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