ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/Changes
Revision: 1.345
Committed: Fri Nov 22 14:32:13 2019 UTC (4 years, 5 months ago) by root
Branch: MAIN
Changes since 1.344: +5 -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: revisit 59.x timer in the light of modern powersaving
4 TODO: maybe use timerfd to detect time jumps on linux
5 TODO: document EV_TSTAMP_T
6
7 4.29
8 - (libev) add io uring autoconf and non-autoconf detection,
9 the latter of which should disable io_uring compilation
10 on old systems.
11
12 4.28 (EV only)
13 - linuxaio backend resulted in random memory corruption
14 when loop is forked.
15 - linuxaio backend might have tried to cancel an iocb
16 multiple times (was unable to trigger this).
17 - linuxaio backend now employs a generation counter to
18 avoid handling spurious events from cancelled requests.
19 - io_cancel can return EINTR, deal with it. also, assume
20 io_submit also returns EINTR.
21 - fix some other minor bugs in linuxaio backend.
22 - ev_tstamp type can now be overriden by defining EV_TSTAMP_T.
23 - cleanup: replace expect_true/false and noinline by their
24 libecb counterparts.
25 - move syscall infrastructure from ev_linuxaio.c to ev.c.
26 - prepare io_uring integration.
27 - tweak ev_floor.
28 - epoll, poll, win32 Sleep and other places that use millisecond
29 reslution now all try to round up times.
30 - solaris port backend didn't compile.
31 - abstract time constants into their macros, for more flexibility.
32
33 4.27 Thu Jun 27 22:43:44 CEST 2019
34 - linux aio backend almost completely rewritten to work around its
35 limitations.
36 - linux aio backend now requires linux 4.19+.
37 - epoll backend now mandatory for linux aio backend.
38 - fail assertions more aggressively on invalid fd's detected
39 in the event loop, do not just silently fd_kill in case of
40 user error.
41 - ev_io_start/ev_io_stop now verify the watcher fd using
42 a syscall when EV_VERIFY is 2 or higher.
43
44 4.26 (EV only)
45 - update to libecb 0x00010006.
46 - new experimental linux aio backend (linux 4.18+).
47 - removed redundant 0-ptr check in ev_once.
48 - updated/extended ev_set_allocator documentation.
49 - replaced EMPTY2 macro by array_needsize_noinit.
50 - minor code cleanups.
51 - epoll backend now uses epoll_create1 also after fork.
52
53 4.25 Fri Dec 21 07:49:20 CET 2018
54 - INCOMPATIBLE CHANGE: EV_THROW was renamed to EV_NOEXCEPT
55 (EV_THROW still provided) and now uses noexcept on C++11 or newer.
56 - move the darwin select workaround higher in ev.c, as newer versions of
57 darwin managed to break their broken select even more.
58 - ANDROID => __ANDROID__ (reported by enh@google.com).
59 - disable epoll_create1 on android because it has broken header files
60 and google is unwilling to fix them (reported by enh@google.com).
61 - avoid a minor compilation warning on win32.
62 - c++: remove deprecated dynamic throw() specifications.
63 - c++: improve the (unsupported) bad_loop exception class.
64 - backport perl ev_periodic example to C, untested.
65 - update libecb, biggets change is to include a memory fence
66 in ECB_MEMORY_FENCE_RELEASE on x86/amd64.
67 - minor autoconf/automake modernisation.
68
69 4.24 Wed Dec 28 05:19:55 CET 2016
70 - bump version to 4.24, as the release tarball inexplicably
71 didn't have the right version in ev.h, even though the cvs-tagged
72 version did have the right one (reported by Ales Teska).
73
74 4.23 Wed Nov 16 18:23:41 CET 2016
75 - move some declarations at the beginning to help certain retarded
76 microsoft compilers, even though their documentation claims
77 otherwise (reported by Ruslan Osmanov).
78
79 4.22 Sun Dec 20 22:11:50 CET 2015
80 - when epoll detects unremovable fds in the fd set, rebuild
81 only the epoll descriptor, not the signal pipe, to avoid
82 SIGPIPE in ev_async_send. This doesn't solve it on fork,
83 so document what needs to be done in ev_loop_fork
84 (analyzed by Benjamin Mahler).
