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

Comparing libev/Changes (file contents):
Revision 1.24 by root, Wed Apr 9 22:33:55 2008 UTC vs.
Revision 1.158 by root, Wed Jan 13 12:44:33 2010 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
3 - applied win32 fixes by Michael Lenaghan.
4 - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu).
5
63.9 Thu Dec 31 07:59:59 CET 2009
7 - signalfd is no longer used by default and has to be requested
8 explicitly - this means that easy to catch bugs become hard to
9 catch race conditions, but the users have spoken.
10 - point out the unspecified signal mask in the documentation, and
11 that this is a race condition regardless of EV_SIGNALFD.
12 - backport inotify code to C89.
13 - inotify file descriptors could leak into child processes.
14 - ev_stat watchers could keep an errornous extra ref on the loop,
15 preventing exit when unregistering all watchers (testcases
16 provided by ry@tinyclouds.org).
17 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
18 symbols to make it easier for apps to do their own fd management.
19 - support EV_IDLE_ENABLE being disabled in ev++.h
20 (patch by Didier Spezia).
21 - take advantage of inotify_init1, if available, to set cloexec/nonblock
22 on fd creation, to avoid races.
23 - the signal handling pipe wasn't always initialised under windows
24 (analysed by lekma).
25 - changed minimum glibc requirement from glibc 2.9 to 2.7, for
26 signalfd.
27 - add missing string.h include (Denis F. Latypoff).
28 - only replace ev_stat.prev when we detect an actual difference,
29 so prev is (almost) always different to attr. this might
30 have caused the probems with 04_stat.t.
31 - add ev::timer->remaining () method to C++ API.
32
333.8 Sun Aug 9 14:30:45 CEST 2009
34 - incompatible change: do not necessarily reset signal handler
35 to SIG_DFL when a sighandler is stopped.
36 - ev_default_destroy did not properly free or zero some members,
37 potentially causing crashes and memory corruption on repated
38 ev_default_destroy/ev_default_loop calls.
39 - take advantage of signalfd on GNU/Linux systems.
40 - document that the signal mask might be in an unspecified
41 state when using libev's signal handling.
42 - take advantage of some GNU/Linux calls to set cloexec/nonblock
43 on fd creation, to avoid race conditions.
44
453.7 Fri Jul 17 16:36:32 CEST 2009
46 - ev_unloop and ev_loop wrongly used a global variable to exit loops,
47 instead of using a per-loop variable (bug caught by accident...).
48 - the ev_set_io_collect_interval interpretation has changed.
49 - add new functionality: ev_set_userdata, ev_userdata,
50 ev_set_invoke_pending_cb, ev_set_loop_release_cb,
51 ev_invoke_pending, ev_pending_count, together with a long example
52 about thread locking.
53 - add ev_timer_remaining (as requested by Denis F. Latypoff).
54 - add ev_loop_depth.
55 - calling ev_unloop in fork/prepare watchers will no longer poll
56 for new events.
57 - Denis F. Latypoff corrected many typos in example code snippets.
58 - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double-
59 check that the syscall number is available before trying to
60 use it (reported by ry@tinyclouds).
61 - use GetSystemTimeAsFileTime instead of _timeb on windows, for
62 slightly higher accuracy.
63 - properly declare ev_loop_verify and ev_now_update even when
64 !EV_MULTIPLICITY.
65 - do not compile in any priority code when EV_MAXPRI == EV_MINPRI.
66 - support EV_MINIMAL==2 for a reduced API.
67 - actually 0-initialise struct sigaction when installing signals.
68 - add section on hibernate and stopped processes to ev_timer docs.
69
703.6 Tue Apr 28 02:49:30 CEST 2009
71 - multiple timers becoming ready within an event loop iteration
72 will be invoked in the "correct" order now.
73 - do not leave the event loop early just because we have no active
74 watchers, fixing a problem when embedding a kqueue loop
75 that has active kernel events but no registered watchers
76 (reported by blacksand blacksand).
77 - correctly zero the idx values for arrays, so destroying and
78 reinitialising the default loop actually works (patch by
79 Malek Hadj-Ali).
80 - implement ev_suspend and ev_resume.
81 - new EV_CUSTOM revents flag for use by applications.
82 - add documentation section about priorites.
83 - add a glossary to the dcoumentation.
84 - extend the ev_fork description slightly.
85 - optimize a jump out of call_pending.
86
873.53 Sun Feb 15 02:38:20 CET 2009
88 - fix a bug in event pipe creation on win32 that would cause a
89 failed assertion on event loop creation (patch by Malek Hadj-Ali).
90 - probe for CLOCK_REALTIME support at runtime as well and fall
91 back to gettimeofday if there is an error, to support older
92 operating systems with newer header files/libraries.
93 - prefer gettimeofday over clock_gettime with USE_CLOCK_SYSCALL
94 (default most everywhere), otherwise not.
95
963.52 Wed Jan 7 21:43:02 CET 2009
97 - fix compilation of select backend in fd_set mode when NFDBITS is
98 missing (to get it to compile on QNX, reported by Rodrigo Campos).
99 - better select-nfds handling when select backend is in fd_set mode.
100 - diagnose fd_set overruns when select backend is in fd_set mode.
101 - due to a thinko, instead of disabling everything but
102 select on the borked OS X platform, everything but select was
103 allowed (reported by Emanuele Giaquinta).
104 - actually verify that local and remote port are matching in
105 libev's socketpair emulation, which makes denial-of-service
106 attacks harder (but not impossible - it's windows). Make sure
107 it even works under vista, which thinks that getpeer/sockname
108 should return fantasy port numbers.
109 - include "libev" in all assertion messages for potentially
110 clearer diagnostics.
111 - event_get_version (libevent compatibility) returned
112 a useless string instead of the expected version string
113 (patch by W.C.A. Wijngaards).
114
1153.51 Wed Dec 24 23:00:11 CET 2008
116 - fix a bug where an inotify watcher was added twice, causing
117 freezes on hash collisions (reported and analysed by Graham Leggett).
118 - new config symbol, EV_USE_CLOCK_SYSCALL, to make libev use
119 a direct syscall - slower, but no dependency on librt et al.
120 - assume negative return values != -1 signals success of port_getn
121 (http://cvs.epicsol.org/cgi/viewcvs.cgi/epic5/source/newio.c?rev=1.52)
122 (no known failure reports, but it doesn't hurt).
123 - fork detection in ev_embed now stops and restarts the watcher
124 automatically.
125 - EXPERIMENTAL: default the method to operator () in ev++.h,
126 to make it nicer to use functors (requested by Benedek László).
127 - fixed const object callbacks in ev++.h.
128 - replaced loop_ref argument of watcher.set (loop) by a direct
129 ev_loop * in ev++.h, to avoid clashes with functor patch.
130 - do not try to watch the empty string via inotify.
131 - inotify watchers could be leaked under certain circumstances.
132 - OS X 10.5 is actually even more broken than earlier versions,
133 so fall back to select on that piece of garbage.
134 - fixed some weirdness in the ev_embed documentation.
135
1363.49 Wed Nov 19 11:26:53 CET 2008
137 - ev_stat watchers will now use inotify as a mere hint on
138 kernels <2.6.25, or if the filesystem is not in the
139 "known to be good" list.
140 - better mingw32 compatibility (it's not as borked as native win32)
141 (analysed by Roger Pack).
142 - include stdio.h in the example program, as too many people are
143 confused by the weird C language otherwise. I guess the next thing
144 I get told is that the "..." ellipses in the examples don't compile
145 with their C compiler.
146
1473.48 Thu Oct 30 09:02:37 CET 2008
148 - further optimise away the EPOLL_CTL_ADD/MOD combo in the epoll
149 backend by assuming the kernel event mask hasn't changed if
150 ADD fails with EEXIST.
151 - work around spurious event notification bugs in epoll by using
152 a 32-bit generation counter. recreate kernel state if we receive
153 spurious notifications or unwanted events. this is very costly,
154 but I didn't come up with this horrible design.
155 - use memset to initialise most arrays now and do away with the
156 init functions.
157 - expand time-out strategies into a "Be smart about timeouts" section.
158 - drop the "struct" from all ev_watcher declarations in the
159 documentation and did other clarifications (yeah, it was a mistake
160 to have a struct AND a function called ev_loop).
161 - fix a bug where ev_default would not initialise the default
162 loop again after it was destroyed with ev_default_destroy.
163 - rename syserr to ev_syserr to avoid name clashes when embedding,
164 do similar changes for event.c.
165
1663.45 Tue Oct 21 21:59:26 CEST 2008
167 - disable inotify usage on linux <2.6.25, as it is broken
168 (reported by Yoann Vandoorselaere).
169 - ev_stat errornously would try to add inotify watchers
170 even when inotify wasn't available (this should only
171 have a performance impact).
172 - ev_once now passes both timeout and io to the callback if both
173 occur concurrently, instead of giving timeouts precedence.
174 - disable EV_USE_INOTIFY when sys/inotify.h is too old.
175
1763.44 Mon Sep 29 05:18:39 CEST 2008
177 - embed watchers now automatically invoke ev_loop_fork on the
178 embedded loop when the parent loop forks.
179 - new function: ev_now_update (loop).
180 - verify_watcher was not marked static.
181 - improve the "associating..." manpage section.
182 - documentation tweaks here and there.
183
1843.43 Sun Jul 6 05:34:41 CEST 2008
185 - include more include files on windows to get struct _stati64
186 (reported by Chris Hulbert, but doesn't quite fix his issue).
187 - add missing #include <io.h> in ev.c on windows (reported by
188 Matt Tolton).
189
1903.42 Tue Jun 17 12:12:07 CEST 2008
191 - work around yet another windows bug: FD_SET actually adds fd's
192 multiple times to the fd_*SET*, despite official MSN docs claiming
193 otherwise. Reported and well-analysed by Matt Tolton.
194 - define NFDBITS to 0 when EV_SELECT_IS_WINSOCKET to make it compile
195 (reported any analysed by Chris Hulbert).
196 - fix a bug in ev_ebadf (this function is only used to catch
197 programming errors in the libev user). reported by Matt Tolton.
198 - fix a bug in fd_intern on win32 (could lead to compile errors
199 under some circumstances, but would work correctly if it compiles).
200 reported by Matt Tolton.
201 - (try to) work around missing lstat on windows.
202 - pass in the write fd set as except fd set under windows. windows
203 is so uncontrollably lame that it requires this. this means that
204 switching off oobinline is not supported (but tcp/ip doesn't
205 have oob, so that would be stupid anyways.
206 - use posix module symbol to auto-detect monotonic clock presence
207 and some other default values.
208
2093.41 Fri May 23 18:42:54 CEST 2008
210 - work around an obscure bug in winsocket select: if you
211 provide only empty fd sets then select returns WSAEINVAL. how sucky.
212 - improve timer scheduling stability and reduce use of time_epsilon.
213 - use 1-based 2-heap for EV_MINIMAL, simplifies code, reduces
214 codesize and makes for better cache-efficiency.
215 - use 3-based 4-heap for !EV_MINIMAL. this makes better use
216 of cpu cache lines and gives better growth behaviour than
217 2-based heaps.
218 - cache timestamp within heap for !EV_MINIMAL, to avoid random
219 memory accesses.
220 - document/add EV_USE_4HEAP and EV_HEAP_CACHE_AT.
221 - fix a potential aliasing issue in ev_timer_again.
222 - add/document ev_periodic_at, retract direct access to ->at.
223 - improve ev_stat docs.
224 - add portability requirements section.
225 - fix manpage headers etc.
226 - normalise WSA error codes to lower range on windows.
227 - add consistency check code that can be called automatically
228 or on demand to check for internal structures (ev_loop_verify).
229
2303.31 Wed Apr 16 20:45:04 CEST 2008
231 - added last minute fix for ev_poll.c by Brandon Black.
232
2333.3 Wed Apr 16 19:04:10 CEST 2008
3 - event_base_loopexit should return 0 on success 234 - event_base_loopexit should return 0 on success
4 (W.C.A. Wijngaards). 235 (W.C.A. Wijngaards).
5 - work around poll() potentially not clearing revents
6 field in ev_poll (Brandon Black).
7 - added linux eventfd support. 236 - added linux eventfd support.
8 - try to autodetect epoll and inotify support 237 - try to autodetect epoll and inotify support
9 by libc header version if not using autoconf. 238 by libc header version if not using autoconf.
10 - new symbols: EV_DEFAULT_UC and EV_DEFAULT_UC_. 239 - new symbols: EV_DEFAULT_UC and EV_DEFAULT_UC_.
11 - declare functions defined in ev.h as inline if 240 - declare functions defined in ev.h as inline if
12 C99 or gcc are available. 241 C99 or gcc are available.
13 - enable inlining with gcc versions 2 and 3. 242 - enable inlining with gcc versions 2 and 3.
243 - work around broken poll implementations potentially
244 not clearing revents field in ev_poll (Brandon Black)
245 (no such systems are known at this time).
14 - work around a bug in realloc on openbsd and darwin, 246 - work around a bug in realloc on openbsd and darwin,
15 also makes the errornous valgrind complaints 247 also makes the errornous valgrind complaints
16 go away (noted by various people). 248 go away (noted by various people).
249 - fix ev_async_pending, add c++ wrapper for ev_async
250 (based on patch sent by Johannes Deisenhofer.
251 - add sensible set method to ev::embed.
252 - made integer constants type int in ev.h.
17 253
183.2 Wed Apr 2 17:11:19 CEST 2008 2543.2 Wed Apr 2 17:11:19 CEST 2008
19 - fix a 64 bit overflow issue in the select backend, 255 - fix a 64 bit overflow issue in the select backend,
20 by using fd_mask instead of int for the mask. 256 by using fd_mask instead of int for the mask.
21 - rename internal sighandler to avoid clash with very old perls. 257 - rename internal sighandler to avoid clash with very old perls.

Diff Legend

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