ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/Changes
Revision: 1.287
Committed: Fri Mar 1 11:13:22 2013 UTC (11 years, 4 months ago) by root
Branch: MAIN
CVS Tags: rel-4_15
Changes since 1.286: +2 -1 lines
Log Message:
4.15

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 TODO: faq, process a thing in each iteration
6 TODO: dbeugging tips, ev_verify, ev_init twice
7 TODO: ev_break for immediate exit (EVBREAK_NOW?)
8 TODO: ev_feed_child_event
9 TODO: document the special problem of signals around fork.
10 TODO: store pid for each signal
11 TODO: document file descriptor usage per loop
12 TODO: store loop pid_t and compare isndie signal handler,store 1 for same, 2 for differign pid, clean up in loop_fork
13 TODO: embed watchers need updating when fd changes
14 TODO: document portbaility requirements for atomic pointer access
15
16 4.15 Fri Mar 1 12:04:50 CET 2013
17 - destroying a non-default loop would stop the global waitpid
18 watcher (Denis Bilenko).
19 - queueing pending watchers of higher priority from a watcher now invokes
20 them in a timely fashion (reported by Denis Bilenko).
21 - add throw() to all libev functions that cannot throw exceptions, for
22 further code size decrease when compiling for C++.
23 - add throw () to callbacks that must not throw exceptions (allocator,
24 syserr, loop acquire/release, periodic reschedule cbs).
25 - fix event_base_loop return code, add event_get_callback, event_base_new,
26 event_base_get_method calls to improve libevent 1.x emulation and add
27 some libevent 2.x functionality (based on a patch by Jeff Davey).
28 - add more memory fences to fix a bug reported by Jeff Davey. Better
29 be overfenced than underprotected.
30 - ev_run now returns a boolean status (true meaning watchers are
31 still active).
32 - ev_once: undef EV_ERROR in ev_kqueue.c, to avoid clashing with
33 libev's EV_ERROR (reported by 191919).
34 - (ecb) add memory fence support for xlC (Darin McBride).
35 - (ecb) add memory fence support for gcc-mips (Anton Kirilov).
36 - (ecb) add memory fence support for gcc-alpha (Christian Weisgerber).
37 - work around some kernels losing file descriptors by leaking
38 the kqueue descriptor in the child.
39 - work around linux inotify not reporting IN_ATTRIB changes for directories
40 in many cases.
41 - include sys/syscall.h instead of plain syscall.h.
42 - check for io watcher loops in ev_verify, check for the most
43 common reported usage bug in ev_io_start.
44 - choose socket vs. WSASocket at compiletime using EV_USE_WSASOCKET.
45 - always use WSASend/WSARecv directly on windows, hoping that this
46 works in all cases (unlike read/write/send/recv...).
47 - try to detect signals around a fork faster (test program by
48 Denis Bilenko).
49 - work around recent glibc versions that leak memory in realloc.
50 - rename ev::embed::set to ev::embed::set_embed to avoid clashing
51 the watcher base set (loop) method.
52 - rewrite the async/signal pipe logic to always keep a valid fd, which
53 simplifies (and hopefully correctifies :) the race checking
54 on fork, at the cost of one extra fd.
55 - add fat, msdos, jffs2, ramfs, ntfs and btrfs to the list of
56 inotify-supporting filesystems.
57 - move orig_CFLAGS assignment to after AC_INIT, as newer autoconf
58 versions ignore it before
59 (https://bugzilla.redhat.com/show_bug.cgi?id=908096).
60 - add some untested android support.
61 - enum expressions must be of type int (reported by Juan Pablo L).
62
63 4.11 Sat Feb 4 19:52:39 CET 2012
64 - INCOMPATIBLE CHANGE: ev_timer_again now clears the pending status, as
65 was documented already, but not implemented in the repeating case.
66 - new compiletime symbols: EV_NO_SMP and EV_NO_THREADS.
67 - fix a race where the workaround against the epoll fork bugs
68 caused signals to not be handled anymore.
69 - correct backend_fudge for most backends, and implement a windows
70 specific workaround to avoid looping because we call both
71 select and Sleep, both with different time resolutions.