85 - remove superfluous sys/timeb.h include on win32
86 (analyzed by Jason Madden).
87 - updated libecb.
88
89 4.20 Sat Jun 20 13:01:43 CEST 2015
90 - prefer noexcept over throw () with C++ 11.
91 - update ecb.h due to incompatibilities with c11.
92 - fix a potential aliasing issue when reading and writing
93 watcher callbacks.
94
95 4.19 Thu Sep 25 08:18:25 CEST 2014
96 - ev.h wasn't valid C++ anymore, which tripped compilers other than
97 clang, msvc or gcc (analyzed by Raphael 'kena' Poss). Unfortunately,
98 C++ doesn't support typedefs for function pointers fully, so the affected
99 declarations have to spell out the types each time.
100 - when not using autoconf, tighten the check for clock_gettime and related
101 functionality.
102
103 4.18 Fri Sep 5 17:55:26 CEST 2014
104 - events on files were not always generated properly with the
105 epoll backend (testcase by Assaf Inbal).
106 - mark event pipe fd as cloexec after a fork (analyzed by Sami Farin).
107 - (ecb) support m68k, m88k and sh (patch by Miod Vallat).
108 - use a reasonable fallback for EV_NSIG instead of erroring out
109 when we can't detect the signal set size.
110 - in the absence of autoconf, do not use the clock syscall
111 on glibc >= 2.17 (avoids the syscall AND -lrt on systems
112 doing clock_gettime in userspace).
113 - ensure extern "C" function pointers are used for externally-visible
114 loop callbacks (not watcher callbacks yet).
115 - (ecb) work around memory barriers and volatile apparently both being
116 broken in visual studio 2008 and later (analysed and patch by Nicolas Noble).
117
118 4.15 Fri Mar 1 12:04:50 CET 2013
119 - destroying a non-default loop would stop the global waitpid
120 watcher (Denis Bilenko).
121 - queueing pending watchers of higher priority from a watcher now invokes
122 them in a timely fashion (reported by Denis Bilenko).
123 - add throw() to all libev functions that cannot throw exceptions, for
124 further code size decrease when compiling for C++.
125 - add throw () to callbacks that must not throw exceptions (allocator,
126 syserr, loop acquire/release, periodic reschedule cbs).
127 - fix event_base_loop return code, add event_get_callback, event_base_new,
128 event_base_get_method calls to improve libevent 1.x emulation and add
129 some libevent 2.x functionality (based on a patch by Jeff Davey).
130 - add more memory fences to fix a bug reported by Jeff Davey. Better
131 be overfenced than underprotected.
132 - ev_run now returns a boolean status (true meaning watchers are
133 still active).
134 - ev_once: undef EV_ERROR in ev_kqueue.c, to avoid clashing with
135 libev's EV_ERROR (reported by 191919).
136 - (ecb) add memory fence support for xlC (Darin McBride).
137 - (ecb) add memory fence support for gcc-mips (Anton Kirilov).
138 - (ecb) add memory fence support for gcc-alpha (Christian Weisgerber).
139 - work around some kernels losing file descriptors by leaking
140 the kqueue descriptor in the child.
141 - work around linux inotify not reporting IN_ATTRIB changes for directories
142 in many cases.
143 - include sys/syscall.h instead of plain syscall.h.
144 - check for io watcher loops in ev_verify, check for the most
145 common reported usage bug in ev_io_start.
146 - choose socket vs. WSASocket at compiletime using EV_USE_WSASOCKET.
147 - always use WSASend/WSARecv directly on windows, hoping that this
148 works in all cases (unlike read/write/send/recv...).
149 - try to detect signals around a fork faster (test program by
150 Denis Bilenko).
151 - work around recent glibc versions that leak memory in realloc.
152 - rename ev::embed::set to ev::embed::set_embed to avoid clashing
153 the watcher base set (loop) method.
154 - rewrite the async/signal pipe logic to always keep a valid fd, which
155 simplifies (and hopefully correctifies :) the race checking
156 on fork, at the cost of one extra fd.
157 - add fat, msdos, jffs2, ramfs, ntfs and btrfs to the list of
158 inotify-supporting filesystems.
