1 | Revision history for libev, a high-performance and full-featured event loop. |
1 | Revision history for libev, a high-performance and full-featured event loop. |
2 | |
2 | |
3 | TODO: ABI??? API????? Changes??? |
3 | 4.25 Fri Dec 21 07:49:20 CET 2018 |
4 | TODO: win32 write() to socket for signal handling |
4 | - INCOMPATIBLE CHANGE: EV_THROW was renamed to EV_NOEXCEPT |
|
|
5 | (EV_THROW sitll provided) and now uses noexcept on C++11 or newer. |
|
|
6 | - move the darwin select workaround highe rin ev.c, as newer versions of |
|
|
7 | darwin managed to break their broken select even more. |
|
|
8 | - ANDROID => __ANDROID__ (reported by enh@google.com). |
|
|
9 | - disable epoll_create1 on android because it has broken header files |
|
|
10 | and google is unwilling to fix them (reported by enh@google.com). |
|
|
11 | - avoid a minor compilation warning on win32. |
|
|
12 | - c++: remove deprecated dynamic throw() specifications. |
|
|
13 | - c++: improve the (unsupported) bad_loop exception class. |
|
|
14 | - backport perl ev_periodic example to C, untested. |
|
|
15 | |
|
|
16 | 4.24 Wed Dec 28 05:19:55 CET 2016 |
|
|
17 | - bump version to 4.24, as the release tarball inexplicably |
|
|
18 | didn't have the right version in ev.h, even though the cvs-tagged |
|
|
19 | version did have the right one (reported by Ales Teska). |
|
|
20 | |
|
|
21 | 4.23 Wed Nov 16 18:23:41 CET 2016 |
|
|
22 | - move some declarations at the beginning to help certain retarded |
|
|
23 | microsoft compilers, even though their documentation claims |
|
|
24 | otherwise (reported by Ruslan Osmanov). |
|
|
25 | |
|
|
26 | 4.22 Sun Dec 20 22:11:50 CET 2015 |
|
|
27 | - when epoll detects unremovable fds in the fd set, rebuild |
|
|
28 | only the epoll descriptor, not the signal pipe, to avoid |
|
|
29 | SIGPIPE in ev_async_send. This doesn't solve it on fork, |
|
|
30 | so document what needs to be done in ev_loop_fork |
|
|
31 | (analyzed by Benjamin Mahler). |
|
|
32 | - remove superfluous sys/timeb.h include on win32 |
|
|
33 | (analyzed by Jason Madden). |
|
|
34 | - updated libecb. |
|
|
35 | |
|
|
36 | 4.20 Sat Jun 20 13:01:43 CEST 2015 |
|
|
37 | - prefer noexcept over throw () with C++ 11. |
|
|
38 | - update ecb.h due to incompatibilities with c11. |
|
|
39 | - fix a potential aliasing issue when reading and writing |
|
|
40 | watcher callbacks. |
|
|
41 | |
|
|
42 | 4.19 Thu Sep 25 08:18:25 CEST 2014 |
|
|
43 | - ev.h wasn't valid C++ anymore, which tripped compilers other than |
|
|
44 | clang, msvc or gcc (analyzed by Raphael 'kena' Poss). Unfortunately, |
|
|
45 | C++ doesn't support typedefs for function pointers fully, so the affected |
|
|
46 | declarations have to spell out the types each time. |
|
|
47 | - when not using autoconf, tighten the check for clock_gettime and related |
|
|
48 | functionality. |
|
|
49 | |
|
|
50 | 4.18 Fri Sep 5 17:55:26 CEST 2014 |
|
|
51 | - events on files were not always generated properly with the |
|
|
52 | epoll backend (testcase by Assaf Inbal). |
|
|
53 | - mark event pipe fd as cloexec after a fork (analyzed by Sami Farin). |
|
|
54 | - (ecb) support m68k, m88k and sh (patch by Miod Vallat). |
|
|
55 | - use a reasonable fallback for EV_NSIG instead of erroring out |
|
|
56 | when we can't detect the signal set size. |
|
|
57 | - in the absence of autoconf, do not use the clock syscall |
|
|
58 | on glibc >= 2.17 (avoids the syscall AND -lrt on systems |
|
|
59 | doing clock_gettime in userspace). |
|
|
60 | - ensure extern "C" function pointers are used for externally-visible |
|
|
61 | loop callbacks (not watcher callbacks yet). |
|
|
62 | - (ecb) work around memory barriers and volatile apparently both being |
|
|
