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

Comparing libev/Changes (file contents):
Revision 1.64 by root, Tue Sep 30 19:51:44 2008 UTC vs.
Revision 1.369 by root, Wed May 19 06:15:50 2021 UTC

1Revision history for libev, a high-performance and full-featured event loop. 1Revision history for libev, a high-performance and full-featured event loop.
2 2
3WISH? monotonic clocks times/GetTickCount for coarse corrections? 3TODO: for next ABI/API change, consider moving EV__IOFDSSET into io->fd instead and provide a getter.
4TODO: make ev_once timeout=0 useful, and make I/O always win. 4TODO: document EV_TSTAMP_T
5
6 - libev did not compile on systems lacking clock_gettime.
7 - due to a logic bug, timerfd availability did not always translate
8 into fewer wakeups - libev is now more aggressive in using
9 longer sleeps, even when timerfd is not in use.
10
114.33 Wed Mar 18 13:22:29 CET 2020
12 - no changes w.r.t. 4.32.
13
144.32 (EV only)
15 - the 4.31 timerfd code wrongly changed the priority of the signal
16 fd watcher, which is usually harmless unless signal fds are
17 also used (found via cpan tester service).
18 - the documentation wrongly claimed that user may modify fd and events
19 members in io watchers when the watcher was stopped
20 (found by b_jonas).
21 - new ev_io_modify mutator which changes only the events member,
22 which can be faster. also added ev::io::set (int events) method
23 to ev++.h.
24 - officially allow a zero events mask for io watchers. this should
25 work with older libev versions as well but was not officially
26 allowed before.
27 - do not wake up every minute when timerfd is used to detect timejumps.
28 - do not wake up every minute when periodics are disabled and we have
29 a monotonic clock.
30 - support a lot more "uncommon" compile time configurations,
31 such as ev_embed enabled but ev_timer disabled.
32 - use a start/stop wrapper class to reduce code duplication in
33 ev++.h and make it needlessly more c++-y.
34 - the linux aio backend is no longer compiled in by default.
35 - update to libecb version 0x00010008.
36
374.31 Fri Dec 20 21:58:29 CET 2019
38 - handle backends with minimum wait time a bit better by not
39 waiting in the presence of already-expired timers
40 (behaviour reported by Felipe Gasper).
41 - new feature: use timerfd to detect timejumps quickly,
42 can be disabled with the new EVFLAG_NOTIMERFD loop flag.
43 - document EV_USE_SIGNALFD feature macro.
44
454.30 (EV only)
46 - change non-autoconf test for __kernel_rwf_t by testing
47 LINUX_VERSION_CODE, the most direct test I could find.
48 - fix a bug in the io_uring backend that polled the wrong
49 backend fd, causing it to not work in many cases.
50
514.29 (EV only)
52 - add io uring autoconf and non-autoconf detection.
53 - disable io_uring when some header files are too old.
54
554.28 (EV only)
56 - linuxaio backend resulted in random memory corruption
57 when loop is forked.
58 - linuxaio backend might have tried to cancel an iocb
59 multiple times (was unable to trigger this).
60 - linuxaio backend now employs a generation counter to
61 avoid handling spurious events from cancelled requests.
62 - io_cancel can return EINTR, deal with it. also, assume
63 io_submit also returns EINTR.
64 - fix some other minor bugs in linuxaio backend.
65 - ev_tstamp type can now be overriden by defining EV_TSTAMP_T.
66 - cleanup: replace expect_true/false and noinline by their
67 libecb counterparts.
68 - move syscall infrastructure from ev_linuxaio.c to ev.c.
69 - prepare io_uring integration.
70 - tweak ev_floor.
71 - epoll, poll, win32 Sleep and other places that use millisecond
72 reslution now all try to round up times.
73 - solaris port backend didn't compile.
74 - abstract time constants into their macros, for more flexibility.
75
764.27 Thu Jun 27 22:43:44 CEST 2019
77 - linux aio backend almost completely rewritten to work around its
78 limitations.
