ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/Changes
Revision: 1.344
Committed: Tue Nov 19 13:09:00 2019 UTC (4 years, 7 months ago) by root
Branch: MAIN
CVS Tags: EV-rel-4_28
Changes since 1.343: +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: 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 4.28 (EV only)
8 - linuxaio backend resulted in random memory corruption
9 when loop is forked.
10 - linuxaio backend might have tried to cancel an iocb
11 multiple times (was unable to trigger this).
12 - linuxaio backend now employs a generation counter to
13 avoid handling spurious events from cancelled requests.
14 - io_cancel can return EINTR, deal with it. also, assume
15 io_submit also returns EINTR.
16 - fix some other minor bugs in linuxaio backend.
17 - ev_tstamp type can now be overriden by defining EV_TSTAMP_T.
18 - cleanup: replace expect_true/false and noinline by their
19 libecb counterparts.
20 - move syscall infrastructure from ev_linuxaio.c to ev.c.
21 - prepare io_uring integration.
22 - tweak ev_floor.
23 - epoll, poll, win32 Sleep and other places that use millisecond
24 reslution now all try to round up times.
25 - solaris port backend didn't compile.
26 - abstract time constants into their macros, for more flexibility.
27
28 4.27 Thu Jun 27 22:43:44 CEST 2019
29 - linux aio backend almost completely rewritten to work around its
30 limitations.
31 - linux aio backend now requires linux 4.19+.
32 - epoll backend now mandatory for linux aio backend.
33 - fail assertions more aggressively on invalid fd's detected
34 in the event loop, do not just silently fd_kill in case of
35 user error.
36 - ev_io_start/ev_io_stop now verify the watcher fd using
37 a syscall when EV_VERIFY is 2 or higher.
38
39 4.26 (EV only)
40 - update to libecb 0x00010006.
41 - new experimental linux aio backend (linux 4.18+).
42 - removed redundant 0-ptr check in ev_once.
43 - updated/extended ev_set_allocator documentation.
44 - replaced EMPTY2 macro by array_needsize_noinit.
45 - minor code cleanups.
46 - epoll backend now uses epoll_create1 also after fork.
47
48 4.25 Fri Dec 21 07:49:20 CET 2018
49 - INCOMPATIBLE CHANGE: EV_THROW was renamed to EV_NOEXCEPT
50 (EV_THROW still provided) and now uses noexcept on C++11 or newer.
51 - move the darwin select workaround higher in ev.c, as newer versions of
52 darwin managed to break their broken select even more.
53 - ANDROID => __ANDROID__ (reported by enh@google.com).
54 - disable epoll_create1 on android because it has broken header files
55 and google is unwilling to fix them (reported by enh@google.com).
56 - avoid a minor compilation warning on win32.
57 - c++: remove deprecated dynamic throw() specifications.
58 - c++: improve the (unsupported) bad_loop exception class.
59 - backport perl ev_periodic example to C, untested.
60 - update libecb, biggets change is to include a memory fence
61 in ECB_MEMORY_FENCE_RELEASE on x86/amd64.
62 - minor autoconf/automake modernisation.
63
64 4.24 Wed Dec 28 05:19:55 CET 2016
65 - bump version to 4.24, as the release tarball inexplicably
66 didn't have the right version in ev.h, even though the cvs-tagged
67 version did have the right one (reported by Ales Teska).
68
69 4.23 Wed Nov 16 18:23:41 CET 2016
70 - move some declarations at the beginning to help certain retarded
71 microsoft compilers, even though their documentation claims
72 otherwise (reported by Ruslan Osmanov).
73
74 4.22 Sun Dec 20 22:11:50 CET 2015
75 - when epoll detects unremovable fds in the fd set, rebuild
76 only the epoll descriptor, not the signal pipe, to avoid
77 SIGPIPE in ev_async_send. This doesn't solve it on fork,
78 so document what needs to be done in ev_loop_fork
79 (analyzed by Benjamin Mahler).
80 - remove superfluous sys/timeb.h include on win32
81 (analyzed by Jason Madden).
82 - updated libecb.
83
84 4.20 Sat Jun 20 13:01:43 CEST 2015
85 - prefer noexcept over throw () with C++ 11.
