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

Comparing libev/Changes (file contents):
Revision 1.174 by root, Sat Mar 27 02:26:27 2010 UTC vs.
Revision 1.357 by root, Wed Jan 22 01:50:42 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: ABI??? API????? Changes??? 3TODO: revisit 59.x timer in the light of modern powersaving
4TODO: win32 write() to socket for signal handling 4TODO: ^ maybe just increase it when timerfd, modern linux only thing that counts?
5TODO: document EV_TSTAMP_T
6TODO: for next ABI/API change, consider moving EV__IOFDSSET into io->fd instead and provide a getter.
7
8TODO: add ev_io_set_events, or ev_io_modify, or ev_io_change, or..., without IOFDSET
9TODO: fix documentation of read-only] [read-write] and so on
10TODO: maybe allow events = 0 for io watchers
11 - the 4.31 timerfd code wrongly changes the priority of the signal
12 fd watcher, which is usually harmless unless signal fds are
13 also used (found via cpan tester service).
14 - do not wake up every minute when timerfd is used to detect timejumps.
15 - support a lot more "uncommon" compile time configurations,
16 such as ev_embed enabled but ev_timer disabled.
17 - the linux aio backend is no longer compiled in by default.
18 - update to libecb version 0x00010008.
19
204.31 Fri Dec 20 21:58:29 CET 2019
21 - handle backends with minimum wait time a bit better by not
22 waiting in the presence of already-expired timers
23 (behaviour reported by Felipe Gasper).
24 - new feature: use timerfd to detect timejumps quickly,
25 can be disabled with the new EVFLAG_NOTIMERFD loop flag.
26 - document EV_USE_SIGNALFD feature macro.
27
284.30 (EV only)
29 - change non-autoconf test for __kernel_rwf_t by testing
30 LINUX_VERSION_CODE, the most direct test I could find.
31 - fix a bug in the io_uring backend that polled the wrong
32 backend fd, causing it to not work in many cases.
33
344.29 (EV only)
35 - add io uring autoconf and non-autoconf detection.
36 - disable io_uring when some header files are too old.
37
384.28 (EV only)
39 - linuxaio backend resulted in random memory corruption
40 when loop is forked.
41 - linuxaio backend might have tried to cancel an iocb
42 multiple times (was unable to trigger this).
43 - linuxaio backend now employs a generation counter to
44 avoid handling spurious events from cancelled requests.
45 - io_cancel can return EINTR, deal with it. also, assume
46 io_submit also returns EINTR.
47 - fix some other minor bugs in linuxaio backend.
48 - ev_tstamp type can now be overriden by defining EV_TSTAMP_T.
49 - cleanup: replace expect_true/false and noinline by their
50 libecb counterparts.
51 - move syscall infrastructure from ev_linuxaio.c to ev.c.
52 - prepare io_uring integration.
53 - tweak ev_floor.
54 - epoll, poll, win32 Sleep and other places that use millisecond
55 reslution now all try to round up times.
56 - solaris port backend didn't compile.
57 - abstract time constants into their macros, for more flexibility.
58
594.27 Thu Jun 27 22:43:44 CEST 2019
60 - linux aio backend almost completely rewritten to work around its
61 limitations.
62 - linux aio backend now requires linux 4.19+.
63 - epoll backend now mandatory for linux aio backend.
64 - fail assertions more aggressively on invalid fd's detected
65 in the event loop, do not just silently fd_kill in case of
66 user error.
67 - ev_io_start/ev_io_stop now verify the watcher fd using
68 a syscall when EV_VERIFY is 2 or higher.
69
704.26 (EV only)
71 - update to libecb 0x00010006.
72 - new experimental linux aio backend (linux 4.18+).
73 - removed redundant 0-ptr check in ev_once.
74 - updated/extended ev_set_allocator documentation.
75 - replaced EMPTY2 macro by array_needsize_noinit.
76 - minor code cleanups.
77 - epoll backend now uses epoll_create1 also after fork.
