5 libev - a high performance full-featured event loop written in C
11 =head2 EXAMPLE PROGRAM
13 // a single header file is required
16 #include <stdio.h> // for puts
18 // every watcher type has its own typedef'd struct
19 // with the name ev_TYPE
21 ev_timer timeout_watcher;
23 // all watcher callbacks have a similar signature
24 // this callback is called when data is readable on stdin
26 stdin_cb (EV_P_ ev_io *w, int revents)
29 // for one-shot events, one must manually stop the watcher
30 // with its corresponding stop function.
33 // this causes all nested ev_run's to stop iterating
34 ev_break (EV_A_ EVBREAK_ALL);
37 // another callback, this time for a time-out
39 timeout_cb (EV_P_ ev_timer *w, int revents)
42 // this causes the innermost ev_run to stop iterating
43 ev_break (EV_A_ EVBREAK_ONE);
49 // use the default event loop unless you have special needs
50 struct ev_loop *loop = EV_DEFAULT;
52 // initialise an io watcher, then start it
53 // this one will watch for stdin to become readable
54 ev_io_init (&stdin_watcher, stdin_cb, /*STDIN_FILENO*/ 0, EV_READ);
55 ev_io_start (loop, &stdin_watcher);
57 // initialise a timer watcher, then start it
58 // simple non-repeating 5.5 second timeout
59 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
60 ev_timer_start (loop, &timeout_watcher);
62 // now wait for events to arrive
65 // break was called, so exit
69 =head1 ABOUT THIS DOCUMENT
71 This document documents the libev software package.
73 The newest version of this document is also available as an html-formatted
74 web page you might find easier to navigate when reading it for the first
75 time: L<http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod>.
77 While this document tries to be as complete as possible in documenting
78 libev, its usage and the rationale behind its design, it is not a tutorial
79 on event-based programming, nor will it introduce event-based programming
82 Familiarity with event based programming techniques in general is assumed
83 throughout this document.
85 =head1 WHAT TO READ WHEN IN A HURRY
87 This manual tries to be very detailed, but unfortunately, this also makes
88 it very long. If you just want to know the basics of libev, I suggest
89 reading L</ANATOMY OF A WATCHER>, then the L</EXAMPLE PROGRAM> above and
90 look up the missing functions in L</GLOBAL FUNCTIONS> and the C<ev_io> and
91 C<ev_timer> sections in L</WATCHER TYPES>.
95 Libev is an event loop: you register interest in certain events (such as a
96 file descriptor being readable or a timeout occurring), and it will manage
97 these event sources and provide your program with events.
99 To do this, it must take more or less complete control over your process
100 (or thread) by executing the I<event loop> handler, and will then
101 communicate events via a callback mechanism.
103 You register interest in certain events by registering so-called I<event
104 watchers>, which are relatively small C structures you initialise with the
105 details of the event, and then hand it over to libev by I<starting> the
110 Libev supports C<select>, C<poll>, the Linux-specific C<epoll>, the
111 BSD-specific C<kqueue> and the Solaris-specific event port mechanisms
112 for file descriptor events (C<ev_io>), the Linux C<inotify> interface
113 (for C<ev_stat>), Linux eventfd/signalfd (for faster and cleaner
114 inter-thread wakeup (C<ev_async>)/signal handling (C<ev_signal>)) relative
115 timers (C<ev_timer>), absolute timers with customised rescheduling
116 (C<ev_periodic>), synchronous signals (C<ev_signal>), process status
117 change events (C<ev_child>), and event watchers dealing with the event
118 loop mechanism itself (C<ev_idle>, C<ev_embed>, C<ev_prepare> and
119 C<ev_check> watchers) as well as file watchers (C<ev_stat>) and even
120 limited support for fork events (C<ev_fork>).
122 It also is quite fast (see this
123 L<benchmark|http://libev.schmorp.de/bench.html> comparing it to libevent
128 Libev is very configurable. In this manual the default (and most common)
129 configuration will be described, which supports multiple event loops. For
130 more info about various configuration options please have a look at
131 B<EMBED> section in this manual. If libev was configured without support
132 for multiple event loops, then all functions taking an initial argument of
133 name C<loop> (which is always of type C<struct ev_loop *>) will not have
136 =head2 TIME REPRESENTATION
138 Libev represents time as a single floating point number, representing
139 the (fractional) number of seconds since the (POSIX) epoch (in practice
140 somewhere near the beginning of 1970, details are complicated, don't
141 ask). This type is called C<ev_tstamp>, which is what you should use
142 too. It usually aliases to the C<double> type in C. When you need to do
143 any calculations on it, you should treat it as some floating point value.
145 Unlike the name component C<stamp> might indicate, it is also used for
146 time differences (e.g. delays) throughout libev.
148 =head1 ERROR HANDLING
150 Libev knows three classes of errors: operating system errors, usage errors
151 and internal errors (bugs).
153 When libev catches an operating system error it cannot handle (for example
154 a system call indicating a condition libev cannot fix), it calls the callback
155 set via C<ev_set_syserr_cb>, which is supposed to fix the problem or
156 abort. The default is to print a diagnostic message and to call C<abort
159 When libev detects a usage error such as a negative timer interval, then
160 it will print a diagnostic message and abort (via the C<assert> mechanism,
161 so C<NDEBUG> will disable this checking): these are programming errors in
162 the libev caller and need to be fixed there.
164 Libev also has a few internal error-checking C<assert>ions, and also has
165 extensive consistency checking code. These do not trigger under normal
166 circumstances, as they indicate either a bug in libev or worse.
169 =head1 GLOBAL FUNCTIONS
171 These functions can be called anytime, even before initialising the
176 =item ev_tstamp ev_time ()
178 Returns the current time as libev would use it. Please note that the
179 C<ev_now> function is usually faster and also often returns the timestamp
180 you actually want to know. Also interesting is the combination of
181 C<ev_now_update> and C<ev_now>.
183 =item ev_sleep (ev_tstamp interval)
185 Sleep for the given interval: The current thread will be blocked
186 until either it is interrupted or the given time interval has
187 passed (approximately - it might return a bit earlier even if not
188 interrupted). Returns immediately if C<< interval <= 0 >>.
190 Basically this is a sub-second-resolution C<sleep ()>.
192 The range of the C<interval> is limited - libev only guarantees to work
193 with sleep times of up to one day (C<< interval <= 86400 >>).
195 =item int ev_version_major ()
197 =item int ev_version_minor ()
199 You can find out the major and minor ABI version numbers of the library
200 you linked against by calling the functions C<ev_version_major> and
201 C<ev_version_minor>. If you want, you can compare against the global
202 symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the
203 version of the library your program was compiled against.
205 These version numbers refer to the ABI version of the library, not the
208 Usually, it's a good idea to terminate if the major versions mismatch,
209 as this indicates an incompatible change. Minor versions are usually
210 compatible to older versions, so a larger minor version alone is usually
213 Example: Make sure we haven't accidentally been linked against the wrong
214 version (note, however, that this will not detect other ABI mismatches,
215 such as LFS or reentrancy).
217 assert (("libev version mismatch",
218 ev_version_major () == EV_VERSION_MAJOR
219 && ev_version_minor () >= EV_VERSION_MINOR));
221 =item unsigned int ev_supported_backends ()
223 Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
224 value) compiled into this binary of libev (independent of their
225 availability on the system you are running on). See C<ev_default_loop> for
226 a description of the set values.
228 Example: make sure we have the epoll method, because yeah this is cool and
229 a must have and can we have a torrent of it please!!!11
231 assert (("sorry, no epoll, no sex",
232 ev_supported_backends () & EVBACKEND_EPOLL));
234 =item unsigned int ev_recommended_backends ()
236 Return the set of all backends compiled into this binary of libev and
237 also recommended for this platform, meaning it will work for most file
238 descriptor types. This set is often smaller than the one returned by
239 C<ev_supported_backends>, as for example kqueue is broken on most BSDs
240 and will not be auto-detected unless you explicitly request it (assuming
241 you know what you are doing). This is the set of backends that libev will
242 probe for if you specify no backends explicitly.
244 =item unsigned int ev_embeddable_backends ()
246 Returns the set of backends that are embeddable in other event loops. This
247 value is platform-specific but can include backends not available on the
248 current system. To find which embeddable backends might be supported on
249 the current system, you would need to look at C<ev_embeddable_backends ()
250 & ev_supported_backends ()>, likewise for recommended ones.
252 See the description of C<ev_embed> watchers for more info.
254 =item ev_set_allocator (void *(*cb)(void *ptr, long size) throw ())
256 Sets the allocation function to use (the prototype is similar - the
257 semantics are identical to the C<realloc> C89/SuS/POSIX function). It is
258 used to allocate and free memory (no surprises here). If it returns zero
259 when memory needs to be allocated (C<size != 0>), the library might abort
260 or take some potentially destructive action.
262 Since some systems (at least OpenBSD and Darwin) fail to implement
263 correct C<realloc> semantics, libev will use a wrapper around the system
264 C<realloc> and C<free> functions by default.
266 You could override this function in high-availability programs to, say,
267 free some memory if it cannot allocate memory, to use a special allocator,
268 or even to sleep a while and retry until some memory is available.
270 Example: Replace the libev allocator with one that waits a bit and then
271 retries (example requires a standards-compliant C<realloc>).
274 persistent_realloc (void *ptr, size_t size)
278 void *newptr = realloc (ptr, size);
288 ev_set_allocator (persistent_realloc);
290 =item ev_set_syserr_cb (void (*cb)(const char *msg) throw ())
292 Set the callback function to call on a retryable system call error (such
293 as failed select, poll, epoll_wait). The message is a printable string
294 indicating the system call or subsystem causing the problem. If this
295 callback is set, then libev will expect it to remedy the situation, no
296 matter what, when it returns. That is, libev will generally retry the
297 requested operation, or, if the condition doesn't go away, do bad stuff
300 Example: This is basically the same thing that libev does internally, too.
303 fatal_error (const char *msg)
310 ev_set_syserr_cb (fatal_error);
312 =item ev_feed_signal (int signum)
314 This function can be used to "simulate" a signal receive. It is completely
315 safe to call this function at any time, from any context, including signal
316 handlers or random threads.
318 Its main use is to customise signal handling in your process, especially
319 in the presence of threads. For example, you could block signals
320 by default in all threads (and specifying C<EVFLAG_NOSIGMASK> when
321 creating any loops), and in one thread, use C<sigwait> or any other
322 mechanism to wait for signals, then "deliver" them to libev by calling
327 =head1 FUNCTIONS CONTROLLING EVENT LOOPS
329 An event loop is described by a C<struct ev_loop *> (the C<struct> is
330 I<not> optional in this case unless libev 3 compatibility is disabled, as
331 libev 3 had an C<ev_loop> function colliding with the struct name).
333 The library knows two types of such loops, the I<default> loop, which
334 supports child process events, and dynamically created event loops which
339 =item struct ev_loop *ev_default_loop (unsigned int flags)
341 This returns the "default" event loop object, which is what you should
342 normally use when you just need "the event loop". Event loop objects and
343 the C<flags> parameter are described in more detail in the entry for
346 If the default loop is already initialised then this function simply
347 returns it (and ignores the flags. If that is troubling you, check
348 C<ev_backend ()> afterwards). Otherwise it will create it with the given
349 flags, which should almost always be C<0>, unless the caller is also the
350 one calling C<ev_run> or otherwise qualifies as "the main program".
352 If you don't know what event loop to use, use the one returned from this
353 function (or via the C<EV_DEFAULT> macro).
355 Note that this function is I<not> thread-safe, so if you want to use it
356 from multiple threads, you have to employ some kind of mutex (note also
357 that this case is unlikely, as loops cannot be shared easily between
360 The default loop is the only loop that can handle C<ev_child> watchers,
361 and to do this, it always registers a handler for C<SIGCHLD>. If this is
362 a problem for your application you can either create a dynamic loop with
363 C<ev_loop_new> which doesn't do that, or you can simply overwrite the
364 C<SIGCHLD> signal handler I<after> calling C<ev_default_init>.
366 Example: This is the most typical usage.
368 if (!ev_default_loop (0))
369 fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
371 Example: Restrict libev to the select and poll backends, and do not allow
372 environment settings to be taken into account:
374 ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
376 =item struct ev_loop *ev_loop_new (unsigned int flags)
378 This will create and initialise a new event loop object. If the loop
379 could not be initialised, returns false.
381 This function is thread-safe, and one common way to use libev with
382 threads is indeed to create one loop per thread, and using the default
383 loop in the "main" or "initial" thread.
385 The flags argument can be used to specify special behaviour or specific
386 backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
388 The following flags are supported:
394 The default flags value. Use this if you have no clue (it's the right
397 =item C<EVFLAG_NOENV>
399 If this flag bit is or'ed into the flag value (or the program runs setuid
400 or setgid) then libev will I<not> look at the environment variable
401 C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
402 override the flags completely if it is found in the environment. This is
403 useful to try out specific backends to test their performance, to work
404 around bugs, or to make libev threadsafe (accessing environment variables
405 cannot be done in a threadsafe way, but usually it works if no other
406 thread modifies them).
408 =item C<EVFLAG_FORKCHECK>
410 Instead of calling C<ev_loop_fork> manually after a fork, you can also
411 make libev check for a fork in each iteration by enabling this flag.
413 This works by calling C<getpid ()> on every iteration of the loop,
414 and thus this might slow down your event loop if you do a lot of loop
415 iterations and little real work, but is usually not noticeable (on my
416 GNU/Linux system for example, C<getpid> is actually a simple 5-insn sequence
417 without a system call and thus I<very> fast, but my GNU/Linux system also has
418 C<pthread_atfork> which is even faster).
420 The big advantage of this flag is that you can forget about fork (and
421 forget about forgetting to tell libev about forking) when you use this
424 This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
425 environment variable.
427 =item C<EVFLAG_NOINOTIFY>
429 When this flag is specified, then libev will not attempt to use the
430 I<inotify> API for its C<ev_stat> watchers. Apart from debugging and
431 testing, this flag can be useful to conserve inotify file descriptors, as
432 otherwise each loop using C<ev_stat> watchers consumes one inotify handle.
434 =item C<EVFLAG_SIGNALFD>
436 When this flag is specified, then libev will attempt to use the
437 I<signalfd> API for its C<ev_signal> (and C<ev_child>) watchers. This API
438 delivers signals synchronously, which makes it both faster and might make
439 it possible to get the queued signal data. It can also simplify signal
440 handling with threads, as long as you properly block signals in your
441 threads that are not interested in handling them.
443 Signalfd will not be used by default as this changes your signal mask, and
444 there are a lot of shoddy libraries and programs (glib's threadpool for
445 example) that can't properly initialise their signal masks.
447 =item C<EVFLAG_NOSIGMASK>
449 When this flag is specified, then libev will avoid to modify the signal
450 mask. Specifically, this means you have to make sure signals are unblocked
451 when you want to receive them.
453 This behaviour is useful when you want to do your own signal handling, or
454 want to handle signals only in specific threads and want to avoid libev
455 unblocking the signals.
457 It's also required by POSIX in a threaded program, as libev calls
458 C<sigprocmask>, whose behaviour is officially unspecified.
460 This flag's behaviour will become the default in future versions of libev.
462 =item C<EVBACKEND_SELECT> (value 1, portable select backend)
464 This is your standard select(2) backend. Not I<completely> standard, as
465 libev tries to roll its own fd_set with no limits on the number of fds,
466 but if that fails, expect a fairly low limit on the number of fds when
467 using this backend. It doesn't scale too well (O(highest_fd)), but its
468 usually the fastest backend for a low number of (low-numbered :) fds.
470 To get good performance out of this backend you need a high amount of
471 parallelism (most of the file descriptors should be busy). If you are
472 writing a server, you should C<accept ()> in a loop to accept as many
473 connections as possible during one iteration. You might also want to have
474 a look at C<ev_set_io_collect_interval ()> to increase the amount of
475 readiness notifications you get per iteration.
477 This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
478 C<writefds> set (and to work around Microsoft Windows bugs, also onto the
479 C<exceptfds> set on that platform).
481 =item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
483 And this is your standard poll(2) backend. It's more complicated
484 than select, but handles sparse fds better and has no artificial
485 limit on the number of fds you can use (except it will slow down
486 considerably with a lot of inactive fds). It scales similarly to select,
487 i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
490 This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
491 C<EV_WRITE> to C<POLLOUT | POLLERR | POLLHUP>.
493 =item C<EVBACKEND_EPOLL> (value 4, Linux)
495 Use the linux-specific epoll(7) interface (for both pre- and post-2.6.9
498 For few fds, this backend is a bit little slower than poll and select, but
499 it scales phenomenally better. While poll and select usually scale like
500 O(total_fds) where total_fds is the total number of fds (or the highest
501 fd), epoll scales either O(1) or O(active_fds).
503 The epoll mechanism deserves honorable mention as the most misdesigned
504 of the more advanced event mechanisms: mere annoyances include silently
505 dropping file descriptors, requiring a system call per change per file
506 descriptor (and unnecessary guessing of parameters), problems with dup,
507 returning before the timeout value, resulting in additional iterations
508 (and only giving 5ms accuracy while select on the same platform gives
509 0.1ms) and so on. The biggest issue is fork races, however - if a program
510 forks then I<both> parent and child process have to recreate the epoll
511 set, which can take considerable time (one syscall per file descriptor)
512 and is of course hard to detect.
514 Epoll is also notoriously buggy - embedding epoll fds I<should> work,
515 but of course I<doesn't>, and epoll just loves to report events for
516 totally I<different> file descriptors (even already closed ones, so
517 one cannot even remove them from the set) than registered in the set
518 (especially on SMP systems). Libev tries to counter these spurious
519 notifications by employing an additional generation counter and comparing
520 that against the events to filter out spurious ones, recreating the set
521 when required. Epoll also erroneously rounds down timeouts, but gives you
522 no way to know when and by how much, so sometimes you have to busy-wait
523 because epoll returns immediately despite a nonzero timeout. And last
524 not least, it also refuses to work with some file descriptors which work
525 perfectly fine with C<select> (files, many character devices...).
527 Epoll is truly the train wreck among event poll mechanisms, a frankenpoll,
528 cobbled together in a hurry, no thought to design or interaction with
529 others. Oh, the pain, will it ever stop...
531 While stopping, setting and starting an I/O watcher in the same iteration
532 will result in some caching, there is still a system call per such
533 incident (because the same I<file descriptor> could point to a different
534 I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
535 file descriptors might not work very well if you register events for both
538 Best performance from this backend is achieved by not unregistering all
539 watchers for a file descriptor until it has been closed, if possible,
540 i.e. keep at least one watcher active per fd at all times. Stopping and
541 starting a watcher (without re-setting it) also usually doesn't cause
542 extra overhead. A fork can both result in spurious notifications as well
543 as in libev having to destroy and recreate the epoll object, which can
544 take considerable time and thus should be avoided.
546 All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
547 faster than epoll for maybe up to a hundred file descriptors, depending on
550 While nominally embeddable in other event loops, this feature is broken in
551 all kernel versions tested so far.
553 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
556 =item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
558 Kqueue deserves special mention, as at the time of this writing, it
559 was broken on all BSDs except NetBSD (usually it doesn't work reliably
560 with anything but sockets and pipes, except on Darwin, where of course
561 it's completely useless). Unlike epoll, however, whose brokenness
562 is by design, these kqueue bugs can (and eventually will) be fixed
563 without API changes to existing programs. For this reason it's not being
564 "auto-detected" unless you explicitly specify it in the flags (i.e. using
565 C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
568 You still can embed kqueue into a normal poll or select backend and use it
569 only for sockets (after having made sure that sockets work with kqueue on
570 the target platform). See C<ev_embed> watchers for more info.
572 It scales in the same way as the epoll backend, but the interface to the
573 kernel is more efficient (which says nothing about its actual speed, of
574 course). While stopping, setting and starting an I/O watcher does never
575 cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
576 two event changes per incident. Support for C<fork ()> is very bad (you
577 might have to leak fd's on fork, but it's more sane than epoll) and it
578 drops fds silently in similarly hard-to-detect cases.
580 This backend usually performs well under most conditions.
582 While nominally embeddable in other event loops, this doesn't work
583 everywhere, so you might need to test for this. And since it is broken
584 almost everywhere, you should only use it when you have a lot of sockets
585 (for which it usually works), by embedding it into another event loop
586 (e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
587 also broken on OS X)) and, did I mention it, using it only for sockets.
589 This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
590 C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
593 =item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
595 This is not implemented yet (and might never be, unless you send me an
596 implementation). According to reports, C</dev/poll> only supports sockets
597 and is not embeddable, which would limit the usefulness of this backend
600 =item C<EVBACKEND_PORT> (value 32, Solaris 10)
602 This uses the Solaris 10 event port mechanism. As with everything on Solaris,
603 it's really slow, but it still scales very well (O(active_fds)).
605 While this backend scales well, it requires one system call per active
606 file descriptor per loop iteration. For small and medium numbers of file
607 descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
608 might perform better.
610 On the positive side, this backend actually performed fully to
611 specification in all tests and is fully embeddable, which is a rare feat
612 among the OS-specific backends (I vastly prefer correctness over speed
615 On the negative side, the interface is I<bizarre> - so bizarre that
616 even sun itself gets it wrong in their code examples: The event polling
617 function sometimes returns events to the caller even though an error
618 occurred, but with no indication whether it has done so or not (yes, it's
619 even documented that way) - deadly for edge-triggered interfaces where you
620 absolutely have to know whether an event occurred or not because you have
621 to re-arm the watcher.
623 Fortunately libev seems to be able to work around these idiocies.
625 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
628 =item C<EVBACKEND_ALL>
630 Try all backends (even potentially broken ones that wouldn't be tried
631 with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
632 C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.
634 It is definitely not recommended to use this flag, use whatever
635 C<ev_recommended_backends ()> returns, or simply do not specify a backend
638 =item C<EVBACKEND_MASK>
640 Not a backend at all, but a mask to select all backend bits from a
641 C<flags> value, in case you want to mask out any backends from a flags
642 value (e.g. when modifying the C<LIBEV_FLAGS> environment variable).
646 If one or more of the backend flags are or'ed into the flags value,
647 then only these backends will be tried (in the reverse order as listed
648 here). If none are specified, all backends in C<ev_recommended_backends
651 Example: Try to create a event loop that uses epoll and nothing else.
653 struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
655 fatal ("no epoll found here, maybe it hides under your chair");
657 Example: Use whatever libev has to offer, but make sure that kqueue is
660 struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_KQUEUE);
662 =item ev_loop_destroy (loop)
664 Destroys an event loop object (frees all memory and kernel state
665 etc.). None of the active event watchers will be stopped in the normal
666 sense, so e.g. C<ev_is_active> might still return true. It is your
667 responsibility to either stop all watchers cleanly yourself I<before>
668 calling this function, or cope with the fact afterwards (which is usually
669 the easiest thing, you can just ignore the watchers and/or C<free ()> them
672 Note that certain global state, such as signal state (and installed signal
673 handlers), will not be freed by this function, and related watchers (such
674 as signal and child watchers) would need to be stopped manually.
676 This function is normally used on loop objects allocated by
677 C<ev_loop_new>, but it can also be used on the default loop returned by
678 C<ev_default_loop>, in which case it is not thread-safe.
680 Note that it is not advisable to call this function on the default loop
681 except in the rare occasion where you really need to free its resources.
682 If you need dynamically allocated loops it is better to use C<ev_loop_new>
683 and C<ev_loop_destroy>.
