ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/Changes
(Generate patch)

Comparing libev/Changes (file contents):
Revision 1.84 by root, Mon Nov 17 03:37:08 2008 UTC vs.
Revision 1.172 by sf-exg, Mon Mar 22 09:57:00 2010 UTC

1Revision history for libev, a high-performance and full-featured event loop. 1Revision history for libev, a high-performance and full-featured event loop.
2 2
3WISH? monotonic clocks times/GetTickCount for coarse corrections? 3TODO: ABI??? API????? Changes???
4TODO: win32 write() to socket for signal handling
5 - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading.
6 - ev_embed_stop did not correctly stop the watcher (very good
7 testcase by Vladimir Timofeev).
8 - applied win32 fixes by Michael Lenaghan (also James Mansion).
9 - replace EV_MINIMAL by EV_FEATURES.
10 - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu).
11 - configure now prepends -O3, not appends it, so one can still
12 override it.
13 - disable poll backend on AIX, the poll header spams the namespace
14 and it's not worth working around dead platforms (reported
15 and analyzed by Aivars Kalvans).
16 - improve header file compatibility of the standalone eventfd code
17 in an obscure case.
18 - implement EV_AVOID_STDIO option.
19 - do not use sscanf to parse linux version number (smaller, faster,
20 no sscanf dependency).
21 - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings.
22 - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs.
23 - add section on accept() problems to the manpage.
24 - rename EV_TIMEOUT to EV_TIMER.
25 - remove loop_ name component from ev_loop_count/depth/verify.
26 - switch to two-digit minor version.
4 27
283.9 Thu Dec 31 07:59:59 CET 2009
29 - signalfd is no longer used by default and has to be requested
30 explicitly - this means that easy to catch bugs become hard to
31 catch race conditions, but the users have spoken.
32 - point out the unspecified signal mask in the documentation, and
33 that this is a race condition regardless of EV_SIGNALFD.
34 - backport inotify code to C89.
35 - inotify file descriptors could leak into child processes.
36 - ev_stat watchers could keep an errornous extra ref on the loop,
37 preventing exit when unregistering all watchers (testcases
38 provided by ry@tinyclouds.org).
39 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
40 symbols to make it easier for apps to do their own fd management.
41 - support EV_IDLE_ENABLE being disabled in ev++.h
42 (patch by Didier Spezia).
43 - take advantage of inotify_init1, if available, to set cloexec/nonblock
44 on fd creation, to avoid races.
45 - the signal handling pipe wasn't always initialised under windows
46 (analysed by lekma).
47 - changed minimum glibc requirement from glibc 2.9 to 2.7, for
48 signalfd.
49 - add missing string.h include (Denis F. Latypoff).
50 - only replace ev_stat.prev when we detect an actual difference,
51 so prev is (almost) always different to attr. this might
52 have caused the probems with 04_stat.t.
53 - add ev::timer->remaining () method to C++ API.
54
553.8 Sun Aug 9 14:30:45 CEST 2009
56 - incompatible change: do not necessarily reset signal handler
57 to SIG_DFL when a sighandler is stopped.
58 - ev_default_destroy did not properly free or zero some members,
59 potentially causing crashes and memory corruption on repated
60 ev_default_destroy/ev_default_loop calls.
61 - take advantage of signalfd on GNU/Linux systems.
62 - document that the signal mask might be in an unspecified
63 state when using libev's signal handling.
64 - take advantage of some GNU/Linux calls to set cloexec/nonblock
65 on fd creation, to avoid race conditions.
66
673.7 Fri Jul 17 16:36:32 CEST 2009
68 - ev_unloop and ev_loop wrongly used a global variable to exit loops,
69 instead of using a per-loop variable (bug caught by accident...).
70 - the ev_set_io_collect_interval interpretation has changed.
71 - add new functionality: ev_set_userdata, ev_userdata,
72 ev_set_invoke_pending_cb, ev_set_loop_release_cb,
73 ev_invoke_pending, ev_pending_count, together with a long example
74 about thread locking.
75 - add ev_timer_remaining (as requested by Denis F. Latypoff).
76 - add ev_loop_depth.
77 - calling ev_unloop in fork/prepare watchers will no longer poll
78 for new events.
79 - Denis F. Latypoff corrected many typos in example code snippets.
80 - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double-
81 check that the syscall number is available before trying to
82 use it (reported by ry@tinyclouds).
83 - use GetSystemTimeAsFileTime instead of _timeb on windows, for
84 slightly higher accuracy.
85 - properly declare ev_loop_verify and ev_now_update even when
86 !EV_MULTIPLICITY.
87 - do not compile in any priority code when EV_MAXPRI == EV_MINPRI.
88 - support EV_MINIMAL==2 for a reduced API.
89 - actually 0-initialise struct sigaction when installing signals.
