ViewVC Help
View File | Revision Log | Show Annotations | Download File
Revision: 1.426
Committed: Sat Feb 23 23:06:40 2013 UTC (11 years, 4 months ago) by root
Branch: MAIN
CVS Tags: rel-4_15
Changes since 1.425: +4 -3 lines
Log Message:
*** empty log message ***

File Contents

# Content
1 =head1 NAME
3 libev - a high performance full-featured event loop written in C
5 =head1 SYNOPSIS
7 #include <ev.h>
11 // a single header file is required
12 #include <ev.h>
14 #include <stdio.h> // for puts
16 // every watcher type has its own typedef'd struct
17 // with the name ev_TYPE
18 ev_io stdin_watcher;
19 ev_timer timeout_watcher;
21 // all watcher callbacks have a similar signature
22 // this callback is called when data is readable on stdin
23 static void
24 stdin_cb (EV_P_ ev_io *w, int revents)
25 {
26 puts ("stdin ready");
27 // for one-shot events, one must manually stop the watcher
28 // with its corresponding stop function.
29 ev_io_stop (EV_A_ w);
31 // this causes all nested ev_run's to stop iterating
32 ev_break (EV_A_ EVBREAK_ALL);
33 }
35 // another callback, this time for a time-out
36 static void
37 timeout_cb (EV_P_ ev_timer *w, int revents)
38 {
39 puts ("timeout");
40 // this causes the innermost ev_run to stop iterating
41 ev_break (EV_A_ EVBREAK_ONE);
42 }
44 int
45 main (void)
46 {
47 // use the default event loop unless you have special needs
48 struct ev_loop *loop = EV_DEFAULT;
50 // initialise an io watcher, then start it
51 // this one will watch for stdin to become readable
52 ev_io_init (&stdin_watcher, stdin_cb, /*STDIN_FILENO*/ 0, EV_READ);
53 ev_io_start (loop, &stdin_watcher);
55 // initialise a timer watcher, then start it
56 // simple non-repeating 5.5 second timeout
57 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
58 ev_timer_start (loop, &timeout_watcher);
60 // now wait for events to arrive
61 ev_run (loop, 0);
63 // break was called, so exit
64 return 0;
65 }
69 This document documents the libev software package.
71 The newest version of this document is also available as an html-formatted
72 web page you might find easier to navigate when reading it for the first
73 time: L<>.
75 While this document tries to be as complete as possible in documenting
76 libev, its usage and the rationale behind its design, it is not a tutorial
77 on event-based programming, nor will it introduce event-based programming
78 with libev.
80 Familiarity with event based programming techniques in general is assumed
81 throughout this document.
85 This manual tries to be very detailed, but unfortunately, this also makes
86 it very long. If you just want to know the basics of libev, I suggest
87 reading L</ANATOMY OF A WATCHER>, then the L</EXAMPLE PROGRAM> above and
88 look up the missing functions in L</GLOBAL FUNCTIONS> and the C<ev_io> and
89 C<ev_timer> sections in L</WATCHER TYPES>.
91 =head1 ABOUT LIBEV
93 Libev is an event loop: you register interest in certain events (such as a
94 file descriptor being readable or a timeout occurring), and it will manage
95 these event sources and provide your program with events.
97 To do this, it must take more or less complete control over your process
98 (or thread) by executing the I<event loop> handler, and will then
99 communicate events via a callback mechanism.
101 You register interest in certain events by registering so-called I<event
102 watchers>, which are relatively small C structures you initialise with the
103 details of the event, and then hand it over to libev by I<starting> the
104 watcher.
106 =head2 FEATURES
108 Libev supports C<select>, C<poll>, the Linux-specific C<epoll>, the
109 BSD-specific C<kqueue> and the Solaris-specific event port mechanisms
110 for file descriptor events (C<ev_io>), the Linux C<inotify> interface
111 (for C<ev_stat>), Linux eventfd/signalfd (for faster and cleaner
112 inter-thread wakeup (C<ev_async>)/signal handling (C<ev_signal>)) relative
113 timers (C<ev_timer>), absolute timers with customised rescheduling
114 (C<ev_periodic>), synchronous signals (C<ev_signal>), process status
115 change events (C<ev_child>), and event watchers dealing with the event
116 loop mechanism itself (C<ev_idle>, C<ev_embed>, C<ev_prepare> and
117 C<ev_check> watchers) as well as file watchers (C<ev_stat>) and even
118 limited support for fork events (C<ev_fork>).
120 It also is quite fast (see this
121 L<benchmark|> comparing it to libevent
122 for example).
124 =head2 CONVENTIONS
126 Libev is very configurable. In this manual the default (and most common)
127 configuration will be described, which supports multiple event loops. For
128 more info about various configuration options please have a look at
129 B<EMBED> section in this manual. If libev was configured without support
130 for multiple event loops, then all functions taking an initial argument of
131 name C<loop> (which is always of type C<struct ev_loop *>) will not have
132 this argument.
136 Libev represents time as a single floating point number, representing
137 the (fractional) number of seconds since the (POSIX) epoch (in practice
138 somewhere near the beginning of 1970, details are complicated, don't
139 ask). This type is called C<ev_tstamp>, which is what you should use
140 too. It usually aliases to the C<double> type in C. When you need to do
141 any calculations on it, you should treat it as some floating point value.
143 Unlike the name component C<stamp> might indicate, it is also used for
144 time differences (e.g. delays) throughout libev.
148 Libev knows three classes of errors: operating system errors, usage errors
149 and internal errors (bugs).
151 When libev catches an operating system error it cannot handle (for example
152 a system call indicating a condition libev cannot fix), it calls the callback
153 set via C<ev_set_syserr_cb>, which is supposed to fix the problem or
154 abort. The default is to print a diagnostic message and to call C<abort
155 ()>.
157 When libev detects a usage error such as a negative timer interval, then
158 it will print a diagnostic message and abort (via the C<assert> mechanism,
159 so C<NDEBUG> will disable this checking): these are programming errors in
160 the libev caller and need to be fixed there.
162 Libev also has a few internal error-checking C<assert>ions, and also has
163 extensive consistency checking code. These do not trigger under normal
164 circumstances, as they indicate either a bug in libev or worse.
169 These functions can be called anytime, even before initialising the
170 library in any way.
172 =over 4
174 =item ev_tstamp ev_time ()
176 Returns the current time as libev would use it. Please note that the
177 C<ev_now> function is usually faster and also often returns the timestamp
178 you actually want to know. Also interesting is the combination of
179 C<ev_now_update> and C<ev_now>.
181 =item ev_sleep (ev_tstamp interval)
183 Sleep for the given interval: The current thread will be blocked
184 until either it is interrupted or the given time interval has
185 passed (approximately - it might return a bit earlier even if not
186 interrupted). Returns immediately if C<< interval <= 0 >>.
188 Basically this is a sub-second-resolution C<sleep ()>.
190 The range of the C<interval> is limited - libev only guarantees to work
191 with sleep times of up to one day (C<< interval <= 86400 >>).
193 =item int ev_version_major ()
195 =item int ev_version_minor ()
197 You can find out the major and minor ABI version numbers of the library
198 you linked against by calling the functions C<ev_version_major> and
199 C<ev_version_minor>. If you want, you can compare against the global
200 symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the
201 version of the library your program was compiled against.
203 These version numbers refer to the ABI version of the library, not the
204 release version.
206 Usually, it's a good idea to terminate if the major versions mismatch,
207 as this indicates an incompatible change. Minor versions are usually
208 compatible to older versions, so a larger minor version alone is usually
209 not a problem.
211 Example: Make sure we haven't accidentally been linked against the wrong
212 version (note, however, that this will not detect other ABI mismatches,
213 such as LFS or reentrancy).
215 assert (("libev version mismatch",
216 ev_version_major () == EV_VERSION_MAJOR
217 && ev_version_minor () >= EV_VERSION_MINOR));
219 =item unsigned int ev_supported_backends ()
221 Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
222 value) compiled into this binary of libev (independent of their
223 availability on the system you are running on). See C<ev_default_loop> for
224 a description of the set values.
226 Example: make sure we have the epoll method, because yeah this is cool and
227 a must have and can we have a torrent of it please!!!11
229 assert (("sorry, no epoll, no sex",
230 ev_supported_backends () & EVBACKEND_EPOLL));
232 =item unsigned int ev_recommended_backends ()
234 Return the set of all backends compiled into this binary of libev and
235 also recommended for this platform, meaning it will work for most file
236 descriptor types. This set is often smaller than the one returned by
237 C<ev_supported_backends>, as for example kqueue is broken on most BSDs
238 and will not be auto-detected unless you explicitly request it (assuming
239 you know what you are doing). This is the set of backends that libev will
240 probe for if you specify no backends explicitly.
242 =item unsigned int ev_embeddable_backends ()
244 Returns the set of backends that are embeddable in other event loops. This
245 value is platform-specific but can include backends not available on the
246 current system. To find which embeddable backends might be supported on
247 the current system, you would need to look at C<ev_embeddable_backends ()
248 & ev_supported_backends ()>, likewise for recommended ones.
250 See the description of C<ev_embed> watchers for more info.
252 =item ev_set_allocator (void *(*cb)(void *ptr, long size) throw ())
254 Sets the allocation function to use (the prototype is similar - the
255 semantics are identical to the C<realloc> C89/SuS/POSIX function). It is
256 used to allocate and free memory (no surprises here). If it returns zero
257 when memory needs to be allocated (C<size != 0>), the library might abort
258 or take some potentially destructive action.
260 Since some systems (at least OpenBSD and Darwin) fail to implement
261 correct C<realloc> semantics, libev will use a wrapper around the system
262 C<realloc> and C<free> functions by default.
264 You could override this function in high-availability programs to, say,
265 free some memory if it cannot allocate memory, to use a special allocator,
266 or even to sleep a while and retry until some memory is available.
268 Example: Replace the libev allocator with one that waits a bit and then
269 retries (example requires a standards-compliant C<realloc>).