685 =item ev_loop_fork (loop)
687 This function sets a flag that causes subsequent C<ev_run> iterations to
688 reinitialise the kernel state for backends that have one. Despite the
689 name, you can call it anytime, but it makes most sense after forking, in
690 the child process. You I<must> call it (or use C<EVFLAG_FORKCHECK>) in the
691 child before resuming or calling C<ev_run>.
693 Again, you I<have> to call it on I<any> loop that you want to re-use after
694 a fork, I<even if you do not plan to use the loop in the parent>. This is
695 because some kernel interfaces *cough* I<kqueue> *cough* do funny things
698 On the other hand, you only need to call this function in the child
699 process if and only if you want to use the event loop in the child. If
700 you just fork+exec or create a new loop in the child, you don't have to
701 call it at all (in fact, C<epoll> is so badly broken that it makes a
702 difference, but libev will usually detect this case on its own and do a
703 costly reset of the backend).
705 The function itself is quite fast and it's usually not a problem to call
706 it just in case after a fork.
708 Example: Automate calling C<ev_loop_fork> on the default loop when
712 post_fork_child (void)
714 ev_loop_fork (EV_DEFAULT);
718 pthread_atfork (0, 0, post_fork_child);
720 =item int ev_is_default_loop (loop)
722 Returns true when the given loop is, in fact, the default loop, and false
725 =item unsigned int ev_iteration (loop)
727 Returns the current iteration count for the event loop, which is identical
728 to the number of times libev did poll for new events. It starts at C<0>
729 and happily wraps around with enough iterations.
731 This value can sometimes be useful as a generation counter of sorts (it
732 "ticks" the number of loop iterations), as it roughly corresponds with
733 C<ev_prepare> and C<ev_check> calls - and is incremented between the
734 prepare and check phases.
736 =item unsigned int ev_depth (loop)
738 Returns the number of times C<ev_run> was entered minus the number of
739 times C<ev_run> was exited normally, in other words, the recursion depth.
741 Outside C<ev_run>, this number is zero. In a callback, this number is
742 C<1>, unless C<ev_run> was invoked recursively (or from another thread),
743 in which case it is higher.
745 Leaving C<ev_run> abnormally (setjmp/longjmp, cancelling the thread,
746 throwing an exception etc.), doesn't count as "exit" - consider this
747 as a hint to avoid such ungentleman-like behaviour unless it's really
748 convenient, in which case it is fully supported.
750 =item unsigned int ev_backend (loop)
752 Returns one of the C<EVBACKEND_*> flags indicating the event backend in
755 =item ev_tstamp ev_now (loop)
757 Returns the current "event loop time", which is the time the event loop
758 received events and started processing them. This timestamp does not
759 change as long as callbacks are being processed, and this is also the base
760 time used for relative timers. You can treat it as the timestamp of the
761 event occurring (or more correctly, libev finding out about it).
763 =item ev_now_update (loop)
765 Establishes the current time by querying the kernel, updating the time
766 returned by C<ev_now ()> in the progress. This is a costly operation and
767 is usually done automatically within C<ev_run ()>.
769 This function is rarely useful, but when some event callback runs for a
770 very long time without entering the event loop, updating libev's idea of
771 the current time is a good idea.
773 See also L</The special problem of time updates> in the C<ev_timer> section.
775 =item ev_suspend (loop)
777 =item ev_resume (loop)
779 These two functions suspend and resume an event loop, for use when the
780 loop is not used for a while and timeouts should not be processed.
782 A typical use case would be an interactive program such as a game: When
783 the user presses C<^Z> to suspend the game and resumes it an hour later it
784 would be best to handle timeouts as if no time had actually passed while
785 the program was suspended. This can be achieved by calling C<ev_suspend>
786 in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
787 C<ev_resume> directly afterwards to resume timer processing.
789 Effectively, all C<ev_timer> watchers will be delayed by the time spend
790 between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
791 will be rescheduled (that is, they will lose any events that would have
792 occurred while suspended).
794 After calling C<ev_suspend> you B<must not> call I<any> function on the
795 given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
796 without a previous call to C<ev_suspend>.
798 Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
799 event loop time (see C<ev_now_update>).
801 =item bool ev_run (loop, int flags)
803 Finally, this is it, the event handler. This function usually is called
804 after you have initialised all your watchers and you want to start
805 handling events. It will ask the operating system for any new events, call
806 the watcher callbacks, and then repeat the whole process indefinitely: This
807 is why event loops are called I<loops>.
809 If the flags argument is specified as C<0>, it will keep handling events
810 until either no event watchers are active anymore or C<ev_break> was
813 The return value is false if there are no more active watchers (which
814 usually means "all jobs done" or "deadlock"), and true in all other cases
815 (which usually means " you should call C<ev_run> again").
817 Please note that an explicit C<ev_break> is usually better than
818 relying on all watchers to be stopped when deciding when a program has
819 finished (especially in interactive programs), but having a program
820 that automatically loops as long as it has to and no longer by virtue
821 of relying on its watchers stopping correctly, that is truly a thing of
824 This function is I<mostly> exception-safe - you can break out of a
825 C<ev_run> call by calling C<longjmp> in a callback, throwing a C++
826 exception and so on. This does not decrement the C<ev_depth> value, nor
827 will it clear any outstanding C<EVBREAK_ONE> breaks.
829 A flags value of C<EVRUN_NOWAIT> will look for new events, will handle
830 those events and any already outstanding ones, but will not wait and
831 block your process in case there are no events and will return after one
832 iteration of the loop. This is sometimes useful to poll and handle new
833 events while doing lengthy calculations, to keep the program responsive.
835 A flags value of C<EVRUN_ONCE> will look for new events (waiting if
836 necessary) and will handle those and any already outstanding ones. It
837 will block your process until at least one new event arrives (which could
838 be an event internal to libev itself, so there is no guarantee that a
839 user-registered callback will be called), and will return after one
840 iteration of the loop.
842 This is useful if you are waiting for some external event in conjunction
843 with something not expressible using other libev watchers (i.e. "roll your
844 own C<ev_run>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
845 usually a better approach for this kind of thing.
847 Here are the gory details of what C<ev_run> does (this is for your
848 understanding, not a guarantee that things will work exactly like this in
851 - Increment loop depth.
852 - Reset the ev_break status.
853 - Before the first iteration, call any pending watchers.
855 - If EVFLAG_FORKCHECK was used, check for a fork.
856 - If a fork was detected (by any means), queue and call all fork watchers.
857 - Queue and call all prepare watchers.
858 - If ev_break was called, goto FINISH.
859 - If we have been forked, detach and recreate the kernel state
860 as to not disturb the other process.
861 - Update the kernel state with all outstanding changes.
862 - Update the "event loop time" (ev_now ()).
863 - Calculate for how long to sleep or block, if at all
864 (active idle watchers, EVRUN_NOWAIT or not having
865 any active watchers at all will result in not sleeping).
866 - Sleep if the I/O and timer collect interval say so.
867 - Increment loop iteration counter.
868 - Block the process, waiting for any events.
869 - Queue all outstanding I/O (fd) events.
870 - Update the "event loop time" (ev_now ()), and do time jump adjustments.
871 - Queue all expired timers.
872 - Queue all expired periodics.
873 - Queue all idle watchers with priority higher than that of pending events.
874 - Queue all check watchers.
875 - Call all queued watchers in reverse order (i.e. check watchers first).
876 Signals and child watchers are implemented as I/O watchers, and will
877 be handled here by queueing them when their watcher gets executed.
878 - If ev_break has been called, or EVRUN_ONCE or EVRUN_NOWAIT
879 were used, or there are no active watchers, goto FINISH, otherwise
880 continue with step LOOP.
882 - Reset the ev_break status iff it was EVBREAK_ONE.
883 - Decrement the loop depth.
886 Example: Queue some jobs and then loop until no events are outstanding
889 ... queue jobs here, make sure they register event watchers as long
890 ... as they still have work to do (even an idle watcher will do..)
892 ... jobs done or somebody called break. yeah!
894 =item ev_break (loop, how)
896 Can be used to make a call to C<ev_run> return early (but only after it
897 has processed all outstanding events). The C<how> argument must be either
898 C<EVBREAK_ONE>, which will make the innermost C<ev_run> call return, or
899 C<EVBREAK_ALL>, which will make all nested C<ev_run> calls return.
901 This "break state" will be cleared on the next call to C<ev_run>.
903 It is safe to call C<ev_break> from outside any C<ev_run> calls, too, in
904 which case it will have no effect.
908 =item ev_unref (loop)
910 Ref/unref can be used to add or remove a reference count on the event
911 loop: Every watcher keeps one reference, and as long as the reference
912 count is nonzero, C<ev_run> will not return on its own.
914 This is useful when you have a watcher that you never intend to
915 unregister, but that nevertheless should not keep C<ev_run> from
916 returning. In such a case, call C<ev_unref> after starting, and C<ev_ref>
919 As an example, libev itself uses this for its internal signal pipe: It
920 is not visible to the libev user and should not keep C<ev_run> from
921 exiting if no event watchers registered by it are active. It is also an
922 excellent way to do this for generic recurring timers or from within
923 third-party libraries. Just remember to I<unref after start> and I<ref
924 before stop> (but only if the watcher wasn't active before, or was active
925 before, respectively. Note also that libev might stop watchers itself
926 (e.g. non-repeating timers) in which case you have to C<ev_ref>
929 Example: Create a signal watcher, but keep it from keeping C<ev_run>
930 running when nothing else is active.
933 ev_signal_init (&exitsig, sig_cb, SIGINT);
934 ev_signal_start (loop, &exitsig);
937 Example: For some weird reason, unregister the above signal handler again.
940 ev_signal_stop (loop, &exitsig);
942 =item ev_set_io_collect_interval (loop, ev_tstamp interval)
944 =item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
946 These advanced functions influence the time that libev will spend waiting
947 for events. Both time intervals are by default C<0>, meaning that libev
948 will try to invoke timer/periodic callbacks and I/O callbacks with minimum
951 Setting these to a higher value (the C<interval> I<must> be >= C<0>)
952 allows libev to delay invocation of I/O and timer/periodic callbacks
953 to increase efficiency of loop iterations (or to increase power-saving
956 The idea is that sometimes your program runs just fast enough to handle
957 one (or very few) event(s) per loop iteration. While this makes the
958 program responsive, it also wastes a lot of CPU time to poll for new
959 events, especially with backends like C<select ()> which have a high
960 overhead for the actual polling but can deliver many events at once.
962 By setting a higher I<io collect interval> you allow libev to spend more
963 time collecting I/O events, so you can handle more events per iteration,
964 at the cost of increasing latency. Timeouts (both C<ev_periodic> and
965 C<ev_timer>) will not be affected. Setting this to a non-null value will
966 introduce an additional C<ev_sleep ()> call into most loop iterations. The
967 sleep time ensures that libev will not poll for I/O events more often then
968 once per this interval, on average (as long as the host time resolution is
971 Likewise, by setting a higher I<timeout collect interval> you allow libev
972 to spend more time collecting timeouts, at the expense of increased
973 latency/jitter/inexactness (the watcher callback will be called
974 later). C<ev_io> watchers will not be affected. Setting this to a non-null
975 value will not introduce any overhead in libev.
977 Many (busy) programs can usually benefit by setting the I/O collect
978 interval to a value near C<0.1> or so, which is often enough for
979 interactive servers (of course not for games), likewise for timeouts. It
980 usually doesn't make much sense to set it to a lower value than C<0.01>,
981 as this approaches the timing granularity of most systems. Note that if
982 you do transactions with the outside world and you can't increase the
983 parallelity, then this setting will limit your transaction rate (if you
984 need to poll once per transaction and the I/O collect interval is 0.01,
985 then you can't do more than 100 transactions per second).
987 Setting the I<timeout collect interval> can improve the opportunity for
988 saving power, as the program will "bundle" timer callback invocations that
989 are "near" in time together, by delaying some, thus reducing the number of
990 times the process sleeps and wakes up again. Another useful technique to
991 reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
992 they fire on, say, one-second boundaries only.
994 Example: we only need 0.1s timeout granularity, and we wish not to poll
995 more often than 100 times per second:
997 ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
998 ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
1000 =item ev_invoke_pending (loop)
1002 This call will simply invoke all pending watchers while resetting their
1003 pending state. Normally, C<ev_run> does this automatically when required,
1004 but when overriding the invoke callback this call comes handy. This
1005 function can be invoked from a watcher - this can be useful for example
1006 when you want to do some lengthy calculation and want to pass further
1007 event handling to another thread (you still have to make sure only one
1008 thread executes within C<ev_invoke_pending> or C<ev_run> of course).
1010 =item int ev_pending_count (loop)
1012 Returns the number of pending watchers - zero indicates that no watchers
1015 =item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
1017 This overrides the invoke pending functionality of the loop: Instead of
1018 invoking all pending watchers when there are any, C<ev_run> will call
1019 this callback instead. This is useful, for example, when you want to
1020 invoke the actual watchers inside another context (another thread etc.).
1022 If you want to reset the callback, use C<ev_invoke_pending> as new
1025 =item ev_set_loop_release_cb (loop, void (*release)(EV_P) throw (), void (*acquire)(EV_P) throw ())
1027 Sometimes you want to share the same loop between multiple threads. This
1028 can be done relatively simply by putting mutex_lock/unlock calls around
1029 each call to a libev function.
1031 However, C<ev_run> can run an indefinite time, so it is not feasible
1032 to wait for it to return. One way around this is to wake up the event
1033 loop via C<ev_break> and C<ev_async_send>, another way is to set these
1034 I<release> and I<acquire> callbacks on the loop.
1036 When set, then C<release> will be called just before the thread is
1037 suspended waiting for new events, and C<acquire> is called just
1040 Ideally, C<release> will just call your mutex_unlock function, and
1041 C<acquire> will just call the mutex_lock function again.
1043 While event loop modifications are allowed between invocations of
1044 C<release> and C<acquire> (that's their only purpose after all), no
1045 modifications done will affect the event loop, i.e. adding watchers will
1046 have no effect on the set of file descriptors being watched, or the time
1047 waited. Use an C<ev_async> watcher to wake up C<ev_run> when you want it
1048 to take note of any changes you made.
1050 In theory, threads executing C<ev_run> will be async-cancel safe between
1051 invocations of C<release> and C<acquire>.
1053 See also the locking example in the C<THREADS> section later in this
1056 =item ev_set_userdata (loop, void *data)
1058 =item void *ev_userdata (loop)
1060 Set and retrieve a single C<void *> associated with a loop. When
1061 C<ev_set_userdata> has never been called, then C<ev_userdata> returns
1064 These two functions can be used to associate arbitrary data with a loop,
1065 and are intended solely for the C<invoke_pending_cb>, C<release> and
1066 C<acquire> callbacks described above, but of course can be (ab-)used for
1067 any other purpose as well.
1069 =item ev_verify (loop)
1071 This function only does something when C<EV_VERIFY> support has been
1072 compiled in, which is the default for non-minimal builds. It tries to go
1073 through all internal structures and checks them for validity. If anything
1074 is found to be inconsistent, it will print an error message to standard
1075 error and call C<abort ()>.
1077 This can be used to catch bugs inside libev itself: under normal
1078 circumstances, this function will never abort as of course libev keeps its
1079 data structures consistent.
1084 =head1 ANATOMY OF A WATCHER
1086 In the following description, uppercase C<TYPE> in names stands for the
1087 watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
1088 watchers and C<ev_io_start> for I/O watchers.
1090 A watcher is an opaque structure that you allocate and register to record
1091 your interest in some event. To make a concrete example, imagine you want
1092 to wait for STDIN to become readable, you would create an C<ev_io> watcher
1095 static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
1098 ev_break (loop, EVBREAK_ALL);
1101 struct ev_loop *loop = ev_default_loop (0);
1103 ev_io stdin_watcher;
1105 ev_init (&stdin_watcher, my_cb);
1106 ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
1107 ev_io_start (loop, &stdin_watcher);
1111 As you can see, you are responsible for allocating the memory for your
1112 watcher structures (and it is I<usually> a bad idea to do this on the
1115 Each watcher has an associated watcher structure (called C<struct ev_TYPE>
1116 or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
1118 Each watcher structure must be initialised by a call to C<ev_init (watcher
1119 *, callback)>, which expects a callback to be provided. This callback is
1120 invoked each time the event occurs (or, in the case of I/O watchers, each
1121 time the event loop detects that the file descriptor given is readable
1124 Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
1125 macro to configure it, with arguments specific to the watcher type. There
1126 is also a macro to combine initialisation and setting in one call: C<<
1127 ev_TYPE_init (watcher *, callback, ...) >>.
1129 To make the watcher actually watch out for events, you have to start it
1130 with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
1131 *) >>), and you can stop watching for events at any time by calling the
1132 corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
1134 As long as your watcher is active (has been started but not stopped) you
1135 must not touch the values stored in it. Most specifically you must never
1136 reinitialise it or call its C<ev_TYPE_set> macro.
1138 Each and every callback receives the event loop pointer as first, the
1139 registered watcher structure as second, and a bitset of received events as
1142 The received events usually include a single bit per event type received
1143 (you can receive multiple events at the same time). The possible bit masks
1152 The file descriptor in the C<ev_io> watcher has become readable and/or
1157 The C<ev_timer> watcher has timed out.
1159 =item C<EV_PERIODIC>
1161 The C<ev_periodic> watcher has timed out.
1165 The signal specified in the C<ev_signal> watcher has been received by a thread.
1169 The pid specified in the C<ev_child> watcher has received a status change.
1173 The path specified in the C<ev_stat> watcher changed its attributes somehow.
1177 The C<ev_idle> watcher has determined that you have nothing better to do.
1183 All C<ev_prepare> watchers are invoked just I<before> C<ev_run> starts to
1184 gather new events, and all C<ev_check> watchers are queued (not invoked)
1185 just after C<ev_run> has gathered them, but before it queues any callbacks
1186 for any received events. That means C<ev_prepare> watchers are the last
1187 watchers invoked before the event loop sleeps or polls for new events, and
1188 C<ev_check> watchers will be invoked before any other watchers of the same
1189 or lower priority within an event loop iteration.
1191 Callbacks of both watcher types can start and stop as many watchers as
1192 they want, and all of them will be taken into account (for example, a
1193 C<ev_prepare> watcher might start an idle watcher to keep C<ev_run> from
1198 The embedded event loop specified in the C<ev_embed> watcher needs attention.
1202 The event loop has been resumed in the child process after fork (see
1207 The event loop is about to be destroyed (see C<ev_cleanup>).
1211 The given async watcher has been asynchronously notified (see C<ev_async>).
1215 Not ever sent (or otherwise used) by libev itself, but can be freely used
1216 by libev users to signal watchers (e.g. via C<ev_feed_event>).
1220 An unspecified error has occurred, the watcher has been stopped. This might
1221 happen because the watcher could not be properly started because libev
1222 ran out of memory, a file descriptor was found to be closed or any other
1223 problem. Libev considers these application bugs.
1225 You best act on it by reporting the problem and somehow coping with the
1226 watcher being stopped. Note that well-written programs should not receive
1227 an error ever, so when your watcher receives it, this usually indicates a
1228 bug in your program.
1230 Libev will usually signal a few "dummy" events together with an error, for
1231 example it might indicate that a fd is readable or writable, and if your
1232 callbacks is well-written it can just attempt the operation and cope with
1233 the error from read() or write(). This will not work in multi-threaded
1234 programs, though, as the fd could already be closed and reused for another
1239 =head2 GENERIC WATCHER FUNCTIONS
1243 =item C<ev_init> (ev_TYPE *watcher, callback)
1245 This macro initialises the generic portion of a watcher. The contents
1246 of the watcher object can be arbitrary (so C<malloc> will do). Only
1247 the generic parts of the watcher are initialised, you I<need> to call
1248 the type-specific C<ev_TYPE_set> macro afterwards to initialise the
1249 type-specific parts. For each type there is also a C<ev_TYPE_init> macro
1250 which rolls both calls into one.
1252 You can reinitialise a watcher at any time as long as it has been stopped
1253 (or never started) and there are no pending events outstanding.
1255 The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
1258 Example: Initialise an C<ev_io> watcher in two steps.
1261 ev_init (&w, my_cb);
1262 ev_io_set (&w, STDIN_FILENO, EV_READ);
1264 =item C<ev_TYPE_set> (ev_TYPE *watcher, [args])
1266 This macro initialises the type-specific parts of a watcher. You need to
1267 call C<ev_init> at least once before you call this macro, but you can
1268 call C<ev_TYPE_set> any number of times. You must not, however, call this
1269 macro on a watcher that is active (it can be pending, however, which is a
1270 difference to the C<ev_init> macro).
1272 Although some watcher types do not have type-specific arguments
1273 (e.g. C<ev_prepare>) you still need to call its C<set> macro.
1275 See C<ev_init>, above, for an example.
1277 =item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
1279 This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
1280 calls into a single call. This is the most convenient method to initialise
1281 a watcher. The same limitations apply, of course.
1283 Example: Initialise and set an C<ev_io> watcher in one step.
1285 ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
1287 =item C<ev_TYPE_start> (loop, ev_TYPE *watcher)
1289 Starts (activates) the given watcher. Only active watchers will receive
1290 events. If the watcher is already active nothing will happen.
1292 Example: Start the C<ev_io> watcher that is being abused as example in this
1295 ev_io_start (EV_DEFAULT_UC, &w);
1297 =item C<ev_TYPE_stop> (loop, ev_TYPE *watcher)
1299 Stops the given watcher if active, and clears the pending status (whether
1300 the watcher was active or not).
1302 It is possible that stopped watchers are pending - for example,
1303 non-repeating timers are being stopped when they become pending - but
1304 calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
1305 pending. If you want to free or reuse the memory used by the watcher it is
1306 therefore a good idea to always call its C<ev_TYPE_stop> function.
1308 =item bool ev_is_active (ev_TYPE *watcher)
1310 Returns a true value iff the watcher is active (i.e. it has been started
1311 and not yet been stopped). As long as a watcher is active you must not modify
1314 =item bool ev_is_pending (ev_TYPE *watcher)
1316 Returns a true value iff the watcher is pending, (i.e. it has outstanding
1317 events but its callback has not yet been invoked). As long as a watcher
1318 is pending (but not active) you must not call an init function on it (but
1319 C<ev_TYPE_set> is safe), you must not change its priority, and you must
1320 make sure the watcher is available to libev (e.g. you cannot C<free ()>
1323 =item callback ev_cb (ev_TYPE *watcher)
1325 Returns the callback currently set on the watcher.
1327 =item ev_set_cb (ev_TYPE *watcher, callback)
1329 Change the callback. You can change the callback at virtually any time
1332 =item ev_set_priority (ev_TYPE *watcher, int priority)
1334 =item int ev_priority (ev_TYPE *watcher)
1336 Set and query the priority of the watcher. The priority is a small
1337 integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
1338 (default: C<-2>). Pending watchers with higher priority will be invoked
1339 before watchers with lower priority, but priority will not keep watchers
1340 from being executed (except for C<ev_idle> watchers).
1342 If you need to suppress invocation when higher priority events are pending
1343 you need to look at C<ev_idle> watchers, which provide this functionality.
1345 You I<must not> change the priority of a watcher as long as it is active or
1348 Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
1349 fine, as long as you do not mind that the priority value you query might
1350 or might not have been clamped to the valid range.
1352 The default priority used by watchers when no priority has been set is
1353 always C<0>, which is supposed to not be too high and not be too low :).
