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

Comparing libev/Changes (file contents):
Revision 1.102 by root, Sat Mar 14 04:45:39 2009 UTC vs.
Revision 1.299 by root, Tue Sep 9 21:51:35 2014 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
3TODO: ev_loop_wakeup
4TODO: EV_STANDALONE == NO_HASSEL (do not use clock_gettime in ev_standalone)
5TODO: faq, process a thing in each iteration
6TODO: dbeugging tips, ev_verify, ev_init twice
7TODO: ev_break for immediate exit (EVBREAK_NOW?)
8TODO: ev_feed_child_event
9TODO: document the special problem of signals around fork.
10TODO: store pid for each signal
11TODO: document file descriptor usage per loop
12TODO: store loop pid_t and compare isndie signal handler,store 1 for same, 2 for differign pid, clean up in loop_fork
13TODO: embed watchers need updating when fd changes
14TODO: document portability requirements for atomic pointer access
15TODO: possible cb aliasing?
16TODO: document requirements for function pointers and calling conventions.
17
18 - ev.h wasn't valid C++ anymore, which tripped compilers other than
19 clang, msvc or gcc (analyzed by Raphael 'kena' Poss). Unfortunately,
20 C++ doesn't support typedefs for function pointers fully, so the affected
21 declarations have to spell out the types each time.
22 - when not using autoconf, tighten the check for clock_gettime and related
23 functionality.
24
254.18 Fri Sep 5 17:55:26 CEST 2014
26 - events on files were not always generated properly with the
27 epoll backend (testcase by Assaf Inbal).
28 - mark event pipe fd as cloexec after a fork (analyzed by Sami Farin).
29 - (ecb) support m68k, m88k and sh (patch by Miod Vallat).
30 - use a reasonable fallback for EV_NSIG instead of erroring out
31 when we can't detect the signal set size.
32 - in the absence of autoconf, do not use the clock syscall
33 on glibc >= 2.17 (avoids the syscall AND -lrt on systems
34 doing clock_gettime in userspace).
35 - ensure extern "C" function pointers are used for externally-visible
36 loop callbacks (not watcher callbacks yet).
37 - (ecb) work around memory barriers and volatile apparently both being
38 broken in visual studio 2008 and later (analysed and patch by Nicolas Noble).
39
404.15 Fri Mar 1 12:04:50 CET 2013
41 - destroying a non-default loop would stop the global waitpid
42 watcher (Denis Bilenko).
43 - queueing pending watchers of higher priority from a watcher now invokes
44 them in a timely fashion (reported by Denis Bilenko).
45 - add throw() to all libev functions that cannot throw exceptions, for
46 further code size decrease when compiling for C++.
47 - add throw () to callbacks that must not throw exceptions (allocator,
48 syserr, loop acquire/release, periodic reschedule cbs).
49 - fix event_base_loop return code, add event_get_callback, event_base_new,
50 event_base_get_method calls to improve libevent 1.x emulation and add
51 some libevent 2.x functionality (based on a patch by Jeff Davey).
52 - add more memory fences to fix a bug reported by Jeff Davey. Better
53 be overfenced than underprotected.
54 - ev_run now returns a boolean status (true meaning watchers are
55 still active).
56 - ev_once: undef EV_ERROR in ev_kqueue.c, to avoid clashing with
57 libev's EV_ERROR (reported by 191919).
58 - (ecb) add memory fence support for xlC (Darin McBride).
59 - (ecb) add memory fence support for gcc-mips (Anton Kirilov).
60 - (ecb) add memory fence support for gcc-alpha (Christian Weisgerber).
61 - work around some kernels losing file descriptors by leaking
62 the kqueue descriptor in the child.
63 - work around linux inotify not reporting IN_ATTRIB changes for directories
64 in many cases.
65 - include sys/syscall.h instead of plain syscall.h.
66 - check for io watcher loops in ev_verify, check for the most
67 common reported usage bug in ev_io_start.
68 - choose socket vs. WSASocket at compiletime using EV_USE_WSASOCKET.
69 - always use WSASend/WSARecv directly on windows, hoping that this
70 works in all cases (unlike read/write/send/recv...).
71 - try to detect signals around a fork faster (test program by
72 Denis Bilenko).
73 - work around recent glibc versions that leak memory in realloc.
74 - rename ev::embed::set to ev::embed::set_embed to avoid clashing
75 the watcher base set (loop) method.
76 - rewrite the async/signal pipe logic to always keep a valid fd, which
77 simplifies (and hopefully correctifies :) the race checking
78 on fork, at the cost of one extra fd.
79 - add fat, msdos, jffs2, ramfs, ntfs and btrfs to the list of
80 inotify-supporting filesystems.
81 - move orig_CFLAGS assignment to after AC_INIT, as newer autoconf
82 versions ignore it before
83 (https://bugzilla.redhat.com/show_bug.cgi?id=908096).
84 - add some untested android support.
85 - enum expressions must be of type int (reported by Juan Pablo L).
86
874.11 Sat Feb 4 19:52:39 CET 2012
88 - INCOMPATIBLE CHANGE: ev_timer_again now clears the pending status, as
89 was documented already, but not implemented in the repeating case.
90 - new compiletime symbols: EV_NO_SMP and EV_NO_THREADS.
91 - fix a race where the workaround against the epoll fork bugs
92 caused signals to not be handled anymore.
93 - correct backend_fudge for most backends, and implement a windows
94 specific workaround to avoid looping because we call both
95 select and Sleep, both with different time resolutions.
96 - document range and guarantees of ev_sleep.
97 - document reasonable ranges for periodics interval and offset.
98 - rename backend_fudge to backend_mintime to avoid future confusion :)
99 - change the default periodic reschedule function to hopefully be more
100 exact and correct even in corner cases or in the far future.
101 - do not rely on -lm anymore: use it when available but use our
102 own floor () if it is missing. This should make it easier to embed,
103 as no external libraries are required.
104 - strategically import macros from libecb and mark rarely-used functions
105 as cache-cold (saving almost 2k code size on typical amd64 setups).
106 - add Symbols.ev and Symbols.event files, that were missing.
107 - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now).
108 - fix #3 "be smart about timeouts" to not "deadlock" when
109 timeout == now, also improve the section overall.
110 - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom.
111 - support new EV_API_STATIC mode to make all libev symbols
112 static.
113 - supply default CFLAGS of -g -O3 with gcc when original CFLAGS
114 were empty.
115
1164.04 Wed Feb 16 09:01:51 CET 2011
117 - fix two problems in the native win32 backend, where reuse of fd's
118 with different underlying handles caused handles not to be removed
119 or added to the select set (analyzed and tested by Bert Belder).
120 - do no rely on ceil() in ev_e?poll.c.
121 - backport libev to HP-UX versions before 11 v3.
122 - configure did not detect nanosleep and clock_gettime properly when
123 they are available in the libc (as opposed to -lrt).
124
1254.03 Tue Jan 11 14:37:25 CET 2011
126 - officially support polling files with all backends.
127 - support files, /dev/zero etc. the same way as select in the epoll
128 backend, by generating events on our own.
129 - ports backend: work around solaris bug 6874410 and many related ones
130 (EINTR, maybe more), with no performance loss (note that the solaris
131 bug report is actually wrong, reality is far more bizarre and broken
132 than that).
133 - define EV_READ/EV_WRITE as macros in event.h, as some programs use
134 #ifdef to test for them.
135 - new (experimental) function: ev_feed_signal.
136 - new (to become default) EVFLAG_NOSIGMASK flag.
137 - new EVBACKEND_MASK symbol.
138 - updated COMMON IDIOMS SECTION.
139
1404.01 Fri Nov 5 21:51:29 CET 2010
141 - automake fucked it up, apparently, --add-missing -f is not quite enough
142 to make it update its files, so 4.00 didn't install ev++.h and
143 event.h on make install. grrr.
144 - ev_loop(count|depth) didn't return anything (Robin Haberkorn).
145 - change EV_UNDEF to 0xffffffff to silence some overzealous compilers.
146 - use "(libev) " prefix for all libev error messages now.
147
1484.00 Mon Oct 25 12:32:12 CEST 2010
149 - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading.
150 - ev_embed_stop did not correctly stop the watcher (very good
151 testcase by Vladimir Timofeev).
152 - ev_run will now always update the current loop time - it erroneously
153 didn't when idle watchers were active, causing timers not to fire.
154 - fix a bug where a timeout of zero caused the timer not to fire
155 in the libevent emulation (testcase by Péter Szabó).
156 - applied win32 fixes by Michael Lenaghan (also James Mansion).
157 - replace EV_MINIMAL by EV_FEATURES.
158 - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it
159 seems the former is *much* faster than the latter.
160 - linux kernel version detection (for inotify bug workarounds)
161 did not work properly.
162 - reduce the number of spurious wake-ups with the ports backend.
163 - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu).