79 - linux aio backend now requires linux 4.19+.
80 - epoll backend now mandatory for linux aio backend.
81 - fail assertions more aggressively on invalid fd's detected
82 in the event loop, do not just silently fd_kill in case of
83 user error.
84 - ev_io_start/ev_io_stop now verify the watcher fd using
85 a syscall when EV_VERIFY is 2 or higher.
86
874.26 (EV only)
88 - update to libecb 0x00010006.
89 - new experimental linux aio backend (linux 4.18+).
90 - removed redundant 0-ptr check in ev_once.
91 - updated/extended ev_set_allocator documentation.
92 - replaced EMPTY2 macro by array_needsize_noinit.
93 - minor code cleanups.
94 - epoll backend now uses epoll_create1 also after fork.
95
964.25 Fri Dec 21 07:49:20 CET 2018
97 - INCOMPATIBLE CHANGE: EV_THROW was renamed to EV_NOEXCEPT
98 (EV_THROW still provided) and now uses noexcept on C++11 or newer.
99 - move the darwin select workaround higher in ev.c, as newer versions of
100 darwin managed to break their broken select even more.
101 - ANDROID => __ANDROID__ (reported by enh@google.com).
102 - disable epoll_create1 on android because it has broken header files
103 and google is unwilling to fix them (reported by enh@google.com).
104 - avoid a minor compilation warning on win32.
105 - c++: remove deprecated dynamic throw() specifications.
106 - c++: improve the (unsupported) bad_loop exception class.
107 - backport perl ev_periodic example to C, untested.
108 - update libecb, biggets change is to include a memory fence
109 in ECB_MEMORY_FENCE_RELEASE on x86/amd64.
110 - minor autoconf/automake modernisation.
111
1124.24 Wed Dec 28 05:19:55 CET 2016
113 - bump version to 4.24, as the release tarball inexplicably
114 didn't have the right version in ev.h, even though the cvs-tagged
115 version did have the right one (reported by Ales Teska).
116
1174.23 Wed Nov 16 18:23:41 CET 2016
118 - move some declarations at the beginning to help certain retarded
119 microsoft compilers, even though their documentation claims
120 otherwise (reported by Ruslan Osmanov).
121
1224.22 Sun Dec 20 22:11:50 CET 2015
123 - when epoll detects unremovable fds in the fd set, rebuild
124 only the epoll descriptor, not the signal pipe, to avoid
125 SIGPIPE in ev_async_send. This doesn't solve it on fork,
126 so document what needs to be done in ev_loop_fork
127 (analyzed by Benjamin Mahler).
128 - remove superfluous sys/timeb.h include on win32
129 (analyzed by Jason Madden).
130 - updated libecb.
131
1324.20 Sat Jun 20 13:01:43 CEST 2015
133 - prefer noexcept over throw () with C++ 11.
134 - update ecb.h due to incompatibilities with c11.
135 - fix a potential aliasing issue when reading and writing
136 watcher callbacks.
137
1384.19 Thu Sep 25 08:18:25 CEST 2014
139 - ev.h wasn't valid C++ anymore, which tripped compilers other than
140 clang, msvc or gcc (analyzed by Raphael 'kena' Poss). Unfortunately,
141 C++ doesn't support typedefs for function pointers fully, so the affected
142 declarations have to spell out the types each time.
143 - when not using autoconf, tighten the check for clock_gettime and related
144 functionality.
145
1464.18 Fri Sep 5 17:55:26 CEST 2014
147 - events on files were not always generated properly with the
148 epoll backend (testcase by Assaf Inbal).
149 - mark event pipe fd as cloexec after a fork (analyzed by Sami Farin).
150 - (ecb) support m68k, m88k and sh (patch by Miod Vallat).
151 - use a reasonable fallback for EV_NSIG instead of erroring out
152 when we can't detect the signal set size.
