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: move some other examples to common idioms? combining watchers, |
4 | TODO: include ev_xyz_start in each example? |
4 | thread usage, coroutine switch? |
5 | TODO: section watcher states/lifetime |
5 | |
|
|
6 | 4.03 |
|
|
7 | - officially support polling files with all backends. |
|
|
8 | - support files, /dev/zero etc. the same way as select in the epoll |
|
|
9 | backend, by generating events on our own. |
|
|
10 | - ports backend: work around solaris bug 6874410 and many related ones |
|
|
11 | (EINTR, maybe more), with no performance loss (note that the solaris |
|
|
12 | bug report is actually wrong, reality is far more bizarre and broken |
|
|
13 | than that). |
|
|
14 | - define EV_READ/EV_WRITE as macros in event.h, as some programs use |
|
|
15 | #ifdef to test for them. |
|
|
16 | - new (experimental) function: ev_feed_signal. |
|
|
17 | - new (to become default) EVFLAG_NOSIGMASK flag. |
|
|
18 | - new EVBACKEND_MASK symbol. |
|
|
19 | |
|
|
20 | 4.01 Fri Nov 5 21:51:29 CET 2010 |
|
|
21 | - automake fucked it up, apparently, --add-missing -f is not quite enough |
|
|
22 | to make it update its files, so 4.00 didn't install ev++.h and |
|
|
23 | event.h on make install. grrr. |
|
|
24 | - ev_loop(count|depth) didn't return anything (Robin Haberkorn). |
|
|
25 | - change EV_UNDEF to 0xffffffff to silence some overzealous compilers. |
|
|
26 | - use "(libev) " prefix for all libev error messages now. |
|
|
27 | |
|
|
28 | 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. |
29 | - "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 |
30 | - ev_embed_stop did not correctly stop the watcher (very good |
8 | testcase by Vladimir Timofeev). |
31 | testcase by Vladimir Timofeev). |
|
|
32 | - ev_run will now always update the current loop time - it erroneously |
|
|
33 | 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 |
34 | - fix a bug where a timeout of zero caused the timer not to fire |
10 | in the libevent emulation (testcase by Péter Szabó). |
35 | in the libevent emulation (testcase by Péter Szabó). |
11 | - applied win32 fixes by Michael Lenaghan (also James Mansion). |
36 | - applied win32 fixes by Michael Lenaghan (also James Mansion). |
12 | - replace EV_MINIMAL by EV_FEATURES. |
37 | - replace EV_MINIMAL by EV_FEATURES. |
13 | - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it |
38 | - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it |
14 | seems the former is *much* faster than the latter. |
39 | seems the former is *much* faster than the latter. |
|
|
40 | - linux kernel version detection (for inotify bug workarounds) |
|
|
41 | did not work properly. |
15 | - reduce the number of spurious wake-ups with the ports backend. |
42 | - reduce the number of spurious wake-ups with the ports backend. |
16 | - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu). |
43 | - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu). |
|
|
44 | - do async init within ev_async_start, not ev_async_set, which avoids |
|
|
45 | an API quirk where the set function must be called in the C++ API |
|
|
46 | even when there is nothing to set. |
17 | - add (undocumented) EV_ENABLE when adding events with kqueue, |
47 | - add (undocumented) EV_ENABLE when adding events with kqueue, |
18 | this might help with OS X, which seems to need it despite documenting |
48 | this might help with OS X, which seems to need it despite documenting |
19 | not to need it (helpfully pointed out by Tilghman Lesher). |
49 | not to need it (helpfully pointed out by Tilghman Lesher). |
20 | - do not use poll by default on freebsd, it's broken (what isn't |
50 | - do not use poll by default on freebsd, it's broken (what isn't |
21 | on freebsd...). |
51 | on freebsd...). |
|
|
52 | - allow to embed epoll on kernels >= 2.6.32. |
22 | - configure now prepends -O3, not appends it, so one can still |
53 | - configure now prepends -O3, not appends it, so one can still |
23 | override it. |
54 | override it. |
24 | - greatly expanded the portability section. |
55 | - ev.pod: greatly expanded the portability section, added a porting |
|
|
56 | section, a description of watcher states and made lots of minor fixes. |
25 | - disable poll backend on AIX, the poll header spams the namespace |
57 | - disable poll backend on AIX, the poll header spams the namespace |
26 | and it's not worth working around dead platforms (reported |
58 | and it's not worth working around dead platforms (reported |
27 | and analyzed by Aivars Kalvans). |
59 | and analyzed by Aivars Kalvans). |
28 | - improve header file compatibility of the standalone eventfd code |
60 | - improve header file compatibility of the standalone eventfd code |
29 | in an obscure case. |
61 | in an obscure case. |
… | |
… | |
32 | no sscanf dependency). |
64 | no sscanf dependency). |
33 | - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings. |
65 | - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings. |
34 | - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs. |
66 | - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs. |
35 | - add section on accept() problems to the manpage. |
67 | - add section on accept() problems to the manpage. |
36 | - rename EV_TIMEOUT to EV_TIMER. |
68 | - rename EV_TIMEOUT to EV_TIMER. |
37 | - rename ev_loop_count/depth/verify. |
69 | - rename ev_loop_count/depth/verify/loop/unloop. |
|
|
70 | - remove ev_default_destroy and ev_default_fork. |
38 | - switch to two-digit minor version. |
71 | - switch to two-digit minor version. |
39 | - work around an apparent gentoo compiler bug. |