1355 See L</WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
1358 =item ev_invoke (loop, ev_TYPE *watcher, int revents)
1360 Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
1361 C<loop> nor C<revents> need to be valid as long as the watcher callback
1362 can deal with that fact, as both are simply passed through to the
1365 =item int ev_clear_pending (loop, ev_TYPE *watcher)
1367 If the watcher is pending, this function clears its pending status and
1368 returns its C<revents> bitset (as if its callback was invoked). If the
1369 watcher isn't pending it does nothing and returns C<0>.
1371 Sometimes it can be useful to "poll" a watcher instead of waiting for its
1372 callback to be invoked, which can be accomplished with this function.
1374 =item ev_feed_event (loop, ev_TYPE *watcher, int revents)
1376 Feeds the given event set into the event loop, as if the specified event
1377 had happened for the specified watcher (which must be a pointer to an
1378 initialised but not necessarily started event watcher). Obviously you must
1379 not free the watcher as long as it has pending events.
1381 Stopping the watcher, letting libev invoke it, or calling
1382 C<ev_clear_pending> will clear the pending event, even if the watcher was
1383 not started in the first place.
1385 See also C<ev_feed_fd_event> and C<ev_feed_signal_event> for related
1386 functions that do not need a watcher.
1390 See also the L</ASSOCIATING CUSTOM DATA WITH A WATCHER> and L</BUILDING YOUR
1391 OWN COMPOSITE WATCHERS> idioms.
1393 =head2 WATCHER STATES
1395 There are various watcher states mentioned throughout this manual -
1396 active, pending and so on. In this section these states and the rules to
1397 transition between them will be described in more detail - and while these
1398 rules might look complicated, they usually do "the right thing".
1404 Before a watcher can be registered with the event loop it has to be
1405 initialised. This can be done with a call to C<ev_TYPE_init>, or calls to
1406 C<ev_init> followed by the watcher-specific C<ev_TYPE_set> function.
1408 In this state it is simply some block of memory that is suitable for
1409 use in an event loop. It can be moved around, freed, reused etc. at
1410 will - as long as you either keep the memory contents intact, or call
1411 C<ev_TYPE_init> again.
1413 =item started/running/active
1415 Once a watcher has been started with a call to C<ev_TYPE_start> it becomes
1416 property of the event loop, and is actively waiting for events. While in
1417 this state it cannot be accessed (except in a few documented ways), moved,
1418 freed or anything else - the only legal thing is to keep a pointer to it,
1419 and call libev functions on it that are documented to work on active watchers.
1423 If a watcher is active and libev determines that an event it is interested
1424 in has occurred (such as a timer expiring), it will become pending. It will
1425 stay in this pending state until either it is stopped or its callback is
1426 about to be invoked, so it is not normally pending inside the watcher
1429 The watcher might or might not be active while it is pending (for example,
1430 an expired non-repeating timer can be pending but no longer active). If it
1431 is stopped, it can be freely accessed (e.g. by calling C<ev_TYPE_set>),
1432 but it is still property of the event loop at this time, so cannot be
1433 moved, freed or reused. And if it is active the rules described in the
1434 previous item still apply.
1436 It is also possible to feed an event on a watcher that is not active (e.g.
1437 via C<ev_feed_event>), in which case it becomes pending without being
1442 A watcher can be stopped implicitly by libev (in which case it might still
1443 be pending), or explicitly by calling its C<ev_TYPE_stop> function. The
1444 latter will clear any pending state the watcher might be in, regardless
1445 of whether it was active or not, so stopping a watcher explicitly before
1446 freeing it is often a good idea.
1448 While stopped (and not pending) the watcher is essentially in the
1449 initialised state, that is, it can be reused, moved, modified in any way
1450 you wish (but when you trash the memory block, you need to C<ev_TYPE_init>
1455 =head2 WATCHER PRIORITY MODELS
1457 Many event loops support I<watcher priorities>, which are usually small
1458 integers that influence the ordering of event callback invocation
1459 between watchers in some way, all else being equal.
1461 In libev, Watcher priorities can be set using C<ev_set_priority>. See its
1462 description for the more technical details such as the actual priority
1465 There are two common ways how these these priorities are being interpreted
1468 In the more common lock-out model, higher priorities "lock out" invocation
1469 of lower priority watchers, which means as long as higher priority
1470 watchers receive events, lower priority watchers are not being invoked.
1472 The less common only-for-ordering model uses priorities solely to order
1473 callback invocation within a single event loop iteration: Higher priority
1474 watchers are invoked before lower priority ones, but they all get invoked
1475 before polling for new events.
1477 Libev uses the second (only-for-ordering) model for all its watchers
1478 except for idle watchers (which use the lock-out model).
1480 The rationale behind this is that implementing the lock-out model for
1481 watchers is not well supported by most kernel interfaces, and most event
1482 libraries will just poll for the same events again and again as long as
1483 their callbacks have not been executed, which is very inefficient in the
1484 common case of one high-priority watcher locking out a mass of lower
1487 Static (ordering) priorities are most useful when you have two or more
1488 watchers handling the same resource: a typical usage example is having an
1489 C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
1490 timeouts. Under load, data might be received while the program handles
1491 other jobs, but since timers normally get invoked first, the timeout
1492 handler will be executed before checking for data. In that case, giving
1493 the timer a lower priority than the I/O watcher ensures that I/O will be
1494 handled first even under adverse conditions (which is usually, but not
1495 always, what you want).
1497 Since idle watchers use the "lock-out" model, meaning that idle watchers
1498 will only be executed when no same or higher priority watchers have
1499 received events, they can be used to implement the "lock-out" model when
1502 For example, to emulate how many other event libraries handle priorities,
1503 you can associate an C<ev_idle> watcher to each such watcher, and in
1504 the normal watcher callback, you just start the idle watcher. The real
1505 processing is done in the idle watcher callback. This causes libev to
1506 continuously poll and process kernel event data for the watcher, but when
1507 the lock-out case is known to be rare (which in turn is rare :), this is
1510 Usually, however, the lock-out model implemented that way will perform
1511 miserably under the type of load it was designed to handle. In that case,
1512 it might be preferable to stop the real watcher before starting the
1513 idle watcher, so the kernel will not have to process the event in case
1514 the actual processing will be delayed for considerable time.
1516 Here is an example of an I/O watcher that should run at a strictly lower
1517 priority than the default, and which should only process data when no
1518 other events are pending:
1520 ev_idle idle; // actual processing watcher
1521 ev_io io; // actual event watcher
1524 io_cb (EV_P_ ev_io *w, int revents)
1526 // stop the I/O watcher, we received the event, but
1527 // are not yet ready to handle it.
1528 ev_io_stop (EV_A_ w);
1530 // start the idle watcher to handle the actual event.
1531 // it will not be executed as long as other watchers
1532 // with the default priority are receiving events.
1533 ev_idle_start (EV_A_ &idle);
1537 idle_cb (EV_P_ ev_idle *w, int revents)
1539 // actual processing
1540 read (STDIN_FILENO, ...);
1542 // have to start the I/O watcher again, as
1543 // we have handled the event
1544 ev_io_start (EV_P_ &io);
1548 ev_idle_init (&idle, idle_cb);
1549 ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
1550 ev_io_start (EV_DEFAULT_ &io);
1552 In the "real" world, it might also be beneficial to start a timer, so that
1553 low-priority connections can not be locked out forever under load. This
1554 enables your program to keep a lower latency for important connections
1555 during short periods of high load, while not completely locking out less
1559 =head1 WATCHER TYPES
1561 This section describes each watcher in detail, but will not repeat
1562 information given in the last section. Any initialisation/set macros,
1563 functions and members specific to the watcher type are explained.
1565 Members are additionally marked with either I<[read-only]>, meaning that,
1566 while the watcher is active, you can look at the member and expect some
1567 sensible content, but you must not modify it (you can modify it while the
1568 watcher is stopped to your hearts content), or I<[read-write]>, which
1569 means you can expect it to have some sensible content while the watcher
1570 is active, but you can also modify it. Modifying it may not do something
1571 sensible or take immediate effect (or do anything at all), but libev will
1572 not crash or malfunction in any way.
1575 =head2 C<ev_io> - is this file descriptor readable or writable?
1577 I/O watchers check whether a file descriptor is readable or writable
1578 in each iteration of the event loop, or, more precisely, when reading
1579 would not block the process and writing would at least be able to write
1580 some data. This behaviour is called level-triggering because you keep
1581 receiving events as long as the condition persists. Remember you can stop
1582 the watcher if you don't want to act on the event and neither want to
1583 receive future events.
1585 In general you can register as many read and/or write event watchers per
1586 fd as you want (as long as you don't confuse yourself). Setting all file
1587 descriptors to non-blocking mode is also usually a good idea (but not
1588 required if you know what you are doing).
1590 Another thing you have to watch out for is that it is quite easy to
1591 receive "spurious" readiness notifications, that is, your callback might
1592 be called with C<EV_READ> but a subsequent C<read>(2) will actually block
1593 because there is no data. It is very easy to get into this situation even
1594 with a relatively standard program structure. Thus it is best to always
1595 use non-blocking I/O: An extra C<read>(2) returning C<EAGAIN> is far
1596 preferable to a program hanging until some data arrives.
1598 If you cannot run the fd in non-blocking mode (for example you should
1599 not play around with an Xlib connection), then you have to separately
1600 re-test whether a file descriptor is really ready with a known-to-be good
1601 interface such as poll (fortunately in the case of Xlib, it already does
1602 this on its own, so its quite safe to use). Some people additionally
1603 use C<SIGALRM> and an interval timer, just to be sure you won't block
1606 But really, best use non-blocking mode.
1608 =head3 The special problem of disappearing file descriptors
1610 Some backends (e.g. kqueue, epoll) need to be told about closing a file
1611 descriptor (either due to calling C<close> explicitly or any other means,
1612 such as C<dup2>). The reason is that you register interest in some file
1613 descriptor, but when it goes away, the operating system will silently drop
1614 this interest. If another file descriptor with the same number then is
1615 registered with libev, there is no efficient way to see that this is, in
1616 fact, a different file descriptor.
1618 To avoid having to explicitly tell libev about such cases, libev follows
1619 the following policy: Each time C<ev_io_set> is being called, libev
1620 will assume that this is potentially a new file descriptor, otherwise
1621 it is assumed that the file descriptor stays the same. That means that
1622 you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
1623 descriptor even if the file descriptor number itself did not change.
1625 This is how one would do it normally anyway, the important point is that
1626 the libev application should not optimise around libev but should leave
1627 optimisations to libev.
1629 =head3 The special problem of dup'ed file descriptors
1631 Some backends (e.g. epoll), cannot register events for file descriptors,
1632 but only events for the underlying file descriptions. That means when you
1633 have C<dup ()>'ed file descriptors or weirder constellations, and register
1634 events for them, only one file descriptor might actually receive events.
1636 There is no workaround possible except not registering events
1637 for potentially C<dup ()>'ed file descriptors, or to resort to
1638 C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
1640 =head3 The special problem of files
1642 Many people try to use C<select> (or libev) on file descriptors
1643 representing files, and expect it to become ready when their program
1644 doesn't block on disk accesses (which can take a long time on their own).
1646 However, this cannot ever work in the "expected" way - you get a readiness
1647 notification as soon as the kernel knows whether and how much data is
1648 there, and in the case of open files, that's always the case, so you
1649 always get a readiness notification instantly, and your read (or possibly
1650 write) will still block on the disk I/O.
1652 Another way to view it is that in the case of sockets, pipes, character
1653 devices and so on, there is another party (the sender) that delivers data
1654 on its own, but in the case of files, there is no such thing: the disk
1655 will not send data on its own, simply because it doesn't know what you
1656 wish to read - you would first have to request some data.
1658 Since files are typically not-so-well supported by advanced notification
1659 mechanism, libev tries hard to emulate POSIX behaviour with respect
1660 to files, even though you should not use it. The reason for this is
1661 convenience: sometimes you want to watch STDIN or STDOUT, which is
1662 usually a tty, often a pipe, but also sometimes files or special devices
1663 (for example, C<epoll> on Linux works with F</dev/random> but not with
1664 F</dev/urandom>), and even though the file might better be served with
1665 asynchronous I/O instead of with non-blocking I/O, it is still useful when
1666 it "just works" instead of freezing.
1668 So avoid file descriptors pointing to files when you know it (e.g. use
1669 libeio), but use them when it is convenient, e.g. for STDIN/STDOUT, or
1670 when you rarely read from a file instead of from a socket, and want to
1671 reuse the same code path.
1673 =head3 The special problem of fork
1675 Some backends (epoll, kqueue) do not support C<fork ()> at all or exhibit
1676 useless behaviour. Libev fully supports fork, but needs to be told about
1677 it in the child if you want to continue to use it in the child.
1679 To support fork in your child processes, you have to call C<ev_loop_fork
1680 ()> after a fork in the child, enable C<EVFLAG_FORKCHECK>, or resort to
1681 C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
1683 =head3 The special problem of SIGPIPE
1685 While not really specific to libev, it is easy to forget about C<SIGPIPE>:
1686 when writing to a pipe whose other end has been closed, your program gets
1687 sent a SIGPIPE, which, by default, aborts your program. For most programs
1688 this is sensible behaviour, for daemons, this is usually undesirable.
1690 So when you encounter spurious, unexplained daemon exits, make sure you
1691 ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
1692 somewhere, as that would have given you a big clue).
1694 =head3 The special problem of accept()ing when you can't
1696 Many implementations of the POSIX C<accept> function (for example,
1697 found in post-2004 Linux) have the peculiar behaviour of not removing a
1698 connection from the pending queue in all error cases.
1700 For example, larger servers often run out of file descriptors (because
1701 of resource limits), causing C<accept> to fail with C<ENFILE> but not
1702 rejecting the connection, leading to libev signalling readiness on
1703 the next iteration again (the connection still exists after all), and
1704 typically causing the program to loop at 100% CPU usage.
1706 Unfortunately, the set of errors that cause this issue differs between
1707 operating systems, there is usually little the app can do to remedy the
1708 situation, and no known thread-safe method of removing the connection to
1709 cope with overload is known (to me).
1711 One of the easiest ways to handle this situation is to just ignore it
1712 - when the program encounters an overload, it will just loop until the
1713 situation is over. While this is a form of busy waiting, no OS offers an
1714 event-based way to handle this situation, so it's the best one can do.
1716 A better way to handle the situation is to log any errors other than
1717 C<EAGAIN> and C<EWOULDBLOCK>, making sure not to flood the log with such
1718 messages, and continue as usual, which at least gives the user an idea of
1719 what could be wrong ("raise the ulimit!"). For extra points one could stop
1720 the C<ev_io> watcher on the listening fd "for a while", which reduces CPU
1723 If your program is single-threaded, then you could also keep a dummy file
1724 descriptor for overload situations (e.g. by opening F</dev/null>), and
1725 when you run into C<ENFILE> or C<EMFILE>, close it, run C<accept>,
1726 close that fd, and create a new dummy fd. This will gracefully refuse
1727 clients under typical overload conditions.
1729 The last way to handle it is to simply log the error and C<exit>, as
1730 is often done with C<malloc> failures, but this results in an easy
1731 opportunity for a DoS attack.
1733 =head3 Watcher-Specific Functions
1737 =item ev_io_init (ev_io *, callback, int fd, int events)
1739 =item ev_io_set (ev_io *, int fd, int events)
1741 Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
1742 receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
1743 C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
1745 =item int fd [read-only]
1747 The file descriptor being watched.
1749 =item int events [read-only]
1751 The events being watched.
1757 Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
1758 readable, but only once. Since it is likely line-buffered, you could
1759 attempt to read a whole line in the callback.
1762 stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
1764 ev_io_stop (loop, w);
1765 .. read from stdin here (or from w->fd) and handle any I/O errors
1769 struct ev_loop *loop = ev_default_init (0);
1770 ev_io stdin_readable;
1771 ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
1772 ev_io_start (loop, &stdin_readable);
1776 =head2 C<ev_timer> - relative and optionally repeating timeouts
1778 Timer watchers are simple relative timers that generate an event after a
1779 given time, and optionally repeating in regular intervals after that.
1781 The timers are based on real time, that is, if you register an event that
1782 times out after an hour and you reset your system clock to January last
1783 year, it will still time out after (roughly) one hour. "Roughly" because
1784 detecting time jumps is hard, and some inaccuracies are unavoidable (the
1785 monotonic clock option helps a lot here).
1787 The callback is guaranteed to be invoked only I<after> its timeout has
1788 passed (not I<at>, so on systems with very low-resolution clocks this
1789 might introduce a small delay, see "the special problem of being too
1790 early", below). If multiple timers become ready during the same loop
1791 iteration then the ones with earlier time-out values are invoked before
1792 ones of the same priority with later time-out values (but this is no
1793 longer true when a callback calls C<ev_run> recursively).
1795 =head3 Be smart about timeouts
1797 Many real-world problems involve some kind of timeout, usually for error
1798 recovery. A typical example is an HTTP request - if the other side hangs,
1799 you want to raise some error after a while.
1801 What follows are some ways to handle this problem, from obvious and
1802 inefficient to smart and efficient.
1804 In the following, a 60 second activity timeout is assumed - a timeout that
1805 gets reset to 60 seconds each time there is activity (e.g. each time some
1806 data or other life sign was received).
1810 =item 1. Use a timer and stop, reinitialise and start it on activity.
1812 This is the most obvious, but not the most simple way: In the beginning,
1815 ev_timer_init (timer, callback, 60., 0.);
1816 ev_timer_start (loop, timer);
1818 Then, each time there is some activity, C<ev_timer_stop> it, initialise it
1821 ev_timer_stop (loop, timer);
1822 ev_timer_set (timer, 60., 0.);
1823 ev_timer_start (loop, timer);
1825 This is relatively simple to implement, but means that each time there is
1826 some activity, libev will first have to remove the timer from its internal
1827 data structure and then add it again. Libev tries to be fast, but it's
1828 still not a constant-time operation.
1830 =item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
1832 This is the easiest way, and involves using C<ev_timer_again> instead of
1835 To implement this, configure an C<ev_timer> with a C<repeat> value
1836 of C<60> and then call C<ev_timer_again> at start and each time you
1837 successfully read or write some data. If you go into an idle state where
1838 you do not expect data to travel on the socket, you can C<ev_timer_stop>
1839 the timer, and C<ev_timer_again> will automatically restart it if need be.
1841 That means you can ignore both the C<ev_timer_start> function and the
1842 C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
1843 member and C<ev_timer_again>.
1847 ev_init (timer, callback);
1848 timer->repeat = 60.;
1849 ev_timer_again (loop, timer);
1851 Each time there is some activity:
1853 ev_timer_again (loop, timer);
1855 It is even possible to change the time-out on the fly, regardless of
1856 whether the watcher is active or not:
1858 timer->repeat = 30.;
1859 ev_timer_again (loop, timer);
1861 This is slightly more efficient then stopping/starting the timer each time
1862 you want to modify its timeout value, as libev does not have to completely
1863 remove and re-insert the timer from/into its internal data structure.
1865 It is, however, even simpler than the "obvious" way to do it.
1867 =item 3. Let the timer time out, but then re-arm it as required.
1869 This method is more tricky, but usually most efficient: Most timeouts are
1870 relatively long compared to the intervals between other activity - in
1871 our example, within 60 seconds, there are usually many I/O events with
1872 associated activity resets.
1874 In this case, it would be more efficient to leave the C<ev_timer> alone,
1875 but remember the time of last activity, and check for a real timeout only
1876 within the callback:
1878 ev_tstamp timeout = 60.;
1879 ev_tstamp last_activity; // time of last activity
1883 callback (EV_P_ ev_timer *w, int revents)
1885 // calculate when the timeout would happen
1886 ev_tstamp after = last_activity - ev_now (EV_A) + timeout;
1888 // if negative, it means we the timeout already occurred
1891 // timeout occurred, take action
1895 // callback was invoked, but there was some recent
1896 // activity. simply restart the timer to time out
1897 // after "after" seconds, which is the earliest time
1898 // the timeout can occur.
1899 ev_timer_set (w, after, 0.);
1900 ev_timer_start (EV_A_ w);
1904 To summarise the callback: first calculate in how many seconds the
1905 timeout will occur (by calculating the absolute time when it would occur,
1906 C<last_activity + timeout>, and subtracting the current time, C<ev_now
1909 If this value is negative, then we are already past the timeout, i.e. we
1910 timed out, and need to do whatever is needed in this case.
1912 Otherwise, we now the earliest time at which the timeout would trigger,
1913 and simply start the timer with this timeout value.
1915 In other words, each time the callback is invoked it will check whether
1916 the timeout occurred. If not, it will simply reschedule itself to check
1917 again at the earliest time it could time out. Rinse. Repeat.
1919 This scheme causes more callback invocations (about one every 60 seconds
1920 minus half the average time between activity), but virtually no calls to
1921 libev to change the timeout.
1923 To start the machinery, simply initialise the watcher and set
1924 C<last_activity> to the current time (meaning there was some activity just
1925 now), then call the callback, which will "do the right thing" and start
1928 last_activity = ev_now (EV_A);
1929 ev_init (&timer, callback);
1930 callback (EV_A_ &timer, 0);
1932 When there is some activity, simply store the current time in
1933 C<last_activity>, no libev calls at all:
1935 if (activity detected)
1936 last_activity = ev_now (EV_A);
1938 When your timeout value changes, then the timeout can be changed by simply
1939 providing a new value, stopping the timer and calling the callback, which
1940 will again do the right thing (for example, time out immediately :).
1942 timeout = new_value;
1943 ev_timer_stop (EV_A_ &timer);
1944 callback (EV_A_ &timer, 0);
1946 This technique is slightly more complex, but in most cases where the
1947 time-out is unlikely to be triggered, much more efficient.
1949 =item 4. Wee, just use a double-linked list for your timeouts.
1951 If there is not one request, but many thousands (millions...), all
1952 employing some kind of timeout with the same timeout value, then one can
1955 When starting the timeout, calculate the timeout value and put the timeout
1956 at the I<end> of the list.
1958 Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
1959 the list is expected to fire (for example, using the technique #3).
1961 When there is some activity, remove the timer from the list, recalculate
1962 the timeout, append it to the end of the list again, and make sure to
1963 update the C<ev_timer> if it was taken from the beginning of the list.
1965 This way, one can manage an unlimited number of timeouts in O(1) time for
1966 starting, stopping and updating the timers, at the expense of a major
1967 complication, and having to use a constant timeout. The constant timeout
1968 ensures that the list stays sorted.
1972 So which method the best?
1974 Method #2 is a simple no-brain-required solution that is adequate in most
1975 situations. Method #3 requires a bit more thinking, but handles many cases
1976 better, and isn't very complicated either. In most case, choosing either
1977 one is fine, with #3 being better in typical situations.
1979 Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
1980 rather complicated, but extremely efficient, something that really pays
1981 off after the first million or so of active timers, i.e. it's usually
1984 =head3 The special problem of being too early
1986 If you ask a timer to call your callback after three seconds, then
1987 you expect it to be invoked after three seconds - but of course, this
1988 cannot be guaranteed to infinite precision. Less obviously, it cannot be
1989 guaranteed to any precision by libev - imagine somebody suspending the
1990 process with a STOP signal for a few hours for example.
1992 So, libev tries to invoke your callback as soon as possible I<after> the
1993 delay has occurred, but cannot guarantee this.