86 - update ecb.h due to incompatibilities with c11.
87 - fix a potential aliasing issue when reading and writing
88 watcher callbacks.
89
90 4.19 Thu Sep 25 08:18:25 CEST 2014
91 - ev.h wasn't valid C++ anymore, which tripped compilers other than
92 clang, msvc or gcc (analyzed by Raphael 'kena' Poss). Unfortunately,
93 C++ doesn't support typedefs for function pointers fully, so the affected
94 declarations have to spell out the types each time.
95 - when not using autoconf, tighten the check for clock_gettime and related
96 functionality.
97
98 4.18 Fri Sep 5 17:55:26 CEST 2014
99 - events on files were not always generated properly with the
100 epoll backend (testcase by Assaf Inbal).
101 - mark event pipe fd as cloexec after a fork (analyzed by Sami Farin).
102 - (ecb) support m68k, m88k and sh (patch by Miod Vallat).
103 - use a reasonable fallback for EV_NSIG instead of erroring out
104 when we can't detect the signal set size.
105 - in the absence of autoconf, do not use the clock syscall
106 on glibc >= 2.17 (avoids the syscall AND -lrt on systems
107 doing clock_gettime in userspace).
108 - ensure extern "C" function pointers are used for externally-visible
109 loop callbacks (not watcher callbacks yet).
110 - (ecb) work around memory barriers and volatile apparently both being
111 broken in visual studio 2008 and later (analysed and patch by Nicolas Noble).
112
113 4.15 Fri Mar 1 12:04:50 CET 2013
114 - destroying a non-default loop would stop the global waitpid
115 watcher (Denis Bilenko).
116 - queueing pending watchers of higher priority from a watcher now invokes
117 them in a timely fashion (reported by Denis Bilenko).
118 - add throw() to all libev functions that cannot throw exceptions, for
119 further code size decrease when compiling for C++.
120 - add throw () to callbacks that must not throw exceptions (allocator,
121 syserr, loop acquire/release, periodic reschedule cbs).
122 - fix event_base_loop return code, add event_get_callback, event_base_new,
123 event_base_get_method calls to improve libevent 1.x emulation and add
124 some libevent 2.x functionality (based on a patch by Jeff Davey).
125 - add more memory fences to fix a bug reported by Jeff Davey. Better
126 be overfenced than underprotected.
127 - ev_run now returns a boolean status (true meaning watchers are
128 still active).
129 - ev_once: undef EV_ERROR in ev_kqueue.c, to avoid clashing with
130 libev's EV_ERROR (reported by 191919).
131 - (ecb) add memory fence support for xlC (Darin McBride).
132 - (ecb) add memory fence support for gcc-mips (Anton Kirilov).
133 - (ecb) add memory fence support for gcc-alpha (Christian Weisgerber).
134 - work around some kernels losing file descriptors by leaking
135 the kqueue descriptor in the child.
136 - work around linux inotify not reporting IN_ATTRIB changes for directories
137 in many cases.
138 - include sys/syscall.h instead of plain syscall.h.
139 - check for io watcher loops in ev_verify, check for the most
140 common reported usage bug in ev_io_start.
141 - choose socket vs. WSASocket at compiletime using EV_USE_WSASOCKET.
142 - always use WSASend/WSARecv directly on windows, hoping that this
143 works in all cases (unlike read/write/send/recv...).
144 - try to detect signals around a fork faster (test program by
145 Denis Bilenko).
146 - work around recent glibc versions that leak memory in realloc.
147 - rename ev::embed::set to ev::embed::set_embed to avoid clashing
148 the watcher base set (loop) method.
149 - rewrite the async/signal pipe logic to always keep a valid fd, which
150 simplifies (and hopefully correctifies :) the race checking
151 on fork, at the cost of one extra fd.
152 - add fat, msdos, jffs2, ramfs, ntfs and btrfs to the list of
153 inotify-supporting filesystems.
154 - move orig_CFLAGS assignment to after AC_INIT, as newer autoconf
155 versions ignore it before
156 (https://bugzilla.redhat.com/show_bug.cgi?id=908096).
157 - add some untested android support.
158 - enum expressions must be of type int (reported by Juan Pablo L).