153 - in the absence of autoconf, do not use the clock syscall
154 on glibc >= 2.17 (avoids the syscall AND -lrt on systems
155 doing clock_gettime in userspace).
156 - ensure extern "C" function pointers are used for externally-visible
157 loop callbacks (not watcher callbacks yet).
158 - (ecb) work around memory barriers and volatile apparently both being
159 broken in visual studio 2008 and later (analysed and patch by Nicolas Noble).
160
1614.15 Fri Mar 1 12:04:50 CET 2013
162 - destroying a non-default loop would stop the global waitpid
163 watcher (Denis Bilenko).
164 - queueing pending watchers of higher priority from a watcher now invokes
165 them in a timely fashion (reported by Denis Bilenko).
166 - add throw() to all libev functions that cannot throw exceptions, for
167 further code size decrease when compiling for C++.
168 - add throw () to callbacks that must not throw exceptions (allocator,
169 syserr, loop acquire/release, periodic reschedule cbs).
170 - fix event_base_loop return code, add event_get_callback, event_base_new,
171 event_base_get_method calls to improve libevent 1.x emulation and add
172 some libevent 2.x functionality (based on a patch by Jeff Davey).
173 - add more memory fences to fix a bug reported by Jeff Davey. Better
174 be overfenced than underprotected.
175 - ev_run now returns a boolean status (true meaning watchers are
176 still active).
177 - ev_once: undef EV_ERROR in ev_kqueue.c, to avoid clashing with
178 libev's EV_ERROR (reported by 191919).
179 - (ecb) add memory fence support for xlC (Darin McBride).
180 - (ecb) add memory fence support for gcc-mips (Anton Kirilov).
181 - (ecb) add memory fence support for gcc-alpha (Christian Weisgerber).
182 - work around some kernels losing file descriptors by leaking
183 the kqueue descriptor in the child.
184 - work around linux inotify not reporting IN_ATTRIB changes for directories
185 in many cases.
186 - include sys/syscall.h instead of plain syscall.h.
187 - check for io watcher loops in ev_verify, check for the most
188 common reported usage bug in ev_io_start.
189 - choose socket vs. WSASocket at compiletime using EV_USE_WSASOCKET.
190 - always use WSASend/WSARecv directly on windows, hoping that this
191 works in all cases (unlike read/write/send/recv...).
192 - try to detect signals around a fork faster (test program by
193 Denis Bilenko).
194 - work around recent glibc versions that leak memory in realloc.
195 - rename ev::embed::set to ev::embed::set_embed to avoid clashing
196 the watcher base set (loop) method.
197 - rewrite the async/signal pipe logic to always keep a valid fd, which
198 simplifies (and hopefully correctifies :) the race checking
199 on fork, at the cost of one extra fd.
200 - add fat, msdos, jffs2, ramfs, ntfs and btrfs to the list of
201 inotify-supporting filesystems.
202 - move orig_CFLAGS assignment to after AC_INIT, as newer autoconf
203 versions ignore it before
204 (https://bugzilla.redhat.com/show_bug.cgi?id=908096).
205 - add some untested android support.
206 - enum expressions must be of type int (reported by Juan Pablo L).
207
2084.11 Sat Feb 4 19:52:39 CET 2012
209 - INCOMPATIBLE CHANGE: ev_timer_again now clears the pending status, as
210 was documented already, but not implemented in the repeating case.
211 - new compiletime symbols: EV_NO_SMP and EV_NO_THREADS.
212 - fix a race where the workaround against the epoll fork bugs
213 caused signals to not be handled anymore.
214 - correct backend_fudge for most backends, and implement a windows
215 specific workaround to avoid looping because we call both
216 select and Sleep, both with different time resolutions.
217 - document range and guarantees of ev_sleep.
218 - document reasonable ranges for periodics interval and offset.
219 - rename backend_fudge to backend_mintime to avoid future confusion :)
220 - change the default periodic reschedule function to hopefully be more
221 exact and correct even in corner cases or in the far future.