271 static void *
272 persistent_realloc (void *ptr, size_t size)
273 {
274 for (;;)
275 {
276 void *newptr = realloc (ptr, size);
278 if (newptr)
279 return newptr;
281 sleep (60);
282 }
283 }
285 ...
286 ev_set_allocator (persistent_realloc);
288 =item ev_set_syserr_cb (void (*cb)(const char *msg) throw ())
290 Set the callback function to call on a retryable system call error (such
291 as failed select, poll, epoll_wait). The message is a printable string
292 indicating the system call or subsystem causing the problem. If this
293 callback is set, then libev will expect it to remedy the situation, no
294 matter what, when it returns. That is, libev will generally retry the
295 requested operation, or, if the condition doesn't go away, do bad stuff
296 (such as abort).
298 Example: This is basically the same thing that libev does internally, too.
300 static void
301 fatal_error (const char *msg)
302 {
303 perror (msg);
304 abort ();
305 }
307 ...
308 ev_set_syserr_cb (fatal_error);
310 =item ev_feed_signal (int signum)
312 This function can be used to "simulate" a signal receive. It is completely
313 safe to call this function at any time, from any context, including signal
314 handlers or random threads.
316 Its main use is to customise signal handling in your process, especially
317 in the presence of threads. For example, you could block signals
318 by default in all threads (and specifying C<EVFLAG_NOSIGMASK> when
319 creating any loops), and in one thread, use C<sigwait> or any other
320 mechanism to wait for signals, then "deliver" them to libev by calling
321 C<ev_feed_signal>.
323 =back
327 An event loop is described by a C<struct ev_loop *> (the C<struct> is
328 I<not> optional in this case unless libev 3 compatibility is disabled, as
329 libev 3 had an C<ev_loop> function colliding with the struct name).
331 The library knows two types of such loops, the I<default> loop, which
332 supports child process events, and dynamically created event loops which
333 do not.
335 =over 4
337 =item struct ev_loop *ev_default_loop (unsigned int flags)
339 This returns the "default" event loop object, which is what you should
340 normally use when you just need "the event loop". Event loop objects and
341 the C<flags> parameter are described in more detail in the entry for
342 C<ev_loop_new>.
344 If the default loop is already initialised then this function simply
345 returns it (and ignores the flags. If that is troubling you, check
346 C<ev_backend ()> afterwards). Otherwise it will create it with the given
347 flags, which should almost always be C<0>, unless the caller is also the
348 one calling C<ev_run> or otherwise qualifies as "the main program".
350 If you don't know what event loop to use, use the one returned from this
351 function (or via the C<EV_DEFAULT> macro).
353 Note that this function is I<not> thread-safe, so if you want to use it
354 from multiple threads, you have to employ some kind of mutex (note also
355 that this case is unlikely, as loops cannot be shared easily between
356 threads anyway).
358 The default loop is the only loop that can handle C<ev_child> watchers,
359 and to do this, it always registers a handler for C<SIGCHLD>. If this is
360 a problem for your application you can either create a dynamic loop with
361 C<ev_loop_new> which doesn't do that, or you can simply overwrite the
362 C<SIGCHLD> signal handler I<after> calling C<ev_default_init>.
364 Example: This is the most typical usage.
366 if (!ev_default_loop (0))
367 fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
369 Example: Restrict libev to the select and poll backends, and do not allow
370 environment settings to be taken into account:
374 =item struct ev_loop *ev_loop_new (unsigned int flags)
376 This will create and initialise a new event loop object. If the loop
377 could not be initialised, returns false.
379 This function is thread-safe, and one common way to use libev with
380 threads is indeed to create one loop per thread, and using the default
381 loop in the "main" or "initial" thread.
383 The flags argument can be used to specify special behaviour or specific
384 backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
386 The following flags are supported:
388 =over 4
390 =item C<EVFLAG_AUTO>
392 The default flags value. Use this if you have no clue (it's the right
393 thing, believe me).
395 =item C<EVFLAG_NOENV>
397 If this flag bit is or'ed into the flag value (or the program runs setuid
398 or setgid) then libev will I<not> look at the environment variable
399 C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
400 override the flags completely if it is found in the environment. This is
401 useful to try out specific backends to test their performance, or to work
402 around bugs.
406 Instead of calling C<ev_loop_fork> manually after a fork, you can also
407 make libev check for a fork in each iteration by enabling this flag.
409 This works by calling C<getpid ()> on every iteration of the loop,
410 and thus this might slow down your event loop if you do a lot of loop
411 iterations and little real work, but is usually not noticeable (on my
412 GNU/Linux system for example, C<getpid> is actually a simple 5-insn sequence
413 without a system call and thus I<very> fast, but my GNU/Linux system also has
414 C<pthread_atfork> which is even faster).
416 The big advantage of this flag is that you can forget about fork (and
417 forget about forgetting to tell libev about forking) when you use this
418 flag.
420 This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
421 environment variable.
425 When this flag is specified, then libev will not attempt to use the
426 I<inotify> API for its C<ev_stat> watchers. Apart from debugging and
427 testing, this flag can be useful to conserve inotify file descriptors, as
428 otherwise each loop using C<ev_stat> watchers consumes one inotify handle.
432 When this flag is specified, then libev will attempt to use the
433 I<signalfd> API for its C<ev_signal> (and C<ev_child>) watchers. This API
434 delivers signals synchronously, which makes it both faster and might make
435 it possible to get the queued signal data. It can also simplify signal
436 handling with threads, as long as you properly block signals in your
437 threads that are not interested in handling them.
439 Signalfd will not be used by default as this changes your signal mask, and
440 there are a lot of shoddy libraries and programs (glib's threadpool for
441 example) that can't properly initialise their signal masks.
445 When this flag is specified, then libev will avoid to modify the signal
446 mask. Specifically, this means you have to make sure signals are unblocked
447 when you want to receive them.
449 This behaviour is useful when you want to do your own signal handling, or
450 want to handle signals only in specific threads and want to avoid libev
451 unblocking the signals.
453 It's also required by POSIX in a threaded program, as libev calls
454 C<sigprocmask>, whose behaviour is officially unspecified.
456 This flag's behaviour will become the default in future versions of libev.
458 =item C<EVBACKEND_SELECT> (value 1, portable select backend)
460 This is your standard select(2) backend. Not I<completely> standard, as
461 libev tries to roll its own fd_set with no limits on the number of fds,
462 but if that fails, expect a fairly low limit on the number of fds when
463 using this backend. It doesn't scale too well (O(highest_fd)), but its
464 usually the fastest backend for a low number of (low-numbered :) fds.
466 To get good performance out of this backend you need a high amount of
467 parallelism (most of the file descriptors should be busy). If you are
468 writing a server, you should C<accept ()> in a loop to accept as many
469 connections as possible during one iteration. You might also want to have
470 a look at C<ev_set_io_collect_interval ()> to increase the amount of
471 readiness notifications you get per iteration.
473 This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
474 C<writefds> set (and to work around Microsoft Windows bugs, also onto the
475 C<exceptfds> set on that platform).
477 =item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
479 And this is your standard poll(2) backend. It's more complicated
480 than select, but handles sparse fds better and has no artificial
481 limit on the number of fds you can use (except it will slow down
482 considerably with a lot of inactive fds). It scales similarly to select,
483 i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
484 performance tips.
486 This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
489 =item C<EVBACKEND_EPOLL> (value 4, Linux)
491 Use the linux-specific epoll(7) interface (for both pre- and post-2.6.9
492 kernels).
494 For few fds, this backend is a bit little slower than poll and select, but
495 it scales phenomenally better. While poll and select usually scale like
496 O(total_fds) where total_fds is the total number of fds (or the highest
497 fd), epoll scales either O(1) or O(active_fds).
499 The epoll mechanism deserves honorable mention as the most misdesigned
500 of the more advanced event mechanisms: mere annoyances include silently
501 dropping file descriptors, requiring a system call per change per file
502 descriptor (and unnecessary guessing of parameters), problems with dup,
503 returning before the timeout value, resulting in additional iterations
504 (and only giving 5ms accuracy while select on the same platform gives
505 0.1ms) and so on. The biggest issue is fork races, however - if a program
506 forks then I<both> parent and child process have to recreate the epoll
507 set, which can take considerable time (one syscall per file descriptor)
508 and is of course hard to detect.
510 Epoll is also notoriously buggy - embedding epoll fds I<should> work,
511 but of course I<doesn't>, and epoll just loves to report events for
512 totally I<different> file descriptors (even already closed ones, so
513 one cannot even remove them from the set) than registered in the set
514 (especially on SMP systems). Libev tries to counter these spurious
515 notifications by employing an additional generation counter and comparing
516 that against the events to filter out spurious ones, recreating the set
517 when required. Epoll also erroneously rounds down timeouts, but gives you
518 no way to know when and by how much, so sometimes you have to busy-wait
519 because epoll returns immediately despite a nonzero timeout. And last
520 not least, it also refuses to work with some file descriptors which work
521 perfectly fine with C<select> (files, many character devices...).
523 Epoll is truly the train wreck among event poll mechanisms, a frankenpoll,
524 cobbled together in a hurry, no thought to design or interaction with
525 others. Oh, the pain, will it ever stop...
527 While stopping, setting and starting an I/O watcher in the same iteration
528 will result in some caching, there is still a system call per such
529 incident (because the same I<file descriptor> could point to a different
530 I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
531 file descriptors might not work very well if you register events for both
532 file descriptors.
534 Best performance from this backend is achieved by not unregistering all
535 watchers for a file descriptor until it has been closed, if possible,
536 i.e. keep at least one watcher active per fd at all times. Stopping and
537 starting a watcher (without re-setting it) also usually doesn't cause
538 extra overhead. A fork can both result in spurious notifications as well
539 as in libev having to destroy and recreate the epoll object, which can
540 take considerable time and thus should be avoided.
542 All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
543 faster than epoll for maybe up to a hundred file descriptors, depending on
544 the usage. So sad.
