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

Comparing libev/Changes (file contents):
Revision 1.178 by root, Wed Apr 21 13:13:17 2010 UTC vs.
Revision 1.339 by root, Sun Jul 7 06:00:31 2019 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 mdoenr powersaving
4TODO: win32 write() to socket for signal handling 4TODO: maybe use timerfd to detect time jumps on linux
5TODO: poll asert 5 - linuxaio backend might have tried to cancel an iocb
6TODO: EV_USE__XXX default with config, document 6 multiple times (was unable to trigger this).
7 - io_cancel can return EINTR, deal with it. also, assume
8 io_asubmit also retursn EINTR.
9 - fix some othe rminor bugs in linuxaio backend.
10 - cleanup: replace expect_true/false and noinline by their
11 libecb counterparts.
12 - move syscall infrastructure from ev_linuxaio.c to ev.c.
13 - prepare io_uring integration.
14 - tweak ev_floor.
15
164.27 Thu Jun 27 22:43:44 CEST 2019
17 - linux aio backend almost completely rewritten to work around its
18 limitations.
19 - linux aio backend now requires linux 4.19+.
20 - epoll backend now mandatory for linux aio backend.
21 - fail assertions more aggressively on invalid fd's detected
22 in the event loop, do not just silently fd_kill in case of
23 user error.
24 - ev_io_start/ev_io_stop now verify the watcher fd using
25 a syscall when EV_VERIFY is 2 or higher.
26
274.26 (EV only)
28 - update to libecb 0x00010006.
29 - new experimental linux aio backend (linux 4.18+).
30 - removed redundant 0-ptr check in ev_once.
31 - updated/extended ev_set_allocator documentation.
32 - replaced EMPTY2 macro by array_needsize_noinit.
33 - minor code cleanups.
34 - epoll backend now uses epoll_create1 also after fork.
35
364.25 Fri Dec 21 07:49:20 CET 2018
37 - INCOMPATIBLE CHANGE: EV_THROW was renamed to EV_NOEXCEPT
38 (EV_THROW still provided) and now uses noexcept on C++11 or newer.
39 - move the darwin select workaround higher in ev.c, as newer versions of
40 darwin managed to break their broken select even more.
41 - ANDROID => __ANDROID__ (reported by enh@google.com).
42 - disable epoll_create1 on android because it has broken header files
43 and google is unwilling to fix them (reported by enh@google.com).
44 - avoid a minor compilation warning on win32.
45 - c++: remove deprecated dynamic throw() specifications.
46 - c++: improve the (unsupported) bad_loop exception class.
47 - backport perl ev_periodic example to C, untested.
48 - update libecb, biggets change is to include a memory fence
49 in ECB_MEMORY_FENCE_RELEASE on x86/amd64.
50 - minor autoconf/automake modernisation.
51
524.24 Wed Dec 28 05:19:55 CET 2016
53 - bump version to 4.24, as the release tarball inexplicably
54 didn't have the right version in ev.h, even though the cvs-tagged
55 version did have the right one (reported by Ales Teska).
56
574.23 Wed Nov 16 18:23:41 CET 2016
58 - move some declarations at the beginning to help certain retarded
59 microsoft compilers, even though their documentation claims
60 otherwise (reported by Ruslan Osmanov).
61
624.22 Sun Dec 20 22:11:50 CET 2015
63 - when epoll detects unremovable fds in the fd set, rebuild
64 only the epoll descriptor, not the signal pipe, to avoid
65 SIGPIPE in ev_async_send. This doesn't solve it on fork,
66 so document what needs to be done in ev_loop_fork
67 (analyzed by Benjamin Mahler).
68 - remove superfluous sys/timeb.h include on win32
69 (analyzed by Jason Madden).
70 - updated libecb.
71
724.20 Sat Jun 20 13:01:43 CEST 2015
73 - prefer noexcept over throw () with C++ 11.
74 - update ecb.h due to incompatibilities with c11.
75 - fix a potential aliasing issue when reading and writing
76 watcher callbacks.
77
784.19 Thu Sep 25 08:18:25 CEST 2014
79 - ev.h wasn't valid C++ anymore, which tripped compilers other than
80 clang, msvc or gcc (analyzed by Raphael 'kena' Poss). Unfortunately,
81 C++ doesn't support typedefs for function pointers fully, so the affected
82 declarations have to spell out the types each time.
