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

Comparing libev/Changes (file contents):
Revision 1.138 by root, Sun Jul 19 07:20:41 2009 UTC vs.
Revision 1.367 by root, Wed Mar 18 12:30:20 2020 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
3TODO: ev_walk 3TODO: for next ABI/API change, consider moving EV__IOFDSSET into io->fd instead and provide a getter.
4TODO: signal handling per loop 4TODO: document EV_TSTAMP_T
5TODO: NSIG etc. 5
6TODO: NOSIGFD/NOEVTFD/NOINOTIFY? 64.33 Wed Mar 18 13:22:29 CET 2020
7TODO: gotsig per-loop 7 - no changes w.r.t. 4.32.
8
94.32 (EV only)
10 - the 4.31 timerfd code wrongly changed the priority of the signal
11 fd watcher, which is usually harmless unless signal fds are
12 also used (found via cpan tester service).
13 - the documentation wrongly claimed that user may modify fd and events
14 members in io watchers when the watcher was stopped
15 (found by b_jonas).
16 - new ev_io_modify mutator which changes only the events member,
17 which can be faster. also added ev::io::set (int events) method
18 to ev++.h.
19 - officially allow a zero events mask for io watchers. this should
20 work with older libev versions as well but was not officially
21 allowed before.
22 - do not wake up every minute when timerfd is used to detect timejumps.
23 - do not wake up every minute when periodics are disabled and we have
24 a monotonic clock.
25 - support a lot more "uncommon" compile time configurations,
26 such as ev_embed enabled but ev_timer disabled.
27 - use a start/stop wrapper class to reduce code duplication in
28 ev++.h and make it needlessly more c++-y.
29 - the linux aio backend is no longer compiled in by default.
30 - update to libecb version 0x00010008.
31
324.31 Fri Dec 20 21:58:29 CET 2019
33 - handle backends with minimum wait time a bit better by not
34 waiting in the presence of already-expired timers
35 (behaviour reported by Felipe Gasper).
36 - new feature: use timerfd to detect timejumps quickly,
37 can be disabled with the new EVFLAG_NOTIMERFD loop flag.
38 - document EV_USE_SIGNALFD feature macro.
39
404.30 (EV only)
41 - change non-autoconf test for __kernel_rwf_t by testing
42 LINUX_VERSION_CODE, the most direct test I could find.
43 - fix a bug in the io_uring backend that polled the wrong
44 backend fd, causing it to not work in many cases.
45
464.29 (EV only)
47 - add io uring autoconf and non-autoconf detection.
48 - disable io_uring when some header files are too old.
49
504.28 (EV only)
51 - linuxaio backend resulted in random memory corruption
52 when loop is forked.
53 - linuxaio backend might have tried to cancel an iocb
54 multiple times (was unable to trigger this).
55 - linuxaio backend now employs a generation counter to
56 avoid handling spurious events from cancelled requests.
57 - io_cancel can return EINTR, deal with it. also, assume
58 io_submit also returns EINTR.
59 - fix some other minor bugs in linuxaio backend.
60 - ev_tstamp type can now be overriden by defining EV_TSTAMP_T.
61 - cleanup: replace expect_true/false and noinline by their
62 libecb counterparts.
63 - move syscall infrastructure from ev_linuxaio.c to ev.c.
64 - prepare io_uring integration.
65 - tweak ev_floor.
66 - epoll, poll, win32 Sleep and other places that use millisecond
67 reslution now all try to round up times.
68 - solaris port backend didn't compile.
69 - abstract time constants into their macros, for more flexibility.
70
714.27 Thu Jun 27 22:43:44 CEST 2019
72 - linux aio backend almost completely rewritten to work around its
73 limitations.
74 - linux aio backend now requires linux 4.19+.
75 - epoll backend now mandatory for linux aio backend.
76 - fail assertions more aggressively on invalid fd's detected
77 in the event loop, do not just silently fd_kill in case of
78 user error.
79 - ev_io_start/ev_io_stop now verify the watcher fd using
80 a syscall when EV_VERIFY is 2 or higher.
81
824.26 (EV only)
83 - update to libecb 0x00010006.
84 - new experimental linux aio backend (linux 4.18+).
85 - removed redundant 0-ptr check in ev_once.
86 - updated/extended ev_set_allocator documentation.
87 - replaced EMPTY2 macro by array_needsize_noinit.
88 - minor code cleanups.
89 - epoll backend now uses epoll_create1 also after fork.