222 - do not rely on -lm anymore: use it when available but use our
223 own floor () if it is missing. This should make it easier to embed,
224 as no external libraries are required.
225 - strategically import macros from libecb and mark rarely-used functions
226 as cache-cold (saving almost 2k code size on typical amd64 setups).
227 - add Symbols.ev and Symbols.event files, that were missing.
228 - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now).
229 - fix #3 "be smart about timeouts" to not "deadlock" when
230 timeout == now, also improve the section overall.
231 - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom.
232 - support new EV_API_STATIC mode to make all libev symbols
233 static.
234 - supply default CFLAGS of -g -O3 with gcc when original CFLAGS
235 were empty.
236
2374.04 Wed Feb 16 09:01:51 CET 2011
238 - fix two problems in the native win32 backend, where reuse of fd's
239 with different underlying handles caused handles not to be removed
240 or added to the select set (analyzed and tested by Bert Belder).
241 - do no rely on ceil() in ev_e?poll.c.
242 - backport libev to HP-UX versions before 11 v3.
243 - configure did not detect nanosleep and clock_gettime properly when
244 they are available in the libc (as opposed to -lrt).
245
2464.03 Tue Jan 11 14:37:25 CET 2011
247 - officially support polling files with all backends.
248 - support files, /dev/zero etc. the same way as select in the epoll
249 backend, by generating events on our own.
250 - ports backend: work around solaris bug 6874410 and many related ones
251 (EINTR, maybe more), with no performance loss (note that the solaris
252 bug report is actually wrong, reality is far more bizarre and broken
253 than that).
254 - define EV_READ/EV_WRITE as macros in event.h, as some programs use
255 #ifdef to test for them.
256 - new (experimental) function: ev_feed_signal.
257 - new (to become default) EVFLAG_NOSIGMASK flag.
258 - new EVBACKEND_MASK symbol.
259 - updated COMMON IDIOMS SECTION.
260
2614.01 Fri Nov 5 21:51:29 CET 2010
262 - automake fucked it up, apparently, --add-missing -f is not quite enough
263 to make it update its files, so 4.00 didn't install ev++.h and
264 event.h on make install. grrr.
265 - ev_loop(count|depth) didn't return anything (Robin Haberkorn).
266 - change EV_UNDEF to 0xffffffff to silence some overzealous compilers.
267 - use "(libev) " prefix for all libev error messages now.
268
2694.00 Mon Oct 25 12:32:12 CEST 2010
270 - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading.
271 - ev_embed_stop did not correctly stop the watcher (very good
272 testcase by Vladimir Timofeev).
273 - ev_run will now always update the current loop time - it erroneously
274 didn't when idle watchers were active, causing timers not to fire.
275 - fix a bug where a timeout of zero caused the timer not to fire
276 in the libevent emulation (testcase by Péter Szabó).
277 - applied win32 fixes by Michael Lenaghan (also James Mansion).
278 - replace EV_MINIMAL by EV_FEATURES.
279 - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it
280 seems the former is *much* faster than the latter.
281 - linux kernel version detection (for inotify bug workarounds)
282 did not work properly.
283 - reduce the number of spurious wake-ups with the ports backend.
284 - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu).
285 - do async init within ev_async_start, not ev_async_set, which avoids
286 an API quirk where the set function must be called in the C++ API
287 even when there is nothing to set.
288 - add (undocumented) EV_ENABLE when adding events with kqueue,
289 this might help with OS X, which seems to need it despite documenting
290 not to need it (helpfully pointed out by Tilghman Lesher).
291 - do not use poll by default on freebsd, it's broken (what isn't
292 on freebsd...).
293 - allow to embed epoll on kernels >= 2.6.32.
294 - configure now prepends -O3, not appends it, so one can still
295 override it.
296 - ev.pod: greatly expanded the portability section, added a porting
297 section, a description of watcher states and made lots of minor fixes.