78
794.25 Fri Dec 21 07:49:20 CET 2018
80 - INCOMPATIBLE CHANGE: EV_THROW was renamed to EV_NOEXCEPT
81 (EV_THROW still provided) and now uses noexcept on C++11 or newer.
82 - move the darwin select workaround higher in ev.c, as newer versions of
83 darwin managed to break their broken select even more.
84 - ANDROID => __ANDROID__ (reported by enh@google.com).
85 - disable epoll_create1 on android because it has broken header files
86 and google is unwilling to fix them (reported by enh@google.com).
87 - avoid a minor compilation warning on win32.
88 - c++: remove deprecated dynamic throw() specifications.
89 - c++: improve the (unsupported) bad_loop exception class.
90 - backport perl ev_periodic example to C, untested.
91 - update libecb, biggets change is to include a memory fence
92 in ECB_MEMORY_FENCE_RELEASE on x86/amd64.
93 - minor autoconf/automake modernisation.
94
954.24 Wed Dec 28 05:19:55 CET 2016
96 - bump version to 4.24, as the release tarball inexplicably
97 didn't have the right version in ev.h, even though the cvs-tagged
98 version did have the right one (reported by Ales Teska).
99
1004.23 Wed Nov 16 18:23:41 CET 2016
101 - move some declarations at the beginning to help certain retarded
102 microsoft compilers, even though their documentation claims
103 otherwise (reported by Ruslan Osmanov).
104
1054.22 Sun Dec 20 22:11:50 CET 2015
106 - when epoll detects unremovable fds in the fd set, rebuild
107 only the epoll descriptor, not the signal pipe, to avoid
108 SIGPIPE in ev_async_send. This doesn't solve it on fork,
109 so document what needs to be done in ev_loop_fork
110 (analyzed by Benjamin Mahler).
111 - remove superfluous sys/timeb.h include on win32
112 (analyzed by Jason Madden).
113 - updated libecb.
114
1154.20 Sat Jun 20 13:01:43 CEST 2015
116 - prefer noexcept over throw () with C++ 11.
117 - update ecb.h due to incompatibilities with c11.
118 - fix a potential aliasing issue when reading and writing
119 watcher callbacks.
120
1214.19 Thu Sep 25 08:18:25 CEST 2014
122 - ev.h wasn't valid C++ anymore, which tripped compilers other than
123 clang, msvc or gcc (analyzed by Raphael 'kena' Poss). Unfortunately,
124 C++ doesn't support typedefs for function pointers fully, so the affected
125 declarations have to spell out the types each time.
126 - when not using autoconf, tighten the check for clock_gettime and related
127 functionality.
128
1294.18 Fri Sep 5 17:55:26 CEST 2014
130 - events on files were not always generated properly with the
131 epoll backend (testcase by Assaf Inbal).
132 - mark event pipe fd as cloexec after a fork (analyzed by Sami Farin).
133 - (ecb) support m68k, m88k and sh (patch by Miod Vallat).
134 - use a reasonable fallback for EV_NSIG instead of erroring out
135 when we can't detect the signal set size.
136 - in the absence of autoconf, do not use the clock syscall
137 on glibc >= 2.17 (avoids the syscall AND -lrt on systems
138 doing clock_gettime in userspace).
139 - ensure extern "C" function pointers are used for externally-visible
140 loop callbacks (not watcher callbacks yet).
141 - (ecb) work around memory barriers and volatile apparently both being
142 broken in visual studio 2008 and later (analysed and patch by Nicolas Noble).
143
1444.15 Fri Mar 1 12:04:50 CET 2013
145 - destroying a non-default loop would stop the global waitpid
146 watcher (Denis Bilenko).
147 - queueing pending watchers of higher priority from a watcher now invokes
148 them in a timely fashion (reported by Denis Bilenko).
149 - add throw() to all libev functions that cannot throw exceptions, for
150 further code size decrease when compiling for C++.
151 - add throw () to callbacks that must not throw exceptions (allocator,
152 syserr, loop acquire/release, periodic reschedule cbs).