159
160 4.11 Sat Feb 4 19:52:39 CET 2012
161 - INCOMPATIBLE CHANGE: ev_timer_again now clears the pending status, as
162 was documented already, but not implemented in the repeating case.
163 - new compiletime symbols: EV_NO_SMP and EV_NO_THREADS.
164 - fix a race where the workaround against the epoll fork bugs
165 caused signals to not be handled anymore.
166 - correct backend_fudge for most backends, and implement a windows
167 specific workaround to avoid looping because we call both
168 select and Sleep, both with different time resolutions.
169 - document range and guarantees of ev_sleep.
170 - document reasonable ranges for periodics interval and offset.
171 - rename backend_fudge to backend_mintime to avoid future confusion :)
172 - change the default periodic reschedule function to hopefully be more
173 exact and correct even in corner cases or in the far future.
174 - do not rely on -lm anymore: use it when available but use our
175 own floor () if it is missing. This should make it easier to embed,
176 as no external libraries are required.
177 - strategically import macros from libecb and mark rarely-used functions
178 as cache-cold (saving almost 2k code size on typical amd64 setups).
179 - add Symbols.ev and Symbols.event files, that were missing.
180 - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now).
181 - fix #3 "be smart about timeouts" to not "deadlock" when
182 timeout == now, also improve the section overall.
183 - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom.
184 - support new EV_API_STATIC mode to make all libev symbols
185 static.
186 - supply default CFLAGS of -g -O3 with gcc when original CFLAGS
187 were empty.
188
189 4.04 Wed Feb 16 09:01:51 CET 2011
190 - fix two problems in the native win32 backend, where reuse of fd's
191 with different underlying handles caused handles not to be removed
192 or added to the select set (analyzed and tested by Bert Belder).
193 - do no rely on ceil() in ev_e?poll.c.
194 - backport libev to HP-UX versions before 11 v3.
195 - configure did not detect nanosleep and clock_gettime properly when
196 they are available in the libc (as opposed to -lrt).
197
198 4.03 Tue Jan 11 14:37:25 CET 2011
199 - officially support polling files with all backends.
200 - support files, /dev/zero etc. the same way as select in the epoll
201 backend, by generating events on our own.
202 - ports backend: work around solaris bug 6874410 and many related ones
203 (EINTR, maybe more), with no performance loss (note that the solaris
204 bug report is actually wrong, reality is far more bizarre and broken
205 than that).
206 - define EV_READ/EV_WRITE as macros in event.h, as some programs use
207 #ifdef to test for them.
208 - new (experimental) function: ev_feed_signal.
209 - new (to become default) EVFLAG_NOSIGMASK flag.
210 - new EVBACKEND_MASK symbol.
211 - updated COMMON IDIOMS SECTION.
212
213 4.01 Fri Nov 5 21:51:29 CET 2010
214 - automake fucked it up, apparently, --add-missing -f is not quite enough
215 to make it update its files, so 4.00 didn't install ev++.h and
216 event.h on make install. grrr.
217 - ev_loop(count|depth) didn't return anything (Robin Haberkorn).
218 - change EV_UNDEF to 0xffffffff to silence some overzealous compilers.
219 - use "(libev) " prefix for all libev error messages now.
220
221 4.00 Mon Oct 25 12:32:12 CEST 2010
222 - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading.
223 - ev_embed_stop did not correctly stop the watcher (very good
224 testcase by Vladimir Timofeev).
225 - ev_run will now always update the current loop time - it erroneously
226 didn't when idle watchers were active, causing timers not to fire.
227 - fix a bug where a timeout of zero caused the timer not to fire
228 in the libevent emulation (testcase by Péter Szabó).
229 - applied win32 fixes by Michael Lenaghan (also James Mansion).
230 - replace EV_MINIMAL by EV_FEATURES.
231 - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it
232 seems the former is *much* faster than the latter.
233 - linux kernel version detection (for inotify bug workarounds)
234 did not work properly.
235 - reduce the number of spurious wake-ups with the ports backend.
236 - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu).
237 - do async init within ev_async_start, not ev_async_set, which avoids
238 an API quirk where the set function must be called in the C++ API
239 even when there is nothing to set.
