ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/Changes
Revision: 1.252
Committed: Sun Feb 12 08:38:31 2012 UTC (12 years, 3 months ago) by root
Branch: MAIN
Changes since 1.251: +3 -0 lines
Log Message:
undef ev_error

File Contents

# Content
1 Revision history for libev, a high-performance and full-featured event loop.
2
3 TODO: ev_loop_wakeup
4 TODO: EV_STANDALONE == NO_HASSEL (do not use clock_gettime in ev_standalone)
5
6 - ev_once: undef EV_ERROR in ev_kqueue.c, to avoid clashing with
7 libev's EV_ERROR (reported by 191919).
8
9 4.11 Sat Feb 4 19:52:39 CET 2012
10 - INCOMPATIBLE CHANGE: ev_timer_again now clears the pending status, as
11 was documented already, but not implemented in the repeating case.
12 - new compiletime symbols: EV_NO_SMP and EV_NO_THREADS.
13 - fix a race where the workaround against the epoll fork bugs
14 caused signals to not be handled anymore.
15 - correct backend_fudge for most backends, and implement a windows
16 specific workaround to avoid looping because we call both
17 select and Sleep, both with different time resolutions.
18 - document range and guarantees of ev_sleep.
19 - document reasonable ranges for periodics interval and offset.
20 - rename backend_fudge to backend_mintime to avoid future confusion :)
21 - change the default periodic reschedule function to hopefully be more
22 exact and correct even in corner cases or in the far future.
23 - do not rely on -lm anymore: use it when available but use our
24 own floor () if it is missing. This should make it easier to embed,
25 as no external libraries are required.
26 - strategically import macros from libecb and mark rarely-used functions
27 as cache-cold (saving almost 2k code size on typical amd64 setups).
28 - add Symbols.ev and Symbols.event files, that were missing.
29 - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now).
30 - fix #3 "be smart about timeouts" to not "deadlock" when
31 timeout == now, also improve the section overall.
32 - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom.
33 - support new EV_API_STATIC mode to make all libev symbols
34 static.
35 - supply default CFLAGS of -g -O3 with gcc when original CFLAGS
36 were empty.
37
38 4.04 Wed Feb 16 09:01:51 CET 2011
39 - fix two problems in the native win32 backend, where reuse of fd's
40 with different underlying handles caused handles not to be removed
41 or added to the select set (analyzed and tested by Bert Belder).
42 - do no rely on ceil() in ev_e?poll.c.
43 - backport libev to HP-UX versions before 11 v3.
44 - configure did not detect nanosleep and clock_gettime properly when
45 they are available in the libc (as opposed to -lrt).
46
47 4.03 Tue Jan 11 14:37:25 CET 2011
48 - officially support polling files with all backends.
49 - support files, /dev/zero etc. the same way as select in the epoll
50 backend, by generating events on our own.
51 - ports backend: work around solaris bug 6874410 and many related ones
52 (EINTR, maybe more), with no performance loss (note that the solaris
53 bug report is actually wrong, reality is far more bizarre and broken
54 than that).
55 - define EV_READ/EV_WRITE as macros in event.h, as some programs use
56 #ifdef to test for them.
57 - new (experimental) function: ev_feed_signal.
58 - new (to become default) EVFLAG_NOSIGMASK flag.
59 - new EVBACKEND_MASK symbol.
60 - updated COMMON IDIOMS SECTION.
61
62 4.01 Fri Nov 5 21:51:29 CET 2010
63 - automake fucked it up, apparently, --add-missing -f is not quite enough
64 to make it update its files, so 4.00 didn't install ev++.h and
65 event.h on make install. grrr.
66 - ev_loop(count|depth) didn't return anything (Robin Haberkorn).
67 - change EV_UNDEF to 0xffffffff to silence some overzealous compilers.
68 - use "(libev) " prefix for all libev error messages now.
69
70 4.00 Mon Oct 25 12:32:12 CEST 2010
71 - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading.
72 - ev_embed_stop did not correctly stop the watcher (very good
73 testcase by Vladimir Timofeev).
74 - ev_run will now always update the current loop time - it erroneously
75 didn't when idle watchers were active, causing timers not to fire.
76 - fix a bug where a timeout of zero caused the timer not to fire
77 in the libevent emulation (testcase by Péter Szabó).
