ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/Changes
Revision: 1.251
Committed: Sat Feb 4 19:09:28 2012 UTC (12 years, 2 months ago) by root
Branch: MAIN
CVS Tags: EV_rel-4_11, rel-4_11
Changes since 1.250: +2 -0 lines
Log Message:
*** empty log message ***

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