298 - disable poll backend on AIX, the poll header spams the namespace
299 and it's not worth working around dead platforms (reported
300 and analyzed by Aivars Kalvans).
301 - improve header file compatibility of the standalone eventfd code
302 in an obscure case.
303 - implement EV_AVOID_STDIO option.
304 - do not use sscanf to parse linux version number (smaller, faster,
305 no sscanf dependency).
306 - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings.
307 - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs.
308 - add section on accept() problems to the manpage.
309 - rename EV_TIMEOUT to EV_TIMER.
310 - rename ev_loop_count/depth/verify/loop/unloop.
311 - remove ev_default_destroy and ev_default_fork.
312 - switch to two-digit minor version.
313 - work around an apparent gentoo compiler bug.
314 - define _DARWIN_UNLIMITED_SELECT. just so.
315 - use enum instead of #define for most constants.
316 - improve compatibility to older C++ compilers.
317 - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now
318 default arguments when compiled as C++.
319 - enable automake dependency tracking.
320 - ev_loop_new no longer leaks memory when loop creation failed.
321 - new ev_cleanup watcher type.
322
3233.9 Thu Dec 31 07:59:59 CET 2009
324 - signalfd is no longer used by default and has to be requested
325 explicitly - this means that easy to catch bugs become hard to
326 catch race conditions, but the users have spoken.
327 - point out the unspecified signal mask in the documentation, and
328 that this is a race condition regardless of EV_SIGNALFD.
329 - backport inotify code to C89.
330 - inotify file descriptors could leak into child processes.
331 - ev_stat watchers could keep an erroneous extra ref on the loop,
332 preventing exit when unregistering all watchers (testcases
333 provided by ry@tinyclouds.org).
334 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
335 symbols to make it easier for apps to do their own fd management.
336 - support EV_IDLE_ENABLE being disabled in ev++.h
337 (patch by Didier Spezia).
338 - take advantage of inotify_init1, if available, to set cloexec/nonblock
339 on fd creation, to avoid races.
340 - the signal handling pipe wasn't always initialised under windows
341 (analysed by lekma).
342 - changed minimum glibc requirement from glibc 2.9 to 2.7, for
343 signalfd.
344 - add missing string.h include (Denis F. Latypoff).
345 - only replace ev_stat.prev when we detect an actual difference,
346 so prev is (almost) always different to attr. this might
347 have caused the problems with 04_stat.t.
348 - add ev::timer->remaining () method to C++ API.
349
3503.8 Sun Aug 9 14:30:45 CEST 2009
351 - incompatible change: do not necessarily reset signal handler
352 to SIG_DFL when a sighandler is stopped.
353 - ev_default_destroy did not properly free or zero some members,
354 potentially causing crashes and memory corruption on repeated
355 ev_default_destroy/ev_default_loop calls.
356 - take advantage of signalfd on GNU/Linux systems.
357 - document that the signal mask might be in an unspecified
358 state when using libev's signal handling.
359 - take advantage of some GNU/Linux calls to set cloexec/nonblock
360 on fd creation, to avoid race conditions.
361
3623.7 Fri Jul 17 16:36:32 CEST 2009
363 - ev_unloop and ev_loop wrongly used a global variable to exit loops,
364 instead of using a per-loop variable (bug caught by accident...).
365 - the ev_set_io_collect_interval interpretation has changed.
366 - add new functionality: ev_set_userdata, ev_userdata,
367 ev_set_invoke_pending_cb, ev_set_loop_release_cb,
368 ev_invoke_pending, ev_pending_count, together with a long example
369 about thread locking.
370 - add ev_timer_remaining (as requested by Denis F. Latypoff).
371 - add ev_loop_depth.
372 - calling ev_unloop in fork/prepare watchers will no longer poll
373 for new events.
374 - Denis F. Latypoff corrected many typos in example code snippets.
375 - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double-
376 check that the syscall number is available before trying to
377 use it (reported by ry@tinyclouds).
378 - use GetSystemTimeAsFileTime instead of _timeb on windows, for
379 slightly higher accuracy.
380 - properly declare ev_loop_verify and ev_now_update even when
381 !EV_MULTIPLICITY.
382 - do not compile in any priority code when EV_MAXPRI == EV_MINPRI.
383 - support EV_MINIMAL==2 for a reduced API.
384 - actually 0-initialise struct sigaction when installing signals.
385 - add section on hibernate and stopped processes to ev_timer docs.
386
3873.6 Tue Apr 28 02:49:30 CEST 2009
388 - multiple timers becoming ready within an event loop iteration
389 will be invoked in the "correct" order now.
390 - do not leave the event loop early just because we have no active
391 watchers, fixing a problem when embedding a kqueue loop
392 that has active kernel events but no registered watchers
393 (reported by blacksand blacksand).
394 - correctly zero the idx values for arrays, so destroying and
395 reinitialising the default loop actually works (patch by
396 Malek Hadj-Ali).
397 - implement ev_suspend and ev_resume.
398 - new EV_CUSTOM revents flag for use by applications.
399 - add documentation section about priorities.
400 - add a glossary to the documentation.
401 - extend the ev_fork description slightly.
402 - optimize a jump out of call_pending.
403
4043.53 Sun Feb 15 02:38:20 CET 2009
405 - fix a bug in event pipe creation on win32 that would cause a
406 failed assertion on event loop creation (patch by Malek Hadj-Ali).
407 - probe for CLOCK_REALTIME support at runtime as well and fall
408 back to gettimeofday if there is an error, to support older
409 operating systems with newer header files/libraries.
410 - prefer gettimeofday over clock_gettime with USE_CLOCK_SYSCALL
411 (default most everywhere), otherwise not.
412
4133.52 Wed Jan 7 21:43:02 CET 2009
414 - fix compilation of select backend in fd_set mode when NFDBITS is
415 missing (to get it to compile on QNX, reported by Rodrigo Campos).
416 - better select-nfds handling when select backend is in fd_set mode.
417 - diagnose fd_set overruns when select backend is in fd_set mode.
418 - due to a thinko, instead of disabling everything but
419 select on the borked OS X platform, everything but select was
420 allowed (reported by Emanuele Giaquinta).
421 - actually verify that local and remote port are matching in
422 libev's socketpair emulation, which makes denial-of-service
423 attacks harder (but not impossible - it's windows). Make sure
424 it even works under vista, which thinks that getpeer/sockname
425 should return fantasy port numbers.
426 - include "libev" in all assertion messages for potentially
427 clearer diagnostics.
428 - event_get_version (libevent compatibility) returned
429 a useless string instead of the expected version string
430 (patch by W.C.A. Wijngaards).
431
4323.51 Wed Dec 24 23:00:11 CET 2008
433 - fix a bug where an inotify watcher was added twice, causing
434 freezes on hash collisions (reported and analysed by Graham Leggett).
435 - new config symbol, EV_USE_CLOCK_SYSCALL, to make libev use
436 a direct syscall - slower, but no dependency on librt et al.
437 - assume negative return values != -1 signals success of port_getn
438 (http://cvs.epicsol.org/cgi/viewcvs.cgi/epic5/source/newio.c?rev=1.52)
439 (no known failure reports, but it doesn't hurt).
440 - fork detection in ev_embed now stops and restarts the watcher
441 automatically.
442 - EXPERIMENTAL: default the method to operator () in ev++.h,
443 to make it nicer to use functors (requested by Benedek László).
444 - fixed const object callbacks in ev++.h.
445 - replaced loop_ref argument of watcher.set (loop) by a direct
446 ev_loop * in ev++.h, to avoid clashes with functor patch.
447 - do not try to watch the empty string via inotify.
448 - inotify watchers could be leaked under certain circumstances.
449 - OS X 10.5 is actually even more broken than earlier versions,
450 so fall back to select on that piece of garbage.
451 - fixed some weirdness in the ev_embed documentation.
452
4533.49 Wed Nov 19 11:26:53 CET 2008
454 - ev_stat watchers will now use inotify as a mere hint on
455 kernels <2.6.25, or if the filesystem is not in the
456 "known to be good" list.
457 - better mingw32 compatibility (it's not as borked as native win32)
458 (analysed by Roger Pack).
459 - include stdio.h in the example program, as too many people are
460 confused by the weird C language otherwise. I guess the next thing
461 I get told is that the "..." ellipses in the examples don't compile
462 with their C compiler.
463
4643.48 Thu Oct 30 09:02:37 CET 2008
465 - further optimise away the EPOLL_CTL_ADD/MOD combo in the epoll
466 backend by assuming the kernel event mask hasn't changed if
467 ADD fails with EEXIST.
468 - work around spurious event notification bugs in epoll by using
469 a 32-bit generation counter. recreate kernel state if we receive
470 spurious notifications or unwanted events. this is very costly,
471 but I didn't come up with this horrible design.
472 - use memset to initialise most arrays now and do away with the
473 init functions.
474 - expand time-out strategies into a "Be smart about timeouts" section.
475 - drop the "struct" from all ev_watcher declarations in the
476 documentation and did other clarifications (yeah, it was a mistake
477 to have a struct AND a function called ev_loop).
478 - fix a bug where ev_default would not initialise the default
479 loop again after it was destroyed with ev_default_destroy.
480 - rename syserr to ev_syserr to avoid name clashes when embedding,
481 do similar changes for event.c.
482
4833.45 Tue Oct 21 21:59:26 CEST 2008
484 - disable inotify usage on linux <2.6.25, as it is broken
485 (reported by Yoann Vandoorselaere).
486 - ev_stat erroneously would try to add inotify watchers
487 even when inotify wasn't available (this should only
488 have a performance impact).
489 - ev_once now passes both timeout and io to the callback if both
490 occur concurrently, instead of giving timeouts precedence.
491 - disable EV_USE_INOTIFY when sys/inotify.h is too old.
492
53.44 Mon Sep 29 05:18:39 CEST 2008 4933.44 Mon Sep 29 05:18:39 CEST 2008
6 - embed watchers now automatically invoke ev_loop_fork on the 494 - embed watchers now automatically invoke ev_loop_fork on the
7 embedded loop when the parent loop forks. 495 embedded loop when the parent loop forks.
8 - new function: ev_now_update (loop). 496 - new function: ev_now_update (loop).
9 - verify_watcher was not marked static. 497 - verify_watcher was not marked static.
… …
71 - enable inlining with gcc versions 2 and 3. 559 - enable inlining with gcc versions 2 and 3.
72 - work around broken poll implementations potentially 560 - work around broken poll implementations potentially
73 not clearing revents field in ev_poll (Brandon Black) 561 not clearing revents field in ev_poll (Brandon Black)
74 (no such systems are known at this time). 562 (no such systems are known at this time).
75 - work around a bug in realloc on openbsd and darwin, 563 - work around a bug in realloc on openbsd and darwin,
76 also makes the errornous valgrind complaints 564 also makes the erroneous valgrind complaints
77 go away (noted by various people). 565 go away (noted by various people).
78 - fix ev_async_pending, add c++ wrapper for ev_async 566 - fix ev_async_pending, add c++ wrapper for ev_async
79 (based on patch sent by Johannes Deisenhofer. 567 (based on patch sent by Johannes Deisenhofer).
80 - add sensible set method to ev::embed. 568 - add sensible set method to ev::embed.
81 - made integer constants type int in ev.h. 569 - made integer constants type int in ev.h.
82 570
833.2 Wed Apr 2 17:11:19 CEST 2008 5713.2 Wed Apr 2 17:11:19 CEST 2008
84 - fix a 64 bit overflow issue in the select backend, 572 - fix a 64 bit overflow issue in the select backend,

Diff Legend

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