78 - applied win32 fixes by Michael Lenaghan (also James Mansion).
79 - replace EV_MINIMAL by EV_FEATURES.
80 - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it
81 seems the former is *much* faster than the latter.
82 - linux kernel version detection (for inotify bug workarounds)
83 did not work properly.
84 - reduce the number of spurious wake-ups with the ports backend.
85 - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu).
86 - do async init within ev_async_start, not ev_async_set, which avoids
87 an API quirk where the set function must be called in the C++ API
88 even when there is nothing to set.
89 - add (undocumented) EV_ENABLE when adding events with kqueue,
90 this might help with OS X, which seems to need it despite documenting
91 not to need it (helpfully pointed out by Tilghman Lesher).
92 - do not use poll by default on freebsd, it's broken (what isn't
93 on freebsd...).
94 - allow to embed epoll on kernels >= 2.6.32.
95 - configure now prepends -O3, not appends it, so one can still
96 override it.
97 - ev.pod: greatly expanded the portability section, added a porting
98 section, a description of watcher states and made lots of minor fixes.
99 - disable poll backend on AIX, the poll header spams the namespace
100 and it's not worth working around dead platforms (reported
101 and analyzed by Aivars Kalvans).
102 - improve header file compatibility of the standalone eventfd code
103 in an obscure case.
104 - implement EV_AVOID_STDIO option.
105 - do not use sscanf to parse linux version number (smaller, faster,
106 no sscanf dependency).
107 - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings.
108 - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs.
109 - add section on accept() problems to the manpage.
110 - rename EV_TIMEOUT to EV_TIMER.
111 - rename ev_loop_count/depth/verify/loop/unloop.
112 - remove ev_default_destroy and ev_default_fork.
113 - switch to two-digit minor version.
114 - work around an apparent gentoo compiler bug.
115 - define _DARWIN_UNLIMITED_SELECT. just so.
116 - use enum instead of #define for most constants.
117 - improve compatibility to older C++ compilers.
118 - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now
119 default arguments when compiled as C++.
120 - enable automake dependency tracking.
121 - ev_loop_new no longer leaks memory when loop creation failed.
122 - new ev_cleanup watcher type.
123
124 3.9 Thu Dec 31 07:59:59 CET 2009
125 - signalfd is no longer used by default and has to be requested
126 explicitly - this means that easy to catch bugs become hard to
127 catch race conditions, but the users have spoken.
128 - point out the unspecified signal mask in the documentation, and
129 that this is a race condition regardless of EV_SIGNALFD.
130 - backport inotify code to C89.
131 - inotify file descriptors could leak into child processes.
132 - ev_stat watchers could keep an erroneous extra ref on the loop,
133 preventing exit when unregistering all watchers (testcases
134 provided by ry@tinyclouds.org).
135 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
136 symbols to make it easier for apps to do their own fd management.
137 - support EV_IDLE_ENABLE being disabled in ev++.h
138 (patch by Didier Spezia).
139 - take advantage of inotify_init1, if available, to set cloexec/nonblock
140 on fd creation, to avoid races.
141 - the signal handling pipe wasn't always initialised under windows
142 (analysed by lekma).
143 - changed minimum glibc requirement from glibc 2.9 to 2.7, for
144 signalfd.
145 - add missing string.h include (Denis F. Latypoff).
146 - only replace ev_stat.prev when we detect an actual difference,
147 so prev is (almost) always different to attr. this might
148 have caused the problems with 04_stat.t.
149 - add ev::timer->remaining () method to C++ API.
150
151 3.8 Sun Aug 9 14:30:45 CEST 2009
152 - incompatible change: do not necessarily reset signal handler
153 to SIG_DFL when a sighandler is stopped.
154 - ev_default_destroy did not properly free or zero some members,
155 potentially causing crashes and memory corruption on repeated
156 ev_default_destroy/ev_default_loop calls.
157 - take advantage of signalfd on GNU/Linux systems.
158 - document that the signal mask might be in an unspecified
159 state when using libev's signal handling.
160 - take advantage of some GNU/Linux calls to set cloexec/nonblock
161 on fd creation, to avoid race conditions.
162
163 3.7 Fri Jul 17 16:36:32 CEST 2009
164 - ev_unloop and ev_loop wrongly used a global variable to exit loops,
165 instead of using a per-loop variable (bug caught by accident...).
166 - the ev_set_io_collect_interval interpretation has changed.
167 - add new functionality: ev_set_userdata, ev_userdata,
168 ev_set_invoke_pending_cb, ev_set_loop_release_cb,
169 ev_invoke_pending, ev_pending_count, together with a long example
170 about thread locking.
171 - add ev_timer_remaining (as requested by Denis F. Latypoff).
172 - add ev_loop_depth.
173 - calling ev_unloop in fork/prepare watchers will no longer poll
174 for new events.
175 - Denis F. Latypoff corrected many typos in example code snippets.
176 - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double-
177 check that the syscall number is available before trying to
178 use it (reported by ry@tinyclouds).
179 - use GetSystemTimeAsFileTime instead of _timeb on windows, for
180 slightly higher accuracy.
181 - properly declare ev_loop_verify and ev_now_update even when
182 !EV_MULTIPLICITY.
183 - do not compile in any priority code when EV_MAXPRI == EV_MINPRI.
184 - support EV_MINIMAL==2 for a reduced API.
185 - actually 0-initialise struct sigaction when installing signals.
186 - add section on hibernate and stopped processes to ev_timer docs.
187
188 3.6 Tue Apr 28 02:49:30 CEST 2009
189 - multiple timers becoming ready within an event loop iteration
190 will be invoked in the "correct" order now.
191 - do not leave the event loop early just because we have no active
192 watchers, fixing a problem when embedding a kqueue loop
193 that has active kernel events but no registered watchers
194 (reported by blacksand blacksand).
195 - correctly zero the idx values for arrays, so destroying and
196 reinitialising the default loop actually works (patch by
197 Malek Hadj-Ali).
198 - implement ev_suspend and ev_resume.
199 - new EV_CUSTOM revents flag for use by applications.
200 - add documentation section about priorities.
201 - add a glossary to the dcoumentation.
202 - extend the ev_fork description slightly.
203 - optimize a jump out of call_pending.
204
205 3.53 Sun Feb 15 02:38:20 CET 2009
206 - fix a bug in event pipe creation on win32 that would cause a
207 failed assertion on event loop creation (patch by Malek Hadj-Ali).
208 - probe for CLOCK_REALTIME support at runtime as well and fall
209 back to gettimeofday if there is an error, to support older
210 operating systems with newer header files/libraries.
211 - prefer gettimeofday over clock_gettime with USE_CLOCK_SYSCALL
212 (default most everywhere), otherwise not.
213
214 3.52 Wed Jan 7 21:43:02 CET 2009
215 - fix compilation of select backend in fd_set mode when NFDBITS is
216 missing (to get it to compile on QNX, reported by Rodrigo Campos).
217 - better select-nfds handling when select backend is in fd_set mode.
218 - diagnose fd_set overruns when select backend is in fd_set mode.
219 - due to a thinko, instead of disabling everything but
220 select on the borked OS X platform, everything but select was
221 allowed (reported by Emanuele Giaquinta).
222 - actually verify that local and remote port are matching in
223 libev's socketpair emulation, which makes denial-of-service
224 attacks harder (but not impossible - it's windows). Make sure
225 it even works under vista, which thinks that getpeer/sockname
226 should return fantasy port numbers.
227 - include "libev" in all assertion messages for potentially
228 clearer diagnostics.
229 - event_get_version (libevent compatibility) returned
230 a useless string instead of the expected version string
231 (patch by W.C.A. Wijngaards).
232
233 3.51 Wed Dec 24 23:00:11 CET 2008
234 - fix a bug where an inotify watcher was added twice, causing
235 freezes on hash collisions (reported and analysed by Graham Leggett).
236 - new config symbol, EV_USE_CLOCK_SYSCALL, to make libev use
237 a direct syscall - slower, but no dependency on librt et al.
238 - assume negative return values != -1 signals success of port_getn
239 (http://cvs.epicsol.org/cgi/viewcvs.cgi/epic5/source/newio.c?rev=1.52)
240 (no known failure reports, but it doesn't hurt).
241 - fork detection in ev_embed now stops and restarts the watcher
242 automatically.
243 - EXPERIMENTAL: default the method to operator () in ev++.h,
244 to make it nicer to use functors (requested by Benedek László).
245 - fixed const object callbacks in ev++.h.
