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

Comparing libev/Changes (file contents):
Revision 1.119 by root, Wed Jul 8 02:46:05 2009 UTC vs.
Revision 1.316 by root, Tue Jul 31 04:45:57 2018 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 3 - move the darwin select workaround highe rin ev.c, as newre versions of
4TODO: child wactehr execution order w.r.t. childsigcb reg. priorities 4 darwin managed to break their broken select even more.
5TODO: remain 5 - ANDROID => __ANDROID__ (reported by enh@google.com).
6 - disable epoll_create1 on android because it has broken header files
7 and google is unwilling to fix them (reported by enh@google.com).
8
94.24 Wed Dec 28 05:19:55 CET 2016
10 - bump version to 4.24, as the release tarball inexplicably
11 didn't have the right version in ev.h, even though the cvs-tagged
12 version did have the right one (reported by Ales Teska).
13
144.23 Wed Nov 16 18:23:41 CET 2016
15 - move some declarations at the beginning to help certain retarded
16 microsoft compilers, even though their documentation claims
17 otherwise (reported by Ruslan Osmanov).
18
194.22 Sun Dec 20 22:11:50 CET 2015
20 - when epoll detects unremovable fds in the fd set, rebuild
21 only the epoll descriptor, not the signal pipe, to avoid
22 SIGPIPE in ev_async_send. This doesn't solve it on fork,
23 so document what needs to be done in ev_loop_fork
24 (analyzed by Benjamin Mahler).
25 - remove superfluous sys/timeb.h include on win32
26 (analyzed by Jason Madden).
27 - updated libecb.
28
294.20 Sat Jun 20 13:01:43 CEST 2015
30 - prefer noexcept over throw () with C++ 11.
31 - update ecb.h due to incompatibilities with c11.
32 - fix a potential aliasing issue when reading and writing
33 watcher callbacks.
34
354.19 Thu Sep 25 08:18:25 CEST 2014
36 - ev.h wasn't valid C++ anymore, which tripped compilers other than
37 clang, msvc or gcc (analyzed by Raphael 'kena' Poss). Unfortunately,
38 C++ doesn't support typedefs for function pointers fully, so the affected
39 declarations have to spell out the types each time.
40 - when not using autoconf, tighten the check for clock_gettime and related
41 functionality.
42
434.18 Fri Sep 5 17:55:26 CEST 2014
44 - events on files were not always generated properly with the
45 epoll backend (testcase by Assaf Inbal).
46 - mark event pipe fd as cloexec after a fork (analyzed by Sami Farin).
47 - (ecb) support m68k, m88k and sh (patch by Miod Vallat).
48 - use a reasonable fallback for EV_NSIG instead of erroring out
49 when we can't detect the signal set size.
50 - in the absence of autoconf, do not use the clock syscall
51 on glibc >= 2.17 (avoids the syscall AND -lrt on systems
52 doing clock_gettime in userspace).
53 - ensure extern "C" function pointers are used for externally-visible
54 loop callbacks (not watcher callbacks yet).
55 - (ecb) work around memory barriers and volatile apparently both being
56 broken in visual studio 2008 and later (analysed and patch by Nicolas Noble).
57
584.15 Fri Mar 1 12:04:50 CET 2013
59 - destroying a non-default loop would stop the global waitpid
60 watcher (Denis Bilenko).
61 - queueing pending watchers of higher priority from a watcher now invokes
62 them in a timely fashion (reported by Denis Bilenko).
63 - add throw() to all libev functions that cannot throw exceptions, for
64 further code size decrease when compiling for C++.
65 - add throw () to callbacks that must not throw exceptions (allocator,
66 syserr, loop acquire/release, periodic reschedule cbs).
67 - fix event_base_loop return code, add event_get_callback, event_base_new,
68 event_base_get_method calls to improve libevent 1.x emulation and add
69 some libevent 2.x functionality (based on a patch by Jeff Davey).