546 While nominally embeddable in other event loops, this feature is broken in
547 all kernel versions tested so far.
549 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
552 =item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
554 Kqueue deserves special mention, as at the time of this writing, it
555 was broken on all BSDs except NetBSD (usually it doesn't work reliably
556 with anything but sockets and pipes, except on Darwin, where of course
557 it's completely useless). Unlike epoll, however, whose brokenness
558 is by design, these kqueue bugs can (and eventually will) be fixed
559 without API changes to existing programs. For this reason it's not being
560 "auto-detected" unless you explicitly specify it in the flags (i.e. using
561 C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
562 system like NetBSD.
564 You still can embed kqueue into a normal poll or select backend and use it
565 only for sockets (after having made sure that sockets work with kqueue on
566 the target platform). See C<ev_embed> watchers for more info.
568 It scales in the same way as the epoll backend, but the interface to the
569 kernel is more efficient (which says nothing about its actual speed, of
570 course). While stopping, setting and starting an I/O watcher does never
571 cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
572 two event changes per incident. Support for C<fork ()> is very bad (you
573 might have to leak fd's on fork, but it's more sane than epoll) and it
574 drops fds silently in similarly hard-to-detect cases.
576 This backend usually performs well under most conditions.
578 While nominally embeddable in other event loops, this doesn't work
579 everywhere, so you might need to test for this. And since it is broken
580 almost everywhere, you should only use it when you have a lot of sockets
581 (for which it usually works), by embedding it into another event loop
582 (e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
583 also broken on OS X)) and, did I mention it, using it only for sockets.
585 This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
586 C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
587 C<NOTE_EOF>.
589 =item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
591 This is not implemented yet (and might never be, unless you send me an
592 implementation). According to reports, C</dev/poll> only supports sockets
593 and is not embeddable, which would limit the usefulness of this backend
594 immensely.
596 =item C<EVBACKEND_PORT> (value 32, Solaris 10)
598 This uses the Solaris 10 event port mechanism. As with everything on Solaris,
599 it's really slow, but it still scales very well (O(active_fds)).
601 While this backend scales well, it requires one system call per active
602 file descriptor per loop iteration. For small and medium numbers of file
603 descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
604 might perform better.
606 On the positive side, this backend actually performed fully to
607 specification in all tests and is fully embeddable, which is a rare feat
608 among the OS-specific backends (I vastly prefer correctness over speed
609 hacks).
611 On the negative side, the interface is I<bizarre> - so bizarre that
612 even sun itself gets it wrong in their code examples: The event polling
613 function sometimes returns events to the caller even though an error
614 occurred, but with no indication whether it has done so or not (yes, it's
615 even documented that way) - deadly for edge-triggered interfaces where you
616 absolutely have to know whether an event occurred or not because you have
617 to re-arm the watcher.
619 Fortunately libev seems to be able to work around these idiocies.
621 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
624 =item C<EVBACKEND_ALL>
626 Try all backends (even potentially broken ones that wouldn't be tried
627 with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
630 It is definitely not recommended to use this flag, use whatever
631 C<ev_recommended_backends ()> returns, or simply do not specify a backend
632 at all.
636 Not a backend at all, but a mask to select all backend bits from a
637 C<flags> value, in case you want to mask out any backends from a flags
638 value (e.g. when modifying the C<LIBEV_FLAGS> environment variable).
640 =back
642 If one or more of the backend flags are or'ed into the flags value,
643 then only these backends will be tried (in the reverse order as listed
644 here). If none are specified, all backends in C<ev_recommended_backends
645 ()> will be tried.
647 Example: Try to create a event loop that uses epoll and nothing else.
649 struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
650 if (!epoller)
651 fatal ("no epoll found here, maybe it hides under your chair");
653 Example: Use whatever libev has to offer, but make sure that kqueue is
654 used if available.
656 struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_KQUEUE);
658 =item ev_loop_destroy (loop)
660 Destroys an event loop object (frees all memory and kernel state
661 etc.). None of the active event watchers will be stopped in the normal
662 sense, so e.g. C<ev_is_active> might still return true. It is your
663 responsibility to either stop all watchers cleanly yourself I<before>
664 calling this function, or cope with the fact afterwards (which is usually
665 the easiest thing, you can just ignore the watchers and/or C<free ()> them
666 for example).
668 Note that certain global state, such as signal state (and installed signal
669 handlers), will not be freed by this function, and related watchers (such
670 as signal and child watchers) would need to be stopped manually.
672 This function is normally used on loop objects allocated by
673 C<ev_loop_new>, but it can also be used on the default loop returned by
674 C<ev_default_loop>, in which case it is not thread-safe.
676 Note that it is not advisable to call this function on the default loop
677 except in the rare occasion where you really need to free its resources.
678 If you need dynamically allocated loops it is better to use C<ev_loop_new>
679 and C<ev_loop_destroy>.
681 =item ev_loop_fork (loop)
683 This function sets a flag that causes subsequent C<ev_run> iterations to
684 reinitialise the kernel state for backends that have one. Despite the
685 name, you can call it anytime, but it makes most sense after forking, in
686 the child process. You I<must> call it (or use C<EVFLAG_FORKCHECK>) in the
687 child before resuming or calling C<ev_run>.
689 Again, you I<have> to call it on I<any> loop that you want to re-use after
690 a fork, I<even if you do not plan to use the loop in the parent>. This is
691 because some kernel interfaces *cough* I<kqueue> *cough* do funny things
692 during fork.
694 On the other hand, you only need to call this function in the child
695 process if and only if you want to use the event loop in the child. If
696 you just fork+exec or create a new loop in the child, you don't have to
697 call it at all (in fact, C<epoll> is so badly broken that it makes a
698 difference, but libev will usually detect this case on its own and do a
699 costly reset of the backend).
701 The function itself is quite fast and it's usually not a problem to call
702 it just in case after a fork.
704 Example: Automate calling C<ev_loop_fork> on the default loop when
705 using pthreads.
707 static void
708 post_fork_child (void)
709 {
710 ev_loop_fork (EV_DEFAULT);
711 }
713 ...
714 pthread_atfork (0, 0, post_fork_child);
716 =item int ev_is_default_loop (loop)
718 Returns true when the given loop is, in fact, the default loop, and false
719 otherwise.
721 =item unsigned int ev_iteration (loop)
723 Returns the current iteration count for the event loop, which is identical
724 to the number of times libev did poll for new events. It starts at C<0>
725 and happily wraps around with enough iterations.
727 This value can sometimes be useful as a generation counter of sorts (it
728 "ticks" the number of loop iterations), as it roughly corresponds with
729 C<ev_prepare> and C<ev_check> calls - and is incremented between the
730 prepare and check phases.
732 =item unsigned int ev_depth (loop)
734 Returns the number of times C<ev_run> was entered minus the number of
735 times C<ev_run> was exited normally, in other words, the recursion depth.
737 Outside C<ev_run>, this number is zero. In a callback, this number is
738 C<1>, unless C<ev_run> was invoked recursively (or from another thread),
739 in which case it is higher.
741 Leaving C<ev_run> abnormally (setjmp/longjmp, cancelling the thread,
742 throwing an exception etc.), doesn't count as "exit" - consider this
743 as a hint to avoid such ungentleman-like behaviour unless it's really
744 convenient, in which case it is fully supported.
746 =item unsigned int ev_backend (loop)
748 Returns one of the C<EVBACKEND_*> flags indicating the event backend in
749 use.
751 =item ev_tstamp ev_now (loop)
753 Returns the current "event loop time", which is the time the event loop
754 received events and started processing them. This timestamp does not
755 change as long as callbacks are being processed, and this is also the base
756 time used for relative timers. You can treat it as the timestamp of the
757 event occurring (or more correctly, libev finding out about it).
759 =item ev_now_update (loop)
761 Establishes the current time by querying the kernel, updating the time
762 returned by C<ev_now ()> in the progress. This is a costly operation and
763 is usually done automatically within C<ev_run ()>.
765 This function is rarely useful, but when some event callback runs for a
766 very long time without entering the event loop, updating libev's idea of
767 the current time is a good idea.
769 See also L</The special problem of time updates> in the C<ev_timer> section.
771 =item ev_suspend (loop)
773 =item ev_resume (loop)
775 These two functions suspend and resume an event loop, for use when the
776 loop is not used for a while and timeouts should not be processed.
778 A typical use case would be an interactive program such as a game: When
779 the user presses C<^Z> to suspend the game and resumes it an hour later it
780 would be best to handle timeouts as if no time had actually passed while
781 the program was suspended. This can be achieved by calling C<ev_suspend>
782 in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
783 C<ev_resume> directly afterwards to resume timer processing.
785 Effectively, all C<ev_timer> watchers will be delayed by the time spend
786 between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
787 will be rescheduled (that is, they will lose any events that would have
788 occurred while suspended).
790 After calling C<ev_suspend> you B<must not> call I<any> function on the
791 given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
792 without a previous call to C<ev_suspend>.
794 Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
795 event loop time (see C<ev_now_update>).
797 =item bool ev_run (loop, int flags)
799 Finally, this is it, the event handler. This function usually is called
800 after you have initialised all your watchers and you want to start
801 handling events. It will ask the operating system for any new events, call
802 the watcher callbacks, and then repeat the whole process indefinitely: This
803 is why event loops are called I<loops>.
805 If the flags argument is specified as C<0>, it will keep handling events
806 until either no event watchers are active anymore or C<ev_break> was
807 called.
809 The return value is false if there are no more active watchers (which
810 usually means "all jobs done" or "deadlock"), and true in all other cases
811 (which usually means " you should call C<ev_run> again").
813 Please note that an explicit C<ev_break> is usually better than
814 relying on all watchers to be stopped when deciding when a program has
815 finished (especially in interactive programs), but having a program
816 that automatically loops as long as it has to and no longer by virtue
817 of relying on its watchers stopping correctly, that is truly a thing of
818 beauty.