246 - replaced loop_ref argument of watcher.set (loop) by a direct
247 ev_loop * in ev++.h, to avoid clashes with functor patch.
248 - do not try to watch the empty string via inotify.
249 - inotify watchers could be leaked under certain circumstances.
250 - OS X 10.5 is actually even more broken than earlier versions,
251 so fall back to select on that piece of garbage.
252 - fixed some weirdness in the ev_embed documentation.
253
254 3.49 Wed Nov 19 11:26:53 CET 2008
255 - ev_stat watchers will now use inotify as a mere hint on
256 kernels <2.6.25, or if the filesystem is not in the
257 "known to be good" list.
258 - better mingw32 compatibility (it's not as borked as native win32)
259 (analysed by Roger Pack).
260 - include stdio.h in the example program, as too many people are
261 confused by the weird C language otherwise. I guess the next thing
262 I get told is that the "..." ellipses in the examples don't compile
263 with their C compiler.
264
265 3.48 Thu Oct 30 09:02:37 CET 2008
266 - further optimise away the EPOLL_CTL_ADD/MOD combo in the epoll
267 backend by assuming the kernel event mask hasn't changed if
268 ADD fails with EEXIST.
269 - work around spurious event notification bugs in epoll by using
270 a 32-bit generation counter. recreate kernel state if we receive
271 spurious notifications or unwanted events. this is very costly,
272 but I didn't come up with this horrible design.
273 - use memset to initialise most arrays now and do away with the
274 init functions.
275 - expand time-out strategies into a "Be smart about timeouts" section.
276 - drop the "struct" from all ev_watcher declarations in the
277 documentation and did other clarifications (yeah, it was a mistake
278 to have a struct AND a function called ev_loop).
279 - fix a bug where ev_default would not initialise the default
280 loop again after it was destroyed with ev_default_destroy.
281 - rename syserr to ev_syserr to avoid name clashes when embedding,
282 do similar changes for event.c.
283
284 3.45 Tue Oct 21 21:59:26 CEST 2008
285 - disable inotify usage on linux <2.6.25, as it is broken
286 (reported by Yoann Vandoorselaere).
287 - ev_stat erroneously would try to add inotify watchers
288 even when inotify wasn't available (this should only
289 have a performance impact).
290 - ev_once now passes both timeout and io to the callback if both
291 occur concurrently, instead of giving timeouts precedence.
292 - disable EV_USE_INOTIFY when sys/inotify.h is too old.
293
294 3.44 Mon Sep 29 05:18:39 CEST 2008
295 - embed watchers now automatically invoke ev_loop_fork on the
296 embedded loop when the parent loop forks.
297 - new function: ev_now_update (loop).
298 - verify_watcher was not marked static.
299 - improve the "associating..." manpage section.
300 - documentation tweaks here and there.
301
302 3.43 Sun Jul 6 05:34:41 CEST 2008
303 - include more include files on windows to get struct _stati64
304 (reported by Chris Hulbert, but doesn't quite fix his issue).
305 - add missing #include <io.h> in ev.c on windows (reported by
306 Matt Tolton).
307
308 3.42 Tue Jun 17 12:12:07 CEST 2008
309 - work around yet another windows bug: FD_SET actually adds fd's
310 multiple times to the fd_*SET*, despite official MSN docs claiming
311 otherwise. Reported and well-analysed by Matt Tolton.
312 - define NFDBITS to 0 when EV_SELECT_IS_WINSOCKET to make it compile
313 (reported any analysed by Chris Hulbert).
314 - fix a bug in ev_ebadf (this function is only used to catch
315 programming errors in the libev user). reported by Matt Tolton.
316 - fix a bug in fd_intern on win32 (could lead to compile errors
317 under some circumstances, but would work correctly if it compiles).
318 reported by Matt Tolton.
319 - (try to) work around missing lstat on windows.
320 - pass in the write fd set as except fd set under windows. windows
321 is so uncontrollably lame that it requires this. this means that
322 switching off oobinline is not supported (but tcp/ip doesn't
323 have oob, so that would be stupid anyways.
324 - use posix module symbol to auto-detect monotonic clock presence
325 and some other default values.
326
327 3.41 Fri May 23 18:42:54 CEST 2008
328 - work around an obscure bug in winsocket select: if you
329 provide only empty fd sets then select returns WSAEINVAL. how sucky.
330 - improve timer scheduling stability and reduce use of time_epsilon.
331 - use 1-based 2-heap for EV_MINIMAL, simplifies code, reduces
332 codesize and makes for better cache-efficiency.
333 - use 3-based 4-heap for !EV_MINIMAL. this makes better use
334 of cpu cache lines and gives better growth behaviour than
335 2-based heaps.
336 - cache timestamp within heap for !EV_MINIMAL, to avoid random
337 memory accesses.
338 - document/add EV_USE_4HEAP and EV_HEAP_CACHE_AT.
339 - fix a potential aliasing issue in ev_timer_again.
340 - add/document ev_periodic_at, retract direct access to ->at.
341 - improve ev_stat docs.
342 - add portability requirements section.
343 - fix manpage headers etc.
344 - normalise WSA error codes to lower range on windows.
345 - add consistency check code that can be called automatically
346 or on demand to check for internal structures (ev_loop_verify).
347
348 3.31 Wed Apr 16 20:45:04 CEST 2008
349 - added last minute fix for ev_poll.c by Brandon Black.
350
351 3.3 Wed Apr 16 19:04:10 CEST 2008
352 - event_base_loopexit should return 0 on success
353 (W.C.A. Wijngaards).
354 - added linux eventfd support.
355 - try to autodetect epoll and inotify support
356 by libc header version if not using autoconf.
357 - new symbols: EV_DEFAULT_UC and EV_DEFAULT_UC_.
358 - declare functions defined in ev.h as inline if
359 C99 or gcc are available.
360 - enable inlining with gcc versions 2 and 3.
361 - work around broken poll implementations potentially
362 not clearing revents field in ev_poll (Brandon Black)
363 (no such systems are known at this time).
364 - work around a bug in realloc on openbsd and darwin,
365 also makes the erroneous valgrind complaints
366 go away (noted by various people).
367 - fix ev_async_pending, add c++ wrapper for ev_async
368 (based on patch sent by Johannes Deisenhofer).
369 - add sensible set method to ev::embed.
370 - made integer constants type int in ev.h.
371
372 3.2 Wed Apr 2 17:11:19 CEST 2008
373 - fix a 64 bit overflow issue in the select backend,
374 by using fd_mask instead of int for the mask.
375 - rename internal sighandler to avoid clash with very old perls.
376 - entering ev_loop will not clear the ONESHOT or NONBLOCKING
377 flags of any outer loops anymore.
378 - add ev_async_pending.
379
380 3.1 Thu Mar 13 13:45:22 CET 2008
381 - implement ev_async watchers.
382 - only initialise signal pipe on demand.
383 - make use of sig_atomic_t configurable.
384 - improved documentation.
385
386 3.0 Mon Jan 28 13:14:47 CET 2008
387 - API/ABI bump to version 3.0.
388 - ev++.h includes "ev.h" by default now, not <ev.h>.
389 - slightly improved documentation.
390 - speed up signal detection after a fork.
391 - only optionally return trace status changed in ev_child
392 watchers.
393 - experimental (and undocumented) loop wrappers for ev++.h.
394
395 2.01 Tue Dec 25 08:04:41 CET 2007
396 - separate Changes file.
397 - fix ev_path_set => ev_stat_set typo.
398 - remove event_compat.h from the libev tarball.
399 - change how include files are found.
400 - doc updates.
401 - update licenses, explicitly allow for GPL relicensing.
402
403 2.0 Sat Dec 22 17:47:03 CET 2007
404 - new ev_sleep, ev_set_(io|timeout)_collect_interval.
405 - removed epoll from embeddable fd set.
406 - fix embed watchers.
407 - renamed ev_embed.loop to other.
408 - added exported Symbol tables.
409 - undefine member wrapper macros at the end of ev.c.
410 - respect EV_H in ev++.h.
411
412 1.86 Tue Dec 18 02:36:57 CET 2007
413 - fix memleak on loop destroy (not relevant for perl).
414
415 1.85 Fri Dec 14 20:32:40 CET 2007
416 - fix some aliasing issues w.r.t. timers and periodics
417 (not relevant for perl).
418
419 (for historic versions refer to EV/Changes, found in the Perl interface)
420
421 0.1 Wed Oct 31 21:31:48 CET 2007
422 - original version; hacked together in <24h.
423