ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/Changes
Revision: 1.144
Committed: Sat Aug 22 20:40:02 2009 UTC (14 years, 8 months ago) by root
Branch: MAIN
Changes since 1.143: +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 - backport inotify code to C89.
4 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
5 symbols to make ti easier for apps to do their own fd management.
6 - support EV_IDLE_ENABLE being disabled in ev++.h
7 (patch by Didier Spezia).
8
9 3.8 Sun Aug 9 14:30:45 CEST 2009
10 - incompatible change: do not necessarily reset signal handler
11 to SIG_DFL when a sighandler is stopped.
12 - ev_default_destroy did not properly free or zero some members,
13 potentially causing crashes and memory corruption on repated
14 ev_default_destroy/ev_default_loop calls.
15 - take advantage of signalfd on GNU/Linux systems.
16 - document that the signal mask might be in an unspecified
17 state when using libev's signal handling.
18 - take advantage of some GNU/Linux calls to set cloexec/nonblock
19 on fd creation, to avoid race conditions.
20
21 3.7 Fri Jul 17 16:36:32 CEST 2009
22 - ev_unloop and ev_loop wrongly used a global variable to exit loops,
23 instead of using a per-loop variable (bug caught by accident...).
24 - the ev_set_io_collect_interval interpretation has changed.
25 - add new functionality: ev_set_userdata, ev_userdata,
26 ev_set_invoke_pending_cb, ev_set_loop_release_cb,
27 ev_invoke_pending, ev_pending_count, together with a long example
28 about thread locking.
29 - add ev_timer_remaining (as requested by Denis F. Latypoff).
30 - add ev_loop_depth.
31 - calling ev_unloop in fork/prepare watchers will no longer poll
32 for new events.
33 - Denis F. Latypoff corrected many typos in example code snippets.
34 - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double-
35 check that the syscall number is available before trying to
36 use it (reported by ry@tinyclouds).
37 - use GetSystemTimeAsFileTime instead of _timeb on windows, for
38 slightly higher accuracy.
39 - properly declare ev_loop_verify and ev_now_update even when
40 !EV_MULTIPLICITY.
41 - do not compile in any priority code when EV_MAXPRI == EV_MINPRI.
42 - support EV_MINIMAL==2 for a reduced API.
43 - actually 0-initialise struct sigaction when installing signals.
44 - add section on hibernate and stopped processes to ev_timer docs.
45
46 3.6 Tue Apr 28 02:49:30 CEST 2009
47 - multiple timers becoming ready within an event loop iteration
48 will be invoked in the "correct" order now.
49 - do not leave the event loop early just because we have no active
50 watchers, fixing a problem when embedding a kqueue loop
51 that has active kernel events but no registered watchers
52 (reported by blacksand blacksand).
53 - correctly zero the idx values for arrays, so destroying and
54 reinitialising the default loop actually works (patch by
55 Malek Hadj-Ali).
56 - implement ev_suspend and ev_resume.
57 - new EV_CUSTOM revents flag for use by applications.
58 - add documentation section about priorites.
59 - add a glossary to the dcoumentation.
60 - extend the ev_fork description slightly.
61 - optimize a jump out of call_pending.
62
63 3.53 Sun Feb 15 02:38:20 CET 2009
64 - fix a bug in event pipe creation on win32 that would cause a
65 failed assertion on event loop creation (patch by Malek Hadj-Ali).
66 - probe for CLOCK_REALTIME support at runtime as well and fall
67 back to gettimeofday if there is an error, to support older
68 operating systems with newer header files/libraries.
69 - prefer gettimeofday over clock_gettime with USE_CLOCK_SYSCALL
70 (default most everywhere), otherwise not.
71
72 3.52 Wed Jan 7 21:43:02 CET 2009
73 - fix compilation of select backend in fd_set mode when NFDBITS is
74 missing (to get it to compile on QNX, reported by Rodrigo Campos).
75 - better select-nfds handling when select backend is in fd_set mode.
76 - diagnose fd_set overruns when select backend is in fd_set mode.
77 - due to a thinko, instead of disabling everything but
78 select on the borked OS X platform, everything but select was
79 allowed (reported by Emanuele Giaquinta).
80 - actually verify that local and remote port are matching in
81 libev's socketpair emulation, which makes denial-of-service
82 attacks harder (but not impossible - it's windows). Make sure
83 it even works under vista, which thinks that getpeer/sockname
84 should return fantasy port numbers.
85 - include "libev" in all assertion messages for potentially
86 clearer diagnostics.
87 - event_get_version (libevent compatibility) returned
88 a useless string instead of the expected version string
89 (patch by W.C.A. Wijngaards).
90
91 3.51 Wed Dec 24 23:00:11 CET 2008
92 - fix a bug where an inotify watcher was added twice, causing
93 freezes on hash collisions (reported and analysed by Graham Leggett).
94 - new config symbol, EV_USE_CLOCK_SYSCALL, to make libev use
95 a direct syscall - slower, but no dependency on librt et al.
96 - assume negative return values != -1 signals success of port_getn
97 (http://cvs.epicsol.org/cgi/viewcvs.cgi/epic5/source/newio.c?rev=1.52)
98 (no known failure reports, but it doesn't hurt).
99 - fork detection in ev_embed now stops and restarts the watcher
100 automatically.
101 - EXPERIMENTAL: default the method to operator () in ev++.h,
102 to make it nicer to use functors (requested by Benedek László).
103 - fixed const object callbacks in ev++.h.
104 - replaced loop_ref argument of watcher.set (loop) by a direct
105 ev_loop * in ev++.h, to avoid clashes with functor patch.
106 - do not try to watch the empty string via inotify.
107 - inotify watchers could be leaked under certain circumstances.
108 - OS X 10.5 is actually even more broken than earlier versions,
109 so fall back to select on that piece of garbage.
110 - fixed some weirdness in the ev_embed documentation.
111
112 3.49 Wed Nov 19 11:26:53 CET 2008
113 - ev_stat watchers will now use inotify as a mere hint on
114 kernels <2.6.25, or if the filesystem is not in the
115 "known to be good" list.
116 - better mingw32 compatibility (it's not as borked as native win32)
117 (analysed by Roger Pack).
118 - include stdio.h in the example program, as too many people are
119 confused by the weird C language otherwise. I guess the next thing
120 I get told is that the "..." ellipses in the examples don't compile
121 with their C compiler.
122
123 3.48 Thu Oct 30 09:02:37 CET 2008
124 - further optimise away the EPOLL_CTL_ADD/MOD combo in the epoll
125 backend by assuming the kernel event mask hasn't changed if
126 ADD fails with EEXIST.
127 - work around spurious event notification bugs in epoll by using
128 a 32-bit generation counter. recreate kernel state if we receive
129 spurious notifications or unwanted events. this is very costly,
130 but I didn't come up with this horrible design.
131 - use memset to initialise most arrays now and do away with the
132 init functions.
133 - expand time-out strategies into a "Be smart about timeouts" section.
134 - drop the "struct" from all ev_watcher declarations in the
135 documentation and did other clarifications (yeah, it was a mistake
136 to have a struct AND a function called ev_loop).
137 - fix a bug where ev_default would not initialise the default
138 loop again after it was destroyed with ev_default_destroy.
139 - rename syserr to ev_syserr to avoid name clashes when embedding,
140 do similar changes for event.c.
141
142 3.45 Tue Oct 21 21:59:26 CEST 2008
143 - disable inotify usage on linux <2.6.25, as it is broken
144 (reported by Yoann Vandoorselaere).
145 - ev_stat errornously would try to add inotify watchers
146 even when inotify wasn't available (this should only
147 have a performance impact).
148 - ev_once now passes both timeout and io to the callback if both
149 occur concurrently, instead of giving timeouts precedence.
150 - disable EV_USE_INOTIFY when sys/inotify.h is too old.
151
152 3.44 Mon Sep 29 05:18:39 CEST 2008
153 - embed watchers now automatically invoke ev_loop_fork on the
154 embedded loop when the parent loop forks.
155 - new function: ev_now_update (loop).
156 - verify_watcher was not marked static.
157 - improve the "associating..." manpage section.
158 - documentation tweaks here and there.
159
160 3.43 Sun Jul 6 05:34:41 CEST 2008
161 - include more include files on windows to get struct _stati64
162 (reported by Chris Hulbert, but doesn't quite fix his issue).
163 - add missing #include <io.h> in ev.c on windows (reported by
164 Matt Tolton).
165
166 3.42 Tue Jun 17 12:12:07 CEST 2008
167 - work around yet another windows bug: FD_SET actually adds fd's
168 multiple times to the fd_*SET*, despite official MSN docs claiming
169 otherwise. Reported and well-analysed by Matt Tolton.
170 - define NFDBITS to 0 when EV_SELECT_IS_WINSOCKET to make it compile
171 (reported any analysed by Chris Hulbert).
172 - fix a bug in ev_ebadf (this function is only used to catch
173 programming errors in the libev user). reported by Matt Tolton.
174 - fix a bug in fd_intern on win32 (could lead to compile errors
175 under some circumstances, but would work correctly if it compiles).
176 reported by Matt Tolton.
177 - (try to) work around missing lstat on windows.
