--- EV/Changes 2008/10/28 08:08:28 1.151 +++ EV/Changes 2010/07/08 00:45:03 1.192 @@ -1,14 +1,152 @@ Revision history for Perl extension EV - - (libev) use a generation counter to detetc spurious epoll events +Changes marked with (libev) are changes in libev, and have more +documentation in the libev Changes file. + +4.00 + - fix a bug where inotify usage would parse the same event + multiple times, causing various forms of breakage. + - greatly reduce stack usage for inotify (8kb to <0.5kb). + - renamed loop_verify to verify, loop_count to iteration and + loop_depth to depth. + - expose ev_depth and ev_verify via the XS API. + - (libev) ev_embed_stop did not correctly stop the watcher. + - (libev) disable poll backend on AIX. + - (libev) rename EV_TIMEOUT to EV_TIMER. + - (libev) add section on accept() problems to the manpage. + - (libev) no child watchers on win32. + - make code more aliasing compliant, in case perl is ever + translated to C. + - ported to minix 3.1.7. + +3.9 Thu Dec 31 07:59:59 CET 2009 + - disable t/07* under automatic testing. + - increase t/09* timeouts as netbsd has *horribly* broken + select/kevent that *usually* sleep >>0.6s instead of 0.3. + - add constants EV::FLAG_NOSIGFD, EV::FLAG_NOINOTIFY and + EV::BACKEND_ALL. + - (libev) signalfd is no longer used by default. + - (libev) backport inotify code to C89. + - (libev) inotify file descriptors could leak into child processes. + - (libev) ev_stat watchers could keep an errornous extra ref on the loop. + - (libev) take advantage of inotify_init1, if available. + - (libev) the signal handling pipe wasn't always initialised under windows. + - changed minimum glibc requirement from glibc 2.9 to 2.7, for signalfd. + - (libev) only replace ev_stat.prev when we detect an actual difference. + +3.8 Sun Aug 9 15:30:10 CEST 2009 + - implement $loop->signal/signal_ns. + - (libev) incompatible change: do not necessarily reset signal + handler to SIG_DFL when a sighandler is stopped. + - (libev) ev_default_destroy did not properly free or zero some + members, potentially causing crashes and memory corruption on + repated ev_default_destroy/ev_default_loop calls. + - (libev) take advantage of signalfd on GNU/Linux systems. + - (libev) document that the signal mask might be in an unspecified + state when using libev's signal handling. + - (libev) take advantage of some GNU/Linux calls to set + cloexec/nonblock on fd creation, to avoid race conditions. + - implement internal glue code to interface more efficiently + with AnyEvent. + +3.7 Fri Jul 17 16:49:16 CEST 2009 + - add EV::loop_verify and EV::loop_depth. + - use output filehandle for i/o watchers waiting + for EV_WRITE, otherwise input filehandle. + - use common schmorp.h header. + - add EV::Timer->remaining. + - allow for subclassing of EV::Loop (see the exciting + EV::Loop::Async module for an example). + - added EV::invoke_pending and EV::pending_count. + + - (libev) ev_unloop and ev_loop wrongly used a global variable to + exit loops, instead of using a per-loop variable. + - (libev) the ev_set_io_collect_interval interpretation has changed. + - add new functionality: EV::invoke_pending, EV::pending_count. + - add $timer->remaining. + - add EV::loop_depth. + - (libev) calling unloop in fork/prepare watchers will no longer poll + for new events. + - (libev) use GetSystemTimeAsFileTime instead of _timeb on windows, + for slightly higher accuracy. + - (libev) actually 0-initialise struct sigaction when installing + signals. + +3.6 Tue Apr 28 02:50:37 CEST 2009 + - keepalive(0) could decrease the refcount of the loop + permanently. + - add ev_suspend/ev_resume, also make ev_now_update + accessible via the XS API. + - most EV:: constants were missing and have been + added :/. + - add EV::VERSION_MAJOR/VERSION_MINOR constants. + - (libev) multiple timers becoming ready within an event loop + iteration will be invoked in the "correct" order now. + - (libev) do not leave the event loop early just because we have + no active watchers, fixing a problem when embedding a kqueue + loop that has active kernel events but no registered watchers + (reported by blacksand blacksand). + - (libev) correctly zero the idx values for arrays, so destroying + and reinitialising the default loop actually works (patch by + Malek Hadj-Ali). + - (libev) new EV::CUSTOM revents flag for use by applications. + - (libev) add documentation section about priorites. + - (libev) add a glossary to the dcoumentation. + - (libev) extend the ev_fork description slightly. + - (libev) optimize a jump out of call_pending. + - t/03_keepalive could fail when there was no actual error. + +3.53 Sun Feb 15 02:38:20 CET 2009 + - (libev) on win32, the event loop creation could randomly fail + due to an initialised variable having the wrong value. + - (libev) probe for CLOCK_REALTIME support at runtime as well and + fall back to gettimeofday if there is an error, to support older + operating systems with newer header files/libraries. + - prefer gettimeofday over clock_gettime by default. + +3.52 Wed Jan 7 21:46:14 CET 2009 + - (libev) fix some issues in the select backend when in fd_set mode. + - (libev) due to a thinko, instead of disabling everything but + select on the borked OS X platform, everything but select was + allowed (reported by Emanuele Giaquinta). + - (libev) actually verify that local and remote port are matching in + libev's socketpair emulation, which makes denial-of-service + attacks harder (but not impossible - it's windows). Make sure + it even works under vista, which thinks that getpeer/sockname + should return fantasy port numbers. + +3.51 Wed Dec 24 23:01:59 CET 2008 + - do not cache the arguments passed to callbacks if the refcount + indicates that the callback has stolen them. + - (libev) try to avoid librt on GNU/Linux. + - (libev) check port_getn return value dfferently, might potentially + avoid problems. + - (libev) fix a bug with stat watchers possibly causing freezes. + - (libev) work around OS X 10.5 breaking poll, now select is the + only viable choice left on that pile of garbage. + - play tester whore: disable some tests that typically fail only + on cpan tester machines. + +3.49 Wed Nov 19 11:26:53 CET 2008 + - fix typos in manpage (Alex Efros). + - increase timing even further, for the ever-overloaded cpan-tester + machines, and to the detriment of everybody else who wants a fast + make test. + - possible 5.6 compatibility. + - (libev) use inotify even on buggy kernels, but don't rely on it. + - (libev) use inotify only as an added hint on network filesystems. + +3.48 Thu Oct 30 09:09:48 CET 2008 + - (libev) use a generation counter to detect spurious epoll events and recreate the kernel in such a case (sorry, it's slow, but I didn't design epoll...). - (libev) optimise away an EPOLL_CTL_ADD/MOD combo in the epoll backend in some cases. - (libev) use memset to initialise most arrays now and do away with the init functions. - - speed up event callback invocation further (~15%) by + - speed up event callback invocation further (~15%), by not creating the object reference each time. + - EV::sleep was documented, but not implemented. fun. 3.45 Tue Oct 21 22:20:39 CEST 2008 - (libev) disable inotify usage on linux <2.6.25 (kernel bug).