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 | |
|
|
10 | TODO: document the special problem of signals around fork. |
|
|
11 | TODO: store pid for each signal |
|
|
12 | TODO: document file descriptor usage per loop |
|
|
13 | - destroying a non-default loop would stop the global waitpid |
|
|
14 | watcher (Denis Bilenko). |
|
|
15 | - queueing pending watchers of higher priority from a watcher now invokes |
|
|
16 | them in a timely fashion (reported by Denis Bilenko). |
|
|
17 | - add throw() to all libev functions that cannot throw exceptions, for |
|
|
18 | further code size decrease when compiling for C++. |
|
|
19 | - add throw () to callbacks that must not throw exceptions (allocator, |
|
|
20 | syserr, loop acquire/release, periodic reschedule cbs). |
|
|
21 | - fix event_base_loop return code, add event_get_callback, event_base_new, |
|
|
22 | event_base_get_method calls to improve libevent 1.x emulation and add |
|
|
23 | some libevent 2.x functionality (based on a patch by Jeff Davey). |
|
|
24 | - add more memory fences to fix a bug reported by Jeff Davey. Better |
|
|
25 | be overfenced than underprotected. |
|
|
26 | - ev_run now returns a boolean status (true meaning watchers are |
|
|
27 | still active). |
|
|
28 | - ev_once: undef EV_ERROR in ev_kqueue.c, to avoid clashing with |
|
|
29 | libev's EV_ERROR (reported by 191919). |
|
|
30 | - (ecb) add memory fence support for xlC (Darin McBride). |
|
|
31 | - (ecb) add memory fence support for gcc-mips (Anton Kirilov). |
|
|
32 | - (ecb) add memory fence support for gcc-alpha (Christian Weisgerber). |
|
|
33 | - work around some kernels losing file descriptors by leaking |
|
|
34 | the kqueue descriptor in the child. |
|
|
35 | - include sys/syscall.h instead of plain syscall.h. |
|
|
36 | - check for io watcher loops in ev_verify, check for the most |
|
|
37 | common reported usage bug in ev_io_start. |
|
|
38 | - chose socket vs. WSASocket at compiletime using EV_USE_WSASOCKET. |
|
|
39 | - always use WSASend/WSARecv directly on windows, hoping that this |
|
|
40 | works in all cases (unlike read/write/send/recv...). |
|
|
41 | - try to detect signals around a fork faster (test program by |
|
|
42 | Denis Bilenko). |
|
|
43 | - work around recent glibc versions that leak memory in realloc. |
|
|
44 | - rename ev::embed::set to ev::embed::set_embed to avoid clashing |
|
|
45 | the watcher base set (loop) method. |
|
|
46 | - rewrite the async/signal pipe logic to always keep a valid fd, which |
|
|
47 | simplifies (and hopefuly correctifies :) the race checking |
|
|
48 | on fork, at the cost of one extra fd. |
|
|
49 | |
|
|
50 | 4.11 Sat Feb 4 19:52:39 CET 2012 |
|
|
51 | - INCOMPATIBLE CHANGE: ev_timer_again now clears the pending status, as |
|
|
52 | was documented already, but not implemented in the repeating case. |
|
|
53 | - new compiletime symbols: EV_NO_SMP and EV_NO_THREADS. |
|
|
54 | - fix a race where the workaround against the epoll fork bugs |
|
|
55 | caused signals to not be handled anymore. |
|
|
56 | - correct backend_fudge for most backends, and implement a windows |
|
|
57 | specific workaround to avoid looping because we call both |
|
|
58 | select and Sleep, both with different time resolutions. |
|
|
59 | - document range and guarantees of ev_sleep. |
|
|
60 | - document reasonable ranges for periodics interval and offset. |
|
|
61 | - rename backend_fudge to backend_mintime to avoid future confusion :) |
|
|
62 | - change the default periodic reschedule function to hopefully be more |
|
|
63 | exact and correct even in corner cases or in the far future. |
|
|
64 | - do not rely on -lm anymore: use it when available but use our |
|
|
65 | own floor () if it is missing. This should make it easier to embed, |
|
|
66 | as no external libraries are required. |
|
|
67 | - strategically import macros from libecb and mark rarely-used functions |
|
|
68 | as cache-cold (saving almost 2k code size on typical amd64 setups). |
|
|
69 | - add Symbols.ev and Symbols.event files, that were missing. |
|
|
70 | - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now). |
|
|
71 | - fix #3 "be smart about timeouts" to not "deadlock" when |
|
|
72 | timeout == now, also improve the section overall. |
|
|
73 | - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom. |
|
|
74 | - support new EV_API_STATIC mode to make all libev symbols |
|
|
75 | static. |
|
|
76 | - supply default CFLAGS of -g -O3 with gcc when original CFLAGS |
|
|
77 | were empty. |
|
|
78 | |
|
|
79 | 4.04 Wed Feb 16 09:01:51 CET 2011 |
|
|
80 | - fix two problems in the native win32 backend, where reuse of fd's |
|
|
81 | with different underlying handles caused handles not to be removed |
|
|
82 | or added to the select set (analyzed and tested by Bert Belder). |
|
|
83 | - do no rely on ceil() in ev_e?poll.c. |
|
|
84 | - backport libev to HP-UX versions before 11 v3. |
|
|
85 | - configure did not detect nanosleep and clock_gettime properly when |
|
|
86 | they are available in the libc (as opposed to -lrt). |
|
|
87 | |
|
|
88 | 4.03 Tue Jan 11 14:37:25 CET 2011 |
|
|
89 | - officially support polling files with all backends. |
|
|
90 | - support files, /dev/zero etc. the same way as select in the epoll |
|
|
91 | backend, by generating events on our own. |
|
|
92 | - ports backend: work around solaris bug 6874410 and many related ones |
|
|
93 | (EINTR, maybe more), with no performance loss (note that the solaris |
|
|
94 | bug report is actually wrong, reality is far more bizarre and broken |
|
|
95 | than that). |
|
|
96 | - define EV_READ/EV_WRITE as macros in event.h, as some programs use |
|
|
97 | #ifdef to test for them. |
|
|
98 | - new (experimental) function: ev_feed_signal. |
|
|
99 | - new (to become default) EVFLAG_NOSIGMASK flag. |
|
|
100 | - new EVBACKEND_MASK symbol. |
|
|
101 | - updated COMMON IDIOMS SECTION. |
|
|
102 | |
|
|
103 | 4.01 Fri Nov 5 21:51:29 CET 2010 |
|
|
104 | - automake fucked it up, apparently, --add-missing -f is not quite enough |
|
|
105 | to make it update its files, so 4.00 didn't install ev++.h and |
|
|
106 | event.h on make install. grrr. |
|
|
107 | - ev_loop(count|depth) didn't return anything (Robin Haberkorn). |
|
|
108 | - change EV_UNDEF to 0xffffffff to silence some overzealous compilers. |
|
|
109 | - use "(libev) " prefix for all libev error messages now. |
|
|
110 | |
|
|
111 | 4.00 Mon Oct 25 12:32:12 CEST 2010 |
|
|
112 | - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading. |
3 | - ev_embed_stop did not correctly stop the watcher (very good |
113 | - ev_embed_stop did not correctly stop the watcher (very good |
4 | testcase by Vladimir Timofeev). |
114 | testcase by Vladimir Timofeev). |
|
|
115 | - ev_run will now always update the current loop time - it erroneously |
|
|
116 | didn't when idle watchers were active, causing timers not to fire. |
|
|
117 | - fix a bug where a timeout of zero caused the timer not to fire |
|
|
118 | in the libevent emulation (testcase by Péter Szabó). |
5 | - applied win32 fixes by Michael Lenaghan. |
119 | - applied win32 fixes by Michael Lenaghan (also James Mansion). |
|
|
120 | - replace EV_MINIMAL by EV_FEATURES. |
|
|
121 | - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it |
|
|
122 | seems the former is *much* faster than the latter. |
|
|
123 | - linux kernel version detection (for inotify bug workarounds) |
|
|
124 | did not work properly. |
|
|
125 | - reduce the number of spurious wake-ups with the ports backend. |
6 | - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu). |
126 | - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu). |
|
|
127 | - do async init within ev_async_start, not ev_async_set, which avoids |
|
|
128 | an API quirk where the set function must be called in the C++ API |
|
|
129 | even when there is nothing to set. |
|
|
130 | - add (undocumented) EV_ENABLE when adding events with kqueue, |
|
|
131 | this might help with OS X, which seems to need it despite documenting |
|
|
132 | not to need it (helpfully pointed out by Tilghman Lesher). |
|
|
133 | - do not use poll by default on freebsd, it's broken (what isn't |
|
|
134 | on freebsd...). |
|
|
135 | - allow to embed epoll on kernels >= 2.6.32. |
7 | - configure now prepends -O3, not appends it, so one can still |
136 | - configure now prepends -O3, not appends it, so one can still |
8 | override it. |
137 | override it. |
|
|
138 | - ev.pod: greatly expanded the portability section, added a porting |
|
|
139 | section, a description of watcher states and made lots of minor fixes. |
9 | - disable poll backend on AIX, the poll header spams the namespace |
140 | - disable poll backend on AIX, the poll header spams the namespace |
10 | and it's not worth working around dead platforms (reported |
141 | and it's not worth working around dead platforms (reported |
11 | and analyzed by Aivars Kalvans). |
142 | and analyzed by Aivars Kalvans). |
12 | - improve header file compatibility of the standalone eventfd code |
143 | - improve header file compatibility of the standalone eventfd code |
13 | in an obscure case. |
144 | in an obscure case. |
14 | - implement EV_AVOID_STDIO option. |
145 | - implement EV_AVOID_STDIO option. |
15 | - do not use sscanf to parse linux version number (smaller, faster, |
146 | - do not use sscanf to parse linux version number (smaller, faster, |
16 | no sscanf dependency). |
147 | no sscanf dependency). |
|
|
148 | - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings. |
|
|
149 | - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs. |
|
|
150 | - add section on accept() problems to the manpage. |
|
|
151 | - rename EV_TIMEOUT to EV_TIMER. |
|
|
152 | - rename ev_loop_count/depth/verify/loop/unloop. |
|
|
153 | - remove ev_default_destroy and ev_default_fork. |
|
|
154 | - switch to two-digit minor version. |
|
|
155 | - work around an apparent gentoo compiler bug. |
|
|
156 | - define _DARWIN_UNLIMITED_SELECT. just so. |
|
|
157 | - use enum instead of #define for most constants. |
|
|
158 | - improve compatibility to older C++ compilers. |
|
|
159 | - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now |
|
|
160 | default arguments when compiled as C++. |
|
|
161 | - enable automake dependency tracking. |
|
|
162 | - ev_loop_new no longer leaks memory when loop creation failed. |
|
|
163 | - new ev_cleanup watcher type. |
17 | |
164 | |
18 | 3.9 Thu Dec 31 07:59:59 CET 2009 |
165 | 3.9 Thu Dec 31 07:59:59 CET 2009 |
19 | - signalfd is no longer used by default and has to be requested |
166 | - signalfd is no longer used by default and has to be requested |
20 | explicitly - this means that easy to catch bugs become hard to |
167 | explicitly - this means that easy to catch bugs become hard to |
21 | catch race conditions, but the users have spoken. |
168 | catch race conditions, but the users have spoken. |
22 | - point out the unspecified signal mask in the documentation, and |
169 | - point out the unspecified signal mask in the documentation, and |
23 | that this is a race condition regardless of EV_SIGNALFD. |
170 | that this is a race condition regardless of EV_SIGNALFD. |
24 | - backport inotify code to C89. |
171 | - backport inotify code to C89. |
25 | - inotify file descriptors could leak into child processes. |
172 | - inotify file descriptors could leak into child processes. |
26 | - ev_stat watchers could keep an errornous extra ref on the loop, |
173 | - ev_stat watchers could keep an erroneous extra ref on the loop, |
27 | preventing exit when unregistering all watchers (testcases |
174 | preventing exit when unregistering all watchers (testcases |
28 | provided by ry@tinyclouds.org). |
175 | provided by ry@tinyclouds.org). |
29 | - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration |
176 | - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration |
30 | symbols to make it easier for apps to do their own fd management. |
177 | symbols to make it easier for apps to do their own fd management. |
31 | - support EV_IDLE_ENABLE being disabled in ev++.h |
178 | - support EV_IDLE_ENABLE being disabled in ev++.h |
… | |
… | |
37 | - changed minimum glibc requirement from glibc 2.9 to 2.7, for |
184 | - changed minimum glibc requirement from glibc 2.9 to 2.7, for |
38 | signalfd. |
185 | signalfd. |
39 | - add missing string.h include (Denis F. Latypoff). |
186 | - add missing string.h include (Denis F. Latypoff). |
40 | - only replace ev_stat.prev when we detect an actual difference, |
187 | - only replace ev_stat.prev when we detect an actual difference, |
41 | so prev is (almost) always different to attr. this might |
188 | so prev is (almost) always different to attr. this might |
42 | have caused the probems with 04_stat.t. |
189 | have caused the problems with 04_stat.t. |
43 | - add ev::timer->remaining () method to C++ API. |
190 | - add ev::timer->remaining () method to C++ API. |
44 | |
191 | |
45 | 3.8 Sun Aug 9 14:30:45 CEST 2009 |
192 | 3.8 Sun Aug 9 14:30:45 CEST 2009 |
46 | - incompatible change: do not necessarily reset signal handler |
193 | - incompatible change: do not necessarily reset signal handler |
47 | to SIG_DFL when a sighandler is stopped. |
194 | to SIG_DFL when a sighandler is stopped. |
48 | - ev_default_destroy did not properly free or zero some members, |
195 | - ev_default_destroy did not properly free or zero some members, |
49 | potentially causing crashes and memory corruption on repated |
196 | potentially causing crashes and memory corruption on repeated |
50 | ev_default_destroy/ev_default_loop calls. |
197 | ev_default_destroy/ev_default_loop calls. |
51 | - take advantage of signalfd on GNU/Linux systems. |
198 | - take advantage of signalfd on GNU/Linux systems. |
52 | - document that the signal mask might be in an unspecified |
199 | - document that the signal mask might be in an unspecified |
53 | state when using libev's signal handling. |
200 | state when using libev's signal handling. |
54 | - take advantage of some GNU/Linux calls to set cloexec/nonblock |
201 | - take advantage of some GNU/Linux calls to set cloexec/nonblock |
… | |
… | |
89 | - correctly zero the idx values for arrays, so destroying and |
236 | - correctly zero the idx values for arrays, so destroying and |
90 | reinitialising the default loop actually works (patch by |
237 | reinitialising the default loop actually works (patch by |
91 | Malek Hadj-Ali). |
238 | Malek Hadj-Ali). |
92 | - implement ev_suspend and ev_resume. |
239 | - implement ev_suspend and ev_resume. |
93 | - new EV_CUSTOM revents flag for use by applications. |
240 | - new EV_CUSTOM revents flag for use by applications. |
94 | - add documentation section about priorites. |
241 | - add documentation section about priorities. |
95 | - add a glossary to the dcoumentation. |
242 | - add a glossary to the dcoumentation. |
96 | - extend the ev_fork description slightly. |
243 | - extend the ev_fork description slightly. |
97 | - optimize a jump out of call_pending. |
244 | - optimize a jump out of call_pending. |
98 | |
245 | |
99 | 3.53 Sun Feb 15 02:38:20 CET 2009 |
246 | 3.53 Sun Feb 15 02:38:20 CET 2009 |
… | |
… | |
176 | do similar changes for event.c. |
323 | do similar changes for event.c. |
177 | |
324 | |
178 | 3.45 Tue Oct 21 21:59:26 CEST 2008 |
325 | 3.45 Tue Oct 21 21:59:26 CEST 2008 |
179 | - disable inotify usage on linux <2.6.25, as it is broken |
326 | - disable inotify usage on linux <2.6.25, as it is broken |
180 | (reported by Yoann Vandoorselaere). |
327 | (reported by Yoann Vandoorselaere). |
181 | - ev_stat errornously would try to add inotify watchers |
328 | - ev_stat erroneously would try to add inotify watchers |
182 | even when inotify wasn't available (this should only |
329 | even when inotify wasn't available (this should only |
183 | have a performance impact). |
330 | have a performance impact). |
184 | - ev_once now passes both timeout and io to the callback if both |
331 | - ev_once now passes both timeout and io to the callback if both |
185 | occur concurrently, instead of giving timeouts precedence. |
332 | occur concurrently, instead of giving timeouts precedence. |
186 | - disable EV_USE_INOTIFY when sys/inotify.h is too old. |
333 | - disable EV_USE_INOTIFY when sys/inotify.h is too old. |
… | |
… | |
254 | - enable inlining with gcc versions 2 and 3. |
401 | - enable inlining with gcc versions 2 and 3. |
255 | - work around broken poll implementations potentially |
402 | - work around broken poll implementations potentially |
256 | not clearing revents field in ev_poll (Brandon Black) |
403 | not clearing revents field in ev_poll (Brandon Black) |
257 | (no such systems are known at this time). |
404 | (no such systems are known at this time). |
258 | - work around a bug in realloc on openbsd and darwin, |
405 | - work around a bug in realloc on openbsd and darwin, |
259 | also makes the errornous valgrind complaints |
406 | also makes the erroneous valgrind complaints |
260 | go away (noted by various people). |
407 | go away (noted by various people). |
261 | - fix ev_async_pending, add c++ wrapper for ev_async |
408 | - fix ev_async_pending, add c++ wrapper for ev_async |
262 | (based on patch sent by Johannes Deisenhofer. |
409 | (based on patch sent by Johannes Deisenhofer). |
263 | - add sensible set method to ev::embed. |
410 | - add sensible set method to ev::embed. |
264 | - made integer constants type int in ev.h. |
411 | - made integer constants type int in ev.h. |
265 | |
412 | |
266 | 3.2 Wed Apr 2 17:11:19 CEST 2008 |
413 | 3.2 Wed Apr 2 17:11:19 CEST 2008 |
267 | - fix a 64 bit overflow issue in the select backend, |
414 | - fix a 64 bit overflow issue in the select backend, |