| 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: include ev_xyz_start in each example? |
|
|
4 | TODO: section watcher states/lifetime |
|
|
5 | TODO: use enums //TODO: enum? |
|
|
6 | - "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 |
|
|
8 | testcase by Vladimir Timofeev). |
|
|
9 | - ev_run will now always update the current loop time - it errornously |
|
|
10 | didn't when idle watchers were active, causing timers not to fire. |
|
|
11 | - fix a bug where a timeout of zero caused the timer not to fire |
|
|
12 | in the libevent emulation (testcase by Péter Szabó). |
|
|
13 | - applied win32 fixes by Michael Lenaghan (also James Mansion). |
|
|
14 | - replace EV_MINIMAL by EV_FEATURES. |
|
|
15 | - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it |
|
|
16 | seems the former is *much* faster than the latter. |
|
|
17 | - reduce the number of spurious wake-ups with the ports backend. |
|
|
18 | - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu). |
|
|
19 | - do async init within ev_async_start, not ev_async_set, which avoids |
|
|
20 | an API quirk where the set function must be called in the C++ API |
|
|
21 | even when there is nothing to set. |
|
|
22 | - add (undocumented) EV_ENABLE when adding events with kqueue, |
|
|
23 | this might help with OS X, which seems to need it despite documenting |
|
|
24 | not to need it (helpfully pointed out by Tilghman Lesher). |
|
|
25 | - do not use poll by default on freebsd, it's broken (what isn't |
|
|
26 | on freebsd...). |
|
|
27 | - configure now prepends -O3, not appends it, so one can still |
|
|
28 | override it. |
|
|
29 | - greatly expanded the portability section. |
|
|
30 | - disable poll backend on AIX, the poll header spams the namespace |
|
|
31 | and it's not worth working around dead platforms (reported |
|
|
32 | and analyzed by Aivars Kalvans). |
|
|
33 | - improve header file compatibility of the standalone eventfd code |
|
|
34 | in an obscure case. |
|
|
35 | - implement EV_AVOID_STDIO option. |
|
|
36 | - do not use sscanf to parse linux version number (smaller, faster, |
|
|
37 | no sscanf dependency). |
|
|
38 | - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings. |
|
|
39 | - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs. |
|
|
40 | - add section on accept() problems to the manpage. |
|
|
41 | - rename EV_TIMEOUT to EV_TIMER. |
|
|
42 | - rename ev_loop_count/depth/verify. |
|
|
43 | - switch to two-digit minor version. |
|
|
44 | - work around an apparent gentoo compiler bug. |
|
|
45 | - define _DARWIN_UNLIMITED_SELECT. just so. |
|
|
46 | - use enum instead of #define for most constants. |
|
|
47 | - improve compatibility to older C++ compilers. |
|
|
48 | |
|
|
49 | 3.9 Thu Dec 31 07:59:59 CET 2009 |
|
|
50 | - signalfd is no longer used by default and has to be requested |
|
|
51 | explicitly - this means that easy to catch bugs become hard to |
|
|
52 | catch race conditions, but the users have spoken. |
|
|
53 | - point out the unspecified signal mask in the documentation, and |
|
|
54 | that this is a race condition regardless of EV_SIGNALFD. |
|
|
55 | - backport inotify code to C89. |
|
|
56 | - inotify file descriptors could leak into child processes. |
|
|
57 | - ev_stat watchers could keep an errornous extra ref on the loop, |
|
|
58 | preventing exit when unregistering all watchers (testcases |
|
|
59 | provided by ry@tinyclouds.org). |
|
|
60 | - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration |
|
|
61 | symbols to make it easier for apps to do their own fd management. |
|
|
62 | - support EV_IDLE_ENABLE being disabled in ev++.h |
|
|
63 | (patch by Didier Spezia). |
|
|
64 | - take advantage of inotify_init1, if available, to set cloexec/nonblock |
|
|
65 | on fd creation, to avoid races. |
|
|
66 | - the signal handling pipe wasn't always initialised under windows |
|
|
67 | (analysed by lekma). |
|
|