820 This function is I<mostly> exception-safe - you can break out of a
821 C<ev_run> call by calling C<longjmp> in a callback, throwing a C++
822 exception and so on. This does not decrement the C<ev_depth> value, nor
823 will it clear any outstanding C<EVBREAK_ONE> breaks.
825 A flags value of C<EVRUN_NOWAIT> will look for new events, will handle
826 those events and any already outstanding ones, but will not wait and
827 block your process in case there are no events and will return after one
828 iteration of the loop. This is sometimes useful to poll and handle new
829 events while doing lengthy calculations, to keep the program responsive.
831 A flags value of C<EVRUN_ONCE> will look for new events (waiting if
832 necessary) and will handle those and any already outstanding ones. It
833 will block your process until at least one new event arrives (which could
834 be an event internal to libev itself, so there is no guarantee that a
835 user-registered callback will be called), and will return after one
836 iteration of the loop.
838 This is useful if you are waiting for some external event in conjunction
839 with something not expressible using other libev watchers (i.e. "roll your
840 own C<ev_run>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
841 usually a better approach for this kind of thing.
843 Here are the gory details of what C<ev_run> does (this is for your
844 understanding, not a guarantee that things will work exactly like this in
845 future versions):
847 - Increment loop depth.
848 - Reset the ev_break status.
849 - Before the first iteration, call any pending watchers.
850 LOOP:
851 - If EVFLAG_FORKCHECK was used, check for a fork.
852 - If a fork was detected (by any means), queue and call all fork watchers.
853 - Queue and call all prepare watchers.
854 - If ev_break was called, goto FINISH.
855 - If we have been forked, detach and recreate the kernel state
856 as to not disturb the other process.
857 - Update the kernel state with all outstanding changes.
858 - Update the "event loop time" (ev_now ()).
859 - Calculate for how long to sleep or block, if at all
860 (active idle watchers, EVRUN_NOWAIT or not having
861 any active watchers at all will result in not sleeping).
862 - Sleep if the I/O and timer collect interval say so.
863 - Increment loop iteration counter.
864 - Block the process, waiting for any events.
865 - Queue all outstanding I/O (fd) events.
866 - Update the "event loop time" (ev_now ()), and do time jump adjustments.
867 - Queue all expired timers.
868 - Queue all expired periodics.
869 - Queue all idle watchers with priority higher than that of pending events.
870 - Queue all check watchers.
871 - Call all queued watchers in reverse order (i.e. check watchers first).
872 Signals and child watchers are implemented as I/O watchers, and will
873 be handled here by queueing them when their watcher gets executed.
874 - If ev_break has been called, or EVRUN_ONCE or EVRUN_NOWAIT
875 were used, or there are no active watchers, goto FINISH, otherwise
876 continue with step LOOP.
878 - Reset the ev_break status iff it was EVBREAK_ONE.
879 - Decrement the loop depth.
880 - Return.
882 Example: Queue some jobs and then loop until no events are outstanding
883 anymore.
885 ... queue jobs here, make sure they register event watchers as long
886 ... as they still have work to do (even an idle watcher will do..)
887 ev_run (my_loop, 0);
888 ... jobs done or somebody called break. yeah!
890 =item ev_break (loop, how)
892 Can be used to make a call to C<ev_run> return early (but only after it
893 has processed all outstanding events). The C<how> argument must be either
894 C<EVBREAK_ONE>, which will make the innermost C<ev_run> call return, or
895 C<EVBREAK_ALL>, which will make all nested C<ev_run> calls return.
897 This "break state" will be cleared on the next call to C<ev_run>.
899 It is safe to call C<ev_break> from outside any C<ev_run> calls, too, in
900 which case it will have no effect.
902 =item ev_ref (loop)
904 =item ev_unref (loop)
906 Ref/unref can be used to add or remove a reference count on the event
907 loop: Every watcher keeps one reference, and as long as the reference
908 count is nonzero, C<ev_run> will not return on its own.
910 This is useful when you have a watcher that you never intend to
911 unregister, but that nevertheless should not keep C<ev_run> from
912 returning. In such a case, call C<ev_unref> after starting, and C<ev_ref>
913 before stopping it.
915 As an example, libev itself uses this for its internal signal pipe: It
916 is not visible to the libev user and should not keep C<ev_run> from
917 exiting if no event watchers registered by it are active. It is also an
918 excellent way to do this for generic recurring timers or from within
919 third-party libraries. Just remember to I<unref after start> and I<ref
920 before stop> (but only if the watcher wasn't active before, or was active
921 before, respectively. Note also that libev might stop watchers itself
922 (e.g. non-repeating timers) in which case you have to C<ev_ref>
923 in the callback).
925 Example: Create a signal watcher, but keep it from keeping C<ev_run>
926 running when nothing else is active.
928 ev_signal exitsig;
929 ev_signal_init (&exitsig, sig_cb, SIGINT);
930 ev_signal_start (loop, &exitsig);
931 ev_unref (loop);
933 Example: For some weird reason, unregister the above signal handler again.
935 ev_ref (loop);
936 ev_signal_stop (loop, &exitsig);
938 =item ev_set_io_collect_interval (loop, ev_tstamp interval)
940 =item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
942 These advanced functions influence the time that libev will spend waiting
943 for events. Both time intervals are by default C<0>, meaning that libev
944 will try to invoke timer/periodic callbacks and I/O callbacks with minimum
945 latency.
947 Setting these to a higher value (the C<interval> I<must> be >= C<0>)
948 allows libev to delay invocation of I/O and timer/periodic callbacks
949 to increase efficiency of loop iterations (or to increase power-saving
950 opportunities).
952 The idea is that sometimes your program runs just fast enough to handle
953 one (or very few) event(s) per loop iteration. While this makes the
954 program responsive, it also wastes a lot of CPU time to poll for new
955 events, especially with backends like C<select ()> which have a high
956 overhead for the actual polling but can deliver many events at once.
958 By setting a higher I<io collect interval> you allow libev to spend more
959 time collecting I/O events, so you can handle more events per iteration,
960 at the cost of increasing latency. Timeouts (both C<ev_periodic> and
961 C<ev_timer>) will not be affected. Setting this to a non-null value will
962 introduce an additional C<ev_sleep ()> call into most loop iterations. The
963 sleep time ensures that libev will not poll for I/O events more often then
964 once per this interval, on average (as long as the host time resolution is
965 good enough).
967 Likewise, by setting a higher I<timeout collect interval> you allow libev
968 to spend more time collecting timeouts, at the expense of increased
969 latency/jitter/inexactness (the watcher callback will be called
970 later). C<ev_io> watchers will not be affected. Setting this to a non-null
971 value will not introduce any overhead in libev.
973 Many (busy) programs can usually benefit by setting the I/O collect
974 interval to a value near C<0.1> or so, which is often enough for
975 interactive servers (of course not for games), likewise for timeouts. It
976 usually doesn't make much sense to set it to a lower value than C<0.01>,
977 as this approaches the timing granularity of most systems. Note that if
978 you do transactions with the outside world and you can't increase the
979 parallelity, then this setting will limit your transaction rate (if you
980 need to poll once per transaction and the I/O collect interval is 0.01,
981 then you can't do more than 100 transactions per second).
983 Setting the I<timeout collect interval> can improve the opportunity for
984 saving power, as the program will "bundle" timer callback invocations that
985 are "near" in time together, by delaying some, thus reducing the number of
986 times the process sleeps and wakes up again. Another useful technique to
987 reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
988 they fire on, say, one-second boundaries only.
990 Example: we only need 0.1s timeout granularity, and we wish not to poll
991 more often than 100 times per second:
993 ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
994 ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
996 =item ev_invoke_pending (loop)
998 This call will simply invoke all pending watchers while resetting their
999 pending state. Normally, C<ev_run> does this automatically when required,
1000 but when overriding the invoke callback this call comes handy. This
1001 function can be invoked from a watcher - this can be useful for example
1002 when you want to do some lengthy calculation and want to pass further
1003 event handling to another thread (you still have to make sure only one
1004 thread executes within C<ev_invoke_pending> or C<ev_run> of course).
1006 =item int ev_pending_count (loop)
1008 Returns the number of pending watchers - zero indicates that no watchers
1009 are pending.
1011 =item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
1013 This overrides the invoke pending functionality of the loop: Instead of
1014 invoking all pending watchers when there are any, C<ev_run> will call
1015 this callback instead. This is useful, for example, when you want to
1016 invoke the actual watchers inside another context (another thread etc.).
1018 If you want to reset the callback, use C<ev_invoke_pending> as new
1019 callback.
1021 =item ev_set_loop_release_cb (loop, void (*release)(EV_P) throw (), void (*acquire)(EV_P) throw ())
1023 Sometimes you want to share the same loop between multiple threads. This
1024 can be done relatively simply by putting mutex_lock/unlock calls around
1025 each call to a libev function.
1027 However, C<ev_run> can run an indefinite time, so it is not feasible
1028 to wait for it to return. One way around this is to wake up the event
1029 loop via C<ev_break> and C<ev_async_send>, another way is to set these
1030 I<release> and I<acquire> callbacks on the loop.
1032 When set, then C<release> will be called just before the thread is
1033 suspended waiting for new events, and C<acquire> is called just
1034 afterwards.
1036 Ideally, C<release> will just call your mutex_unlock function, and
1037 C<acquire> will just call the mutex_lock function again.
1039 While event loop modifications are allowed between invocations of
1040 C<release> and C<acquire> (that's their only purpose after all), no
1041 modifications done will affect the event loop, i.e. adding watchers will
1042 have no effect on the set of file descriptors being watched, or the time
1043 waited. Use an C<ev_async> watcher to wake up C<ev_run> when you want it
1044 to take note of any changes you made.
1046 In theory, threads executing C<ev_run> will be async-cancel safe between
1047 invocations of C<release> and C<acquire>.
1049 See also the locking example in the C<THREADS> section later in this
1050 document.
1052 =item ev_set_userdata (loop, void *data)
1054 =item void *ev_userdata (loop)
1056 Set and retrieve a single C<void *> associated with a loop. When
1057 C<ev_set_userdata> has never been called, then C<ev_userdata> returns
1058 C<0>.
1060 These two functions can be used to associate arbitrary data with a loop,
1061 and are intended solely for the C<invoke_pending_cb>, C<release> and
1062 C<acquire> callbacks described above, but of course can be (ab-)used for
1063 any other purpose as well.
1065 =item ev_verify (loop)
1067 This function only does something when C<EV_VERIFY> support has been
1068 compiled in, which is the default for non-minimal builds. It tries to go
1069 through all internal structures and checks them for validity. If anything
1070 is found to be inconsistent, it will print an error message to standard
1071 error and call C<abort ()>.
1073 This can be used to catch bugs inside libev itself: under normal
1074 circumstances, this function will never abort as of course libev keeps its
1075 data structures consistent.
1077 =back
1082 In the following description, uppercase C<TYPE> in names stands for the
1083 watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
1084 watchers and C<ev_io_start> for I/O watchers.
1086 A watcher is an opaque structure that you allocate and register to record
1087 your interest in some event. To make a concrete example, imagine you want
1088 to wait for STDIN to become readable, you would create an C<ev_io> watcher
1089 for that:
1091 static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
1092 {
1093 ev_io_stop (w);
1094 ev_break (loop, EVBREAK_ALL);
1095 }
1097 struct ev_loop *loop = ev_default_loop (0);
1099 ev_io stdin_watcher;
1101 ev_init (&stdin_watcher, my_cb);
1102 ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
1103 ev_io_start (loop, &stdin_watcher);
1105 ev_run (loop, 0);
1107 As you can see, you are responsible for allocating the memory for your
1108 watcher structures (and it is I<usually> a bad idea to do this on the
1109 stack).
1111 Each watcher has an associated watcher structure (called C<struct ev_TYPE>
1112 or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
1114 Each watcher structure must be initialised by a call to C<ev_init (watcher
1115 *, callback)>, which expects a callback to be provided. This callback is
1116 invoked each time the event occurs (or, in the case of I/O watchers, each
1117 time the event loop detects that the file descriptor given is readable
1118 and/or writable).
1120 Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
1121 macro to configure it, with arguments specific to the watcher type. There
1122 is also a macro to combine initialisation and setting in one call: C<<
1123 ev_TYPE_init (watcher *, callback, ...) >>.
1125 To make the watcher actually watch out for events, you have to start it
1126 with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
1127 *) >>), and you can stop watching for events at any time by calling the
1128 corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
1130 As long as your watcher is active (has been started but not stopped) you
1131 must not touch the values stored in it. Most specifically you must never
1132 reinitialise it or call its C<ev_TYPE_set> macro.
1134 Each and every callback receives the event loop pointer as first, the
1135 registered watcher structure as second, and a bitset of received events as
1136 third argument.
1138 The received events usually include a single bit per event type received
1139 (you can receive multiple events at the same time). The possible bit masks
1140 are:
1142 =over 4
1144 =item C<EV_READ>
1146 =item C<EV_WRITE>
1148 The file descriptor in the C<ev_io> watcher has become readable and/or
1149 writable.
1151 =item C<EV_TIMER>
1153 The C<ev_timer> watcher has timed out.
1155 =item C<EV_PERIODIC>
1157 The C<ev_periodic> watcher has timed out.
1159 =item C<EV_SIGNAL>
1161 The signal specified in the C<ev_signal> watcher has been received by a thread.
1163 =item C<EV_CHILD>
1165 The pid specified in the C<ev_child> watcher has received a status change.
1167 =item C<EV_STAT>
1169 The path specified in the C<ev_stat> watcher changed its attributes somehow.
1171 =item C<EV_IDLE>
1173 The C<ev_idle> watcher has determined that you have nothing better to do.
1175 =item C<EV_PREPARE>
1177 =item C<EV_CHECK>
1179 All C<ev_prepare> watchers are invoked just I<before> C<ev_run> starts to
1180 gather new events, and all C<ev_check> watchers are queued (not invoked)
1181 just after C<ev_run> has gathered them, but before it queues any callbacks
1182 for any received events. That means C<ev_prepare> watchers are the last
1183 watchers invoked before the event loop sleeps or polls for new events, and
1184 C<ev_check> watchers will be invoked before any other watchers of the same
1185 or lower priority within an event loop iteration.
1187 Callbacks of both watcher types can start and stop as many watchers as
1188 they want, and all of them will be taken into account (for example, a
1189 C<ev_prepare> watcher might start an idle watcher to keep C<ev_run> from
1190 blocking).
1192 =item C<EV_EMBED>
1194 The embedded event loop specified in the C<ev_embed> watcher needs attention.
1196 =item C<EV_FORK>
1198 The event loop has been resumed in the child process after fork (see
1199 C<ev_fork>).
1201 =item C<EV_CLEANUP>
1203 The event loop is about to be destroyed (see C<ev_cleanup>).
1205 =item C<EV_ASYNC>
1207 The given async watcher has been asynchronously notified (see C<ev_async>).
1209 =item C<EV_CUSTOM>
1211 Not ever sent (or otherwise used) by libev itself, but can be freely used
1212 by libev users to signal watchers (e.g. via C<ev_feed_event>).
1214 =item C<EV_ERROR>
1216 An unspecified error has occurred, the watcher has been stopped. This might
1217 happen because the watcher could not be properly started because libev
1218 ran out of memory, a file descriptor was found to be closed or any other
1219 problem. Libev considers these application bugs.
1221 You best act on it by reporting the problem and somehow coping with the
1222 watcher being stopped. Note that well-written programs should not receive
1223 an error ever, so when your watcher receives it, this usually indicates a
1224 bug in your program.
1226 Libev will usually signal a few "dummy" events together with an error, for
1227 example it might indicate that a fd is readable or writable, and if your
1228 callbacks is well-written it can just attempt the operation and cope with
1229 the error from read() or write(). This will not work in multi-threaded
1230 programs, though, as the fd could already be closed and reused for another
1231 thing, so beware.
1233 =back
1237 =over 4
1239 =item C<ev_init> (ev_TYPE *watcher, callback)
1241 This macro initialises the generic portion of a watcher. The contents
1242 of the watcher object can be arbitrary (so C<malloc> will do). Only
1243 the generic parts of the watcher are initialised, you I<need> to call
1244 the type-specific C<ev_TYPE_set> macro afterwards to initialise the
1245 type-specific parts. For each type there is also a C<ev_TYPE_init> macro
1246 which rolls both calls into one.
1248 You can reinitialise a watcher at any time as long as it has been stopped
1249 (or never started) and there are no pending events outstanding.
1251 The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
1252 int revents)>.
1254 Example: Initialise an C<ev_io> watcher in two steps.
1256 ev_io w;
1257 ev_init (&w, my_cb);
1258 ev_io_set (&w, STDIN_FILENO, EV_READ);
1260 =item C<ev_TYPE_set> (ev_TYPE *watcher, [args])
1262 This macro initialises the type-specific parts of a watcher. You need to
1263 call C<ev_init> at least once before you call this macro, but you can
1264 call C<ev_TYPE_set> any number of times. You must not, however, call this
1265 macro on a watcher that is active (it can be pending, however, which is a
1266 difference to the C<ev_init> macro).
1268 Although some watcher types do not have type-specific arguments
1269 (e.g. C<ev_prepare>) you still need to call its C<set> macro.
1271 See C<ev_init>, above, for an example.
1273 =item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
1275 This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
1276 calls into a single call. This is the most convenient method to initialise
1277 a watcher. The same limitations apply, of course.
1279 Example: Initialise and set an C<ev_io> watcher in one step.
1281 ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
1283 =item C<ev_TYPE_start> (loop, ev_TYPE *watcher)
1285 Starts (activates) the given watcher. Only active watchers will receive
1286 events. If the watcher is already active nothing will happen.
1288 Example: Start the C<ev_io> watcher that is being abused as example in this
1289 whole section.
1291 ev_io_start (EV_DEFAULT_UC, &w);
1293 =item C<ev_TYPE_stop> (loop, ev_TYPE *watcher)
1295 Stops the given watcher if active, and clears the pending status (whether
1296 the watcher was active or not).
1298 It is possible that stopped watchers are pending - for example,
1299 non-repeating timers are being stopped when they become pending - but
1300 calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
1301 pending. If you want to free or reuse the memory used by the watcher it is
1302 therefore a good idea to always call its C<ev_TYPE_stop> function.
1304 =item bool ev_is_active (ev_TYPE *watcher)
1306 Returns a true value iff the watcher is active (i.e. it has been started
1307 and not yet been stopped). As long as a watcher is active you must not modify
1308 it.
1310 =item bool ev_is_pending (ev_TYPE *watcher)
1312 Returns a true value iff the watcher is pending, (i.e. it has outstanding
1313 events but its callback has not yet been invoked). As long as a watcher
1314 is pending (but not active) you must not call an init function on it (but
1315 C<ev_TYPE_set> is safe), you must not change its priority, and you must
1316 make sure the watcher is available to libev (e.g. you cannot C<free ()>
1317 it).
1319 =item callback ev_cb (ev_TYPE *watcher)
1321 Returns the callback currently set on the watcher.
1323 =item ev_set_cb (ev_TYPE *watcher, callback)
1325 Change the callback. You can change the callback at virtually any time
1326 (modulo threads).
1328 =item ev_set_priority (ev_TYPE *watcher, int priority)
1330 =item int ev_priority (ev_TYPE *watcher)
1332 Set and query the priority of the watcher. The priority is a small
1333 integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
1334 (default: C<-2>). Pending watchers with higher priority will be invoked
1335 before watchers with lower priority, but priority will not keep watchers
1336 from being executed (except for C<ev_idle> watchers).
1338 If you need to suppress invocation when higher priority events are pending
1339 you need to look at C<ev_idle> watchers, which provide this functionality.