1995 A less obvious failure mode is calling your callback too early: many event
1996 loops compare timestamps with a "elapsed delay >= requested delay", but
1997 this can cause your callback to be invoked much earlier than you would
2000 To see why, imagine a system with a clock that only offers full second
2001 resolution (think windows if you can't come up with a broken enough OS
2002 yourself). If you schedule a one-second timer at the time 500.9, then the
2003 event loop will schedule your timeout to elapse at a system time of 500
2004 (500.9 truncated to the resolution) + 1, or 501.
2006 If an event library looks at the timeout 0.1s later, it will see "501 >=
2007 501" and invoke the callback 0.1s after it was started, even though a
2008 one-second delay was requested - this is being "too early", despite best
2011 This is the reason why libev will never invoke the callback if the elapsed
2012 delay equals the requested delay, but only when the elapsed delay is
2013 larger than the requested delay. In the example above, libev would only invoke
2014 the callback at system time 502, or 1.1s after the timer was started.
2016 So, while libev cannot guarantee that your callback will be invoked
2017 exactly when requested, it I<can> and I<does> guarantee that the requested
2018 delay has actually elapsed, or in other words, it always errs on the "too
2019 late" side of things.
2021 =head3 The special problem of time updates
2023 Establishing the current time is a costly operation (it usually takes
2024 at least one system call): EV therefore updates its idea of the current
2025 time only before and after C<ev_run> collects new events, which causes a
2026 growing difference between C<ev_now ()> and C<ev_time ()> when handling
2027 lots of events in one iteration.
2029 The relative timeouts are calculated relative to the C<ev_now ()>
2030 time. This is usually the right thing as this timestamp refers to the time
2031 of the event triggering whatever timeout you are modifying/starting. If
2032 you suspect event processing to be delayed and you I<need> to base the
2033 timeout on the current time, use something like this to adjust for this:
2035 ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);
2037 If the event loop is suspended for a long time, you can also force an
2038 update of the time returned by C<ev_now ()> by calling C<ev_now_update
2041 =head3 The special problem of unsynchronised clocks
2043 Modern systems have a variety of clocks - libev itself uses the normal
2044 "wall clock" clock and, if available, the monotonic clock (to avoid time
2047 Neither of these clocks is synchronised with each other or any other clock
2048 on the system, so C<ev_time ()> might return a considerably different time
2049 than C<gettimeofday ()> or C<time ()>. On a GNU/Linux system, for example,
2050 a call to C<gettimeofday> might return a second count that is one higher
2051 than a directly following call to C<time>.
2053 The moral of this is to only compare libev-related timestamps with
2054 C<ev_time ()> and C<ev_now ()>, at least if you want better precision than
2057 One more problem arises due to this lack of synchronisation: if libev uses
2058 the system monotonic clock and you compare timestamps from C<ev_time>
2059 or C<ev_now> from when you started your timer and when your callback is
2060 invoked, you will find that sometimes the callback is a bit "early".
2062 This is because C<ev_timer>s work in real time, not wall clock time, so
2063 libev makes sure your callback is not invoked before the delay happened,
2064 I<measured according to the real time>, not the system clock.
2066 If your timeouts are based on a physical timescale (e.g. "time out this
2067 connection after 100 seconds") then this shouldn't bother you as it is
2068 exactly the right behaviour.
2070 If you want to compare wall clock/system timestamps to your timers, then
2071 you need to use C<ev_periodic>s, as these are based on the wall clock
2072 time, where your comparisons will always generate correct results.
2074 =head3 The special problems of suspended animation
2076 When you leave the server world it is quite customary to hit machines that
2077 can suspend/hibernate - what happens to the clocks during such a suspend?
2079 Some quick tests made with a Linux 2.6.28 indicate that a suspend freezes
2080 all processes, while the clocks (C<times>, C<CLOCK_MONOTONIC>) continue
2081 to run until the system is suspended, but they will not advance while the
2082 system is suspended. That means, on resume, it will be as if the program
2083 was frozen for a few seconds, but the suspend time will not be counted
2084 towards C<ev_timer> when a monotonic clock source is used. The real time
2085 clock advanced as expected, but if it is used as sole clocksource, then a
2086 long suspend would be detected as a time jump by libev, and timers would
2087 be adjusted accordingly.
2089 I would not be surprised to see different behaviour in different between
2090 operating systems, OS versions or even different hardware.
2092 The other form of suspend (job control, or sending a SIGSTOP) will see a
2093 time jump in the monotonic clocks and the realtime clock. If the program
2094 is suspended for a very long time, and monotonic clock sources are in use,
2095 then you can expect C<ev_timer>s to expire as the full suspension time
2096 will be counted towards the timers. When no monotonic clock source is in
2097 use, then libev will again assume a timejump and adjust accordingly.
2099 It might be beneficial for this latter case to call C<ev_suspend>
2100 and C<ev_resume> in code that handles C<SIGTSTP>, to at least get
2101 deterministic behaviour in this case (you can do nothing against
2104 =head3 Watcher-Specific Functions and Data Members
2108 =item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
2110 =item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
2112 Configure the timer to trigger after C<after> seconds. If C<repeat>
2113 is C<0.>, then it will automatically be stopped once the timeout is
2114 reached. If it is positive, then the timer will automatically be
2115 configured to trigger again C<repeat> seconds later, again, and again,
2116 until stopped manually.
2118 The timer itself will do a best-effort at avoiding drift, that is, if
2119 you configure a timer to trigger every 10 seconds, then it will normally
2120 trigger at exactly 10 second intervals. If, however, your program cannot
2121 keep up with the timer (because it takes longer than those 10 seconds to
2122 do stuff) the timer will not fire more than once per event loop iteration.
2124 =item ev_timer_again (loop, ev_timer *)
2126 This will act as if the timer timed out, and restarts it again if it is
2127 repeating. It basically works like calling C<ev_timer_stop>, updating the
2128 timeout to the C<repeat> value and calling C<ev_timer_start>.
2130 The exact semantics are as in the following rules, all of which will be
2131 applied to the watcher:
2135 =item If the timer is pending, the pending status is always cleared.
2137 =item If the timer is started but non-repeating, stop it (as if it timed
2138 out, without invoking it).
2140 =item If the timer is repeating, make the C<repeat> value the new timeout
2141 and start the timer, if necessary.
2145 This sounds a bit complicated, see L</Be smart about timeouts>, above, for a
2148 =item ev_tstamp ev_timer_remaining (loop, ev_timer *)
2150 Returns the remaining time until a timer fires. If the timer is active,
2151 then this time is relative to the current event loop time, otherwise it's
2152 the timeout value currently configured.
2154 That is, after an C<ev_timer_set (w, 5, 7)>, C<ev_timer_remaining> returns
2155 C<5>. When the timer is started and one second passes, C<ev_timer_remaining>
2156 will return C<4>. When the timer expires and is restarted, it will return
2157 roughly C<7> (likely slightly less as callback invocation takes some time,
2160 =item ev_tstamp repeat [read-write]
2162 The current C<repeat> value. Will be used each time the watcher times out
2163 or C<ev_timer_again> is called, and determines the next timeout (if any),
2164 which is also when any modifications are taken into account.
2170 Example: Create a timer that fires after 60 seconds.
2173 one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
2175 .. one minute over, w is actually stopped right here
2179 ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
2180 ev_timer_start (loop, &mytimer);
2182 Example: Create a timeout timer that times out after 10 seconds of
2186 timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
2188 .. ten seconds without any activity
2192 ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
2193 ev_timer_again (&mytimer); /* start timer */
2196 // and in some piece of code that gets executed on any "activity":
2197 // reset the timeout to start ticking again at 10 seconds
2198 ev_timer_again (&mytimer);
2201 =head2 C<ev_periodic> - to cron or not to cron?
2203 Periodic watchers are also timers of a kind, but they are very versatile
2204 (and unfortunately a bit complex).
2206 Unlike C<ev_timer>, periodic watchers are not based on real time (or
2207 relative time, the physical time that passes) but on wall clock time
2208 (absolute time, the thing you can read on your calender or clock). The
2209 difference is that wall clock time can run faster or slower than real
2210 time, and time jumps are not uncommon (e.g. when you adjust your
2213 You can tell a periodic watcher to trigger after some specific point
2214 in time: for example, if you tell a periodic watcher to trigger "in 10
2215 seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
2216 not a delay) and then reset your system clock to January of the previous
2217 year, then it will take a year or more to trigger the event (unlike an
2218 C<ev_timer>, which would still trigger roughly 10 seconds after starting
2219 it, as it uses a relative timeout).
2221 C<ev_periodic> watchers can also be used to implement vastly more complex
2222 timers, such as triggering an event on each "midnight, local time", or
2223 other complicated rules. This cannot be done with C<ev_timer> watchers, as
2224 those cannot react to time jumps.
2226 As with timers, the callback is guaranteed to be invoked only when the
2227 point in time where it is supposed to trigger has passed. If multiple
2228 timers become ready during the same loop iteration then the ones with
2229 earlier time-out values are invoked before ones with later time-out values
2230 (but this is no longer true when a callback calls C<ev_run> recursively).
2232 =head3 Watcher-Specific Functions and Data Members
2236 =item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
2238 =item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
2240 Lots of arguments, let's sort it out... There are basically three modes of
2241 operation, and we will explain them from simplest to most complex:
2245 =item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
2247 In this configuration the watcher triggers an event after the wall clock
2248 time C<offset> has passed. It will not repeat and will not adjust when a
2249 time jump occurs, that is, if it is to be run at January 1st 2011 then it
2250 will be stopped and invoked when the system clock reaches or surpasses
2253 =item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
2255 In this mode the watcher will always be scheduled to time out at the next
2256 C<offset + N * interval> time (for some integer N, which can also be
2257 negative) and then repeat, regardless of any time jumps. The C<offset>
2258 argument is merely an offset into the C<interval> periods.
2260 This can be used to create timers that do not drift with respect to the
2261 system clock, for example, here is an C<ev_periodic> that triggers each
2262 hour, on the hour (with respect to UTC):
2264 ev_periodic_set (&periodic, 0., 3600., 0);
2266 This doesn't mean there will always be 3600 seconds in between triggers,
2267 but only that the callback will be called when the system time shows a
2268 full hour (UTC), or more correctly, when the system time is evenly divisible
2271 Another way to think about it (for the mathematically inclined) is that
2272 C<ev_periodic> will try to run the callback in this mode at the next possible
2273 time where C<time = offset (mod interval)>, regardless of any time jumps.
2275 The C<interval> I<MUST> be positive, and for numerical stability, the
2276 interval value should be higher than C<1/8192> (which is around 100
2277 microseconds) and C<offset> should be higher than C<0> and should have
2278 at most a similar magnitude as the current time (say, within a factor of
2279 ten). Typical values for offset are, in fact, C<0> or something between
2280 C<0> and C<interval>, which is also the recommended range.
2282 Note also that there is an upper limit to how often a timer can fire (CPU
2283 speed for example), so if C<interval> is very small then timing stability
2284 will of course deteriorate. Libev itself tries to be exact to be about one
2285 millisecond (if the OS supports it and the machine is fast enough).
2287 =item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
2289 In this mode the values for C<interval> and C<offset> are both being
2290 ignored. Instead, each time the periodic watcher gets scheduled, the
2291 reschedule callback will be called with the watcher as first, and the
2292 current time as second argument.
2294 NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, ever,
2295 or make ANY other event loop modifications whatsoever, unless explicitly
2296 allowed by documentation here>.
2298 If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
2299 it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
2300 only event loop modification you are allowed to do).
2302 The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
2303 *w, ev_tstamp now)>, e.g.:
2306 my_rescheduler (ev_periodic *w, ev_tstamp now)
2311 It must return the next time to trigger, based on the passed time value
2312 (that is, the lowest time value larger than to the second argument). It
2313 will usually be called just before the callback will be triggered, but
2314 might be called at other times, too.
2316 NOTE: I<< This callback must always return a time that is higher than or
2317 equal to the passed C<now> value >>.
2319 This can be used to create very complex timers, such as a timer that
2320 triggers on "next midnight, local time". To do this, you would calculate the
2321 next midnight after C<now> and return the timestamp value for this. How
2322 you do this is, again, up to you (but it is not trivial, which is the main
2323 reason I omitted it as an example).
2327 =item ev_periodic_again (loop, ev_periodic *)
2329 Simply stops and restarts the periodic watcher again. This is only useful
2330 when you changed some parameters or the reschedule callback would return
2331 a different time than the last time it was called (e.g. in a crond like
2332 program when the crontabs have changed).
2334 =item ev_tstamp ev_periodic_at (ev_periodic *)
2336 When active, returns the absolute time that the watcher is supposed
2337 to trigger next. This is not the same as the C<offset> argument to
2338 C<ev_periodic_set>, but indeed works even in interval and manual
2341 =item ev_tstamp offset [read-write]
2343 When repeating, this contains the offset value, otherwise this is the
2344 absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
2345 although libev might modify this value for better numerical stability).
2347 Can be modified any time, but changes only take effect when the periodic
2348 timer fires or C<ev_periodic_again> is being called.
2350 =item ev_tstamp interval [read-write]
2352 The current interval value. Can be modified any time, but changes only
2353 take effect when the periodic timer fires or C<ev_periodic_again> is being
2356 =item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
2358 The current reschedule callback, or C<0>, if this functionality is
2359 switched off. Can be changed any time, but changes only take effect when
2360 the periodic timer fires or C<ev_periodic_again> is being called.
2366 Example: Call a callback every hour, or, more precisely, whenever the
2367 system time is divisible by 3600. The callback invocation times have
2368 potentially a lot of jitter, but good long-term stability.
2371 clock_cb (struct ev_loop *loop, ev_periodic *w, int revents)
2373 ... its now a full hour (UTC, or TAI or whatever your clock follows)
2376 ev_periodic hourly_tick;
2377 ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
2378 ev_periodic_start (loop, &hourly_tick);
2380 Example: The same as above, but use a reschedule callback to do it:
2385 my_scheduler_cb (ev_periodic *w, ev_tstamp now)
2387 return now + (3600. - fmod (now, 3600.));
2390 ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
2392 Example: Call a callback every hour, starting now:
2394 ev_periodic hourly_tick;
2395 ev_periodic_init (&hourly_tick, clock_cb,
2396 fmod (ev_now (loop), 3600.), 3600., 0);
2397 ev_periodic_start (loop, &hourly_tick);
2400 =head2 C<ev_signal> - signal me when a signal gets signalled!
2402 Signal watchers will trigger an event when the process receives a specific
2403 signal one or more times. Even though signals are very asynchronous, libev
2404 will try its best to deliver signals synchronously, i.e. as part of the
2405 normal event processing, like any other event.
2407 If you want signals to be delivered truly asynchronously, just use
2408 C<sigaction> as you would do without libev and forget about sharing
2409 the signal. You can even use C<ev_async> from a signal handler to
2410 synchronously wake up an event loop.
2412 You can configure as many watchers as you like for the same signal, but
2413 only within the same loop, i.e. you can watch for C<SIGINT> in your
2414 default loop and for C<SIGIO> in another loop, but you cannot watch for
2415 C<SIGINT> in both the default loop and another loop at the same time. At
2416 the moment, C<SIGCHLD> is permanently tied to the default loop.
2418 Only after the first watcher for a signal is started will libev actually
2419 register something with the kernel. It thus coexists with your own signal
2420 handlers as long as you don't register any with libev for the same signal.
2422 If possible and supported, libev will install its handlers with
2423 C<SA_RESTART> (or equivalent) behaviour enabled, so system calls should
2424 not be unduly interrupted. If you have a problem with system calls getting
2425 interrupted by signals you can block all signals in an C<ev_check> watcher
2426 and unblock them in an C<ev_prepare> watcher.
2428 =head3 The special problem of inheritance over fork/execve/pthread_create
2430 Both the signal mask (C<sigprocmask>) and the signal disposition
2431 (C<sigaction>) are unspecified after starting a signal watcher (and after
2432 stopping it again), that is, libev might or might not block the signal,
2433 and might or might not set or restore the installed signal handler (but
2434 see C<EVFLAG_NOSIGMASK>).
2436 While this does not matter for the signal disposition (libev never
2437 sets signals to C<SIG_IGN>, so handlers will be reset to C<SIG_DFL> on
2438 C<execve>), this matters for the signal mask: many programs do not expect
2439 certain signals to be blocked.
2441 This means that before calling C<exec> (from the child) you should reset
2442 the signal mask to whatever "default" you expect (all clear is a good
2445 The simplest way to ensure that the signal mask is reset in the child is
2446 to install a fork handler with C<pthread_atfork> that resets it. That will
2447 catch fork calls done by libraries (such as the libc) as well.
2449 In current versions of libev, the signal will not be blocked indefinitely
2450 unless you use the C<signalfd> API (C<EV_SIGNALFD>). While this reduces
2451 the window of opportunity for problems, it will not go away, as libev
2452 I<has> to modify the signal mask, at least temporarily.
2454 So I can't stress this enough: I<If you do not reset your signal mask when
2455 you expect it to be empty, you have a race condition in your code>. This
2456 is not a libev-specific thing, this is true for most event libraries.
2458 =head3 The special problem of threads signal handling
2460 POSIX threads has problematic signal handling semantics, specifically,
2461 a lot of functionality (sigfd, sigwait etc.) only really works if all
2462 threads in a process block signals, which is hard to achieve.
2464 When you want to use sigwait (or mix libev signal handling with your own
2465 for the same signals), you can tackle this problem by globally blocking
2466 all signals before creating any threads (or creating them with a fully set
2467 sigprocmask) and also specifying the C<EVFLAG_NOSIGMASK> when creating
2468 loops. Then designate one thread as "signal receiver thread" which handles
2469 these signals. You can pass on any signals that libev might be interested
2470 in by calling C<ev_feed_signal>.
2472 =head3 Watcher-Specific Functions and Data Members
2476 =item ev_signal_init (ev_signal *, callback, int signum)
2478 =item ev_signal_set (ev_signal *, int signum)
2480 Configures the watcher to trigger on the given signal number (usually one
2481 of the C<SIGxxx> constants).
2483 =item int signum [read-only]
2485 The signal the watcher watches out for.
2491 Example: Try to exit cleanly on SIGINT.
2494 sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
2496 ev_break (loop, EVBREAK_ALL);
2499 ev_signal signal_watcher;
2500 ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
2501 ev_signal_start (loop, &signal_watcher);
2504 =head2 C<ev_child> - watch out for process status changes
2506 Child watchers trigger when your process receives a SIGCHLD in response to
2507 some child status changes (most typically when a child of yours dies or
2508 exits). It is permissible to install a child watcher I<after> the child
2509 has been forked (which implies it might have already exited), as long
2510 as the event loop isn't entered (or is continued from a watcher), i.e.,
2511 forking and then immediately registering a watcher for the child is fine,
2512 but forking and registering a watcher a few event loop iterations later or
2513 in the next callback invocation is not.
2515 Only the default event loop is capable of handling signals, and therefore
2516 you can only register child watchers in the default event loop.
2518 Due to some design glitches inside libev, child watchers will always be
2519 handled at maximum priority (their priority is set to C<EV_MAXPRI> by
2522 =head3 Process Interaction
2524 Libev grabs C<SIGCHLD> as soon as the default event loop is
2525 initialised. This is necessary to guarantee proper behaviour even if the
2526 first child watcher is started after the child exits. The occurrence
2527 of C<SIGCHLD> is recorded asynchronously, but child reaping is done
2528 synchronously as part of the event loop processing. Libev always reaps all
2529 children, even ones not watched.
2531 =head3 Overriding the Built-In Processing
2533 Libev offers no special support for overriding the built-in child
2534 processing, but if your application collides with libev's default child
2535 handler, you can override it easily by installing your own handler for
2536 C<SIGCHLD> after initialising the default loop, and making sure the
2537 default loop never gets destroyed. You are encouraged, however, to use an
2538 event-based approach to child reaping and thus use libev's support for
2539 that, so other libev users can use C<ev_child> watchers freely.
2541 =head3 Stopping the Child Watcher
2543 Currently, the child watcher never gets stopped, even when the
2544 child terminates, so normally one needs to stop the watcher in the
2545 callback. Future versions of libev might stop the watcher automatically
2546 when a child exit is detected (calling C<ev_child_stop> twice is not a
2549 =head3 Watcher-Specific Functions and Data Members
2553 =item ev_child_init (ev_child *, callback, int pid, int trace)
2555 =item ev_child_set (ev_child *, int pid, int trace)
2557 Configures the watcher to wait for status changes of process C<pid> (or
2558 I<any> process if C<pid> is specified as C<0>). The callback can look
2559 at the C<rstatus> member of the C<ev_child> watcher structure to see
2560 the status word (use the macros from C<sys/wait.h> and see your systems
2561 C<waitpid> documentation). The C<rpid> member contains the pid of the
2562 process causing the status change. C<trace> must be either C<0> (only
2563 activate the watcher when the process terminates) or C<1> (additionally
2564 activate the watcher when the process is stopped or continued).
2566 =item int pid [read-only]
2568 The process id this watcher watches out for, or C<0>, meaning any process id.
2570 =item int rpid [read-write]
2572 The process id that detected a status change.
2574 =item int rstatus [read-write]
2576 The process exit/trace status caused by C<rpid> (see your systems
2577 C<waitpid> and C<sys/wait.h> documentation for details).
2583 Example: C<fork()> a new process and install a child handler to wait for
2589 child_cb (EV_P_ ev_child *w, int revents)
2591 ev_child_stop (EV_A_ w);
2592 printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
2595 pid_t pid = fork ();
2601 // the forked child executes here
2606 ev_child_init (&cw, child_cb, pid, 0);
2607 ev_child_start (EV_DEFAULT_ &cw);
2611 =head2 C<ev_stat> - did the file attributes just change?
2613 This watches a file system path for attribute changes. That is, it calls
2614 C<stat> on that path in regular intervals (or when the OS says it changed)
2615 and sees if it changed compared to the last time, invoking the callback
2616 if it did. Starting the watcher C<stat>'s the file, so only changes that
2617 happen after the watcher has been started will be reported.
2619 The path does not need to exist: changing from "path exists" to "path does
2620 not exist" is a status change like any other. The condition "path does not
2621 exist" (or more correctly "path cannot be stat'ed") is signified by the
2622 C<st_nlink> field being zero (which is otherwise always forced to be at
2623 least one) and all the other fields of the stat buffer having unspecified
2626 The path I<must not> end in a slash or contain special components such as
2627 C<.> or C<..>. The path I<should> be absolute: If it is relative and
2628 your working directory changes, then the behaviour is undefined.
2630 Since there is no portable change notification interface available, the
2631 portable implementation simply calls C<stat(2)> regularly on the path
2632 to see if it changed somehow. You can specify a recommended polling
2633 interval for this case. If you specify a polling interval of C<0> (highly
2634 recommended!) then a I<suitable, unspecified default> value will be used
2635 (which you can expect to be around five seconds, although this might
2636 change dynamically). Libev will also impose a minimum interval which is
2637 currently around C<0.1>, but that's usually overkill.
2639 This watcher type is not meant for massive numbers of stat watchers,
2640 as even with OS-supported change notifications, this can be
2643 At the time of this writing, the only OS-specific interface implemented
2644 is the Linux inotify interface (implementing kqueue support is left as an
2645 exercise for the reader. Note, however, that the author sees no way of
2646 implementing C<ev_stat> semantics with kqueue, except as a hint).
2648 =head3 ABI Issues (Largefile Support)
2650 Libev by default (unless the user overrides this) uses the default
2651 compilation environment, which means that on systems with large file
2652 support disabled by default, you get the 32 bit version of the stat
2653 structure. When using the library from programs that change the ABI to
2654 use 64 bit file offsets the programs will fail. In that case you have to
2655 compile libev with the same flags to get binary compatibility. This is
2656 obviously the case with any flags that change the ABI, but the problem is
2657 most noticeably displayed with ev_stat and large file support.