63 | broken in visual studio 2008 and later (analysed and patch by Nicolas Noble). |
|
|
64 | |
|
|
65 | 4.15 Fri Mar 1 12:04:50 CET 2013 |
|
|
66 | - destroying a non-default loop would stop the global waitpid |
|
|
67 | watcher (Denis Bilenko). |
|
|
68 | - queueing pending watchers of higher priority from a watcher now invokes |
|
|
69 | them in a timely fashion (reported by Denis Bilenko). |
|
|
70 | - add throw() to all libev functions that cannot throw exceptions, for |
|
|
71 | further code size decrease when compiling for C++. |
|
|
72 | - add throw () to callbacks that must not throw exceptions (allocator, |
|
|
73 | syserr, loop acquire/release, periodic reschedule cbs). |
|
|
74 | - fix event_base_loop return code, add event_get_callback, event_base_new, |
|
|
75 | event_base_get_method calls to improve libevent 1.x emulation and add |
|
|
76 | some libevent 2.x functionality (based on a patch by Jeff Davey). |
|
|
77 | - add more memory fences to fix a bug reported by Jeff Davey. Better |
|
|
78 | be overfenced than underprotected. |
|
|
79 | - ev_run now returns a boolean status (true meaning watchers are |
|
|
80 | still active). |
|
|
81 | - ev_once: undef EV_ERROR in ev_kqueue.c, to avoid clashing with |
|
|
82 | libev's EV_ERROR (reported by 191919). |
|
|
83 | - (ecb) add memory fence support for xlC (Darin McBride). |
|
|
84 | - (ecb) add memory fence support for gcc-mips (Anton Kirilov). |
|
|
85 | - (ecb) add memory fence support for gcc-alpha (Christian Weisgerber). |
|
|
86 | - work around some kernels losing file descriptors by leaking |
|
|
87 | the kqueue descriptor in the child. |
|
|
88 | - work around linux inotify not reporting IN_ATTRIB changes for directories |
|
|
89 | in many cases. |
|
|
90 | - include sys/syscall.h instead of plain syscall.h. |
|
|
91 | - check for io watcher loops in ev_verify, check for the most |
|
|
92 | common reported usage bug in ev_io_start. |
|
|
93 | - choose socket vs. WSASocket at compiletime using EV_USE_WSASOCKET. |
|
|
94 | - always use WSASend/WSARecv directly on windows, hoping that this |
|
|
95 | works in all cases (unlike read/write/send/recv...). |
|
|
96 | - try to detect signals around a fork faster (test program by |
|
|
97 | Denis Bilenko). |
|
|
98 | - work around recent glibc versions that leak memory in realloc. |
|
|
99 | - rename ev::embed::set to ev::embed::set_embed to avoid clashing |
|
|
100 | the watcher base set (loop) method. |
|
|
101 | - rewrite the async/signal pipe logic to always keep a valid fd, which |
|
|
102 | simplifies (and hopefully correctifies :) the race checking |
|
|
103 | on fork, at the cost of one extra fd. |
|
|
104 | - add fat, msdos, jffs2, ramfs, ntfs and btrfs to the list of |
|
|
105 | inotify-supporting filesystems. |
|
|
106 | - move orig_CFLAGS assignment to after AC_INIT, as newer autoconf |
|
|
107 | versions ignore it before |
|
|
108 | (https://bugzilla.redhat.com/show_bug.cgi?id=908096). |
|
|
109 | - add some untested android support. |
|
|
110 | - enum expressions must be of type int (reported by Juan Pablo L). |
|
|
111 | |
|
|
112 | 4.11 Sat Feb 4 19:52:39 CET 2012 |
|
|
113 | - INCOMPATIBLE CHANGE: ev_timer_again now clears the pending status, as |
|
|
114 | was documented already, but not implemented in the repeating case. |
|
|
115 | - new compiletime symbols: EV_NO_SMP and EV_NO_THREADS. |
|
|
116 | - fix a race where the workaround against the epoll fork bugs |
|
|
117 | caused signals to not be handled anymore. |
|
|
118 | - correct backend_fudge for most backends, and implement a windows |
|
|
119 | specific workaround to avoid looping because we call both |
|
|
120 | select and Sleep, both with different time resolutions. |
|
|
121 | - document range and guarantees of ev_sleep. |
|
|
122 | - document reasonable ranges for periodics interval and offset. |
|
|
123 | - rename backend_fudge to backend_mintime to avoid future confusion :) |
|
|
124 | - change the default periodic reschedule function to hopefully be more |
|
|
125 | exact and correct even in corner cases or in the far future. |
|
|
126 | - do not rely on -lm anymore: use it when available but use our |
|
|
127 | own floor () if it is missing. This should make it easier to embed, |
|
|
128 | as no external libraries are required. |
|
|
129 | - strategically import macros from libecb and mark rarely-used functions |
|
|
130 | as cache-cold (saving almost 2k code size on typical amd64 setups). |
|
|
131 | - add Symbols.ev and Symbols.event files, that were missing. |
|
|
132 | - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now). |
|
|
133 | - fix #3 "be smart about timeouts" to not "deadlock" when |
|
|
134 | timeout == now, also improve the section overall. |
|
|
135 | - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom. |
|
|
136 | - support new EV_API_STATIC mode to make all libev symbols |
|
|
137 | static. |
|
|
138 | - supply default CFLAGS of -g -O3 with gcc when original CFLAGS |
|
|
139 | were empty. |
|
|
140 | |
|
|
141 | 4.04 Wed Feb 16 09:01:51 CET 2011 |
|
|
142 | - fix two problems in the native win32 backend, where reuse of fd's |
|
|
143 | with different underlying handles caused handles not to be removed |
|
|
144 | or added to the select set (analyzed and tested by Bert Belder). |
|
|
145 | - do no rely on ceil() in ev_e?poll.c. |
|
|
146 | - backport libev to HP-UX versions before 11 v3. |
|
|
147 | - configure did not detect nanosleep and clock_gettime properly when |
|
|
148 | they are available in the libc (as opposed to -lrt). |
|
|
149 | |
|
|
150 | 4.03 Tue Jan 11 14:37:25 CET 2011 |
|
|
151 | - officially support polling files with all backends. |
|
|
152 | - support files, /dev/zero etc. the same way as select in the epoll |
|
|
153 | backend, by generating events on our own. |
|
|
154 | - ports backend: work around solaris bug 6874410 and many related ones |
|
|
155 | (EINTR, maybe more), with no performance loss (note that the solaris |
|
|
156 | bug report is actually wrong, reality is far more bizarre and broken |
|
|
157 | than that). |
|
|
158 | - define EV_READ/EV_WRITE as macros in event.h, as some programs use |
|
|
159 | #ifdef to test for them. |
|
|
160 | - new (experimental) function: ev_feed_signal. |
|
|
161 | - new (to become default) EVFLAG_NOSIGMASK flag. |
|
|
162 | - new EVBACKEND_MASK symbol. |
|
|
163 | - updated COMMON IDIOMS SECTION. |
|
|
164 | |
|
|
165 | 4.01 Fri Nov 5 21:51:29 CET 2010 |
|
|
166 | - automake fucked it up, apparently, --add-missing -f is not quite enough |
|
|
167 | to make it update its files, so 4.00 didn't install ev++.h and |
|
|
168 | event.h on make install. grrr. |
|
|
169 | - ev_loop(count|depth) didn't return anything (Robin Haberkorn). |
|
|
170 | - change EV_UNDEF to 0xffffffff to silence some overzealous compilers. |
|
|
171 | - use "(libev) " prefix for all libev error messages now. |
|
|
172 | |
|
|
173 | 4.00 Mon Oct 25 12:32:12 CEST 2010 |
5 | - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading. |
174 | - "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 |
175 | - ev_embed_stop did not correctly stop the watcher (very good |
7 | testcase by Vladimir Timofeev). |
176 | testcase by Vladimir Timofeev). |
|
|
177 | - ev_run will now always update the current loop time - it erroneously |
|
|
178 | didn't when idle watchers were active, causing timers not to fire. |
|
|
179 | - fix a bug where a timeout of zero caused the timer not to fire |
|
|
180 | in the libevent emulation (testcase by Péter Szabó). |
8 | - applied win32 fixes by Michael Lenaghan (also James Mansion). |
181 | - applied win32 fixes by Michael Lenaghan (also James Mansion). |
9 | - replace EV_MINIMAL by EV_FEATURES. |
182 | - replace EV_MINIMAL by EV_FEATURES. |
|
|
183 | - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it |
|
|
184 | seems the former is *much* faster than the latter. |
|
|
185 | - linux kernel version detection (for inotify bug workarounds) |
|
|
186 | did not work properly. |
|
|
187 | - reduce the number of spurious wake-ups with the ports backend. |
10 | - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu). |
188 | - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu). |
|
|
189 | - do async init within ev_async_start, not ev_async_set, which avoids |
|
|
190 | an API quirk where the set function must be called in the C++ API |
|
|
191 | even when there is nothing to set. |
11 | - add (undocumented) EV_ENABLE when adding events with kqueue, |
192 | - add (undocumented) EV_ENABLE when adding events with kqueue, |
12 | this might help with OS X, which seems to need it despite documenting |
193 | this might help with OS X, which seems to need it despite documenting |
13 | not to need it (helpfully pointed out by Tilghman Lesher). |
194 | not to need it (helpfully pointed out by Tilghman Lesher). |
|
|
195 | - do not use poll by default on freebsd, it's broken (what isn't |
|
|
196 | on freebsd...). |
|
|
197 | - allow to embed epoll on kernels >= 2.6.32. |
14 | - configure now prepends -O3, not appends it, so one can still |
198 | - configure now prepends -O3, not appends it, so one can still |
15 | override it. |
199 | override it. |
|
|
200 | - ev.pod: greatly expanded the portability section, added a porting |
|
|
201 | section, a description of watcher states and made lots of minor fixes. |
16 | - disable poll backend on AIX, the poll header spams the namespace |
202 | - disable poll backend on AIX, the poll header spams the namespace |
17 | and it's not worth working around dead platforms (reported |
203 | and it's not worth working around dead platforms (reported |
18 | and analyzed by Aivars Kalvans). |
204 | and analyzed by Aivars Kalvans). |
19 | - improve header file compatibility of the standalone eventfd code |
205 | - improve header file compatibility of the standalone eventfd code |
20 | in an obscure case. |
206 | in an obscure case. |
… | |
… | |
23 | no sscanf dependency). |
209 | no sscanf dependency). |
24 | - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings. |
210 | - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings. |
25 | - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs. |
211 | - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs. |
26 | - add section on accept() problems to the manpage. |
212 | - add section on accept() problems to the manpage. |
27 | - rename EV_TIMEOUT to EV_TIMER. |
213 | - rename EV_TIMEOUT to EV_TIMER. |
28 | - rename ev_loop_count/depth/verify. |
214 | - rename ev_loop_count/depth/verify/loop/unloop. |
|
|
215 | - remove ev_default_destroy and ev_default_fork. |
29 | - switch to two-digit minor version. |
216 | - switch to two-digit minor version. |
|
|
217 | - work around an apparent gentoo compiler bug. |
|
|
218 | - define _DARWIN_UNLIMITED_SELECT. just so. |
|
|
219 | - use enum instead of #define for most constants. |
|
|
220 | - improve compatibility to older C++ compilers. |
|
|
221 | - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now |
|
|
222 | default arguments when compiled as C++. |
|
|
223 | - enable automake dependency tracking. |
|
|
224 | - ev_loop_new no longer leaks memory when loop creation failed. |
|
|
225 | - new ev_cleanup watcher type. |
30 | |
226 | |
