ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/Changes
(Generate patch)

Comparing libev/Changes (file contents):
Revision 1.42 by root, Wed May 21 23:35:01 2008 UTC vs.
Revision 1.211 by root, Tue Jan 4 01:28:56 2011 UTC

1Revision history for libev, a high-performance and full-featured event loop. 1Revision history for libev, a high-performance and full-featured event loop.
2 2
33.4 3TODO: move some other examples to common idioms? combining watchers,
4thread usage, coroutine switch?
5
6TODO: /dev/null epoll breakage, maybe catch EPERM?
74.03
8 - define EV_READ/EV_WRITE as macros in event.h, as some programs use
9 #ifdef to test for them.
10
114.01 Fri Nov 5 21:51:29 CET 2010
12 - automake fucked it up, apparently, --add-missing -f is not quite enough
13 to make it update its files, so 4.00 didn't install ev++.h and
14 event.h on make install. grrr.
15 - ev_loop(count|depth) didn't return anything (Robin Haberkorn).
16 - change EV_UNDEF to 0xffffffff to silence some overzealous compilers.
17 - use "(libev) " prefix for all libev error messages now.
18
194.00 Mon Oct 25 12:32:12 CEST 2010
20 - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading.
21 - ev_embed_stop did not correctly stop the watcher (very good
22 testcase by Vladimir Timofeev).
23 - ev_run will now always update the current loop time - it erroneously
24 didn't when idle watchers were active, causing timers not to fire.
25 - fix a bug where a timeout of zero caused the timer not to fire
26 in the libevent emulation (testcase by Péter Szabó).
27 - applied win32 fixes by Michael Lenaghan (also James Mansion).
28 - replace EV_MINIMAL by EV_FEATURES.
29 - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it
30 seems the former is *much* faster than the latter.
31 - linux kernel version detection (for inotify bug workarounds)
32 did not work properly.
33 - reduce the number of spurious wake-ups with the ports backend.
34 - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu).
35 - do async init within ev_async_start, not ev_async_set, which avoids
36 an API quirk where the set function must be called in the C++ API
37 even when there is nothing to set.
38 - add (undocumented) EV_ENABLE when adding events with kqueue,
39 this might help with OS X, which seems to need it despite documenting
40 not to need it (helpfully pointed out by Tilghman Lesher).
41 - do not use poll by default on freebsd, it's broken (what isn't
42 on freebsd...).
43 - allow to embed epoll on kernels >= 2.6.32.
44 - configure now prepends -O3, not appends it, so one can still
45 override it.
46 - ev.pod: greatly expanded the portability section, added a porting
47 section, a description of watcher states and made lots of minor fixes.
48 - disable poll backend on AIX, the poll header spams the namespace
49 and it's not worth working around dead platforms (reported
50 and analyzed by Aivars Kalvans).
51 - improve header file compatibility of the standalone eventfd code
52 in an obscure case.
53 - implement EV_AVOID_STDIO option.
54 - do not use sscanf to parse linux version number (smaller, faster,
55 no sscanf dependency).
56 - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings.
57 - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs.
58 - add section on accept() problems to the manpage.
59 - rename EV_TIMEOUT to EV_TIMER.
60 - rename ev_loop_count/depth/verify/loop/unloop.
61 - remove ev_default_destroy and ev_default_fork.
62 - switch to two-digit minor version.
63 - work around an apparent gentoo compiler bug.
64 - define _DARWIN_UNLIMITED_SELECT. just so.
65 - use enum instead of #define for most constants.
66 - improve compatibility to older C++ compilers.
67 - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now
68 default arguments when compiled as C++.
69 - enable automake dependency tracking.
70 - ev_loop_new no longer leaks memory when loop creation failed.
71 - new ev_cleanup watcher type.
72
733.9 Thu Dec 31 07:59:59 CET 2009
74 - signalfd is no longer used by default and has to be requested
75 explicitly - this means that easy to catch bugs become hard to
76 catch race conditions, but the users have spoken.
77 - point out the unspecified signal mask in the documentation, and
78 that this is a race condition regardless of EV_SIGNALFD.
79 - backport inotify code to C89.
80 - inotify file descriptors could leak into child processes.
81 - ev_stat watchers could keep an errornous extra ref on the loop,
82 preventing exit when unregistering all watchers (testcases
83 provided by ry@tinyclouds.org).
84 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
85 symbols to make it easier for apps to do their own fd management.
86 - support EV_IDLE_ENABLE being disabled in ev++.h
87 (patch by Didier Spezia).
88 - take advantage of inotify_init1, if available, to set cloexec/nonblock
89 on fd creation, to avoid races.
90 - the signal handling pipe wasn't always initialised under windows
91 (analysed by lekma).
92 - changed minimum glibc requirement from glibc 2.9 to 2.7, for
93 signalfd.
94 - add missing string.h include (Denis F. Latypoff).
95 - only replace ev_stat.prev when we detect an actual difference,
96 so prev is (almost) always different to attr. this might
97 have caused the problems with 04_stat.t.
98 - add ev::timer->remaining () method to C++ API.
99
1003.8 Sun Aug 9 14:30:45 CEST 2009
101 - incompatible change: do not necessarily reset signal handler
102 to SIG_DFL when a sighandler is stopped.
103 - ev_default_destroy did not properly free or zero some members,
104 potentially causing crashes and memory corruption on repeated
105 ev_default_destroy/ev_default_loop calls.
106 - take advantage of signalfd on GNU/Linux systems.
107 - document that the signal mask might be in an unspecified
108 state when using libev's signal handling.
109 - take advantage of some GNU/Linux calls to set cloexec/nonblock
110 on fd creation, to avoid race conditions.
111
1123.7 Fri Jul 17 16:36:32 CEST 2009
113 - ev_unloop and ev_loop wrongly used a global variable to exit loops,
114 instead of using a per-loop variable (bug caught by accident...).
115 - the ev_set_io_collect_interval interpretation has changed.
116 - add new functionality: ev_set_userdata, ev_userdata,
117 ev_set_invoke_pending_cb, ev_set_loop_release_cb,
118 ev_invoke_pending, ev_pending_count, together with a long example
119 about thread locking.
120 - add ev_timer_remaining (as requested by Denis F. Latypoff).
121 - add ev_loop_depth.
122 - calling ev_unloop in fork/prepare watchers will no longer poll
123 for new events.
124 - Denis F. Latypoff corrected many typos in example code snippets.
125 - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double-
126 check that the syscall number is available before trying to
127 use it (reported by ry@tinyclouds).
128 - use GetSystemTimeAsFileTime instead of _timeb on windows, for
129 slightly higher accuracy.
130 - properly declare ev_loop_verify and ev_now_update even when
131 !EV_MULTIPLICITY.
132 - do not compile in any priority code when EV_MAXPRI == EV_MINPRI.
133 - support EV_MINIMAL==2 for a reduced API.
134 - actually 0-initialise struct sigaction when installing signals.
135 - add section on hibernate and stopped processes to ev_timer docs.
136
1373.6 Tue Apr 28 02:49:30 CEST 2009
138 - multiple timers becoming ready within an event loop iteration
139 will be invoked in the "correct" order now.
140 - do not leave the event loop early just because we have no active
141 watchers, fixing a problem when embedding a kqueue loop
142 that has active kernel events but no registered watchers
143 (reported by blacksand blacksand).
144 - correctly zero the idx values for arrays, so destroying and
145 reinitialising the default loop actually works (patch by
146 Malek Hadj-Ali).
147 - implement ev_suspend and ev_resume.
148 - new EV_CUSTOM revents flag for use by applications.
149 - add documentation section about priorites.
150 - add a glossary to the dcoumentation.
151 - extend the ev_fork description slightly.
152 - optimize a jump out of call_pending.
153
1543.53 Sun Feb 15 02:38:20 CET 2009
155 - fix a bug in event pipe creation on win32 that would cause a
156 failed assertion on event loop creation (patch by Malek Hadj-Ali).
157 - probe for CLOCK_REALTIME support at runtime as well and fall
158 back to gettimeofday if there is an error, to support older
159 operating systems with newer header files/libraries.
160 - prefer gettimeofday over clock_gettime with USE_CLOCK_SYSCALL
161 (default most everywhere), otherwise not.
162
1633.52 Wed Jan 7 21:43:02 CET 2009
164 - fix compilation of select backend in fd_set mode when NFDBITS is
165 missing (to get it to compile on QNX, reported by Rodrigo Campos).
166 - better select-nfds handling when select backend is in fd_set mode.
167 - diagnose fd_set overruns when select backend is in fd_set mode.
168 - due to a thinko, instead of disabling everything but
169 select on the borked OS X platform, everything but select was
170 allowed (reported by Emanuele Giaquinta).
