allows sharing of cctxs between coro and state
348 objects, for added memory savings and speed increases.
349 - bumped $Coro::POOL_RSS up to 32kb by default.
350 - no longer set the optype to OP_CUSTOM, as B::* understandably
351 doesn't like this very much (and we *are* a type of entersub).
352 - implement state cloning, just to prove that call/cc can be done.
353 - automatically load Coro::AnyEvent in Coro::Handle.
354 - wrap ->cancel calls in eval inside Coro::Handle as EV watchers
355 do not have this method (and don't need it either).
356 - speed up generic anyevent methods in Coro::Handle by using rouse
357 callbacks.
358 - allow coroutines in $Coro::IDLE, speeding up Coro::AnyEvent and
359 others. It also makes the debugger happier, as you can trace
360 through the idle threads now.
361 - add comppad_name* and hints ($^H, %^H) to per-thread variables.
362 - eg/event was pretty broken.
363 - better 5.8.6 compatibility.
364
3655.0 Thu Nov 20 10:35:05 CET 2008
366 - NEW ARCHITECTURE: use the latest 4.x version if you experience
367 stability issues.
10 - bump API version to 7 - all dependents must be recompiled. 368 - bump API version to 7 - all dependents must be recompiled.
369 - removed timed_* functions - they were not being used anyways
370 and should be replaced by a more generic mechanism -
371 and were annoying to support anyways :)
372 - removed SemaphoreSet's waiter method - use sem method instead.
11 - Coro::Semaphore->adjust didn't correctly wake up enough waiters. 373 - Coro::Semaphore->adjust didn't correctly wake up enough waiters.
12 - async_pool did free a scalar value twice 374 - async_pool did free a scalar value twice
13 ("Attempt to unreference..."). 375 ("Attempt to unreference...").
376 - fix a longstanding bug where calling terminate on a coro that
377 was waiting for a semaphore that was just becoming available
378 would cause a deadlock (semaphore would get into a state where
379 it was available but waiters were still blocked).
380 - calling throw on a coroutine that is waiting for a semaphore will
381 no longer make it acquire the semaphore (and thus leak a count).
14 - perl's process emulation is now not even theoretically supported 382 - perl's process emulation is now not even theoretically supported
15 anymore. 383 anymore.
384 - new functions Coro::rouse_cb and Coro::rouse_wait for easier
385 conversion of callback-style to blocking-style.
386 - new methods $coro->schedule_to and ->cede_to, to specifically
387 schedule or cede to a specific coroutine.
388 - new function Coro::Semaphore::wait.
389 - use named constants in Coro::Channel (Richard Hundt).
16 - directly patch the entersub opcode calling SLF functions (cede, 390 - directly patch the entersub opcode calling SLF functions (cede,
17 transfer and so on). this does speed up context switching, but 391 transfer and so on). this does speed up context switching, but
18 more importanly, it frees us from the hardcoded behaviour of 392 more importanly, it frees us from the hardcoded behaviour of
19 entersub, so we might actually be able to return something from 393 entersub, so we might actually be able to return something from
20 those functions and atcually create new ones. 394 those functions and atcually create new ones.
21 - take advantage of __builtin_frame_address on gcc. 395 - take advantage of __builtin_frame_address on gcc.
22 - expose THX in coroapi (not sure whether this was a wise decision, 396 - expose THX in coroapi (not sure whether this was a wise decision,
23 as "threaded" perls are running at half speed anyways). 397 as "threaded" perls are running at half speed anyways).
24 - implement execute_slf (schedule-like-function) interface that makes 398 - implement execute_slf (schedule-like-function) interface that makes
25 it possible to implement schedule-like-functions in XS. 399 it possible to implement schedule-like-functions in XS.
26 - use new SLF interface to massively speed up Coro::EV by roughly a factor 400 - use new SLF interface to massively speed up Coro::EV by roughly a
27 of two. 401 factor of two.
28 - removed timed_down and timed_guard functions - they were not being used
29 anyways and should be replaced by a more generic mechanism - and were
30 annoying to support anyways :
31 - used new SLF interface to massively speed up Coro::Semaphore by a factor 402 - used new SLF interface to massively speed up Coro::Semaphore by a
32 of three. 403 factor of three.
404 - used new SLF interface to speed up Coro::AIO by roughly a factor of
405 four and reduce its memory usage considerably.
406 - implement Coro::SemaphoreSet purely in terms of Coro::Semaphore,
407 for a nice speedup and vastly more correct behaviour. Also implement
408 a new method "sem" to get at the underlying semaphore object.
409 - implement Coro::Channel in terms of Coro::Semaphore, for a moderate
410 (in comparison) 20-40% speedup.
411 - used new SLF interface to reimplement Coro::Signal gaining
412 some unknown (because I was too lazy), but certain, speedup, and also
413 making signals reliable for the first time.
414 - used new SLF interface and other optimisations to speed up async_pool
415 by a factor of two. It also doesn't rely on perl's exception mechanism
416 to exit anymore. The overhead for terminating an async_pool, coro over
417 a normal async is now very small.
418 - sped up coroutine creation/destruction by 40%.
419 - forgot to include Coro/libcoro/README in the dist for all these years.
420 - work around a freebsd pthreads bug (manual testcancel is required as
421 pthread_cond_wait isn't a cancellation point on freebsd).
422 - use new rouse functions to speed up and simplify Coro::BDB.
423 - make "prefer perl native functions" work with threaded perls.
424 - condense Coro::Debug ps output, hint at v and w flags.
425 - (libcoro) lots of minor cleanups and portability improvements.
426
4274.914 Wed Nov 19 12:54:18 CET 2008
428 - fix a disastrous bug in the readline optimisation
429 introduced in 4.801.
33 430
344.913 Sat Nov 15 07:58:28 CET 2008 4314.913 Sat Nov 15 07:58:28 CET 2008
35 - async_pool did free a scalar value twice 432 - async_pool did free a scalar value twice
36 ("Attempt to unreference..."). 433 ("Attempt to unreference...").
37 434

Diff Legend

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