72 | - work around an apparent gentoo compiler bug. |
40 | - define _DARWIN_UNLIMITED_SELECT. just so. |
73 | - define _DARWIN_UNLIMITED_SELECT. just so. |
|
|
74 | - use enum instead of #define for most constants. |
|
|
75 | - improve compatibility to older C++ compilers. |
|
|
76 | - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now |
|
|
77 | default arguments when compiled as C++. |
|
|
78 | - enable automake dependency tracking. |
|
|
79 | - ev_loop_new no longer leaks memory when loop creation failed. |
|
|
80 | - new ev_cleanup watcher type. |
41 | |
81 | |
42 | 3.9 Thu Dec 31 07:59:59 CET 2009 |
82 | 3.9 Thu Dec 31 07:59:59 CET 2009 |
43 | - signalfd is no longer used by default and has to be requested |
83 | - signalfd is no longer used by default and has to be requested |
44 | explicitly - this means that easy to catch bugs become hard to |
84 | explicitly - this means that easy to catch bugs become hard to |
45 | catch race conditions, but the users have spoken. |
85 | catch race conditions, but the users have spoken. |
… | |
… | |
61 | - changed minimum glibc requirement from glibc 2.9 to 2.7, for |
101 | - changed minimum glibc requirement from glibc 2.9 to 2.7, for |
62 | signalfd. |
102 | signalfd. |
63 | - add missing string.h include (Denis F. Latypoff). |
103 | - add missing string.h include (Denis F. Latypoff). |
64 | - only replace ev_stat.prev when we detect an actual difference, |
104 | - only replace ev_stat.prev when we detect an actual difference, |
65 | so prev is (almost) always different to attr. this might |
105 | so prev is (almost) always different to attr. this might |
66 | have caused the probems with 04_stat.t. |
106 | have caused the problems with 04_stat.t. |
67 | - add ev::timer->remaining () method to C++ API. |
107 | - add ev::timer->remaining () method to C++ API. |
68 | |
108 | |
69 | 3.8 Sun Aug 9 14:30:45 CEST 2009 |
109 | 3.8 Sun Aug 9 14:30:45 CEST 2009 |
70 | - incompatible change: do not necessarily reset signal handler |
110 | - incompatible change: do not necessarily reset signal handler |
71 | to SIG_DFL when a sighandler is stopped. |
111 | to SIG_DFL when a sighandler is stopped. |
72 | - ev_default_destroy did not properly free or zero some members, |
112 | - ev_default_destroy did not properly free or zero some members, |
73 | potentially causing crashes and memory corruption on repated |
113 | potentially causing crashes and memory corruption on repeated |
74 | ev_default_destroy/ev_default_loop calls. |
114 | ev_default_destroy/ev_default_loop calls. |
75 | - take advantage of signalfd on GNU/Linux systems. |
115 | - take advantage of signalfd on GNU/Linux systems. |
76 | - document that the signal mask might be in an unspecified |
116 | - document that the signal mask might be in an unspecified |
77 | state when using libev's signal handling. |
117 | state when using libev's signal handling. |
78 | - take advantage of some GNU/Linux calls to set cloexec/nonblock |
118 | - take advantage of some GNU/Linux calls to set cloexec/nonblock |
… | |
… | |
200 | do similar changes for event.c. |
240 | do similar changes for event.c. |
201 | |
241 | |
202 | 3.45 Tue Oct 21 21:59:26 CEST 2008 |
242 | 3.45 Tue Oct 21 21:59:26 CEST 2008 |
203 | - disable inotify usage on linux <2.6.25, as it is broken |
243 | - disable inotify usage on linux <2.6.25, as it is broken |
204 | (reported by Yoann Vandoorselaere). |
244 | (reported by Yoann Vandoorselaere). |
205 | - ev_stat errornously would try to add inotify watchers |
245 | - ev_stat erroneously would try to add inotify watchers |
206 | even when inotify wasn't available (this should only |
246 | even when inotify wasn't available (this should only |
207 | have a performance impact). |
247 | have a performance impact). |
208 | - ev_once now passes both timeout and io to the callback if both |
248 | - ev_once now passes both timeout and io to the callback if both |
209 | occur concurrently, instead of giving timeouts precedence. |
249 | occur concurrently, instead of giving timeouts precedence. |
210 | - disable EV_USE_INOTIFY when sys/inotify.h is too old. |
250 | - disable EV_USE_INOTIFY when sys/inotify.h is too old. |
… | |
… | |
278 | - enable inlining with gcc versions 2 and 3. |
318 | - enable inlining with gcc versions 2 and 3. |
279 | - work around broken poll implementations potentially |
319 | - work around broken poll implementations potentially |
280 | not clearing revents field in ev_poll (Brandon Black) |
320 | not clearing revents field in ev_poll (Brandon Black) |
281 | (no such systems are known at this time). |
321 | (no such systems are known at this time). |
282 | - work around a bug in realloc on openbsd and darwin, |
322 | - work around a bug in realloc on openbsd and darwin, |
283 | also makes the errornous valgrind complaints |
323 | also makes the erroneous valgrind complaints |
284 | go away (noted by various people). |
324 | go away (noted by various people). |
285 | - fix ev_async_pending, add c++ wrapper for ev_async |
325 | - fix ev_async_pending, add c++ wrapper for ev_async |
286 | (based on patch sent by Johannes Deisenhofer. |
326 | (based on patch sent by Johannes Deisenhofer). |
287 | - add sensible set method to ev::embed. |
327 | - add sensible set method to ev::embed. |
288 | - made integer constants type int in ev.h. |
328 | - made integer constants type int in ev.h. |
289 | |
329 | |
290 | 3.2 Wed Apr 2 17:11:19 CEST 2008 |
330 | 3.2 Wed Apr 2 17:11:19 CEST 2008 |
291 | - fix a 64 bit overflow issue in the select backend, |
331 | - fix a 64 bit overflow issue in the select backend, |