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

Comparing libev/Changes (file contents):
Revision 1.75 by root, Mon Oct 27 11:08:29 2008 UTC vs.
Revision 1.247 by root, Tue Dec 20 20:04:51 2011 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: ev_loop_wakeup
4 4
5TODO: EV_NO_THREADS/NO_SMP and/or EV_USE_PTHREADS or so. also document.
6TODO: EV_STANDALONE == NO_HASSEL (do not use clock_gettime in ev_standalone)
74.10
8 - fix a race where the workaround against the epoll fork bugs
9 caused signals to not be handled anymore.
10 - correct backend_fudge for most backends, and implement a windows
11 specific workaround to avoid looping because we call both
12 select and Sleep, both with different time resolutions.
13 - document range and guarantees of ev_sleep.
14 - document reasonable ranges for periodics interval and offset.
15 - rename backend_fudge to backend_mintime to avoid future confusion :)
16 - change the default periodic reschedule function to hopefully be more
17 exact and correct even in corner cases or in the far future.
18 - do not rely on -lm anymore: use it when available but use our
19 own floor () if it is missing. This should make it easier to embed,
20 as no external libraries are required.
21 - strategically import macros from libecb and mark rarely-used functions
22 as cache-cold (saving almost 2k code size on typical amd64 setups).
23 - add Symbols.ev and Symbols.event files, that were missing.
24 - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now).
25 - fix #3 "be smart about timeouts" to not "deadlock" when
26 timeout == now, also improve the section overall.
27 - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom.
28 - support new EV_API_STATIC mode to make all libev symbols
29 static.
30
314.04 Wed Feb 16 09:01:51 CET 2011
32 - fix two problems in the native win32 backend, where reuse of fd's
33 with different underlying handles caused handles not to be removed
34 or added to the select set (analyzed and tested by Bert Belder).
35 - do no rely on ceil() in ev_e?poll.c.
36 - backport libev to HP-UX versions before 11 v3.
37 - configure did not detect nanosleep and clock_gettime properly when
38 they are available in the libc (as opposed to -lrt).
39
404.03 Tue Jan 11 14:37:25 CET 2011
41 - officially support polling files with all backends.
42 - support files, /dev/zero etc. the same way as select in the epoll
43 backend, by generating events on our own.
44 - ports backend: work around solaris bug 6874410 and many related ones
45 (EINTR, maybe more), with no performance loss (note that the solaris
46 bug report is actually wrong, reality is far more bizarre and broken
47 than that).
48 - define EV_READ/EV_WRITE as macros in event.h, as some programs use
49 #ifdef to test for them.
50 - new (experimental) function: ev_feed_signal.
51 - new (to become default) EVFLAG_NOSIGMASK flag.
52 - new EVBACKEND_MASK symbol.
53 - updated COMMON IDIOMS SECTION.
54
554.01 Fri Nov 5 21:51:29 CET 2010
56 - automake fucked it up, apparently, --add-missing -f is not quite enough
57 to make it update its files, so 4.00 didn't install ev++.h and
58 event.h on make install. grrr.
59 - ev_loop(count|depth) didn't return anything (Robin Haberkorn).
60 - change EV_UNDEF to 0xffffffff to silence some overzealous compilers.
61 - use "(libev) " prefix for all libev error messages now.
62
634.00 Mon Oct 25 12:32:12 CEST 2010
64 - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading.
65 - ev_embed_stop did not correctly stop the watcher (very good
66 testcase by Vladimir Timofeev).
67 - ev_run will now always update the current loop time - it erroneously
68 didn't when idle watchers were active, causing timers not to fire.
69 - fix a bug where a timeout of zero caused the timer not to fire
70 in the libevent emulation (testcase by Péter Szabó).
71 - applied win32 fixes by Michael Lenaghan (also James Mansion).
72 - replace EV_MINIMAL by EV_FEATURES.
73 - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it
74 seems the former is *much* faster than the latter.
75 - linux kernel version detection (for inotify bug workarounds)
76 did not work properly.
77 - reduce the number of spurious wake-ups with the ports backend.
78 - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu).
79 - do async init within ev_async_start, not ev_async_set, which avoids
80 an API quirk where the set function must be called in the C++ API
81 even when there is nothing to set.
82 - add (undocumented) EV_ENABLE when adding events with kqueue,
83 this might help with OS X, which seems to need it despite documenting
84 not to need it (helpfully pointed out by Tilghman Lesher).
85 - do not use poll by default on freebsd, it's broken (what isn't
86 on freebsd...).
87 - allow to embed epoll on kernels >= 2.6.32.
88 - configure now prepends -O3, not appends it, so one can still
89 override it.
90 - ev.pod: greatly expanded the portability section, added a porting
91 section, a description of watcher states and made lots of minor fixes.
92 - disable poll backend on AIX, the poll header spams the namespace
93 and it's not worth working around dead platforms (reported
94 and analyzed by Aivars Kalvans).
95 - improve header file compatibility of the standalone eventfd code
96 in an obscure case.
97 - implement EV_AVOID_STDIO option.
98 - do not use sscanf to parse linux version number (smaller, faster,
99 no sscanf dependency).
100 - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings.
101 - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs.
102 - add section on accept() problems to the manpage.
103 - rename EV_TIMEOUT to EV_TIMER.
104 - rename ev_loop_count/depth/verify/loop/unloop.
105 - remove ev_default_destroy and ev_default_fork.
106 - switch to two-digit minor version.
107 - work around an apparent gentoo compiler bug.
108 - define _DARWIN_UNLIMITED_SELECT. just so.
109 - use enum instead of #define for most constants.
110 - improve compatibility to older C++ compilers.
111 - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now
112 default arguments when compiled as C++.
113 - enable automake dependency tracking.
114 - ev_loop_new no longer leaks memory when loop creation failed.
115 - new ev_cleanup watcher type.
116
1173.9 Thu Dec 31 07:59:59 CET 2009
118 - signalfd is no longer used by default and has to be requested
119 explicitly - this means that easy to catch bugs become hard to
120 catch race conditions, but the users have spoken.
121 - point out the unspecified signal mask in the documentation, and
122 that this is a race condition regardless of EV_SIGNALFD.
123 - backport inotify code to C89.
124 - inotify file descriptors could leak into child processes.
125 - ev_stat watchers could keep an erroneous extra ref on the loop,
126 preventing exit when unregistering all watchers (testcases
127 provided by ry@tinyclouds.org).
128 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
129 symbols to make it easier for apps to do their own fd management.
130 - support EV_IDLE_ENABLE being disabled in ev++.h
131 (patch by Didier Spezia).
132 - take advantage of inotify_init1, if available, to set cloexec/nonblock
133 on fd creation, to avoid races.
134 - the signal handling pipe wasn't always initialised under windows
135 (analysed by lekma).
136 - changed minimum glibc requirement from glibc 2.9 to 2.7, for
137 signalfd.
138 - add missing string.h include (Denis F. Latypoff).
139 - only replace ev_stat.prev when we detect an actual difference,
140 so prev is (almost) always different to attr. this might
141 have caused the problems with 04_stat.t.
142 - add ev::timer->remaining () method to C++ API.
143
1443.8 Sun Aug 9 14:30:45 CEST 2009
145 - incompatible change: do not necessarily reset signal handler
146 to SIG_DFL when a sighandler is stopped.
147 - ev_default_destroy did not properly free or zero some members,
148 potentially causing crashes and memory corruption on repeated
149 ev_default_destroy/ev_default_loop calls.
150 - take advantage of signalfd on GNU/Linux systems.
151 - document that the signal mask might be in an unspecified
152 state when using libev's signal handling.
153 - take advantage of some GNU/Linux calls to set cloexec/nonblock
154 on fd creation, to avoid race conditions.
155
1563.7 Fri Jul 17 16:36:32 CEST 2009
157 - ev_unloop and ev_loop wrongly used a global variable to exit loops,
158 instead of using a per-loop variable (bug caught by accident...).
159 - the ev_set_io_collect_interval interpretation has changed.
160 - add new functionality: ev_set_userdata, ev_userdata,
161 ev_set_invoke_pending_cb, ev_set_loop_release_cb,
162 ev_invoke_pending, ev_pending_count, together with a long example
163 about thread locking.
164 - add ev_timer_remaining (as requested by Denis F. Latypoff).
165 - add ev_loop_depth.
166 - calling ev_unloop in fork/prepare watchers will no longer poll
167 for new events.
168 - Denis F. Latypoff corrected many typos in example code snippets.