153 - fix event_base_loop return code, add event_get_callback, event_base_new,
154 event_base_get_method calls to improve libevent 1.x emulation and add
155 some libevent 2.x functionality (based on a patch by Jeff Davey).
156 - add more memory fences to fix a bug reported by Jeff Davey. Better
157 be overfenced than underprotected.
158 - ev_run now returns a boolean status (true meaning watchers are
159 still active).
160 - ev_once: undef EV_ERROR in ev_kqueue.c, to avoid clashing with
161 libev's EV_ERROR (reported by 191919).
162 - (ecb) add memory fence support for xlC (Darin McBride).
163 - (ecb) add memory fence support for gcc-mips (Anton Kirilov).
164 - (ecb) add memory fence support for gcc-alpha (Christian Weisgerber).
165 - work around some kernels losing file descriptors by leaking
166 the kqueue descriptor in the child.
167 - work around linux inotify not reporting IN_ATTRIB changes for directories
168 in many cases.
169 - include sys/syscall.h instead of plain syscall.h.
170 - check for io watcher loops in ev_verify, check for the most
171 common reported usage bug in ev_io_start.
172 - choose socket vs. WSASocket at compiletime using EV_USE_WSASOCKET.
173 - always use WSASend/WSARecv directly on windows, hoping that this
174 works in all cases (unlike read/write/send/recv...).
175 - try to detect signals around a fork faster (test program by
176 Denis Bilenko).
177 - work around recent glibc versions that leak memory in realloc.
178 - rename ev::embed::set to ev::embed::set_embed to avoid clashing
179 the watcher base set (loop) method.
180 - rewrite the async/signal pipe logic to always keep a valid fd, which
181 simplifies (and hopefully correctifies :) the race checking
182 on fork, at the cost of one extra fd.
183 - add fat, msdos, jffs2, ramfs, ntfs and btrfs to the list of
184 inotify-supporting filesystems.
185 - move orig_CFLAGS assignment to after AC_INIT, as newer autoconf
186 versions ignore it before
187 (https://bugzilla.redhat.com/show_bug.cgi?id=908096).
188 - add some untested android support.
189 - enum expressions must be of type int (reported by Juan Pablo L).
190
1914.11 Sat Feb 4 19:52:39 CET 2012
192 - INCOMPATIBLE CHANGE: ev_timer_again now clears the pending status, as
193 was documented already, but not implemented in the repeating case.
194 - new compiletime symbols: EV_NO_SMP and EV_NO_THREADS.
195 - fix a race where the workaround against the epoll fork bugs
196 caused signals to not be handled anymore.
197 - correct backend_fudge for most backends, and implement a windows
198 specific workaround to avoid looping because we call both
199 select and Sleep, both with different time resolutions.
200 - document range and guarantees of ev_sleep.
201 - document reasonable ranges for periodics interval and offset.
202 - rename backend_fudge to backend_mintime to avoid future confusion :)
203 - change the default periodic reschedule function to hopefully be more
204 exact and correct even in corner cases or in the far future.
205 - do not rely on -lm anymore: use it when available but use our
206 own floor () if it is missing. This should make it easier to embed,
207 as no external libraries are required.
208 - strategically import macros from libecb and mark rarely-used functions
209 as cache-cold (saving almost 2k code size on typical amd64 setups).
210 - add Symbols.ev and Symbols.event files, that were missing.
211 - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now).
212 - fix #3 "be smart about timeouts" to not "deadlock" when
213 timeout == now, also improve the section overall.
214 - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom.
215 - support new EV_API_STATIC mode to make all libev symbols
216 static.
217 - supply default CFLAGS of -g -O3 with gcc when original CFLAGS
218 were empty.
219
2204.04 Wed Feb 16 09:01:51 CET 2011
221 - fix two problems in the native win32 backend, where reuse of fd's
222 with different underlying handles caused handles not to be removed
223 or added to the select set (analyzed and tested by Bert Belder).