240 - add (undocumented) EV_ENABLE when adding events with kqueue,
241 this might help with OS X, which seems to need it despite documenting
242 not to need it (helpfully pointed out by Tilghman Lesher).
243 - do not use poll by default on freebsd, it's broken (what isn't
244 on freebsd...).
245 - allow to embed epoll on kernels >= 2.6.32.
246 - configure now prepends -O3, not appends it, so one can still
247 override it.
248 - ev.pod: greatly expanded the portability section, added a porting
249 section, a description of watcher states and made lots of minor fixes.
250 - disable poll backend on AIX, the poll header spams the namespace
251 and it's not worth working around dead platforms (reported
252 and analyzed by Aivars Kalvans).
253 - improve header file compatibility of the standalone eventfd code
254 in an obscure case.
255 - implement EV_AVOID_STDIO option.
256 - do not use sscanf to parse linux version number (smaller, faster,
257 no sscanf dependency).
258 - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings.
259 - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs.
260 - add section on accept() problems to the manpage.
261 - rename EV_TIMEOUT to EV_TIMER.
262 - rename ev_loop_count/depth/verify/loop/unloop.
263 - remove ev_default_destroy and ev_default_fork.
264 - switch to two-digit minor version.
265 - work around an apparent gentoo compiler bug.
266 - define _DARWIN_UNLIMITED_SELECT. just so.
267 - use enum instead of #define for most constants.
268 - improve compatibility to older C++ compilers.
269 - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now
270 default arguments when compiled as C++.
271 - enable automake dependency tracking.
272 - ev_loop_new no longer leaks memory when loop creation failed.
273 - new ev_cleanup watcher type.
274
275 3.9 Thu Dec 31 07:59:59 CET 2009
276 - signalfd is no longer used by default and has to be requested
277 explicitly - this means that easy to catch bugs become hard to
278 catch race conditions, but the users have spoken.
279 - point out the unspecified signal mask in the documentation, and
280 that this is a race condition regardless of EV_SIGNALFD.
281 - backport inotify code to C89.
282 - inotify file descriptors could leak into child processes.
283 - ev_stat watchers could keep an erroneous extra ref on the loop,
284 preventing exit when unregistering all watchers (testcases
285 provided by ry@tinyclouds.org).
286 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
287 symbols to make it easier for apps to do their own fd management.
288 - support EV_IDLE_ENABLE being disabled in ev++.h
289 (patch by Didier Spezia).
290 - take advantage of inotify_init1, if available, to set cloexec/nonblock
291 on fd creation, to avoid races.
292 - the signal handling pipe wasn't always initialised under windows
293 (analysed by lekma).
294 - changed minimum glibc requirement from glibc 2.9 to 2.7, for
295 signalfd.
296 - add missing string.h include (Denis F. Latypoff).
297 - only replace ev_stat.prev when we detect an actual difference,
298 so prev is (almost) always different to attr. this might
299 have caused the problems with 04_stat.t.
300 - add ev::timer->remaining () method to C++ API.
301
302 3.8 Sun Aug 9 14:30:45 CEST 2009
303 - incompatible change: do not necessarily reset signal handler
304 to SIG_DFL when a sighandler is stopped.
305 - ev_default_destroy did not properly free or zero some members,
306 potentially causing crashes and memory corruption on repeated
307 ev_default_destroy/ev_default_loop calls.
308 - take advantage of signalfd on GNU/Linux systems.
309 - document that the signal mask might be in an unspecified
310 state when using libev's signal handling.
311 - take advantage of some GNU/Linux calls to set cloexec/nonblock
312 on fd creation, to avoid race conditions.
313
314 3.7 Fri Jul 17 16:36:32 CEST 2009
315 - ev_unloop and ev_loop wrongly used a global variable to exit loops,
316 instead of using a per-loop variable (bug caught by accident...).
317 - the ev_set_io_collect_interval interpretation has changed.
318 - add new functionality: ev_set_userdata, ev_userdata,
319 ev_set_invoke_pending_cb, ev_set_loop_release_cb,
320 ev_invoke_pending, ev_pending_count, together with a long example
321 about thread locking.
322 - add ev_timer_remaining (as requested by Denis F. Latypoff).
323 - add ev_loop_depth.
324 - calling ev_unloop in fork/prepare watchers will no longer poll
325 for new events.
326 - Denis F. Latypoff corrected many typos in example code snippets.
327 - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double-
328 check that the syscall number is available before trying to
329 use it (reported by ry@tinyclouds).
330 - use GetSystemTimeAsFileTime instead of _timeb on windows, for
331 slightly higher accuracy.
332 - properly declare ev_loop_verify and ev_now_update even when
333 !EV_MULTIPLICITY.
334 - do not compile in any priority code when EV_MAXPRI == EV_MINPRI.
335 - support EV_MINIMAL==2 for a reduced API.
336 - actually 0-initialise struct sigaction when installing signals.
337 - add section on hibernate and stopped processes to ev_timer docs.
338
339 3.6 Tue Apr 28 02:49:30 CEST 2009
340 - multiple timers becoming ready within an event loop iteration
341 will be invoked in the "correct" order now.
342 - do not leave the event loop early just because we have no active
343 watchers, fixing a problem when embedding a kqueue loop
344 that has active kernel events but no registered watchers
345 (reported by blacksand blacksand).
346 - correctly zero the idx values for arrays, so destroying and
347 reinitialising the default loop actually works (patch by
348 Malek Hadj-Ali).
349 - implement ev_suspend and ev_resume.
350 - new EV_CUSTOM revents flag for use by applications.
351 - add documentation section about priorities.
352 - add a glossary to the documentation.
353 - extend the ev_fork description slightly.
354 - optimize a jump out of call_pending.
355
356 3.53 Sun Feb 15 02:38:20 CET 2009
357 - fix a bug in event pipe creation on win32 that would cause a
358 failed assertion on event loop creation (patch by Malek Hadj-Ali).
359 - probe for CLOCK_REALTIME support at runtime as well and fall
360 back to gettimeofday if there is an error, to support older
361 operating systems with newer header files/libraries.
362 - prefer gettimeofday over clock_gettime with USE_CLOCK_SYSCALL
363 (default most everywhere), otherwise not.
364
365 3.52 Wed Jan 7 21:43:02 CET 2009
366 - fix compilation of select backend in fd_set mode when NFDBITS is
367 missing (to get it to compile on QNX, reported by Rodrigo Campos).
368 - better select-nfds handling when select backend is in fd_set mode.
369 - diagnose fd_set overruns when select backend is in fd_set mode.
370 - due to a thinko, instead of disabling everything but
371 select on the borked OS X platform, everything but select was
372 allowed (reported by Emanuele Giaquinta).
373 - actually verify that local and remote port are matching in
374 libev's socketpair emulation, which makes denial-of-service
375 attacks harder (but not impossible - it's windows). Make sure
376 it even works under vista, which thinks that getpeer/sockname
377 should return fantasy port numbers.
378 - include "libev" in all assertion messages for potentially
379 clearer diagnostics.
380 - event_get_version (libevent compatibility) returned
381 a useless string instead of the expected version string
382 (patch by W.C.A. Wijngaards).
383
384 3.51 Wed Dec 24 23:00:11 CET 2008
385 - fix a bug where an inotify watcher was added twice, causing
386 freezes on hash collisions (reported and analysed by Graham Leggett).
387 - new config symbol, EV_USE_CLOCK_SYSCALL, to make libev use
388 a direct syscall - slower, but no dependency on librt et al.
389 - assume negative return values != -1 signals success of port_getn
390 (http://cvs.epicsol.org/cgi/viewcvs.cgi/epic5/source/newio.c?rev=1.52)
391 (no known failure reports, but it doesn't hurt).
392 - fork detection in ev_embed now stops and restarts the watcher
393 automatically.
394 - EXPERIMENTAL: default the method to operator () in ev++.h,
395 to make it nicer to use functors (requested by Benedek László).
396 - fixed const object callbacks in ev++.h.
397 - replaced loop_ref argument of watcher.set (loop) by a direct
398 ev_loop * in ev++.h, to avoid clashes with functor patch.
399 - do not try to watch the empty string via inotify.
400 - inotify watchers could be leaked under certain circumstances.