68 | - changed minimum glibc requirement from glibc 2.9 to 2.7, for |
|
|
69 | signalfd. |
|
|
70 | - add missing string.h include (Denis F. Latypoff). |
|
|
71 | - only replace ev_stat.prev when we detect an actual difference, |
|
|
72 | so prev is (almost) always different to attr. this might |
|
|
73 | have caused the probems with 04_stat.t. |
|
|
74 | - add ev::timer->remaining () method to C++ API. |
|
|
75 | |
|
|
76 | 3.8 Sun Aug 9 14:30:45 CEST 2009 |
|
|
77 | - incompatible change: do not necessarily reset signal handler |
|
|
78 | to SIG_DFL when a sighandler is stopped. |
|
|
79 | - ev_default_destroy did not properly free or zero some members, |
|
|
80 | potentially causing crashes and memory corruption on repated |
|
|
81 | ev_default_destroy/ev_default_loop calls. |
|
|
82 | - take advantage of signalfd on GNU/Linux systems. |
|
|
83 | - document that the signal mask might be in an unspecified |
|
|
84 | state when using libev's signal handling. |
|
|
85 | - take advantage of some GNU/Linux calls to set cloexec/nonblock |
|
|
86 | on fd creation, to avoid race conditions. |
|
|
87 | |
|
|
88 | 3.7 Fri Jul 17 16:36:32 CEST 2009 |
|
|
89 | - ev_unloop and ev_loop wrongly used a global variable to exit loops, |
|
|
90 | instead of using a per-loop variable (bug caught by accident...). |
|
|
91 | - the ev_set_io_collect_interval interpretation has changed. |
|
|
92 | - add new functionality: ev_set_userdata, ev_userdata, |
|
|
93 | ev_set_invoke_pending_cb, ev_set_loop_release_cb, |
|
|
94 | ev_invoke_pending, ev_pending_count, together with a long example |
|
|
95 | about thread locking. |
|
|
96 | - add ev_timer_remaining (as requested by Denis F. Latypoff). |
|
|
97 | - add ev_loop_depth. |
|
|
98 | - calling ev_unloop in fork/prepare watchers will no longer poll |
|
|
99 | for new events. |
|
|
100 | - Denis F. Latypoff corrected many typos in example code snippets. |
|
|
101 | - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double- |
|
|
102 | check that the syscall number is available before trying to |
|
|
103 | use it (reported by ry@tinyclouds). |
|
|
104 | - use GetSystemTimeAsFileTime instead of _timeb on windows, for |
|
|
105 | slightly higher accuracy. |
|
|
106 | - properly declare ev_loop_verify and ev_now_update even when |
|
|
107 | !EV_MULTIPLICITY. |
|
|
108 | - do not compile in any priority code when EV_MAXPRI == EV_MINPRI. |
|
|
109 | - support EV_MINIMAL==2 for a reduced API. |
|
|
110 | - actually 0-initialise struct sigaction when installing signals. |
|
|
111 | - add section on hibernate and stopped processes to ev_timer docs. |
|
|
112 | |
|
|
113 | 3.6 Tue Apr 28 02:49:30 CEST 2009 |
|
|
114 | - multiple timers becoming ready within an event loop iteration |
|
|
115 | will be invoked in the "correct" order now. |
|
|
116 | - do not leave the event loop early just because we have no active |
|
|
117 | watchers, fixing a problem when embedding a kqueue loop |
|
|
118 | that has active kernel events but no registered watchers |
|
|
119 | (reported by blacksand blacksand). |
|
|
120 | - correctly zero the idx values for arrays, so destroying and |
|
|
121 | reinitialising the default loop actually works (patch by |
|
|
122 | Malek Hadj-Ali). |
|
|
123 | - implement ev_suspend and ev_resume. |
|
|
124 | - new EV_CUSTOM revents flag for use by applications. |
|
|
125 | - add documentation section about priorites. |
|
|
126 | - add a glossary to the dcoumentation. |
|
|
127 | - extend the ev_fork description slightly. |
|
|
128 | - optimize a jump out of call_pending. |
|
|
129 | |
|
|
130 | 3.53 Sun Feb 15 02:38:20 CET 2009 |
|
|
131 | - fix a bug in event pipe creation on win32 that would cause a |
|
|
132 | failed assertion on event loop creation (patch by Malek Hadj-Ali). |
|
|
133 | - probe for CLOCK_REALTIME support at runtime as well and fall |
|
|
134 | back to gettimeofday if there is an error, to support older |
|
|
