| 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 | |
|
|
3 | TODO: revisit 59.x timer in the light of modern powersaving |
|
|
4 | TODO: maybe use timerfd to detect time jumps on linux |
|
|
5 | TODO: document EV_TSTAMP_T |
|
|
6 | - linuxaio backend resulted in random memory corruption |
|
|
7 | when loop is forked. |
|
|
8 | - linuxaio backend might have tried to cancel an iocb |
|
|
9 | multiple times (was unable to trigger this). |
|
|
10 | - linuxaio backend now employs a generation counter to |
|
|
11 | avoid handling spurious events from cancelled requests. |
|
|
12 | - io_cancel can return EINTR, deal with it. also, assume |
|
|
13 | io_submit also returns EINTR. |
|
|
14 | - fix some other minor bugs in linuxaio backend. |
|
|
15 | - ev_tstamp type can now be overriden by defining EV_TSTAMP_T. |
|
|
16 | - cleanup: replace expect_true/false and noinline by their |
|
|
17 | libecb counterparts. |
|
|
18 | - move syscall infrastructure from ev_linuxaio.c to ev.c. |
|
|
19 | - prepare io_uring integration. |
|
|
20 | - tweak ev_floor. |
|
|
21 | - epoll, poll, win32 Sleep and other places that use millisecond |
|
|
22 | reslution now all try to round up times. |
|
|
23 | |
|
|
24 | 4.27 Thu Jun 27 22:43:44 CEST 2019 |
|
|
25 | - linux aio backend almost completely rewritten to work around its |
|
|
26 | limitations. |
|
|
27 | - linux aio backend now requires linux 4.19+. |
|
|
28 | - epoll backend now mandatory for linux aio backend. |
|
|
29 | - fail assertions more aggressively on invalid fd's detected |
|
|
30 | in the event loop, do not just silently fd_kill in case of |
|
|
31 | user error. |
|
|
32 | - ev_io_start/ev_io_stop now verify the watcher fd using |
|
|
33 | a syscall when EV_VERIFY is 2 or higher. |
|
|
34 | |
|
|
35 | 4.26 (EV only) |
|
|
36 | - update to libecb 0x00010006. |
|
|
37 | - new experimental linux aio backend (linux 4.18+). |
|
|
38 | - removed redundant 0-ptr check in ev_once. |
|
|
39 | - updated/extended ev_set_allocator documentation. |
|
|
40 | - replaced EMPTY2 macro by array_needsize_noinit. |
|
|
41 | - minor code cleanups. |
|
|
42 | - epoll backend now uses epoll_create1 also after fork. |
|
|
43 | |
|
|
44 | 4.25 Fri Dec 21 07:49:20 CET 2018 |
|
|
45 | - INCOMPATIBLE CHANGE: EV_THROW was renamed to EV_NOEXCEPT |
|
|
46 | (EV_THROW still provided) and now uses noexcept on C++11 or newer. |
|
|
47 | - move the darwin select workaround higher in ev.c, as newer versions of |
|
|
48 | darwin managed to break their broken select even more. |
|
|
49 | - ANDROID => __ANDROID__ (reported by enh@google.com). |
|
|
50 | - disable epoll_create1 on android because it has broken header files |
|
|
51 | and google is unwilling to fix them (reported by enh@google.com). |
|
|
52 | - avoid a minor compilation warning on win32. |
|
|
53 | - c++: remove deprecated dynamic throw() specifications. |
|
|
54 | - c++: improve the (unsupported) bad_loop exception class. |
|
|
55 | - backport perl ev_periodic example to C, untested. |
|
|
56 | - update libecb, biggets change is to include a memory fence |
|
|
57 | in ECB_MEMORY_FENCE_RELEASE on x86/amd64. |
|
|
58 | - minor autoconf/automake modernisation. |
|
|
59 | |
|
|
60 | 4.24 Wed Dec 28 05:19:55 CET 2016 |
|
|
61 | - bump version to 4.24, as the release tarball inexplicably |
|
|
62 | didn't have the right version in ev.h, even though the cvs-tagged |
|
|
63 | version did have the right one (reported by Ales Teska). |
|
|
64 | |
|
|
65 | 4.23 Wed Nov 16 18:23:41 CET 2016 |
|
|
66 | - move some declarations at the beginning to help certain retarded |
|
|
67 | microsoft compilers, even though their documentation claims |
|
|
68 | otherwise (reported by Ruslan Osmanov). |
|
|
69 | |
|
|
70 | 4.22 Sun Dec 20 22:11:50 CET 2015 |
|
|
71 | - when epoll detects unremovable fds in the fd set, rebuild |
|
|
72 | only the epoll descriptor, not the signal pipe, to avoid |
|
|
73 | SIGPIPE in ev_async_send. This doesn't solve it on fork, |
|
|
74 | so document what needs to be done in ev_loop_fork |
|
|
75 | (analyzed by Benjamin Mahler). |
|
|
76 | - remove superfluous sys/timeb.h include on win32 |
|
|
77 | (analyzed by Jason Madden). |
|
|
78 | - updated libecb. |
|
|
79 | |
|
|
80 | 4.20 Sat Jun 20 13:01:43 CEST 2015 |
|
|
81 | - prefer noexcept over throw () with C++ 11. |
|
|
82 | - update ecb.h due to incompatibilities with c11. |
|
|
83 | - fix a potential aliasing issue when reading and writing |
|
|
84 | watcher callbacks. |
|
|
85 | |
|
|
86 | 4.19 Thu Sep 25 08:18:25 CEST 2014 |
|
|
87 | - ev.h wasn't valid C++ anymore, which tripped compilers other than |
|
|
88 | clang, msvc or gcc (analyzed by Raphael 'kena' Poss). Unfortunately, |
|
|
89 | C++ doesn't support typedefs for function pointers fully, so the affected |
|
|
90 | declarations have to spell out the types each time. |
|
|
91 | - when not using autoconf, tighten the check for clock_gettime and related |
|
|
92 | functionality. |
|
|
93 | |
|
|
94 | 4.18 Fri Sep 5 17:55:26 CEST 2014 |
|
|
95 | - events on files were not always generated properly with the |
|
|
96 | epoll backend (testcase by Assaf Inbal). |
|
|
97 | - mark event pipe fd as cloexec after a fork (analyzed by Sami Farin). |
|
|
98 | - (ecb) support m68k, m88k and sh (patch by Miod Vallat). |
|
|
99 | - use a reasonable fallback for EV_NSIG instead of erroring out |
|
|
100 | when we can't detect the signal set size. |
|
|
101 | - in the absence of autoconf, do not use the clock syscall |
|
|
102 | on glibc >= 2.17 (avoids the syscall AND -lrt on systems |
|
|
103 | doing clock_gettime in userspace). |
|
|
104 | - ensure extern "C" function pointers are used for externally-visible |
|
|
105 | loop callbacks (not watcher callbacks yet). |
|
|
106 | - (ecb) work around memory barriers and volatile apparently both being |
|
|
107 | broken in visual studio 2008 and later (analysed and patch by Nicolas Noble). |
|
|
108 | |
|
|
109 | 4.15 Fri Mar 1 12:04:50 CET 2013 |
|
|
110 | - destroying a non-default loop would stop the global waitpid |
|
|
111 | watcher (Denis Bilenko). |
|
|
112 | - queueing pending watchers of higher priority from a watcher now invokes |
|
|
113 | them in a timely fashion (reported by Denis Bilenko). |
|
|
114 | - add throw() to all libev functions that cannot throw exceptions, for |
|
|
115 | further code size decrease when compiling for C++. |
|
|
116 | - add throw () to callbacks that must not throw exceptions (allocator, |
|
|
117 | syserr, loop acquire/release, periodic reschedule cbs). |
|
|
118 | - fix event_base_loop return code, add event_get_callback, event_base_new, |
|
|
119 | event_base_get_method calls to improve libevent 1.x emulation and add |
|
|
120 | some libevent 2.x functionality (based on a patch by Jeff Davey). |
|
|
121 | - add more memory fences to fix a bug reported by Jeff Davey. Better |
|
|
122 | be overfenced than underprotected. |
|
|
123 | - ev_run now returns a boolean status (true meaning watchers are |
|
|
124 | still active). |
|
|
125 | - ev_once: undef EV_ERROR in ev_kqueue.c, to avoid clashing with |
|
|
126 | libev's EV_ERROR (reported by 191919). |
|
|
127 | - (ecb) add memory fence support for xlC (Darin McBride). |
|
|
128 | - (ecb) add memory fence support for gcc-mips (Anton Kirilov). |
|
|
129 | - (ecb) add memory fence support for gcc-alpha (Christian Weisgerber). |
|
|
130 | - work around some kernels losing file descriptors by leaking |
|
|
131 | the kqueue descriptor in the child. |
|
|
132 | - work around linux inotify not reporting IN_ATTRIB changes for directories |
|
|
133 | in many cases. |
|
|
134 | - include sys/syscall.h instead of plain syscall.h. |
|
|
