ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/Changes
Revision: 1.117
Committed: Mon Jun 29 11:53:48 2009 UTC (14 years, 10 months ago) by root
Branch: MAIN
Changes since 1.116: +1 -1 lines
Log Message:
*** empty log message ***

File Contents

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