159 - move orig_CFLAGS assignment to after AC_INIT, as newer autoconf
160 versions ignore it before
161 (https://bugzilla.redhat.com/show_bug.cgi?id=908096).
162 - add some untested android support.
163 - enum expressions must be of type int (reported by Juan Pablo L).
164
165 4.11 Sat Feb 4 19:52:39 CET 2012
166 - INCOMPATIBLE CHANGE: ev_timer_again now clears the pending status, as
167 was documented already, but not implemented in the repeating case.
168 - new compiletime symbols: EV_NO_SMP and EV_NO_THREADS.
169 - fix a race where the workaround against the epoll fork bugs
170 caused signals to not be handled anymore.
171 - correct backend_fudge for most backends, and implement a windows
172 specific workaround to avoid looping because we call both
173 select and Sleep, both with different time resolutions.
174 - document range and guarantees of ev_sleep.
175 - document reasonable ranges for periodics interval and offset.
176 - rename backend_fudge to backend_mintime to avoid future confusion :)
177 - change the default periodic reschedule function to hopefully be more
178 exact and correct even in corner cases or in the far future.
179 - do not rely on -lm anymore: use it when available but use our
180 own floor () if it is missing. This should make it easier to embed,
181 as no external libraries are required.
182 - strategically import macros from libecb and mark rarely-used functions
183 as cache-cold (saving almost 2k code size on typical amd64 setups).
184 - add Symbols.ev and Symbols.event files, that were missing.
185 - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now).
186 - fix #3 "be smart about timeouts" to not "deadlock" when
187 timeout == now, also improve the section overall.
188 - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom.
189 - support new EV_API_STATIC mode to make all libev symbols
190 static.
191 - supply default CFLAGS of -g -O3 with gcc when original CFLAGS
192 were empty.
193
194 4.04 Wed Feb 16 09:01:51 CET 2011
195 - fix two problems in the native win32 backend, where reuse of fd's
196 with different underlying handles caused handles not to be removed
197 or added to the select set (analyzed and tested by Bert Belder).
198 - do no rely on ceil() in ev_e?poll.c.
199 - backport libev to HP-UX versions before 11 v3.
200 - configure did not detect nanosleep and clock_gettime properly when
201 they are available in the libc (as opposed to -lrt).
202
203 4.03 Tue Jan 11 14:37:25 CET 2011
204 - officially support polling files with all backends.
205 - support files, /dev/zero etc. the same way as select in the epoll
206 backend, by generating events on our own.
207 - ports backend: work around solaris bug 6874410 and many related ones
208 (EINTR, maybe more), with no performance loss (note that the solaris
209 bug report is actually wrong, reality is far more bizarre and broken
210 than that).
211 - define EV_READ/EV_WRITE as macros in event.h, as some programs use
212 #ifdef to test for them.
213 - new (experimental) function: ev_feed_signal.
214 - new (to become default) EVFLAG_NOSIGMASK flag.
215 - new EVBACKEND_MASK symbol.
216 - updated COMMON IDIOMS SECTION.
217
218 4.01 Fri Nov 5 21:51:29 CET 2010
219 - automake fucked it up, apparently, --add-missing -f is not quite enough
220 to make it update its files, so 4.00 didn't install ev++.h and
221 event.h on make install. grrr.
222 - ev_loop(count|depth) didn't return anything (Robin Haberkorn).
223 - change EV_UNDEF to 0xffffffff to silence some overzealous compilers.
224 - use "(libev) " prefix for all libev error messages now.
225
226 4.00 Mon Oct 25 12:32:12 CEST 2010
227 - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading.
228 - ev_embed_stop did not correctly stop the watcher (very good
229 testcase by Vladimir Timofeev).
230 - ev_run will now always update the current loop time - it erroneously
231 didn't when idle watchers were active, causing timers not to fire.
232 - fix a bug where a timeout of zero caused the timer not to fire
233 in the libevent emulation (testcase by Péter Szabó).
234 - applied win32 fixes by Michael Lenaghan (also James Mansion).
235 - replace EV_MINIMAL by EV_FEATURES.
236 - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it
237 seems the former is *much* faster than the latter.
238 - linux kernel version detection (for inotify bug workarounds)
239 did not work properly.
240 - reduce the number of spurious wake-ups with the ports backend.
241 - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu).