83 - when not using autoconf, tighten the check for clock_gettime and related
84 functionality.
85
864.18 Fri Sep 5 17:55:26 CEST 2014
87 - events on files were not always generated properly with the
88 epoll backend (testcase by Assaf Inbal).
89 - mark event pipe fd as cloexec after a fork (analyzed by Sami Farin).
90 - (ecb) support m68k, m88k and sh (patch by Miod Vallat).
91 - use a reasonable fallback for EV_NSIG instead of erroring out
92 when we can't detect the signal set size.
93 - in the absence of autoconf, do not use the clock syscall
94 on glibc >= 2.17 (avoids the syscall AND -lrt on systems
95 doing clock_gettime in userspace).
96 - ensure extern "C" function pointers are used for externally-visible
97 loop callbacks (not watcher callbacks yet).
98 - (ecb) work around memory barriers and volatile apparently both being
99 broken in visual studio 2008 and later (analysed and patch by Nicolas Noble).
100
1014.15 Fri Mar 1 12:04:50 CET 2013
102 - destroying a non-default loop would stop the global waitpid
103 watcher (Denis Bilenko).
104 - queueing pending watchers of higher priority from a watcher now invokes
105 them in a timely fashion (reported by Denis Bilenko).
106 - add throw() to all libev functions that cannot throw exceptions, for
107 further code size decrease when compiling for C++.
108 - add throw () to callbacks that must not throw exceptions (allocator,
109 syserr, loop acquire/release, periodic reschedule cbs).
110 - fix event_base_loop return code, add event_get_callback, event_base_new,
111 event_base_get_method calls to improve libevent 1.x emulation and add
112 some libevent 2.x functionality (based on a patch by Jeff Davey).
113 - add more memory fences to fix a bug reported by Jeff Davey. Better
114 be overfenced than underprotected.
115 - ev_run now returns a boolean status (true meaning watchers are
116 still active).
117 - ev_once: undef EV_ERROR in ev_kqueue.c, to avoid clashing with
118 libev's EV_ERROR (reported by 191919).
119 - (ecb) add memory fence support for xlC (Darin McBride).
120 - (ecb) add memory fence support for gcc-mips (Anton Kirilov).
121 - (ecb) add memory fence support for gcc-alpha (Christian Weisgerber).
122 - work around some kernels losing file descriptors by leaking
123 the kqueue descriptor in the child.
124 - work around linux inotify not reporting IN_ATTRIB changes for directories
125 in many cases.
126 - include sys/syscall.h instead of plain syscall.h.
127 - check for io watcher loops in ev_verify, check for the most
128 common reported usage bug in ev_io_start.
129 - choose socket vs. WSASocket at compiletime using EV_USE_WSASOCKET.
130 - always use WSASend/WSARecv directly on windows, hoping that this
131 works in all cases (unlike read/write/send/recv...).
132 - try to detect signals around a fork faster (test program by
133 Denis Bilenko).
134 - work around recent glibc versions that leak memory in realloc.
135 - rename ev::embed::set to ev::embed::set_embed to avoid clashing
136 the watcher base set (loop) method.
137 - rewrite the async/signal pipe logic to always keep a valid fd, which
138 simplifies (and hopefully correctifies :) the race checking
139 on fork, at the cost of one extra fd.
140 - add fat, msdos, jffs2, ramfs, ntfs and btrfs to the list of
141 inotify-supporting filesystems.
142 - move orig_CFLAGS assignment to after AC_INIT, as newer autoconf
143 versions ignore it before
144 (https://bugzilla.redhat.com/show_bug.cgi?id=908096).
145 - add some untested android support.
146 - enum expressions must be of type int (reported by Juan Pablo L).
147
1484.11 Sat Feb 4 19:52:39 CET 2012
149 - INCOMPATIBLE CHANGE: ev_timer_again now clears the pending status, as
150 was documented already, but not implemented in the repeating case.
151 - new compiletime symbols: EV_NO_SMP and EV_NO_THREADS.
152 - fix a race where the workaround against the epoll fork bugs
153 caused signals to not be handled anymore.
154 - correct backend_fudge for most backends, and implement a windows
155 specific workaround to avoid looping because we call both
156 select and Sleep, both with different time resolutions.
157 - document range and guarantees of ev_sleep.
158 - document reasonable ranges for periodics interval and offset.
