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

Comparing libev/Changes (file contents):
Revision 1.88 by root, Fri Nov 21 02:10:34 2008 UTC vs.
Revision 1.280 by root, Sun Sep 23 21:21:58 2012 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_loop_wakeup
4TODO: EV_STANDALONE == NO_HASSEL (do not use clock_gettime in ev_standalone)
5TODO: faq, process a thing in each iteration
6TODO: dbeugging tips, ev_verify, ev_init twice
7TODO: ev_break for immediate exit (EVBREAK_NOW?)
8TODO: ev_feed_child_event
9
10TODO: document the special problem of signals around fork.
11TODO: store pid for each signal
12TODO: document file descriptor usage per loop
13TODO: store loop pid_t and compare isndie signal handler,store 1 for same, 2 for differign pid, clean up in loop_fork
14TODO: embed watchers need updating when fd changes
15 - destroying a non-default loop would stop the global waitpid
16 watcher (Denis Bilenko).
17 - queueing pending watchers of higher priority from a watcher now invokes
18 them in a timely fashion (reported by Denis Bilenko).
19 - add throw() to all libev functions that cannot throw exceptions, for
20 further code size decrease when compiling for C++.
21 - add throw () to callbacks that must not throw exceptions (allocator,
22 syserr, loop acquire/release, periodic reschedule cbs).
23 - fix event_base_loop return code, add event_get_callback, event_base_new,
24 event_base_get_method calls to improve libevent 1.x emulation and add
25 some libevent 2.x functionality (based on a patch by Jeff Davey).
26 - add more memory fences to fix a bug reported by Jeff Davey. Better
27 be overfenced than underprotected.
28 - ev_run now returns a boolean status (true meaning watchers are
29 still active).
30 - ev_once: undef EV_ERROR in ev_kqueue.c, to avoid clashing with
31 libev's EV_ERROR (reported by 191919).
32 - (ecb) add memory fence support for xlC (Darin McBride).
33 - (ecb) add memory fence support for gcc-mips (Anton Kirilov).
34 - (ecb) add memory fence support for gcc-alpha (Christian Weisgerber).
35 - work around some kernels losing file descriptors by leaking
36 the kqueue descriptor in the child.
37 - include sys/syscall.h instead of plain syscall.h.
38 - check for io watcher loops in ev_verify, check for the most
39 common reported usage bug in ev_io_start.
40 - chose socket vs. WSASocket at compiletime using EV_USE_WSASOCKET.
41 - always use WSASend/WSARecv directly on windows, hoping that this
42 works in all cases (unlike read/write/send/recv...).
43 - try to detect signals around a fork faster (test program by
44 Denis Bilenko).
45 - work around recent glibc versions that leak memory in realloc.
46 - rename ev::embed::set to ev::embed::set_embed to avoid clashing
47 the watcher base set (loop) method.
48 - rewrite the async/signal pipe logic to always keep a valid fd, which
49 simplifies (and hopefuly correctifies :) the race checking
50 on fork, at the cost of one extra fd.
51
524.11 Sat Feb 4 19:52:39 CET 2012
53 - INCOMPATIBLE CHANGE: ev_timer_again now clears the pending status, as
54 was documented already, but not implemented in the repeating case.
55 - new compiletime symbols: EV_NO_SMP and EV_NO_THREADS.
56 - fix a race where the workaround against the epoll fork bugs
57 caused signals to not be handled anymore.
58 - correct backend_fudge for most backends, and implement a windows
59 specific workaround to avoid looping because we call both
60 select and Sleep, both with different time resolutions.
61 - document range and guarantees of ev_sleep.
62 - document reasonable ranges for periodics interval and offset.
63 - rename backend_fudge to backend_mintime to avoid future confusion :)
64 - change the default periodic reschedule function to hopefully be more
65 exact and correct even in corner cases or in the far future.
66 - do not rely on -lm anymore: use it when available but use our
67 own floor () if it is missing. This should make it easier to embed,
68 as no external libraries are required.
69 - strategically import macros from libecb and mark rarely-used functions
70 as cache-cold (saving almost 2k code size on typical amd64 setups).
71 - add Symbols.ev and Symbols.event files, that were missing.
72 - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now).
73 - fix #3 "be smart about timeouts" to not "deadlock" when
74 timeout == now, also improve the section overall.
75 - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom.
76 - support new EV_API_STATIC mode to make all libev symbols
77 static.
78 - supply default CFLAGS of -g -O3 with gcc when original CFLAGS
79 were empty.
80
814.04 Wed Feb 16 09:01:51 CET 2011
82 - fix two problems in the native win32 backend, where reuse of fd's
83 with different underlying handles caused handles not to be removed
84 or added to the select set (analyzed and tested by Bert Belder).