70 - add more memory fences to fix a bug reported by Jeff Davey. Better
71 be overfenced than underprotected.
72 - ev_run now returns a boolean status (true meaning watchers are
73 still active).
74 - ev_once: undef EV_ERROR in ev_kqueue.c, to avoid clashing with
75 libev's EV_ERROR (reported by 191919).
76 - (ecb) add memory fence support for xlC (Darin McBride).
77 - (ecb) add memory fence support for gcc-mips (Anton Kirilov).
78 - (ecb) add memory fence support for gcc-alpha (Christian Weisgerber).
79 - work around some kernels losing file descriptors by leaking
80 the kqueue descriptor in the child.
81 - work around linux inotify not reporting IN_ATTRIB changes for directories
82 in many cases.
83 - include sys/syscall.h instead of plain syscall.h.
84 - check for io watcher loops in ev_verify, check for the most
85 common reported usage bug in ev_io_start.
86 - choose socket vs. WSASocket at compiletime using EV_USE_WSASOCKET.
87 - always use WSASend/WSARecv directly on windows, hoping that this
88 works in all cases (unlike read/write/send/recv...).
89 - try to detect signals around a fork faster (test program by
90 Denis Bilenko).
91 - work around recent glibc versions that leak memory in realloc.
92 - rename ev::embed::set to ev::embed::set_embed to avoid clashing
93 the watcher base set (loop) method.
94 - rewrite the async/signal pipe logic to always keep a valid fd, which
95 simplifies (and hopefully correctifies :) the race checking
96 on fork, at the cost of one extra fd.
97 - add fat, msdos, jffs2, ramfs, ntfs and btrfs to the list of
98 inotify-supporting filesystems.
99 - move orig_CFLAGS assignment to after AC_INIT, as newer autoconf
100 versions ignore it before
101 (https://bugzilla.redhat.com/show_bug.cgi?id=908096).
102 - add some untested android support.
103 - enum expressions must be of type int (reported by Juan Pablo L).
104
1054.11 Sat Feb 4 19:52:39 CET 2012
106 - INCOMPATIBLE CHANGE: ev_timer_again now clears the pending status, as
107 was documented already, but not implemented in the repeating case.
108 - new compiletime symbols: EV_NO_SMP and EV_NO_THREADS.
109 - fix a race where the workaround against the epoll fork bugs
110 caused signals to not be handled anymore.
111 - correct backend_fudge for most backends, and implement a windows
112 specific workaround to avoid looping because we call both
113 select and Sleep, both with different time resolutions.
114 - document range and guarantees of ev_sleep.
115 - document reasonable ranges for periodics interval and offset.
116 - rename backend_fudge to backend_mintime to avoid future confusion :)
117 - change the default periodic reschedule function to hopefully be more
118 exact and correct even in corner cases or in the far future.
119 - do not rely on -lm anymore: use it when available but use our
120 own floor () if it is missing. This should make it easier to embed,
121 as no external libraries are required.
122 - strategically import macros from libecb and mark rarely-used functions
123 as cache-cold (saving almost 2k code size on typical amd64 setups).
124 - add Symbols.ev and Symbols.event files, that were missing.
125 - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now).
126 - fix #3 "be smart about timeouts" to not "deadlock" when
127 timeout == now, also improve the section overall.
128 - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom.
129 - support new EV_API_STATIC mode to make all libev symbols
130 static.
131 - supply default CFLAGS of -g -O3 with gcc when original CFLAGS
132 were empty.
133
1344.04 Wed Feb 16 09:01:51 CET 2011
135 - fix two problems in the native win32 backend, where reuse of fd's
136 with different underlying handles caused handles not to be removed
137 or added to the select set (analyzed and tested by Bert Belder).
138 - do no rely on ceil() in ev_e?poll.c.
139 - backport libev to HP-UX versions before 11 v3.
140 - configure did not detect nanosleep and clock_gettime properly when
141 they are available in the libc (as opposed to -lrt).