178 - pass in the write fd set as except fd set under windows. windows
179 is so uncontrollably lame that it requires this. this means that
180 switching off oobinline is not supported (but tcp/ip doesn't
181 have oob, so that would be stupid anyways.
182 - use posix module symbol to auto-detect monotonic clock presence
183 and some other default values.
184
185 3.41 Fri May 23 18:42:54 CEST 2008
186 - work around an obscure bug in winsocket select: if you
187 provide only empty fd sets then select returns WSAEINVAL. how sucky.
188 - improve timer scheduling stability and reduce use of time_epsilon.
189 - use 1-based 2-heap for EV_MINIMAL, simplifies code, reduces
190 codesize and makes for better cache-efficiency.
191 - use 3-based 4-heap for !EV_MINIMAL. this makes better use
192 of cpu cache lines and gives better growth behaviour than
193 2-based heaps.
194 - cache timestamp within heap for !EV_MINIMAL, to avoid random
195 memory accesses.
196 - document/add EV_USE_4HEAP and EV_HEAP_CACHE_AT.
197 - fix a potential aliasing issue in ev_timer_again.
198 - add/document ev_periodic_at, retract direct access to ->at.
199 - improve ev_stat docs.
200 - add portability requirements section.
201 - fix manpage headers etc.
202 - normalise WSA error codes to lower range on windows.
203 - add consistency check code that can be called automatically
204 or on demand to check for internal structures (ev_loop_verify).
205
206 3.31 Wed Apr 16 20:45:04 CEST 2008
207 - added last minute fix for ev_poll.c by Brandon Black.
208
209 3.3 Wed Apr 16 19:04:10 CEST 2008
210 - event_base_loopexit should return 0 on success
211 (W.C.A. Wijngaards).
212 - added linux eventfd support.
213 - try to autodetect epoll and inotify support
214 by libc header version if not using autoconf.
215 - new symbols: EV_DEFAULT_UC and EV_DEFAULT_UC_.
216 - declare functions defined in ev.h as inline if
217 C99 or gcc are available.
218 - enable inlining with gcc versions 2 and 3.
219 - work around broken poll implementations potentially
220 not clearing revents field in ev_poll (Brandon Black)
221 (no such systems are known at this time).
222 - work around a bug in realloc on openbsd and darwin,
223 also makes the errornous valgrind complaints
224 go away (noted by various people).
225 - fix ev_async_pending, add c++ wrapper for ev_async
226 (based on patch sent by Johannes Deisenhofer.
227 - add sensible set method to ev::embed.
228 - made integer constants type int in ev.h.
229
230 3.2 Wed Apr 2 17:11:19 CEST 2008
231 - fix a 64 bit overflow issue in the select backend,
232 by using fd_mask instead of int for the mask.
233 - rename internal sighandler to avoid clash with very old perls.
234 - entering ev_loop will not clear the ONESHOT or NONBLOCKING
235 flags of any outer loops anymore.
236 - add ev_async_pending.
237
238 3.1 Thu Mar 13 13:45:22 CET 2008
239 - implement ev_async watchers.
240 - only initialise signal pipe on demand.
241 - make use of sig_atomic_t configurable.
242 - improved documentation.
243
244 3.0 Mon Jan 28 13:14:47 CET 2008
245 - API/ABI bump to version 3.0.
246 - ev++.h includes "ev.h" by default now, not <ev.h>.
247 - slightly improved documentation.
248 - speed up signal detection after a fork.
249 - only optionally return trace status changed in ev_child
250 watchers.
251 - experimental (and undocumented) loop wrappers for ev++.h.
252
253 2.01 Tue Dec 25 08:04:41 CET 2007
254 - separate Changes file.
255 - fix ev_path_set => ev_stat_set typo.
256 - remove event_compat.h from the libev tarball.
257 - change how include files are found.
258 - doc updates.
259 - update licenses, explicitly allow for GPL relicensing.
260
261 2.0 Sat Dec 22 17:47:03 CET 2007
262 - new ev_sleep, ev_set_(io|timeout)_collect_interval.
263 - removed epoll from embeddable fd set.
264 - fix embed watchers.
265 - renamed ev_embed.loop to other.
266 - added exported Symbol tables.
267 - undefine member wrapper macros at the end of ev.c.
268 - respect EV_H in ev++.h.
269
270 1.86 Tue Dec 18 02:36:57 CET 2007
271 - fix memleak on loop destroy (not relevant for perl).
272
273 1.85 Fri Dec 14 20:32:40 CET 2007
274 - fix some aliasing issues w.r.t. timers and periodics
275 (not relevant for perl).
276
277 (for historic versions refer to EV/Changes, found in the Perl interface)
278
279 0.1 Wed Oct 31 21:31:48 CET 2007
280 - original version; hacked together in <24h.
281