90
914.25 Fri Dec 21 07:49:20 CET 2018
92 - INCOMPATIBLE CHANGE: EV_THROW was renamed to EV_NOEXCEPT
93 (EV_THROW still provided) and now uses noexcept on C++11 or newer.
94 - move the darwin select workaround higher in ev.c, as newer versions of
95 darwin managed to break their broken select even more.
96 - ANDROID => __ANDROID__ (reported by enh@google.com).
97 - disable epoll_create1 on android because it has broken header files
98 and google is unwilling to fix them (reported by enh@google.com).
99 - avoid a minor compilation warning on win32.
100 - c++: remove deprecated dynamic throw() specifications.
101 - c++: improve the (unsupported) bad_loop exception class.
102 - backport perl ev_periodic example to C, untested.
103 - update libecb, biggets change is to include a memory fence
104 in ECB_MEMORY_FENCE_RELEASE on x86/amd64.
105 - minor autoconf/automake modernisation.
106
1074.24 Wed Dec 28 05:19:55 CET 2016
108 - bump version to 4.24, as the release tarball inexplicably
109 didn't have the right version in ev.h, even though the cvs-tagged
110 version did have the right one (reported by Ales Teska).
111
1124.23 Wed Nov 16 18:23:41 CET 2016
113 - move some declarations at the beginning to help certain retarded
114 microsoft compilers, even though their documentation claims
115 otherwise (reported by Ruslan Osmanov).
116
1174.22 Sun Dec 20 22:11:50 CET 2015
118 - when epoll detects unremovable fds in the fd set, rebuild
119 only the epoll descriptor, not the signal pipe, to avoid
120 SIGPIPE in ev_async_send. This doesn't solve it on fork,
121 so document what needs to be done in ev_loop_fork
122 (analyzed by Benjamin Mahler).
123 - remove superfluous sys/timeb.h include on win32
124 (analyzed by Jason Madden).
125 - updated libecb.
126
1274.20 Sat Jun 20 13:01:43 CEST 2015
128 - prefer noexcept over throw () with C++ 11.
129 - update ecb.h due to incompatibilities with c11.
130 - fix a potential aliasing issue when reading and writing
131 watcher callbacks.
132
1334.19 Thu Sep 25 08:18:25 CEST 2014
134 - ev.h wasn't valid C++ anymore, which tripped compilers other than
135 clang, msvc or gcc (analyzed by Raphael 'kena' Poss). Unfortunately,
136 C++ doesn't support typedefs for function pointers fully, so the affected
137 declarations have to spell out the types each time.
138 - when not using autoconf, tighten the check for clock_gettime and related
139 functionality.
140
1414.18 Fri Sep 5 17:55:26 CEST 2014
142 - events on files were not always generated properly with the
143 epoll backend (testcase by Assaf Inbal).
144 - mark event pipe fd as cloexec after a fork (analyzed by Sami Farin).
145 - (ecb) support m68k, m88k and sh (patch by Miod Vallat).
146 - use a reasonable fallback for EV_NSIG instead of erroring out
147 when we can't detect the signal set size.
148 - in the absence of autoconf, do not use the clock syscall
149 on glibc >= 2.17 (avoids the syscall AND -lrt on systems
150 doing clock_gettime in userspace).
151 - ensure extern "C" function pointers are used for externally-visible
152 loop callbacks (not watcher callbacks yet).
153 - (ecb) work around memory barriers and volatile apparently both being
154 broken in visual studio 2008 and later (analysed and patch by Nicolas Noble).
155
1564.15 Fri Mar 1 12:04:50 CET 2013
157 - destroying a non-default loop would stop the global waitpid
158 watcher (Denis Bilenko).
159 - queueing pending watchers of higher priority from a watcher now invokes
160 them in a timely fashion (reported by Denis Bilenko).
161 - add throw() to all libev functions that cannot throw exceptions, for
162 further code size decrease when compiling for C++.
163 - add throw () to callbacks that must not throw exceptions (allocator,
164 syserr, loop acquire/release, periodic reschedule cbs).
165 - fix event_base_loop return code, add event_get_callback, event_base_new,
166 event_base_get_method calls to improve libevent 1.x emulation and add
167 some libevent 2.x functionality (based on a patch by Jeff Davey).
168 - add more memory fences to fix a bug reported by Jeff Davey. Better
169 be overfenced than underprotected.
170 - ev_run now returns a boolean status (true meaning watchers are
171 still active).
172 - ev_once: undef EV_ERROR in ev_kqueue.c, to avoid clashing with
173 libev's EV_ERROR (reported by 191919).