2659 The solution for this is to lobby your distribution maker to make large
2660 file interfaces available by default (as e.g. FreeBSD does) and not
2661 optional. Libev cannot simply switch on large file support because it has
2662 to exchange stat structures with application programs compiled using the
2663 default compilation environment.
2665 =head3 Inotify and Kqueue
2667 When C<inotify (7)> support has been compiled into libev and present at
2668 runtime, it will be used to speed up change detection where possible. The
2669 inotify descriptor will be created lazily when the first C<ev_stat>
2670 watcher is being started.
2672 Inotify presence does not change the semantics of C<ev_stat> watchers
2673 except that changes might be detected earlier, and in some cases, to avoid
2674 making regular C<stat> calls. Even in the presence of inotify support
2675 there are many cases where libev has to resort to regular C<stat> polling,
2676 but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
2677 many bugs), the path exists (i.e. stat succeeds), and the path resides on
2678 a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
2679 xfs are fully working) libev usually gets away without polling.
2681 There is no support for kqueue, as apparently it cannot be used to
2682 implement this functionality, due to the requirement of having a file
2683 descriptor open on the object at all times, and detecting renames, unlinks
2686 =head3 C<stat ()> is a synchronous operation
2688 Libev doesn't normally do any kind of I/O itself, and so is not blocking
2689 the process. The exception are C<ev_stat> watchers - those call C<stat
2690 ()>, which is a synchronous operation.
2692 For local paths, this usually doesn't matter: unless the system is very
2693 busy or the intervals between stat's are large, a stat call will be fast,
2694 as the path data is usually in memory already (except when starting the
2697 For networked file systems, calling C<stat ()> can block an indefinite
2698 time due to network issues, and even under good conditions, a stat call
2699 often takes multiple milliseconds.
2701 Therefore, it is best to avoid using C<ev_stat> watchers on networked
2702 paths, although this is fully supported by libev.
2704 =head3 The special problem of stat time resolution
2706 The C<stat ()> system call only supports full-second resolution portably,
2707 and even on systems where the resolution is higher, most file systems
2708 still only support whole seconds.
2710 That means that, if the time is the only thing that changes, you can
2711 easily miss updates: on the first update, C<ev_stat> detects a change and
2712 calls your callback, which does something. When there is another update
2713 within the same second, C<ev_stat> will be unable to detect unless the
2714 stat data does change in other ways (e.g. file size).
2716 The solution to this is to delay acting on a change for slightly more
2717 than a second (or till slightly after the next full second boundary), using
2718 a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
2719 ev_timer_again (loop, w)>).
2721 The C<.02> offset is added to work around small timing inconsistencies
2722 of some operating systems (where the second counter of the current time
2723 might be be delayed. One such system is the Linux kernel, where a call to
2724 C<gettimeofday> might return a timestamp with a full second later than
2725 a subsequent C<time> call - if the equivalent of C<time ()> is used to
2726 update file times then there will be a small window where the kernel uses
2727 the previous second to update file times but libev might already execute
2728 the timer callback).
2730 =head3 Watcher-Specific Functions and Data Members
2734 =item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
2736 =item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
2738 Configures the watcher to wait for status changes of the given
2739 C<path>. The C<interval> is a hint on how quickly a change is expected to
2740 be detected and should normally be specified as C<0> to let libev choose
2741 a suitable value. The memory pointed to by C<path> must point to the same
2742 path for as long as the watcher is active.
2744 The callback will receive an C<EV_STAT> event when a change was detected,
2745 relative to the attributes at the time the watcher was started (or the
2746 last change was detected).
2748 =item ev_stat_stat (loop, ev_stat *)
2750 Updates the stat buffer immediately with new values. If you change the
2751 watched path in your callback, you could call this function to avoid
2752 detecting this change (while introducing a race condition if you are not
2753 the only one changing the path). Can also be useful simply to find out the
2756 =item ev_statdata attr [read-only]
2758 The most-recently detected attributes of the file. Although the type is
2759 C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
2760 suitable for your system, but you can only rely on the POSIX-standardised
2761 members to be present. If the C<st_nlink> member is C<0>, then there was
2762 some error while C<stat>ing the file.
2764 =item ev_statdata prev [read-only]
2766 The previous attributes of the file. The callback gets invoked whenever
2767 C<prev> != C<attr>, or, more precisely, one or more of these members
2768 differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
2769 C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
2771 =item ev_tstamp interval [read-only]
2773 The specified interval.
2775 =item const char *path [read-only]
2777 The file system path that is being watched.
2783 Example: Watch C</etc/passwd> for attribute changes.
2786 passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
2788 /* /etc/passwd changed in some way */
2789 if (w->attr.st_nlink)
2791 printf ("passwd current size %ld\n", (long)w->attr.st_size);
2792 printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
2793 printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
2796 /* you shalt not abuse printf for puts */
2797 puts ("wow, /etc/passwd is not there, expect problems. "
2798 "if this is windows, they already arrived\n");
2804 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
2805 ev_stat_start (loop, &passwd);
2807 Example: Like above, but additionally use a one-second delay so we do not
2808 miss updates (however, frequent updates will delay processing, too, so
2809 one might do the work both on C<ev_stat> callback invocation I<and> on
2810 C<ev_timer> callback invocation).
2812 static ev_stat passwd;
2813 static ev_timer timer;
2816 timer_cb (EV_P_ ev_timer *w, int revents)
2818 ev_timer_stop (EV_A_ w);
2820 /* now it's one second after the most recent passwd change */
2824 stat_cb (EV_P_ ev_stat *w, int revents)
2826 /* reset the one-second timer */
2827 ev_timer_again (EV_A_ &timer);
2831 ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
2832 ev_stat_start (loop, &passwd);
2833 ev_timer_init (&timer, timer_cb, 0., 1.02);
2836 =head2 C<ev_idle> - when you've got nothing better to do...
2838 Idle watchers trigger events when no other events of the same or higher
2839 priority are pending (prepare, check and other idle watchers do not count
2840 as receiving "events").
2842 That is, as long as your process is busy handling sockets or timeouts
2843 (or even signals, imagine) of the same or higher priority it will not be
2844 triggered. But when your process is idle (or only lower-priority watchers
2845 are pending), the idle watchers are being called once per event loop
2846 iteration - until stopped, that is, or your process receives more events
2847 and becomes busy again with higher priority stuff.
2849 The most noteworthy effect is that as long as any idle watchers are
2850 active, the process will not block when waiting for new events.
2852 Apart from keeping your process non-blocking (which is a useful
2853 effect on its own sometimes), idle watchers are a good place to do
2854 "pseudo-background processing", or delay processing stuff to after the
2855 event loop has handled all outstanding events.
2857 =head3 Abusing an C<ev_idle> watcher for its side-effect
2859 As long as there is at least one active idle watcher, libev will never
2860 sleep unnecessarily. Or in other words, it will loop as fast as possible.
2861 For this to work, the idle watcher doesn't need to be invoked at all - the
2862 lowest priority will do.
2864 This mode of operation can be useful together with an C<ev_check> watcher,
2865 to do something on each event loop iteration - for example to balance load
2866 between different connections.
2868 See L</Abusing an ev_check watcher for its side-effect> for a longer
2871 =head3 Watcher-Specific Functions and Data Members
2875 =item ev_idle_init (ev_idle *, callback)
2877 Initialises and configures the idle watcher - it has no parameters of any
2878 kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
2885 Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
2886 callback, free it. Also, use no error checking, as usual.
2889 idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
2892 ev_idle_stop (loop, w);
2894 // now we can free it
2897 // now do something you wanted to do when the program has
2898 // no longer anything immediate to do.
2901 ev_idle *idle_watcher = malloc (sizeof (ev_idle));
2902 ev_idle_init (idle_watcher, idle_cb);
2903 ev_idle_start (loop, idle_watcher);
2906 =head2 C<ev_prepare> and C<ev_check> - customise your event loop!
2908 Prepare and check watchers are often (but not always) used in pairs:
2909 prepare watchers get invoked before the process blocks and check watchers
2912 You I<must not> call C<ev_run> or similar functions that enter
2913 the current event loop from either C<ev_prepare> or C<ev_check>
2914 watchers. Other loops than the current one are fine, however. The
2915 rationale behind this is that you do not need to check for recursion in
2916 those watchers, i.e. the sequence will always be C<ev_prepare>, blocking,
2917 C<ev_check> so if you have one watcher of each kind they will always be
2918 called in pairs bracketing the blocking call.
2920 Their main purpose is to integrate other event mechanisms into libev and
2921 their use is somewhat advanced. They could be used, for example, to track
2922 variable changes, implement your own watchers, integrate net-snmp or a
2923 coroutine library and lots more. They are also occasionally useful if
2924 you cache some data and want to flush it before blocking (for example,
2925 in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
2928 This is done by examining in each prepare call which file descriptors
2929 need to be watched by the other library, registering C<ev_io> watchers
2930 for them and starting an C<ev_timer> watcher for any timeouts (many
2931 libraries provide exactly this functionality). Then, in the check watcher,
2932 you check for any events that occurred (by checking the pending status
2933 of all watchers and stopping them) and call back into the library. The
2934 I/O and timer callbacks will never actually be called (but must be valid
2935 nevertheless, because you never know, you know?).
2937 As another example, the Perl Coro module uses these hooks to integrate
2938 coroutines into libev programs, by yielding to other active coroutines
2939 during each prepare and only letting the process block if no coroutines
2940 are ready to run (it's actually more complicated: it only runs coroutines
2941 with priority higher than or equal to the event loop and one coroutine
2942 of lower priority, but only once, using idle watchers to keep the event
2943 loop from blocking if lower-priority coroutines are active, thus mapping
2944 low-priority coroutines to idle/background tasks).
2946 When used for this purpose, it is recommended to give C<ev_check> watchers
2947 highest (C<EV_MAXPRI>) priority, to ensure that they are being run before
2948 any other watchers after the poll (this doesn't matter for C<ev_prepare>
2951 Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
2952 activate ("feed") events into libev. While libev fully supports this, they
2953 might get executed before other C<ev_check> watchers did their job. As
2954 C<ev_check> watchers are often used to embed other (non-libev) event
2955 loops those other event loops might be in an unusable state until their
2956 C<ev_check> watcher ran (always remind yourself to coexist peacefully with
2959 =head3 Abusing an C<ev_check> watcher for its side-effect
2961 C<ev_check> (and less often also C<ev_prepare>) watchers can also be
2962 useful because they are called once per event loop iteration. For
2963 example, if you want to handle a large number of connections fairly, you
2964 normally only do a bit of work for each active connection, and if there
2965 is more work to do, you wait for the next event loop iteration, so other
2966 connections have a chance of making progress.
2968 Using an C<ev_check> watcher is almost enough: it will be called on the
2969 next event loop iteration. However, that isn't as soon as possible -
2970 without external events, your C<ev_check> watcher will not be invoked.
2972 This is where C<ev_idle> watchers come in handy - all you need is a
2973 single global idle watcher that is active as long as you have one active
2974 C<ev_check> watcher. The C<ev_idle> watcher makes sure the event loop
2975 will not sleep, and the C<ev_check> watcher makes sure a callback gets
2976 invoked. Neither watcher alone can do that.
2978 =head3 Watcher-Specific Functions and Data Members
2982 =item ev_prepare_init (ev_prepare *, callback)
2984 =item ev_check_init (ev_check *, callback)
2986 Initialises and configures the prepare or check watcher - they have no
2987 parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
2988 macros, but using them is utterly, utterly, utterly and completely
2995 There are a number of principal ways to embed other event loops or modules
2996 into libev. Here are some ideas on how to include libadns into libev
2997 (there is a Perl module named C<EV::ADNS> that does this, which you could
2998 use as a working example. Another Perl module named C<EV::Glib> embeds a
2999 Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
3002 Method 1: Add IO watchers and a timeout watcher in a prepare handler,
3003 and in a check watcher, destroy them and call into libadns. What follows
3004 is pseudo-code only of course. This requires you to either use a low
3005 priority for the check watcher or use C<ev_clear_pending> explicitly, as
3006 the callbacks for the IO/timeout watchers might not have been called yet.
3008 static ev_io iow [nfd];
3012 io_cb (struct ev_loop *loop, ev_io *w, int revents)
3016 // create io watchers for each fd and a timer before blocking
3018 adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
3020 int timeout = 3600000;
3021 struct pollfd fds [nfd];
3022 // actual code will need to loop here and realloc etc.
3023 adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
3025 /* the callback is illegal, but won't be called as we stop during check */
3026 ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
3027 ev_timer_start (loop, &tw);
3029 // create one ev_io per pollfd
3030 for (int i = 0; i < nfd; ++i)
3032 ev_io_init (iow + i, io_cb, fds [i].fd,
3033 ((fds [i].events & POLLIN ? EV_READ : 0)
3034 | (fds [i].events & POLLOUT ? EV_WRITE : 0)));
3036 fds [i].revents = 0;
3037 ev_io_start (loop, iow + i);
3041 // stop all watchers after blocking
3043 adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
3045 ev_timer_stop (loop, &tw);
3047 for (int i = 0; i < nfd; ++i)
3049 // set the relevant poll flags
3050 // could also call adns_processreadable etc. here
3051 struct pollfd *fd = fds + i;
3052 int revents = ev_clear_pending (iow + i);
3053 if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
3054 if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
3056 // now stop the watcher
3057 ev_io_stop (loop, iow + i);
3060 adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
3063 Method 2: This would be just like method 1, but you run C<adns_afterpoll>
3064 in the prepare watcher and would dispose of the check watcher.
3066 Method 3: If the module to be embedded supports explicit event
3067 notification (libadns does), you can also make use of the actual watcher
3068 callbacks, and only destroy/create the watchers in the prepare watcher.
3071 timer_cb (EV_P_ ev_timer *w, int revents)
3073 adns_state ads = (adns_state)w->data;
3076 adns_processtimeouts (ads, &tv_now);
3080 io_cb (EV_P_ ev_io *w, int revents)
3082 adns_state ads = (adns_state)w->data;
3085 if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
3086 if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
3089 // do not ever call adns_afterpoll
3091 Method 4: Do not use a prepare or check watcher because the module you
3092 want to embed is not flexible enough to support it. Instead, you can
3093 override their poll function. The drawback with this solution is that the
3094 main loop is now no longer controllable by EV. The C<Glib::EV> module uses
3095 this approach, effectively embedding EV as a client into the horrible
3099 event_poll_func (GPollFD *fds, guint nfds, gint timeout)
3103 for (n = 0; n < nfds; ++n)
3104 // create/start io watcher that sets the relevant bits in fds[n] and increment got_events
3107 // create/start timer
3114 ev_timer_stop (EV_A_ &to);
3116 // stop io watchers again - their callbacks should have set
3117 for (n = 0; n < nfds; ++n)
3118 ev_io_stop (EV_A_ iow [n]);
3124 =head2 C<ev_embed> - when one backend isn't enough...
3126 This is a rather advanced watcher type that lets you embed one event loop
3127 into another (currently only C<ev_io> events are supported in the embedded
3128 loop, other types of watchers might be handled in a delayed or incorrect
3129 fashion and must not be used).
3131 There are primarily two reasons you would want that: work around bugs and
3134 As an example for a bug workaround, the kqueue backend might only support
3135 sockets on some platform, so it is unusable as generic backend, but you
3136 still want to make use of it because you have many sockets and it scales
3137 so nicely. In this case, you would create a kqueue-based loop and embed
3138 it into your default loop (which might use e.g. poll). Overall operation
3139 will be a bit slower because first libev has to call C<poll> and then
3140 C<kevent>, but at least you can use both mechanisms for what they are
3141 best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
3143 As for prioritising I/O: under rare circumstances you have the case where
3144 some fds have to be watched and handled very quickly (with low latency),
3145 and even priorities and idle watchers might have too much overhead. In
3146 this case you would put all the high priority stuff in one loop and all
3147 the rest in a second one, and embed the second one in the first.
3149 As long as the watcher is active, the callback will be invoked every
3150 time there might be events pending in the embedded loop. The callback
3151 must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
3152 sweep and invoke their callbacks (the callback doesn't need to invoke the
3153 C<ev_embed_sweep> function directly, it could also start an idle watcher
3154 to give the embedded loop strictly lower priority for example).
3156 You can also set the callback to C<0>, in which case the embed watcher
3157 will automatically execute the embedded loop sweep whenever necessary.
3159 Fork detection will be handled transparently while the C<ev_embed> watcher
3160 is active, i.e., the embedded loop will automatically be forked when the
3161 embedding loop forks. In other cases, the user is responsible for calling
3162 C<ev_loop_fork> on the embedded loop.
3164 Unfortunately, not all backends are embeddable: only the ones returned by
3165 C<ev_embeddable_backends> are, which, unfortunately, does not include any
3168 So when you want to use this feature you will always have to be prepared
3169 that you cannot get an embeddable loop. The recommended way to get around
3170 this is to have a separate variables for your embeddable loop, try to
3171 create it, and if that fails, use the normal loop for everything.
3173 =head3 C<ev_embed> and fork
3175 While the C<ev_embed> watcher is running, forks in the embedding loop will
3176 automatically be applied to the embedded loop as well, so no special
3177 fork handling is required in that case. When the watcher is not running,
3178 however, it is still the task of the libev user to call C<ev_loop_fork ()>
3181 =head3 Watcher-Specific Functions and Data Members
3185 =item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
3187 =item ev_embed_set (ev_embed *, struct ev_loop *embedded_loop)
3189 Configures the watcher to embed the given loop, which must be
3190 embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
3191 invoked automatically, otherwise it is the responsibility of the callback
3192 to invoke it (it will continue to be called until the sweep has been done,
3193 if you do not want that, you need to temporarily stop the embed watcher).
3195 =item ev_embed_sweep (loop, ev_embed *)
3197 Make a single, non-blocking sweep over the embedded loop. This works
3198 similarly to C<ev_run (embedded_loop, EVRUN_NOWAIT)>, but in the most
3199 appropriate way for embedded loops.
3201 =item struct ev_loop *other [read-only]
3203 The embedded event loop.
3209 Example: Try to get an embeddable event loop and embed it into the default
3210 event loop. If that is not possible, use the default loop. The default
3211 loop is stored in C<loop_hi>, while the embeddable loop is stored in
3212 C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
3215 struct ev_loop *loop_hi = ev_default_init (0);
3216 struct ev_loop *loop_lo = 0;
3219 // see if there is a chance of getting one that works
3220 // (remember that a flags value of 0 means autodetection)
3221 loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
3222 ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
3225 // if we got one, then embed it, otherwise default to loop_hi
3228 ev_embed_init (&embed, 0, loop_lo);
3229 ev_embed_start (loop_hi, &embed);
3234 Example: Check if kqueue is available but not recommended and create
3235 a kqueue backend for use with sockets (which usually work with any
3236 kqueue implementation). Store the kqueue/socket-only event loop in
3237 C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
3239 struct ev_loop *loop = ev_default_init (0);
3240 struct ev_loop *loop_socket = 0;
3243 if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
3244 if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
3246 ev_embed_init (&embed, 0, loop_socket);
3247 ev_embed_start (loop, &embed);
3253 // now use loop_socket for all sockets, and loop for everything else
3256 =head2 C<ev_fork> - the audacity to resume the event loop after a fork
3258 Fork watchers are called when a C<fork ()> was detected (usually because
3259 whoever is a good citizen cared to tell libev about it by calling
3260 C<ev_loop_fork>). The invocation is done before the event loop blocks next
3261 and before C<ev_check> watchers are being called, and only in the child
3262 after the fork. If whoever good citizen calling C<ev_default_fork> cheats
3263 and calls it in the wrong process, the fork handlers will be invoked, too,
3266 =head3 The special problem of life after fork - how is it possible?
3268 Most uses of C<fork()> consist of forking, then some simple calls to set
3269 up/change the process environment, followed by a call to C<exec()>. This
3270 sequence should be handled by libev without any problems.
3272 This changes when the application actually wants to do event handling
3273 in the child, or both parent in child, in effect "continuing" after the
3276 The default mode of operation (for libev, with application help to detect
3277 forks) is to duplicate all the state in the child, as would be expected
3278 when I<either> the parent I<or> the child process continues.
3280 When both processes want to continue using libev, then this is usually the
3281 wrong result. In that case, usually one process (typically the parent) is
3282 supposed to continue with all watchers in place as before, while the other
3283 process typically wants to start fresh, i.e. without any active watchers.
3285 The cleanest and most efficient way to achieve that with libev is to
3286 simply create a new event loop, which of course will be "empty", and
3287 use that for new watchers. This has the advantage of not touching more
3288 memory than necessary, and thus avoiding the copy-on-write, and the
3289 disadvantage of having to use multiple event loops (which do not support
3292 When this is not possible, or you want to use the default loop for
3293 other reasons, then in the process that wants to start "fresh", call
3294 C<ev_loop_destroy (EV_DEFAULT)> followed by C<ev_default_loop (...)>.
3295 Destroying the default loop will "orphan" (not stop) all registered
3296 watchers, so you have to be careful not to execute code that modifies
3297 those watchers. Note also that in that case, you have to re-register any
3300 =head3 Watcher-Specific Functions and Data Members
3304 =item ev_fork_init (ev_fork *, callback)
3306 Initialises and configures the fork watcher - it has no parameters of any
3307 kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
3313 =head2 C<ev_cleanup> - even the best things end
3315 Cleanup watchers are called just before the event loop is being destroyed
3316 by a call to C<ev_loop_destroy>.
3318 While there is no guarantee that the event loop gets destroyed, cleanup
3319 watchers provide a convenient method to install cleanup hooks for your
3320 program, worker threads and so on - you just to make sure to destroy the
3321 loop when you want them to be invoked.
3323 Cleanup watchers are invoked in the same way as any other watcher. Unlike
3324 all other watchers, they do not keep a reference to the event loop (which
3325 makes a lot of sense if you think about it). Like all other watchers, you
3326 can call libev functions in the callback, except C<ev_cleanup_start>.
3328 =head3 Watcher-Specific Functions and Data Members
3332 =item ev_cleanup_init (ev_cleanup *, callback)
3334 Initialises and configures the cleanup watcher - it has no parameters of
3335 any kind. There is a C<ev_cleanup_set> macro, but using it is utterly
3336 pointless, I assure you.
3340 Example: Register an atexit handler to destroy the default loop, so any
3341 cleanup functions are called.
3344 program_exits (void)
3346 ev_loop_destroy (EV_DEFAULT_UC);
3350 atexit (program_exits);
3353 =head2 C<ev_async> - how to wake up an event loop
3355 In general, you cannot use an C<ev_loop> from multiple threads or other
3356 asynchronous sources such as signal handlers (as opposed to multiple event
3357 loops - those are of course safe to use in different threads).
3359 Sometimes, however, you need to wake up an event loop you do not control,
3360 for example because it belongs to another thread. This is what C<ev_async>
3361 watchers do: as long as the C<ev_async> watcher is active, you can signal
3362 it by calling C<ev_async_send>, which is thread- and signal safe.
3364 This functionality is very similar to C<ev_signal> watchers, as signals,
3365 too, are asynchronous in nature, and signals, too, will be compressed
3366 (i.e. the number of callback invocations may be less than the number of
3367 C<ev_async_send> calls). In fact, you could use signal watchers as a kind
3368 of "global async watchers" by using a watcher on an otherwise unused
3369 signal, and C<ev_feed_signal> to signal this watcher from another thread,
3370 even without knowing which loop owns the signal.