72 - document range and guarantees of ev_sleep.
73 - document reasonable ranges for periodics interval and offset.
74 - rename backend_fudge to backend_mintime to avoid future confusion :)
75 - change the default periodic reschedule function to hopefully be more
76 exact and correct even in corner cases or in the far future.
77 - do not rely on -lm anymore: use it when available but use our
78 own floor () if it is missing. This should make it easier to embed,
79 as no external libraries are required.
80 - strategically import macros from libecb and mark rarely-used functions
81 as cache-cold (saving almost 2k code size on typical amd64 setups).
82 - add Symbols.ev and Symbols.event files, that were missing.
83 - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now).
84 - fix #3 "be smart about timeouts" to not "deadlock" when
85 timeout == now, also improve the section overall.
86 - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom.
87 - support new EV_API_STATIC mode to make all libev symbols
88 static.
89 - supply default CFLAGS of -g -O3 with gcc when original CFLAGS
90 were empty.
91
92 4.04 Wed Feb 16 09:01:51 CET 2011
93 - fix two problems in the native win32 backend, where reuse of fd's
94 with different underlying handles caused handles not to be removed
95 or added to the select set (analyzed and tested by Bert Belder).
96 - do no rely on ceil() in ev_e?poll.c.
97 - backport libev to HP-UX versions before 11 v3.
98 - configure did not detect nanosleep and clock_gettime properly when
99 they are available in the libc (as opposed to -lrt).
100
101 4.03 Tue Jan 11 14:37:25 CET 2011
102 - officially support polling files with all backends.
103 - support files, /dev/zero etc. the same way as select in the epoll
104 backend, by generating events on our own.
105 - ports backend: work around solaris bug 6874410 and many related ones
106 (EINTR, maybe more), with no performance loss (note that the solaris
107 bug report is actually wrong, reality is far more bizarre and broken
108 than that).
109 - define EV_READ/EV_WRITE as macros in event.h, as some programs use
110 #ifdef to test for them.
111 - new (experimental) function: ev_feed_signal.
112 - new (to become default) EVFLAG_NOSIGMASK flag.
113 - new EVBACKEND_MASK symbol.
114 - updated COMMON IDIOMS SECTION.
115
116 4.01 Fri Nov 5 21:51:29 CET 2010
117 - automake fucked it up, apparently, --add-missing -f is not quite enough
118 to make it update its files, so 4.00 didn't install ev++.h and
119 event.h on make install. grrr.
120 - ev_loop(count|depth) didn't return anything (Robin Haberkorn).
121 - change EV_UNDEF to 0xffffffff to silence some overzealous compilers.
122 - use "(libev) " prefix for all libev error messages now.
123
124 4.00 Mon Oct 25 12:32:12 CEST 2010
125 - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading.
126 - ev_embed_stop did not correctly stop the watcher (very good
127 testcase by Vladimir Timofeev).
128 - ev_run will now always update the current loop time - it erroneously
129 didn't when idle watchers were active, causing timers not to fire.
130 - fix a bug where a timeout of zero caused the timer not to fire
131 in the libevent emulation (testcase by Péter Szabó).
132 - applied win32 fixes by Michael Lenaghan (also James Mansion).
133 - replace EV_MINIMAL by EV_FEATURES.
134 - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it
135 seems the former is *much* faster than the latter.
136 - linux kernel version detection (for inotify bug workarounds)
137 did not work properly.
138 - reduce the number of spurious wake-ups with the ports backend.
139 - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu).
140 - do async init within ev_async_start, not ev_async_set, which avoids
141 an API quirk where the set function must be called in the C++ API
142 even when there is nothing to set.
143 - add (undocumented) EV_ENABLE when adding events with kqueue,
144 this might help with OS X, which seems to need it despite documenting
145 not to need it (helpfully pointed out by Tilghman Lesher).
146 - do not use poll by default on freebsd, it's broken (what isn't
147 on freebsd...).
148 - allow to embed epoll on kernels >= 2.6.32.
149 - configure now prepends -O3, not appends it, so one can still
150 override it.
151 - ev.pod: greatly expanded the portability section, added a porting
152 section, a description of watcher states and made lots of minor fixes.
153 - disable poll backend on AIX, the poll header spams the namespace
154 and it's not worth working around dead platforms (reported
155 and analyzed by Aivars Kalvans).
156 - improve header file compatibility of the standalone eventfd code
157 in an obscure case.
158 - implement EV_AVOID_STDIO option.
159 - do not use sscanf to parse linux version number (smaller, faster,
160 no sscanf dependency).
161 - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings.
162 - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs.
163 - add section on accept() problems to the manpage.
164 - rename EV_TIMEOUT to EV_TIMER.
165 - rename ev_loop_count/depth/verify/loop/unloop.
166 - remove ev_default_destroy and ev_default_fork.
167 - switch to two-digit minor version.
168 - work around an apparent gentoo compiler bug.
169 - define _DARWIN_UNLIMITED_SELECT. just so.
170 - use enum instead of #define for most constants.
171 - improve compatibility to older C++ compilers.
172 - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now
173 default arguments when compiled as C++.
174 - enable automake dependency tracking.
175 - ev_loop_new no longer leaks memory when loop creation failed.
176 - new ev_cleanup watcher type.
177
178 3.9 Thu Dec 31 07:59:59 CET 2009
179 - signalfd is no longer used by default and has to be requested
180 explicitly - this means that easy to catch bugs become hard to
181 catch race conditions, but the users have spoken.
182 - point out the unspecified signal mask in the documentation, and
183 that this is a race condition regardless of EV_SIGNALFD.
184 - backport inotify code to C89.
185 - inotify file descriptors could leak into child processes.
186 - ev_stat watchers could keep an erroneous extra ref on the loop,
187 preventing exit when unregistering all watchers (testcases
188 provided by ry@tinyclouds.org).
189 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
190 symbols to make it easier for apps to do their own fd management.
191 - support EV_IDLE_ENABLE being disabled in ev++.h
192 (patch by Didier Spezia).
193 - take advantage of inotify_init1, if available, to set cloexec/nonblock
194 on fd creation, to avoid races.
195 - the signal handling pipe wasn't always initialised under windows
196 (analysed by lekma).
197 - changed minimum glibc requirement from glibc 2.9 to 2.7, for
198 signalfd.
199 - add missing string.h include (Denis F. Latypoff).
200 - only replace ev_stat.prev when we detect an actual difference,
201 so prev is (almost) always different to attr. this might
202 have caused the problems with 04_stat.t.
203 - add ev::timer->remaining () method to C++ API.
204
205 3.8 Sun Aug 9 14:30:45 CEST 2009
206 - incompatible change: do not necessarily reset signal handler
207 to SIG_DFL when a sighandler is stopped.
208 - ev_default_destroy did not properly free or zero some members,
209 potentially causing crashes and memory corruption on repeated
210 ev_default_destroy/ev_default_loop calls.
211 - take advantage of signalfd on GNU/Linux systems.
212 - document that the signal mask might be in an unspecified
213 state when using libev's signal handling.
214 - take advantage of some GNU/Linux calls to set cloexec/nonblock
215 on fd creation, to avoid race conditions.
216
217 3.7 Fri Jul 17 16:36:32 CEST 2009
218 - ev_unloop and ev_loop wrongly used a global variable to exit loops,
219 instead of using a per-loop variable (bug caught by accident...).
220 - the ev_set_io_collect_interval interpretation has changed.
221 - add new functionality: ev_set_userdata, ev_userdata,
222 ev_set_invoke_pending_cb, ev_set_loop_release_cb,
223 ev_invoke_pending, ev_pending_count, together with a long example
224 about thread locking.
225 - add ev_timer_remaining (as requested by Denis F. Latypoff).
226 - add ev_loop_depth.
227 - calling ev_unloop in fork/prepare watchers will no longer poll
228 for new events.
229 - Denis F. Latypoff corrected many typos in example code snippets.
230 - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double-
231 check that the syscall number is available before trying to
232 use it (reported by ry@tinyclouds).