174 - (ecb) add memory fence support for xlC (Darin McBride).
175 - (ecb) add memory fence support for gcc-mips (Anton Kirilov).
176 - (ecb) add memory fence support for gcc-alpha (Christian Weisgerber).
177 - work around some kernels losing file descriptors by leaking
178 the kqueue descriptor in the child.
179 - work around linux inotify not reporting IN_ATTRIB changes for directories
180 in many cases.
181 - include sys/syscall.h instead of plain syscall.h.
182 - check for io watcher loops in ev_verify, check for the most
183 common reported usage bug in ev_io_start.
184 - choose socket vs. WSASocket at compiletime using EV_USE_WSASOCKET.
185 - always use WSASend/WSARecv directly on windows, hoping that this
186 works in all cases (unlike read/write/send/recv...).
187 - try to detect signals around a fork faster (test program by
188 Denis Bilenko).
189 - work around recent glibc versions that leak memory in realloc.
190 - rename ev::embed::set to ev::embed::set_embed to avoid clashing
191 the watcher base set (loop) method.
192 - rewrite the async/signal pipe logic to always keep a valid fd, which
193 simplifies (and hopefully correctifies :) the race checking
194 on fork, at the cost of one extra fd.
195 - add fat, msdos, jffs2, ramfs, ntfs and btrfs to the list of
196 inotify-supporting filesystems.
197 - move orig_CFLAGS assignment to after AC_INIT, as newer autoconf
198 versions ignore it before
199 (https://bugzilla.redhat.com/show_bug.cgi?id=908096).
200 - add some untested android support.
201 - enum expressions must be of type int (reported by Juan Pablo L).
202
2034.11 Sat Feb 4 19:52:39 CET 2012
204 - INCOMPATIBLE CHANGE: ev_timer_again now clears the pending status, as
205 was documented already, but not implemented in the repeating case.
206 - new compiletime symbols: EV_NO_SMP and EV_NO_THREADS.
207 - fix a race where the workaround against the epoll fork bugs
208 caused signals to not be handled anymore.
209 - correct backend_fudge for most backends, and implement a windows
210 specific workaround to avoid looping because we call both
211 select and Sleep, both with different time resolutions.
212 - document range and guarantees of ev_sleep.
213 - document reasonable ranges for periodics interval and offset.
214 - rename backend_fudge to backend_mintime to avoid future confusion :)
215 - change the default periodic reschedule function to hopefully be more
216 exact and correct even in corner cases or in the far future.
217 - do not rely on -lm anymore: use it when available but use our
218 own floor () if it is missing. This should make it easier to embed,
219 as no external libraries are required.
220 - strategically import macros from libecb and mark rarely-used functions
221 as cache-cold (saving almost 2k code size on typical amd64 setups).
222 - add Symbols.ev and Symbols.event files, that were missing.
223 - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now).
224 - fix #3 "be smart about timeouts" to not "deadlock" when
225 timeout == now, also improve the section overall.
226 - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom.
227 - support new EV_API_STATIC mode to make all libev symbols
228 static.
229 - supply default CFLAGS of -g -O3 with gcc when original CFLAGS
230 were empty.
231
2324.04 Wed Feb 16 09:01:51 CET 2011
233 - fix two problems in the native win32 backend, where reuse of fd's
234 with different underlying handles caused handles not to be removed
235 or added to the select set (analyzed and tested by Bert Belder).
236 - do no rely on ceil() in ev_e?poll.c.
237 - backport libev to HP-UX versions before 11 v3.
238 - configure did not detect nanosleep and clock_gettime properly when
239 they are available in the libc (as opposed to -lrt).
240
2414.03 Tue Jan 11 14:37:25 CET 2011
242 - officially support polling files with all backends.
243 - support files, /dev/zero etc. the same way as select in the epoll
244 backend, by generating events on our own.
245 - ports backend: work around solaris bug 6874410 and many related ones
246 (EINTR, maybe more), with no performance loss (note that the solaris
247 bug report is actually wrong, reality is far more bizarre and broken
248 than that).
249 - define EV_READ/EV_WRITE as macros in event.h, as some programs use
250 #ifdef to test for them.
251 - new (experimental) function: ev_feed_signal.
252 - new (to become default) EVFLAG_NOSIGMASK flag.
253 - new EVBACKEND_MASK symbol.
254 - updated COMMON IDIOMS SECTION.