401 - OS X 10.5 is actually even more broken than earlier versions,
402 so fall back to select on that piece of garbage.
403 - fixed some weirdness in the ev_embed documentation.
404
405 3.49 Wed Nov 19 11:26:53 CET 2008
406 - ev_stat watchers will now use inotify as a mere hint on
407 kernels <2.6.25, or if the filesystem is not in the
408 "known to be good" list.
409 - better mingw32 compatibility (it's not as borked as native win32)
410 (analysed by Roger Pack).
411 - include stdio.h in the example program, as too many people are
412 confused by the weird C language otherwise. I guess the next thing
413 I get told is that the "..." ellipses in the examples don't compile
414 with their C compiler.
415
416 3.48 Thu Oct 30 09:02:37 CET 2008
417 - further optimise away the EPOLL_CTL_ADD/MOD combo in the epoll
418 backend by assuming the kernel event mask hasn't changed if
419 ADD fails with EEXIST.
420 - work around spurious event notification bugs in epoll by using
421 a 32-bit generation counter. recreate kernel state if we receive
422 spurious notifications or unwanted events. this is very costly,
423 but I didn't come up with this horrible design.
424 - use memset to initialise most arrays now and do away with the
425 init functions.
426 - expand time-out strategies into a "Be smart about timeouts" section.
427 - drop the "struct" from all ev_watcher declarations in the
428 documentation and did other clarifications (yeah, it was a mistake
429 to have a struct AND a function called ev_loop).
430 - fix a bug where ev_default would not initialise the default
431 loop again after it was destroyed with ev_default_destroy.
432 - rename syserr to ev_syserr to avoid name clashes when embedding,
433 do similar changes for event.c.
434
435 3.45 Tue Oct 21 21:59:26 CEST 2008
436 - disable inotify usage on linux <2.6.25, as it is broken
437 (reported by Yoann Vandoorselaere).
438 - ev_stat erroneously would try to add inotify watchers
439 even when inotify wasn't available (this should only
440 have a performance impact).
441 - ev_once now passes both timeout and io to the callback if both
442 occur concurrently, instead of giving timeouts precedence.
443 - disable EV_USE_INOTIFY when sys/inotify.h is too old.
444
445 3.44 Mon Sep 29 05:18:39 CEST 2008
446 - embed watchers now automatically invoke ev_loop_fork on the
447 embedded loop when the parent loop forks.
448 - new function: ev_now_update (loop).
449 - verify_watcher was not marked static.
450 - improve the "associating..." manpage section.
451 - documentation tweaks here and there.
452
453 3.43 Sun Jul 6 05:34:41 CEST 2008
454 - include more include files on windows to get struct _stati64
455 (reported by Chris Hulbert, but doesn't quite fix his issue).
456 - add missing #include <io.h> in ev.c on windows (reported by
457 Matt Tolton).
458
459 3.42 Tue Jun 17 12:12:07 CEST 2008
460 - work around yet another windows bug: FD_SET actually adds fd's
461 multiple times to the fd_*SET*, despite official MSN docs claiming
462 otherwise. Reported and well-analysed by Matt Tolton.
463 - define NFDBITS to 0 when EV_SELECT_IS_WINSOCKET to make it compile
464 (reported any analysed by Chris Hulbert).
465 - fix a bug in ev_ebadf (this function is only used to catch
466 programming errors in the libev user). reported by Matt Tolton.
467 - fix a bug in fd_intern on win32 (could lead to compile errors
468 under some circumstances, but would work correctly if it compiles).
469 reported by Matt Tolton.
470 - (try to) work around missing lstat on windows.
471 - pass in the write fd set as except fd set under windows. windows
472 is so uncontrollably lame that it requires this. this means that
473 switching off oobinline is not supported (but tcp/ip doesn't
474 have oob, so that would be stupid anyways.
475 - use posix module symbol to auto-detect monotonic clock presence
476 and some other default values.
477
478 3.41 Fri May 23 18:42:54 CEST 2008
479 - work around an obscure bug in winsocket select: if you
480 provide only empty fd sets then select returns WSAEINVAL. how sucky.
481 - improve timer scheduling stability and reduce use of time_epsilon.
482 - use 1-based 2-heap for EV_MINIMAL, simplifies code, reduces
483 codesize and makes for better cache-efficiency.
484 - use 3-based 4-heap for !EV_MINIMAL. this makes better use
485 of cpu cache lines and gives better growth behaviour than
486 2-based heaps.
487 - cache timestamp within heap for !EV_MINIMAL, to avoid random
488 memory accesses.
489 - document/add EV_USE_4HEAP and EV_HEAP_CACHE_AT.
490 - fix a potential aliasing issue in ev_timer_again.
491 - add/document ev_periodic_at, retract direct access to ->at.
492 - improve ev_stat docs.
493 - add portability requirements section.
494 - fix manpage headers etc.
495 - normalise WSA error codes to lower range on windows.
496 - add consistency check code that can be called automatically
497 or on demand to check for internal structures (ev_loop_verify).
498
499 3.31 Wed Apr 16 20:45:04 CEST 2008
500 - added last minute fix for ev_poll.c by Brandon Black.
501
502 3.3 Wed Apr 16 19:04:10 CEST 2008
503 - event_base_loopexit should return 0 on success
504 (W.C.A. Wijngaards).
505 - added linux eventfd support.
506 - try to autodetect epoll and inotify support
507 by libc header version if not using autoconf.
508 - new symbols: EV_DEFAULT_UC and EV_DEFAULT_UC_.
509 - declare functions defined in ev.h as inline if
510 C99 or gcc are available.
511 - enable inlining with gcc versions 2 and 3.
512 - work around broken poll implementations potentially
513 not clearing revents field in ev_poll (Brandon Black)
514 (no such systems are known at this time).
515 - work around a bug in realloc on openbsd and darwin,
516 also makes the erroneous valgrind complaints
517 go away (noted by various people).
518 - fix ev_async_pending, add c++ wrapper for ev_async
519 (based on patch sent by Johannes Deisenhofer).
520 - add sensible set method to ev::embed.
521 - made integer constants type int in ev.h.
522
523 3.2 Wed Apr 2 17:11:19 CEST 2008
524 - fix a 64 bit overflow issue in the select backend,
525 by using fd_mask instead of int for the mask.
526 - rename internal sighandler to avoid clash with very old perls.
527 - entering ev_loop will not clear the ONESHOT or NONBLOCKING
528 flags of any outer loops anymore.
529 - add ev_async_pending.
530
531 3.1 Thu Mar 13 13:45:22 CET 2008
532 - implement ev_async watchers.
533 - only initialise signal pipe on demand.
534 - make use of sig_atomic_t configurable.
535 - improved documentation.
536
537 3.0 Mon Jan 28 13:14:47 CET 2008
538 - API/ABI bump to version 3.0.
539 - ev++.h includes "ev.h" by default now, not <ev.h>.
540 - slightly improved documentation.
541 - speed up signal detection after a fork.
542 - only optionally return trace status changed in ev_child
543 watchers.
544 - experimental (and undocumented) loop wrappers for ev++.h.
545
546 2.01 Tue Dec 25 08:04:41 CET 2007
547 - separate Changes file.
548 - fix ev_path_set => ev_stat_set typo.
549 - remove event_compat.h from the libev tarball.
550 - change how include files are found.
551 - doc updates.
552 - update licenses, explicitly allow for GPL relicensing.
553
554 2.0 Sat Dec 22 17:47:03 CET 2007
555 - new ev_sleep, ev_set_(io|timeout)_collect_interval.
556 - removed epoll from embeddable fd set.
557 - fix embed watchers.
558 - renamed ev_embed.loop to other.
559 - added exported Symbol tables.
560 - undefine member wrapper macros at the end of ev.c.
561 - respect EV_H in ev++.h.
562
563 1.86 Tue Dec 18 02:36:57 CET 2007
564 - fix memleak on loop destroy (not relevant for perl).
565
566 1.85 Fri Dec 14 20:32:40 CET 2007
567 - fix some aliasing issues w.r.t. timers and periodics
568 (not relevant for perl).
569
570 (for historic versions refer to EV/Changes, found in the Perl interface)
571
572 0.1 Wed Oct 31 21:31:48 CET 2007
573 - original version; hacked together in <24h.
574