233 - use GetSystemTimeAsFileTime instead of _timeb on windows, for
234 slightly higher accuracy.
235 - properly declare ev_loop_verify and ev_now_update even when
236 !EV_MULTIPLICITY.
237 - do not compile in any priority code when EV_MAXPRI == EV_MINPRI.
238 - support EV_MINIMAL==2 for a reduced API.
239 - actually 0-initialise struct sigaction when installing signals.
240 - add section on hibernate and stopped processes to ev_timer docs.
241
242 3.6 Tue Apr 28 02:49:30 CEST 2009
243 - multiple timers becoming ready within an event loop iteration
244 will be invoked in the "correct" order now.
245 - do not leave the event loop early just because we have no active
246 watchers, fixing a problem when embedding a kqueue loop
247 that has active kernel events but no registered watchers
248 (reported by blacksand blacksand).
249 - correctly zero the idx values for arrays, so destroying and
250 reinitialising the default loop actually works (patch by
251 Malek Hadj-Ali).
252 - implement ev_suspend and ev_resume.
253 - new EV_CUSTOM revents flag for use by applications.
254 - add documentation section about priorities.
255 - add a glossary to the documentation.
256 - extend the ev_fork description slightly.
257 - optimize a jump out of call_pending.
258
259 3.53 Sun Feb 15 02:38:20 CET 2009
260 - fix a bug in event pipe creation on win32 that would cause a
261 failed assertion on event loop creation (patch by Malek Hadj-Ali).
262 - probe for CLOCK_REALTIME support at runtime as well and fall
263 back to gettimeofday if there is an error, to support older
264 operating systems with newer header files/libraries.
265 - prefer gettimeofday over clock_gettime with USE_CLOCK_SYSCALL
266 (default most everywhere), otherwise not.
267
268 3.52 Wed Jan 7 21:43:02 CET 2009
269 - fix compilation of select backend in fd_set mode when NFDBITS is
270 missing (to get it to compile on QNX, reported by Rodrigo Campos).
271 - better select-nfds handling when select backend is in fd_set mode.
272 - diagnose fd_set overruns when select backend is in fd_set mode.
273 - due to a thinko, instead of disabling everything but
274 select on the borked OS X platform, everything but select was
275 allowed (reported by Emanuele Giaquinta).
276 - actually verify that local and remote port are matching in
277 libev's socketpair emulation, which makes denial-of-service
278 attacks harder (but not impossible - it's windows). Make sure
279 it even works under vista, which thinks that getpeer/sockname
280 should return fantasy port numbers.
281 - include "libev" in all assertion messages for potentially
282 clearer diagnostics.
283 - event_get_version (libevent compatibility) returned
284 a useless string instead of the expected version string
285 (patch by W.C.A. Wijngaards).
286
287 3.51 Wed Dec 24 23:00:11 CET 2008
288 - fix a bug where an inotify watcher was added twice, causing
289 freezes on hash collisions (reported and analysed by Graham Leggett).
290 - new config symbol, EV_USE_CLOCK_SYSCALL, to make libev use
291 a direct syscall - slower, but no dependency on librt et al.
292 - assume negative return values != -1 signals success of port_getn
293 (http://cvs.epicsol.org/cgi/viewcvs.cgi/epic5/source/newio.c?rev=1.52)
294 (no known failure reports, but it doesn't hurt).
295 - fork detection in ev_embed now stops and restarts the watcher
296 automatically.
297 - EXPERIMENTAL: default the method to operator () in ev++.h,
298 to make it nicer to use functors (requested by Benedek László).
299 - fixed const object callbacks in ev++.h.
300 - replaced loop_ref argument of watcher.set (loop) by a direct
301 ev_loop * in ev++.h, to avoid clashes with functor patch.
302 - do not try to watch the empty string via inotify.
303 - inotify watchers could be leaked under certain circumstances.
304 - OS X 10.5 is actually even more broken than earlier versions,
305 so fall back to select on that piece of garbage.
306 - fixed some weirdness in the ev_embed documentation.