31 | 3.9 Thu Dec 31 07:59:59 CET 2009 |
227 | 3.9 Thu Dec 31 07:59:59 CET 2009 |
32 | - signalfd is no longer used by default and has to be requested |
228 | - signalfd is no longer used by default and has to be requested |
33 | explicitly - this means that easy to catch bugs become hard to |
229 | explicitly - this means that easy to catch bugs become hard to |
34 | catch race conditions, but the users have spoken. |
230 | catch race conditions, but the users have spoken. |
35 | - point out the unspecified signal mask in the documentation, and |
231 | - point out the unspecified signal mask in the documentation, and |
36 | that this is a race condition regardless of EV_SIGNALFD. |
232 | that this is a race condition regardless of EV_SIGNALFD. |
37 | - backport inotify code to C89. |
233 | - backport inotify code to C89. |
38 | - inotify file descriptors could leak into child processes. |
234 | - inotify file descriptors could leak into child processes. |
39 | - ev_stat watchers could keep an errornous extra ref on the loop, |
235 | - ev_stat watchers could keep an erroneous extra ref on the loop, |
40 | preventing exit when unregistering all watchers (testcases |
236 | preventing exit when unregistering all watchers (testcases |
41 | provided by ry@tinyclouds.org). |
237 | provided by ry@tinyclouds.org). |
42 | - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration |
238 | - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration |
43 | symbols to make it easier for apps to do their own fd management. |
239 | symbols to make it easier for apps to do their own fd management. |
44 | - support EV_IDLE_ENABLE being disabled in ev++.h |
240 | - support EV_IDLE_ENABLE being disabled in ev++.h |
… | |
… | |
50 | - changed minimum glibc requirement from glibc 2.9 to 2.7, for |
246 | - changed minimum glibc requirement from glibc 2.9 to 2.7, for |
51 | signalfd. |
247 | signalfd. |
52 | - add missing string.h include (Denis F. Latypoff). |
248 | - add missing string.h include (Denis F. Latypoff). |
53 | - only replace ev_stat.prev when we detect an actual difference, |
249 | - only replace ev_stat.prev when we detect an actual difference, |
54 | so prev is (almost) always different to attr. this might |
250 | so prev is (almost) always different to attr. this might |
55 | have caused the probems with 04_stat.t. |
251 | have caused the problems with 04_stat.t. |
56 | - add ev::timer->remaining () method to C++ API. |
252 | - add ev::timer->remaining () method to C++ API. |
57 | |
253 | |
58 | 3.8 Sun Aug 9 14:30:45 CEST 2009 |
254 | 3.8 Sun Aug 9 14:30:45 CEST 2009 |
59 | - incompatible change: do not necessarily reset signal handler |
255 | - incompatible change: do not necessarily reset signal handler |
60 | to SIG_DFL when a sighandler is stopped. |
256 | to SIG_DFL when a sighandler is stopped. |
61 | - ev_default_destroy did not properly free or zero some members, |
257 | - ev_default_destroy did not properly free or zero some members, |
62 | potentially causing crashes and memory corruption on repated |
258 | potentially causing crashes and memory corruption on repeated |
63 | ev_default_destroy/ev_default_loop calls. |
259 | ev_default_destroy/ev_default_loop calls. |
64 | - take advantage of signalfd on GNU/Linux systems. |
260 | - take advantage of signalfd on GNU/Linux systems. |
65 | - document that the signal mask might be in an unspecified |
261 | - document that the signal mask might be in an unspecified |
66 | state when using libev's signal handling. |
262 | state when using libev's signal handling. |
67 | - take advantage of some GNU/Linux calls to set cloexec/nonblock |
263 | - take advantage of some GNU/Linux calls to set cloexec/nonblock |
… | |
… | |
102 | - correctly zero the idx values for arrays, so destroying and |
298 | - correctly zero the idx values for arrays, so destroying and |
103 | reinitialising the default loop actually works (patch by |
299 | reinitialising the default loop actually works (patch by |
104 | Malek Hadj-Ali). |
300 | Malek Hadj-Ali). |
105 | - implement ev_suspend and ev_resume. |
301 | - implement ev_suspend and ev_resume. |
106 | - new EV_CUSTOM revents flag for use by applications. |
302 | - new EV_CUSTOM revents flag for use by applications. |
107 | - add documentation section about priorites. |
303 | - add documentation section about priorities. |
108 | - add a glossary to the dcoumentation. |
304 | - add a glossary to the documentation. |
109 | - extend the ev_fork description slightly. |
305 | - extend the ev_fork description slightly. |
110 | - optimize a jump out of call_pending. |
306 | - optimize a jump out of call_pending. |
111 | |
307 | |
112 | 3.53 Sun Feb 15 02:38:20 CET 2009 |
308 | 3.53 Sun Feb 15 02:38:20 CET 2009 |
113 | - fix a bug in event pipe creation on win32 that would cause a |
309 | - fix a bug in event pipe creation on win32 that would cause a |
… | |
… | |
189 | do similar changes for event.c. |
385 | do similar changes for event.c. |
190 | |
386 | |
191 | 3.45 Tue Oct 21 21:59:26 CEST 2008 |
387 | 3.45 Tue Oct 21 21:59:26 CEST 2008 |
192 | - disable inotify usage on linux <2.6.25, as it is broken |
388 | - disable inotify usage on linux <2.6.25, as it is broken |
193 | (reported by Yoann Vandoorselaere). |
389 | (reported by Yoann Vandoorselaere). |
194 | - ev_stat errornously would try to add inotify watchers |
390 | - ev_stat erroneously would try to add inotify watchers |
195 | even when inotify wasn't available (this should only |
391 | even when inotify wasn't available (this should only |
196 | have a performance impact). |
392 | have a performance impact). |
197 | - ev_once now passes both timeout and io to the callback if both |
393 | - ev_once now passes both timeout and io to the callback if both |
198 | occur concurrently, instead of giving timeouts precedence. |
394 | occur concurrently, instead of giving timeouts precedence. |
199 | - disable EV_USE_INOTIFY when sys/inotify.h is too old. |
395 | - disable EV_USE_INOTIFY when sys/inotify.h is too old. |
… | |
… | |
267 | - enable inlining with gcc versions 2 and 3. |
463 | - enable inlining with gcc versions 2 and 3. |
268 | - work around broken poll implementations potentially |
464 | - work around broken poll implementations potentially |
269 | not clearing revents field in ev_poll (Brandon Black) |
465 | not clearing revents field in ev_poll (Brandon Black) |
270 | (no such systems are known at this time). |
466 | (no such systems are known at this time). |
271 | - work around a bug in realloc on openbsd and darwin, |
467 | - work around a bug in realloc on openbsd and darwin, |
272 | also makes the errornous valgrind complaints |
468 | also makes the erroneous valgrind complaints |
273 | go away (noted by various people). |
469 | go away (noted by various people). |
274 | - fix ev_async_pending, add c++ wrapper for ev_async |
470 | - fix ev_async_pending, add c++ wrapper for ev_async |
275 | (based on patch sent by Johannes Deisenhofer. |
471 | (based on patch sent by Johannes Deisenhofer). |
276 | - add sensible set method to ev::embed. |
472 | - add sensible set method to ev::embed. |
277 | - made integer constants type int in ev.h. |
473 | - made integer constants type int in ev.h. |
278 | |
474 | |
279 | 3.2 Wed Apr 2 17:11:19 CEST 2008 |
475 | 3.2 Wed Apr 2 17:11:19 CEST 2008 |
280 | - fix a 64 bit overflow issue in the select backend, |
476 | - fix a 64 bit overflow issue in the select backend, |