/src/libev/Changes

  1Revision history for libev, a high-performance and full-featured event loop.
  2
  34.03 Tue Jan 11 14:37:25 CET 2011
  4	- officially support polling files with all backends.
  5	- support files, /dev/zero etc. the same way as select in the epoll
  6          backend, by generating events on our own.
  7        - ports backend: work around solaris bug 6874410 and many related ones
  8          (EINTR, maybe more), with no performance loss (note that the solaris
  9          bug report is actually wrong, reality is far more bizarre and broken
 10          than that).
 11	- define EV_READ/EV_WRITE as macros in event.h, as some programs use
 12          #ifdef to test for them.
 13        - new (experimental) function: ev_feed_signal.
 14        - new (to become default) EVFLAG_NOSIGMASK flag.
 15        - new EVBACKEND_MASK symbol.
 16        - updated COMMON IDIOMS SECTION.
 17
 184.01 Fri Nov  5 21:51:29 CET 2010
 19        - automake fucked it up, apparently, --add-missing -f is not quite enough
 20          to make it update its files, so 4.00 didn't install ev++.h and
 21          event.h on make install. grrr.
 22        - ev_loop(count|depth) didn't return anything (Robin Haberkorn).
 23        - change EV_UNDEF to 0xffffffff to silence some overzealous compilers.
 24        - use "(libev) " prefix for all libev error messages now.
 25
 264.00 Mon Oct 25 12:32:12 CEST 2010
 27	- "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading.
 28	- ev_embed_stop did not correctly stop the watcher (very good
 29          testcase by Vladimir Timofeev).
 30        - ev_run will now always update the current loop time - it erroneously
 31          didn't when idle watchers were active, causing timers not to fire.
 32        - fix a bug where a timeout of zero caused the timer not to fire
 33          in the libevent emulation (testcase by Péter Szabó).
 34	- applied win32 fixes by Michael Lenaghan (also James Mansion).
 35	- replace EV_MINIMAL by EV_FEATURES.
 36        - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it
 37          seems the former is *much* faster than the latter.
 38        - linux kernel version detection (for inotify bug workarounds)
 39          did not work properly.
 40        - reduce the number of spurious wake-ups with the ports backend.
 41        - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu).
 42        - do async init within ev_async_start, not ev_async_set, which avoids
 43          an API quirk where the set function must be called in the C++ API
 44          even when there is nothing to set.
 45        - add (undocumented) EV_ENABLE when adding events with kqueue,
 46          this might help with OS X, which seems to need it despite documenting
 47          not to need it (helpfully pointed out by Tilghman Lesher).
 48        - do not use poll by default on freebsd, it's broken (what isn't
 49          on freebsd...).
 50        - allow to embed epoll on kernels >= 2.6.32.
 51        - configure now prepends -O3, not appends it, so one can still
 52          override it.
 53        - ev.pod: greatly expanded the portability section, added a porting
 54          section, a description of watcher states and made lots of minor fixes.
 55        - disable poll backend on AIX, the poll header spams the namespace
 56          and it's not worth working around dead platforms (reported
 57          and analyzed by Aivars Kalvans).
 58        - improve header file compatibility of the standalone eventfd code
 59          in an obscure case.
 60        - implement EV_AVOID_STDIO option.
 61        - do not use sscanf to parse linux version number (smaller, faster,
 62          no sscanf dependency).
 63        - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings.
 64        - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs.
 65        - add section on accept() problems to the manpage.
 66        - rename EV_TIMEOUT to EV_TIMER.
 67        - rename ev_loop_count/depth/verify/loop/unloop.
 68        - remove ev_default_destroy and ev_default_fork.
 69        - switch to two-digit minor version.
 70        - work around an apparent gentoo compiler bug.
 71        - define _DARWIN_UNLIMITED_SELECT. just so.
 72        - use enum instead of #define for most constants.
 73        - improve compatibility to older C++ compilers.
 74        - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now
 75          default arguments when compiled as C++.
 76        - enable automake dependency tracking.
 77        - ev_loop_new no longer leaks memory when loop creation failed.
 78        - new ev_cleanup watcher type.
 79
 803.9  Thu Dec 31 07:59:59 CET 2009
 81	- signalfd is no longer used by default and has to be requested
 82          explicitly - this means that easy to catch bugs become hard to
 83          catch race conditions, but the users have spoken.
 84        - point out the unspecified signal mask in the documentation, and
 85          that this is a race condition regardless of EV_SIGNALFD.
 86	- backport inotify code to C89.
 87        - inotify file descriptors could leak into child processes.
 88        - ev_stat watchers could keep an errornous extra ref on the loop,
 89          preventing exit when unregistering all watchers (testcases
 90          provided by ry@tinyclouds.org).
 91        - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
 92          symbols to make it easier for apps to do their own fd management.
 93        - support EV_IDLE_ENABLE being disabled in ev++.h
 94          (patch by Didier Spezia).
 95        - take advantage of inotify_init1, if available, to set cloexec/nonblock
 96          on fd creation, to avoid races.
 97        - the signal handling pipe wasn't always initialised under windows
 98          (analysed by lekma).
 99        - changed minimum glibc requirement from glibc 2.9 to 2.7, for
100          signalfd.
101        - add missing string.h include (Denis F. Latypoff).
102        - only replace ev_stat.prev when we detect an actual difference,
103          so prev is (almost) always different to attr. this might
104          have caused the problems with 04_stat.t.
105        - add ev::timer->remaining () method to C++ API.
106
1073.8  Sun Aug  9 14:30:45 CEST 2009
108	- incompatible change: do not necessarily reset signal handler
109          to SIG_DFL when a sighandler is stopped.
110        - ev_default_destroy did not properly free or zero some members,
111          potentially causing crashes and memory corruption on repeated
112          ev_default_destroy/ev_default_loop calls.
113	- take advantage of signalfd on GNU/Linux systems.
114	- document that the signal mask might be in an unspecified
115          state when using libev's signal handling.
116        - take advantage of some GNU/Linux calls to set cloexec/nonblock
117          on fd creation, to avoid race conditions.
118
1193.7  Fri Jul 17 16:36:32 CEST 2009
120	- ev_unloop and ev_loop wrongly used a global variable to exit loops,
121          instead of using a per-loop variable (bug caught by accident...).
122	- the ev_set_io_collect_interval interpretation has changed.
123        - add new functionality: ev_set_userdata, ev_userdata,
124          ev_set_invoke_pending_cb, ev_set_loop_release_cb,
125          ev_invoke_pending, ev_pending_count, together with a long example
126          about thread locking.
127        - add ev_timer_remaining (as requested by Denis F. Latypoff).
128        - add ev_loop_depth.
129        - calling ev_unloop in fork/prepare watchers will no longer poll
130          for new events.
131	- Denis F. Latypoff corrected many typos in example code snippets.
132        - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double-
133          check that the syscall number is available before trying to
134          use it (reported by ry@tinyclouds).
135        - use GetSystemTimeAsFileTime instead of _timeb on windows, for
136          slightly higher accuracy.
137        - properly declare ev_loop_verify and ev_now_update even when
138          !EV_MULTIPLICITY.
139        - do not compile in any priority code when EV_MAXPRI == EV_MINPRI.
140        - support EV_MINIMAL==2 for a reduced API.
141        - actually 0-initialise struct sigaction when installing signals.
142        - add section on hibernate and stopped processes to ev_timer docs.
143
1443.6  Tue Apr 28 02:49:30 CEST 2009
145	- multiple timers becoming ready within an event loop iteration
146          will be invoked in the "correct" order now.
147	- do not leave the event loop early just because we have no active
148          watchers, fixing a problem when embedding a kqueue loop
149          that has active kernel events but no registered watchers
150          (reported by blacksand blacksand).
151	- correctly zero the idx values for arrays, so destroying and
152          reinitialising the default loop actually works (patch by
153          Malek Hadj-Ali).
154        - implement ev_suspend and ev_resume.
155        - new EV_CUSTOM revents flag for use by applications.
156        - add documentation section about priorites.
157        - add a glossary to the dcoumentation.
158        - extend the ev_fork description slightly.
159        - optimize a jump out of call_pending.
160
1613.53 Sun Feb 15 02:38:20 CET 2009
162	- fix a bug in event pipe creation on win32 that would cause a
163          failed assertion on event loop creation (patch by Malek Hadj-Ali).
164	- probe for CLOCK_REALTIME support at runtime as well and fall
165          back to gettimeofday if there is an error, to support older
166          operating systems with newer header files/libraries.
167        - prefer gettimeofday over clock_gettime with USE_CLOCK_SYSCALL
168          (default most everywhere), otherwise not.
169
1703.52 Wed Jan  7 21:43:02 CET 2009
171	- fix compilation of select backend in fd_set mode when NFDBITS is
172          missing (to get it to compile on QNX, reported by Rodrigo Campos).
173        - better select-nfds handling when select backend is in fd_set mode.
174        - diagnose fd_set overruns when select backend is in fd_set mode.
175        - due to a thinko, instead of disabling everything but
176          select on the borked OS X platform, everything but select was
177          allowed (reported by Emanuele Giaquinta).
178        - actually verify that local and remote port are matching in
179          libev's socketpair emulation, which makes denial-of-service
180          attacks harder (but not impossible - it's windows). Make sure
181          it even works under vista, which thinks that getpeer/sockname
182          should return fantasy port numbers.
183        - include "libev" in all assertion messages for potentially
184          clearer diagnostics.
185        - event_get_version (libevent compatibility) returned
186          a useless string instead of the expected version string
187          (patch by W.C.A. Wijngaards).
188
1893.51 Wed Dec 24 23:00:11 CET 2008
190        - fix a bug where an inotify watcher was added twice, causing
191          freezes on hash collisions (reported and analysed by Graham Leggett).
192	- new config symbol, EV_USE_CLOCK_SYSCALL, to make libev use
193          a direct syscall - slower, but no dependency on librt et al.
194        - assume negative return values != -1 signals success of port_getn
195          (http://cvs.epicsol.org/cgi/viewcvs.cgi/epic5/source/newio.c?rev=1.52)
196          (no known failure reports, but it doesn't hurt).
197        - fork detection in ev_embed now stops and restarts the watcher
198          automatically.
199        - EXPERIMENTAL: default the method to operator () in ev++.h,
200          to make it nicer to use functors (requested by Benedek László).
201        - fixed const object callbacks in ev++.h.
202        - replaced loop_ref argument of watcher.set (loop) by a direct
203          ev_loop * in ev++.h, to avoid clashes with functor patch.
204        - do not try to watch the empty string via inotify.
205        - inotify watchers could be leaked under certain circumstances.
206        - OS X 10.5 is actually even more broken than earlier versions,
207          so fall back to select on that piece of garbage.
208        - fixed some weirdness in the ev_embed documentation.
209
2103.49 Wed Nov 19 11:26:53 CET 2008
211	- ev_stat watchers will now use inotify as a mere hint on
212          kernels <2.6.25, or if the filesystem is not in the
213          "known to be good" list.
214        - better mingw32 compatibility (it's not as borked as native win32)
215          (analysed by Roger Pack).
216        - include stdio.h in the example program, as too many people are
217          confused by the weird C language otherwise. I guess the next thing
218          I get told is that the "..." ellipses in the examples don't compile
219          with their C compiler.
220
2213.48 Thu Oct 30 09:02:37 CET 2008
222	- further optimise away the EPOLL_CTL_ADD/MOD combo in the epoll
223          backend by assuming the kernel event mask hasn't changed if
224          ADD fails with EEXIST.
225        - work around spurious event notification bugs in epoll by using
226          a 32-bit generation counter. recreate kernel state if we receive
227          spurious notifications or unwanted events. this is very costly,
228          but I didn't come up with this horrible design.
229        - use memset to initialise most arrays now and do away with the
230          init functions.
231        - expand time-out strategies into a "Be smart about timeouts" section.
232        - drop the "struct" from all ev_watcher declarations in the
233          documentation and did other clarifications (yeah, it was a mistake
234          to have a struct AND a function called ev_loop).
235	- fix a bug where ev_default would not initialise the default
236          loop again after it was destroyed with ev_default_destroy.
237        - rename syserr to ev_syserr to avoid name clashes when embedding,
238          do similar changes for event.c.
239
2403.45 Tue Oct 21 21:59:26 CEST 2008
241	- disable inotify usage on linux <2.6.25, as it is broken
242          (reported by Yoann Vandoorselaere).
243        - ev_stat erroneously would try to add inotify watchers
244          even when inotify wasn't available (this should only
245          have a performance impact).
246	- ev_once now passes both timeout and io to the callback if both
247          occur concurrently, instead of giving timeouts precedence.
248	- disable EV_USE_INOTIFY when sys/inotify.h is too old.
249
2503.44 Mon Sep 29 05:18:39 CEST 2008
251	- embed watchers now automatically invoke ev_loop_fork on the
252          embedded loop when the parent loop forks.
253	- new function: ev_now_update (loop).
254	- verify_watcher was not marked static.
255        - improve the "associating..." manpage section.
256        - documentation tweaks here and there.
257
2583.43 Sun Jul  6 05:34:41 CEST 2008
259	- include more include files on windows to get struct _stati64
260          (reported by Chris Hulbert, but doesn't quite fix his issue).
261	- add missing #include <io.h> in ev.c on windows (reported by
262          Matt Tolton).
263
2643.42 Tue Jun 17 12:12:07 CEST 2008
265	- work around yet another windows bug: FD_SET actually adds fd's
266          multiple times to the fd_*SET*, despite official MSN docs claiming
267          otherwise. Reported and well-analysed by Matt Tolton.
268	- define NFDBITS to 0 when EV_SELECT_IS_WINSOCKET to make it compile
269          (reported any analysed by Chris Hulbert).
270        - fix a bug in ev_ebadf (this function is only used to catch
271          programming errors in the libev user). reported by Matt Tolton.
272        - fix a bug in fd_intern on win32 (could lead to compile errors
273          under some circumstances, but would work correctly if it compiles).
274          reported by Matt Tolton.
275        - (try to) work around missing lstat on windows.
276	- pass in the write fd set as except fd set under windows. windows
277          is so uncontrollably lame that it requires this. this means that
278          switching off oobinline is not supported (but tcp/ip doesn't
279          have oob, so that would be stupid anyways.
280        - use posix module symbol to auto-detect monotonic clock presence
281          and some other default values.
282
2833.41 Fri May 23 18:42:54 CEST 2008
284	- work around an obscure bug in winsocket select: if you
285          provide only empty fd sets then select returns WSAEINVAL. how sucky.
286        - improve timer scheduling stability and reduce use of time_epsilon.
287        - use 1-based 2-heap for EV_MINIMAL, simplifies code, reduces
288          codesize and makes for better cache-efficiency.
289        - use 3-based 4-heap for !EV_MINIMAL. this makes better use
290          of cpu cache lines and gives better growth behaviour than
291          2-based heaps.
292        - cache timestamp within heap for !EV_MINIMAL, to avoid random
293          memory accesses.
294        - document/add EV_USE_4HEAP and EV_HEAP_CACHE_AT.
295        - fix a potential aliasing issue in ev_timer_again.
296        - add/document ev_periodic_at, retract direct access to ->at.
297        - improve ev_stat docs.
298        - add portability requirements section.
299	- fix manpage headers etc.
300        - normalise WSA error codes to lower range on windows.
301        - add consistency check code that can be called automatically
302          or on demand to check for internal structures (ev_loop_verify).
303
3043.31 Wed Apr 16 20:45:04 CEST 2008
305	- added last minute fix for ev_poll.c by Brandon Black.
306
3073.3  Wed Apr 16 19:04:10 CEST 2008
308        - event_base_loopexit should return 0 on success
309          (W.C.A. Wijngaards).
310	- added linux eventfd support.
311        - try to autodetect epoll and inotify support
312          by libc header version if not using autoconf.
313        - new symbols: EV_DEFAULT_UC and EV_DEFAULT_UC_.
314        - declare functions defined in ev.h as inline if
315          C99 or gcc are available.
316        - enable inlining with gcc versions 2 and 3.
317        - work around broken poll implementations potentially
318          not clearing revents field in ev_poll (Brandon Black)
319          (no such systems are known at this time).
320        - work around a bug in realloc on openbsd and darwin,
321          also makes the erroneous valgrind complaints
322          go away (noted by various people).
323        - fix ev_async_pending, add c++ wrapper for ev_async
324          (based on patch sent by Johannes Deisenhofer).
325        - add sensible set method to ev::embed.
326        - made integer constants type int in ev.h.
327
3283.2  Wed Apr  2 17:11:19 CEST 2008
329	- fix a 64 bit overflow issue in the select backend,
330          by using fd_mask instead of int for the mask.
331        - rename internal sighandler to avoid clash with very old perls.
332        - entering ev_loop will not clear the ONESHOT or NONBLOCKING
333          flags of any outer loops anymore.
334        - add ev_async_pending.
335
3363.1  Thu Mar 13 13:45:22 CET 2008
337	- implement ev_async watchers.
338        - only initialise signal pipe on demand.
339	- make use of sig_atomic_t configurable.
340        - improved documentation.
341
3423.0  Mon Jan 28 13:14:47 CET 2008
343	- API/ABI bump to version 3.0.
344	- ev++.h includes "ev.h" by default now, not <ev.h>.
345	- slightly improved documentation.
346	- speed up signal detection after a fork.
347        - only optionally return trace status changed in ev_child
348          watchers.
349        - experimental (and undocumented) loop wrappers for ev++.h.
350
3512.01 Tue Dec 25 08:04:41 CET 2007
352	- separate Changes file.
353	- fix ev_path_set => ev_stat_set typo.
354        - remove event_compat.h from the libev tarball.
355        - change how include files are found.
356        - doc updates.
357        - update licenses, explicitly allow for GPL relicensing.
358
3592.0  Sat Dec 22 17:47:03 CET 2007
360        - new ev_sleep, ev_set_(io|timeout)_collect_interval.
361        - removed epoll from embeddable fd set.
362        - fix embed watchers.
363	- renamed ev_embed.loop to other.
364	- added exported Symbol tables.
365        - undefine member wrapper macros at the end of ev.c.
366        - respect EV_H in ev++.h.
367
3681.86 Tue Dec 18 02:36:57 CET 2007
369	- fix memleak on loop destroy (not relevant for perl).
370
3711.85 Fri Dec 14 20:32:40 CET 2007
372        - fix some aliasing issues w.r.t. timers and periodics
373          (not relevant for perl).
374
375(for historic versions refer to EV/Changes, found in the Perl interface)
376
3770.1  Wed Oct 31 21:31:48 CET 2007
378	- original version; hacked together in <24h.
379