135 | - check for io watcher loops in ev_verify, check for the most |
|
|
136 | common reported usage bug in ev_io_start. |
|
|
137 | - choose socket vs. WSASocket at compiletime using EV_USE_WSASOCKET. |
|
|
138 | - always use WSASend/WSARecv directly on windows, hoping that this |
|
|
139 | works in all cases (unlike read/write/send/recv...). |
|
|
140 | - try to detect signals around a fork faster (test program by |
|
|
141 | Denis Bilenko). |
|
|
142 | - work around recent glibc versions that leak memory in realloc. |
|
|
143 | - rename ev::embed::set to ev::embed::set_embed to avoid clashing |
|
|
144 | the watcher base set (loop) method. |
|
|
145 | - rewrite the async/signal pipe logic to always keep a valid fd, which |
|
|
146 | simplifies (and hopefully correctifies :) the race checking |
|
|
147 | on fork, at the cost of one extra fd. |
|
|
148 | - add fat, msdos, jffs2, ramfs, ntfs and btrfs to the list of |
|
|
149 | inotify-supporting filesystems. |
|
|
150 | - move orig_CFLAGS assignment to after AC_INIT, as newer autoconf |
|
|
151 | versions ignore it before |
|
|
152 | (https://bugzilla.redhat.com/show_bug.cgi?id=908096). |
|
|
153 | - add some untested android support. |
|
|
154 | - enum expressions must be of type int (reported by Juan Pablo L). |
|
|
155 | |
|
|
156 | 4.11 Sat Feb 4 19:52:39 CET 2012 |
|
|
157 | - INCOMPATIBLE CHANGE: ev_timer_again now clears the pending status, as |
|
|
158 | was documented already, but not implemented in the repeating case. |
|
|
159 | - new compiletime symbols: EV_NO_SMP and EV_NO_THREADS. |
|
|
160 | - fix a race where the workaround against the epoll fork bugs |
|
|
161 | caused signals to not be handled anymore. |
|
|
162 | - correct backend_fudge for most backends, and implement a windows |
|
|
163 | specific workaround to avoid looping because we call both |
|
|
164 | select and Sleep, both with different time resolutions. |
|
|
165 | - document range and guarantees of ev_sleep. |
|
|
166 | - document reasonable ranges for periodics interval and offset. |
|
|
167 | - rename backend_fudge to backend_mintime to avoid future confusion :) |
|
|
168 | - change the default periodic reschedule function to hopefully be more |
|
|
169 | exact and correct even in corner cases or in the far future. |
|
|
170 | - do not rely on -lm anymore: use it when available but use our |
|
|
171 | own floor () if it is missing. This should make it easier to embed, |
|
|
172 | as no external libraries are required. |
|
|
173 | - strategically import macros from libecb and mark rarely-used functions |
|
|
174 | as cache-cold (saving almost 2k code size on typical amd64 setups). |
|
|
175 | - add Symbols.ev and Symbols.event files, that were missing. |
|
|
176 | - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now). |
|
|
177 | - fix #3 "be smart about timeouts" to not "deadlock" when |
|
|
178 | timeout == now, also improve the section overall. |
|
|
179 | - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom. |
|
|
180 | - support new EV_API_STATIC mode to make all libev symbols |
|
|
181 | static. |
|
|
182 | - supply default CFLAGS of -g -O3 with gcc when original CFLAGS |
|
|
183 | were empty. |
|
|
184 | |
|
|
185 | 4.04 Wed Feb 16 09:01:51 CET 2011 |
|
|
186 | - fix two problems in the native win32 backend, where reuse of fd's |
|
|
187 | with different underlying handles caused handles not to be removed |
|
|
188 | or added to the select set (analyzed and tested by Bert Belder). |
|
|
189 | - do no rely on ceil() in ev_e?poll.c. |
|
|
190 | - backport libev to HP-UX versions before 11 v3. |
|
|
191 | - configure did not detect nanosleep and clock_gettime properly when |
|
|
192 | they are available in the libc (as opposed to -lrt). |
|
|
193 | |
|
|
194 | 4.03 Tue Jan 11 14:37:25 CET 2011 |
|
|
195 | - officially support polling files with all backends. |
|
|
196 | - support files, /dev/zero etc. the same way as select in the epoll |
|
|
197 | backend, by generating events on our own. |
|
|
198 | - ports backend: work around solaris bug 6874410 and many related ones |
|
|
199 | (EINTR, maybe more), with no performance loss (note that the solaris |
|
|
200 | bug report is actually wrong, reality is far more bizarre and broken |
|
|
201 | than that). |
|
|
202 | - define EV_READ/EV_WRITE as macros in event.h, as some programs use |
|
|
203 | #ifdef to test for them. |
|
|
204 | - new (experimental) function: ev_feed_signal. |
|
|
205 | - new (to become default) EVFLAG_NOSIGMASK flag. |
|
|
206 | - new EVBACKEND_MASK symbol. |
|
|
207 | - updated COMMON IDIOMS SECTION. |
| 2 | |
208 | |
| 3 | 4.01 Fri Nov 5 21:51:29 CET 2010 |
209 | 4.01 Fri Nov 5 21:51:29 CET 2010 |
| 4 | - automake fucked it up, apparently, --add-missing -f is not quite enough |
210 | - automake fucked it up, apparently, --add-missing -f is not quite enough |
| 5 | to make it update its files, so 4.00 didn't install ev++.h and |
211 | to make it update its files, so 4.00 didn't install ev++.h and |
| 6 | event.h on make install. grrr. |
212 | event.h on make install. grrr. |
| … | |
… | |
| 68 | catch race conditions, but the users have spoken. |
274 | catch race conditions, but the users have spoken. |
| 69 | - point out the unspecified signal mask in the documentation, and |
275 | - point out the unspecified signal mask in the documentation, and |
| 70 | that this is a race condition regardless of EV_SIGNALFD. |
276 | that this is a race condition regardless of EV_SIGNALFD. |
| 71 | - backport inotify code to C89. |
277 | - backport inotify code to C89. |
| 72 | - inotify file descriptors could leak into child processes. |
278 | - inotify file descriptors could leak into child processes. |
| 73 | - ev_stat watchers could keep an errornous extra ref on the loop, |
279 | - ev_stat watchers could keep an erroneous extra ref on the loop, |
| 74 | preventing exit when unregistering all watchers (testcases |
280 | preventing exit when unregistering all watchers (testcases |
| 75 | provided by ry@tinyclouds.org). |
281 | provided by ry@tinyclouds.org). |
| 76 | - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration |
282 | - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration |
| 77 | symbols to make it easier for apps to do their own fd management. |
283 | symbols to make it easier for apps to do their own fd management. |
| 78 | - support EV_IDLE_ENABLE being disabled in ev++.h |
284 | - support EV_IDLE_ENABLE being disabled in ev++.h |
| … | |
… | |
| 136 | - correctly zero the idx values for arrays, so destroying and |
342 | - correctly zero the idx values for arrays, so destroying and |
| 137 | reinitialising the default loop actually works (patch by |
343 | reinitialising the default loop actually works (patch by |
| 138 | Malek Hadj-Ali). |
344 | Malek Hadj-Ali). |
| 139 | - implement ev_suspend and ev_resume. |
345 | - implement ev_suspend and ev_resume. |
| 140 | - new EV_CUSTOM revents flag for use by applications. |
346 | - new EV_CUSTOM revents flag for use by applications. |
| 141 | - add documentation section about priorites. |
347 | - add documentation section about priorities. |
| 142 | - add a glossary to the dcoumentation. |
348 | - add a glossary to the documentation. |
| 143 | - extend the ev_fork description slightly. |
349 | - extend the ev_fork description slightly. |
| 144 | - optimize a jump out of call_pending. |
350 | - optimize a jump out of call_pending. |
| 145 | |
351 | |
| 146 | 3.53 Sun Feb 15 02:38:20 CET 2009 |
352 | 3.53 Sun Feb 15 02:38:20 CET 2009 |
| 147 | - fix a bug in event pipe creation on win32 that would cause a |
353 | - fix a bug in event pipe creation on win32 that would cause a |
