--- libev/ev.pod 2008/10/29 14:12:34 1.209 +++ libev/ev.pod 2009/03/04 14:33:10 1.227 @@ -11,6 +11,8 @@ // a single header file is required #include + #include // for puts + // every watcher type has its own typedef'd struct // with the name ev_TYPE ev_io stdin_watcher; @@ -43,7 +45,7 @@ main (void) { // use the default event loop unless you have special needs - ev_loop *loop = ev_default_loop (0); + struct ev_loop *loop = ev_default_loop (0); // initialise an io watcher, then start it // this one will watch for stdin to become readable @@ -388,27 +390,29 @@ like O(total_fds) where n is the total number of fds (or the highest fd), epoll scales either O(1) or O(active_fds). -The epoll syscalls are the most misdesigned of the more advanced event -mechanisms: problems include silently dropping fds, requiring a system -call per change per fd (and unnecessary guessing of parameters), problems -with dup and so on. The biggest issue is fork races, however - if a -program forks then I parent and child process have to recreate the -epoll set, which can take considerable time (one syscall per fd) and is of -course hard to detect. +The epoll mechanism deserves honorable mention as the most misdesigned +of the more advanced event mechanisms: mere annoyances include silently +dropping file descriptors, requiring a system call per change per file +descriptor (and unnecessary guessing of parameters), problems with dup and +so on. The biggest issue is fork races, however - if a program forks then +I parent and child process have to recreate the epoll set, which can +take considerable time (one syscall per file descriptor) and is of course +hard to detect. -Epoll is also notoriously buggy - embedding epoll fds should work, but -of course doesn't, and epoll just loves to report events for totally +Epoll is also notoriously buggy - embedding epoll fds I work, but +of course I, and epoll just loves to report events for totally I file descriptors (even already closed ones, so one cannot even remove them from the set) than registered in the set (especially on SMP systems). Libev tries to counter these spurious notifications by employing an additional generation counter and comparing that against the -events to filter out spurious ones. +events to filter out spurious ones, recreating the set when required. While stopping, setting and starting an I/O watcher in the same iteration -will result in some caching, there is still a system call per such incident -(because the fd could point to a different file description now), so its -best to avoid that. Also, C'ed file descriptors might not work -very well if you register events for both fds. +will result in some caching, there is still a system call per such +incident (because the same I could point to a different +I now), so its best to avoid that. Also, C'ed +file descriptors might not work very well if you register events for both +file descriptors. Best performance from this backend is achieved by not unregistering all watchers for a file descriptor until it has been closed, if possible, @@ -418,6 +422,10 @@ as in libev having to destroy and recreate the epoll object, which can take considerable time and thus should be avoided. +All this means that, in practice, C can be as fast or +faster than epoll for maybe up to a hundred file descriptors, depending on +the usage. So sad. + While nominally embeddable in other event loops, this feature is broken in all kernel versions tested so far. @@ -426,12 +434,15 @@ =item C (value 8, most BSD clones) -Kqueue deserves special mention, as at the time of this writing, it was -broken on all BSDs except NetBSD (usually it doesn't work reliably with -anything but sockets and pipes, except on Darwin, where of course it's -completely useless). For this reason it's not being "auto-detected" unless -you explicitly specify it in the flags (i.e. using C) or -libev was compiled on a known-to-be-good (-enough) system like NetBSD. +Kqueue deserves special mention, as at the time of this writing, it +was broken on all BSDs except NetBSD (usually it doesn't work reliably +with anything but sockets and pipes, except on Darwin, where of course +it's completely useless). Unlike epoll, however, whose brokenness +is by design, these kqueue bugs can (and eventually will) be fixed +without API changes to existing programs. For this reason it's not being +"auto-detected" unless you explicitly specify it in the flags (i.e. using +C) or libev was compiled on a known-to-be-good (-enough) +system like NetBSD. You still can embed kqueue into a normal poll or select backend and use it only for sockets (after having made sure that sockets work with kqueue on @@ -451,8 +462,8 @@ everywhere, so you might need to test for this. And since it is broken almost everywhere, you should only use it when you have a lot of sockets (for which it usually works), by embedding it into another event loop -(e.g. C or C) and, did I mention it, -using it only for sockets. +(e.g. C or C (but C is of course +also broken on OS X)) and, did I mention it, using it only for sockets. This backend maps C into an C kevent with C, and C into an C kevent with @@ -1412,7 +1423,7 @@ // callback was invoked, but there was some activity, re-arm // the watcher to fire in last_activity + 60, which is // guaranteed to be in the future, so "again" is positive: - w->again = timeout - now; + w->repeat = timeout - now; ev_timer_again (EV_A_ w); } } @@ -1587,52 +1598,62 @@ Periodic watchers are also timers of a kind, but they are very versatile (and unfortunately a bit complex). -Unlike C's, they are not based on real time (or relative time) -but on wall clock time (absolute time). You can tell a periodic watcher -to trigger after some specific point in time. For example, if you tell a -periodic watcher to trigger in 10 seconds (by specifying e.g. C, that is, an absolute time not a delay) and then reset your system -clock to January of the previous year, then it will take more than year -to trigger the event (unlike an C, which would still trigger -roughly 10 seconds later as it uses a relative timeout). - -Cs can also be used to implement vastly more complex timers, -such as triggering an event on each "midnight, local time", or other -complicated rules. +Unlike C, periodic watchers are not based on real time (or +relative time, the physical time that passes) but on wall clock time +(absolute time, the thing you can read on your calender or clock). The +difference is that wall clock time can run faster or slower than real +time, and time jumps are not uncommon (e.g. when you adjust your +wrist-watch). + +You can tell a periodic watcher to trigger after some specific point +in time: for example, if you tell a periodic watcher to trigger "in 10 +seconds" (by specifying e.g. C, that is, an absolute time +not a delay) and then reset your system clock to January of the previous +year, then it will take a year or more to trigger the event (unlike an +C, which would still trigger roughly 10 seconds after starting +it, as it uses a relative timeout). + +C watchers can also be used to implement vastly more complex +timers, such as triggering an event on each "midnight, local time", or +other complicated rules. This cannot be done with C watchers, as +those cannot react to time jumps. As with timers, the callback is guaranteed to be invoked only when the -time (C) has passed, but if multiple periodic timers become ready -during the same loop iteration, then order of execution is undefined. +point in time where it is supposed to trigger has passed, but if multiple +periodic timers become ready during the same loop iteration, then order of +execution is undefined. =head3 Watcher-Specific Functions and Data Members =over 4 -=item ev_periodic_init (ev_periodic *, callback, ev_tstamp at, ev_tstamp interval, reschedule_cb) +=item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb) -=item ev_periodic_set (ev_periodic *, ev_tstamp after, ev_tstamp repeat, reschedule_cb) +=item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb) -Lots of arguments, lets sort it out... There are basically three modes of +Lots of arguments, let's sort it out... There are basically three modes of operation, and we will explain them from simplest to most complex: =over 4 -=item * absolute timer (at = time, interval = reschedule_cb = 0) +=item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0) In this configuration the watcher triggers an event after the wall clock -time C has passed. It will not repeat and will not adjust when a time -jump occurs, that is, if it is to be run at January 1st 2011 then it will -only run when the system clock reaches or surpasses this time. +time C has passed. It will not repeat and will not adjust when a +time jump occurs, that is, if it is to be run at January 1st 2011 then it +will be stopped and invoked when the system clock reaches or surpasses +this point in time. -=item * repeating interval timer (at = offset, interval > 0, reschedule_cb = 0) +=item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0) In this mode the watcher will always be scheduled to time out at the next -C time (for some integer N, which can also be negative) -and then repeat, regardless of any time jumps. +C time (for some integer N, which can also be +negative) and then repeat, regardless of any time jumps. The C +argument is merely an offset into the C periods. This can be used to create timers that do not drift with respect to the -system clock, for example, here is a C that triggers each -hour, on the hour: +system clock, for example, here is an C that triggers each +hour, on the hour (with respect to UTC): ev_periodic_set (&periodic, 0., 3600., 0); @@ -1643,9 +1664,9 @@ Another way to think about it (for the mathematically inclined) is that C will try to run the callback in this mode at the next possible -time where C