1341 You I<must not> change the priority of a watcher as long as it is active or
1342 pending.
1344 Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
1345 fine, as long as you do not mind that the priority value you query might
1346 or might not have been clamped to the valid range.
1348 The default priority used by watchers when no priority has been set is
1349 always C<0>, which is supposed to not be too high and not be too low :).
1351 See L</WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
1352 priorities.
1354 =item ev_invoke (loop, ev_TYPE *watcher, int revents)
1356 Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
1357 C<loop> nor C<revents> need to be valid as long as the watcher callback
1358 can deal with that fact, as both are simply passed through to the
1359 callback.
1361 =item int ev_clear_pending (loop, ev_TYPE *watcher)
1363 If the watcher is pending, this function clears its pending status and
1364 returns its C<revents> bitset (as if its callback was invoked). If the
1365 watcher isn't pending it does nothing and returns C<0>.
1367 Sometimes it can be useful to "poll" a watcher instead of waiting for its
1368 callback to be invoked, which can be accomplished with this function.
1370 =item ev_feed_event (loop, ev_TYPE *watcher, int revents)
1372 Feeds the given event set into the event loop, as if the specified event
1373 had happened for the specified watcher (which must be a pointer to an
1374 initialised but not necessarily started event watcher). Obviously you must
1375 not free the watcher as long as it has pending events.
1377 Stopping the watcher, letting libev invoke it, or calling
1378 C<ev_clear_pending> will clear the pending event, even if the watcher was
1379 not started in the first place.
1381 See also C<ev_feed_fd_event> and C<ev_feed_signal_event> for related
1382 functions that do not need a watcher.
1384 =back
1389 =head2 WATCHER STATES
1391 There are various watcher states mentioned throughout this manual -
1392 active, pending and so on. In this section these states and the rules to
1393 transition between them will be described in more detail - and while these
1394 rules might look complicated, they usually do "the right thing".
1396 =over 4
1398 =item initialised
1400 Before a watcher can be registered with the event loop it has to be
1401 initialised. This can be done with a call to C<ev_TYPE_init>, or calls to
1402 C<ev_init> followed by the watcher-specific C<ev_TYPE_set> function.
1404 In this state it is simply some block of memory that is suitable for
1405 use in an event loop. It can be moved around, freed, reused etc. at
1406 will - as long as you either keep the memory contents intact, or call
1407 C<ev_TYPE_init> again.
1409 =item started/running/active
1411 Once a watcher has been started with a call to C<ev_TYPE_start> it becomes
1412 property of the event loop, and is actively waiting for events. While in
1413 this state it cannot be accessed (except in a few documented ways), moved,
1414 freed or anything else - the only legal thing is to keep a pointer to it,
1415 and call libev functions on it that are documented to work on active watchers.
1417 =item pending
1419 If a watcher is active and libev determines that an event it is interested
1420 in has occurred (such as a timer expiring), it will become pending. It will
1421 stay in this pending state until either it is stopped or its callback is
1422 about to be invoked, so it is not normally pending inside the watcher
1423 callback.
1425 The watcher might or might not be active while it is pending (for example,
1426 an expired non-repeating timer can be pending but no longer active). If it
1427 is stopped, it can be freely accessed (e.g. by calling C<ev_TYPE_set>),
1428 but it is still property of the event loop at this time, so cannot be
1429 moved, freed or reused. And if it is active the rules described in the
1430 previous item still apply.
1432 It is also possible to feed an event on a watcher that is not active (e.g.
1433 via C<ev_feed_event>), in which case it becomes pending without being
1434 active.
1436 =item stopped
1438 A watcher can be stopped implicitly by libev (in which case it might still
1439 be pending), or explicitly by calling its C<ev_TYPE_stop> function. The
1440 latter will clear any pending state the watcher might be in, regardless
1441 of whether it was active or not, so stopping a watcher explicitly before
1442 freeing it is often a good idea.
1444 While stopped (and not pending) the watcher is essentially in the
1445 initialised state, that is, it can be reused, moved, modified in any way
1446 you wish (but when you trash the memory block, you need to C<ev_TYPE_init>
1447 it again).
1449 =back
1453 Many event loops support I<watcher priorities>, which are usually small
1454 integers that influence the ordering of event callback invocation
1455 between watchers in some way, all else being equal.
1457 In libev, Watcher priorities can be set using C<ev_set_priority>. See its
1458 description for the more technical details such as the actual priority
1459 range.
1461 There are two common ways how these these priorities are being interpreted
1462 by event loops:
1464 In the more common lock-out model, higher priorities "lock out" invocation
1465 of lower priority watchers, which means as long as higher priority
1466 watchers receive events, lower priority watchers are not being invoked.
1468 The less common only-for-ordering model uses priorities solely to order
1469 callback invocation within a single event loop iteration: Higher priority
1470 watchers are invoked before lower priority ones, but they all get invoked
1471 before polling for new events.
1473 Libev uses the second (only-for-ordering) model for all its watchers
1474 except for idle watchers (which use the lock-out model).
1476 The rationale behind this is that implementing the lock-out model for
1477 watchers is not well supported by most kernel interfaces, and most event
1478 libraries will just poll for the same events again and again as long as
1479 their callbacks have not been executed, which is very inefficient in the
1480 common case of one high-priority watcher locking out a mass of lower
1481 priority ones.
1483 Static (ordering) priorities are most useful when you have two or more
1484 watchers handling the same resource: a typical usage example is having an
1485 C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
1486 timeouts. Under load, data might be received while the program handles
1487 other jobs, but since timers normally get invoked first, the timeout
1488 handler will be executed before checking for data. In that case, giving
1489 the timer a lower priority than the I/O watcher ensures that I/O will be
1490 handled first even under adverse conditions (which is usually, but not
1491 always, what you want).
1493 Since idle watchers use the "lock-out" model, meaning that idle watchers
1494 will only be executed when no same or higher priority watchers have
1495 received events, they can be used to implement the "lock-out" model when
1496 required.
1498 For example, to emulate how many other event libraries handle priorities,
1499 you can associate an C<ev_idle> watcher to each such watcher, and in
1500 the normal watcher callback, you just start the idle watcher. The real
1501 processing is done in the idle watcher callback. This causes libev to
1502 continuously poll and process kernel event data for the watcher, but when
1503 the lock-out case is known to be rare (which in turn is rare :), this is
1504 workable.
1506 Usually, however, the lock-out model implemented that way will perform
1507 miserably under the type of load it was designed to handle. In that case,
1508 it might be preferable to stop the real watcher before starting the
1509 idle watcher, so the kernel will not have to process the event in case
1510 the actual processing will be delayed for considerable time.
1512 Here is an example of an I/O watcher that should run at a strictly lower
1513 priority than the default, and which should only process data when no
1514 other events are pending:
1516 ev_idle idle; // actual processing watcher
1517 ev_io io; // actual event watcher
1519 static void
1520 io_cb (EV_P_ ev_io *w, int revents)
1521 {
1522 // stop the I/O watcher, we received the event, but
1523 // are not yet ready to handle it.
1524 ev_io_stop (EV_A_ w);
1526 // start the idle watcher to handle the actual event.
1527 // it will not be executed as long as other watchers
1528 // with the default priority are receiving events.
1529 ev_idle_start (EV_A_ &idle);
1530 }
1532 static void
1533 idle_cb (EV_P_ ev_idle *w, int revents)
1534 {
1535 // actual processing
1536 read (STDIN_FILENO, ...);
1538 // have to start the I/O watcher again, as
1539 // we have handled the event
1540 ev_io_start (EV_P_ &io);
1541 }
1543 // initialisation
1544 ev_idle_init (&idle, idle_cb);
1545 ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
1546 ev_io_start (EV_DEFAULT_ &io);
1548 In the "real" world, it might also be beneficial to start a timer, so that
1549 low-priority connections can not be locked out forever under load. This
1550 enables your program to keep a lower latency for important connections
1551 during short periods of high load, while not completely locking out less
1552 important ones.
1555 =head1 WATCHER TYPES
1557 This section describes each watcher in detail, but will not repeat
1558 information given in the last section. Any initialisation/set macros,
1559 functions and members specific to the watcher type are explained.
1561 Members are additionally marked with either I<[read-only]>, meaning that,
1562 while the watcher is active, you can look at the member and expect some
1563 sensible content, but you must not modify it (you can modify it while the
1564 watcher is stopped to your hearts content), or I<[read-write]>, which
1565 means you can expect it to have some sensible content while the watcher
1566 is active, but you can also modify it. Modifying it may not do something
1567 sensible or take immediate effect (or do anything at all), but libev will
1568 not crash or malfunction in any way.
1571 =head2 C<ev_io> - is this file descriptor readable or writable?
1573 I/O watchers check whether a file descriptor is readable or writable
1574 in each iteration of the event loop, or, more precisely, when reading
1575 would not block the process and writing would at least be able to write
1576 some data. This behaviour is called level-triggering because you keep
1577 receiving events as long as the condition persists. Remember you can stop
1578 the watcher if you don't want to act on the event and neither want to
1579 receive future events.
1581 In general you can register as many read and/or write event watchers per
1582 fd as you want (as long as you don't confuse yourself). Setting all file
1583 descriptors to non-blocking mode is also usually a good idea (but not
1584 required if you know what you are doing).
1586 Another thing you have to watch out for is that it is quite easy to
1587 receive "spurious" readiness notifications, that is, your callback might
1588 be called with C<EV_READ> but a subsequent C<read>(2) will actually block
1589 because there is no data. It is very easy to get into this situation even
1590 with a relatively standard program structure. Thus it is best to always
1591 use non-blocking I/O: An extra C<read>(2) returning C<EAGAIN> is far
1592 preferable to a program hanging until some data arrives.
