ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/Changes
Revision: 1.115
Committed: Mon Jun 29 04:41:34 2009 UTC (14 years, 10 months ago) by root
Branch: MAIN
Changes since 1.114: +7 -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 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
11 3.6 Tue Apr 28 02:49:30 CEST 2009
12 - multiple timers becoming ready within an event loop iteration
13 will be invoked in the "correct" order now.
14 - do not leave the event loop early just because we have no active
15 watchers, fixing a problem when embedding a kqueue loop
16 that has active kernel events but no registered watchers
17 (reported by blacksand blacksand).
18 - correctly zero the idx values for arrays, so destroying and
19 reinitialising the default loop actually works (patch by
20 Malek Hadj-Ali).
21 - implement ev_suspend and ev_resume.
22 - new EV_CUSTOM revents flag for use by applications.
23 - add documentation section about priorites.
24 - add a glossary to the dcoumentation.
25 - extend the ev_fork description slightly.
26 - optimize a jump out of call_pending.
27
28 3.53 Sun Feb 15 02:38:20 CET 2009
29 - fix a bug in event pipe creation on win32 that would cause a
30 failed assertion on event loop creation (patch by Malek Hadj-Ali).
31 - probe for CLOCK_REALTIME support at runtime as well and fall
32 back to gettimeofday if there is an error, to support older
33 operating systems with newer header files/libraries.
34 - prefer gettimeofday over clock_gettime with USE_CLOCK_SYSCALL
35 (default most everywhere), otherwise not.
36
37 3.52 Wed Jan 7 21:43:02 CET 2009
38 - fix compilation of select backend in fd_set mode when NFDBITS is
39 missing (to get it to compile on QNX, reported by Rodrigo Campos).
40 - better select-nfds handling when select backend is in fd_set mode.
41 - diagnose fd_set overruns when select backend is in fd_set mode.
42 - due to a thinko, instead of disabling everything but
43 select on the borked OS X platform, everything but select was
44 allowed (reported by Emanuele Giaquinta).
45 - actually verify that local and remote port are matching in
46 libev's socketpair emulation, which makes denial-of-service
47 attacks harder (but not impossible - it's windows). Make sure
48 it even works under vista, which thinks that getpeer/sockname
49 should return fantasy port numbers.
50 - include "libev" in all assertion messages for potentially
51 clearer diagnostics.
52 - event_get_version (libevent compatibility) returned
53 a useless string instead of the expected version string
54 (patch by W.C.A. Wijngaards).
55
56 3.51 Wed Dec 24 23:00:11 CET 2008
57 - fix a bug where an inotify watcher was added twice, causing
58 freezes on hash collisions (reported and analysed by Graham Leggett).
59 - new config symbol, EV_USE_CLOCK_SYSCALL, to make libev use
60 a direct syscall - slower, but no dependency on librt et al.
61 - assume negative return values != -1 signals success of port_getn
62 (http://cvs.epicsol.org/cgi/viewcvs.cgi/epic5/source/newio.c?rev=1.52)
63 (no known failure reports, but it doesn't hurt).
64 - fork detection in ev_embed now stops and restarts the watcher
65 automatically.
66 - EXPERIMENTAL: default the method to operator () in ev++.h,
67 to make it nicer to use functors (requested by Benedek László).
68 - fixed const object callbacks in ev++.h.
69 - replaced loop_ref argument of watcher.set (loop) by a direct
70 ev_loop * in ev++.h, to avoid clashes with functor patch.
71 - do not try to watch the empty string via inotify.
72 - inotify watchers could be leaked under certain circumstances.
73 - OS X 10.5 is actually even more broken than earlier versions,
74 so fall back to select on that piece of garbage.
75 - fixed some weirdness in the ev_embed documentation.
76
77 3.49 Wed Nov 19 11:26:53 CET 2008
78 - ev_stat watchers will now use inotify as a mere hint on
79 kernels <2.6.25, or if the filesystem is not in the
80 "known to be good" list.
81 - better mingw32 compatibility (it's not as borked as native win32)
82 (analysed by Roger Pack).
83 - include stdio.h in the example program, as too many people are
84 confused by the weird C language otherwise. I guess the next thing
85 I get told is that the "..." ellipses in the examples don't compile
86 with their C compiler.
87
88 3.48 Thu Oct 30 09:02:37 CET 2008
89 - further optimise away the EPOLL_CTL_ADD/MOD combo in the epoll
90 backend by assuming the kernel event mask hasn't changed if
91 ADD fails with EEXIST.
92 - work around spurious event notification bugs in epoll by using
93 a 32-bit generation counter. recreate kernel state if we receive
94 spurious notifications or unwanted events. this is very costly,
95 but I didn't come up with this horrible design.
96 - use memset to initialise most arrays now and do away with the
97 init functions.
98 - expand time-out strategies into a "Be smart about timeouts" section.
99 - drop the "struct" from all ev_watcher declarations in the
100 documentation and did other clarifications (yeah, it was a mistake
101 to have a struct AND a function called ev_loop).
102 - fix a bug where ev_default would not initialise the default
103 loop again after it was destroyed with ev_default_destroy.
104 - rename syserr to ev_syserr to avoid name clashes when embedding,
105 do similar changes for event.c.
106
107 3.45 Tue Oct 21 21:59:26 CEST 2008
108 - disable inotify usage on linux <2.6.25, as it is broken
109 (reported by Yoann Vandoorselaere).
110 - ev_stat errornously would try to add inotify watchers
111 even when inotify wasn't available (this should only
112 have a performance impact).
113 - ev_once now passes both timeout and io to the callback if both
114 occur concurrently, instead of giving timeouts precedence.
115 - disable EV_USE_INOTIFY when sys/inotify.h is too old.
116
117 3.44 Mon Sep 29 05:18:39 CEST 2008
118 - embed watchers now automatically invoke ev_loop_fork on the
119 embedded loop when the parent loop forks.
120 - new function: ev_now_update (loop).
121 - verify_watcher was not marked static.
122 - improve the "associating..." manpage section.
123 - documentation tweaks here and there.
124
125 3.43 Sun Jul 6 05:34:41 CEST 2008
126 - include more include files on windows to get struct _stati64
127 (reported by Chris Hulbert, but doesn't quite fix his issue).
128 - add missing #include <io.h> in ev.c on windows (reported by
129 Matt Tolton).
130
131 3.42 Tue Jun 17 12:12:07 CEST 2008
132 - work around yet another windows bug: FD_SET actually adds fd's
133 multiple times to the fd_*SET*, despite official MSN docs claiming
134 otherwise. Reported and well-analysed by Matt Tolton.
135 - define NFDBITS to 0 when EV_SELECT_IS_WINSOCKET to make it compile
136 (reported any analysed by Chris Hulbert).
137 - fix a bug in ev_ebadf (this function is only used to catch
138 programming errors in the libev user). reported by Matt Tolton.
139 - fix a bug in fd_intern on win32 (could lead to compile errors
140 under some circumstances, but would work correctly if it compiles).
141 reported by Matt Tolton.
142 - (try to) work around missing lstat on windows.
143 - pass in the write fd set as except fd set under windows. windows
144 is so uncontrollably lame that it requires this. this means that
145 switching off oobinline is not supported (but tcp/ip doesn't
146 have oob, so that would be stupid anyways.
147 - use posix module symbol to auto-detect monotonic clock presence
148 and some other default values.
149
150 3.41 Fri May 23 18:42:54 CEST 2008
151 - work around an obscure bug in winsocket select: if you
152 provide only empty fd sets then select returns WSAEINVAL. how sucky.
153 - improve timer scheduling stability and reduce use of time_epsilon.
154 - use 1-based 2-heap for EV_MINIMAL, simplifies code, reduces
155 codesize and makes for better cache-efficiency.
156 - use 3-based 4-heap for !EV_MINIMAL. this makes better use
157 of cpu cache lines and gives better growth behaviour than
158 2-based heaps.