142
1434.03 Tue Jan 11 14:37:25 CET 2011
144 - officially support polling files with all backends.
145 - support files, /dev/zero etc. the same way as select in the epoll
146 backend, by generating events on our own.
147 - ports backend: work around solaris bug 6874410 and many related ones
148 (EINTR, maybe more), with no performance loss (note that the solaris
149 bug report is actually wrong, reality is far more bizarre and broken
150 than that).
151 - define EV_READ/EV_WRITE as macros in event.h, as some programs use
152 #ifdef to test for them.
153 - new (experimental) function: ev_feed_signal.
154 - new (to become default) EVFLAG_NOSIGMASK flag.
155 - new EVBACKEND_MASK symbol.
156 - updated COMMON IDIOMS SECTION.
157
1584.01 Fri Nov 5 21:51:29 CET 2010
159 - automake fucked it up, apparently, --add-missing -f is not quite enough
160 to make it update its files, so 4.00 didn't install ev++.h and
161 event.h on make install. grrr.
162 - ev_loop(count|depth) didn't return anything (Robin Haberkorn).
163 - change EV_UNDEF to 0xffffffff to silence some overzealous compilers.
164 - use "(libev) " prefix for all libev error messages now.
165
1664.00 Mon Oct 25 12:32:12 CEST 2010
167 - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading.
168 - ev_embed_stop did not correctly stop the watcher (very good
169 testcase by Vladimir Timofeev).
170 - ev_run will now always update the current loop time - it erroneously
171 didn't when idle watchers were active, causing timers not to fire.
172 - fix a bug where a timeout of zero caused the timer not to fire
173 in the libevent emulation (testcase by Péter Szabó).
174 - applied win32 fixes by Michael Lenaghan (also James Mansion).
175 - replace EV_MINIMAL by EV_FEATURES.
176 - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it
177 seems the former is *much* faster than the latter.
178 - linux kernel version detection (for inotify bug workarounds)
179 did not work properly.
180 - reduce the number of spurious wake-ups with the ports backend.
181 - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu).
182 - do async init within ev_async_start, not ev_async_set, which avoids
183 an API quirk where the set function must be called in the C++ API
184 even when there is nothing to set.
185 - add (undocumented) EV_ENABLE when adding events with kqueue,
186 this might help with OS X, which seems to need it despite documenting
187 not to need it (helpfully pointed out by Tilghman Lesher).
188 - do not use poll by default on freebsd, it's broken (what isn't
189 on freebsd...).
190 - allow to embed epoll on kernels >= 2.6.32.
191 - configure now prepends -O3, not appends it, so one can still
192 override it.
193 - ev.pod: greatly expanded the portability section, added a porting
194 section, a description of watcher states and made lots of minor fixes.
195 - disable poll backend on AIX, the poll header spams the namespace
196 and it's not worth working around dead platforms (reported
197 and analyzed by Aivars Kalvans).
198 - improve header file compatibility of the standalone eventfd code
199 in an obscure case.
200 - implement EV_AVOID_STDIO option.
201 - do not use sscanf to parse linux version number (smaller, faster,
202 no sscanf dependency).
203 - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings.
204 - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs.
205 - add section on accept() problems to the manpage.
206 - rename EV_TIMEOUT to EV_TIMER.
207 - rename ev_loop_count/depth/verify/loop/unloop.
208 - remove ev_default_destroy and ev_default_fork.
209 - switch to two-digit minor version.
210 - work around an apparent gentoo compiler bug.
211 - define _DARWIN_UNLIMITED_SELECT. just so.
212 - use enum instead of #define for most constants.
213 - improve compatibility to older C++ compilers.
214 - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now
215 default arguments when compiled as C++.
216 - enable automake dependency tracking.
217 - ev_loop_new no longer leaks memory when loop creation failed.
218 - new ev_cleanup watcher type.
