ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/Changes
Revision: 1.156
Committed: Thu Dec 31 07:04:33 2009 UTC (14 years, 6 months ago) by root
Branch: MAIN
CVS Tags: rel-3_9
Changes since 1.155: +1 -0 lines
Log Message:
3.9

File Contents

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