1594 If you cannot run the fd in non-blocking mode (for example you should
1595 not play around with an Xlib connection), then you have to separately
1596 re-test whether a file descriptor is really ready with a known-to-be good
1597 interface such as poll (fortunately in the case of Xlib, it already does
1598 this on its own, so its quite safe to use). Some people additionally
1599 use C<SIGALRM> and an interval timer, just to be sure you won't block
1600 indefinitely.
1602 But really, best use non-blocking mode.
1604 =head3 The special problem of disappearing file descriptors
1606 Some backends (e.g. kqueue, epoll) need to be told about closing a file
1607 descriptor (either due to calling C<close> explicitly or any other means,
1608 such as C<dup2>). The reason is that you register interest in some file
1609 descriptor, but when it goes away, the operating system will silently drop
1610 this interest. If another file descriptor with the same number then is
1611 registered with libev, there is no efficient way to see that this is, in
1612 fact, a different file descriptor.
1614 To avoid having to explicitly tell libev about such cases, libev follows
1615 the following policy: Each time C<ev_io_set> is being called, libev
1616 will assume that this is potentially a new file descriptor, otherwise
1617 it is assumed that the file descriptor stays the same. That means that
1618 you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
1619 descriptor even if the file descriptor number itself did not change.
1621 This is how one would do it normally anyway, the important point is that
1622 the libev application should not optimise around libev but should leave
1623 optimisations to libev.
1625 =head3 The special problem of dup'ed file descriptors
1627 Some backends (e.g. epoll), cannot register events for file descriptors,
1628 but only events for the underlying file descriptions. That means when you
1629 have C<dup ()>'ed file descriptors or weirder constellations, and register
1630 events for them, only one file descriptor might actually receive events.
1632 There is no workaround possible except not registering events
1633 for potentially C<dup ()>'ed file descriptors, or to resort to
1636 =head3 The special problem of files
1638 Many people try to use C<select> (or libev) on file descriptors
1639 representing files, and expect it to become ready when their program
1640 doesn't block on disk accesses (which can take a long time on their own).
1642 However, this cannot ever work in the "expected" way - you get a readiness
1643 notification as soon as the kernel knows whether and how much data is
1644 there, and in the case of open files, that's always the case, so you
1645 always get a readiness notification instantly, and your read (or possibly
1646 write) will still block on the disk I/O.
1648 Another way to view it is that in the case of sockets, pipes, character
1649 devices and so on, there is another party (the sender) that delivers data
1650 on its own, but in the case of files, there is no such thing: the disk
1651 will not send data on its own, simply because it doesn't know what you
1652 wish to read - you would first have to request some data.
1654 Since files are typically not-so-well supported by advanced notification
1655 mechanism, libev tries hard to emulate POSIX behaviour with respect
1656 to files, even though you should not use it. The reason for this is
1657 convenience: sometimes you want to watch STDIN or STDOUT, which is
1658 usually a tty, often a pipe, but also sometimes files or special devices
1659 (for example, C<epoll> on Linux works with F</dev/random> but not with
1660 F</dev/urandom>), and even though the file might better be served with
1661 asynchronous I/O instead of with non-blocking I/O, it is still useful when
1662 it "just works" instead of freezing.
1664 So avoid file descriptors pointing to files when you know it (e.g. use
1665 libeio), but use them when it is convenient, e.g. for STDIN/STDOUT, or
1666 when you rarely read from a file instead of from a socket, and want to
1667 reuse the same code path.
1669 =head3 The special problem of fork
1671 Some backends (epoll, kqueue) do not support C<fork ()> at all or exhibit
1672 useless behaviour. Libev fully supports fork, but needs to be told about
1673 it in the child if you want to continue to use it in the child.
1675 To support fork in your child processes, you have to call C<ev_loop_fork
1676 ()> after a fork in the child, enable C<EVFLAG_FORKCHECK>, or resort to
1679 =head3 The special problem of SIGPIPE
1681 While not really specific to libev, it is easy to forget about C<SIGPIPE>:
1682 when writing to a pipe whose other end has been closed, your program gets
1683 sent a SIGPIPE, which, by default, aborts your program. For most programs
1684 this is sensible behaviour, for daemons, this is usually undesirable.
1686 So when you encounter spurious, unexplained daemon exits, make sure you
1687 ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
1688 somewhere, as that would have given you a big clue).
1690 =head3 The special problem of accept()ing when you can't
1692 Many implementations of the POSIX C<accept> function (for example,
1693 found in post-2004 Linux) have the peculiar behaviour of not removing a
1694 connection from the pending queue in all error cases.
1696 For example, larger servers often run out of file descriptors (because
1697 of resource limits), causing C<accept> to fail with C<ENFILE> but not
1698 rejecting the connection, leading to libev signalling readiness on
1699 the next iteration again (the connection still exists after all), and
1700 typically causing the program to loop at 100% CPU usage.
1702 Unfortunately, the set of errors that cause this issue differs between
1703 operating systems, there is usually little the app can do to remedy the
1704 situation, and no known thread-safe method of removing the connection to
1705 cope with overload is known (to me).
1707 One of the easiest ways to handle this situation is to just ignore it
1708 - when the program encounters an overload, it will just loop until the
1709 situation is over. While this is a form of busy waiting, no OS offers an
1710 event-based way to handle this situation, so it's the best one can do.
1712 A better way to handle the situation is to log any errors other than
1713 C<EAGAIN> and C<EWOULDBLOCK>, making sure not to flood the log with such
1714 messages, and continue as usual, which at least gives the user an idea of
1715 what could be wrong ("raise the ulimit!"). For extra points one could stop
1716 the C<ev_io> watcher on the listening fd "for a while", which reduces CPU
1717 usage.
1719 If your program is single-threaded, then you could also keep a dummy file
1720 descriptor for overload situations (e.g. by opening F</dev/null>), and
1721 when you run into C<ENFILE> or C<EMFILE>, close it, run C<accept>,
1722 close that fd, and create a new dummy fd. This will gracefully refuse
1723 clients under typical overload conditions.
1725 The last way to handle it is to simply log the error and C<exit>, as
1726 is often done with C<malloc> failures, but this results in an easy
1727 opportunity for a DoS attack.
1729 =head3 Watcher-Specific Functions
1731 =over 4
1733 =item ev_io_init (ev_io *, callback, int fd, int events)
1735 =item ev_io_set (ev_io *, int fd, int events)
1737 Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
1738 receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
1739 C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
1741 =item int fd [read-only]
1743 The file descriptor being watched.
1745 =item int events [read-only]
1747 The events being watched.
1749 =back
1751 =head3 Examples
1753 Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
1754 readable, but only once. Since it is likely line-buffered, you could
1755 attempt to read a whole line in the callback.
1757 static void
1758 stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
1759 {
1760 ev_io_stop (loop, w);
1761 .. read from stdin here (or from w->fd) and handle any I/O errors
1762 }
1764 ...
1765 struct ev_loop *loop = ev_default_init (0);
1766 ev_io stdin_readable;
1767 ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
1768 ev_io_start (loop, &stdin_readable);
1769 ev_run (loop, 0);
1772 =head2 C<ev_timer> - relative and optionally repeating timeouts
1774 Timer watchers are simple relative timers that generate an event after a
1775 given time, and optionally repeating in regular intervals after that.
1777 The timers are based on real time, that is, if you register an event that
1778 times out after an hour and you reset your system clock to January last
1779 year, it will still time out after (roughly) one hour. "Roughly" because
1780 detecting time jumps is hard, and some inaccuracies are unavoidable (the
1781 monotonic clock option helps a lot here).
1783 The callback is guaranteed to be invoked only I<after> its timeout has
1784 passed (not I<at>, so on systems with very low-resolution clocks this
1785 might introduce a small delay, see "the special problem of being too
1786 early", below). If multiple timers become ready during the same loop
1787 iteration then the ones with earlier time-out values are invoked before
1788 ones of the same priority with later time-out values (but this is no
1789 longer true when a callback calls C<ev_run> recursively).
1791 =head3 Be smart about timeouts
1793 Many real-world problems involve some kind of timeout, usually for error
1794 recovery. A typical example is an HTTP request - if the other side hangs,
1795 you want to raise some error after a while.
1797 What follows are some ways to handle this problem, from obvious and
1798 inefficient to smart and efficient.
1800 In the following, a 60 second activity timeout is assumed - a timeout that
1801 gets reset to 60 seconds each time there is activity (e.g. each time some
1802 data or other life sign was received).
1804 =over 4
1806 =item 1. Use a timer and stop, reinitialise and start it on activity.
1808 This is the most obvious, but not the most simple way: In the beginning,
1809 start the watcher:
1811 ev_timer_init (timer, callback, 60., 0.);
1812 ev_timer_start (loop, timer);
1814 Then, each time there is some activity, C<ev_timer_stop> it, initialise it
1815 and start it again:
1817 ev_timer_stop (loop, timer);
1818 ev_timer_set (timer, 60., 0.);
1819 ev_timer_start (loop, timer);
1821 This is relatively simple to implement, but means that each time there is
1822 some activity, libev will first have to remove the timer from its internal
1823 data structure and then add it again. Libev tries to be fast, but it's
1824 still not a constant-time operation.
1826 =item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
1828 This is the easiest way, and involves using C<ev_timer_again> instead of
1829 C<ev_timer_start>.
1831 To implement this, configure an C<ev_timer> with a C<repeat> value
1832 of C<60> and then call C<ev_timer_again> at start and each time you
1833 successfully read or write some data. If you go into an idle state where
1834 you do not expect data to travel on the socket, you can C<ev_timer_stop>
1835 the timer, and C<ev_timer_again> will automatically restart it if need be.
1837 That means you can ignore both the C<ev_timer_start> function and the
1838 C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
1839 member and C<ev_timer_again>.
1841 At start:
1843 ev_init (timer, callback);
1844 timer->repeat = 60.;
1845 ev_timer_again (loop, timer);
1847 Each time there is some activity:
1849 ev_timer_again (loop, timer);
1851 It is even possible to change the time-out on the fly, regardless of
1852 whether the watcher is active or not:
1854 timer->repeat = 30.;
1855 ev_timer_again (loop, timer);
1857 This is slightly more efficient then stopping/starting the timer each time
1858 you want to modify its timeout value, as libev does not have to completely
1859 remove and re-insert the timer from/into its internal data structure.