85 - do no rely on ceil() in ev_e?poll.c.
86 - backport libev to HP-UX versions before 11 v3.
87 - configure did not detect nanosleep and clock_gettime properly when
88 they are available in the libc (as opposed to -lrt).
89
904.03 Tue Jan 11 14:37:25 CET 2011
91 - officially support polling files with all backends.
92 - support files, /dev/zero etc. the same way as select in the epoll
93 backend, by generating events on our own.
94 - ports backend: work around solaris bug 6874410 and many related ones
95 (EINTR, maybe more), with no performance loss (note that the solaris
96 bug report is actually wrong, reality is far more bizarre and broken
97 than that).
98 - define EV_READ/EV_WRITE as macros in event.h, as some programs use
99 #ifdef to test for them.
100 - new (experimental) function: ev_feed_signal.
101 - new (to become default) EVFLAG_NOSIGMASK flag.
102 - new EVBACKEND_MASK symbol.
103 - updated COMMON IDIOMS SECTION.
104
1054.01 Fri Nov 5 21:51:29 CET 2010
106 - automake fucked it up, apparently, --add-missing -f is not quite enough
107 to make it update its files, so 4.00 didn't install ev++.h and
108 event.h on make install. grrr.
109 - ev_loop(count|depth) didn't return anything (Robin Haberkorn).
110 - change EV_UNDEF to 0xffffffff to silence some overzealous compilers.
111 - use "(libev) " prefix for all libev error messages now.
112
1134.00 Mon Oct 25 12:32:12 CEST 2010
114 - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading.
115 - ev_embed_stop did not correctly stop the watcher (very good
116 testcase by Vladimir Timofeev).
117 - ev_run will now always update the current loop time - it erroneously
118 didn't when idle watchers were active, causing timers not to fire.
119 - fix a bug where a timeout of zero caused the timer not to fire
120 in the libevent emulation (testcase by Péter Szabó).
121 - applied win32 fixes by Michael Lenaghan (also James Mansion).
122 - replace EV_MINIMAL by EV_FEATURES.
123 - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it
124 seems the former is *much* faster than the latter.
125 - linux kernel version detection (for inotify bug workarounds)
126 did not work properly.
127 - reduce the number of spurious wake-ups with the ports backend.
128 - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu).
129 - do async init within ev_async_start, not ev_async_set, which avoids
130 an API quirk where the set function must be called in the C++ API
131 even when there is nothing to set.
132 - add (undocumented) EV_ENABLE when adding events with kqueue,
133 this might help with OS X, which seems to need it despite documenting
134 not to need it (helpfully pointed out by Tilghman Lesher).
135 - do not use poll by default on freebsd, it's broken (what isn't
136 on freebsd...).
137 - allow to embed epoll on kernels >= 2.6.32.
138 - configure now prepends -O3, not appends it, so one can still
139 override it.
140 - ev.pod: greatly expanded the portability section, added a porting
141 section, a description of watcher states and made lots of minor fixes.
142 - disable poll backend on AIX, the poll header spams the namespace
143 and it's not worth working around dead platforms (reported
144 and analyzed by Aivars Kalvans).
145 - improve header file compatibility of the standalone eventfd code
146 in an obscure case.
147 - implement EV_AVOID_STDIO option.
148 - do not use sscanf to parse linux version number (smaller, faster,
149 no sscanf dependency).
150 - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings.
151 - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs.
152 - add section on accept() problems to the manpage.
153 - rename EV_TIMEOUT to EV_TIMER.
154 - rename ev_loop_count/depth/verify/loop/unloop.
155 - remove ev_default_destroy and ev_default_fork.
156 - switch to two-digit minor version.
157 - work around an apparent gentoo compiler bug.
158 - define _DARWIN_UNLIMITED_SELECT. just so.
159 - use enum instead of #define for most constants.
160 - improve compatibility to older C++ compilers.
161 - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now
162 default arguments when compiled as C++.
163 - enable automake dependency tracking.
164 - ev_loop_new no longer leaks memory when loop creation failed.
165 - new ev_cleanup watcher type.
166
1673.9 Thu Dec 31 07:59:59 CET 2009
168 - signalfd is no longer used by default and has to be requested
169 explicitly - this means that easy to catch bugs become hard to
170 catch race conditions, but the users have spoken.
171 - point out the unspecified signal mask in the documentation, and
172 that this is a race condition regardless of EV_SIGNALFD.
173 - backport inotify code to C89.
174 - inotify file descriptors could leak into child processes.
175 - ev_stat watchers could keep an erroneous extra ref on the loop,
176 preventing exit when unregistering all watchers (testcases
177 provided by ry@tinyclouds.org).
178 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
179 symbols to make it easier for apps to do their own fd management.