224 - do no rely on ceil() in ev_e?poll.c.
225 - backport libev to HP-UX versions before 11 v3.
226 - configure did not detect nanosleep and clock_gettime properly when
227 they are available in the libc (as opposed to -lrt).
228
2294.03 Tue Jan 11 14:37:25 CET 2011
230 - officially support polling files with all backends.
231 - support files, /dev/zero etc. the same way as select in the epoll
232 backend, by generating events on our own.
233 - ports backend: work around solaris bug 6874410 and many related ones
234 (EINTR, maybe more), with no performance loss (note that the solaris
235 bug report is actually wrong, reality is far more bizarre and broken
236 than that).
237 - define EV_READ/EV_WRITE as macros in event.h, as some programs use
238 #ifdef to test for them.
239 - new (experimental) function: ev_feed_signal.
240 - new (to become default) EVFLAG_NOSIGMASK flag.
241 - new EVBACKEND_MASK symbol.
242 - updated COMMON IDIOMS SECTION.
243
2444.01 Fri Nov 5 21:51:29 CET 2010
245 - automake fucked it up, apparently, --add-missing -f is not quite enough
246 to make it update its files, so 4.00 didn't install ev++.h and
247 event.h on make install. grrr.
248 - ev_loop(count|depth) didn't return anything (Robin Haberkorn).
249 - change EV_UNDEF to 0xffffffff to silence some overzealous compilers.
250 - use "(libev) " prefix for all libev error messages now.
251
2524.00 Mon Oct 25 12:32:12 CEST 2010
5 - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading. 253 - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading.
6 - ev_embed_stop did not correctly stop the watcher (very good 254 - ev_embed_stop did not correctly stop the watcher (very good
7 testcase by Vladimir Timofeev). 255 testcase by Vladimir Timofeev).
256 - ev_run will now always update the current loop time - it erroneously
257 didn't when idle watchers were active, causing timers not to fire.
258 - fix a bug where a timeout of zero caused the timer not to fire
259 in the libevent emulation (testcase by Péter Szabó).
8 - applied win32 fixes by Michael Lenaghan (also James Mansion). 260 - applied win32 fixes by Michael Lenaghan (also James Mansion).
9 - replace EV_MINIMAL by EV_FEATURES. 261 - replace EV_MINIMAL by EV_FEATURES.
262 - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it
263 seems the former is *much* faster than the latter.
264 - linux kernel version detection (for inotify bug workarounds)
265 did not work properly.
266 - reduce the number of spurious wake-ups with the ports backend.
10 - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu). 267 - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu).
268 - do async init within ev_async_start, not ev_async_set, which avoids
269 an API quirk where the set function must be called in the C++ API
270 even when there is nothing to set.
11 - add (undocumented) EV_ENABLE when adding events with kqueue, 271 - add (undocumented) EV_ENABLE when adding events with kqueue,
272 this might help with OS X, which seems to need it despite documenting
12 this might help (helpfully pointed out by Tilghman Lesher). 273 not to need it (helpfully pointed out by Tilghman Lesher).
274 - do not use poll by default on freebsd, it's broken (what isn't
275 on freebsd...).
276 - allow to embed epoll on kernels >= 2.6.32.
13 - configure now prepends -O3, not appends it, so one can still 277 - configure now prepends -O3, not appends it, so one can still
14 override it. 278 override it.
279 - ev.pod: greatly expanded the portability section, added a porting
280 section, a description of watcher states and made lots of minor fixes.
15 - disable poll backend on AIX, the poll header spams the namespace 281 - disable poll backend on AIX, the poll header spams the namespace
16 and it's not worth working around dead platforms (reported 282 and it's not worth working around dead platforms (reported
17 and analyzed by Aivars Kalvans). 283 and analyzed by Aivars Kalvans).
18 - improve header file compatibility of the standalone eventfd code 284 - improve header file compatibility of the standalone eventfd code
19 in an obscure case. 285 in an obscure case.