159 - cache timestamp within heap for !EV_MINIMAL, to avoid random
160 memory accesses.
161 - document/add EV_USE_4HEAP and EV_HEAP_CACHE_AT.
162 - fix a potential aliasing issue in ev_timer_again.
163 - add/document ev_periodic_at, retract direct access to ->at.
164 - improve ev_stat docs.
165 - add portability requirements section.
166 - fix manpage headers etc.
167 - normalise WSA error codes to lower range on windows.
168 - add consistency check code that can be called automatically
169 or on demand to check for internal structures (ev_loop_verify).
170
171 3.31 Wed Apr 16 20:45:04 CEST 2008
172 - added last minute fix for ev_poll.c by Brandon Black.
173
174 3.3 Wed Apr 16 19:04:10 CEST 2008
175 - event_base_loopexit should return 0 on success
176 (W.C.A. Wijngaards).
177 - added linux eventfd support.
178 - try to autodetect epoll and inotify support
179 by libc header version if not using autoconf.
180 - new symbols: EV_DEFAULT_UC and EV_DEFAULT_UC_.
181 - declare functions defined in ev.h as inline if
182 C99 or gcc are available.
183 - enable inlining with gcc versions 2 and 3.
184 - work around broken poll implementations potentially
185 not clearing revents field in ev_poll (Brandon Black)
186 (no such systems are known at this time).
187 - work around a bug in realloc on openbsd and darwin,
188 also makes the errornous valgrind complaints
189 go away (noted by various people).
190 - fix ev_async_pending, add c++ wrapper for ev_async
191 (based on patch sent by Johannes Deisenhofer.
192 - add sensible set method to ev::embed.
193 - made integer constants type int in ev.h.
194
195 3.2 Wed Apr 2 17:11:19 CEST 2008
196 - fix a 64 bit overflow issue in the select backend,
197 by using fd_mask instead of int for the mask.
198 - rename internal sighandler to avoid clash with very old perls.
199 - entering ev_loop will not clear the ONESHOT or NONBLOCKING
200 flags of any outer loops anymore.
201 - add ev_async_pending.
202
203 3.1 Thu Mar 13 13:45:22 CET 2008
204 - implement ev_async watchers.
205 - only initialise signal pipe on demand.
206 - make use of sig_atomic_t configurable.
207 - improved documentation.
208
209 3.0 Mon Jan 28 13:14:47 CET 2008
210 - API/ABI bump to version 3.0.
211 - ev++.h includes "ev.h" by default now, not <ev.h>.
212 - slightly improved documentation.
213 - speed up signal detection after a fork.
214 - only optionally return trace status changed in ev_child
215 watchers.
216 - experimental (and undocumented) loop wrappers for ev++.h.
217
218 2.01 Tue Dec 25 08:04:41 CET 2007
219 - separate Changes file.
220 - fix ev_path_set => ev_stat_set typo.
221 - remove event_compat.h from the libev tarball.
222 - change how include files are found.
223 - doc updates.
224 - update licenses, explicitly allow for GPL relicensing.
225
226 2.0 Sat Dec 22 17:47:03 CET 2007
227 - new ev_sleep, ev_set_(io|timeout)_collect_interval.
228 - removed epoll from embeddable fd set.
229 - fix embed watchers.
230 - renamed ev_embed.loop to other.
231 - added exported Symbol tables.
232 - undefine member wrapper macros at the end of ev.c.
233 - respect EV_H in ev++.h.
234
235 1.86 Tue Dec 18 02:36:57 CET 2007
236 - fix memleak on loop destroy (not relevant for perl).
237
238 1.85 Fri Dec 14 20:32:40 CET 2007
239 - fix some aliasing issues w.r.t. timers and periodics
240 (not relevant for perl).
241
242 (for historic versions refer to EV/Changes, found in the Perl interface)
243
244 0.1 Wed Oct 31 21:31:48 CET 2007
245 - original version; hacked together in <24h.
246