242 - do async init within ev_async_start, not ev_async_set, which avoids
243 an API quirk where the set function must be called in the C++ API
244 even when there is nothing to set.
245 - add (undocumented) EV_ENABLE when adding events with kqueue,
246 this might help with OS X, which seems to need it despite documenting
247 not to need it (helpfully pointed out by Tilghman Lesher).
248 - do not use poll by default on freebsd, it's broken (what isn't
249 on freebsd...).
250 - allow to embed epoll on kernels >= 2.6.32.
251 - configure now prepends -O3, not appends it, so one can still
252 override it.
253 - ev.pod: greatly expanded the portability section, added a porting
254 section, a description of watcher states and made lots of minor fixes.
255 - disable poll backend on AIX, the poll header spams the namespace
256 and it's not worth working around dead platforms (reported
257 and analyzed by Aivars Kalvans).
258 - improve header file compatibility of the standalone eventfd code
259 in an obscure case.
260 - implement EV_AVOID_STDIO option.
261 - do not use sscanf to parse linux version number (smaller, faster,
262 no sscanf dependency).
263 - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings.
264 - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs.
265 - add section on accept() problems to the manpage.
266 - rename EV_TIMEOUT to EV_TIMER.
267 - rename ev_loop_count/depth/verify/loop/unloop.
268 - remove ev_default_destroy and ev_default_fork.
269 - switch to two-digit minor version.
270 - work around an apparent gentoo compiler bug.
271 - define _DARWIN_UNLIMITED_SELECT. just so.
272 - use enum instead of #define for most constants.
273 - improve compatibility to older C++ compilers.
274 - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now
275 default arguments when compiled as C++.
276 - enable automake dependency tracking.
277 - ev_loop_new no longer leaks memory when loop creation failed.
278 - new ev_cleanup watcher type.
279
280 3.9 Thu Dec 31 07:59:59 CET 2009
281 - signalfd is no longer used by default and has to be requested
282 explicitly - this means that easy to catch bugs become hard to
283 catch race conditions, but the users have spoken.
284 - point out the unspecified signal mask in the documentation, and
285 that this is a race condition regardless of EV_SIGNALFD.
286 - backport inotify code to C89.
287 - inotify file descriptors could leak into child processes.
288 - ev_stat watchers could keep an erroneous extra ref on the loop,
289 preventing exit when unregistering all watchers (testcases
290 provided by ry@tinyclouds.org).
291 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
292 symbols to make it easier for apps to do their own fd management.
293 - support EV_IDLE_ENABLE being disabled in ev++.h
294 (patch by Didier Spezia).
295 - take advantage of inotify_init1, if available, to set cloexec/nonblock
296 on fd creation, to avoid races.
297 - the signal handling pipe wasn't always initialised under windows
298 (analysed by lekma).
299 - changed minimum glibc requirement from glibc 2.9 to 2.7, for
300 signalfd.
301 - add missing string.h include (Denis F. Latypoff).
302 - only replace ev_stat.prev when we detect an actual difference,
303 so prev is (almost) always different to attr. this might
304 have caused the problems with 04_stat.t.
305 - add ev::timer->remaining () method to C++ API.
306
307 3.8 Sun Aug 9 14:30:45 CEST 2009
308 - incompatible change: do not necessarily reset signal handler
309 to SIG_DFL when a sighandler is stopped.
310 - ev_default_destroy did not properly free or zero some members,
311 potentially causing crashes and memory corruption on repeated
312 ev_default_destroy/ev_default_loop calls.
313 - take advantage of signalfd on GNU/Linux systems.
314 - document that the signal mask might be in an unspecified
315 state when using libev's signal handling.
316 - take advantage of some GNU/Linux calls to set cloexec/nonblock
317 on fd creation, to avoid race conditions.
318
319 3.7 Fri Jul 17 16:36:32 CEST 2009
320 - ev_unloop and ev_loop wrongly used a global variable to exit loops,
321 instead of using a per-loop variable (bug caught by accident...).
322 - the ev_set_io_collect_interval interpretation has changed.
323 - add new functionality: ev_set_userdata, ev_userdata,
324 ev_set_invoke_pending_cb, ev_set_loop_release_cb,
325 ev_invoke_pending, ev_pending_count, together with a long example
326 about thread locking.