… …
22 no sscanf dependency). 288 no sscanf dependency).
23 - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings. 289 - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings.
24 - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs. 290 - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs.
25 - add section on accept() problems to the manpage. 291 - add section on accept() problems to the manpage.
26 - rename EV_TIMEOUT to EV_TIMER. 292 - rename EV_TIMEOUT to EV_TIMER.
27 - rename ev_loop_count/depth/verify. 293 - rename ev_loop_count/depth/verify/loop/unloop.
294 - remove ev_default_destroy and ev_default_fork.
28 - switch to two-digit minor version. 295 - switch to two-digit minor version.
296 - work around an apparent gentoo compiler bug.
297 - define _DARWIN_UNLIMITED_SELECT. just so.
298 - use enum instead of #define for most constants.
299 - improve compatibility to older C++ compilers.
300 - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now
301 default arguments when compiled as C++.
302 - enable automake dependency tracking.
303 - ev_loop_new no longer leaks memory when loop creation failed.
304 - new ev_cleanup watcher type.
29 305
303.9 Thu Dec 31 07:59:59 CET 2009 3063.9 Thu Dec 31 07:59:59 CET 2009
31 - signalfd is no longer used by default and has to be requested 307 - signalfd is no longer used by default and has to be requested
32 explicitly - this means that easy to catch bugs become hard to 308 explicitly - this means that easy to catch bugs become hard to
33 catch race conditions, but the users have spoken. 309 catch race conditions, but the users have spoken.
34 - point out the unspecified signal mask in the documentation, and 310 - point out the unspecified signal mask in the documentation, and
35 that this is a race condition regardless of EV_SIGNALFD. 311 that this is a race condition regardless of EV_SIGNALFD.
36 - backport inotify code to C89. 312 - backport inotify code to C89.
37 - inotify file descriptors could leak into child processes. 313 - inotify file descriptors could leak into child processes.
38 - ev_stat watchers could keep an errornous extra ref on the loop, 314 - ev_stat watchers could keep an erroneous extra ref on the loop,
39 preventing exit when unregistering all watchers (testcases 315 preventing exit when unregistering all watchers (testcases
40 provided by ry@tinyclouds.org). 316 provided by ry@tinyclouds.org).
41 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration 317 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
42 symbols to make it easier for apps to do their own fd management. 318 symbols to make it easier for apps to do their own fd management.
43 - support EV_IDLE_ENABLE being disabled in ev++.h 319 - support EV_IDLE_ENABLE being disabled in ev++.h
… …
49 - changed minimum glibc requirement from glibc 2.9 to 2.7, for 325 - changed minimum glibc requirement from glibc 2.9 to 2.7, for
50 signalfd. 326 signalfd.
51 - add missing string.h include (Denis F. Latypoff). 327 - add missing string.h include (Denis F. Latypoff).
52 - only replace ev_stat.prev when we detect an actual difference, 328 - only replace ev_stat.prev when we detect an actual difference,
53 so prev is (almost) always different to attr. this might 329 so prev is (almost) always different to attr. this might
54 have caused the probems with 04_stat.t. 330 have caused the problems with 04_stat.t.
55 - add ev::timer->remaining () method to C++ API. 331 - add ev::timer->remaining () method to C++ API.
56 332
573.8 Sun Aug 9 14:30:45 CEST 2009 3333.8 Sun Aug 9 14:30:45 CEST 2009
58 - incompatible change: do not necessarily reset signal handler 334 - incompatible change: do not necessarily reset signal handler
59 to SIG_DFL when a sighandler is stopped. 335 to SIG_DFL when a sighandler is stopped.
60 - ev_default_destroy did not properly free or zero some members, 336 - ev_default_destroy did not properly free or zero some members,
61 potentially causing crashes and memory corruption on repated 337 potentially causing crashes and memory corruption on repeated
62 ev_default_destroy/ev_default_loop calls. 338 ev_default_destroy/ev_default_loop calls.
63 - take advantage of signalfd on GNU/Linux systems. 339 - take advantage of signalfd on GNU/Linux systems.
64 - document that the signal mask might be in an unspecified 340 - document that the signal mask might be in an unspecified
65 state when using libev's signal handling. 341 state when using libev's signal handling.
66 - take advantage of some GNU/Linux calls to set cloexec/nonblock 342 - take advantage of some GNU/Linux calls to set cloexec/nonblock
… …
101 - correctly zero the idx values for arrays, so destroying and 377 - correctly zero the idx values for arrays, so destroying and
102 reinitialising the default loop actually works (patch by 378 reinitialising the default loop actually works (patch by
103 Malek Hadj-Ali). 379 Malek Hadj-Ali).
104 - implement ev_suspend and ev_resume. 380 - implement ev_suspend and ev_resume.
105 - new EV_CUSTOM revents flag for use by applications. 381 - new EV_CUSTOM revents flag for use by applications.
106 - add documentation section about priorites. 382 - add documentation section about priorities.
107 - add a glossary to the dcoumentation. 383 - add a glossary to the documentation.
108 - extend the ev_fork description slightly. 384 - extend the ev_fork description slightly.
109 - optimize a jump out of call_pending. 385 - optimize a jump out of call_pending.
110 386
1113.53 Sun Feb 15 02:38:20 CET 2009 3873.53 Sun Feb 15 02:38:20 CET 2009
112 - fix a bug in event pipe creation on win32 that would cause a 388 - fix a bug in event pipe creation on win32 that would cause a
… …
188 do similar changes for event.c. 464 do similar changes for event.c.
189 465
1903.45 Tue Oct 21 21:59:26 CEST 2008 4663.45 Tue Oct 21 21:59:26 CEST 2008
191 - disable inotify usage on linux <2.6.25, as it is broken 467 - disable inotify usage on linux <2.6.25, as it is broken
192 (reported by Yoann Vandoorselaere). 468 (reported by Yoann Vandoorselaere).
193 - ev_stat errornously would try to add inotify watchers 469 - ev_stat erroneously would try to add inotify watchers
194 even when inotify wasn't available (this should only 470 even when inotify wasn't available (this should only
195 have a performance impact). 471 have a performance impact).
196 - ev_once now passes both timeout and io to the callback if both 472 - ev_once now passes both timeout and io to the callback if both
197 occur concurrently, instead of giving timeouts precedence. 473 occur concurrently, instead of giving timeouts precedence.
198 - disable EV_USE_INOTIFY when sys/inotify.h is too old. 474 - disable EV_USE_INOTIFY when sys/inotify.h is too old.
… …
266 - enable inlining with gcc versions 2 and 3. 542 - enable inlining with gcc versions 2 and 3.
267 - work around broken poll implementations potentially 543 - work around broken poll implementations potentially
268 not clearing revents field in ev_poll (Brandon Black) 544 not clearing revents field in ev_poll (Brandon Black)
269 (no such systems are known at this time). 545 (no such systems are known at this time).
270 - work around a bug in realloc on openbsd and darwin, 546 - work around a bug in realloc on openbsd and darwin,
271 also makes the errornous valgrind complaints 547 also makes the erroneous valgrind complaints
272 go away (noted by various people). 548 go away (noted by various people).
273 - fix ev_async_pending, add c++ wrapper for ev_async 549 - fix ev_async_pending, add c++ wrapper for ev_async
274 (based on patch sent by Johannes Deisenhofer. 550 (based on patch sent by Johannes Deisenhofer).
275 - add sensible set method to ev::embed. 551 - add sensible set method to ev::embed.
276 - made integer constants type int in ev.h. 552 - made integer constants type int in ev.h.
277 553
2783.2 Wed Apr 2 17:11:19 CEST 2008 5543.2 Wed Apr 2 17:11:19 CEST 2008
279 - fix a 64 bit overflow issue in the select backend, 555 - fix a 64 bit overflow issue in the select backend,

Diff Legend

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