169 - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double-
170 check that the syscall number is available before trying to
171 use it (reported by ry@tinyclouds).
172 - use GetSystemTimeAsFileTime instead of _timeb on windows, for
173 slightly higher accuracy.
174 - properly declare ev_loop_verify and ev_now_update even when
175 !EV_MULTIPLICITY.
176 - do not compile in any priority code when EV_MAXPRI == EV_MINPRI.
177 - support EV_MINIMAL==2 for a reduced API.
178 - actually 0-initialise struct sigaction when installing signals.
179 - add section on hibernate and stopped processes to ev_timer docs.
180
1813.6 Tue Apr 28 02:49:30 CEST 2009
182 - multiple timers becoming ready within an event loop iteration
183 will be invoked in the "correct" order now.
184 - do not leave the event loop early just because we have no active
185 watchers, fixing a problem when embedding a kqueue loop
186 that has active kernel events but no registered watchers
187 (reported by blacksand blacksand).
188 - correctly zero the idx values for arrays, so destroying and
189 reinitialising the default loop actually works (patch by
190 Malek Hadj-Ali).
191 - implement ev_suspend and ev_resume.
192 - new EV_CUSTOM revents flag for use by applications.
193 - add documentation section about priorities.
194 - add a glossary to the dcoumentation.
195 - extend the ev_fork description slightly.
196 - optimize a jump out of call_pending.
197
1983.53 Sun Feb 15 02:38:20 CET 2009
199 - fix a bug in event pipe creation on win32 that would cause a
200 failed assertion on event loop creation (patch by Malek Hadj-Ali).
201 - probe for CLOCK_REALTIME support at runtime as well and fall
202 back to gettimeofday if there is an error, to support older
203 operating systems with newer header files/libraries.
204 - prefer gettimeofday over clock_gettime with USE_CLOCK_SYSCALL
205 (default most everywhere), otherwise not.
206
2073.52 Wed Jan 7 21:43:02 CET 2009
208 - fix compilation of select backend in fd_set mode when NFDBITS is
209 missing (to get it to compile on QNX, reported by Rodrigo Campos).
210 - better select-nfds handling when select backend is in fd_set mode.
211 - diagnose fd_set overruns when select backend is in fd_set mode.
212 - due to a thinko, instead of disabling everything but
213 select on the borked OS X platform, everything but select was
214 allowed (reported by Emanuele Giaquinta).
215 - actually verify that local and remote port are matching in
216 libev's socketpair emulation, which makes denial-of-service
217 attacks harder (but not impossible - it's windows). Make sure
218 it even works under vista, which thinks that getpeer/sockname
219 should return fantasy port numbers.
220 - include "libev" in all assertion messages for potentially
221 clearer diagnostics.
222 - event_get_version (libevent compatibility) returned
223 a useless string instead of the expected version string
224 (patch by W.C.A. Wijngaards).
225
2263.51 Wed Dec 24 23:00:11 CET 2008
227 - fix a bug where an inotify watcher was added twice, causing
228 freezes on hash collisions (reported and analysed by Graham Leggett).
229 - new config symbol, EV_USE_CLOCK_SYSCALL, to make libev use
230 a direct syscall - slower, but no dependency on librt et al.
231 - assume negative return values != -1 signals success of port_getn
232 (http://cvs.epicsol.org/cgi/viewcvs.cgi/epic5/source/newio.c?rev=1.52)
233 (no known failure reports, but it doesn't hurt).
234 - fork detection in ev_embed now stops and restarts the watcher
235 automatically.
236 - EXPERIMENTAL: default the method to operator () in ev++.h,
237 to make it nicer to use functors (requested by Benedek László).
238 - fixed const object callbacks in ev++.h.
239 - replaced loop_ref argument of watcher.set (loop) by a direct
240 ev_loop * in ev++.h, to avoid clashes with functor patch.
241 - do not try to watch the empty string via inotify.
242 - inotify watchers could be leaked under certain circumstances.
243 - OS X 10.5 is actually even more broken than earlier versions,
244 so fall back to select on that piece of garbage.
245 - fixed some weirdness in the ev_embed documentation.
246
2473.49 Wed Nov 19 11:26:53 CET 2008
248 - ev_stat watchers will now use inotify as a mere hint on
249 kernels <2.6.25, or if the filesystem is not in the
250 "known to be good" list.