327 - add ev_timer_remaining (as requested by Denis F. Latypoff).
328 - add ev_loop_depth.
329 - calling ev_unloop in fork/prepare watchers will no longer poll
330 for new events.
331 - Denis F. Latypoff corrected many typos in example code snippets.
332 - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double-
333 check that the syscall number is available before trying to
334 use it (reported by ry@tinyclouds).
335 - use GetSystemTimeAsFileTime instead of _timeb on windows, for
336 slightly higher accuracy.
337 - properly declare ev_loop_verify and ev_now_update even when
338 !EV_MULTIPLICITY.
339 - do not compile in any priority code when EV_MAXPRI == EV_MINPRI.
340 - support EV_MINIMAL==2 for a reduced API.
341 - actually 0-initialise struct sigaction when installing signals.
342 - add section on hibernate and stopped processes to ev_timer docs.
343
344 3.6 Tue Apr 28 02:49:30 CEST 2009
345 - multiple timers becoming ready within an event loop iteration
346 will be invoked in the "correct" order now.
347 - do not leave the event loop early just because we have no active
348 watchers, fixing a problem when embedding a kqueue loop
349 that has active kernel events but no registered watchers
350 (reported by blacksand blacksand).
351 - correctly zero the idx values for arrays, so destroying and
352 reinitialising the default loop actually works (patch by
353 Malek Hadj-Ali).
354 - implement ev_suspend and ev_resume.
355 - new EV_CUSTOM revents flag for use by applications.
356 - add documentation section about priorities.
357 - add a glossary to the documentation.
358 - extend the ev_fork description slightly.
359 - optimize a jump out of call_pending.
360
361 3.53 Sun Feb 15 02:38:20 CET 2009
362 - fix a bug in event pipe creation on win32 that would cause a
363 failed assertion on event loop creation (patch by Malek Hadj-Ali).
364 - probe for CLOCK_REALTIME support at runtime as well and fall
365 back to gettimeofday if there is an error, to support older
366 operating systems with newer header files/libraries.
367 - prefer gettimeofday over clock_gettime with USE_CLOCK_SYSCALL
368 (default most everywhere), otherwise not.
369
370 3.52 Wed Jan 7 21:43:02 CET 2009
371 - fix compilation of select backend in fd_set mode when NFDBITS is
372 missing (to get it to compile on QNX, reported by Rodrigo Campos).
373 - better select-nfds handling when select backend is in fd_set mode.
374 - diagnose fd_set overruns when select backend is in fd_set mode.
375 - due to a thinko, instead of disabling everything but
376 select on the borked OS X platform, everything but select was
377 allowed (reported by Emanuele Giaquinta).
378 - actually verify that local and remote port are matching in
379 libev's socketpair emulation, which makes denial-of-service
380 attacks harder (but not impossible - it's windows). Make sure
381 it even works under vista, which thinks that getpeer/sockname
382 should return fantasy port numbers.
383 - include "libev" in all assertion messages for potentially
384 clearer diagnostics.
385 - event_get_version (libevent compatibility) returned
386 a useless string instead of the expected version string
387 (patch by W.C.A. Wijngaards).
388
389 3.51 Wed Dec 24 23:00:11 CET 2008
390 - fix a bug where an inotify watcher was added twice, causing
391 freezes on hash collisions (reported and analysed by Graham Leggett).
392 - new config symbol, EV_USE_CLOCK_SYSCALL, to make libev use
393 a direct syscall - slower, but no dependency on librt et al.
394 - assume negative return values != -1 signals success of port_getn
395 (http://cvs.epicsol.org/cgi/viewcvs.cgi/epic5/source/newio.c?rev=1.52)
396 (no known failure reports, but it doesn't hurt).
397 - fork detection in ev_embed now stops and restarts the watcher
398 automatically.
399 - EXPERIMENTAL: default the method to operator () in ev++.h,
400 to make it nicer to use functors (requested by Benedek László).
401 - fixed const object callbacks in ev++.h.
402 - replaced loop_ref argument of watcher.set (loop) by a direct
403 ev_loop * in ev++.h, to avoid clashes with functor patch.
404 - do not try to watch the empty string via inotify.
