| 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 | TODO: ev_loop_wakeup |
| 4 | TODO: include ev_xyz_start in each example? |
4 | TODO: EV_STANDALONE == NO_HASSEL (do not use clock_gettime in ev_standalone) |
| 5 | TODO: section watcher states/lifetime |
5 | |
|
|
6 | TODO: document WSA_EV_USE_SOCKET in win32 part |
|
|
7 | TODO: ^ OR use WSASend/WSARecv on the handle, which always works |
|
|
8 | TODO: faq, process a thing in each iteration |
|
|
9 | - queueing pending watchers of higher priority from a watcher now invokes |
|
|
10 | them in a timely fashion (reported by Denis Bilenko). |
|
|
11 | - add throw() to all libev functions that cannot throw exceptions, for |
|
|
12 | further code size decrease when compiling for C++. |
|
|
13 | - add throw () to callbacks that must not throw exceptions (allocator, |
|
|
14 | syserr, loop acquire/release, periodic reschedule cbs). |
|
|
15 | - fix event_base_loop return code, add event_get_callback, event_base_new, |
|
|
16 | event_base_get_method calls to improve libevent 1.x emulation and add |
|
|
17 | some libevent 2.x funcitonality (based on a patch by Jeff Davey). |
|
|
18 | - add more memory fences to fix a bug reported by Jeff Davey. Better |
|
|
19 | be overfenced than underprotected. |
|
|
20 | - ev_run now returns a boolean status (true meaning watchers are |
|
|
21 | still active). |
|
|
22 | - ev_once: undef EV_ERROR in ev_kqueue.c, to avoid clashing with |
|
|
23 | libev's EV_ERROR (reported by 191919). |
|
|
24 | - (ecb) add memory fence support for xlC (Darin McBride). |
|
|
25 | - (ecb) add memory fence support for gcc-mips (Anton Kirilov). |
|
|
26 | - (ecb) add memory fence support for gcc-alpha (Christian Weisgerber). |
|
|
27 | - work around some kernels losing file descriptors by leaking |
|
|
28 | the kqueue descriptor in the child. |
|
|
29 | - include sys/syscall.h instead of plain syscall.h. |
|
|
30 | |
|
|
31 | 4.11 Sat Feb 4 19:52:39 CET 2012 |
|
|
32 | - INCOMPATIBLE CHANGE: ev_timer_again now clears the pending status, as |
|
|
33 | was documented already, but not implemented in the repeating case. |
|
|
34 | - new compiletime symbols: EV_NO_SMP and EV_NO_THREADS. |
|
|
35 | - fix a race where the workaround against the epoll fork bugs |
|
|
36 | caused signals to not be handled anymore. |
|
|
37 | - correct backend_fudge for most backends, and implement a windows |
|
|
38 | specific workaround to avoid looping because we call both |
|
|
39 | select and Sleep, both with different time resolutions. |
|
|
40 | - document range and guarantees of ev_sleep. |
|
|
41 | - document reasonable ranges for periodics interval and offset. |
|
|
42 | - rename backend_fudge to backend_mintime to avoid future confusion :) |
|
|
43 | - change the default periodic reschedule function to hopefully be more |
|
|
44 | exact and correct even in corner cases or in the far future. |
|
|
45 | - do not rely on -lm anymore: use it when available but use our |
|
|
46 | own floor () if it is missing. This should make it easier to embed, |
|
|
47 | as no external libraries are required. |
|
|
48 | - strategically import macros from libecb and mark rarely-used functions |
|
|
49 | as cache-cold (saving almost 2k code size on typical amd64 setups). |
|
|
50 | - add Symbols.ev and Symbols.event files, that were missing. |
|
|
51 | - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now). |
|
|
52 | - fix #3 "be smart about timeouts" to not "deadlock" when |
|
|
53 | timeout == now, also improve the section overall. |
|
|
54 | - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom. |
|
|
55 | - support new EV_API_STATIC mode to make all libev symbols |
|
|
56 | static. |
|
|
57 | - supply default CFLAGS of -g -O3 with gcc when original CFLAGS |
|
|
58 | were empty. |
|
|
59 | |
|
|
60 | 4.04 Wed Feb 16 09:01:51 CET 2011 |
|
|
61 | - fix two problems in the native win32 backend, where reuse of fd's |
|
|
62 | with different underlying handles caused handles not to be removed |
|
|
63 | or added to the select set (analyzed and tested by Bert Belder). |
|
|
64 | - do no rely on ceil() in ev_e?poll.c. |
|
|
65 | - backport libev to HP-UX versions before 11 v3. |
|
|
66 | - configure did not detect nanosleep and clock_gettime properly when |
|
|
67 | they are available in the libc (as opposed to -lrt). |
|
|
68 | |
|
|
69 | 4.03 Tue Jan 11 14:37:25 CET 2011 |
|
|
70 | - officially support polling files with all backends. |
|
|
71 | - support files, /dev/zero etc. the same way as select in the epoll |
|
|
72 | backend, by generating events on our own. |
|
|
73 | - ports backend: work around solaris bug 6874410 and many related ones |
|
|
74 | (EINTR, maybe more), with no performance loss (note that the solaris |
|
|
75 | bug report is actually wrong, reality is far more bizarre and broken |
|
|
76 | than that). |
|
|
77 | - define EV_READ/EV_WRITE as macros in event.h, as some programs use |
|
|
78 | #ifdef to test for them. |
|
|
79 | - new (experimental) function: ev_feed_signal. |
|
|
80 | - new (to become default) EVFLAG_NOSIGMASK flag. |
|
|
81 | - new EVBACKEND_MASK symbol. |
|
|
82 | - updated COMMON IDIOMS SECTION. |
|
|
83 | |
|
|
84 | 4.01 Fri Nov 5 21:51:29 CET 2010 |
|
|
85 | - automake fucked it up, apparently, --add-missing -f is not quite enough |
|
|
86 | to make it update its files, so 4.00 didn't install ev++.h and |
|
|
87 | event.h on make install. grrr. |
|
|
88 | - ev_loop(count|depth) didn't return anything (Robin Haberkorn). |
|
|
89 | - change EV_UNDEF to 0xffffffff to silence some overzealous compilers. |
|
|
90 | - use "(libev) " prefix for all libev error messages now. |
|
|
91 | |
|
|
92 | 4.00 Mon Oct 25 12:32:12 CEST 2010 |
| 6 | - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading. |
93 | - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading. |
| 7 | - ev_embed_stop did not correctly stop the watcher (very good |
94 | - ev_embed_stop did not correctly stop the watcher (very good |
| 8 | testcase by Vladimir Timofeev). |
95 | testcase by Vladimir Timofeev). |
|
|
96 | - ev_run will now always update the current loop time - it erroneously |
|
|
97 | didn't when idle watchers were active, causing timers not to fire. |
| 9 | - fix a bug where a timeout of zero caused the timer not to fire |
98 | - fix a bug where a timeout of zero caused the timer not to fire |
| 10 | in the libevent emulation (testcase by Péter Szabó). |
99 | in the libevent emulation (testcase by Péter Szabó). |
| 11 | - applied win32 fixes by Michael Lenaghan (also James Mansion). |
100 | - applied win32 fixes by Michael Lenaghan (also James Mansion). |
| 12 | - replace EV_MINIMAL by EV_FEATURES. |
101 | - replace EV_MINIMAL by EV_FEATURES. |
| 13 | - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it |
102 | - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it |
| 14 | seems the former is *much* faster than the latter. |
103 | seems the former is *much* faster than the latter. |
|
|
104 | - linux kernel version detection (for inotify bug workarounds) |
|
|
105 | did not work properly. |
|
|
106 | - reduce the number of spurious wake-ups with the ports backend. |
| 15 | - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu). |
107 | - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu). |
|
|
108 | - do async init within ev_async_start, not ev_async_set, which avoids |
|
|
109 | an API quirk where the set function must be called in the C++ API |
|
|
110 | even when there is nothing to set. |
| 16 | - add (undocumented) EV_ENABLE when adding events with kqueue, |
111 | - add (undocumented) EV_ENABLE when adding events with kqueue, |
| 17 | this might help with OS X, which seems to need it despite documenting |
112 | this might help with OS X, which seems to need it despite documenting |
| 18 | not to need it (helpfully pointed out by Tilghman Lesher). |
113 | not to need it (helpfully pointed out by Tilghman Lesher). |
| 19 | - do not use poll by default on freebsd, it's broken (what isn't |
114 | - do not use poll by default on freebsd, it's broken (what isn't |
| 20 | on freebsd...). |
115 | on freebsd...). |
|
|
116 | - allow to embed epoll on kernels >= 2.6.32. |
| 21 | - configure now prepends -O3, not appends it, so one can still |
117 | - configure now prepends -O3, not appends it, so one can still |
| 22 | override it. |
118 | override it. |
| 23 | - greatly expanded the portability section. |
119 | - ev.pod: greatly expanded the portability section, added a porting |
|
|
120 | section, a description of watcher states and made lots of minor fixes. |
| 24 | - disable poll backend on AIX, the poll header spams the namespace |
121 | - disable poll backend on AIX, the poll header spams the namespace |
| 25 | and it's not worth working around dead platforms (reported |
122 | and it's not worth working around dead platforms (reported |
| 26 | and analyzed by Aivars Kalvans). |
123 | and analyzed by Aivars Kalvans). |
| 27 | - improve header file compatibility of the standalone eventfd code |
124 | - improve header file compatibility of the standalone eventfd code |
| 28 | in an obscure case. |
125 | in an obscure case. |
| … | |
… | |
| 31 | no sscanf dependency). |
128 | no sscanf dependency). |
| 32 | - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings. |
129 | - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings. |
| 33 | - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs. |
130 | - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs. |
| 34 | - add section on accept() problems to the manpage. |
131 | - add section on accept() problems to the manpage. |
| 35 | - rename EV_TIMEOUT to EV_TIMER. |
132 | - rename EV_TIMEOUT to EV_TIMER. |
| 36 | - rename ev_loop_count/depth/verify. |
133 | - rename ev_loop_count/depth/verify/loop/unloop. |
|
|
134 | - remove ev_default_destroy and ev_default_fork. |
| 37 | - switch to two-digit minor version. |
135 | - switch to two-digit minor version. |
| 38 | - work around an apparent gentoo compiler bug. |
136 | - work around an apparent gentoo compiler bug. |
| 39 | - define _DARWIN_UNLIMITED_SELECT. just so. |
137 | - define _DARWIN_UNLIMITED_SELECT. just so. |
|
|
138 | - use enum instead of #define for most constants. |
|
|
139 | - improve compatibility to older C++ compilers. |
|
|
140 | - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now |
|
|
141 | default arguments when compiled as C++. |
|
|
142 | - enable automake dependency tracking. |
|
|
143 | - ev_loop_new no longer leaks memory when loop creation failed. |
|
|
144 | - new ev_cleanup watcher type. |
| 40 | |
145 | |
| 41 | 3.9 Thu Dec 31 07:59:59 CET 2009 |
146 | 3.9 Thu Dec 31 07:59:59 CET 2009 |
| 42 | - signalfd is no longer used by default and has to be requested |
147 | - signalfd is no longer used by default and has to be requested |
| 43 | explicitly - this means that easy to catch bugs become hard to |
148 | explicitly - this means that easy to catch bugs become hard to |
| 44 | catch race conditions, but the users have spoken. |
149 | catch race conditions, but the users have spoken. |
| 45 | - point out the unspecified signal mask in the documentation, and |
150 | - point out the unspecified signal mask in the documentation, and |
| 46 | that this is a race condition regardless of EV_SIGNALFD. |
151 | that this is a race condition regardless of EV_SIGNALFD. |
| 47 | - backport inotify code to C89. |
152 | - backport inotify code to C89. |
| 48 | - inotify file descriptors could leak into child processes. |
153 | - inotify file descriptors could leak into child processes. |
| 49 | - ev_stat watchers could keep an errornous extra ref on the loop, |
154 | - ev_stat watchers could keep an erroneous extra ref on the loop, |
| 50 | preventing exit when unregistering all watchers (testcases |
155 | preventing exit when unregistering all watchers (testcases |
| 51 | provided by ry@tinyclouds.org). |
156 | provided by ry@tinyclouds.org). |
| 52 | - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration |
157 | - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration |
| 53 | symbols to make it easier for apps to do their own fd management. |
158 | symbols to make it easier for apps to do their own fd management. |
| 54 | - support EV_IDLE_ENABLE being disabled in ev++.h |
159 | - support EV_IDLE_ENABLE being disabled in ev++.h |
| … | |
… | |
| 60 | - changed minimum glibc requirement from glibc 2.9 to 2.7, for |
165 | - changed minimum glibc requirement from glibc 2.9 to 2.7, for |
| 61 | signalfd. |
166 | signalfd. |
| 62 | - add missing string.h include (Denis F. Latypoff). |
167 | - add missing string.h include (Denis F. Latypoff). |
| 63 | - only replace ev_stat.prev when we detect an actual difference, |
168 | - only replace ev_stat.prev when we detect an actual difference, |
| 64 | so prev is (almost) always different to attr. this might |
169 | so prev is (almost) always different to attr. this might |
| 65 | have caused the probems with 04_stat.t. |
170 | have caused the problems with 04_stat.t. |
| 66 | - add ev::timer->remaining () method to C++ API. |
171 | - add ev::timer->remaining () method to C++ API. |
| 67 | |
172 | |
| 68 | 3.8 Sun Aug 9 14:30:45 CEST 2009 |
173 | 3.8 Sun Aug 9 14:30:45 CEST 2009 |
| 69 | - incompatible change: do not necessarily reset signal handler |
174 | - incompatible change: do not necessarily reset signal handler |
| 70 | to SIG_DFL when a sighandler is stopped. |
175 | to SIG_DFL when a sighandler is stopped. |
| 71 | - ev_default_destroy did not properly free or zero some members, |
176 | - ev_default_destroy did not properly free or zero some members, |
| 72 | potentially causing crashes and memory corruption on repated |
177 | potentially causing crashes and memory corruption on repeated |
| 73 | ev_default_destroy/ev_default_loop calls. |
178 | ev_default_destroy/ev_default_loop calls. |
| 74 | - take advantage of signalfd on GNU/Linux systems. |
179 | - take advantage of signalfd on GNU/Linux systems. |
| 75 | - document that the signal mask might be in an unspecified |
180 | - document that the signal mask might be in an unspecified |
| 76 | state when using libev's signal handling. |
181 | state when using libev's signal handling. |
| 77 | - take advantage of some GNU/Linux calls to set cloexec/nonblock |
182 | - take advantage of some GNU/Linux calls to set cloexec/nonblock |
| … | |
… | |
| 112 | - correctly zero the idx values for arrays, so destroying and |
217 | - correctly zero the idx values for arrays, so destroying and |
| 113 | reinitialising the default loop actually works (patch by |
218 | reinitialising the default loop actually works (patch by |
| 114 | Malek Hadj-Ali). |
219 | Malek Hadj-Ali). |
| 115 | - implement ev_suspend and ev_resume. |
220 | - implement ev_suspend and ev_resume. |
| 116 | - new EV_CUSTOM revents flag for use by applications. |
221 | - new EV_CUSTOM revents flag for use by applications. |
| 117 | - add documentation section about priorites. |
222 | - add documentation section about priorities. |
| 118 | - add a glossary to the dcoumentation. |
223 | - add a glossary to the dcoumentation. |
| 119 | - extend the ev_fork description slightly. |
224 | - extend the ev_fork description slightly. |
| 120 | - optimize a jump out of call_pending. |
225 | - optimize a jump out of call_pending. |
| 121 | |
226 | |
| 122 | 3.53 Sun Feb 15 02:38:20 CET 2009 |
227 | 3.53 Sun Feb 15 02:38:20 CET 2009 |
| … | |
… | |
| 199 | do similar changes for event.c. |
304 | do similar changes for event.c. |
| 200 | |
305 | |
| 201 | 3.45 Tue Oct 21 21:59:26 CEST 2008 |
306 | 3.45 Tue Oct 21 21:59:26 CEST 2008 |
| 202 | - disable inotify usage on linux <2.6.25, as it is broken |
307 | - disable inotify usage on linux <2.6.25, as it is broken |
| 203 | (reported by Yoann Vandoorselaere). |
308 | (reported by Yoann Vandoorselaere). |
| 204 | - ev_stat errornously would try to add inotify watchers |
309 | - ev_stat erroneously would try to add inotify watchers |
| 205 | even when inotify wasn't available (this should only |
310 | even when inotify wasn't available (this should only |
| 206 | have a performance impact). |
311 | have a performance impact). |
| 207 | - ev_once now passes both timeout and io to the callback if both |
312 | - ev_once now passes both timeout and io to the callback if both |
| 208 | occur concurrently, instead of giving timeouts precedence. |
313 | occur concurrently, instead of giving timeouts precedence. |
| 209 | - disable EV_USE_INOTIFY when sys/inotify.h is too old. |
314 | - disable EV_USE_INOTIFY when sys/inotify.h is too old. |
| … | |
… | |
| 277 | - enable inlining with gcc versions 2 and 3. |
382 | - enable inlining with gcc versions 2 and 3. |
| 278 | - work around broken poll implementations potentially |
383 | - work around broken poll implementations potentially |
| 279 | not clearing revents field in ev_poll (Brandon Black) |
384 | not clearing revents field in ev_poll (Brandon Black) |
| 280 | (no such systems are known at this time). |
385 | (no such systems are known at this time). |
| 281 | - work around a bug in realloc on openbsd and darwin, |
386 | - work around a bug in realloc on openbsd and darwin, |
| 282 | also makes the errornous valgrind complaints |
387 | also makes the erroneous valgrind complaints |
| 283 | go away (noted by various people). |
388 | go away (noted by various people). |
| 284 | - fix ev_async_pending, add c++ wrapper for ev_async |
389 | - fix ev_async_pending, add c++ wrapper for ev_async |
| 285 | (based on patch sent by Johannes Deisenhofer. |
390 | (based on patch sent by Johannes Deisenhofer). |
| 286 | - add sensible set method to ev::embed. |
391 | - add sensible set method to ev::embed. |
| 287 | - made integer constants type int in ev.h. |
392 | - made integer constants type int in ev.h. |
| 288 | |
393 | |
| 289 | 3.2 Wed Apr 2 17:11:19 CEST 2008 |
394 | 3.2 Wed Apr 2 17:11:19 CEST 2008 |
| 290 | - fix a 64 bit overflow issue in the select backend, |
395 | - fix a 64 bit overflow issue in the select backend, |