171 - actually verify that local and remote port are matching in
172 libev's socketpair emulation, which makes denial-of-service
173 attacks harder (but not impossible - it's windows). Make sure
174 it even works under vista, which thinks that getpeer/sockname
175 should return fantasy port numbers.
176 - include "libev" in all assertion messages for potentially
177 clearer diagnostics.
178 - event_get_version (libevent compatibility) returned
179 a useless string instead of the expected version string
180 (patch by W.C.A. Wijngaards).
181
1823.51 Wed Dec 24 23:00:11 CET 2008
183 - fix a bug where an inotify watcher was added twice, causing
184 freezes on hash collisions (reported and analysed by Graham Leggett).
185 - new config symbol, EV_USE_CLOCK_SYSCALL, to make libev use
186 a direct syscall - slower, but no dependency on librt et al.
187 - assume negative return values != -1 signals success of port_getn
188 (http://cvs.epicsol.org/cgi/viewcvs.cgi/epic5/source/newio.c?rev=1.52)
189 (no known failure reports, but it doesn't hurt).
190 - fork detection in ev_embed now stops and restarts the watcher
191 automatically.
192 - EXPERIMENTAL: default the method to operator () in ev++.h,
193 to make it nicer to use functors (requested by Benedek László).
194 - fixed const object callbacks in ev++.h.
195 - replaced loop_ref argument of watcher.set (loop) by a direct
196 ev_loop * in ev++.h, to avoid clashes with functor patch.
197 - do not try to watch the empty string via inotify.
198 - inotify watchers could be leaked under certain circumstances.
199 - OS X 10.5 is actually even more broken than earlier versions,
200 so fall back to select on that piece of garbage.
201 - fixed some weirdness in the ev_embed documentation.
202
2033.49 Wed Nov 19 11:26:53 CET 2008
204 - ev_stat watchers will now use inotify as a mere hint on
205 kernels <2.6.25, or if the filesystem is not in the
206 "known to be good" list.
207 - better mingw32 compatibility (it's not as borked as native win32)
208 (analysed by Roger Pack).
209 - include stdio.h in the example program, as too many people are
210 confused by the weird C language otherwise. I guess the next thing
211 I get told is that the "..." ellipses in the examples don't compile
212 with their C compiler.
213
2143.48 Thu Oct 30 09:02:37 CET 2008
215 - further optimise away the EPOLL_CTL_ADD/MOD combo in the epoll
216 backend by assuming the kernel event mask hasn't changed if
217 ADD fails with EEXIST.
218 - work around spurious event notification bugs in epoll by using
219 a 32-bit generation counter. recreate kernel state if we receive
220 spurious notifications or unwanted events. this is very costly,
221 but I didn't come up with this horrible design.
222 - use memset to initialise most arrays now and do away with the
223 init functions.
224 - expand time-out strategies into a "Be smart about timeouts" section.
225 - drop the "struct" from all ev_watcher declarations in the
226 documentation and did other clarifications (yeah, it was a mistake
227 to have a struct AND a function called ev_loop).
228 - fix a bug where ev_default would not initialise the default
229 loop again after it was destroyed with ev_default_destroy.
230 - rename syserr to ev_syserr to avoid name clashes when embedding,
231 do similar changes for event.c.
232
2333.45 Tue Oct 21 21:59:26 CEST 2008
234 - disable inotify usage on linux <2.6.25, as it is broken
235 (reported by Yoann Vandoorselaere).
236 - ev_stat erroneously would try to add inotify watchers
237 even when inotify wasn't available (this should only
238 have a performance impact).
239 - ev_once now passes both timeout and io to the callback if both
240 occur concurrently, instead of giving timeouts precedence.
241 - disable EV_USE_INOTIFY when sys/inotify.h is too old.
242
2433.44 Mon Sep 29 05:18:39 CEST 2008
244 - embed watchers now automatically invoke ev_loop_fork on the
245 embedded loop when the parent loop forks.
246 - new function: ev_now_update (loop).
247 - verify_watcher was not marked static.
248 - improve the "associating..." manpage section.
249 - documentation tweaks here and there.
250
2513.43 Sun Jul 6 05:34:41 CEST 2008
252 - include more include files on windows to get struct _stati64
253 (reported by Chris Hulbert, but doesn't quite fix his issue).