405 - inotify watchers could be leaked under certain circumstances.
406 - OS X 10.5 is actually even more broken than earlier versions,
407 so fall back to select on that piece of garbage.
408 - fixed some weirdness in the ev_embed documentation.
409
410 3.49 Wed Nov 19 11:26:53 CET 2008
411 - ev_stat watchers will now use inotify as a mere hint on
412 kernels <2.6.25, or if the filesystem is not in the
413 "known to be good" list.
414 - better mingw32 compatibility (it's not as borked as native win32)
415 (analysed by Roger Pack).
416 - include stdio.h in the example program, as too many people are
417 confused by the weird C language otherwise. I guess the next thing
418 I get told is that the "..." ellipses in the examples don't compile
419 with their C compiler.
420
421 3.48 Thu Oct 30 09:02:37 CET 2008
422 - further optimise away the EPOLL_CTL_ADD/MOD combo in the epoll
423 backend by assuming the kernel event mask hasn't changed if
424 ADD fails with EEXIST.
425 - work around spurious event notification bugs in epoll by using
426 a 32-bit generation counter. recreate kernel state if we receive
427 spurious notifications or unwanted events. this is very costly,
428 but I didn't come up with this horrible design.
429 - use memset to initialise most arrays now and do away with the
430 init functions.
431 - expand time-out strategies into a "Be smart about timeouts" section.
432 - drop the "struct" from all ev_watcher declarations in the
433 documentation and did other clarifications (yeah, it was a mistake
434 to have a struct AND a function called ev_loop).
435 - fix a bug where ev_default would not initialise the default
436 loop again after it was destroyed with ev_default_destroy.
437 - rename syserr to ev_syserr to avoid name clashes when embedding,
438 do similar changes for event.c.
439
440 3.45 Tue Oct 21 21:59:26 CEST 2008
441 - disable inotify usage on linux <2.6.25, as it is broken
442 (reported by Yoann Vandoorselaere).
443 - ev_stat erroneously would try to add inotify watchers
444 even when inotify wasn't available (this should only
445 have a performance impact).
446 - ev_once now passes both timeout and io to the callback if both
447 occur concurrently, instead of giving timeouts precedence.
448 - disable EV_USE_INOTIFY when sys/inotify.h is too old.
449
450 3.44 Mon Sep 29 05:18:39 CEST 2008
451 - embed watchers now automatically invoke ev_loop_fork on the
452 embedded loop when the parent loop forks.
453 - new function: ev_now_update (loop).
454 - verify_watcher was not marked static.
455 - improve the "associating..." manpage section.
456 - documentation tweaks here and there.
457
458 3.43 Sun Jul 6 05:34:41 CEST 2008
459 - include more include files on windows to get struct _stati64
460 (reported by Chris Hulbert, but doesn't quite fix his issue).
461 - add missing #include <io.h> in ev.c on windows (reported by
462 Matt Tolton).
463
464 3.42 Tue Jun 17 12:12:07 CEST 2008
465 - work around yet another windows bug: FD_SET actually adds fd's
466 multiple times to the fd_*SET*, despite official MSN docs claiming
467 otherwise. Reported and well-analysed by Matt Tolton.
468 - define NFDBITS to 0 when EV_SELECT_IS_WINSOCKET to make it compile
469 (reported any analysed by Chris Hulbert).
470 - fix a bug in ev_ebadf (this function is only used to catch
471 programming errors in the libev user). reported by Matt Tolton.
472 - fix a bug in fd_intern on win32 (could lead to compile errors
473 under some circumstances, but would work correctly if it compiles).
474 reported by Matt Tolton.
475 - (try to) work around missing lstat on windows.
476 - pass in the write fd set as except fd set under windows. windows
477 is so uncontrollably lame that it requires this. this means that
478 switching off oobinline is not supported (but tcp/ip doesn't
479 have oob, so that would be stupid anyways.
480 - use posix module symbol to auto-detect monotonic clock presence
481 and some other default values.
482
483 3.41 Fri May 23 18:42:54 CEST 2008
484 - work around an obscure bug in winsocket select: if you
485 provide only empty fd sets then select returns WSAEINVAL. how sucky.
486 - improve timer scheduling stability and reduce use of time_epsilon.