307
308 3.49 Wed Nov 19 11:26:53 CET 2008
309 - ev_stat watchers will now use inotify as a mere hint on
310 kernels <2.6.25, or if the filesystem is not in the
311 "known to be good" list.
312 - better mingw32 compatibility (it's not as borked as native win32)
313 (analysed by Roger Pack).
314 - include stdio.h in the example program, as too many people are
315 confused by the weird C language otherwise. I guess the next thing
316 I get told is that the "..." ellipses in the examples don't compile
317 with their C compiler.
318
319 3.48 Thu Oct 30 09:02:37 CET 2008
320 - further optimise away the EPOLL_CTL_ADD/MOD combo in the epoll
321 backend by assuming the kernel event mask hasn't changed if
322 ADD fails with EEXIST.
323 - work around spurious event notification bugs in epoll by using
324 a 32-bit generation counter. recreate kernel state if we receive
325 spurious notifications or unwanted events. this is very costly,
326 but I didn't come up with this horrible design.
327 - use memset to initialise most arrays now and do away with the
328 init functions.
329 - expand time-out strategies into a "Be smart about timeouts" section.
330 - drop the "struct" from all ev_watcher declarations in the
331 documentation and did other clarifications (yeah, it was a mistake
332 to have a struct AND a function called ev_loop).
333 - fix a bug where ev_default would not initialise the default
334 loop again after it was destroyed with ev_default_destroy.
335 - rename syserr to ev_syserr to avoid name clashes when embedding,
336 do similar changes for event.c.
337
338 3.45 Tue Oct 21 21:59:26 CEST 2008
339 - disable inotify usage on linux <2.6.25, as it is broken
340 (reported by Yoann Vandoorselaere).
341 - ev_stat erroneously would try to add inotify watchers
342 even when inotify wasn't available (this should only
343 have a performance impact).
344 - ev_once now passes both timeout and io to the callback if both
345 occur concurrently, instead of giving timeouts precedence.
346 - disable EV_USE_INOTIFY when sys/inotify.h is too old.
347
348 3.44 Mon Sep 29 05:18:39 CEST 2008
349 - embed watchers now automatically invoke ev_loop_fork on the
350 embedded loop when the parent loop forks.
351 - new function: ev_now_update (loop).
352 - verify_watcher was not marked static.
353 - improve the "associating..." manpage section.
354 - documentation tweaks here and there.
355
356 3.43 Sun Jul 6 05:34:41 CEST 2008
357 - include more include files on windows to get struct _stati64
358 (reported by Chris Hulbert, but doesn't quite fix his issue).
359 - add missing #include <io.h> in ev.c on windows (reported by
360 Matt Tolton).
361
362 3.42 Tue Jun 17 12:12:07 CEST 2008
363 - work around yet another windows bug: FD_SET actually adds fd's
364 multiple times to the fd_*SET*, despite official MSN docs claiming
365 otherwise. Reported and well-analysed by Matt Tolton.
366 - define NFDBITS to 0 when EV_SELECT_IS_WINSOCKET to make it compile
367 (reported any analysed by Chris Hulbert).
368 - fix a bug in ev_ebadf (this function is only used to catch
369 programming errors in the libev user). reported by Matt Tolton.
370 - fix a bug in fd_intern on win32 (could lead to compile errors
371 under some circumstances, but would work correctly if it compiles).
372 reported by Matt Tolton.
373 - (try to) work around missing lstat on windows.
374 - pass in the write fd set as except fd set under windows. windows
375 is so uncontrollably lame that it requires this. this means that
376 switching off oobinline is not supported (but tcp/ip doesn't
377 have oob, so that would be stupid anyways.
378 - use posix module symbol to auto-detect monotonic clock presence
379 and some other default values.
380
381 3.41 Fri May 23 18:42:54 CEST 2008
382 - work around an obscure bug in winsocket select: if you
383 provide only empty fd sets then select returns WSAEINVAL. how sucky.
384 - improve timer scheduling stability and reduce use of time_epsilon.
385 - use 1-based 2-heap for EV_MINIMAL, simplifies code, reduces
386 codesize and makes for better cache-efficiency.
387 - use 3-based 4-heap for !EV_MINIMAL. this makes better use
388 of cpu cache lines and gives better growth behaviour than
389 2-based heaps.
390 - cache timestamp within heap for !EV_MINIMAL, to avoid random
391 memory accesses.
392 - document/add EV_USE_4HEAP and EV_HEAP_CACHE_AT.
393 - fix a potential aliasing issue in ev_timer_again.
394 - add/document ev_periodic_at, retract direct access to ->at.
395 - improve ev_stat docs.
396 - add portability requirements section.
397 - fix manpage headers etc.
398 - normalise WSA error codes to lower range on windows.
399 - add consistency check code that can be called automatically
400 or on demand to check for internal structures (ev_loop_verify).
401
402 3.31 Wed Apr 16 20:45:04 CEST 2008
403 - added last minute fix for ev_poll.c by Brandon Black.
404
405 3.3 Wed Apr 16 19:04:10 CEST 2008
406 - event_base_loopexit should return 0 on success
407 (W.C.A. Wijngaards).
408 - added linux eventfd support.
409 - try to autodetect epoll and inotify support
410 by libc header version if not using autoconf.
411 - new symbols: EV_DEFAULT_UC and EV_DEFAULT_UC_.
412 - declare functions defined in ev.h as inline if
413 C99 or gcc are available.
414 - enable inlining with gcc versions 2 and 3.
415 - work around broken poll implementations potentially
416 not clearing revents field in ev_poll (Brandon Black)
417 (no such systems are known at this time).
418 - work around a bug in realloc on openbsd and darwin,
419 also makes the erroneous valgrind complaints
420 go away (noted by various people).
421 - fix ev_async_pending, add c++ wrapper for ev_async
422 (based on patch sent by Johannes Deisenhofer).
423 - add sensible set method to ev::embed.
424 - made integer constants type int in ev.h.
425
426 3.2 Wed Apr 2 17:11:19 CEST 2008
427 - fix a 64 bit overflow issue in the select backend,
428 by using fd_mask instead of int for the mask.
429 - rename internal sighandler to avoid clash with very old perls.
430 - entering ev_loop will not clear the ONESHOT or NONBLOCKING
431 flags of any outer loops anymore.
432 - add ev_async_pending.
433
434 3.1 Thu Mar 13 13:45:22 CET 2008
435 - implement ev_async watchers.
436 - only initialise signal pipe on demand.
437 - make use of sig_atomic_t configurable.
438 - improved documentation.
439
440 3.0 Mon Jan 28 13:14:47 CET 2008
441 - API/ABI bump to version 3.0.
442 - ev++.h includes "ev.h" by default now, not <ev.h>.
443 - slightly improved documentation.
444 - speed up signal detection after a fork.
445 - only optionally return trace status changed in ev_child
446 watchers.
447 - experimental (and undocumented) loop wrappers for ev++.h.
448
449 2.01 Tue Dec 25 08:04:41 CET 2007
450 - separate Changes file.
451 - fix ev_path_set => ev_stat_set typo.
452 - remove event_compat.h from the libev tarball.
453 - change how include files are found.
454 - doc updates.
455 - update licenses, explicitly allow for GPL relicensing.
456
457 2.0 Sat Dec 22 17:47:03 CET 2007
458 - new ev_sleep, ev_set_(io|timeout)_collect_interval.
459 - removed epoll from embeddable fd set.
460 - fix embed watchers.
461 - renamed ev_embed.loop to other.
462 - added exported Symbol tables.
463 - undefine member wrapper macros at the end of ev.c.
464 - respect EV_H in ev++.h.
465
466 1.86 Tue Dec 18 02:36:57 CET 2007
467 - fix memleak on loop destroy (not relevant for perl).
468
469 1.85 Fri Dec 14 20:32:40 CET 2007
470 - fix some aliasing issues w.r.t. timers and periodics
471 (not relevant for perl).
472
473 (for historic versions refer to EV/Changes, found in the Perl interface)
474
475 0.1 Wed Oct 31 21:31:48 CET 2007
476 - original version; hacked together in <24h.
477