254 - add missing #include <io.h> in ev.c on windows (reported by
255 Matt Tolton).
256
2573.42 Tue Jun 17 12:12:07 CEST 2008
258 - work around yet another windows bug: FD_SET actually adds fd's
259 multiple times to the fd_*SET*, despite official MSN docs claiming
260 otherwise. Reported and well-analysed by Matt Tolton.
261 - define NFDBITS to 0 when EV_SELECT_IS_WINSOCKET to make it compile
262 (reported any analysed by Chris Hulbert).
263 - fix a bug in ev_ebadf (this function is only used to catch
264 programming errors in the libev user). reported by Matt Tolton.
265 - fix a bug in fd_intern on win32 (could lead to compile errors
266 under some circumstances, but would work correctly if it compiles).
267 reported by Matt Tolton.
268 - (try to) work around missing lstat on windows.
269 - pass in the write fd set as except fd set under windows. windows
270 is so uncontrollably lame that it requires this. this means that
271 switching off oobinline is not supported (but tcp/ip doesn't
272 have oob, so that would be stupid anyways.
273 - use posix module symbol to auto-detect monotonic clock presence
274 and some other default values.
275
2763.41 Fri May 23 18:42:54 CEST 2008
4 - work around an (undocumented) bug in winsocket select: if you 277 - work around an obscure bug in winsocket select: if you
5 provide only empty fd sets then select returns WSAEINVAL. how sucky. 278 provide only empty fd sets then select returns WSAEINVAL. how sucky.
6 - improve timer scheduling stability and reduce use of time_epsilon. 279 - improve timer scheduling stability and reduce use of time_epsilon.
7 - use 1-based 2-heap for EV_MINIMAL, simplifies code, reduces 280 - use 1-based 2-heap for EV_MINIMAL, simplifies code, reduces
8 codesize and makes for better cache-efficiency. 281 codesize and makes for better cache-efficiency.
9 - use 3-based 4-heap for !EV_MINIMAL. this makes better use 282 - use 3-based 4-heap for !EV_MINIMAL. this makes better use
… …
16 - add/document ev_periodic_at, retract direct access to ->at. 289 - add/document ev_periodic_at, retract direct access to ->at.
17 - improve ev_stat docs. 290 - improve ev_stat docs.
18 - add portability requirements section. 291 - add portability requirements section.
19 - fix manpage headers etc. 292 - fix manpage headers etc.
20 - normalise WSA error codes to lower range on windows. 293 - normalise WSA error codes to lower range on windows.
21 - start some (internal only) consistency check code that 294 - add consistency check code that can be called automatically
22 can be called frequently to catch bugs. 295 or on demand to check for internal structures (ev_loop_verify).
23 296
243.31 Wed Apr 16 20:45:04 CEST 2008 2973.31 Wed Apr 16 20:45:04 CEST 2008
25 - added last minute fix for ev_poll.c by Brandon Black. 298 - added last minute fix for ev_poll.c by Brandon Black.
26 299
273.3 Wed Apr 16 19:04:10 CEST 2008 3003.3 Wed Apr 16 19:04:10 CEST 2008
… …
36 - enable inlining with gcc versions 2 and 3. 309 - enable inlining with gcc versions 2 and 3.
37 - work around broken poll implementations potentially 310 - work around broken poll implementations potentially
38 not clearing revents field in ev_poll (Brandon Black) 311 not clearing revents field in ev_poll (Brandon Black)
39 (no such systems are known at this time). 312 (no such systems are known at this time).
40 - work around a bug in realloc on openbsd and darwin, 313 - work around a bug in realloc on openbsd and darwin,
41 also makes the errornous valgrind complaints 314 also makes the erroneous valgrind complaints
42 go away (noted by various people). 315 go away (noted by various people).
43 - fix ev_async_pending, add c++ wrapper for ev_async 316 - fix ev_async_pending, add c++ wrapper for ev_async
44 (based on patch sent by Johannes Deisenhofer. 317 (based on patch sent by Johannes Deisenhofer).
45 - add sensible set method to ev::embed. 318 - add sensible set method to ev::embed.
46 - made integer constants type int in ev.h. 319 - made integer constants type int in ev.h.
47 320
483.2 Wed Apr 2 17:11:19 CEST 2008 3213.2 Wed Apr 2 17:11:19 CEST 2008
49 - fix a 64 bit overflow issue in the select backend, 322 - fix a 64 bit overflow issue in the select backend,

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines