| 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_prefork hint? |
|
|
4 | TODO: ev_loop_wakeup |
|
|
5 | TODO: confusion about threads ongiong |
|
|
6 | TODO: not_blocked flag. |
|
|
7 | |
|
|
8 | - correct backend_fudge for most backends, and implement a windows |
|
|
9 | specific workaround to avoid looping because we call both |
|
|
10 | select and Sleep, both with different time resolutions. |
|
|
11 | - change the default periodic reschedule function to hopefully be more |
|
|
12 | exact and correct even in corner cases or in the far future. |
|
|
13 | - document reasonable ranges for interval and offset. |
|
|
14 | - do not rely on -lm anymore: use it when available but use our |
|
|
15 | own floor () if it is missing. This should make it easier to embed. |
|
|
16 | |
|
|
17 | 4.04 Wed Feb 16 09:01:51 CET 2011 |
|
|
18 | - fix two problems in the native win32 backend, where reuse of fd's |
|
|
19 | with different underlying handles caused handles not to be removed |
|
|
20 | or added to the select set (analyzed and tested by Bert Belder). |
|
|
21 | - do no rely on ceil() in ev_e?poll.c. |
|
|
22 | - backport libev to HP-UX versions before 11 v3. |
|
|
23 | - configure did not detect nanosleep and clock_gettime properly when |
|
|
24 | they are available in the libc (as opposed to -lrt). |
|
|
25 | |
|
|
26 | 4.03 Tue Jan 11 14:37:25 CET 2011 |
|
|
27 | - officially support polling files with all backends. |
|
|
28 | - support files, /dev/zero etc. the same way as select in the epoll |
|
|
29 | backend, by generating events on our own. |
|
|
30 | - ports backend: work around solaris bug 6874410 and many related ones |
|
|
31 | (EINTR, maybe more), with no performance loss (note that the solaris |
|
|
32 | bug report is actually wrong, reality is far more bizarre and broken |
|
|
33 | than that). |
|
|
34 | - define EV_READ/EV_WRITE as macros in event.h, as some programs use |
|
|
35 | #ifdef to test for them. |
|
|
36 | - new (experimental) function: ev_feed_signal. |
|
|
37 | - new (to become default) EVFLAG_NOSIGMASK flag. |
|
|
38 | - new EVBACKEND_MASK symbol. |
|
|
39 | - updated COMMON IDIOMS SECTION. |
|
|
40 | |
|
|
41 | 4.01 Fri Nov 5 21:51:29 CET 2010 |
|
|
42 | - automake fucked it up, apparently, --add-missing -f is not quite enough |
|
|
43 | to make it update its files, so 4.00 didn't install ev++.h and |
|
|
44 | event.h on make install. grrr. |
|
|
45 | - ev_loop(count|depth) didn't return anything (Robin Haberkorn). |
|
|
46 | - change EV_UNDEF to 0xffffffff to silence some overzealous compilers. |
|
|
47 | - use "(libev) " prefix for all libev error messages now. |
|
|
48 | |
|
|
49 | 4.00 Mon Oct 25 12:32:12 CEST 2010 |
|
|
50 | - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading. |
|
|
51 | - ev_embed_stop did not correctly stop the watcher (very good |
|
|
52 | testcase by Vladimir Timofeev). |
|
|
53 | - ev_run will now always update the current loop time - it erroneously |
|
|
54 | didn't when idle watchers were active, causing timers not to fire. |
|
|
55 | - fix a bug where a timeout of zero caused the timer not to fire |
|
|
56 | in the libevent emulation (testcase by Péter Szabó). |
|
|
57 | - applied win32 fixes by Michael Lenaghan (also James Mansion). |
|
|
58 | - replace EV_MINIMAL by EV_FEATURES. |
|
|
59 | - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it |
|
|
60 | seems the former is *much* faster than the latter. |
|
|
61 | - linux kernel version detection (for inotify bug workarounds) |
|
|
62 | did not work properly. |
|
|
63 | - reduce the number of spurious wake-ups with the ports backend. |
|
|
64 | - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu). |
|
|
65 | - do async init within ev_async_start, not ev_async_set, which avoids |
|
|
66 | an API quirk where the set function must be called in the C++ API |
|
|
67 | even when there is nothing to set. |
|
|
68 | - add (undocumented) EV_ENABLE when adding events with kqueue, |
|
|
69 | this might help with OS X, which seems to need it despite documenting |
|
|
70 | not to need it (helpfully pointed out by Tilghman Lesher). |
|
|
71 | - do not use poll by default on freebsd, it's broken (what isn't |
|
|
72 | on freebsd...). |
|
|
73 | - allow to embed epoll on kernels >= 2.6.32. |
|
|
74 | - configure now prepends -O3, not appends it, so one can still |
|
|
75 | override it. |
|
|
76 | - ev.pod: greatly expanded the portability section, added a porting |
|
|
77 | section, a description of watcher states and made lots of minor fixes. |
|
|
78 | - disable poll backend on AIX, the poll header spams the namespace |
|
|
79 | and it's not worth working around dead platforms (reported |
|
|
80 | and analyzed by Aivars Kalvans). |
|
|
81 | - improve header file compatibility of the standalone eventfd code |
|
|
82 | in an obscure case. |
|
|
83 | - implement EV_AVOID_STDIO option. |
|
|
84 | - do not use sscanf to parse linux version number (smaller, faster, |
|
|
85 | no sscanf dependency). |
|
|
86 | - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings. |
|
|
87 | - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs. |
|
|
88 | - add section on accept() problems to the manpage. |
|
|
89 | - rename EV_TIMEOUT to EV_TIMER. |
|
|
90 | - rename ev_loop_count/depth/verify/loop/unloop. |
|
|
91 | - remove ev_default_destroy and ev_default_fork. |
|
|
92 | - switch to two-digit minor version. |
|
|
93 | - work around an apparent gentoo compiler bug. |
|
|
94 | - define _DARWIN_UNLIMITED_SELECT. just so. |
|
|
95 | - use enum instead of #define for most constants. |
|
|
96 | - improve compatibility to older C++ compilers. |
|
|
97 | - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now |
|
|
98 | default arguments when compiled as C++. |
|
|
99 | - enable automake dependency tracking. |
|
|
100 | - ev_loop_new no longer leaks memory when loop creation failed. |
|
|
101 | - new ev_cleanup watcher type. |
|
|
102 | |
|
|
103 | 3.9 Thu Dec 31 07:59:59 CET 2009 |
| 3 | - signalfd is no longer used by default and has to be requested |
104 | - signalfd is no longer used by default and has to be requested |
| 4 | explicitly - this means that easy to catch bugs become hard to |
105 | explicitly - this means that easy to catch bugs become hard to |
| 5 | catch race conditions, but the users have spoken. |
106 | catch race conditions, but the users have spoken. |
| 6 | - point out the unspecified signal mask in the documentation, and |
107 | - point out the unspecified signal mask in the documentation, and |
| 7 | that this is a race condition regardless of EV_SIGNALFD. |
108 | that this is a race condition regardless of EV_SIGNALFD. |
| … | |
… | |
| 21 | - changed minimum glibc requirement from glibc 2.9 to 2.7, for |
122 | - changed minimum glibc requirement from glibc 2.9 to 2.7, for |
| 22 | signalfd. |
123 | signalfd. |
| 23 | - add missing string.h include (Denis F. Latypoff). |
124 | - add missing string.h include (Denis F. Latypoff). |
| 24 | - only replace ev_stat.prev when we detect an actual difference, |
125 | - only replace ev_stat.prev when we detect an actual difference, |
| 25 | so prev is (almost) always different to attr. this might |
126 | so prev is (almost) always different to attr. this might |
| 26 | have caused the probems with 04_stat.t. |
127 | have caused the problems with 04_stat.t. |
| 27 | - add ev::timer->remaining () method to C++ API. |
128 | - add ev::timer->remaining () method to C++ API. |
| 28 | |
129 | |
| 29 | 3.8 Sun Aug 9 14:30:45 CEST 2009 |
130 | 3.8 Sun Aug 9 14:30:45 CEST 2009 |
| 30 | - incompatible change: do not necessarily reset signal handler |
131 | - incompatible change: do not necessarily reset signal handler |
| 31 | to SIG_DFL when a sighandler is stopped. |
132 | to SIG_DFL when a sighandler is stopped. |
| 32 | - ev_default_destroy did not properly free or zero some members, |
133 | - ev_default_destroy did not properly free or zero some members, |
| 33 | potentially causing crashes and memory corruption on repated |
134 | potentially causing crashes and memory corruption on repeated |
| 34 | ev_default_destroy/ev_default_loop calls. |
135 | ev_default_destroy/ev_default_loop calls. |
| 35 | - take advantage of signalfd on GNU/Linux systems. |
136 | - take advantage of signalfd on GNU/Linux systems. |
| 36 | - document that the signal mask might be in an unspecified |
137 | - document that the signal mask might be in an unspecified |
| 37 | state when using libev's signal handling. |
138 | state when using libev's signal handling. |
| 38 | - take advantage of some GNU/Linux calls to set cloexec/nonblock |
139 | - take advantage of some GNU/Linux calls to set cloexec/nonblock |
| … | |
… | |
| 160 | do similar changes for event.c. |
261 | do similar changes for event.c. |
| 161 | |
262 | |
| 162 | 3.45 Tue Oct 21 21:59:26 CEST 2008 |
263 | 3.45 Tue Oct 21 21:59:26 CEST 2008 |
| 163 | - disable inotify usage on linux <2.6.25, as it is broken |
264 | - disable inotify usage on linux <2.6.25, as it is broken |
| 164 | (reported by Yoann Vandoorselaere). |
265 | (reported by Yoann Vandoorselaere). |
| 165 | - ev_stat errornously would try to add inotify watchers |
266 | - ev_stat erroneously would try to add inotify watchers |
| 166 | even when inotify wasn't available (this should only |
267 | even when inotify wasn't available (this should only |
| 167 | have a performance impact). |
268 | have a performance impact). |
| 168 | - ev_once now passes both timeout and io to the callback if both |
269 | - ev_once now passes both timeout and io to the callback if both |
| 169 | occur concurrently, instead of giving timeouts precedence. |
270 | occur concurrently, instead of giving timeouts precedence. |
| 170 | - disable EV_USE_INOTIFY when sys/inotify.h is too old. |
271 | - disable EV_USE_INOTIFY when sys/inotify.h is too old. |
| … | |
… | |
| 238 | - enable inlining with gcc versions 2 and 3. |
339 | - enable inlining with gcc versions 2 and 3. |
| 239 | - work around broken poll implementations potentially |
340 | - work around broken poll implementations potentially |
| 240 | not clearing revents field in ev_poll (Brandon Black) |
341 | not clearing revents field in ev_poll (Brandon Black) |
| 241 | (no such systems are known at this time). |
342 | (no such systems are known at this time). |
| 242 | - work around a bug in realloc on openbsd and darwin, |
343 | - work around a bug in realloc on openbsd and darwin, |
| 243 | also makes the errornous valgrind complaints |
344 | also makes the erroneous valgrind complaints |
| 244 | go away (noted by various people). |
345 | go away (noted by various people). |
| 245 | - fix ev_async_pending, add c++ wrapper for ev_async |
346 | - fix ev_async_pending, add c++ wrapper for ev_async |
| 246 | (based on patch sent by Johannes Deisenhofer. |
347 | (based on patch sent by Johannes Deisenhofer). |
| 247 | - add sensible set method to ev::embed. |
348 | - add sensible set method to ev::embed. |
| 248 | - made integer constants type int in ev.h. |
349 | - made integer constants type int in ev.h. |
| 249 | |
350 | |
| 250 | 3.2 Wed Apr 2 17:11:19 CEST 2008 |
351 | 3.2 Wed Apr 2 17:11:19 CEST 2008 |
| 251 | - fix a 64 bit overflow issue in the select backend, |
352 | - fix a 64 bit overflow issue in the select backend, |