164 - do async init within ev_async_start, not ev_async_set, which avoids
165 an API quirk where the set function must be called in the C++ API
166 even when there is nothing to set.
167 - add (undocumented) EV_ENABLE when adding events with kqueue,
168 this might help with OS X, which seems to need it despite documenting
169 not to need it (helpfully pointed out by Tilghman Lesher).
170 - do not use poll by default on freebsd, it's broken (what isn't
171 on freebsd...).
172 - allow to embed epoll on kernels >= 2.6.32.
173 - configure now prepends -O3, not appends it, so one can still
174 override it.
175 - ev.pod: greatly expanded the portability section, added a porting
176 section, a description of watcher states and made lots of minor fixes.
177 - disable poll backend on AIX, the poll header spams the namespace
178 and it's not worth working around dead platforms (reported
179 and analyzed by Aivars Kalvans).
180 - improve header file compatibility of the standalone eventfd code
181 in an obscure case.
182 - implement EV_AVOID_STDIO option.
183 - do not use sscanf to parse linux version number (smaller, faster,
184 no sscanf dependency).
185 - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings.
186 - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs.
187 - add section on accept() problems to the manpage.
188 - rename EV_TIMEOUT to EV_TIMER.
189 - rename ev_loop_count/depth/verify/loop/unloop.
190 - remove ev_default_destroy and ev_default_fork.
191 - switch to two-digit minor version.
192 - work around an apparent gentoo compiler bug.
193 - define _DARWIN_UNLIMITED_SELECT. just so.
194 - use enum instead of #define for most constants.
195 - improve compatibility to older C++ compilers.
196 - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now
197 default arguments when compiled as C++.
198 - enable automake dependency tracking.
199 - ev_loop_new no longer leaks memory when loop creation failed.
200 - new ev_cleanup watcher type.
201
2023.9 Thu Dec 31 07:59:59 CET 2009
203 - signalfd is no longer used by default and has to be requested
204 explicitly - this means that easy to catch bugs become hard to
205 catch race conditions, but the users have spoken.
206 - point out the unspecified signal mask in the documentation, and
207 that this is a race condition regardless of EV_SIGNALFD.
208 - backport inotify code to C89.
209 - inotify file descriptors could leak into child processes.
210 - ev_stat watchers could keep an erroneous extra ref on the loop,
211 preventing exit when unregistering all watchers (testcases
212 provided by ry@tinyclouds.org).
213 - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
214 symbols to make it easier for apps to do their own fd management.
215 - support EV_IDLE_ENABLE being disabled in ev++.h
216 (patch by Didier Spezia).
217 - take advantage of inotify_init1, if available, to set cloexec/nonblock
218 on fd creation, to avoid races.
219 - the signal handling pipe wasn't always initialised under windows
220 (analysed by lekma).
221 - changed minimum glibc requirement from glibc 2.9 to 2.7, for
222 signalfd.
223 - add missing string.h include (Denis F. Latypoff).
224 - only replace ev_stat.prev when we detect an actual difference,
225 so prev is (almost) always different to attr. this might
226 have caused the problems with 04_stat.t.
227 - add ev::timer->remaining () method to C++ API.
228
2293.8 Sun Aug 9 14:30:45 CEST 2009
230 - incompatible change: do not necessarily reset signal handler
231 to SIG_DFL when a sighandler is stopped.
232 - ev_default_destroy did not properly free or zero some members,
233 potentially causing crashes and memory corruption on repeated
234 ev_default_destroy/ev_default_loop calls.
235 - take advantage of signalfd on GNU/Linux systems.
236 - document that the signal mask might be in an unspecified
237 state when using libev's signal handling.
238 - take advantage of some GNU/Linux calls to set cloexec/nonblock
239 on fd creation, to avoid race conditions.
240
2413.7 Fri Jul 17 16:36:32 CEST 2009
242 - ev_unloop and ev_loop wrongly used a global variable to exit loops,
243 instead of using a per-loop variable (bug caught by accident...).
244 - the ev_set_io_collect_interval interpretation has changed.
245 - add new functionality: ev_set_userdata, ev_userdata,
246 ev_set_invoke_pending_cb, ev_set_loop_release_cb,
247 ev_invoke_pending, ev_pending_count, together with a long example
248 about thread locking.
249 - add ev_timer_remaining (as requested by Denis F. Latypoff).
250 - add ev_loop_depth.
251 - calling ev_unloop in fork/prepare watchers will no longer poll
252 for new events.
253 - Denis F. Latypoff corrected many typos in example code snippets.