159 - rename backend_fudge to backend_mintime to avoid future confusion :)
160 - change the default periodic reschedule function to hopefully be more
161 exact and correct even in corner cases or in the far future.
162 - do not rely on -lm anymore: use it when available but use our
163 own floor () if it is missing. This should make it easier to embed,
164 as no external libraries are required.
165 - strategically import macros from libecb and mark rarely-used functions
166 as cache-cold (saving almost 2k code size on typical amd64 setups).
167 - add Symbols.ev and Symbols.event files, that were missing.
168 - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now).
169 - fix #3 "be smart about timeouts" to not "deadlock" when
170 timeout == now, also improve the section overall.
171 - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom.
172 - support new EV_API_STATIC mode to make all libev symbols
173 static.
174 - supply default CFLAGS of -g -O3 with gcc when original CFLAGS
175 were empty.
176
1774.04 Wed Feb 16 09:01:51 CET 2011
178 - fix two problems in the native win32 backend, where reuse of fd's
179 with different underlying handles caused handles not to be removed
180 or added to the select set (analyzed and tested by Bert Belder).
181 - do no rely on ceil() in ev_e?poll.c.
182 - backport libev to HP-UX versions before 11 v3.
183 - configure did not detect nanosleep and clock_gettime properly when
184 they are available in the libc (as opposed to -lrt).
185
1864.03 Tue Jan 11 14:37:25 CET 2011
187 - officially support polling files with all backends.
188 - support files, /dev/zero etc. the same way as select in the epoll
189 backend, by generating events on our own.
190 - ports backend: work around solaris bug 6874410 and many related ones
191 (EINTR, maybe more), with no performance loss (note that the solaris
192 bug report is actually wrong, reality is far more bizarre and broken
193 than that).
194 - define EV_READ/EV_WRITE as macros in event.h, as some programs use
195 #ifdef to test for them.
196 - new (experimental) function: ev_feed_signal.
197 - new (to become default) EVFLAG_NOSIGMASK flag.
198 - new EVBACKEND_MASK symbol.
199 - updated COMMON IDIOMS SECTION.
200
2014.01 Fri Nov 5 21:51:29 CET 2010
202 - automake fucked it up, apparently, --add-missing -f is not quite enough
203 to make it update its files, so 4.00 didn't install ev++.h and
204 event.h on make install. grrr.
205 - ev_loop(count|depth) didn't return anything (Robin Haberkorn).
206 - change EV_UNDEF to 0xffffffff to silence some overzealous compilers.
207 - use "(libev) " prefix for all libev error messages now.
208
2094.00 Mon Oct 25 12:32:12 CEST 2010
7 - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading. 210 - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading.
8 - ev_embed_stop did not correctly stop the watcher (very good 211 - ev_embed_stop did not correctly stop the watcher (very good
9 testcase by Vladimir Timofeev). 212 testcase by Vladimir Timofeev).
213 - ev_run will now always update the current loop time - it erroneously
214 didn't when idle watchers were active, causing timers not to fire.
215 - fix a bug where a timeout of zero caused the timer not to fire
216 in the libevent emulation (testcase by Péter Szabó).
10 - applied win32 fixes by Michael Lenaghan (also James Mansion). 217 - applied win32 fixes by Michael Lenaghan (also James Mansion).
11 - replace EV_MINIMAL by EV_FEATURES. 218 - replace EV_MINIMAL by EV_FEATURES.
12 - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it 219 - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it
13 seems the former is *much* faster than the latter. 220 seems the former is *much* faster than the latter.
221 - linux kernel version detection (for inotify bug workarounds)
222 did not work properly.
223 - reduce the number of spurious wake-ups with the ports backend.
14 - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu). 224 - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu).
225 - do async init within ev_async_start, not ev_async_set, which avoids
226 an API quirk where the set function must be called in the C++ API
227 even when there is nothing to set.
15 - add (undocumented) EV_ENABLE when adding events with kqueue, 228 - add (undocumented) EV_ENABLE when adding events with kqueue,
16 this might help with OS X, which seems to need it despite documenting 229 this might help with OS X, which seems to need it despite documenting
17 not to need it (helpfully pointed out by Tilghman Lesher). 230 not to need it (helpfully pointed out by Tilghman Lesher).
18 - do not use poll by default on freebsd, it's broken (what isn't 231 - do not use poll by default on freebsd, it's broken (what isn't
19 on freebsd...). 232 on freebsd...).
233 - allow to embed epoll on kernels >= 2.6.32.
20 - configure now prepends -O3, not appends it, so one can still 234 - configure now prepends -O3, not appends it, so one can still
21 override it. 235 override it.
236 - ev.pod: greatly expanded the portability section, added a porting
237 section, a description of watcher states and made lots of minor fixes.
22 - disable poll backend on AIX, the poll header spams the namespace 238 - disable poll backend on AIX, the poll header spams the namespace
23 and it's not worth working around dead platforms (reported 239 and it's not worth working around dead platforms (reported
24 and analyzed by Aivars Kalvans). 240 and analyzed by Aivars Kalvans).
25 - improve header file compatibility of the standalone eventfd code 241 - improve header file compatibility of the standalone eventfd code
26 in an obscure case. 242 in an obscure case.
… …
29 no sscanf dependency). 245 no sscanf dependency).
30 - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings. 246 - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings.
31 - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs. 247 - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs.
32 - add section on accept() problems to the manpage. 248 - add section on accept() problems to the manpage.
33 - rename EV_TIMEOUT to EV_TIMER. 249 - rename EV_TIMEOUT to EV_TIMER.
34 - rename ev_loop_count/depth/verify. 250 - rename ev_loop_count/depth/verify/loop/unloop.
251 - remove ev_default_destroy and ev_default_fork.
35 - switch to two-digit minor version. 252 - switch to two-digit minor version.
253 - work around an apparent gentoo compiler bug.
254 - define _DARWIN_UNLIMITED_SELECT. just so.
255 - use enum instead of #define for most constants.
256 - improve compatibility to older C++ compilers.
257 - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now
258 default arguments when compiled as C++.
259 - enable automake dependency tracking.
260 - ev_loop_new no longer leaks memory when loop creation failed.
261 - new ev_cleanup watcher type.
36 262
373.9 Thu Dec 31 07:59:59 CET 2009 2633.9 Thu Dec 31 07:59:59 CET 2009
38 - signalfd is no longer used by default and has to be requested 264 - signalfd is no longer used by default and has to be requested
39 explicitly - this means that easy to catch bugs become hard to 265 explicitly - this means that easy to catch bugs become hard to
40 catch race conditions, but the users have spoken. 266 catch race conditions, but the users have spoken.
41 - point out the unspecified signal mask in the documentation, and 267 - point out the unspecified signal mask in the documentation, and
42 that this is a race condition regardless of EV_SIGNALFD. 268 that this is a race condition regardless of EV_SIGNALFD.
43 - backport inotify code to C89. 269 - backport inotify code to C89.
44 - inotify file descriptors could leak into child processes. 270 - inotify file descriptors could leak into child processes.
45 - ev_stat watchers could keep an errornous extra ref on the loop, 271 - ev_stat watchers could keep an erroneous extra ref on the loop,
46 preventing exit when unregistering all watchers (testcases 272 preventing exit when unregistering all watchers (testcases
47 provided by ry@tinyclouds.org). 273 provided by ry@tinyclouds.org).
48 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration 274 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
49 symbols to make it easier for apps to do their own fd management. 275 symbols to make it easier for apps to do their own fd management.
50 - support EV_IDLE_ENABLE being disabled in ev++.h 276 - support EV_IDLE_ENABLE being disabled in ev++.h
… …
56 - changed minimum glibc requirement from glibc 2.9 to 2.7, for 282 - changed minimum glibc requirement from glibc 2.9 to 2.7, for
57 signalfd. 283 signalfd.
58 - add missing string.h include (Denis F. Latypoff). 284 - add missing string.h include (Denis F. Latypoff).
59 - only replace ev_stat.prev when we detect an actual difference, 285 - only replace ev_stat.prev when we detect an actual difference,
60 so prev is (almost) always different to attr. this might 286 so prev is (almost) always different to attr. this might
61 have caused the probems with 04_stat.t. 287 have caused the problems with 04_stat.t.
62 - add ev::timer->remaining () method to C++ API. 288 - add ev::timer->remaining () method to C++ API.
63 289
643.8 Sun Aug 9 14:30:45 CEST 2009 2903.8 Sun Aug 9 14:30:45 CEST 2009
65 - incompatible change: do not necessarily reset signal handler 291 - incompatible change: do not necessarily reset signal handler
66 to SIG_DFL when a sighandler is stopped. 292 to SIG_DFL when a sighandler is stopped.
67 - ev_default_destroy did not properly free or zero some members, 293 - ev_default_destroy did not properly free or zero some members,
68 potentially causing crashes and memory corruption on repated 294 potentially causing crashes and memory corruption on repeated
69 ev_default_destroy/ev_default_loop calls. 295 ev_default_destroy/ev_default_loop calls.
70 - take advantage of signalfd on GNU/Linux systems. 296 - take advantage of signalfd on GNU/Linux systems.
71 - document that the signal mask might be in an unspecified 297 - document that the signal mask might be in an unspecified
72 state when using libev's signal handling. 298 state when using libev's signal handling.
73 - take advantage of some GNU/Linux calls to set cloexec/nonblock 299 - take advantage of some GNU/Linux calls to set cloexec/nonblock
… …
108 - correctly zero the idx values for arrays, so destroying and 334 - correctly zero the idx values for arrays, so destroying and
109 reinitialising the default loop actually works (patch by 335 reinitialising the default loop actually works (patch by
110 Malek Hadj-Ali). 336 Malek Hadj-Ali).
111 - implement ev_suspend and ev_resume. 337 - implement ev_suspend and ev_resume.
112 - new EV_CUSTOM revents flag for use by applications. 338 - new EV_CUSTOM revents flag for use by applications.
113 - add documentation section about priorites. 339 - add documentation section about priorities.
114 - add a glossary to the dcoumentation. 340 - add a glossary to the documentation.
115 - extend the ev_fork description slightly. 341 - extend the ev_fork description slightly.
116 - optimize a jump out of call_pending. 342 - optimize a jump out of call_pending.
117 343
1183.53 Sun Feb 15 02:38:20 CET 2009 3443.53 Sun Feb 15 02:38:20 CET 2009
119 - fix a bug in event pipe creation on win32 that would cause a 345 - fix a bug in event pipe creation on win32 that would cause a
… …
195 do similar changes for event.c. 421 do similar changes for event.c.
196 422
1973.45 Tue Oct 21 21:59:26 CEST 2008 4233.45 Tue Oct 21 21:59:26 CEST 2008
198 - disable inotify usage on linux <2.6.25, as it is broken 424 - disable inotify usage on linux <2.6.25, as it is broken
199 (reported by Yoann Vandoorselaere). 425 (reported by Yoann Vandoorselaere).
200 - ev_stat errornously would try to add inotify watchers 426 - ev_stat erroneously would try to add inotify watchers
201 even when inotify wasn't available (this should only 427 even when inotify wasn't available (this should only
202 have a performance impact). 428 have a performance impact).
203 - ev_once now passes both timeout and io to the callback if both 429 - ev_once now passes both timeout and io to the callback if both
204 occur concurrently, instead of giving timeouts precedence. 430 occur concurrently, instead of giving timeouts precedence.
205 - disable EV_USE_INOTIFY when sys/inotify.h is too old. 431 - disable EV_USE_INOTIFY when sys/inotify.h is too old.
… …
273 - enable inlining with gcc versions 2 and 3. 499 - enable inlining with gcc versions 2 and 3.
274 - work around broken poll implementations potentially 500 - work around broken poll implementations potentially
275 not clearing revents field in ev_poll (Brandon Black) 501 not clearing revents field in ev_poll (Brandon Black)
276 (no such systems are known at this time). 502 (no such systems are known at this time).
277 - work around a bug in realloc on openbsd and darwin, 503 - work around a bug in realloc on openbsd and darwin,
278 also makes the errornous valgrind complaints 504 also makes the erroneous valgrind complaints
279 go away (noted by various people). 505 go away (noted by various people).
280 - fix ev_async_pending, add c++ wrapper for ev_async 506 - fix ev_async_pending, add c++ wrapper for ev_async
281 (based on patch sent by Johannes Deisenhofer. 507 (based on patch sent by Johannes Deisenhofer).
282 - add sensible set method to ev::embed. 508 - add sensible set method to ev::embed.
283 - made integer constants type int in ev.h. 509 - made integer constants type int in ev.h.
284 510
2853.2 Wed Apr 2 17:11:19 CEST 2008 5113.2 Wed Apr 2 17:11:19 CEST 2008
286 - fix a 64 bit overflow issue in the select backend, 512 - fix a 64 bit overflow issue in the select backend,

Diff Legend

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