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