3374 C<ev_async> does not support queueing of data in any way. The reason
3375 is that the author does not know of a simple (or any) algorithm for a
3376 multiple-writer-single-reader queue that works in all cases and doesn't
3377 need elaborate support such as pthreads or unportable memory access
3380 That means that if you want to queue data, you have to provide your own
3381 queue. But at least I can tell you how to implement locking around your
3386 =item queueing from a signal handler context
3388 To implement race-free queueing, you simply add to the queue in the signal
3389 handler but you block the signal handler in the watcher callback. Here is
3390 an example that does that for some fictitious SIGUSR1 handler:
3392 static ev_async mysig;
3395 sigusr1_handler (void)
3401 ev_async_send (EV_DEFAULT_ &mysig);
3405 mysig_cb (EV_P_ ev_async *w, int revents)
3408 sigset_t block, prev;
3410 sigemptyset (&block);
3411 sigaddset (&block, SIGUSR1);
3412 sigprocmask (SIG_BLOCK, &block, &prev);
3414 while (queue_get (&data))
3417 if (sigismember (&prev, SIGUSR1)
3418 sigprocmask (SIG_UNBLOCK, &block, 0);
3421 (Note: pthreads in theory requires you to use C<pthread_setmask>
3422 instead of C<sigprocmask> when you use threads, but libev doesn't do it
3425 =item queueing from a thread context
3427 The strategy for threads is different, as you cannot (easily) block
3428 threads but you can easily preempt them, so to queue safely you need to
3429 employ a traditional mutex lock, such as in this pthread example:
3431 static ev_async mysig;
3432 static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
3437 // only need to lock the actual queueing operation
3438 pthread_mutex_lock (&mymutex);
3440 pthread_mutex_unlock (&mymutex);
3442 ev_async_send (EV_DEFAULT_ &mysig);
3446 mysig_cb (EV_P_ ev_async *w, int revents)
3448 pthread_mutex_lock (&mymutex);
3450 while (queue_get (&data))
3453 pthread_mutex_unlock (&mymutex);
3459 =head3 Watcher-Specific Functions and Data Members
3463 =item ev_async_init (ev_async *, callback)
3465 Initialises and configures the async watcher - it has no parameters of any
3466 kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
3469 =item ev_async_send (loop, ev_async *)
3471 Sends/signals/activates the given C<ev_async> watcher, that is, feeds
3472 an C<EV_ASYNC> event on the watcher into the event loop, and instantly
3475 Unlike C<ev_feed_event>, this call is safe to do from other threads,
3476 signal or similar contexts (see the discussion of C<EV_ATOMIC_T> in the
3477 embedding section below on what exactly this means).
3479 Note that, as with other watchers in libev, multiple events might get
3480 compressed into a single callback invocation (another way to look at
3481 this is that C<ev_async> watchers are level-triggered: they are set on
3482 C<ev_async_send>, reset when the event loop detects that).
3484 This call incurs the overhead of at most one extra system call per event
3485 loop iteration, if the event loop is blocked, and no syscall at all if
3486 the event loop (or your program) is processing events. That means that
3487 repeated calls are basically free (there is no need to avoid calls for
3488 performance reasons) and that the overhead becomes smaller (typically
3491 =item bool = ev_async_pending (ev_async *)
3493 Returns a non-zero value when C<ev_async_send> has been called on the
3494 watcher but the event has not yet been processed (or even noted) by the
3497 C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
3498 the loop iterates next and checks for the watcher to have become active,
3499 it will reset the flag again. C<ev_async_pending> can be used to very
3500 quickly check whether invoking the loop might be a good idea.
3502 Not that this does I<not> check whether the watcher itself is pending,
3503 only whether it has been requested to make this watcher pending: there
3504 is a time window between the event loop checking and resetting the async
3505 notification, and the callback being invoked.
3510 =head1 OTHER FUNCTIONS
3512 There are some other functions of possible interest. Described. Here. Now.
3516 =item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)
3518 This function combines a simple timer and an I/O watcher, calls your
3519 callback on whichever event happens first and automatically stops both
3520 watchers. This is useful if you want to wait for a single event on an fd
3521 or timeout without having to allocate/configure/start/stop/free one or
3522 more watchers yourself.
3524 If C<fd> is less than 0, then no I/O watcher will be started and the
3525 C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
3526 the given C<fd> and C<events> set will be created and started.
3528 If C<timeout> is less than 0, then no timeout watcher will be
3529 started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
3530 repeat = 0) will be started. C<0> is a valid timeout.
3532 The callback has the type C<void (*cb)(int revents, void *arg)> and is
3533 passed an C<revents> set like normal event callbacks (a combination of
3534 C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMER>) and the C<arg>
3535 value passed to C<ev_once>. Note that it is possible to receive I<both>
3536 a timeout and an io event at the same time - you probably should give io
3539 Example: wait up to ten seconds for data to appear on STDIN_FILENO.
3541 static void stdin_ready (int revents, void *arg)
3543 if (revents & EV_READ)
3544 /* stdin might have data for us, joy! */;
3545 else if (revents & EV_TIMER)
3546 /* doh, nothing entered */;
3549 ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
3551 =item ev_feed_fd_event (loop, int fd, int revents)
3553 Feed an event on the given fd, as if a file descriptor backend detected
3556 =item ev_feed_signal_event (loop, int signum)
3558 Feed an event as if the given signal occurred. See also C<ev_feed_signal>,
3559 which is async-safe.
3564 =head1 COMMON OR USEFUL IDIOMS (OR BOTH)
3566 This section explains some common idioms that are not immediately
3567 obvious. Note that examples are sprinkled over the whole manual, and this
3568 section only contains stuff that wouldn't fit anywhere else.
3570 =head2 ASSOCIATING CUSTOM DATA WITH A WATCHER
3572 Each watcher has, by default, a C<void *data> member that you can read
3573 or modify at any time: libev will completely ignore it. This can be used
3574 to associate arbitrary data with your watcher. If you need more data and
3575 don't want to allocate memory separately and store a pointer to it in that
3576 data member, you can also "subclass" the watcher type and provide your own
3584 struct whatever *mostinteresting;
3589 ev_io_init (&w.io, my_cb, fd, EV_READ);
3591 And since your callback will be called with a pointer to the watcher, you
3592 can cast it back to your own type:
3594 static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
3596 struct my_io *w = (struct my_io *)w_;
3600 More interesting and less C-conformant ways of casting your callback
3601 function type instead have been omitted.
3603 =head2 BUILDING YOUR OWN COMPOSITE WATCHERS
3605 Another common scenario is to use some data structure with multiple
3606 embedded watchers, in effect creating your own watcher that combines
3607 multiple libev event sources into one "super-watcher":
3616 In this case getting the pointer to C<my_biggy> is a bit more
3617 complicated: Either you store the address of your C<my_biggy> struct in
3618 the C<data> member of the watcher (for woozies or C++ coders), or you need
3619 to use some pointer arithmetic using C<offsetof> inside your watchers (for
3625 t1_cb (EV_P_ ev_timer *w, int revents)
3627 struct my_biggy big = (struct my_biggy *)
3628 (((char *)w) - offsetof (struct my_biggy, t1));
3632 t2_cb (EV_P_ ev_timer *w, int revents)
3634 struct my_biggy big = (struct my_biggy *)
3635 (((char *)w) - offsetof (struct my_biggy, t2));
3638 =head2 AVOIDING FINISHING BEFORE RETURNING
3640 Often you have structures like this in event-based programs:
3647 request = start_new_request (..., callback);
3649 The intent is to start some "lengthy" operation. The C<request> could be
3650 used to cancel the operation, or do other things with it.
3652 It's not uncommon to have code paths in C<start_new_request> that
3653 immediately invoke the callback, for example, to report errors. Or you add
3654 some caching layer that finds that it can skip the lengthy aspects of the
3655 operation and simply invoke the callback with the result.
3657 The problem here is that this will happen I<before> C<start_new_request>
3658 has returned, so C<request> is not set.
3660 Even if you pass the request by some safer means to the callback, you
3661 might want to do something to the request after starting it, such as
3662 canceling it, which probably isn't working so well when the callback has
3663 already been invoked.
3665 A common way around all these issues is to make sure that
3666 C<start_new_request> I<always> returns before the callback is invoked. If
3667 C<start_new_request> immediately knows the result, it can artificially
3668 delay invoking the callback by using a C<prepare> or C<idle> watcher for
3669 example, or more sneakily, by reusing an existing (stopped) watcher and
3670 pushing it into the pending queue:
3672 ev_set_cb (watcher, callback);
3673 ev_feed_event (EV_A_ watcher, 0);
3675 This way, C<start_new_request> can safely return before the callback is
3676 invoked, while not delaying callback invocation too much.
3678 =head2 MODEL/NESTED EVENT LOOP INVOCATIONS AND EXIT CONDITIONS
3680 Often (especially in GUI toolkits) there are places where you have
3681 I<modal> interaction, which is most easily implemented by recursively
3684 This brings the problem of exiting - a callback might want to finish the
3685 main C<ev_run> call, but not the nested one (e.g. user clicked "Quit", but
3686 a modal "Are you sure?" dialog is still waiting), or just the nested one
3687 and not the main one (e.g. user clocked "Ok" in a modal dialog), or some
3688 other combination: In these cases, a simple C<ev_break> will not work.
3690 The solution is to maintain "break this loop" variable for each C<ev_run>
3691 invocation, and use a loop around C<ev_run> until the condition is
3692 triggered, using C<EVRUN_ONCE>:
3695 int exit_main_loop = 0;
3697 while (!exit_main_loop)
3698 ev_run (EV_DEFAULT_ EVRUN_ONCE);
3700 // in a modal watcher
3701 int exit_nested_loop = 0;
3703 while (!exit_nested_loop)
3704 ev_run (EV_A_ EVRUN_ONCE);
3706 To exit from any of these loops, just set the corresponding exit variable:
3709 exit_nested_loop = 1;
3711 // exit main program, after modal loop is finished
3715 exit_main_loop = exit_nested_loop = 1;
3717 =head2 THREAD LOCKING EXAMPLE
3719 Here is a fictitious example of how to run an event loop in a different
3720 thread from where callbacks are being invoked and watchers are
3721 created/added/removed.
3723 For a real-world example, see the C<EV::Loop::Async> perl module,
3724 which uses exactly this technique (which is suited for many high-level
3727 The example uses a pthread mutex to protect the loop data, a condition
3728 variable to wait for callback invocations, an async watcher to notify the
3729 event loop thread and an unspecified mechanism to wake up the main thread.
3731 First, you need to associate some data with the event loop:
3734 mutex_t lock; /* global loop lock */
3740 void prepare_loop (EV_P)
3742 // for simplicity, we use a static userdata struct.
3745 ev_async_init (&u->async_w, async_cb);
3746 ev_async_start (EV_A_ &u->async_w);
3748 pthread_mutex_init (&u->lock, 0);
3749 pthread_cond_init (&u->invoke_cv, 0);
3751 // now associate this with the loop
3752 ev_set_userdata (EV_A_ u);
3753 ev_set_invoke_pending_cb (EV_A_ l_invoke);
3754 ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
3756 // then create the thread running ev_run
3757 pthread_create (&u->tid, 0, l_run, EV_A);
3760 The callback for the C<ev_async> watcher does nothing: the watcher is used
3761 solely to wake up the event loop so it takes notice of any new watchers
3762 that might have been added:
3765 async_cb (EV_P_ ev_async *w, int revents)
3767 // just used for the side effects
3770 The C<l_release> and C<l_acquire> callbacks simply unlock/lock the mutex
3771 protecting the loop data, respectively.
3776 userdata *u = ev_userdata (EV_A);
3777 pthread_mutex_unlock (&u->lock);
3783 userdata *u = ev_userdata (EV_A);
3784 pthread_mutex_lock (&u->lock);
3787 The event loop thread first acquires the mutex, and then jumps straight
3791 l_run (void *thr_arg)
3793 struct ev_loop *loop = (struct ev_loop *)thr_arg;
3796 pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
3803 Instead of invoking all pending watchers, the C<l_invoke> callback will
3804 signal the main thread via some unspecified mechanism (signals? pipe
3805 writes? C<Async::Interrupt>?) and then waits until all pending watchers
3806 have been called (in a while loop because a) spurious wakeups are possible
3807 and b) skipping inter-thread-communication when there are no pending
3808 watchers is very beneficial):
3813 userdata *u = ev_userdata (EV_A);
3815 while (ev_pending_count (EV_A))
3817 wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
3818 pthread_cond_wait (&u->invoke_cv, &u->lock);
3822 Now, whenever the main thread gets told to invoke pending watchers, it
3823 will grab the lock, call C<ev_invoke_pending> and then signal the loop
3827 real_invoke_pending (EV_P)
3829 userdata *u = ev_userdata (EV_A);
3831 pthread_mutex_lock (&u->lock);
3832 ev_invoke_pending (EV_A);
3833 pthread_cond_signal (&u->invoke_cv);
3834 pthread_mutex_unlock (&u->lock);
3837 Whenever you want to start/stop a watcher or do other modifications to an
3838 event loop, you will now have to lock:
3840 ev_timer timeout_watcher;
3841 userdata *u = ev_userdata (EV_A);
3843 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
3845 pthread_mutex_lock (&u->lock);
3846 ev_timer_start (EV_A_ &timeout_watcher);
3847 ev_async_send (EV_A_ &u->async_w);
3848 pthread_mutex_unlock (&u->lock);
3850 Note that sending the C<ev_async> watcher is required because otherwise
3851 an event loop currently blocking in the kernel will have no knowledge
3852 about the newly added timer. By waking up the loop it will pick up any new
3853 watchers in the next event loop iteration.
3855 =head2 THREADS, COROUTINES, CONTINUATIONS, QUEUES... INSTEAD OF CALLBACKS
3857 While the overhead of a callback that e.g. schedules a thread is small, it
3858 is still an overhead. If you embed libev, and your main usage is with some
3859 kind of threads or coroutines, you might want to customise libev so that
3860 doesn't need callbacks anymore.
3862 Imagine you have coroutines that you can switch to using a function
3863 C<switch_to (coro)>, that libev runs in a coroutine called C<libev_coro>
3864 and that due to some magic, the currently active coroutine is stored in a
3865 global called C<current_coro>. Then you can build your own "wait for libev
3866 event" primitive by changing C<EV_CB_DECLARE> and C<EV_CB_INVOKE> (note
3867 the differing C<;> conventions):
3869 #define EV_CB_DECLARE(type) struct my_coro *cb;
3870 #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
3872 That means instead of having a C callback function, you store the
3873 coroutine to switch to in each watcher, and instead of having libev call
3874 your callback, you instead have it switch to that coroutine.
3876 A coroutine might now wait for an event with a function called
3877 C<wait_for_event>. (the watcher needs to be started, as always, but it doesn't
3878 matter when, or whether the watcher is active or not when this function is
3882 wait_for_event (ev_watcher *w)
3884 ev_set_cb (w, current_coro);
3885 switch_to (libev_coro);
3888 That basically suspends the coroutine inside C<wait_for_event> and
3889 continues the libev coroutine, which, when appropriate, switches back to
3890 this or any other coroutine.
3892 You can do similar tricks if you have, say, threads with an event queue -
3893 instead of storing a coroutine, you store the queue object and instead of
3894 switching to a coroutine, you push the watcher onto the queue and notify
3897 To embed libev, see L</EMBEDDING>, but in short, it's easiest to create two
3898 files, F<my_ev.h> and F<my_ev.c> that include the respective libev files:
3901 #define EV_CB_DECLARE(type) struct my_coro *cb;
3902 #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb);
3903 #include "../libev/ev.h"
3906 #define EV_H "my_ev.h"
3907 #include "../libev/ev.c"
3909 And then use F<my_ev.h> when you would normally use F<ev.h>, and compile
3910 F<my_ev.c> into your project. When properly specifying include paths, you
3911 can even use F<ev.h> as header file name directly.
3914 =head1 LIBEVENT EMULATION
3916 Libev offers a compatibility emulation layer for libevent. It cannot
3917 emulate the internals of libevent, so here are some usage hints:
3921 =item * Only the libevent-1.4.1-beta API is being emulated.
3923 This was the newest libevent version available when libev was implemented,
3924 and is still mostly unchanged in 2010.
3926 =item * Use it by including <event.h>, as usual.
3928 =item * The following members are fully supported: ev_base, ev_callback,
3929 ev_arg, ev_fd, ev_res, ev_events.
3931 =item * Avoid using ev_flags and the EVLIST_*-macros, while it is
3932 maintained by libev, it does not work exactly the same way as in libevent (consider
3935 =item * Priorities are not currently supported. Initialising priorities
3936 will fail and all watchers will have the same priority, even though there
3939 =item * In libevent, the last base created gets the signals, in libev, the
3940 base that registered the signal gets the signals.
3942 =item * Other members are not supported.
3944 =item * The libev emulation is I<not> ABI compatible to libevent, you need
3945 to use the libev header file and library.
3953 The normal C API should work fine when used from C++: both ev.h and the
3954 libev sources can be compiled as C++. Therefore, code that uses the C API
3957 Proper exception specifications might have to be added to callbacks passed
3958 to libev: exceptions may be thrown only from watcher callbacks, all
3959 other callbacks (allocator, syserr, loop acquire/release and periodic
3960 reschedule callbacks) must not throw exceptions, and might need a C<throw
3961 ()> specification. If you have code that needs to be compiled as both C
3962 and C++ you can use the C<EV_THROW> macro for this:
3965 fatal_error (const char *msg) EV_THROW
3972 ev_set_syserr_cb (fatal_error);
3974 The only API functions that can currently throw exceptions are C<ev_run>,
3975 C<ev_invoke>, C<ev_invoke_pending> and C<ev_loop_destroy> (the latter
3976 because it runs cleanup watchers).
3978 Throwing exceptions in watcher callbacks is only supported if libev itself
3979 is compiled with a C++ compiler or your C and C++ environments allow
3980 throwing exceptions through C libraries (most do).
3984 Libev comes with some simplistic wrapper classes for C++ that mainly allow
3985 you to use some convenience methods to start/stop watchers and also change
3986 the callback model to a model using method callbacks on objects.
3992 This automatically includes F<ev.h> and puts all of its definitions (many
3993 of them macros) into the global namespace. All C++ specific things are
3994 put into the C<ev> namespace. It should support all the same embedding
3995 options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
3997 Care has been taken to keep the overhead low. The only data member the C++
3998 classes add (compared to plain C-style watchers) is the event loop pointer
3999 that the watcher is associated with (or no additional members at all if
4000 you disable C<EV_MULTIPLICITY> when embedding libev).
4002 Currently, functions, static and non-static member functions and classes
4003 with C<operator ()> can be used as callbacks. Other types should be easy
4004 to add as long as they only need one additional pointer for context. If
4005 you need support for other types of functors please contact the author
4006 (preferably after implementing it).
4008 For all this to work, your C++ compiler either has to use the same calling
4009 conventions as your C compiler (for static member functions), or you have
4010 to embed libev and compile libev itself as C++.
4012 Here is a list of things available in the C<ev> namespace:
4016 =item C<ev::READ>, C<ev::WRITE> etc.
4018 These are just enum values with the same values as the C<EV_READ> etc.
4019 macros from F<ev.h>.
4021 =item C<ev::tstamp>, C<ev::now>
4023 Aliases to the same types/functions as with the C<ev_> prefix.
4025 =item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
4027 For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
4028 the same name in the C<ev> namespace, with the exception of C<ev_signal>
4029 which is called C<ev::sig> to avoid clashes with the C<signal> macro
4030 defined by many implementations.
4032 All of those classes have these methods:
4036 =item ev::TYPE::TYPE ()
4038 =item ev::TYPE::TYPE (loop)
4040 =item ev::TYPE::~TYPE
4042 The constructor (optionally) takes an event loop to associate the watcher
4043 with. If it is omitted, it will use C<EV_DEFAULT>.
4045 The constructor calls C<ev_init> for you, which means you have to call the
4046 C<set> method before starting it.
4048 It will not set a callback, however: You have to call the templated C<set>
4049 method to set a callback before you can start the watcher.
4051 (The reason why you have to use a method is a limitation in C++ which does
4052 not allow explicit template arguments for constructors).
4054 The destructor automatically stops the watcher if it is active.
4056 =item w->set<class, &class::method> (object *)
4058 This method sets the callback method to call. The method has to have a
4059 signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
4060 first argument and the C<revents> as second. The object must be given as
4061 parameter and is stored in the C<data> member of the watcher.
4063 This method synthesizes efficient thunking code to call your method from
4064 the C callback that libev requires. If your compiler can inline your
4065 callback (i.e. it is visible to it at the place of the C<set> call and
4066 your compiler is good :), then the method will be fully inlined into the
4067 thunking function, making it as fast as a direct C callback.
4069 Example: simple class declaration and watcher initialisation
4073 void io_cb (ev::io &w, int revents) { }
4078 iow.set <myclass, &myclass::io_cb> (&obj);
4080 =item w->set (object *)
4082 This is a variation of a method callback - leaving out the method to call
4083 will default the method to C<operator ()>, which makes it possible to use
4084 functor objects without having to manually specify the C<operator ()> all
4085 the time. Incidentally, you can then also leave out the template argument
4088 The C<operator ()> method prototype must be C<void operator ()(watcher &w,
4091 See the method-C<set> above for more details.
4093 Example: use a functor object as callback.
4097 void operator() (ev::io &w, int revents)
4108 =item w->set<function> (void *data = 0)
4110 Also sets a callback, but uses a static method or plain function as
4111 callback. The optional C<data> argument will be stored in the watcher's
4112 C<data> member and is free for you to use.
4114 The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
4116 See the method-C<set> above for more details.
4118 Example: Use a plain function as callback.
4120 static void io_cb (ev::io &w, int revents) { }
4125 Associates a different C<struct ev_loop> with this watcher. You can only
4126 do this when the watcher is inactive (and not pending either).
4128 =item w->set ([arguments])
4130 Basically the same as C<ev_TYPE_set> (except for C<ev::embed> watchers>),
4131 with the same arguments. Either this method or a suitable start method
4132 must be called at least once. Unlike the C counterpart, an active watcher
4133 gets automatically stopped and restarted when reconfiguring it with this
4136 For C<ev::embed> watchers this method is called C<set_embed>, to avoid
4137 clashing with the C<set (loop)> method.
4141 Starts the watcher. Note that there is no C<loop> argument, as the
4142 constructor already stores the event loop.
4144 =item w->start ([arguments])
4146 Instead of calling C<set> and C<start> methods separately, it is often
4147 convenient to wrap them in one call. Uses the same type of arguments as
4148 the configure C<set> method of the watcher.
4152 Stops the watcher if it is active. Again, no C<loop> argument.
4154 =item w->again () (C<ev::timer>, C<ev::periodic> only)
4156 For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
4157 C<ev_TYPE_again> function.
4159 =item w->sweep () (C<ev::embed> only)
4161 Invokes C<ev_embed_sweep>.
4163 =item w->update () (C<ev::stat> only)
4165 Invokes C<ev_stat_stat>.
4171 Example: Define a class with two I/O and idle watchers, start the I/O
4172 watchers in the constructor.