180 - support EV_IDLE_ENABLE being disabled in ev++.h
181 (patch by Didier Spezia).
182 - take advantage of inotify_init1, if available, to set cloexec/nonblock
183 on fd creation, to avoid races.
184 - the signal handling pipe wasn't always initialised under windows
185 (analysed by lekma).
186 - changed minimum glibc requirement from glibc 2.9 to 2.7, for
187 signalfd.
188 - add missing string.h include (Denis F. Latypoff).
189 - only replace ev_stat.prev when we detect an actual difference,
190 so prev is (almost) always different to attr. this might
191 have caused the problems with 04_stat.t.
192 - add ev::timer->remaining () method to C++ API.
193
1943.8 Sun Aug 9 14:30:45 CEST 2009
195 - incompatible change: do not necessarily reset signal handler
196 to SIG_DFL when a sighandler is stopped.
197 - ev_default_destroy did not properly free or zero some members,
198 potentially causing crashes and memory corruption on repeated
199 ev_default_destroy/ev_default_loop calls.
200 - take advantage of signalfd on GNU/Linux systems.
201 - document that the signal mask might be in an unspecified
202 state when using libev's signal handling.
203 - take advantage of some GNU/Linux calls to set cloexec/nonblock
204 on fd creation, to avoid race conditions.
205
2063.7 Fri Jul 17 16:36:32 CEST 2009
207 - ev_unloop and ev_loop wrongly used a global variable to exit loops,
208 instead of using a per-loop variable (bug caught by accident...).
209 - the ev_set_io_collect_interval interpretation has changed.
210 - add new functionality: ev_set_userdata, ev_userdata,
211 ev_set_invoke_pending_cb, ev_set_loop_release_cb,
212 ev_invoke_pending, ev_pending_count, together with a long example
213 about thread locking.
214 - add ev_timer_remaining (as requested by Denis F. Latypoff).
215 - add ev_loop_depth.
216 - calling ev_unloop in fork/prepare watchers will no longer poll
217 for new events.
218 - Denis F. Latypoff corrected many typos in example code snippets.
219 - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double-
220 check that the syscall number is available before trying to
221 use it (reported by ry@tinyclouds).
222 - use GetSystemTimeAsFileTime instead of _timeb on windows, for
223 slightly higher accuracy.
224 - properly declare ev_loop_verify and ev_now_update even when
225 !EV_MULTIPLICITY.
226 - do not compile in any priority code when EV_MAXPRI == EV_MINPRI.
227 - support EV_MINIMAL==2 for a reduced API.
228 - actually 0-initialise struct sigaction when installing signals.
229 - add section on hibernate and stopped processes to ev_timer docs.
230
2313.6 Tue Apr 28 02:49:30 CEST 2009
232 - multiple timers becoming ready within an event loop iteration
233 will be invoked in the "correct" order now.
234 - do not leave the event loop early just because we have no active
235 watchers, fixing a problem when embedding a kqueue loop
236 that has active kernel events but no registered watchers
237 (reported by blacksand blacksand).
238 - correctly zero the idx values for arrays, so destroying and
239 reinitialising the default loop actually works (patch by
240 Malek Hadj-Ali).
241 - implement ev_suspend and ev_resume.
242 - new EV_CUSTOM revents flag for use by applications.
243 - add documentation section about priorities.
244 - add a glossary to the dcoumentation.
245 - extend the ev_fork description slightly.
246 - optimize a jump out of call_pending.
247
2483.53 Sun Feb 15 02:38:20 CET 2009
249 - fix a bug in event pipe creation on win32 that would cause a
250 failed assertion on event loop creation (patch by Malek Hadj-Ali).
251 - probe for CLOCK_REALTIME support at runtime as well and fall
252 back to gettimeofday if there is an error, to support older
253 operating systems with newer header files/libraries.
254 - prefer gettimeofday over clock_gettime with USE_CLOCK_SYSCALL
255 (default most everywhere), otherwise not.
256
2573.52 Wed Jan 7 21:43:02 CET 2009
258 - fix compilation of select backend in fd_set mode when NFDBITS is
259 missing (to get it to compile on QNX, reported by Rodrigo Campos).
260 - better select-nfds handling when select backend is in fd_set mode.
261 - diagnose fd_set overruns when select backend is in fd_set mode.
262 - due to a thinko, instead of disabling everything but
263 select on the borked OS X platform, everything but select was
264 allowed (reported by Emanuele Giaquinta).
265 - actually verify that local and remote port are matching in
266 libev's socketpair emulation, which makes denial-of-service
267 attacks harder (but not impossible - it's windows). Make sure
268 it even works under vista, which thinks that getpeer/sockname
269 should return fantasy port numbers.