135 | operating systems with newer header files/libraries. |
|
|
136 | - prefer gettimeofday over clock_gettime with USE_CLOCK_SYSCALL |
|
|
137 | (default most everywhere), otherwise not. |
|
|
138 | |
|
|
139 | 3.52 Wed Jan 7 21:43:02 CET 2009 |
|
|
140 | - fix compilation of select backend in fd_set mode when NFDBITS is |
|
|
141 | missing (to get it to compile on QNX, reported by Rodrigo Campos). |
|
|
142 | - better select-nfds handling when select backend is in fd_set mode. |
|
|
143 | - diagnose fd_set overruns when select backend is in fd_set mode. |
|
|
144 | - due to a thinko, instead of disabling everything but |
|
|
145 | select on the borked OS X platform, everything but select was |
|
|
146 | allowed (reported by Emanuele Giaquinta). |
|
|
147 | - actually verify that local and remote port are matching in |
|
|
148 | libev's socketpair emulation, which makes denial-of-service |
|
|
149 | attacks harder (but not impossible - it's windows). Make sure |
|
|
150 | it even works under vista, which thinks that getpeer/sockname |
|
|
151 | should return fantasy port numbers. |
|
|
152 | - include "libev" in all assertion messages for potentially |
|
|
153 | clearer diagnostics. |
|
|
154 | - event_get_version (libevent compatibility) returned |
|
|
155 | a useless string instead of the expected version string |
|
|
156 | (patch by W.C.A. Wijngaards). |
|
|
157 | |
|
|
158 | 3.51 Wed Dec 24 23:00:11 CET 2008 |
|
|
159 | - fix a bug where an inotify watcher was added twice, causing |
|
|
160 | freezes on hash collisions (reported and analysed by Graham Leggett). |
|
|
161 | - new config symbol, EV_USE_CLOCK_SYSCALL, to make libev use |
|
|
162 | a direct syscall - slower, but no dependency on librt et al. |
|
|
163 | - assume negative return values != -1 signals success of port_getn |
|
|
164 | (http://cvs.epicsol.org/cgi/viewcvs.cgi/epic5/source/newio.c?rev=1.52) |
|
|
165 | (no known failure reports, but it doesn't hurt). |
|
|
166 | - fork detection in ev_embed now stops and restarts the watcher |
|
|
167 | automatically. |
|
|
168 | - EXPERIMENTAL: default the method to operator () in ev++.h, |
|
|
169 | to make it nicer to use functors (requested by Benedek László). |
|
|
170 | - fixed const object callbacks in ev++.h. |
|
|
171 | - replaced loop_ref argument of watcher.set (loop) by a direct |
|
|
172 | ev_loop * in ev++.h, to avoid clashes with functor patch. |
|
|
173 | - do not try to watch the empty string via inotify. |
|
|
174 | - inotify watchers could be leaked under certain circumstances. |
|
|
175 | - OS X 10.5 is actually even more broken than earlier versions, |
|
|
176 | so fall back to select on that piece of garbage. |
|
|
177 | - fixed some weirdness in the ev_embed documentation. |
|
|
178 | |
|
|
179 | 3.49 Wed Nov 19 11:26:53 CET 2008 |
|
|
180 | - ev_stat watchers will now use inotify as a mere hint on |
|
|
181 | kernels <2.6.25, or if the filesystem is not in the |
|
|
182 | "known to be good" list. |
|
|
183 | - better mingw32 compatibility (it's not as borked as native win32) |
|
|
184 | (analysed by Roger Pack). |
|
|
185 | - include stdio.h in the example program, as too many people are |
|
|
186 | confused by the weird C language otherwise. I guess the next thing |
|
|
187 | I get told is that the "..." ellipses in the examples don't compile |
|
|
188 | with their C compiler. |
|
|
189 | |
|
|
190 | 3.48 Thu Oct 30 09:02:37 CET 2008 |
|
|
191 | - further optimise away the EPOLL_CTL_ADD/MOD combo in the epoll |
|
|
192 | backend by assuming the kernel event mask hasn't changed if |
|
|
193 | ADD fails with EEXIST. |
|
|
194 | - work around spurious event notification bugs in epoll by using |
|
|
195 | a 32-bit generation counter. recreate kernel state if we receive |
|
|
196 | spurious notifications or unwanted events. this is very costly, |
|
|
197 | but I didn't come up with this horrible design. |
|
|
198 | - use memset to initialise most arrays now and do away with the |
|
|
199 | init functions. |
|
|
200 | - expand time-out strategies into a "Be smart about timeouts" section. |
|
|
201 | - drop the "struct" from all ev_watcher declarations in the |
|
|
202 | documentation and did other clarifications (yeah, it was a mistake |
|
|
203 | to have a struct AND a function called ev_loop). |
|
|
204 | - fix a bug where ev_default would not initialise the default |
|
|
205 | loop again after it was destroyed with ev_default_destroy. |
|
|
206 | - rename syserr to ev_syserr to avoid name clashes when embedding, |
|
|
207 | do similar changes for event.c. |
|
|
208 | |
|
|
209 | 3.45 Tue Oct 21 21:59:26 CEST 2008 |
|
|
210 | - disable inotify usage on linux <2.6.25, as it is broken |
|
|
211 | (reported by Yoann Vandoorselaere). |
|
|
212 | - ev_stat errornously would try to add inotify watchers |
|
|
213 | even when inotify wasn't available (this should only |
|
|
214 | have a performance impact). |
|
|
215 | - ev_once now passes both timeout and io to the callback if both |
|
|
216 | occur concurrently, instead of giving timeouts precedence. |
|
|
217 | - disable EV_USE_INOTIFY when sys/inotify.h is too old. |
|
|
218 | |
|
|
219 | 3.44 Mon Sep 29 05:18:39 CEST 2008 |
|
|
220 | - embed watchers now automatically invoke ev_loop_fork on the |
|
|
221 | embedded loop when the parent loop forks. |
|
|
222 | - new function: ev_now_update (loop). |
|
|
223 | - verify_watcher was not marked static. |
|
|
224 | - improve the "associating..." manpage section. |
|
|
225 | - documentation tweaks here and there. |
|
|
226 | |
|
|
227 | 3.43 Sun Jul 6 05:34:41 CEST 2008 |
|
|
228 | - include more include files on windows to get struct _stati64 |
|
|
229 | (reported by Chris Hulbert, but doesn't quite fix his issue). |
|
|
230 | - add missing #include <io.h> in ev.c on windows (reported by |
|
|
231 | Matt Tolton). |
|
|
232 | |
|
|
233 | 3.42 Tue Jun 17 12:12:07 CEST 2008 |
|
|
234 | - work around yet another windows bug: FD_SET actually adds fd's |
|
|
235 | multiple times to the fd_*SET*, despite official MSN docs claiming |
|
|
236 | otherwise. Reported and well-analysed by Matt Tolton. |
|
|
237 | - define NFDBITS to 0 when EV_SELECT_IS_WINSOCKET to make it compile |
|
|
238 | (reported any analysed by Chris Hulbert). |
|
|
239 | - fix a bug in ev_ebadf (this function is only used to catch |
|
|
240 | programming errors in the libev user). reported by Matt Tolton. |
|
|
241 | - fix a bug in fd_intern on win32 (could lead to compile errors |
|
|
242 | under some circumstances, but would work correctly if it compiles). |
|
|
243 | reported by Matt Tolton. |
|
|
244 | - (try to) work around missing lstat on windows. |
| 3 | - pass in the write fd set as except fd set under windows. windows |
245 | - pass in the write fd set as except fd set under windows. windows |
| 4 | is so uncontrollably lame that it requires this. this means that |
246 | is so uncontrollably lame that it requires this. this means that |
| 5 | switching off oobinline is not supported (but tcp/ip doesn't |
247 | switching off oobinline is not supported (but tcp/ip doesn't |
| 6 | have oob, so that would be stupid anyways. |
248 | have oob, so that would be stupid anyways. |
|
|
249 | - use posix module symbol to auto-detect monotonic clock presence |
|
|
250 | and some other default values. |
| 7 | |
251 | |
| 8 | 3.41 Fri May 23 18:42:54 CEST 2008 |
252 | 3.41 Fri May 23 18:42:54 CEST 2008 |
| 9 | - work around an obscure bug in winsocket select: if you |
253 | - work around an obscure bug in winsocket select: if you |
| 10 | provide only empty fd sets then select returns WSAEINVAL. how sucky. |
254 | provide only empty fd sets then select returns WSAEINVAL. how sucky. |
| 11 | - improve timer scheduling stability and reduce use of time_epsilon. |
255 | - improve timer scheduling stability and reduce use of time_epsilon. |