487 - use 1-based 2-heap for EV_MINIMAL, simplifies code, reduces
488 codesize and makes for better cache-efficiency.
489 - use 3-based 4-heap for !EV_MINIMAL. this makes better use
490 of cpu cache lines and gives better growth behaviour than
491 2-based heaps.
492 - cache timestamp within heap for !EV_MINIMAL, to avoid random
493 memory accesses.
494 - document/add EV_USE_4HEAP and EV_HEAP_CACHE_AT.
495 - fix a potential aliasing issue in ev_timer_again.
496 - add/document ev_periodic_at, retract direct access to ->at.
497 - improve ev_stat docs.
498 - add portability requirements section.
499 - fix manpage headers etc.
500 - normalise WSA error codes to lower range on windows.
501 - add consistency check code that can be called automatically
502 or on demand to check for internal structures (ev_loop_verify).
503
504 3.31 Wed Apr 16 20:45:04 CEST 2008
505 - added last minute fix for ev_poll.c by Brandon Black.
506
507 3.3 Wed Apr 16 19:04:10 CEST 2008
508 - event_base_loopexit should return 0 on success
509 (W.C.A. Wijngaards).
510 - added linux eventfd support.
511 - try to autodetect epoll and inotify support
512 by libc header version if not using autoconf.
513 - new symbols: EV_DEFAULT_UC and EV_DEFAULT_UC_.
514 - declare functions defined in ev.h as inline if
515 C99 or gcc are available.
516 - enable inlining with gcc versions 2 and 3.
517 - work around broken poll implementations potentially
518 not clearing revents field in ev_poll (Brandon Black)
519 (no such systems are known at this time).
520 - work around a bug in realloc on openbsd and darwin,
521 also makes the erroneous valgrind complaints
522 go away (noted by various people).
523 - fix ev_async_pending, add c++ wrapper for ev_async
524 (based on patch sent by Johannes Deisenhofer).
525 - add sensible set method to ev::embed.
526 - made integer constants type int in ev.h.
527
528 3.2 Wed Apr 2 17:11:19 CEST 2008
529 - fix a 64 bit overflow issue in the select backend,
530 by using fd_mask instead of int for the mask.
531 - rename internal sighandler to avoid clash with very old perls.
532 - entering ev_loop will not clear the ONESHOT or NONBLOCKING
533 flags of any outer loops anymore.
534 - add ev_async_pending.
535
536 3.1 Thu Mar 13 13:45:22 CET 2008
537 - implement ev_async watchers.
538 - only initialise signal pipe on demand.
539 - make use of sig_atomic_t configurable.
540 - improved documentation.
541
542 3.0 Mon Jan 28 13:14:47 CET 2008
543 - API/ABI bump to version 3.0.
544 - ev++.h includes "ev.h" by default now, not <ev.h>.
545 - slightly improved documentation.
546 - speed up signal detection after a fork.
547 - only optionally return trace status changed in ev_child
548 watchers.
549 - experimental (and undocumented) loop wrappers for ev++.h.
550
551 2.01 Tue Dec 25 08:04:41 CET 2007
552 - separate Changes file.
553 - fix ev_path_set => ev_stat_set typo.
554 - remove event_compat.h from the libev tarball.
555 - change how include files are found.
556 - doc updates.
557 - update licenses, explicitly allow for GPL relicensing.
558
559 2.0 Sat Dec 22 17:47:03 CET 2007
560 - new ev_sleep, ev_set_(io|timeout)_collect_interval.
561 - removed epoll from embeddable fd set.
562 - fix embed watchers.
563 - renamed ev_embed.loop to other.
564 - added exported Symbol tables.
565 - undefine member wrapper macros at the end of ev.c.
566 - respect EV_H in ev++.h.
567
568 1.86 Tue Dec 18 02:36:57 CET 2007
569 - fix memleak on loop destroy (not relevant for perl).
570
571 1.85 Fri Dec 14 20:32:40 CET 2007
572 - fix some aliasing issues w.r.t. timers and periodics
573 (not relevant for perl).
574
575 (for historic versions refer to EV/Changes, found in the Perl interface)
576
577 0.1 Wed Oct 31 21:31:48 CET 2007
578 - original version; hacked together in <24h.
579