270 - include "libev" in all assertion messages for potentially
271 clearer diagnostics.
272 - event_get_version (libevent compatibility) returned
273 a useless string instead of the expected version string
274 (patch by W.C.A. Wijngaards).
275
2763.51 Wed Dec 24 23:00:11 CET 2008
277 - fix a bug where an inotify watcher was added twice, causing
278 freezes on hash collisions (reported and analysed by Graham Leggett).
3 - new config symbol, EV_USE_CLOCK_SYSCALL, to make libev use 279 - new config symbol, EV_USE_CLOCK_SYSCALL, to make libev use
4 a direct syscall - slower, but no dependency on librt et al. 280 a direct syscall - slower, but no dependency on librt et al.
5 - assume negative return values != -1 signals success of port_getn 281 - assume negative return values != -1 signals success of port_getn
6 (http://cvs.epicsol.org/cgi/viewcvs.cgi/epic5/source/newio.c?rev=1.52) 282 (http://cvs.epicsol.org/cgi/viewcvs.cgi/epic5/source/newio.c?rev=1.52)
7 (no known failure reports, but it doesn't hurt). 283 (no known failure reports, but it doesn't hurt).
284 - fork detection in ev_embed now stops and restarts the watcher
285 automatically.
286 - EXPERIMENTAL: default the method to operator () in ev++.h,
287 to make it nicer to use functors (requested by Benedek László).
288 - fixed const object callbacks in ev++.h.
289 - replaced loop_ref argument of watcher.set (loop) by a direct
290 ev_loop * in ev++.h, to avoid clashes with functor patch.
291 - do not try to watch the empty string via inotify.
292 - inotify watchers could be leaked under certain circumstances.
293 - OS X 10.5 is actually even more broken than earlier versions,
294 so fall back to select on that piece of garbage.
295 - fixed some weirdness in the ev_embed documentation.
8 296
93.49 Wed Nov 19 11:26:53 CET 2008 2973.49 Wed Nov 19 11:26:53 CET 2008
10 - ev_stat watchers will now use inotify as a mere hint on 298 - ev_stat watchers will now use inotify as a mere hint on
11 kernels <2.6.25, or if the filesystem is not in the 299 kernels <2.6.25, or if the filesystem is not in the
12 "known to be good" list. 300 "known to be good" list.
… …
37 do similar changes for event.c. 325 do similar changes for event.c.
38 326
393.45 Tue Oct 21 21:59:26 CEST 2008 3273.45 Tue Oct 21 21:59:26 CEST 2008
40 - disable inotify usage on linux <2.6.25, as it is broken 328 - disable inotify usage on linux <2.6.25, as it is broken
41 (reported by Yoann Vandoorselaere). 329 (reported by Yoann Vandoorselaere).
42 - ev_stat errornously would try to add inotify watchers 330 - ev_stat erroneously would try to add inotify watchers
43 even when inotify wasn't available (this should only 331 even when inotify wasn't available (this should only
44 have a performance impact). 332 have a performance impact).
45 - ev_once now passes both timeout and io to the callback if both 333 - ev_once now passes both timeout and io to the callback if both
46 occur concurrently, instead of giving timeouts precedence. 334 occur concurrently, instead of giving timeouts precedence.
47 - disable EV_USE_INOTIFY when sys/inotify.h is too old. 335 - disable EV_USE_INOTIFY when sys/inotify.h is too old.
… …
115 - enable inlining with gcc versions 2 and 3. 403 - enable inlining with gcc versions 2 and 3.
116 - work around broken poll implementations potentially 404 - work around broken poll implementations potentially
117 not clearing revents field in ev_poll (Brandon Black) 405 not clearing revents field in ev_poll (Brandon Black)
118 (no such systems are known at this time). 406 (no such systems are known at this time).
119 - work around a bug in realloc on openbsd and darwin, 407 - work around a bug in realloc on openbsd and darwin,
120 also makes the errornous valgrind complaints 408 also makes the erroneous valgrind complaints
121 go away (noted by various people). 409 go away (noted by various people).
122 - fix ev_async_pending, add c++ wrapper for ev_async 410 - fix ev_async_pending, add c++ wrapper for ev_async
123 (based on patch sent by Johannes Deisenhofer. 411 (based on patch sent by Johannes Deisenhofer).
124 - add sensible set method to ev::embed. 412 - add sensible set method to ev::embed.
125 - made integer constants type int in ev.h. 413 - made integer constants type int in ev.h.
126 414
1273.2 Wed Apr 2 17:11:19 CEST 2008 4153.2 Wed Apr 2 17:11:19 CEST 2008
128 - fix a 64 bit overflow issue in the select backend, 416 - fix a 64 bit overflow issue in the select backend,

Diff Legend

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