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: ev_loop_wakeup |
|
|
4 | TODO: EV_STANDALONE == NO_HASSEL (do not use clock_gettime in ev_standalone) |
|
|
5 | TODO: faq, process a thing in each iteration |
|
|
6 | TODO: dbeugging tips, ev_verify, ev_init twice |
|
|
7 | TODO: ev_break for immediate exit (EVBREAK_NOW?) |
|
|
8 | TODO: ev_feed_child_event |
|
|
9 | TODO: document the special problem of signals around fork. |
|
|
10 | TODO: store pid for each signal |
|
|
11 | TODO: document file descriptor usage per loop |
|
|
12 | TODO: store loop pid_t and compare isndie signal handler,store 1 for same, 2 for differign pid, clean up in loop_fork |
|
|
13 | TODO: embed watchers need updating when fd changes |
|
|
14 | TODO: document portability requirements for atomic pointer access |
|
|
15 | TODO: document requirements for function pointers and calling conventions. |
|
|
16 | |
|
|
17 | 4.22 |
|
|
18 | - when epoll detects unremovable fds in the fd set, rebuild |
|
|
19 | only the epoll descriptor, not the signal pipe, to avoid |
|
|
20 | SIGPIPE in ev_async_send. This doesn't solve it on fork, |
|
|
21 | so document what needs to be done in ev_loop_fork |
|
|
22 | (analyzed by Benjamin Mahler). |
|
|
23 | - remove superfluous sys/timeb.h include on win32 |
|
|
24 | (analyzed by Jason Madden). |
|
|
25 | - updated libecb. |
|
|
26 | |
|
|
27 | 4.20 Sat Jun 20 13:01:43 CEST 2015 |
|
|
28 | - prefer noexcept over throw () with C++ 11. |
|
|
29 | - update ecb.h due to incompatibilities with c11. |
|
|
30 | - fix a potential aliasing issue when reading and writing |
|
|
31 | watcher callbacks. |
|
|
32 | |
|
|
33 | 4.19 Thu Sep 25 08:18:25 CEST 2014 |
|
|
34 | - ev.h wasn't valid C++ anymore, which tripped compilers other than |
|
|
35 | clang, msvc or gcc (analyzed by Raphael 'kena' Poss). Unfortunately, |
|
|
36 | C++ doesn't support typedefs for function pointers fully, so the affected |
|
|
37 | declarations have to spell out the types each time. |
|
|
38 | - when not using autoconf, tighten the check for clock_gettime and related |
|
|
39 | functionality. |
|
|
40 | |
|
|
41 | 4.18 Fri Sep 5 17:55:26 CEST 2014 |
|
|
42 | - events on files were not always generated properly with the |
|
|
43 | epoll backend (testcase by Assaf Inbal). |
|
|
44 | - mark event pipe fd as cloexec after a fork (analyzed by Sami Farin). |
|
|
45 | - (ecb) support m68k, m88k and sh (patch by Miod Vallat). |
|
|
46 | - use a reasonable fallback for EV_NSIG instead of erroring out |
|
|
47 | when we can't detect the signal set size. |
|
|
48 | - in the absence of autoconf, do not use the clock syscall |
|
|
49 | on glibc >= 2.17 (avoids the syscall AND -lrt on systems |
|
|
50 | doing clock_gettime in userspace). |
|
|
51 | - ensure extern "C" function pointers are used for externally-visible |
|
|
52 | loop callbacks (not watcher callbacks yet). |
|
|
53 | - (ecb) work around memory barriers and volatile apparently both being |
|
|
54 | broken in visual studio 2008 and later (analysed and patch by Nicolas Noble). |
|
|
55 | |
|
|
56 | 4.15 Fri Mar 1 12:04:50 CET 2013 |
|
|
57 | - destroying a non-default loop would stop the global waitpid |
|
|
58 | watcher (Denis Bilenko). |
|
|
59 | - queueing pending watchers of higher priority from a watcher now invokes |
|
|
60 | them in a timely fashion (reported by Denis Bilenko). |
|
|
61 | - add throw() to all libev functions that cannot throw exceptions, for |
|
|
62 | further code size decrease when compiling for C++. |
|
|
63 | - add throw () to callbacks that must not throw exceptions (allocator, |
|
|
64 | syserr, loop acquire/release, periodic reschedule cbs). |
|
|
65 | - fix event_base_loop return code, add event_get_callback, event_base_new, |
|
|
66 | event_base_get_method calls to improve libevent 1.x emulation and add |
|
|
67 | some libevent 2.x functionality (based on a patch by Jeff Davey). |
|
|
68 | - add more memory fences to fix a bug reported by Jeff Davey. Better |
|
|
69 | be overfenced than underprotected. |
|
|
70 | - ev_run now returns a boolean status (true meaning watchers are |
|
|
71 | still active). |
|
|
72 | - ev_once: undef EV_ERROR in ev_kqueue.c, to avoid clashing with |
|
|
73 | libev's EV_ERROR (reported by 191919). |
|
|
74 | - (ecb) add memory fence support for xlC (Darin McBride). |
|
|
75 | - (ecb) add memory fence support for gcc-mips (Anton Kirilov). |
|
|
76 | - (ecb) add memory fence support for gcc-alpha (Christian Weisgerber). |
|
|
77 | - work around some kernels losing file descriptors by leaking |
|
|
78 | the kqueue descriptor in the child. |
|
|
79 | - work around linux inotify not reporting IN_ATTRIB changes for directories |
|
|
80 | in many cases. |
|
|
81 | - include sys/syscall.h instead of plain syscall.h. |
|
|
82 | - check for io watcher loops in ev_verify, check for the most |
|
|
83 | common reported usage bug in ev_io_start. |
|
|
84 | - choose socket vs. WSASocket at compiletime using EV_USE_WSASOCKET. |
|
|
85 | - always use WSASend/WSARecv directly on windows, hoping that this |
|
|
86 | works in all cases (unlike read/write/send/recv...). |
|
|
87 | - try to detect signals around a fork faster (test program by |
|
|
88 | Denis Bilenko). |
|
|
89 | - work around recent glibc versions that leak memory in realloc. |
|
|
90 | - rename ev::embed::set to ev::embed::set_embed to avoid clashing |
|
|
91 | the watcher base set (loop) method. |
|
|
92 | - rewrite the async/signal pipe logic to always keep a valid fd, which |
|
|
93 | simplifies (and hopefully correctifies :) the race checking |
|
|
94 | on fork, at the cost of one extra fd. |
|
|
95 | - add fat, msdos, jffs2, ramfs, ntfs and btrfs to the list of |
|
|
96 | inotify-supporting filesystems. |
|
|
97 | - move orig_CFLAGS assignment to after AC_INIT, as newer autoconf |
|
|
98 | versions ignore it before |
|
|
99 | (https://bugzilla.redhat.com/show_bug.cgi?id=908096). |
|
|
100 | - add some untested android support. |
|
|
101 | - enum expressions must be of type int (reported by Juan Pablo L). |
|
|
102 | |
|
|
103 | 4.11 Sat Feb 4 19:52:39 CET 2012 |
|
|
104 | - INCOMPATIBLE CHANGE: ev_timer_again now clears the pending status, as |
|
|
105 | was documented already, but not implemented in the repeating case. |
|
|
106 | - new compiletime symbols: EV_NO_SMP and EV_NO_THREADS. |
|
|
107 | - fix a race where the workaround against the epoll fork bugs |
|
|
108 | caused signals to not be handled anymore. |
|
|
109 | - correct backend_fudge for most backends, and implement a windows |
|
|
110 | specific workaround to avoid looping because we call both |
|
|
111 | select and Sleep, both with different time resolutions. |
|
|
112 | - document range and guarantees of ev_sleep. |
|
|
113 | - document reasonable ranges for periodics interval and offset. |
|
|
114 | - rename backend_fudge to backend_mintime to avoid future confusion :) |
|
|
115 | - change the default periodic reschedule function to hopefully be more |
|
|
116 | exact and correct even in corner cases or in the far future. |
|
|
117 | - do not rely on -lm anymore: use it when available but use our |
|
|
118 | own floor () if it is missing. This should make it easier to embed, |
|
|
119 | as no external libraries are required. |
|
|
120 | - strategically import macros from libecb and mark rarely-used functions |
|
|
121 | as cache-cold (saving almost 2k code size on typical amd64 setups). |
|
|
122 | - add Symbols.ev and Symbols.event files, that were missing. |
|
|
123 | - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now). |
|
|
124 | - fix #3 "be smart about timeouts" to not "deadlock" when |
|
|
125 | timeout == now, also improve the section overall. |
|
|
126 | - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom. |
|
|
127 | - support new EV_API_STATIC mode to make all libev symbols |
|
|
128 | static. |
|
|
129 | - supply default CFLAGS of -g -O3 with gcc when original CFLAGS |
|
|
130 | were empty. |
|
|
131 | |
|
|
132 | 4.04 Wed Feb 16 09:01:51 CET 2011 |
|
|
133 | - fix two problems in the native win32 backend, where reuse of fd's |
|
|
134 | with different underlying handles caused handles not to be removed |
|
|
135 | or added to the select set (analyzed and tested by Bert Belder). |
|
|
136 | - do no rely on ceil() in ev_e?poll.c. |
|
|
137 | - backport libev to HP-UX versions before 11 v3. |
|
|
138 | - configure did not detect nanosleep and clock_gettime properly when |
|
|
139 | they are available in the libc (as opposed to -lrt). |
|
|
140 | |
|
|
141 | 4.03 Tue Jan 11 14:37:25 CET 2011 |
|
|
142 | - officially support polling files with all backends. |
|
|
143 | - support files, /dev/zero etc. the same way as select in the epoll |
|
|
144 | backend, by generating events on our own. |
|
|
145 | - ports backend: work around solaris bug 6874410 and many related ones |
|
|
146 | (EINTR, maybe more), with no performance loss (note that the solaris |
|
|
147 | bug report is actually wrong, reality is far more bizarre and broken |
|
|
148 | than that). |
|
|
149 | - define EV_READ/EV_WRITE as macros in event.h, as some programs use |
|
|
150 | #ifdef to test for them. |
|
|
151 | - new (experimental) function: ev_feed_signal. |
|
|
152 | - new (to become default) EVFLAG_NOSIGMASK flag. |
|
|
153 | - new EVBACKEND_MASK symbol. |
|
|
154 | - updated COMMON IDIOMS SECTION. |
|
|
155 | |
|
|
156 | 4.01 Fri Nov 5 21:51:29 CET 2010 |
|
|
157 | - automake fucked it up, apparently, --add-missing -f is not quite enough |
|
|
158 | to make it update its files, so 4.00 didn't install ev++.h and |
|
|
159 | event.h on make install. grrr. |
|
|
160 | - ev_loop(count|depth) didn't return anything (Robin Haberkorn). |
|
|
161 | - change EV_UNDEF to 0xffffffff to silence some overzealous compilers. |
|
|
162 | - use "(libev) " prefix for all libev error messages now. |
|
|
163 | |
|
|
164 | 4.00 Mon Oct 25 12:32:12 CEST 2010 |
|
|
165 | - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading. |
|
|
166 | - ev_embed_stop did not correctly stop the watcher (very good |
|
|
167 | testcase by Vladimir Timofeev). |
|
|
168 | - ev_run will now always update the current loop time - it erroneously |
|
|
169 | didn't when idle watchers were active, causing timers not to fire. |
|
|
170 | - fix a bug where a timeout of zero caused the timer not to fire |
|
|
171 | in the libevent emulation (testcase by Péter Szabó). |
|
|
172 | - applied win32 fixes by Michael Lenaghan (also James Mansion). |
|
|
173 | - replace EV_MINIMAL by EV_FEATURES. |
|
|
174 | - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it |
|
|
175 | seems the former is *much* faster than the latter. |
|
|
176 | - linux kernel version detection (for inotify bug workarounds) |
|
|
177 | did not work properly. |
|
|
178 | - reduce the number of spurious wake-ups with the ports backend. |
|
|
179 | - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu). |
|
|
180 | - do async init within ev_async_start, not ev_async_set, which avoids |
|
|
181 | an API quirk where the set function must be called in the C++ API |
|
|
182 | even when there is nothing to set. |
|
|
183 | - add (undocumented) EV_ENABLE when adding events with kqueue, |
|
|
184 | this might help with OS X, which seems to need it despite documenting |
|
|
185 | not to need it (helpfully pointed out by Tilghman Lesher). |
|
|
186 | - do not use poll by default on freebsd, it's broken (what isn't |
|
|
187 | on freebsd...). |
|
|
188 | - allow to embed epoll on kernels >= 2.6.32. |
|
|
189 | - configure now prepends -O3, not appends it, so one can still |
|
|
190 | override it. |
|
|
191 | - ev.pod: greatly expanded the portability section, added a porting |
|
|
192 | section, a description of watcher states and made lots of minor fixes. |
|
|
193 | - disable poll backend on AIX, the poll header spams the namespace |
|
|
194 | and it's not worth working around dead platforms (reported |
|
|
195 | and analyzed by Aivars Kalvans). |
|
|
196 | - improve header file compatibility of the standalone eventfd code |
|
|
197 | in an obscure case. |
|
|
198 | - implement EV_AVOID_STDIO option. |
|
|
199 | - do not use sscanf to parse linux version number (smaller, faster, |
|
|
200 | no sscanf dependency). |
|
|
201 | - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings. |
|
|
202 | - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs. |
|
|
203 | - add section on accept() problems to the manpage. |
|
|
204 | - rename EV_TIMEOUT to EV_TIMER. |
|
|
205 | - rename ev_loop_count/depth/verify/loop/unloop. |
|
|
206 | - remove ev_default_destroy and ev_default_fork. |
|
|
207 | - switch to two-digit minor version. |
|
|
208 | - work around an apparent gentoo compiler bug. |
|
|
209 | - define _DARWIN_UNLIMITED_SELECT. just so. |
|
|
210 | - use enum instead of #define for most constants. |
|
|
211 | - improve compatibility to older C++ compilers. |
|
|
212 | - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now |
|
|
213 | default arguments when compiled as C++. |
|
|
214 | - enable automake dependency tracking. |
|
|
215 | - ev_loop_new no longer leaks memory when loop creation failed. |
|
|
216 | - new ev_cleanup watcher type. |
|
|
217 | |
|
|
218 | 3.9 Thu Dec 31 07:59:59 CET 2009 |
|
|
219 | - signalfd is no longer used by default and has to be requested |
|
|
220 | explicitly - this means that easy to catch bugs become hard to |
|
|
221 | catch race conditions, but the users have spoken. |
|
|
222 | - point out the unspecified signal mask in the documentation, and |
|
|
223 | that this is a race condition regardless of EV_SIGNALFD. |
3 | - backport inotify code to C89. |
224 | - backport inotify code to C89. |
|
|
225 | - inotify file descriptors could leak into child processes. |
|
|
226 | - ev_stat watchers could keep an erroneous extra ref on the loop, |
|
|
227 | preventing exit when unregistering all watchers (testcases |
|
|
228 | provided by ry@tinyclouds.org). |
4 | - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration |
229 | - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration |
5 | symbols to make it easier for apps to do their own fd management. |
230 | symbols to make it easier for apps to do their own fd management. |
6 | - support EV_IDLE_ENABLE being disabled in ev++.h |
231 | - support EV_IDLE_ENABLE being disabled in ev++.h |
7 | (patch by Didier Spezia). |
232 | (patch by Didier Spezia). |
8 | - point out the unspecified signal mask in the documentation. |
233 | - take advantage of inotify_init1, if available, to set cloexec/nonblock |
|
|
234 | on fd creation, to avoid races. |
|
|
235 | - the signal handling pipe wasn't always initialised under windows |
|
|
236 | (analysed by lekma). |
|
|
237 | - changed minimum glibc requirement from glibc 2.9 to 2.7, for |
|
|
238 | signalfd. |
|
|
239 | - add missing string.h include (Denis F. Latypoff). |
|
|
240 | - only replace ev_stat.prev when we detect an actual difference, |
|
|
241 | so prev is (almost) always different to attr. this might |
|
|
242 | have caused the problems with 04_stat.t. |
|
|
243 | - add ev::timer->remaining () method to C++ API. |
9 | |
244 | |
10 | 3.8 Sun Aug 9 14:30:45 CEST 2009 |
245 | 3.8 Sun Aug 9 14:30:45 CEST 2009 |
11 | - incompatible change: do not necessarily reset signal handler |
246 | - incompatible change: do not necessarily reset signal handler |
12 | to SIG_DFL when a sighandler is stopped. |
247 | to SIG_DFL when a sighandler is stopped. |
13 | - ev_default_destroy did not properly free or zero some members, |
248 | - ev_default_destroy did not properly free or zero some members, |
14 | potentially causing crashes and memory corruption on repated |
249 | potentially causing crashes and memory corruption on repeated |
15 | ev_default_destroy/ev_default_loop calls. |
250 | ev_default_destroy/ev_default_loop calls. |
16 | - take advantage of signalfd on GNU/Linux systems. |
251 | - take advantage of signalfd on GNU/Linux systems. |
17 | - document that the signal mask might be in an unspecified |
252 | - document that the signal mask might be in an unspecified |
18 | state when using libev's signal handling. |
253 | state when using libev's signal handling. |
19 | - take advantage of some GNU/Linux calls to set cloexec/nonblock |
254 | - take advantage of some GNU/Linux calls to set cloexec/nonblock |
… | |
… | |
54 | - correctly zero the idx values for arrays, so destroying and |
289 | - correctly zero the idx values for arrays, so destroying and |
55 | reinitialising the default loop actually works (patch by |
290 | reinitialising the default loop actually works (patch by |
56 | Malek Hadj-Ali). |
291 | Malek Hadj-Ali). |
57 | - implement ev_suspend and ev_resume. |
292 | - implement ev_suspend and ev_resume. |
58 | - new EV_CUSTOM revents flag for use by applications. |
293 | - new EV_CUSTOM revents flag for use by applications. |
59 | - add documentation section about priorites. |
294 | - add documentation section about priorities. |
60 | - add a glossary to the dcoumentation. |
295 | - add a glossary to the documentation. |
61 | - extend the ev_fork description slightly. |
296 | - extend the ev_fork description slightly. |
62 | - optimize a jump out of call_pending. |
297 | - optimize a jump out of call_pending. |
63 | |
298 | |
64 | 3.53 Sun Feb 15 02:38:20 CET 2009 |
299 | 3.53 Sun Feb 15 02:38:20 CET 2009 |
65 | - fix a bug in event pipe creation on win32 that would cause a |
300 | - fix a bug in event pipe creation on win32 that would cause a |
… | |
… | |
141 | do similar changes for event.c. |
376 | do similar changes for event.c. |
142 | |
377 | |
143 | 3.45 Tue Oct 21 21:59:26 CEST 2008 |
378 | 3.45 Tue Oct 21 21:59:26 CEST 2008 |
144 | - disable inotify usage on linux <2.6.25, as it is broken |
379 | - disable inotify usage on linux <2.6.25, as it is broken |
145 | (reported by Yoann Vandoorselaere). |
380 | (reported by Yoann Vandoorselaere). |
146 | - ev_stat errornously would try to add inotify watchers |
381 | - ev_stat erroneously would try to add inotify watchers |
147 | even when inotify wasn't available (this should only |
382 | even when inotify wasn't available (this should only |
148 | have a performance impact). |
383 | have a performance impact). |
149 | - ev_once now passes both timeout and io to the callback if both |
384 | - ev_once now passes both timeout and io to the callback if both |
150 | occur concurrently, instead of giving timeouts precedence. |
385 | occur concurrently, instead of giving timeouts precedence. |
151 | - disable EV_USE_INOTIFY when sys/inotify.h is too old. |
386 | - disable EV_USE_INOTIFY when sys/inotify.h is too old. |
… | |
… | |
219 | - enable inlining with gcc versions 2 and 3. |
454 | - enable inlining with gcc versions 2 and 3. |
220 | - work around broken poll implementations potentially |
455 | - work around broken poll implementations potentially |
221 | not clearing revents field in ev_poll (Brandon Black) |
456 | not clearing revents field in ev_poll (Brandon Black) |
222 | (no such systems are known at this time). |
457 | (no such systems are known at this time). |
223 | - work around a bug in realloc on openbsd and darwin, |
458 | - work around a bug in realloc on openbsd and darwin, |
224 | also makes the errornous valgrind complaints |
459 | also makes the erroneous valgrind complaints |
225 | go away (noted by various people). |
460 | go away (noted by various people). |
226 | - fix ev_async_pending, add c++ wrapper for ev_async |
461 | - fix ev_async_pending, add c++ wrapper for ev_async |
227 | (based on patch sent by Johannes Deisenhofer. |
462 | (based on patch sent by Johannes Deisenhofer). |
228 | - add sensible set method to ev::embed. |
463 | - add sensible set method to ev::embed. |
229 | - made integer constants type int in ev.h. |
464 | - made integer constants type int in ev.h. |
230 | |
465 | |
231 | 3.2 Wed Apr 2 17:11:19 CEST 2008 |
466 | 3.2 Wed Apr 2 17:11:19 CEST 2008 |
232 | - fix a 64 bit overflow issue in the select backend, |
467 | - fix a 64 bit overflow issue in the select backend, |