ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/Changes
Revision: 1.161
Committed: Sun Feb 14 19:23:19 2010 UTC (14 years, 3 months ago) by root
Branch: MAIN
Changes since 1.160: +2 -0 lines
Log Message:
forgot to ev_stop in ev_embed_stop

File Contents

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