255
2564.01 Fri Nov 5 21:51:29 CET 2010
257 - automake fucked it up, apparently, --add-missing -f is not quite enough
258 to make it update its files, so 4.00 didn't install ev++.h and
259 event.h on make install. grrr.
260 - ev_loop(count|depth) didn't return anything (Robin Haberkorn).
261 - change EV_UNDEF to 0xffffffff to silence some overzealous compilers.
262 - use "(libev) " prefix for all libev error messages now.
263
2644.00 Mon Oct 25 12:32:12 CEST 2010
265 - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading.
266 - ev_embed_stop did not correctly stop the watcher (very good
267 testcase by Vladimir Timofeev).
268 - ev_run will now always update the current loop time - it erroneously
269 didn't when idle watchers were active, causing timers not to fire.
270 - fix a bug where a timeout of zero caused the timer not to fire
271 in the libevent emulation (testcase by Péter Szabó).
272 - applied win32 fixes by Michael Lenaghan (also James Mansion).
273 - replace EV_MINIMAL by EV_FEATURES.
274 - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it
275 seems the former is *much* faster than the latter.
276 - linux kernel version detection (for inotify bug workarounds)
277 did not work properly.
278 - reduce the number of spurious wake-ups with the ports backend.
279 - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu).
280 - do async init within ev_async_start, not ev_async_set, which avoids
281 an API quirk where the set function must be called in the C++ API
282 even when there is nothing to set.
283 - add (undocumented) EV_ENABLE when adding events with kqueue,
284 this might help with OS X, which seems to need it despite documenting
285 not to need it (helpfully pointed out by Tilghman Lesher).
286 - do not use poll by default on freebsd, it's broken (what isn't
287 on freebsd...).
288 - allow to embed epoll on kernels >= 2.6.32.
289 - configure now prepends -O3, not appends it, so one can still
290 override it.
291 - ev.pod: greatly expanded the portability section, added a porting
292 section, a description of watcher states and made lots of minor fixes.
293 - disable poll backend on AIX, the poll header spams the namespace
294 and it's not worth working around dead platforms (reported
295 and analyzed by Aivars Kalvans).
296 - improve header file compatibility of the standalone eventfd code
297 in an obscure case.
298 - implement EV_AVOID_STDIO option.
299 - do not use sscanf to parse linux version number (smaller, faster,
300 no sscanf dependency).
301 - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings.
302 - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs.
303 - add section on accept() problems to the manpage.
304 - rename EV_TIMEOUT to EV_TIMER.
305 - rename ev_loop_count/depth/verify/loop/unloop.
306 - remove ev_default_destroy and ev_default_fork.
307 - switch to two-digit minor version.
308 - work around an apparent gentoo compiler bug.
309 - define _DARWIN_UNLIMITED_SELECT. just so.
310 - use enum instead of #define for most constants.
311 - improve compatibility to older C++ compilers.
312 - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now
313 default arguments when compiled as C++.
314 - enable automake dependency tracking.
315 - ev_loop_new no longer leaks memory when loop creation failed.
316 - new ev_cleanup watcher type.
317
3183.9 Thu Dec 31 07:59:59 CET 2009
319 - signalfd is no longer used by default and has to be requested
320 explicitly - this means that easy to catch bugs become hard to
321 catch race conditions, but the users have spoken.
322 - point out the unspecified signal mask in the documentation, and
323 that this is a race condition regardless of EV_SIGNALFD.
324 - backport inotify code to C89.
325 - inotify file descriptors could leak into child processes.
326 - ev_stat watchers could keep an erroneous extra ref on the loop,
327 preventing exit when unregistering all watchers (testcases
328 provided by ry@tinyclouds.org).
329 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
330 symbols to make it easier for apps to do their own fd management.
331 - support EV_IDLE_ENABLE being disabled in ev++.h
332 (patch by Didier Spezia).
333 - take advantage of inotify_init1, if available, to set cloexec/nonblock
334 on fd creation, to avoid races.
335 - the signal handling pipe wasn't always initialised under windows
336 (analysed by lekma).
337 - changed minimum glibc requirement from glibc 2.9 to 2.7, for
338 signalfd.
339 - add missing string.h include (Denis F. Latypoff).
340 - only replace ev_stat.prev when we detect an actual difference,
341 so prev is (almost) always different to attr. this might
342 have caused the problems with 04_stat.t.
343 - add ev::timer->remaining () method to C++ API.
344
3453.8 Sun Aug 9 14:30:45 CEST 2009
8 - incompatible change: do not necessarily reset signal handler 346 - incompatible change: do not necessarily reset signal handler
9 to SIG_DFL when a sighandler is stopped. 347 to SIG_DFL when a sighandler is stopped.
10 - ev_default_destroy did not properly free or zero some members, 348 - ev_default_destroy did not properly free or zero some members,
11 potentially causing crashes and memory corruption on repated 349 potentially causing crashes and memory corruption on repeated
12 ev_default_destroy/ev_default_loop calls. 350 ev_default_destroy/ev_default_loop calls.
13 - take advantage of signalfd on GNU/Linux systems. 351 - take advantage of signalfd on GNU/Linux systems.
14 - document that the signal mask might be in an unspecified 352 - document that the signal mask might be in an unspecified
15 state when using libev's signal handling. 353 state when using libev's signal handling.
16 - take advantage of some GNU/Linux calls to set cloexec/nonblock 354 - take advantage of some GNU/Linux calls to set cloexec/nonblock
… …
51 - correctly zero the idx values for arrays, so destroying and 389 - correctly zero the idx values for arrays, so destroying and
52 reinitialising the default loop actually works (patch by 390 reinitialising the default loop actually works (patch by
53 Malek Hadj-Ali). 391 Malek Hadj-Ali).
54 - implement ev_suspend and ev_resume. 392 - implement ev_suspend and ev_resume.
55 - new EV_CUSTOM revents flag for use by applications. 393 - new EV_CUSTOM revents flag for use by applications.
56 - add documentation section about priorites. 394 - add documentation section about priorities.
57 - add a glossary to the dcoumentation. 395 - add a glossary to the documentation.
58 - extend the ev_fork description slightly. 396 - extend the ev_fork description slightly.
59 - optimize a jump out of call_pending. 397 - optimize a jump out of call_pending.
60 398
613.53 Sun Feb 15 02:38:20 CET 2009 3993.53 Sun Feb 15 02:38:20 CET 2009
62 - fix a bug in event pipe creation on win32 that would cause a 400 - fix a bug in event pipe creation on win32 that would cause a
… …
138 do similar changes for event.c. 476 do similar changes for event.c.
139 477
1403.45 Tue Oct 21 21:59:26 CEST 2008 4783.45 Tue Oct 21 21:59:26 CEST 2008
141 - disable inotify usage on linux <2.6.25, as it is broken 479 - disable inotify usage on linux <2.6.25, as it is broken
142 (reported by Yoann Vandoorselaere). 480 (reported by Yoann Vandoorselaere).
143 - ev_stat errornously would try to add inotify watchers 481 - ev_stat erroneously would try to add inotify watchers
144 even when inotify wasn't available (this should only 482 even when inotify wasn't available (this should only
145 have a performance impact). 483 have a performance impact).
146 - ev_once now passes both timeout and io to the callback if both 484 - ev_once now passes both timeout and io to the callback if both
147 occur concurrently, instead of giving timeouts precedence. 485 occur concurrently, instead of giving timeouts precedence.
148 - disable EV_USE_INOTIFY when sys/inotify.h is too old. 486 - disable EV_USE_INOTIFY when sys/inotify.h is too old.
… …
216 - enable inlining with gcc versions 2 and 3. 554 - enable inlining with gcc versions 2 and 3.
217 - work around broken poll implementations potentially 555 - work around broken poll implementations potentially
218 not clearing revents field in ev_poll (Brandon Black) 556 not clearing revents field in ev_poll (Brandon Black)
219 (no such systems are known at this time). 557 (no such systems are known at this time).
220 - work around a bug in realloc on openbsd and darwin, 558 - work around a bug in realloc on openbsd and darwin,
221 also makes the errornous valgrind complaints 559 also makes the erroneous valgrind complaints
222 go away (noted by various people). 560 go away (noted by various people).
223 - fix ev_async_pending, add c++ wrapper for ev_async 561 - fix ev_async_pending, add c++ wrapper for ev_async
224 (based on patch sent by Johannes Deisenhofer. 562 (based on patch sent by Johannes Deisenhofer).
225 - add sensible set method to ev::embed. 563 - add sensible set method to ev::embed.
226 - made integer constants type int in ev.h. 564 - made integer constants type int in ev.h.
227 565
2283.2 Wed Apr 2 17:11:19 CEST 2008 5663.2 Wed Apr 2 17:11:19 CEST 2008
229 - fix a 64 bit overflow issue in the select backend, 567 - fix a 64 bit overflow issue in the select backend,

Diff Legend

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