219
2203.9 Thu Dec 31 07:59:59 CET 2009
221 - signalfd is no longer used by default and has to be requested
222 explicitly - this means that easy to catch bugs become hard to
223 catch race conditions, but the users have spoken.
224 - point out the unspecified signal mask in the documentation, and
225 that this is a race condition regardless of EV_SIGNALFD.
226 - backport inotify code to C89.
227 - inotify file descriptors could leak into child processes.
228 - ev_stat watchers could keep an erroneous extra ref on the loop,
229 preventing exit when unregistering all watchers (testcases
230 provided by ry@tinyclouds.org).
231 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
232 symbols to make it easier for apps to do their own fd management.
233 - support EV_IDLE_ENABLE being disabled in ev++.h
234 (patch by Didier Spezia).
235 - take advantage of inotify_init1, if available, to set cloexec/nonblock
236 on fd creation, to avoid races.
237 - the signal handling pipe wasn't always initialised under windows
238 (analysed by lekma).
239 - changed minimum glibc requirement from glibc 2.9 to 2.7, for
240 signalfd.
241 - add missing string.h include (Denis F. Latypoff).
242 - only replace ev_stat.prev when we detect an actual difference,
243 so prev is (almost) always different to attr. this might
244 have caused the problems with 04_stat.t.
245 - add ev::timer->remaining () method to C++ API.
246
2473.8 Sun Aug 9 14:30:45 CEST 2009
248 - incompatible change: do not necessarily reset signal handler
249 to SIG_DFL when a sighandler is stopped.
250 - ev_default_destroy did not properly free or zero some members,
251 potentially causing crashes and memory corruption on repeated
252 ev_default_destroy/ev_default_loop calls.
253 - take advantage of signalfd on GNU/Linux systems.
254 - document that the signal mask might be in an unspecified
255 state when using libev's signal handling.
256 - take advantage of some GNU/Linux calls to set cloexec/nonblock
257 on fd creation, to avoid race conditions.
258
2593.7 Fri Jul 17 16:36:32 CEST 2009
6 - ev_unloop and ev_loop wrongly used a global variable to exit loops, 260 - ev_unloop and ev_loop wrongly used a global variable to exit loops,
7 instead of using a per-loop variable (bug caught by accident...). 261 instead of using a per-loop variable (bug caught by accident...).
262 - the ev_set_io_collect_interval interpretation has changed.
263 - add new functionality: ev_set_userdata, ev_userdata,
264 ev_set_invoke_pending_cb, ev_set_loop_release_cb,
265 ev_invoke_pending, ev_pending_count, together with a long example
266 about thread locking.
267 - add ev_timer_remaining (as requested by Denis F. Latypoff).
268 - add ev_loop_depth.
269 - calling ev_unloop in fork/prepare watchers will no longer poll
270 for new events.
8 - Denis F. Latypoff corrected many typos in example code snippets. 271 - Denis F. Latypoff corrected many typos in example code snippets.
9 - the ev_set_io_collect_interval interpretation has changed.
10 - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double- 272 - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double-
11 check that the syscall number is available before trying to 273 check that the syscall number is available before trying to
12 use it (reported by ry@tinyclouds). 274 use it (reported by ry@tinyclouds).
13 - use GetSystemTimeAsFileTime instead of _timeb on windows, for 275 - use GetSystemTimeAsFileTime instead of _timeb on windows, for
14 slightly higher accuracy. 276 slightly higher accuracy.
15 - properly declare ev_loop_verify even when !EV_MULTIPLICITY. 277 - properly declare ev_loop_verify and ev_now_update even when
16 - implement ev_loop_depth. 278 !EV_MULTIPLICITY.
279 - do not compile in any priority code when EV_MAXPRI == EV_MINPRI.
280 - support EV_MINIMAL==2 for a reduced API.
281 - actually 0-initialise struct sigaction when installing signals.
282 - add section on hibernate and stopped processes to ev_timer docs.
17 283
183.6 Tue Apr 28 02:49:30 CEST 2009 2843.6 Tue Apr 28 02:49:30 CEST 2009
19 - multiple timers becoming ready within an event loop iteration 285 - multiple timers becoming ready within an event loop iteration
20 will be invoked in the "correct" order now. 286 will be invoked in the "correct" order now.
21 - do not leave the event loop early just because we have no active 287 - do not leave the event loop early just because we have no active
… …
25 - correctly zero the idx values for arrays, so destroying and 291 - correctly zero the idx values for arrays, so destroying and
26 reinitialising the default loop actually works (patch by 292 reinitialising the default loop actually works (patch by
27 Malek Hadj-Ali). 293 Malek Hadj-Ali).
28 - implement ev_suspend and ev_resume. 294 - implement ev_suspend and ev_resume.
29 - new EV_CUSTOM revents flag for use by applications. 295 - new EV_CUSTOM revents flag for use by applications.
30 - add documentation section about priorites. 296 - add documentation section about priorities.
31 - add a glossary to the dcoumentation. 297 - add a glossary to the documentation.
32 - extend the ev_fork description slightly. 298 - extend the ev_fork description slightly.
33 - optimize a jump out of call_pending. 299 - optimize a jump out of call_pending.
34 300
353.53 Sun Feb 15 02:38:20 CET 2009 3013.53 Sun Feb 15 02:38:20 CET 2009
36 - fix a bug in event pipe creation on win32 that would cause a 302 - fix a bug in event pipe creation on win32 that would cause a
… …
112 do similar changes for event.c. 378 do similar changes for event.c.
113 379
1143.45 Tue Oct 21 21:59:26 CEST 2008 3803.45 Tue Oct 21 21:59:26 CEST 2008
115 - disable inotify usage on linux <2.6.25, as it is broken 381 - disable inotify usage on linux <2.6.25, as it is broken
116 (reported by Yoann Vandoorselaere). 382 (reported by Yoann Vandoorselaere).
117 - ev_stat errornously would try to add inotify watchers 383 - ev_stat erroneously would try to add inotify watchers
118 even when inotify wasn't available (this should only 384 even when inotify wasn't available (this should only
119 have a performance impact). 385 have a performance impact).
120 - ev_once now passes both timeout and io to the callback if both 386 - ev_once now passes both timeout and io to the callback if both
121 occur concurrently, instead of giving timeouts precedence. 387 occur concurrently, instead of giving timeouts precedence.
122 - disable EV_USE_INOTIFY when sys/inotify.h is too old. 388 - disable EV_USE_INOTIFY when sys/inotify.h is too old.
… …
190 - enable inlining with gcc versions 2 and 3. 456 - enable inlining with gcc versions 2 and 3.
191 - work around broken poll implementations potentially 457 - work around broken poll implementations potentially
192 not clearing revents field in ev_poll (Brandon Black) 458 not clearing revents field in ev_poll (Brandon Black)
193 (no such systems are known at this time). 459 (no such systems are known at this time).
194 - work around a bug in realloc on openbsd and darwin, 460 - work around a bug in realloc on openbsd and darwin,
195 also makes the errornous valgrind complaints 461 also makes the erroneous valgrind complaints
196 go away (noted by various people). 462 go away (noted by various people).
197 - fix ev_async_pending, add c++ wrapper for ev_async 463 - fix ev_async_pending, add c++ wrapper for ev_async
198 (based on patch sent by Johannes Deisenhofer. 464 (based on patch sent by Johannes Deisenhofer).
199 - add sensible set method to ev::embed. 465 - add sensible set method to ev::embed.
200 - made integer constants type int in ev.h. 466 - made integer constants type int in ev.h.
201 467
2023.2 Wed Apr 2 17:11:19 CEST 2008 4683.2 Wed Apr 2 17:11:19 CEST 2008
203 - fix a 64 bit overflow issue in the select backend, 469 - fix a 64 bit overflow issue in the select backend,

Diff Legend

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