251 - better mingw32 compatibility (it's not as borked as native win32)
252 (analysed by Roger Pack).
253 - include stdio.h in the example program, as too many people are
254 confused by the weird C language otherwise. I guess the next thing
255 I get told is that the "..." ellipses in the examples don't compile
256 with their C compiler.
257
2583.48 Thu Oct 30 09:02:37 CET 2008
5 - further optimise away the EPOLL_CTL_ADD/MOD combo in the epoll 259 - further optimise away the EPOLL_CTL_ADD/MOD combo in the epoll
6 backend by assuming the kernel event mask hasn't changed if 260 backend by assuming the kernel event mask hasn't changed if
7 ADD fails with EEXIST. 261 ADD fails with EEXIST.
8 - work around spurious event notification bugs in epoll by using 262 - work around spurious event notification bugs in epoll by using
9 an 8-bit generation counter. 263 a 32-bit generation counter. recreate kernel state if we receive
264 spurious notifications or unwanted events. this is very costly,
265 but I didn't come up with this horrible design.
10 - use memset to initialise most arrays now and do away with the 266 - use memset to initialise most arrays now and do away with the
11 init functions. 267 init functions.
12 - expand time-out strategies into a "Be smart about timeouts" section. 268 - expand time-out strategies into a "Be smart about timeouts" section.
13 - drop the "struct" from all ev_watcher declarations in the 269 - drop the "struct" from all ev_watcher declarations in the
14 documentation and did other clarifications (yeah, it was a mistake 270 documentation and did other clarifications (yeah, it was a mistake
15 to have a struct AND a function called ev_loop). 271 to have a struct AND a function called ev_loop).
16 - fix a bug where ev_default would not initialise the default 272 - fix a bug where ev_default would not initialise the default
17 loop again after it was destroyed with ev_default_destroy. 273 loop again after it was destroyed with ev_default_destroy.
274 - rename syserr to ev_syserr to avoid name clashes when embedding,
275 do similar changes for event.c.
18 276
193.45 Tue Oct 21 21:59:26 CEST 2008 2773.45 Tue Oct 21 21:59:26 CEST 2008
20 - disable inotify usage on linux <2.6.25, as it is broken 278 - disable inotify usage on linux <2.6.25, as it is broken
21 (reported by Yoann Vandoorselaere). 279 (reported by Yoann Vandoorselaere).
22 - ev_stat errornously would try to add inotify watchers 280 - ev_stat erroneously would try to add inotify watchers
23 even when inotify wasn't available (this should only 281 even when inotify wasn't available (this should only
24 have a performance impact). 282 have a performance impact).
25 - ev_once now passes both timeout and io to the callback if both 283 - ev_once now passes both timeout and io to the callback if both
26 occur concurrently, instead of giving timeouts precedence. 284 occur concurrently, instead of giving timeouts precedence.
27 - disable EV_USE_INOTIFY when sys/inotify.h is too old. 285 - disable EV_USE_INOTIFY when sys/inotify.h is too old.
… …
95 - enable inlining with gcc versions 2 and 3. 353 - enable inlining with gcc versions 2 and 3.
96 - work around broken poll implementations potentially 354 - work around broken poll implementations potentially
97 not clearing revents field in ev_poll (Brandon Black) 355 not clearing revents field in ev_poll (Brandon Black)
98 (no such systems are known at this time). 356 (no such systems are known at this time).
99 - work around a bug in realloc on openbsd and darwin, 357 - work around a bug in realloc on openbsd and darwin,
100 also makes the errornous valgrind complaints 358 also makes the erroneous valgrind complaints
101 go away (noted by various people). 359 go away (noted by various people).
102 - fix ev_async_pending, add c++ wrapper for ev_async 360 - fix ev_async_pending, add c++ wrapper for ev_async
103 (based on patch sent by Johannes Deisenhofer. 361 (based on patch sent by Johannes Deisenhofer).
104 - add sensible set method to ev::embed. 362 - add sensible set method to ev::embed.
105 - made integer constants type int in ev.h. 363 - made integer constants type int in ev.h.
106 364
1073.2 Wed Apr 2 17:11:19 CEST 2008 3653.2 Wed Apr 2 17:11:19 CEST 2008
108 - fix a 64 bit overflow issue in the select backend, 366 - fix a 64 bit overflow issue in the select backend,

Diff Legend

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