Mirror of :pserver:anonymous@cvs.schmorp.de/schmorpforge libev http://software.schmorp.de/pkg/libev.html
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

599 lines
31 KiB

14 years ago
1 year ago
2 years ago
2 years ago
3 years ago
4 years ago
5 years ago
6 years ago
7 years ago
7 years ago
8 years ago
8 years ago
8 years ago
8 years ago
9 years ago
10 years ago
10 years ago
10 years ago
10 years ago
11 years ago
11 years ago
11 years ago
11 years ago
11 years ago
11 years ago
12 years ago
10 years ago
12 years ago
11 years ago
12 years ago
11 years ago
12 years ago
12 years ago
12 years ago
12 years ago
12 years ago
12 years ago
12 years ago
10 years ago
8 years ago
12 years ago
13 years ago
13 years ago
13 years ago
13 years ago
13 years ago
13 years ago
13 years ago
11 years ago
13 years ago
13 years ago
13 years ago
11 years ago
11 years ago
14 years ago
  1. Revision history for libev, a high-performance and full-featured event loop.
  2. TODO: revisit 59.x timer in the light of modern powersaving
  3. TODO: ^ maybe just increase it when timerfd, modern linux only thing that counts?
  4. TODO: document EV_TSTAMP_T
  5. - the 4.31 timerfd code wrongly changes the priority of the signal
  6. fd watcher, which is usually harmless unless signal fds are
  7. also used (found via cpan tester service).
  8. - support a lot more "uncommon" compile time configurations,
  9. such as ev_embed enabled but ev_timer disabled.
  10. - the linux aio backend is no longer compiled in by default.
  11. 4.31 Fri Dec 20 21:58:29 CET 2019
  12. - handle backends with minimum wait time a bit better by not
  13. waiting in the presence of already-expired timers
  14. (behaviour reported by Felipe Gasper).
  15. - new feature: use timerfd to detect timejumps quickly,
  16. can be disabled with the new EVFLAG_NOTIMERFD loop flag.
  17. - document EV_USE_SIGNALFD feature macro.
  18. 4.30 (EV only)
  19. - change non-autoconf test for __kernel_rwf_t by testing
  20. LINUX_VERSION_CODE, the most direct test I could find.
  21. - fix a bug in the io_uring backend that polled the wrong
  22. backend fd, causing it to not work in many cases.
  23. 4.29 (EV only)
  24. - add io uring autoconf and non-autoconf detection.
  25. - disable io_uring when some header files are too old.
  26. 4.28 (EV only)
  27. - linuxaio backend resulted in random memory corruption
  28. when loop is forked.
  29. - linuxaio backend might have tried to cancel an iocb
  30. multiple times (was unable to trigger this).
  31. - linuxaio backend now employs a generation counter to
  32. avoid handling spurious events from cancelled requests.
  33. - io_cancel can return EINTR, deal with it. also, assume
  34. io_submit also returns EINTR.
  35. - fix some other minor bugs in linuxaio backend.
  36. - ev_tstamp type can now be overriden by defining EV_TSTAMP_T.
  37. - cleanup: replace expect_true/false and noinline by their
  38. libecb counterparts.
  39. - move syscall infrastructure from ev_linuxaio.c to ev.c.
  40. - prepare io_uring integration.
  41. - tweak ev_floor.
  42. - epoll, poll, win32 Sleep and other places that use millisecond
  43. reslution now all try to round up times.
  44. - solaris port backend didn't compile.
  45. - abstract time constants into their macros, for more flexibility.
  46. 4.27 Thu Jun 27 22:43:44 CEST 2019
  47. - linux aio backend almost completely rewritten to work around its
  48. limitations.
  49. - linux aio backend now requires linux 4.19+.
  50. - epoll backend now mandatory for linux aio backend.
  51. - fail assertions more aggressively on invalid fd's detected
  52. in the event loop, do not just silently fd_kill in case of
  53. user error.
  54. - ev_io_start/ev_io_stop now verify the watcher fd using
  55. a syscall when EV_VERIFY is 2 or higher.
  56. 4.26 (EV only)
  57. - update to libecb 0x00010006.
  58. - new experimental linux aio backend (linux 4.18+).
  59. - removed redundant 0-ptr check in ev_once.
  60. - updated/extended ev_set_allocator documentation.
  61. - replaced EMPTY2 macro by array_needsize_noinit.
  62. - minor code cleanups.
  63. - epoll backend now uses epoll_create1 also after fork.
  64. 4.25 Fri Dec 21 07:49:20 CET 2018
  65. - INCOMPATIBLE CHANGE: EV_THROW was renamed to EV_NOEXCEPT
  66. (EV_THROW still provided) and now uses noexcept on C++11 or newer.
  67. - move the darwin select workaround higher in ev.c, as newer versions of
  68. darwin managed to break their broken select even more.
  69. - ANDROID => __ANDROID__ (reported by enh@google.com).
  70. - disable epoll_create1 on android because it has broken header files
  71. and google is unwilling to fix them (reported by enh@google.com).
  72. - avoid a minor compilation warning on win32.
  73. - c++: remove deprecated dynamic throw() specifications.
  74. - c++: improve the (unsupported) bad_loop exception class.
  75. - backport perl ev_periodic example to C, untested.
  76. - update libecb, biggets change is to include a memory fence
  77. in ECB_MEMORY_FENCE_RELEASE on x86/amd64.
  78. - minor autoconf/automake modernisation.
  79. 4.24 Wed Dec 28 05:19:55 CET 2016
  80. - bump version to 4.24, as the release tarball inexplicably
  81. didn't have the right version in ev.h, even though the cvs-tagged
  82. version did have the right one (reported by Ales Teska).
  83. 4.23 Wed Nov 16 18:23:41 CET 2016
  84. - move some declarations at the beginning to help certain retarded
  85. microsoft compilers, even though their documentation claims
  86. otherwise (reported by Ruslan Osmanov).
  87. 4.22 Sun Dec 20 22:11:50 CET 2015
  88. - when epoll detects unremovable fds in the fd set, rebuild
  89. only the epoll descriptor, not the signal pipe, to avoid
  90. SIGPIPE in ev_async_send. This doesn't solve it on fork,
  91. so document what needs to be done in ev_loop_fork
  92. (analyzed by Benjamin Mahler).
  93. - remove superfluous sys/timeb.h include on win32
  94. (analyzed by Jason Madden).
  95. - updated libecb.
  96. 4.20 Sat Jun 20 13:01:43 CEST 2015
  97. - prefer noexcept over throw () with C++ 11.
  98. - update ecb.h due to incompatibilities with c11.
  99. - fix a potential aliasing issue when reading and writing
  100. watcher callbacks.
  101. 4.19 Thu Sep 25 08:18:25 CEST 2014
  102. - ev.h wasn't valid C++ anymore, which tripped compilers other than
  103. clang, msvc or gcc (analyzed by Raphael 'kena' Poss). Unfortunately,
  104. C++ doesn't support typedefs for function pointers fully, so the affected
  105. declarations have to spell out the types each time.
  106. - when not using autoconf, tighten the check for clock_gettime and related
  107. functionality.
  108. 4.18 Fri Sep 5 17:55:26 CEST 2014
  109. - events on files were not always generated properly with the
  110. epoll backend (testcase by Assaf Inbal).
  111. - mark event pipe fd as cloexec after a fork (analyzed by Sami Farin).
  112. - (ecb) support m68k, m88k and sh (patch by Miod Vallat).
  113. - use a reasonable fallback for EV_NSIG instead of erroring out
  114. when we can't detect the signal set size.
  115. - in the absence of autoconf, do not use the clock syscall
  116. on glibc >= 2.17 (avoids the syscall AND -lrt on systems
  117. doing clock_gettime in userspace).
  118. - ensure extern "C" function pointers are used for externally-visible
  119. loop callbacks (not watcher callbacks yet).
  120. - (ecb) work around memory barriers and volatile apparently both being
  121. broken in visual studio 2008 and later (analysed and patch by Nicolas Noble).
  122. 4.15 Fri Mar 1 12:04:50 CET 2013
  123. - destroying a non-default loop would stop the global waitpid
  124. watcher (Denis Bilenko).
  125. - queueing pending watchers of higher priority from a watcher now invokes
  126. them in a timely fashion (reported by Denis Bilenko).
  127. - add throw() to all libev functions that cannot throw exceptions, for
  128. further code size decrease when compiling for C++.
  129. - add throw () to callbacks that must not throw exceptions (allocator,
  130. syserr, loop acquire/release, periodic reschedule cbs).
  131. - fix event_base_loop return code, add event_get_callback, event_base_new,
  132. event_base_get_method calls to improve libevent 1.x emulation and add
  133. some libevent 2.x functionality (based on a patch by Jeff Davey).
  134. - add more memory fences to fix a bug reported by Jeff Davey. Better
  135. be overfenced than underprotected.
  136. - ev_run now returns a boolean status (true meaning watchers are
  137. still active).
  138. - ev_once: undef EV_ERROR in ev_kqueue.c, to avoid clashing with
  139. libev's EV_ERROR (reported by 191919).
  140. - (ecb) add memory fence support for xlC (Darin McBride).
  141. - (ecb) add memory fence support for gcc-mips (Anton Kirilov).
  142. - (ecb) add memory fence support for gcc-alpha (Christian Weisgerber).
  143. - work around some kernels losing file descriptors by leaking
  144. the kqueue descriptor in the child.
  145. - work around linux inotify not reporting IN_ATTRIB changes for directories
  146. in many cases.
  147. - include sys/syscall.h instead of plain syscall.h.
  148. - check for io watcher loops in ev_verify, check for the most
  149. common reported usage bug in ev_io_start.
  150. - choose socket vs. WSASocket at compiletime using EV_USE_WSASOCKET.
  151. - always use WSASend/WSARecv directly on windows, hoping that this
  152. works in all cases (unlike read/write/send/recv...).
  153. - try to detect signals around a fork faster (test program by
  154. Denis Bilenko).
  155. - work around recent glibc versions that leak memory in realloc.
  156. - rename ev::embed::set to ev::embed::set_embed to avoid clashing
  157. the watcher base set (loop) method.
  158. - rewrite the async/signal pipe logic to always keep a valid fd, which
  159. simplifies (and hopefully correctifies :) the race checking
  160. on fork, at the cost of one extra fd.
  161. - add fat, msdos, jffs2, ramfs, ntfs and btrfs to the list of
  162. inotify-supporting filesystems.
  163. - move orig_CFLAGS assignment to after AC_INIT, as newer autoconf
  164. versions ignore it before
  165. (https://bugzilla.redhat.com/show_bug.cgi?id=908096).
  166. - add some untested android support.
  167. - enum expressions must be of type int (reported by Juan Pablo L).
  168. 4.11 Sat Feb 4 19:52:39 CET 2012
  169. - INCOMPATIBLE CHANGE: ev_timer_again now clears the pending status, as
  170. was documented already, but not implemented in the repeating case.
  171. - new compiletime symbols: EV_NO_SMP and EV_NO_THREADS.
  172. - fix a race where the workaround against the epoll fork bugs
  173. caused signals to not be handled anymore.
  174. - correct backend_fudge for most backends, and implement a windows
  175. specific workaround to avoid looping because we call both
  176. select and Sleep, both with different time resolutions.
  177. - document range and guarantees of ev_sleep.
  178. - document reasonable ranges for periodics interval and offset.
  179. - rename backend_fudge to backend_mintime to avoid future confusion :)
  180. - change the default periodic reschedule function to hopefully be more
  181. exact and correct even in corner cases or in the far future.
  182. - do not rely on -lm anymore: use it when available but use our
  183. own floor () if it is missing. This should make it easier to embed,
  184. as no external libraries are required.
  185. - strategically import macros from libecb and mark rarely-used functions
  186. as cache-cold (saving almost 2k code size on typical amd64 setups).
  187. - add Symbols.ev and Symbols.event files, that were missing.
  188. - fix backend_mintime value for epoll (was 1/1024, is 1/1000 now).
  189. - fix #3 "be smart about timeouts" to not "deadlock" when
  190. timeout == now, also improve the section overall.
  191. - avoid "AVOIDING FINISHING BEFORE RETURNING" idiom.
  192. - support new EV_API_STATIC mode to make all libev symbols
  193. static.
  194. - supply default CFLAGS of -g -O3 with gcc when original CFLAGS
  195. were empty.
  196. 4.04 Wed Feb 16 09:01:51 CET 2011
  197. - fix two problems in the native win32 backend, where reuse of fd's
  198. with different underlying handles caused handles not to be removed
  199. or added to the select set (analyzed and tested by Bert Belder).
  200. - do no rely on ceil() in ev_e?poll.c.
  201. - backport libev to HP-UX versions before 11 v3.
  202. - configure did not detect nanosleep and clock_gettime properly when
  203. they are available in the libc (as opposed to -lrt).
  204. 4.03 Tue Jan 11 14:37:25 CET 2011
  205. - officially support polling files with all backends.
  206. - support files, /dev/zero etc. the same way as select in the epoll
  207. backend, by generating events on our own.
  208. - ports backend: work around solaris bug 6874410 and many related ones
  209. (EINTR, maybe more), with no performance loss (note that the solaris
  210. bug report is actually wrong, reality is far more bizarre and broken
  211. than that).
  212. - define EV_READ/EV_WRITE as macros in event.h, as some programs use
  213. #ifdef to test for them.
  214. - new (experimental) function: ev_feed_signal.
  215. - new (to become default) EVFLAG_NOSIGMASK flag.
  216. - new EVBACKEND_MASK symbol.
  217. - updated COMMON IDIOMS SECTION.
  218. 4.01 Fri Nov 5 21:51:29 CET 2010
  219. - automake fucked it up, apparently, --add-missing -f is not quite enough
  220. to make it update its files, so 4.00 didn't install ev++.h and
  221. event.h on make install. grrr.
  222. - ev_loop(count|depth) didn't return anything (Robin Haberkorn).
  223. - change EV_UNDEF to 0xffffffff to silence some overzealous compilers.
  224. - use "(libev) " prefix for all libev error messages now.
  225. 4.00 Mon Oct 25 12:32:12 CEST 2010
  226. - "PORTING FROM LIBEV 3.X TO 4.X" (in ev.pod) is recommended reading.
  227. - ev_embed_stop did not correctly stop the watcher (very good
  228. testcase by Vladimir Timofeev).
  229. - ev_run will now always update the current loop time - it erroneously
  230. didn't when idle watchers were active, causing timers not to fire.
  231. - fix a bug where a timeout of zero caused the timer not to fire
  232. in the libevent emulation (testcase by Péter Szabó).
  233. - applied win32 fixes by Michael Lenaghan (also James Mansion).
  234. - replace EV_MINIMAL by EV_FEATURES.
  235. - prefer EPOLL_CTL_ADD over EPOLL_CTL_MOD in some more cases, as it
  236. seems the former is *much* faster than the latter.
  237. - linux kernel version detection (for inotify bug workarounds)
  238. did not work properly.
  239. - reduce the number of spurious wake-ups with the ports backend.
  240. - remove dependency on sys/queue.h on freebsd (patch by Vanilla Hsu).
  241. - do async init within ev_async_start, not ev_async_set, which avoids
  242. an API quirk where the set function must be called in the C++ API
  243. even when there is nothing to set.
  244. - add (undocumented) EV_ENABLE when adding events with kqueue,
  245. this might help with OS X, which seems to need it despite documenting
  246. not to need it (helpfully pointed out by Tilghman Lesher).
  247. - do not use poll by default on freebsd, it's broken (what isn't
  248. on freebsd...).
  249. - allow to embed epoll on kernels >= 2.6.32.
  250. - configure now prepends -O3, not appends it, so one can still
  251. override it.
  252. - ev.pod: greatly expanded the portability section, added a porting
  253. section, a description of watcher states and made lots of minor fixes.
  254. - disable poll backend on AIX, the poll header spams the namespace
  255. and it's not worth working around dead platforms (reported
  256. and analyzed by Aivars Kalvans).
  257. - improve header file compatibility of the standalone eventfd code
  258. in an obscure case.
  259. - implement EV_AVOID_STDIO option.
  260. - do not use sscanf to parse linux version number (smaller, faster,
  261. no sscanf dependency).
  262. - new EV_CHILD_ENABLE and EV_SIGNAL_ENABLE configurable settings.
  263. - update libev.m4 HAVE_CLOCK_SYSCALL test for newer glibcs.
  264. - add section on accept() problems to the manpage.
  265. - rename EV_TIMEOUT to EV_TIMER.
  266. - rename ev_loop_count/depth/verify/loop/unloop.
  267. - remove ev_default_destroy and ev_default_fork.
  268. - switch to two-digit minor version.
  269. - work around an apparent gentoo compiler bug.
  270. - define _DARWIN_UNLIMITED_SELECT. just so.
  271. - use enum instead of #define for most constants.
  272. - improve compatibility to older C++ compilers.
  273. - (experimental) ev_run/ev_default_loop/ev_break/ev_loop_new have now
  274. default arguments when compiled as C++.
  275. - enable automake dependency tracking.
  276. - ev_loop_new no longer leaks memory when loop creation failed.
  277. - new ev_cleanup watcher type.
  278. 3.9 Thu Dec 31 07:59:59 CET 2009
  279. - signalfd is no longer used by default and has to be requested
  280. explicitly - this means that easy to catch bugs become hard to
  281. catch race conditions, but the users have spoken.
  282. - point out the unspecified signal mask in the documentation, and
  283. that this is a race condition regardless of EV_SIGNALFD.
  284. - backport inotify code to C89.
  285. - inotify file descriptors could leak into child processes.
  286. - ev_stat watchers could keep an erroneous extra ref on the loop,
  287. preventing exit when unregistering all watchers (testcases
  288. provided by ry@tinyclouds.org).
  289. - implement EV_WIN32_HANDLE_TO_FD and EV_WIN32_CLOSE_FD configuration
  290. symbols to make it easier for apps to do their own fd management.
  291. - support EV_IDLE_ENABLE being disabled in ev++.h
  292. (patch by Didier Spezia).
  293. - take advantage of inotify_init1, if available, to set cloexec/nonblock
  294. on fd creation, to avoid races.
  295. - the signal handling pipe wasn't always initialised under windows
  296. (analysed by lekma).
  297. - changed minimum glibc requirement from glibc 2.9 to 2.7, for
  298. signalfd.
  299. - add missing string.h include (Denis F. Latypoff).
  300. - only replace ev_stat.prev when we detect an actual difference,
  301. so prev is (almost) always different to attr. this might
  302. have caused the problems with 04_stat.t.
  303. - add ev::timer->remaining () method to C++ API.
  304. 3.8 Sun Aug 9 14:30:45 CEST 2009
  305. - incompatible change: do not necessarily reset signal handler
  306. to SIG_DFL when a sighandler is stopped.
  307. - ev_default_destroy did not properly free or zero some members,
  308. potentially causing crashes and memory corruption on repeated
  309. ev_default_destroy/ev_default_loop calls.
  310. - take advantage of signalfd on GNU/Linux systems.
  311. - document that the signal mask might be in an unspecified
  312. state when using libev's signal handling.
  313. - take advantage of some GNU/Linux calls to set cloexec/nonblock
  314. on fd creation, to avoid race conditions.
  315. 3.7 Fri Jul 17 16:36:32 CEST 2009
  316. - ev_unloop and ev_loop wrongly used a global variable to exit loops,
  317. instead of using a per-loop variable (bug caught by accident...).
  318. - the ev_set_io_collect_interval interpretation has changed.
  319. - add new functionality: ev_set_userdata, ev_userdata,
  320. ev_set_invoke_pending_cb, ev_set_loop_release_cb,
  321. ev_invoke_pending, ev_pending_count, together with a long example
  322. about thread locking.
  323. - add ev_timer_remaining (as requested by Denis F. Latypoff).
  324. - add ev_loop_depth.
  325. - calling ev_unloop in fork/prepare watchers will no longer poll
  326. for new events.
  327. - Denis F. Latypoff corrected many typos in example code snippets.
  328. - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double-
  329. check that the syscall number is available before trying to
  330. use it (reported by ry@tinyclouds).
  331. - use GetSystemTimeAsFileTime instead of _timeb on windows, for
  332. slightly higher accuracy.
  333. - properly declare ev_loop_verify and ev_now_update even when
  334. !EV_MULTIPLICITY.
  335. - do not compile in any priority code when EV_MAXPRI == EV_MINPRI.
  336. - support EV_MINIMAL==2 for a reduced API.
  337. - actually 0-initialise struct sigaction when installing signals.
  338. - add section on hibernate and stopped processes to ev_timer docs.
  339. 3.6 Tue Apr 28 02:49:30 CEST 2009
  340. - multiple timers becoming ready within an event loop iteration
  341. will be invoked in the "correct" order now.
  342. - do not leave the event loop early just because we have no active
  343. watchers, fixing a problem when embedding a kqueue loop
  344. that has active kernel events but no registered watchers
  345. (reported by blacksand blacksand).
  346. - correctly zero the idx values for arrays, so destroying and
  347. reinitialising the default loop actually works (patch by
  348. Malek Hadj-Ali).
  349. - implement ev_suspend and ev_resume.
  350. - new EV_CUSTOM revents flag for use by applications.
  351. - add documentation section about priorities.
  352. - add a glossary to the documentation.
  353. - extend the ev_fork description slightly.
  354. - optimize a jump out of call_pending.
  355. 3.53 Sun Feb 15 02:38:20 CET 2009
  356. - fix a bug in event pipe creation on win32 that would cause a
  357. failed assertion on event loop creation (patch by Malek Hadj-Ali).
  358. - probe for CLOCK_REALTIME support at runtime as well and fall
  359. back to gettimeofday if there is an error, to support older
  360. operating systems with newer header files/libraries.
  361. - prefer gettimeofday over clock_gettime with USE_CLOCK_SYSCALL
  362. (default most everywhere), otherwise not.
  363. 3.52 Wed Jan 7 21:43:02 CET 2009
  364. - fix compilation of select backend in fd_set mode when NFDBITS is
  365. missing (to get it to compile on QNX, reported by Rodrigo Campos).
  366. - better select-nfds handling when select backend is in fd_set mode.
  367. - diagnose fd_set overruns when select backend is in fd_set mode.
  368. - due to a thinko, instead of disabling everything but
  369. select on the borked OS X platform, everything but select was
  370. allowed (reported by Emanuele Giaquinta).
  371. - actually verify that local and remote port are matching in
  372. libev's socketpair emulation, which makes denial-of-service
  373. attacks harder (but not impossible - it's windows). Make sure
  374. it even works under vista, which thinks that getpeer/sockname
  375. should return fantasy port numbers.
  376. - include "libev" in all assertion messages for potentially
  377. clearer diagnostics.
  378. - event_get_version (libevent compatibility) returned
  379. a useless string instead of the expected version string
  380. (patch by W.C.A. Wijngaards).
  381. 3.51 Wed Dec 24 23:00:11 CET 2008
  382. - fix a bug where an inotify watcher was added twice, causing
  383. freezes on hash collisions (reported and analysed by Graham Leggett).
  384. - new config symbol, EV_USE_CLOCK_SYSCALL, to make libev use
  385. a direct syscall - slower, but no dependency on librt et al.
  386. - assume negative return values != -1 signals success of port_getn
  387. (http://cvs.epicsol.org/cgi/viewcvs.cgi/epic5/source/newio.c?rev=1.52)
  388. (no known failure reports, but it doesn't hurt).
  389. - fork detection in ev_embed now stops and restarts the watcher
  390. automatically.
  391. - EXPERIMENTAL: default the method to operator () in ev++.h,
  392. to make it nicer to use functors (requested by Benedek László).
  393. - fixed const object callbacks in ev++.h.
  394. - replaced loop_ref argument of watcher.set (loop) by a direct
  395. ev_loop * in ev++.h, to avoid clashes with functor patch.
  396. - do not try to watch the empty string via inotify.
  397. - inotify watchers could be leaked under certain circumstances.
  398. - OS X 10.5 is actually even more broken than earlier versions,
  399. so fall back to select on that piece of garbage.
  400. - fixed some weirdness in the ev_embed documentation.
  401. 3.49 Wed Nov 19 11:26:53 CET 2008
  402. - ev_stat watchers will now use inotify as a mere hint on
  403. kernels <2.6.25, or if the filesystem is not in the
  404. "known to be good" list.
  405. - better mingw32 compatibility (it's not as borked as native win32)
  406. (analysed by Roger Pack).
  407. - include stdio.h in the example program, as too many people are
  408. confused by the weird C language otherwise. I guess the next thing
  409. I get told is that the "..." ellipses in the examples don't compile
  410. with their C compiler.
  411. 3.48 Thu Oct 30 09:02:37 CET 2008
  412. - further optimise away the EPOLL_CTL_ADD/MOD combo in the epoll
  413. backend by assuming the kernel event mask hasn't changed if
  414. ADD fails with EEXIST.
  415. - work around spurious event notification bugs in epoll by using
  416. a 32-bit generation counter. recreate kernel state if we receive
  417. spurious notifications or unwanted events. this is very costly,
  418. but I didn't come up with this horrible design.
  419. - use memset to initialise most arrays now and do away with the
  420. init functions.
  421. - expand time-out strategies into a "Be smart about timeouts" section.
  422. - drop the "struct" from all ev_watcher declarations in the
  423. documentation and did other clarifications (yeah, it was a mistake
  424. to have a struct AND a function called ev_loop).
  425. - fix a bug where ev_default would not initialise the default
  426. loop again after it was destroyed with ev_default_destroy.
  427. - rename syserr to ev_syserr to avoid name clashes when embedding,
  428. do similar changes for event.c.
  429. 3.45 Tue Oct 21 21:59:26 CEST 2008
  430. - disable inotify usage on linux <2.6.25, as it is broken
  431. (reported by Yoann Vandoorselaere).
  432. - ev_stat erroneously would try to add inotify watchers
  433. even when inotify wasn't available (this should only
  434. have a performance impact).
  435. - ev_once now passes both timeout and io to the callback if both
  436. occur concurrently, instead of giving timeouts precedence.
  437. - disable EV_USE_INOTIFY when sys/inotify.h is too old.
  438. 3.44 Mon Sep 29 05:18:39 CEST 2008
  439. - embed watchers now automatically invoke ev_loop_fork on the
  440. embedded loop when the parent loop forks.
  441. - new function: ev_now_update (loop).
  442. - verify_watcher was not marked static.
  443. - improve the "associating..." manpage section.
  444. - documentation tweaks here and there.
  445. 3.43 Sun Jul 6 05:34:41 CEST 2008
  446. - include more include files on windows to get struct _stati64
  447. (reported by Chris Hulbert, but doesn't quite fix his issue).
  448. - add missing #include <io.h> in ev.c on windows (reported by
  449. Matt Tolton).
  450. 3.42 Tue Jun 17 12:12:07 CEST 2008
  451. - work around yet another windows bug: FD_SET actually adds fd's
  452. multiple times to the fd_*SET*, despite official MSN docs claiming
  453. otherwise. Reported and well-analysed by Matt Tolton.
  454. - define NFDBITS to 0 when EV_SELECT_IS_WINSOCKET to make it compile
  455. (reported any analysed by Chris Hulbert).
  456. - fix a bug in ev_ebadf (this function is only used to catch
  457. programming errors in the libev user). reported by Matt Tolton.
  458. - fix a bug in fd_intern on win32 (could lead to compile errors
  459. under some circumstances, but would work correctly if it compiles).
  460. reported by Matt Tolton.
  461. - (try to) work around missing lstat on windows.
  462. - pass in the write fd set as except fd set under windows. windows
  463. is so uncontrollably lame that it requires this. this means that
  464. switching off oobinline is not supported (but tcp/ip doesn't
  465. have oob, so that would be stupid anyways.
  466. - use posix module symbol to auto-detect monotonic clock presence
  467. and some other default values.
  468. 3.41 Fri May 23 18:42:54 CEST 2008
  469. - work around an obscure bug in winsocket select: if you
  470. provide only empty fd sets then select returns WSAEINVAL. how sucky.
  471. - improve timer scheduling stability and reduce use of time_epsilon.
  472. - use 1-based 2-heap for EV_MINIMAL, simplifies code, reduces
  473. codesize and makes for better cache-efficiency.
  474. - use 3-based 4-heap for !EV_MINIMAL. this makes better use
  475. of cpu cache lines and gives better growth behaviour than
  476. 2-based heaps.
  477. - cache timestamp within heap for !EV_MINIMAL, to avoid random
  478. memory accesses.
  479. - document/add EV_USE_4HEAP and EV_HEAP_CACHE_AT.
  480. - fix a potential aliasing issue in ev_timer_again.
  481. - add/document ev_periodic_at, retract direct access to ->at.
  482. - improve ev_stat docs.
  483. - add portability requirements section.
  484. - fix manpage headers etc.
  485. - normalise WSA error codes to lower range on windows.
  486. - add consistency check code that can be called automatically
  487. or on demand to check for internal structures (ev_loop_verify).
  488. 3.31 Wed Apr 16 20:45:04 CEST 2008
  489. - added last minute fix for ev_poll.c by Brandon Black.
  490. 3.3 Wed Apr 16 19:04:10 CEST 2008
  491. - event_base_loopexit should return 0 on success
  492. (W.C.A. Wijngaards).
  493. - added linux eventfd support.
  494. - try to autodetect epoll and inotify support
  495. by libc header version if not using autoconf.
  496. - new symbols: EV_DEFAULT_UC and EV_DEFAULT_UC_.
  497. - declare functions defined in ev.h as inline if
  498. C99 or gcc are available.
  499. - enable inlining with gcc versions 2 and 3.
  500. - work around broken poll implementations potentially
  501. not clearing revents field in ev_poll (Brandon Black)
  502. (no such systems are known at this time).
  503. - work around a bug in realloc on openbsd and darwin,
  504. also makes the erroneous valgrind complaints
  505. go away (noted by various people).
  506. - fix ev_async_pending, add c++ wrapper for ev_async
  507. (based on patch sent by Johannes Deisenhofer).
  508. - add sensible set method to ev::embed.
  509. - made integer constants type int in ev.h.
  510. 3.2 Wed Apr 2 17:11:19 CEST 2008
  511. - fix a 64 bit overflow issue in the select backend,
  512. by using fd_mask instead of int for the mask.
  513. - rename internal sighandler to avoid clash with very old perls.
  514. - entering ev_loop will not clear the ONESHOT or NONBLOCKING
  515. flags of any outer loops anymore.
  516. - add ev_async_pending.
  517. 3.1 Thu Mar 13 13:45:22 CET 2008
  518. - implement ev_async watchers.
  519. - only initialise signal pipe on demand.
  520. - make use of sig_atomic_t configurable.
  521. - improved documentation.
  522. 3.0 Mon Jan 28 13:14:47 CET 2008
  523. - API/ABI bump to version 3.0.
  524. - ev++.h includes "ev.h" by default now, not <ev.h>.
  525. - slightly improved documentation.
  526. - speed up signal detection after a fork.
  527. - only optionally return trace status changed in ev_child
  528. watchers.
  529. - experimental (and undocumented) loop wrappers for ev++.h.
  530. 2.01 Tue Dec 25 08:04:41 CET 2007
  531. - separate Changes file.
  532. - fix ev_path_set => ev_stat_set typo.
  533. - remove event_compat.h from the libev tarball.
  534. - change how include files are found.
  535. - doc updates.
  536. - update licenses, explicitly allow for GPL relicensing.
  537. 2.0 Sat Dec 22 17:47:03 CET 2007
  538. - new ev_sleep, ev_set_(io|timeout)_collect_interval.
  539. - removed epoll from embeddable fd set.
  540. - fix embed watchers.
  541. - renamed ev_embed.loop to other.
  542. - added exported Symbol tables.
  543. - undefine member wrapper macros at the end of ev.c.
  544. - respect EV_H in ev++.h.
  545. 1.86 Tue Dec 18 02:36:57 CET 2007
  546. - fix memleak on loop destroy (not relevant for perl).
  547. 1.85 Fri Dec 14 20:32:40 CET 2007
  548. - fix some aliasing issues w.r.t. timers and periodics
  549. (not relevant for perl).
  550. (for historic versions refer to EV/Changes, found in the Perl interface)
  551. 0.1 Wed Oct 31 21:31:48 CET 2007
  552. - original version; hacked together in <24h.