4176 ev::io io ; void io_cb (ev::io &w, int revents);
4177 ev::io io2 ; void io2_cb (ev::io &w, int revents);
4178 ev::idle idle; void idle_cb (ev::idle &w, int revents);
4182 io .set <myclass, &myclass::io_cb > (this);
4183 io2 .set <myclass, &myclass::io2_cb > (this);
4184 idle.set <myclass, &myclass::idle_cb> (this);
4186 io.set (fd, ev::WRITE); // configure the watcher
4187 io.start (); // start it whenever convenient
4189 io2.start (fd, ev::READ); // set + start in one call
4194 =head1 OTHER LANGUAGE BINDINGS
4196 Libev does not offer other language bindings itself, but bindings for a
4197 number of languages exist in the form of third-party packages. If you know
4198 any interesting language binding in addition to the ones listed here, drop
4205 The EV module implements the full libev API and is actually used to test
4206 libev. EV is developed together with libev. Apart from the EV core module,
4207 there are additional modules that implement libev-compatible interfaces
4208 to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
4209 C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
4212 It can be found and installed via CPAN, its homepage is at
4213 L<http://software.schmorp.de/pkg/EV>.
4217 Python bindings can be found at L<http://code.google.com/p/pyev/>. It
4218 seems to be quite complete and well-documented.
4222 Tony Arcieri has written a ruby extension that offers access to a subset
4223 of the libev API and adds file handle abstractions, asynchronous DNS and
4224 more on top of it. It can be found via gem servers. Its homepage is at
4225 L<http://rev.rubyforge.org/>.
4227 Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
4228 makes rev work even on mingw.
4232 A haskell binding to libev is available at
4233 L<http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
4237 Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
4238 be found at L<http://www.llucax.com.ar/proj/ev.d/index.html>.
4242 Erkki Seppala has written Ocaml bindings for libev, to be found at
4243 L<http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
4247 Brian Maher has written a partial interface to libev for lua (at the
4248 time of this writing, only C<ev_io> and C<ev_timer>), to be found at
4249 L<http://github.com/brimworks/lua-ev>.
4253 Node.js (L<http://nodejs.org>) uses libev as the underlying event library.
4257 There are others, and I stopped counting.
4264 Libev can be compiled with a variety of options, the most fundamental
4265 of which is C<EV_MULTIPLICITY>. This option determines whether (most)
4266 functions and callbacks have an initial C<struct ev_loop *> argument.
4268 To make it easier to write programs that cope with either variant, the
4269 following macros are defined:
4273 =item C<EV_A>, C<EV_A_>
4275 This provides the loop I<argument> for functions, if one is required ("ev
4276 loop argument"). The C<EV_A> form is used when this is the sole argument,
4277 C<EV_A_> is used when other arguments are following. Example:
4280 ev_timer_add (EV_A_ watcher);
4283 It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
4284 which is often provided by the following macro.
4286 =item C<EV_P>, C<EV_P_>
4288 This provides the loop I<parameter> for functions, if one is required ("ev
4289 loop parameter"). The C<EV_P> form is used when this is the sole parameter,
4290 C<EV_P_> is used when other parameters are following. Example:
4292 // this is how ev_unref is being declared
4293 static void ev_unref (EV_P);
4295 // this is how you can declare your typical callback
4296 static void cb (EV_P_ ev_timer *w, int revents)
4298 It declares a parameter C<loop> of type C<struct ev_loop *>, quite
4299 suitable for use with C<EV_A>.
4301 =item C<EV_DEFAULT>, C<EV_DEFAULT_>
4303 Similar to the other two macros, this gives you the value of the default
4304 loop, if multiple loops are supported ("ev loop default"). The default loop
4305 will be initialised if it isn't already initialised.
4307 For non-multiplicity builds, these macros do nothing, so you always have
4308 to initialise the loop somewhere.
4310 =item C<EV_DEFAULT_UC>, C<EV_DEFAULT_UC_>
4312 Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
4313 default loop has been initialised (C<UC> == unchecked). Their behaviour
4314 is undefined when the default loop has not been initialised by a previous
4315 execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
4317 It is often prudent to use C<EV_DEFAULT> when initialising the first
4318 watcher in a function but use C<EV_DEFAULT_UC> afterwards.
4322 Example: Declare and initialise a check watcher, utilising the above
4323 macros so it will work regardless of whether multiple loops are supported
4327 check_cb (EV_P_ ev_timer *w, int revents)
4329 ev_check_stop (EV_A_ w);
4333 ev_check_init (&check, check_cb);
4334 ev_check_start (EV_DEFAULT_ &check);
4335 ev_run (EV_DEFAULT_ 0);
4339 Libev can (and often is) directly embedded into host
4340 applications. Examples of applications that embed it include the Deliantra
4341 Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
4344 The goal is to enable you to just copy the necessary files into your
4345 source directory without having to change even a single line in them, so
4346 you can easily upgrade by simply copying (or having a checked-out copy of
4347 libev somewhere in your source tree).
4351 Depending on what features you need you need to include one or more sets of files
4352 in your application.
4354 =head3 CORE EVENT LOOP
4356 To include only the libev core (all the C<ev_*> functions), with manual
4357 configuration (no autoconf):
4359 #define EV_STANDALONE 1
4362 This will automatically include F<ev.h>, too, and should be done in a
4363 single C source file only to provide the function implementations. To use
4364 it, do the same for F<ev.h> in all files wishing to use this API (best
4365 done by writing a wrapper around F<ev.h> that you can include instead and
4366 where you can put other configuration options):
4368 #define EV_STANDALONE 1
4371 Both header files and implementation files can be compiled with a C++
4372 compiler (at least, that's a stated goal, and breakage will be treated
4375 You need the following files in your source tree, or in a directory
4376 in your include path (e.g. in libev/ when using -Ilibev):
4383 ev_win32.c required on win32 platforms only
4385 ev_select.c only when select backend is enabled (which is enabled by default)
4386 ev_poll.c only when poll backend is enabled (disabled by default)
4387 ev_epoll.c only when the epoll backend is enabled (disabled by default)
4388 ev_kqueue.c only when the kqueue backend is enabled (disabled by default)
4389 ev_port.c only when the solaris port backend is enabled (disabled by default)
4391 F<ev.c> includes the backend files directly when enabled, so you only need
4392 to compile this single file.
4394 =head3 LIBEVENT COMPATIBILITY API
4396 To include the libevent compatibility API, also include:
4400 in the file including F<ev.c>, and:
4404 in the files that want to use the libevent API. This also includes F<ev.h>.
4406 You need the following additional files for this:
4411 =head3 AUTOCONF SUPPORT
4413 Instead of using C<EV_STANDALONE=1> and providing your configuration in
4414 whatever way you want, you can also C<m4_include([libev.m4])> in your
4415 F<configure.ac> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
4416 include F<config.h> and configure itself accordingly.
4418 For this of course you need the m4 file:
4422 =head2 PREPROCESSOR SYMBOLS/MACROS
4424 Libev can be configured via a variety of preprocessor symbols you have to
4425 define before including (or compiling) any of its files. The default in
4426 the absence of autoconf is documented for every option.
4428 Symbols marked with "(h)" do not change the ABI, and can have different
4429 values when compiling libev vs. including F<ev.h>, so it is permissible
4430 to redefine them before including F<ev.h> without breaking compatibility
4431 to a compiled library. All other symbols change the ABI, which means all
4432 users of libev and the libev code itself must be compiled with compatible
4437 =item EV_COMPAT3 (h)
4439 Backwards compatibility is a major concern for libev. This is why this
4440 release of libev comes with wrappers for the functions and symbols that
4441 have been renamed between libev version 3 and 4.
4443 You can disable these wrappers (to test compatibility with future
4444 versions) by defining C<EV_COMPAT3> to C<0> when compiling your
4445 sources. This has the additional advantage that you can drop the C<struct>
4446 from C<struct ev_loop> declarations, as libev will provide an C<ev_loop>
4447 typedef in that case.
4449 In some future version, the default for C<EV_COMPAT3> will become C<0>,
4450 and in some even more future version the compatibility code will be
4453 =item EV_STANDALONE (h)
4455 Must always be C<1> if you do not use autoconf configuration, which
4456 keeps libev from including F<config.h>, and it also defines dummy
4457 implementations for some libevent functions (such as logging, which is not
4458 supported). It will also not define any of the structs usually found in
4459 F<event.h> that are not directly supported by the libev core alone.
4461 In standalone mode, libev will still try to automatically deduce the
4462 configuration, but has to be more conservative.
4466 If defined to be C<1>, libev will use the C<floor ()> function for its
4467 periodic reschedule calculations, otherwise libev will fall back on a
4468 portable (slower) implementation. If you enable this, you usually have to
4469 link against libm or something equivalent. Enabling this when the C<floor>
4470 function is not available will fail, so the safe default is to not enable
4473 =item EV_USE_MONOTONIC
4475 If defined to be C<1>, libev will try to detect the availability of the
4476 monotonic clock option at both compile time and runtime. Otherwise no
4477 use of the monotonic clock option will be attempted. If you enable this,
4478 you usually have to link against librt or something similar. Enabling it
4479 when the functionality isn't available is safe, though, although you have
4480 to make sure you link against any libraries where the C<clock_gettime>
4481 function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
4483 =item EV_USE_REALTIME
4485 If defined to be C<1>, libev will try to detect the availability of the
4486 real-time clock option at compile time (and assume its availability
4487 at runtime if successful). Otherwise no use of the real-time clock
4488 option will be attempted. This effectively replaces C<gettimeofday>
4489 by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
4490 correctness. See the note about libraries in the description of
4491 C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
4492 C<EV_USE_CLOCK_SYSCALL>.
4494 =item EV_USE_CLOCK_SYSCALL
4496 If defined to be C<1>, libev will try to use a direct syscall instead
4497 of calling the system-provided C<clock_gettime> function. This option
4498 exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
4499 unconditionally pulls in C<libpthread>, slowing down single-threaded
4500 programs needlessly. Using a direct syscall is slightly slower (in
4501 theory), because no optimised vdso implementation can be used, but avoids
4502 the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
4503 higher, as it simplifies linking (no need for C<-lrt>).
4505 =item EV_USE_NANOSLEEP
4507 If defined to be C<1>, libev will assume that C<nanosleep ()> is available
4508 and will use it for delays. Otherwise it will use C<select ()>.
4510 =item EV_USE_EVENTFD
4512 If defined to be C<1>, then libev will assume that C<eventfd ()> is
4513 available and will probe for kernel support at runtime. This will improve
4514 C<ev_signal> and C<ev_async> performance and reduce resource consumption.
4515 If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
4516 2.7 or newer, otherwise disabled.
4520 If undefined or defined to be C<1>, libev will compile in support for the
4521 C<select>(2) backend. No attempt at auto-detection will be done: if no
4522 other method takes over, select will be it. Otherwise the select backend
4523 will not be compiled in.
4525 =item EV_SELECT_USE_FD_SET
4527 If defined to C<1>, then the select backend will use the system C<fd_set>
4528 structure. This is useful if libev doesn't compile due to a missing
4529 C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
4530 on exotic systems. This usually limits the range of file descriptors to
4531 some low limit such as 1024 or might have other limitations (winsocket
4532 only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
4533 configures the maximum size of the C<fd_set>.
4535 =item EV_SELECT_IS_WINSOCKET
4537 When defined to C<1>, the select backend will assume that
4538 select/socket/connect etc. don't understand file descriptors but
4539 wants osf handles on win32 (this is the case when the select to
4540 be used is the winsock select). This means that it will call
4541 C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
4542 it is assumed that all these functions actually work on fds, even
4543 on win32. Should not be defined on non-win32 platforms.
4545 =item EV_FD_TO_WIN32_HANDLE(fd)
4547 If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
4548 file descriptors to socket handles. When not defining this symbol (the
4549 default), then libev will call C<_get_osfhandle>, which is usually
4550 correct. In some cases, programs use their own file descriptor management,
4551 in which case they can provide this function to map fds to socket handles.
4553 =item EV_WIN32_HANDLE_TO_FD(handle)
4555 If C<EV_SELECT_IS_WINSOCKET> then libev maps handles to file descriptors
4556 using the standard C<_open_osfhandle> function. For programs implementing
4557 their own fd to handle mapping, overwriting this function makes it easier
4558 to do so. This can be done by defining this macro to an appropriate value.
4560 =item EV_WIN32_CLOSE_FD(fd)
4562 If programs implement their own fd to handle mapping on win32, then this
4563 macro can be used to override the C<close> function, useful to unregister
4564 file descriptors again. Note that the replacement function has to close
4565 the underlying OS handle.
4567 =item EV_USE_WSASOCKET
4569 If defined to be C<1>, libev will use C<WSASocket> to create its internal
4570 communication socket, which works better in some environments. Otherwise,
4571 the normal C<socket> function will be used, which works better in other
4576 If defined to be C<1>, libev will compile in support for the C<poll>(2)
4577 backend. Otherwise it will be enabled on non-win32 platforms. It
4578 takes precedence over select.
4582 If defined to be C<1>, libev will compile in support for the Linux
4583 C<epoll>(7) backend. Its availability will be detected at runtime,
4584 otherwise another method will be used as fallback. This is the preferred
4585 backend for GNU/Linux systems. If undefined, it will be enabled if the
4586 headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
4590 If defined to be C<1>, libev will compile in support for the BSD style
4591 C<kqueue>(2) backend. Its actual availability will be detected at runtime,
4592 otherwise another method will be used as fallback. This is the preferred
4593 backend for BSD and BSD-like systems, although on most BSDs kqueue only
4594 supports some types of fds correctly (the only platform we found that
4595 supports ptys for example was NetBSD), so kqueue might be compiled in, but
4596 not be used unless explicitly requested. The best way to use it is to find
4597 out whether kqueue supports your type of fd properly and use an embedded
4602 If defined to be C<1>, libev will compile in support for the Solaris
4603 10 port style backend. Its availability will be detected at runtime,
4604 otherwise another method will be used as fallback. This is the preferred
4605 backend for Solaris 10 systems.
4607 =item EV_USE_DEVPOLL
4609 Reserved for future expansion, works like the USE symbols above.
4611 =item EV_USE_INOTIFY
4613 If defined to be C<1>, libev will compile in support for the Linux inotify
4614 interface to speed up C<ev_stat> watchers. Its actual availability will
4615 be detected at runtime. If undefined, it will be enabled if the headers
4616 indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
4620 If defined to be C<1>, libev will assume that memory is always coherent
4621 between threads, that is, threads can be used, but threads never run on
4622 different cpus (or different cpu cores). This reduces dependencies
4623 and makes libev faster.
4627 If defined to be C<1>, libev will assume that it will never be called from
4628 different threads (that includes signal handlers), which is a stronger
4629 assumption than C<EV_NO_SMP>, above. This reduces dependencies and makes
4634 Libev requires an integer type (suitable for storing C<0> or C<1>) whose
4635 access is atomic with respect to other threads or signal contexts. No
4636 such type is easily found in the C language, so you can provide your own
4637 type that you know is safe for your purposes. It is used both for signal
4638 handler "locking" as well as for signal and thread safety in C<ev_async>
4641 In the absence of this define, libev will use C<sig_atomic_t volatile>
4642 (from F<signal.h>), which is usually good enough on most platforms.
4646 The name of the F<ev.h> header file used to include it. The default if
4647 undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
4648 used to virtually rename the F<ev.h> header file in case of conflicts.
4650 =item EV_CONFIG_H (h)
4652 If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
4653 F<ev.c>'s idea of where to find the F<config.h> file, similarly to
4656 =item EV_EVENT_H (h)
4658 Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
4659 of how the F<event.h> header can be found, the default is C<"event.h">.
4661 =item EV_PROTOTYPES (h)
4663 If defined to be C<0>, then F<ev.h> will not define any function
4664 prototypes, but still define all the structs and other symbols. This is
4665 occasionally useful if you want to provide your own wrapper functions
4666 around libev functions.
4668 =item EV_MULTIPLICITY
4670 If undefined or defined to C<1>, then all event-loop-specific functions
4671 will have the C<struct ev_loop *> as first argument, and you can create
4672 additional independent event loops. Otherwise there will be no support
4673 for multiple event loops and there is no first event loop pointer
4674 argument. Instead, all functions act on the single default loop.
4676 Note that C<EV_DEFAULT> and C<EV_DEFAULT_> will no longer provide a
4677 default loop when multiplicity is switched off - you always have to
4678 initialise the loop manually in this case.
4684 The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
4685 C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
4686 provide for more priorities by overriding those symbols (usually defined
4687 to be C<-2> and C<2>, respectively).
4689 When doing priority-based operations, libev usually has to linearly search
4690 all the priorities, so having many of them (hundreds) uses a lot of space
4691 and time, so using the defaults of five priorities (-2 .. +2) is usually
4694 If your embedding application does not need any priorities, defining these
4695 both to C<0> will save some memory and CPU.
4697 =item EV_PERIODIC_ENABLE, EV_IDLE_ENABLE, EV_EMBED_ENABLE, EV_STAT_ENABLE,
4698 EV_PREPARE_ENABLE, EV_CHECK_ENABLE, EV_FORK_ENABLE, EV_SIGNAL_ENABLE,
4699 EV_ASYNC_ENABLE, EV_CHILD_ENABLE.
4701 If undefined or defined to be C<1> (and the platform supports it), then
4702 the respective watcher type is supported. If defined to be C<0>, then it
4703 is not. Disabling watcher types mainly saves code size.
4707 If you need to shave off some kilobytes of code at the expense of some
4708 speed (but with the full API), you can define this symbol to request
4709 certain subsets of functionality. The default is to enable all features
4710 that can be enabled on the platform.
4712 A typical way to use this symbol is to define it to C<0> (or to a bitset
4713 with some broad features you want) and then selectively re-enable
4714 additional parts you want, for example if you want everything minimal,
4715 but multiple event loop support, async and child watchers and the poll
4718 #define EV_FEATURES 0
4719 #define EV_MULTIPLICITY 1
4720 #define EV_USE_POLL 1
4721 #define EV_CHILD_ENABLE 1
4722 #define EV_ASYNC_ENABLE 1
4724 The actual value is a bitset, it can be a combination of the following
4725 values (by default, all of these are enabled):
4729 =item C<1> - faster/larger code
4731 Use larger code to speed up some operations.
4733 Currently this is used to override some inlining decisions (enlarging the
4734 code size by roughly 30% on amd64).
4736 When optimising for size, use of compiler flags such as C<-Os> with
4737 gcc is recommended, as well as C<-DNDEBUG>, as libev contains a number of
4740 The default is off when C<__OPTIMIZE_SIZE__> is defined by your compiler
4741 (e.g. gcc with C<-Os>).
4743 =item C<2> - faster/larger data structures
4745 Replaces the small 2-heap for timer management by a faster 4-heap, larger
4746 hash table sizes and so on. This will usually further increase code size
4747 and can additionally have an effect on the size of data structures at
4750 The default is off when C<__OPTIMIZE_SIZE__> is defined by your compiler
4751 (e.g. gcc with C<-Os>).
4753 =item C<4> - full API configuration
4755 This enables priorities (sets C<EV_MAXPRI>=2 and C<EV_MINPRI>=-2), and
4756 enables multiplicity (C<EV_MULTIPLICITY>=1).
4758 =item C<8> - full API
4760 This enables a lot of the "lesser used" API functions. See C<ev.h> for
4761 details on which parts of the API are still available without this
4762 feature, and do not complain if this subset changes over time.
4764 =item C<16> - enable all optional watcher types
4766 Enables all optional watcher types. If you want to selectively enable
4767 only some watcher types other than I/O and timers (e.g. prepare,
4768 embed, async, child...) you can enable them manually by defining
4769 C<EV_watchertype_ENABLE> to C<1> instead.
4771 =item C<32> - enable all backends
4773 This enables all backends - without this feature, you need to enable at
4774 least one backend manually (C<EV_USE_SELECT> is a good choice).
4776 =item C<64> - enable OS-specific "helper" APIs
4778 Enable inotify, eventfd, signalfd and similar OS-specific helper APIs by
4783 Compiling with C<gcc -Os -DEV_STANDALONE -DEV_USE_EPOLL=1 -DEV_FEATURES=0>
4784 reduces the compiled size of libev from 24.7Kb code/2.8Kb data to 6.5Kb
4785 code/0.3Kb data on my GNU/Linux amd64 system, while still giving you I/O
4786 watchers, timers and monotonic clock support.
4788 With an intelligent-enough linker (gcc+binutils are intelligent enough
4789 when you use C<-Wl,--gc-sections -ffunction-sections>) functions unused by
4790 your program might be left out as well - a binary starting a timer and an
4791 I/O watcher then might come out at only 5Kb.
4795 If this symbol is defined (by default it is not), then all identifiers
4796 will have static linkage. This means that libev will not export any
4797 identifiers, and you cannot link against libev anymore. This can be useful
4798 when you embed libev, only want to use libev functions in a single file,
4799 and do not want its identifiers to be visible.
4801 To use this, define C<EV_API_STATIC> and include F<ev.c> in the file that
4804 This option only works when libev is compiled with a C compiler, as C++
4805 doesn't support the required declaration syntax.
4807 =item EV_AVOID_STDIO
4809 If this is set to C<1> at compiletime, then libev will avoid using stdio
4810 functions (printf, scanf, perror etc.). This will increase the code size
4811 somewhat, but if your program doesn't otherwise depend on stdio and your
4812 libc allows it, this avoids linking in the stdio library which is quite
4815 Note that error messages might become less precise when this option is
4820 The highest supported signal number, +1 (or, the number of
4821 signals): Normally, libev tries to deduce the maximum number of signals
4822 automatically, but sometimes this fails, in which case it can be
4823 specified. Also, using a lower number than detected (C<32> should be
4824 good for about any system in existence) can save some memory, as libev
4825 statically allocates some 12-24 bytes per signal number.
4827 =item EV_PID_HASHSIZE
4829 C<ev_child> watchers use a small hash table to distribute workload by
4830 pid. The default size is C<16> (or C<1> with C<EV_FEATURES> disabled),
4831 usually more than enough. If you need to manage thousands of children you
4832 might want to increase this value (I<must> be a power of two).
4834 =item EV_INOTIFY_HASHSIZE
4836 C<ev_stat> watchers use a small hash table to distribute workload by
4837 inotify watch id. The default size is C<16> (or C<1> with C<EV_FEATURES>
4838 disabled), usually more than enough. If you need to manage thousands of
4839 C<ev_stat> watchers you might want to increase this value (I<must> be a
4844 Heaps are not very cache-efficient. To improve the cache-efficiency of the
4845 timer and periodics heaps, libev uses a 4-heap when this symbol is defined
4846 to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
4847 faster performance with many (thousands) of watchers.
4849 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4852 =item EV_HEAP_CACHE_AT
4854 Heaps are not very cache-efficient. To improve the cache-efficiency of the
4855 timer and periodics heaps, libev can cache the timestamp (I<at>) within
4856 the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
4857 which uses 8-12 bytes more per watcher and a few hundred bytes more code,
4858 but avoids random read accesses on heap changes. This improves performance
4859 noticeably with many (hundreds) of watchers.
4861 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4866 Controls how much internal verification (see C<ev_verify ()>) will
4867 be done: If set to C<0>, no internal verification code will be compiled
4868 in. If set to C<1>, then verification code will be compiled in, but not
4869 called. If set to C<2>, then the internal verification code will be
4870 called once per loop, which can slow down libev. If set to C<3>, then the
4871 verification code will be called very frequently, which will slow down
4874 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4879 By default, all watchers have a C<void *data> member. By redefining
4880 this macro to something else you can include more and other types of
4881 members. You have to define it each time you include one of the files,
4882 though, and it must be identical each time.
4884 For example, the perl EV module uses something like this:
4887 SV *self; /* contains this struct */ \
4888 SV *cb_sv, *fh /* note no trailing ";" */
4890 =item EV_CB_DECLARE (type)
4892 =item EV_CB_INVOKE (watcher, revents)
4894 =item ev_set_cb (ev, cb)
4896 Can be used to change the callback member declaration in each watcher,
4897 and the way callbacks are invoked and set. Must expand to a struct member
4898 definition and a statement, respectively. See the F<ev.h> header file for
4899 their default definitions. One possible use for overriding these is to
4900 avoid the C<struct ev_loop *> as first argument in all cases, or to use
4901 method calls instead of plain function calls in C++.
4905 =head2 EXPORTED API SYMBOLS
4907 If you need to re-export the API (e.g. via a DLL) and you need a list of
4908 exported symbols, you can use the provided F<Symbol.*> files which list
4909 all public symbols, one per line:
4911 Symbols.ev for libev proper
4912 Symbols.event for the libevent emulation
4914 This can also be used to rename all public symbols to avoid clashes with
4915 multiple versions of libev linked together (which is obviously bad in
4916 itself, but sometimes it is inconvenient to avoid this).
4918 A sed command like this will create wrapper C<#define>'s that you need to
4919 include before including F<ev.h>:
4921 <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
4923 This would create a file F<wrap.h> which essentially looks like this:
4925 #define ev_backend myprefix_ev_backend
4926 #define ev_check_start myprefix_ev_check_start
4927 #define ev_check_stop myprefix_ev_check_stop
4932 For a real-world example of a program the includes libev
4933 verbatim, you can have a look at the EV perl module
4934 (L<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
4935 the F<libev/> subdirectory and includes them in the F<EV/EVAPI.h> (public
4936 interface) and F<EV.xs> (implementation) files. Only the F<EV.xs> file
4937 will be compiled. It is pretty complex because it provides its own header
4940 The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
4941 that everybody includes and which overrides some configure choices:
4943 #define EV_FEATURES 8
4944 #define EV_USE_SELECT 1
4945 #define EV_PREPARE_ENABLE 1
4946 #define EV_IDLE_ENABLE 1
4947 #define EV_SIGNAL_ENABLE 1
4948 #define EV_CHILD_ENABLE 1
4949 #define EV_USE_STDEXCEPT 0
4950 #define EV_CONFIG_H <config.h>
4954 And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
4959 =head1 INTERACTION WITH OTHER PROGRAMS, LIBRARIES OR THE ENVIRONMENT
4961 =head2 THREADS AND COROUTINES
4965 All libev functions are reentrant and thread-safe unless explicitly
4966 documented otherwise, but libev implements no locking itself. This means
4967 that you can use as many loops as you want in parallel, as long as there
4968 are no concurrent calls into any libev function with the same loop
4969 parameter (C<ev_default_*> calls have an implicit default loop parameter,
4970 of course): libev guarantees that different event loops share no data
4971 structures that need any locking.
4973 Or to put it differently: calls with different loop parameters can be done
4974 concurrently from multiple threads, calls with the same loop parameter
4975 must be done serially (but can be done from different threads, as long as
4976 only one thread ever is inside a call at any point in time, e.g. by using
4979 Specifically to support threads (and signal handlers), libev implements
4980 so-called C<ev_async> watchers, which allow some limited form of
4981 concurrency on the same event loop, namely waking it up "from the
4984 If you want to know which design (one loop, locking, or multiple loops
4985 without or something else still) is best for your problem, then I cannot
4986 help you, but here is some generic advice:
4990 =item * most applications have a main thread: use the default libev loop
4991 in that thread, or create a separate thread running only the default loop.
4993 This helps integrating other libraries or software modules that use libev
4994 themselves and don't care/know about threading.
4996 =item * one loop per thread is usually a good model.
4998 Doing this is almost never wrong, sometimes a better-performance model
4999 exists, but it is always a good start.
5001 =item * other models exist, such as the leader/follower pattern, where one
5002 loop is handed through multiple threads in a kind of round-robin fashion.
5004 Choosing a model is hard - look around, learn, know that usually you can do
5005 better than you currently do :-)
5007 =item * often you need to talk to some other thread which blocks in the
5010 C<ev_async> watchers can be used to wake them up from other threads safely
5011 (or from signal contexts...).
5013 An example use would be to communicate signals or other events that only
5014 work in the default loop by registering the signal watcher with the
5015 default loop and triggering an C<ev_async> watcher from the default loop
5016 watcher callback into the event loop interested in the signal.
5020 See also L</THREAD LOCKING EXAMPLE>.
5024 Libev is very accommodating to coroutines ("cooperative threads"):
5025 libev fully supports nesting calls to its functions from different
5026 coroutines (e.g. you can call C<ev_run> on the same loop from two
5027 different coroutines, and switch freely between both coroutines running
5028 the loop, as long as you don't confuse yourself). The only exception is
5029 that you must not do this from C<ev_periodic> reschedule callbacks.
5031 Care has been taken to ensure that libev does not keep local state inside
5032 C<ev_run>, and other calls do not usually allow for coroutine switches as
5033 they do not call any callbacks.
5035 =head2 COMPILER WARNINGS
5037 Depending on your compiler and compiler settings, you might get no or a
5038 lot of warnings when compiling libev code. Some people are apparently
5041 However, these are unavoidable for many reasons. For one, each compiler
5042 has different warnings, and each user has different tastes regarding
5043 warning options. "Warn-free" code therefore cannot be a goal except when
5044 targeting a specific compiler and compiler-version.
5046 Another reason is that some compiler warnings require elaborate
5047 workarounds, or other changes to the code that make it less clear and less
5050 And of course, some compiler warnings are just plain stupid, or simply
5051 wrong (because they don't actually warn about the condition their message
5052 seems to warn about). For example, certain older gcc versions had some
5053 warnings that resulted in an extreme number of false positives. These have
5054 been fixed, but some people still insist on making code warn-free with
5055 such buggy versions.
5057 While libev is written to generate as few warnings as possible,
5058 "warn-free" code is not a goal, and it is recommended not to build libev
5059 with any compiler warnings enabled unless you are prepared to cope with
5060 them (e.g. by ignoring them). Remember that warnings are just that:
5061 warnings, not errors, or proof of bugs.
5066 Valgrind has a special section here because it is a popular tool that is
5067 highly useful. Unfortunately, valgrind reports are very hard to interpret.
5069 If you think you found a bug (memory leak, uninitialised data access etc.)
5070 in libev, then check twice: If valgrind reports something like:
5072 ==2274== definitely lost: 0 bytes in 0 blocks.
5073 ==2274== possibly lost: 0 bytes in 0 blocks.
5074 ==2274== still reachable: 256 bytes in 1 blocks.
5076 Then there is no memory leak, just as memory accounted to global variables
5077 is not a memleak - the memory is still being referenced, and didn't leak.
5079 Similarly, under some circumstances, valgrind might report kernel bugs
5080 as if it were a bug in libev (e.g. in realloc or in the poll backend,
5081 although an acceptable workaround has been found here), or it might be
5084 Keep in mind that valgrind is a very good tool, but only a tool. Don't
5085 make it into some kind of religion.
5087 If you are unsure about something, feel free to contact the mailing list
5088 with the full valgrind report and an explanation on why you think this
5089 is a bug in libev (best check the archives, too :). However, don't be
5090 annoyed when you get a brisk "this is no bug" answer and take the chance
5091 of learning how to interpret valgrind properly.
5093 If you need, for some reason, empty reports from valgrind for your project
5094 I suggest using suppression lists.
5097 =head1 PORTABILITY NOTES
5099 =head2 GNU/LINUX 32 BIT LIMITATIONS
5101 GNU/Linux is the only common platform that supports 64 bit file/large file
5102 interfaces but I<disables> them by default.
5104 That means that libev compiled in the default environment doesn't support
5105 files larger than 2GiB or so, which mainly affects C<ev_stat> watchers.
5107 Unfortunately, many programs try to work around this GNU/Linux issue
5108 by enabling the large file API, which makes them incompatible with the
5109 standard libev compiled for their system.
5111 Likewise, libev cannot enable the large file API itself as this would
5112 suddenly make it incompatible to the default compile time environment,
5113 i.e. all programs not using special compile switches.
5115 =head2 OS/X AND DARWIN BUGS
5117 The whole thing is a bug if you ask me - basically any system interface
5118 you touch is broken, whether it is locales, poll, kqueue or even the
5121 =head3 C<kqueue> is buggy
5123 The kqueue syscall is broken in all known versions - most versions support
5124 only sockets, many support pipes.
5126 Libev tries to work around this by not using C<kqueue> by default on this
5127 rotten platform, but of course you can still ask for it when creating a
5128 loop - embedding a socket-only kqueue loop into a select-based one is
5129 probably going to work well.
5131 =head3 C<poll> is buggy
5133 Instead of fixing C<kqueue>, Apple replaced their (working) C<poll>
5134 implementation by something calling C<kqueue> internally around the 10.5.6
5135 release, so now C<kqueue> I<and> C<poll> are broken.
5137 Libev tries to work around this by not using C<poll> by default on
5138 this rotten platform, but of course you can still ask for it when creating
5141 =head3 C<select> is buggy
5143 All that's left is C<select>, and of course Apple found a way to fuck this
5144 one up as well: On OS/X, C<select> actively limits the number of file
5145 descriptors you can pass in to 1024 - your program suddenly crashes when
5148 There is an undocumented "workaround" for this - defining
5149 C<_DARWIN_UNLIMITED_SELECT>, which libev tries to use, so select I<should>
5152 =head2 SOLARIS PROBLEMS AND WORKAROUNDS
5154 =head3 C<errno> reentrancy
5156 The default compile environment on Solaris is unfortunately so
5157 thread-unsafe that you can't even use components/libraries compiled
5158 without C<-D_REENTRANT> in a threaded program, which, of course, isn't
5159 defined by default. A valid, if stupid, implementation choice.
5161 If you want to use libev in threaded environments you have to make sure
5162 it's compiled with C<_REENTRANT> defined.
5164 =head3 Event port backend
5166 The scalable event interface for Solaris is called "event
5167 ports". Unfortunately, this mechanism is very buggy in all major
5168 releases. If you run into high CPU usage, your program freezes or you get
5169 a large number of spurious wakeups, make sure you have all the relevant
5170 and latest kernel patches applied. No, I don't know which ones, but there
5171 are multiple ones to apply, and afterwards, event ports actually work
5174 If you can't get it to work, you can try running the program by setting
5175 the environment variable C<LIBEV_FLAGS=3> to only allow C<poll> and
5180 AIX unfortunately has a broken C<poll.h> header. Libev works around
5181 this by trying to avoid the poll backend altogether (i.e. it's not even
5182 compiled in), which normally isn't a big problem as C<select> works fine
5183 with large bitsets on AIX, and AIX is dead anyway.
5185 =head2 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
5187 =head3 General issues
5189 Win32 doesn't support any of the standards (e.g. POSIX) that libev
5190 requires, and its I/O model is fundamentally incompatible with the POSIX
5191 model. Libev still offers limited functionality on this platform in
5192 the form of the C<EVBACKEND_SELECT> backend, and only supports socket
5193 descriptors. This only applies when using Win32 natively, not when using
5194 e.g. cygwin. Actually, it only applies to the microsofts own compilers,
5195 as every compiler comes with a slightly differently broken/incompatible
5198 Lifting these limitations would basically require the full
5199 re-implementation of the I/O system. If you are into this kind of thing,
5200 then note that glib does exactly that for you in a very portable way (note
5201 also that glib is the slowest event library known to man).
5203 There is no supported compilation method available on windows except
5204 embedding it into other applications.
5206 Sensible signal handling is officially unsupported by Microsoft - libev
5207 tries its best, but under most conditions, signals will simply not work.
5209 Not a libev limitation but worth mentioning: windows apparently doesn't
5210 accept large writes: instead of resulting in a partial write, windows will
5211 either accept everything or return C<ENOBUFS> if the buffer is too large,
5212 so make sure you only write small amounts into your sockets (less than a
5213 megabyte seems safe, but this apparently depends on the amount of memory
5216 Due to the many, low, and arbitrary limits on the win32 platform and
5217 the abysmal performance of winsockets, using a large number of sockets
5218 is not recommended (and not reasonable). If your program needs to use
5219 more than a hundred or so sockets, then likely it needs to use a totally
5220 different implementation for windows, as libev offers the POSIX readiness
5221 notification model, which cannot be implemented efficiently on windows
5222 (due to Microsoft monopoly games).
5224 A typical way to use libev under windows is to embed it (see the embedding
5225 section for details) and use the following F<evwrap.h> header file instead
5228 #define EV_STANDALONE /* keeps ev from requiring config.h */
5229 #define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
5233 And compile the following F<evwrap.c> file into your project (make sure
5234 you do I<not> compile the F<ev.c> or any other embedded source files!):
5239 =head3 The winsocket C<select> function
5241 The winsocket C<select> function doesn't follow POSIX in that it
5242 requires socket I<handles> and not socket I<file descriptors> (it is
5243 also extremely buggy). This makes select very inefficient, and also
5244 requires a mapping from file descriptors to socket handles (the Microsoft
5245 C runtime provides the function C<_open_osfhandle> for this). See the
5246 discussion of the C<EV_SELECT_USE_FD_SET>, C<EV_SELECT_IS_WINSOCKET> and
5247 C<EV_FD_TO_WIN32_HANDLE> preprocessor symbols for more info.
5249 The configuration for a "naked" win32 using the Microsoft runtime
5250 libraries and raw winsocket select is:
5252 #define EV_USE_SELECT 1
5253 #define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
5255 Note that winsockets handling of fd sets is O(n), so you can easily get a
5256 complexity in the O(n²) range when using win32.
5258 =head3 Limited number of file descriptors
5260 Windows has numerous arbitrary (and low) limits on things.
5262 Early versions of winsocket's select only supported waiting for a maximum
5263 of C<64> handles (probably owning to the fact that all windows kernels
5264 can only wait for C<64> things at the same time internally; Microsoft
5265 recommends spawning a chain of threads and wait for 63 handles and the
5266 previous thread in each. Sounds great!).
5268 Newer versions support more handles, but you need to define C<FD_SETSIZE>
5269 to some high number (e.g. C<2048>) before compiling the winsocket select
5270 call (which might be in libev or elsewhere, for example, perl and many
5271 other interpreters do their own select emulation on windows).
5273 Another limit is the number of file descriptors in the Microsoft runtime
5274 libraries, which by default is C<64> (there must be a hidden I<64>
5275 fetish or something like this inside Microsoft). You can increase this
5276 by calling C<_setmaxstdio>, which can increase this limit to C<2048>
5277 (another arbitrary limit), but is broken in many versions of the Microsoft
5278 runtime libraries. This might get you to about C<512> or C<2048> sockets
5279 (depending on windows version and/or the phase of the moon). To get more,
5280 you need to wrap all I/O functions and provide your own fd management, but
5281 the cost of calling select (O(n²)) will likely make this unworkable.
5283 =head2 PORTABILITY REQUIREMENTS
5285 In addition to a working ISO-C implementation and of course the
5286 backend-specific APIs, libev relies on a few additional extensions:
5290 =item C<void (*)(ev_watcher_type *, int revents)> must have compatible
5291 calling conventions regardless of C<ev_watcher_type *>.
5293 Libev assumes not only that all watcher pointers have the same internal
5294 structure (guaranteed by POSIX but not by ISO C for example), but it also
5295 assumes that the same (machine) code can be used to call any watcher
5296 callback: The watcher callbacks have different type signatures, but libev
5297 calls them using an C<ev_watcher *> internally.
5299 =item pointer accesses must be thread-atomic
5301 Accessing a pointer value must be atomic, it must both be readable and
5302 writable in one piece - this is the case on all current architectures.
5304 =item C<sig_atomic_t volatile> must be thread-atomic as well
5306 The type C<sig_atomic_t volatile> (or whatever is defined as
5307 C<EV_ATOMIC_T>) must be atomic with respect to accesses from different
5308 threads. This is not part of the specification for C<sig_atomic_t>, but is
5309 believed to be sufficiently portable.
5311 =item C<sigprocmask> must work in a threaded environment
5313 Libev uses C<sigprocmask> to temporarily block signals. This is not
5314 allowed in a threaded program (C<pthread_sigmask> has to be used). Typical
5315 pthread implementations will either allow C<sigprocmask> in the "main
5316 thread" or will block signals process-wide, both behaviours would
5317 be compatible with libev. Interaction between C<sigprocmask> and
5318 C<pthread_sigmask> could complicate things, however.
5320 The most portable way to handle signals is to block signals in all threads
5321 except the initial one, and run the signal handling loop in the initial
5324 =item C<long> must be large enough for common memory allocation sizes
5326 To improve portability and simplify its API, libev uses C<long> internally
5327 instead of C<size_t> when allocating its data structures. On non-POSIX
5328 systems (Microsoft...) this might be unexpectedly low, but is still at
5329 least 31 bits everywhere, which is enough for hundreds of millions of
5332 =item C<double> must hold a time value in seconds with enough accuracy
5334 The type C<double> is used to represent timestamps. It is required to
5335 have at least 51 bits of mantissa (and 9 bits of exponent), which is
5336 good enough for at least into the year 4000 with millisecond accuracy
5337 (the design goal for libev). This requirement is overfulfilled by
5338 implementations using IEEE 754, which is basically all existing ones.
5340 With IEEE 754 doubles, you get microsecond accuracy until at least the
5341 year 2255 (and millisecond accuracy till the year 287396 - by then, libev
5342 is either obsolete or somebody patched it to use C<long double> or
5343 something like that, just kidding).
5347 If you know of other additional requirements drop me a note.
5350 =head1 ALGORITHMIC COMPLEXITIES
5352 In this section the complexities of (many of) the algorithms used inside
5353 libev will be documented. For complexity discussions about backends see
5354 the documentation for C<ev_default_init>.
5356 All of the following are about amortised time: If an array needs to be
5357 extended, libev needs to realloc and move the whole array, but this
5358 happens asymptotically rarer with higher number of elements, so O(1) might
5359 mean that libev does a lengthy realloc operation in rare cases, but on
5360 average it is much faster and asymptotically approaches constant time.
5364 =item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
5366 This means that, when you have a watcher that triggers in one hour and
5367 there are 100 watchers that would trigger before that, then inserting will
5368 have to skip roughly seven (C<ld 100>) of these watchers.
5370 =item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
5372 That means that changing a timer costs less than removing/adding them,
5373 as only the relative motion in the event queue has to be paid for.
5375 =item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
5377 These just add the watcher into an array or at the head of a list.
5379 =item Stopping check/prepare/idle/fork/async watchers: O(1)
5381 =item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
5383 These watchers are stored in lists, so they need to be walked to find the
5384 correct watcher to remove. The lists are usually short (you don't usually
5385 have many watchers waiting for the same fd or signal: one is typical, two
5388 =item Finding the next timer in each loop iteration: O(1)
5390 By virtue of using a binary or 4-heap, the next timer is always found at a
5391 fixed position in the storage array.
5393 =item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
5395 A change means an I/O watcher gets started or stopped, which requires
5396 libev to recalculate its status (and possibly tell the kernel, depending
5397 on backend and whether C<ev_io_set> was used).
5399 =item Activating one watcher (putting it into the pending state): O(1)
5401 =item Priority handling: O(number_of_priorities)
5403 Priorities are implemented by allocating some space for each
5404 priority. When doing priority-based operations, libev usually has to
5405 linearly search all the priorities, but starting/stopping and activating
5406 watchers becomes O(1) with respect to priority handling.
5408 =item Sending an ev_async: O(1)
5410 =item Processing ev_async_send: O(number_of_async_watchers)
5412 =item Processing signals: O(max_signal_number)
5414 Sending involves a system call I<iff> there were no other C<ev_async_send>
5415 calls in the current loop iteration and the loop is currently
5416 blocked. Checking for async and signal events involves iterating over all
5417 running async watchers or all signal numbers.
5422 =head1 PORTING FROM LIBEV 3.X TO 4.X
5424 The major version 4 introduced some incompatible changes to the API.
5426 At the moment, the C<ev.h> header file provides compatibility definitions
5427 for all changes, so most programs should still compile. The compatibility
5428 layer might be removed in later versions of libev, so better update to the
5429 new API early than late.
5433 =item C<EV_COMPAT3> backwards compatibility mechanism
5435 The backward compatibility mechanism can be controlled by
5436 C<EV_COMPAT3>. See L</"PREPROCESSOR SYMBOLS/MACROS"> in the L</EMBEDDING>
5439 =item C<ev_default_destroy> and C<ev_default_fork> have been removed
5441 These calls can be replaced easily by their C<ev_loop_xxx> counterparts:
5443 ev_loop_destroy (EV_DEFAULT_UC);
5444 ev_loop_fork (EV_DEFAULT);
5446 =item function/symbol renames
5448 A number of functions and symbols have been renamed:
5451 EVLOOP_NONBLOCK => EVRUN_NOWAIT
5452 EVLOOP_ONESHOT => EVRUN_ONCE
5454 ev_unloop => ev_break
5455 EVUNLOOP_CANCEL => EVBREAK_CANCEL
5456 EVUNLOOP_ONE => EVBREAK_ONE
5457 EVUNLOOP_ALL => EVBREAK_ALL
5459 EV_TIMEOUT => EV_TIMER
5461 ev_loop_count => ev_iteration
5462 ev_loop_depth => ev_depth
5463 ev_loop_verify => ev_verify
5465 Most functions working on C<struct ev_loop> objects don't have an
5466 C<ev_loop_> prefix, so it was removed; C<ev_loop>, C<ev_unloop> and
5467 associated constants have been renamed to not collide with the C<struct
5468 ev_loop> anymore and C<EV_TIMER> now follows the same naming scheme
5469 as all other watcher types. Note that C<ev_loop_fork> is still called
5470 C<ev_loop_fork> because it would otherwise clash with the C<ev_fork>
5473 =item C<EV_MINIMAL> mechanism replaced by C<EV_FEATURES>
5475 The preprocessor symbol C<EV_MINIMAL> has been replaced by a different
5476 mechanism, C<EV_FEATURES>. Programs using C<EV_MINIMAL> usually compile
5477 and work, but the library code will of course be larger.
5488 A watcher is active as long as it has been started and not yet stopped.
5489 See L</WATCHER STATES> for details.
5493 In this document, an application is whatever is using libev.
5497 The part of the code dealing with the operating system interfaces.
5501 The address of a function that is called when some event has been
5502 detected. Callbacks are being passed the event loop, the watcher that
5503 received the event, and the actual event bitset.
5505 =item callback/watcher invocation
5507 The act of calling the callback associated with a watcher.
5511 A change of state of some external event, such as data now being available
5512 for reading on a file descriptor, time having passed or simply not having
5513 any other events happening anymore.
5515 In libev, events are represented as single bits (such as C<EV_READ> or
5520 A software package implementing an event model and loop.
5524 An entity that handles and processes external events and converts them
5525 into callback invocations.
5529 The model used to describe how an event loop handles and processes
5530 watchers and events.
5534 A watcher is pending as soon as the corresponding event has been
5535 detected. See L</WATCHER STATES> for details.
5539 The physical time that is observed. It is apparently strictly monotonic :)
5541 =item wall-clock time
5543 The time and date as shown on clocks. Unlike real time, it can actually
5544 be wrong and jump forwards and backwards, e.g. when you adjust your
5549 A data structure that describes interest in certain events. Watchers need
5550 to be started (attached to an event loop) before they can receive events.
5556 Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael
5557 Magnusson and Emanuele Giaquinta, and minor corrections by many others.