90 - add section on hibernate and stopped processes to ev_timer docs.
91
923.6 Tue Apr 28 02:49:30 CEST 2009
93 - multiple timers becoming ready within an event loop iteration
94 will be invoked in the "correct" order now.
95 - do not leave the event loop early just because we have no active
96 watchers, fixing a problem when embedding a kqueue loop
97 that has active kernel events but no registered watchers
98 (reported by blacksand blacksand).
99 - correctly zero the idx values for arrays, so destroying and
100 reinitialising the default loop actually works (patch by
101 Malek Hadj-Ali).
102 - implement ev_suspend and ev_resume.
103 - new EV_CUSTOM revents flag for use by applications.
104 - add documentation section about priorites.
105 - add a glossary to the dcoumentation.
106 - extend the ev_fork description slightly.
107 - optimize a jump out of call_pending.
108
1093.53 Sun Feb 15 02:38:20 CET 2009
110 - fix a bug in event pipe creation on win32 that would cause a
111 failed assertion on event loop creation (patch by Malek Hadj-Ali).
112 - probe for CLOCK_REALTIME support at runtime as well and fall
113 back to gettimeofday if there is an error, to support older
114 operating systems with newer header files/libraries.
115 - prefer gettimeofday over clock_gettime with USE_CLOCK_SYSCALL
116 (default most everywhere), otherwise not.
117
1183.52 Wed Jan 7 21:43:02 CET 2009
119 - fix compilation of select backend in fd_set mode when NFDBITS is
120 missing (to get it to compile on QNX, reported by Rodrigo Campos).
121 - better select-nfds handling when select backend is in fd_set mode.
122 - diagnose fd_set overruns when select backend is in fd_set mode.
123 - due to a thinko, instead of disabling everything but
124 select on the borked OS X platform, everything but select was
125 allowed (reported by Emanuele Giaquinta).
126 - actually verify that local and remote port are matching in
127 libev's socketpair emulation, which makes denial-of-service
128 attacks harder (but not impossible - it's windows). Make sure
129 it even works under vista, which thinks that getpeer/sockname
130 should return fantasy port numbers.
131 - include "libev" in all assertion messages for potentially
132 clearer diagnostics.
133 - event_get_version (libevent compatibility) returned
134 a useless string instead of the expected version string
135 (patch by W.C.A. Wijngaards).
136
1373.51 Wed Dec 24 23:00:11 CET 2008
138 - fix a bug where an inotify watcher was added twice, causing
139 freezes on hash collisions (reported and analysed by Graham Leggett).
140 - new config symbol, EV_USE_CLOCK_SYSCALL, to make libev use
141 a direct syscall - slower, but no dependency on librt et al.
142 - assume negative return values != -1 signals success of port_getn
143 (http://cvs.epicsol.org/cgi/viewcvs.cgi/epic5/source/newio.c?rev=1.52)
144 (no known failure reports, but it doesn't hurt).
145 - fork detection in ev_embed now stops and restarts the watcher
146 automatically.
147 - EXPERIMENTAL: default the method to operator () in ev++.h,
148 to make it nicer to use functors (requested by Benedek László).
149 - fixed const object callbacks in ev++.h.
150 - replaced loop_ref argument of watcher.set (loop) by a direct
151 ev_loop * in ev++.h, to avoid clashes with functor patch.
152 - do not try to watch the empty string via inotify.
153 - inotify watchers could be leaked under certain circumstances.
154 - OS X 10.5 is actually even more broken than earlier versions,
155 so fall back to select on that piece of garbage.
156 - fixed some weirdness in the ev_embed documentation.
157
1583.49 Wed Nov 19 11:26:53 CET 2008
5 - ev_stat watchers will now use inotify as a mere hint on 159 - ev_stat watchers will now use inotify as a mere hint on
6 kernels <2.6.25, or if the filesystem is not in the 160 kernels <2.6.25, or if the filesystem is not in the
7 "known to be good" list. 161 "known to be good" list.
8 - better mingw32 compatibility (it's not as borked as native win32) 162 - better mingw32 compatibility (it's not as borked as native win32)
9 (analysed by Roger Pack). 163 (analysed by Roger Pack).
164 - include stdio.h in the example program, as too many people are
165 confused by the weird C language otherwise. I guess the next thing
166 I get told is that the "..." ellipses in the examples don't compile
167 with their C compiler.
10 168
113.48 Thu Oct 30 09:02:37 CET 2008 1693.48 Thu Oct 30 09:02:37 CET 2008
12 - further optimise away the EPOLL_CTL_ADD/MOD combo in the epoll 170 - further optimise away the EPOLL_CTL_ADD/MOD combo in the epoll
13 backend by assuming the kernel event mask hasn't changed if 171 backend by assuming the kernel event mask hasn't changed if
14 ADD fails with EEXIST. 172 ADD fails with EEXIST.

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines