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.

274 lines
13 KiB

14 years ago
12 years ago
12 years ago
12 years ago
12 years ago
12 years ago
12 years ago
12 years ago
12 years ago
12 years ago
12 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
13 years ago
14 years ago
  1. Revision history for libev, a high-performance and full-featured event loop.
  2. TODO: ev_walk
  3. TODO: signal handling per loop
  4. TODO: signalfd
  5. TODO: document signalfd procmask
  6. - incompatible change: do not necessarily reset signal handler
  7. to SIG_DFL when a sighandler is stopped.
  8. - take advantage of signalfd on GNU/Linux systems.
  9. - document that the signal mask might be in an unspecified
  10. state when using libev's signal handling.
  11. 3.7 Fri Jul 17 16:36:32 CEST 2009
  12. - ev_unloop and ev_loop wrongly used a global variable to exit loops,
  13. instead of using a per-loop variable (bug caught by accident...).
  14. - the ev_set_io_collect_interval interpretation has changed.
  15. - add new functionality: ev_set_userdata, ev_userdata,
  16. ev_set_invoke_pending_cb, ev_set_loop_release_cb,
  17. ev_invoke_pending, ev_pending_count, together with a long example
  18. about thread locking.
  19. - add ev_timer_remaining (as requested by Denis F. Latypoff).
  20. - add ev_loop_depth.
  21. - calling ev_unloop in fork/prepare watchers will no longer poll
  22. for new events.
  23. - Denis F. Latypoff corrected many typos in example code snippets.
  24. - honor autoconf detection of EV_USE_CLOCK_SYSCALL, also double-
  25. check that the syscall number is available before trying to
  26. use it (reported by ry@tinyclouds).
  27. - use GetSystemTimeAsFileTime instead of _timeb on windows, for
  28. slightly higher accuracy.
  29. - properly declare ev_loop_verify and ev_now_update even when
  30. !EV_MULTIPLICITY.
  31. - do not compile in any priority code when EV_MAXPRI == EV_MINPRI.
  32. - support EV_MINIMAL==2 for a reduced API.
  33. - actually 0-initialise struct sigaction when installing signals.
  34. - add section on hibernate and stopped processes to ev_timer docs.
  35. 3.6 Tue Apr 28 02:49:30 CEST 2009
  36. - multiple timers becoming ready within an event loop iteration
  37. will be invoked in the "correct" order now.
  38. - do not leave the event loop early just because we have no active
  39. watchers, fixing a problem when embedding a kqueue loop
  40. that has active kernel events but no registered watchers
  41. (reported by blacksand blacksand).
  42. - correctly zero the idx values for arrays, so destroying and
  43. reinitialising the default loop actually works (patch by
  44. Malek Hadj-Ali).
  45. - implement ev_suspend and ev_resume.
  46. - new EV_CUSTOM revents flag for use by applications.
  47. - add documentation section about priorites.
  48. - add a glossary to the dcoumentation.
  49. - extend the ev_fork description slightly.
  50. - optimize a jump out of call_pending.
  51. 3.53 Sun Feb 15 02:38:20 CET 2009
  52. - fix a bug in event pipe creation on win32 that would cause a
  53. failed assertion on event loop creation (patch by Malek Hadj-Ali).
  54. - probe for CLOCK_REALTIME support at runtime as well and fall
  55. back to gettimeofday if there is an error, to support older
  56. operating systems with newer header files/libraries.
  57. - prefer gettimeofday over clock_gettime with USE_CLOCK_SYSCALL
  58. (default most everywhere), otherwise not.
  59. 3.52 Wed Jan 7 21:43:02 CET 2009
  60. - fix compilation of select backend in fd_set mode when NFDBITS is
  61. missing (to get it to compile on QNX, reported by Rodrigo Campos).
  62. - better select-nfds handling when select backend is in fd_set mode.
  63. - diagnose fd_set overruns when select backend is in fd_set mode.
  64. - due to a thinko, instead of disabling everything but
  65. select on the borked OS X platform, everything but select was
  66. allowed (reported by Emanuele Giaquinta).
  67. - actually verify that local and remote port are matching in
  68. libev's socketpair emulation, which makes denial-of-service
  69. attacks harder (but not impossible - it's windows). Make sure
  70. it even works under vista, which thinks that getpeer/sockname
  71. should return fantasy port numbers.
  72. - include "libev" in all assertion messages for potentially
  73. clearer diagnostics.
  74. - event_get_version (libevent compatibility) returned
  75. a useless string instead of the expected version string
  76. (patch by W.C.A. Wijngaards).
  77. 3.51 Wed Dec 24 23:00:11 CET 2008
  78. - fix a bug where an inotify watcher was added twice, causing
  79. freezes on hash collisions (reported and analysed by Graham Leggett).
  80. - new config symbol, EV_USE_CLOCK_SYSCALL, to make libev use
  81. a direct syscall - slower, but no dependency on librt et al.
  82. - assume negative return values != -1 signals success of port_getn
  83. (http://cvs.epicsol.org/cgi/viewcvs.cgi/epic5/source/newio.c?rev=1.52)
  84. (no known failure reports, but it doesn't hurt).
  85. - fork detection in ev_embed now stops and restarts the watcher
  86. automatically.
  87. - EXPERIMENTAL: default the method to operator () in ev++.h,
  88. to make it nicer to use functors (requested by Benedek László).
  89. - fixed const object callbacks in ev++.h.
  90. - replaced loop_ref argument of watcher.set (loop) by a direct
  91. ev_loop * in ev++.h, to avoid clashes with functor patch.
  92. - do not try to watch the empty string via inotify.
  93. - inotify watchers could be leaked under certain circumstances.
  94. - OS X 10.5 is actually even more broken than earlier versions,
  95. so fall back to select on that piece of garbage.
  96. - fixed some weirdness in the ev_embed documentation.
  97. 3.49 Wed Nov 19 11:26:53 CET 2008
  98. - ev_stat watchers will now use inotify as a mere hint on
  99. kernels <2.6.25, or if the filesystem is not in the
  100. "known to be good" list.
  101. - better mingw32 compatibility (it's not as borked as native win32)
  102. (analysed by Roger Pack).
  103. - include stdio.h in the example program, as too many people are
  104. confused by the weird C language otherwise. I guess the next thing
  105. I get told is that the "..." ellipses in the examples don't compile
  106. with their C compiler.
  107. 3.48 Thu Oct 30 09:02:37 CET 2008
  108. - further optimise away the EPOLL_CTL_ADD/MOD combo in the epoll
  109. backend by assuming the kernel event mask hasn't changed if
  110. ADD fails with EEXIST.
  111. - work around spurious event notification bugs in epoll by using
  112. a 32-bit generation counter. recreate kernel state if we receive
  113. spurious notifications or unwanted events. this is very costly,
  114. but I didn't come up with this horrible design.
  115. - use memset to initialise most arrays now and do away with the
  116. init functions.
  117. - expand time-out strategies into a "Be smart about timeouts" section.
  118. - drop the "struct" from all ev_watcher declarations in the
  119. documentation and did other clarifications (yeah, it was a mistake
  120. to have a struct AND a function called ev_loop).
  121. - fix a bug where ev_default would not initialise the default
  122. loop again after it was destroyed with ev_default_destroy.
  123. - rename syserr to ev_syserr to avoid name clashes when embedding,
  124. do similar changes for event.c.
  125. 3.45 Tue Oct 21 21:59:26 CEST 2008
  126. - disable inotify usage on linux <2.6.25, as it is broken
  127. (reported by Yoann Vandoorselaere).
  128. - ev_stat errornously would try to add inotify watchers
  129. even when inotify wasn't available (this should only
  130. have a performance impact).
  131. - ev_once now passes both timeout and io to the callback if both
  132. occur concurrently, instead of giving timeouts precedence.
  133. - disable EV_USE_INOTIFY when sys/inotify.h is too old.
  134. 3.44 Mon Sep 29 05:18:39 CEST 2008
  135. - embed watchers now automatically invoke ev_loop_fork on the
  136. embedded loop when the parent loop forks.
  137. - new function: ev_now_update (loop).
  138. - verify_watcher was not marked static.
  139. - improve the "associating..." manpage section.
  140. - documentation tweaks here and there.
  141. 3.43 Sun Jul 6 05:34:41 CEST 2008
  142. - include more include files on windows to get struct _stati64
  143. (reported by Chris Hulbert, but doesn't quite fix his issue).
  144. - add missing #include <io.h> in ev.c on windows (reported by
  145. Matt Tolton).
  146. 3.42 Tue Jun 17 12:12:07 CEST 2008
  147. - work around yet another windows bug: FD_SET actually adds fd's
  148. multiple times to the fd_*SET*, despite official MSN docs claiming
  149. otherwise. Reported and well-analysed by Matt Tolton.
  150. - define NFDBITS to 0 when EV_SELECT_IS_WINSOCKET to make it compile
  151. (reported any analysed by Chris Hulbert).
  152. - fix a bug in ev_ebadf (this function is only used to catch
  153. programming errors in the libev user). reported by Matt Tolton.
  154. - fix a bug in fd_intern on win32 (could lead to compile errors
  155. under some circumstances, but would work correctly if it compiles).
  156. reported by Matt Tolton.
  157. - (try to) work around missing lstat on windows.
  158. - pass in the write fd set as except fd set under windows. windows
  159. is so uncontrollably lame that it requires this. this means that
  160. switching off oobinline is not supported (but tcp/ip doesn't
  161. have oob, so that would be stupid anyways.
  162. - use posix module symbol to auto-detect monotonic clock presence
  163. and some other default values.
  164. 3.41 Fri May 23 18:42:54 CEST 2008
  165. - work around an obscure bug in winsocket select: if you
  166. provide only empty fd sets then select returns WSAEINVAL. how sucky.
  167. - improve timer scheduling stability and reduce use of time_epsilon.
  168. - use 1-based 2-heap for EV_MINIMAL, simplifies code, reduces
  169. codesize and makes for better cache-efficiency.
  170. - use 3-based 4-heap for !EV_MINIMAL. this makes better use
  171. of cpu cache lines and gives better growth behaviour than
  172. 2-based heaps.
  173. - cache timestamp within heap for !EV_MINIMAL, to avoid random
  174. memory accesses.
  175. - document/add EV_USE_4HEAP and EV_HEAP_CACHE_AT.
  176. - fix a potential aliasing issue in ev_timer_again.
  177. - add/document ev_periodic_at, retract direct access to ->at.
  178. - improve ev_stat docs.
  179. - add portability requirements section.
  180. - fix manpage headers etc.
  181. - normalise WSA error codes to lower range on windows.
  182. - add consistency check code that can be called automatically
  183. or on demand to check for internal structures (ev_loop_verify).
  184. 3.31 Wed Apr 16 20:45:04 CEST 2008
  185. - added last minute fix for ev_poll.c by Brandon Black.
  186. 3.3 Wed Apr 16 19:04:10 CEST 2008
  187. - event_base_loopexit should return 0 on success
  188. (W.C.A. Wijngaards).
  189. - added linux eventfd support.
  190. - try to autodetect epoll and inotify support
  191. by libc header version if not using autoconf.
  192. - new symbols: EV_DEFAULT_UC and EV_DEFAULT_UC_.
  193. - declare functions defined in ev.h as inline if
  194. C99 or gcc are available.
  195. - enable inlining with gcc versions 2 and 3.
  196. - work around broken poll implementations potentially
  197. not clearing revents field in ev_poll (Brandon Black)
  198. (no such systems are known at this time).
  199. - work around a bug in realloc on openbsd and darwin,
  200. also makes the errornous valgrind complaints
  201. go away (noted by various people).
  202. - fix ev_async_pending, add c++ wrapper for ev_async
  203. (based on patch sent by Johannes Deisenhofer.
  204. - add sensible set method to ev::embed.
  205. - made integer constants type int in ev.h.
  206. 3.2 Wed Apr 2 17:11:19 CEST 2008
  207. - fix a 64 bit overflow issue in the select backend,
  208. by using fd_mask instead of int for the mask.
  209. - rename internal sighandler to avoid clash with very old perls.
  210. - entering ev_loop will not clear the ONESHOT or NONBLOCKING
  211. flags of any outer loops anymore.
  212. - add ev_async_pending.
  213. 3.1 Thu Mar 13 13:45:22 CET 2008
  214. - implement ev_async watchers.
  215. - only initialise signal pipe on demand.
  216. - make use of sig_atomic_t configurable.
  217. - improved documentation.
  218. 3.0 Mon Jan 28 13:14:47 CET 2008
  219. - API/ABI bump to version 3.0.
  220. - ev++.h includes "ev.h" by default now, not <ev.h>.
  221. - slightly improved documentation.
  222. - speed up signal detection after a fork.
  223. - only optionally return trace status changed in ev_child
  224. watchers.
  225. - experimental (and undocumented) loop wrappers for ev++.h.
  226. 2.01 Tue Dec 25 08:04:41 CET 2007
  227. - separate Changes file.
  228. - fix ev_path_set => ev_stat_set typo.
  229. - remove event_compat.h from the libev tarball.
  230. - change how include files are found.
  231. - doc updates.
  232. - update licenses, explicitly allow for GPL relicensing.
  233. 2.0 Sat Dec 22 17:47:03 CET 2007
  234. - new ev_sleep, ev_set_(io|timeout)_collect_interval.
  235. - removed epoll from embeddable fd set.
  236. - fix embed watchers.
  237. - renamed ev_embed.loop to other.
  238. - added exported Symbol tables.
  239. - undefine member wrapper macros at the end of ev.c.
  240. - respect EV_H in ev++.h.
  241. 1.86 Tue Dec 18 02:36:57 CET 2007
  242. - fix memleak on loop destroy (not relevant for perl).
  243. 1.85 Fri Dec 14 20:32:40 CET 2007
  244. - fix some aliasing issues w.r.t. timers and periodics
  245. (not relevant for perl).
  246. (for historic versions refer to EV/Changes, found in the Perl interface)
  247. 0.1 Wed Oct 31 21:31:48 CET 2007
  248. - original version; hacked together in <24h.