254 - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double-
255 check that the syscall number is available before trying to
256 use it (reported by ry@tinyclouds).
257 - use GetSystemTimeAsFileTime instead of _timeb on windows, for
258 slightly higher accuracy.
259 - properly declare ev_loop_verify and ev_now_update even when
260 !EV_MULTIPLICITY.
261 - do not compile in any priority code when EV_MAXPRI == EV_MINPRI.
262 - support EV_MINIMAL==2 for a reduced API.
263 - actually 0-initialise struct sigaction when installing signals.
264 - add section on hibernate and stopped processes to ev_timer docs.
265
2663.6 Tue Apr 28 02:49:30 CEST 2009
267 - multiple timers becoming ready within an event loop iteration
268 will be invoked in the "correct" order now.
269 - do not leave the event loop early just because we have no active
270 watchers, fixing a problem when embedding a kqueue loop
271 that has active kernel events but no registered watchers
272 (reported by blacksand blacksand).
3 - correctly zero the idx values for arrays, so destroying and 273 - correctly zero the idx values for arrays, so destroying and
4 reinitialising the default loop actually works (patch by 274 reinitialising the default loop actually works (patch by
5 Malek Hadj-Ali). 275 Malek Hadj-Ali).
276 - implement ev_suspend and ev_resume.
277 - new EV_CUSTOM revents flag for use by applications.
278 - add documentation section about priorities.
279 - add a glossary to the documentation.
280 - extend the ev_fork description slightly.
281 - optimize a jump out of call_pending.
6 282
73.53 Sun Feb 15 02:38:20 CET 2009 2833.53 Sun Feb 15 02:38:20 CET 2009
8 - fix a bug in event pipe creation on win32 that would cause a 284 - fix a bug in event pipe creation on win32 that would cause a
9 failed assertion on event loop creation (patch by Malek Hadj-Ali). 285 failed assertion on event loop creation (patch by Malek Hadj-Ali).
10 - probe for CLOCK_REALTIME support at runtime as well and fall 286 - probe for CLOCK_REALTIME support at runtime as well and fall
84 do similar changes for event.c. 360 do similar changes for event.c.
85 361
863.45 Tue Oct 21 21:59:26 CEST 2008 3623.45 Tue Oct 21 21:59:26 CEST 2008
87 - disable inotify usage on linux <2.6.25, as it is broken 363 - disable inotify usage on linux <2.6.25, as it is broken
88 (reported by Yoann Vandoorselaere). 364 (reported by Yoann Vandoorselaere).
89 - ev_stat errornously would try to add inotify watchers 365 - ev_stat erroneously would try to add inotify watchers
90 even when inotify wasn't available (this should only 366 even when inotify wasn't available (this should only
91 have a performance impact). 367 have a performance impact).
92 - ev_once now passes both timeout and io to the callback if both 368 - ev_once now passes both timeout and io to the callback if both
93 occur concurrently, instead of giving timeouts precedence. 369 occur concurrently, instead of giving timeouts precedence.
94 - disable EV_USE_INOTIFY when sys/inotify.h is too old. 370 - disable EV_USE_INOTIFY when sys/inotify.h is too old.
162 - enable inlining with gcc versions 2 and 3. 438 - enable inlining with gcc versions 2 and 3.
163 - work around broken poll implementations potentially 439 - work around broken poll implementations potentially
164 not clearing revents field in ev_poll (Brandon Black) 440 not clearing revents field in ev_poll (Brandon Black)
165 (no such systems are known at this time). 441 (no such systems are known at this time).
166 - work around a bug in realloc on openbsd and darwin, 442 - work around a bug in realloc on openbsd and darwin,
167 also makes the errornous valgrind complaints 443 also makes the erroneous valgrind complaints
168 go away (noted by various people). 444 go away (noted by various people).
169 - fix ev_async_pending, add c++ wrapper for ev_async 445 - fix ev_async_pending, add c++ wrapper for ev_async
170 (based on patch sent by Johannes Deisenhofer. 446 (based on patch sent by Johannes Deisenhofer).
171 - add sensible set method to ev::embed. 447 - add sensible set method to ev::embed.
172 - made integer constants type int in ev.h. 448 - made integer constants type int in ev.h.
173 449
1743.2 Wed Apr 2 17:11:19 CEST 2008 4503.2 Wed Apr 2 17:11:19 CEST 2008
175 - fix a 64 bit overflow issue in the select backend, 451 - fix a 64 bit overflow issue in the select backend,

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines