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