ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/Changes
Revision: 1.113
Committed: Tue Apr 28 00:50:19 2009 UTC (15 years, 2 months ago) by root
Branch: MAIN
CVS Tags: rel-3_6
Changes since 1.112: +1 -3 lines
Log Message:
3.6

File Contents

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