1861 It is, however, even simpler than the "obvious" way to do it.
1863 =item 3. Let the timer time out, but then re-arm it as required.
1865 This method is more tricky, but usually most efficient: Most timeouts are
1866 relatively long compared to the intervals between other activity - in
1867 our example, within 60 seconds, there are usually many I/O events with
1868 associated activity resets.
1870 In this case, it would be more efficient to leave the C<ev_timer> alone,
1871 but remember the time of last activity, and check for a real timeout only
1872 within the callback:
1874 ev_tstamp timeout = 60.;
1875 ev_tstamp last_activity; // time of last activity
1876 ev_timer timer;
1878 static void
1879 callback (EV_P_ ev_timer *w, int revents)
1880 {
1881 // calculate when the timeout would happen
1882 ev_tstamp after = last_activity - ev_now (EV_A) + timeout;
1884 // if negative, it means we the timeout already occurred
1885 if (after < 0.)
1886 {
1887 // timeout occurred, take action
1888 }
1889 else
1890 {
1891 // callback was invoked, but there was some recent
1892 // activity. simply restart the timer to time out
1893 // after "after" seconds, which is the earliest time
1894 // the timeout can occur.
1895 ev_timer_set (w, after, 0.);
1896 ev_timer_start (EV_A_ w);
1897 }
1898 }
1900 To summarise the callback: first calculate in how many seconds the
1901 timeout will occur (by calculating the absolute time when it would occur,
1902 C<last_activity + timeout>, and subtracting the current time, C<ev_now
1903 (EV_A)> from that).
1905 If this value is negative, then we are already past the timeout, i.e. we
1906 timed out, and need to do whatever is needed in this case.
1908 Otherwise, we now the earliest time at which the timeout would trigger,
1909 and simply start the timer with this timeout value.
1911 In other words, each time the callback is invoked it will check whether
1912 the timeout occurred. If not, it will simply reschedule itself to check
1913 again at the earliest time it could time out. Rinse. Repeat.
1915 This scheme causes more callback invocations (about one every 60 seconds
1916 minus half the average time between activity), but virtually no calls to
1917 libev to change the timeout.
1919 To start the machinery, simply initialise the watcher and set
1920 C<last_activity> to the current time (meaning there was some activity just
1921 now), then call the callback, which will "do the right thing" and start
1922 the timer:
1924 last_activity = ev_now (EV_A);
1925 ev_init (&timer, callback);
1926 callback (EV_A_ &timer, 0);
1928 When there is some activity, simply store the current time in
1929 C<last_activity>, no libev calls at all:
1931 if (activity detected)
1932 last_activity = ev_now (EV_A);
1934 When your timeout value changes, then the timeout can be changed by simply
1935 providing a new value, stopping the timer and calling the callback, which
1936 will again do the right thing (for example, time out immediately :).
1938 timeout = new_value;
1939 ev_timer_stop (EV_A_ &timer);
1940 callback (EV_A_ &timer, 0);
1942 This technique is slightly more complex, but in most cases where the
1943 time-out is unlikely to be triggered, much more efficient.
1945 =item 4. Wee, just use a double-linked list for your timeouts.
1947 If there is not one request, but many thousands (millions...), all
1948 employing some kind of timeout with the same timeout value, then one can
1949 do even better:
1951 When starting the timeout, calculate the timeout value and put the timeout
1952 at the I<end> of the list.
1954 Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
1955 the list is expected to fire (for example, using the technique #3).
1957 When there is some activity, remove the timer from the list, recalculate
1958 the timeout, append it to the end of the list again, and make sure to
1959 update the C<ev_timer> if it was taken from the beginning of the list.
1961 This way, one can manage an unlimited number of timeouts in O(1) time for
1962 starting, stopping and updating the timers, at the expense of a major
1963 complication, and having to use a constant timeout. The constant timeout
1964 ensures that the list stays sorted.
1966 =back
1968 So which method the best?
1970 Method #2 is a simple no-brain-required solution that is adequate in most
1971 situations. Method #3 requires a bit more thinking, but handles many cases
1972 better, and isn't very complicated either. In most case, choosing either
1973 one is fine, with #3 being better in typical situations.
1975 Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
1976 rather complicated, but extremely efficient, something that really pays
1977 off after the first million or so of active timers, i.e. it's usually
1978 overkill :)
1980 =head3 The special problem of being too early
1982 If you ask a timer to call your callback after three seconds, then
1983 you expect it to be invoked after three seconds - but of course, this
1984 cannot be guaranteed to infinite precision. Less obviously, it cannot be
1985 guaranteed to any precision by libev - imagine somebody suspending the
1986 process with a STOP signal for a few hours for example.
1988 So, libev tries to invoke your callback as soon as possible I<after> the
1989 delay has occurred, but cannot guarantee this.
1991 A less obvious failure mode is calling your callback too early: many event
1992 loops compare timestamps with a "elapsed delay >= requested delay", but
1993 this can cause your callback to be invoked much earlier than you would
1994 expect.
1996 To see why, imagine a system with a clock that only offers full second
1997 resolution (think windows if you can't come up with a broken enough OS
1998 yourself). If you schedule a one-second timer at the time 500.9, then the
1999 event loop will schedule your timeout to elapse at a system time of 500
2000 (500.9 truncated to the resolution) + 1, or 501.
2002 If an event library looks at the timeout 0.1s later, it will see "501 >=
2003 501" and invoke the callback 0.1s after it was started, even though a
2004 one-second delay was requested - this is being "too early", despite best
2005 intentions.
2007 This is the reason why libev will never invoke the callback if the elapsed
2008 delay equals the requested delay, but only when the elapsed delay is
2009 larger than the requested delay. In the example above, libev would only invoke
2010 the callback at system time 502, or 1.1s after the timer was started.
2012 So, while libev cannot guarantee that your callback will be invoked
2013 exactly when requested, it I<can> and I<does> guarantee that the requested
2014 delay has actually elapsed, or in other words, it always errs on the "too
2015 late" side of things.
2017 =head3 The special problem of time updates
2019 Establishing the current time is a costly operation (it usually takes
2020 at least one system call): EV therefore updates its idea of the current
2021 time only before and after C<ev_run> collects new events, which causes a
2022 growing difference between C<ev_now ()> and C<ev_time ()> when handling
2023 lots of events in one iteration.
2025 The relative timeouts are calculated relative to the C<ev_now ()>
2026 time. This is usually the right thing as this timestamp refers to the time
2027 of the event triggering whatever timeout you are modifying/starting. If
2028 you suspect event processing to be delayed and you I<need> to base the
2029 timeout on the current time, use something like this to adjust for this:
2031 ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);
2033 If the event loop is suspended for a long time, you can also force an
2034 update of the time returned by C<ev_now ()> by calling C<ev_now_update
2035 ()>.
2037 =head3 The special problem of unsynchronised clocks
2039 Modern systems have a variety of clocks - libev itself uses the normal
2040 "wall clock" clock and, if available, the monotonic clock (to avoid time
2041 jumps).
2043 Neither of these clocks is synchronised with each other or any other clock
2044 on the system, so C<ev_time ()> might return a considerably different time
2045 than C<gettimeofday ()> or C<time ()>. On a GNU/Linux system, for example,
2046 a call to C<gettimeofday> might return a second count that is one higher
2047 than a directly following call to C<time>.
2049 The moral of this is to only compare libev-related timestamps with
2050 C<ev_time ()> and C<ev_now ()>, at least if you want better precision than
2051 a second or so.
2053 One more problem arises due to this lack of synchronisation: if libev uses
2054 the system monotonic clock and you compare timestamps from C<ev_time>
2055 or C<ev_now> from when you started your timer and when your callback is
2056 invoked, you will find that sometimes the callback is a bit "early".
2058 This is because C<ev_timer>s work in real time, not wall clock time, so
2059 libev makes sure your callback is not invoked before the delay happened,
2060 I<measured according to the real time>, not the system clock.
2062 If your timeouts are based on a physical timescale (e.g. "time out this
2063 connection after 100 seconds") then this shouldn't bother you as it is
2064 exactly the right behaviour.
2066 If you want to compare wall clock/system timestamps to your timers, then
2067 you need to use C<ev_periodic>s, as these are based on the wall clock
2068 time, where your comparisons will always generate correct results.
2070 =head3 The special problems of suspended animation
2072 When you leave the server world it is quite customary to hit machines that
2073 can suspend/hibernate - what happens to the clocks during such a suspend?
2075 Some quick tests made with a Linux 2.6.28 indicate that a suspend freezes
2076 all processes, while the clocks (C<times>, C<CLOCK_MONOTONIC>) continue
2077 to run until the system is suspended, but they will not advance while the
2078 system is suspended. That means, on resume, it will be as if the program
2079 was frozen for a few seconds, but the suspend time will not be counted
2080 towards C<ev_timer> when a monotonic clock source is used. The real time
2081 clock advanced as expected, but if it is used as sole clocksource, then a
2082 long suspend would be detected as a time jump by libev, and timers would
2083 be adjusted accordingly.
2085 I would not be surprised to see different behaviour in different between
2086 operating systems, OS versions or even different hardware.
2088 The other form of suspend (job control, or sending a SIGSTOP) will see a
2089 time jump in the monotonic clocks and the realtime clock. If the program
2090 is suspended for a very long time, and monotonic clock sources are in use,
2091 then you can expect C<ev_timer>s to expire as the full suspension time
2092 will be counted towards the timers. When no monotonic clock source is in
2093 use, then libev will again assume a timejump and adjust accordingly.
2095 It might be beneficial for this latter case to call C<ev_suspend>
2096 and C<ev_resume> in code that handles C<SIGTSTP>, to at least get
2097 deterministic behaviour in this case (you can do nothing against
2098 C<SIGSTOP>).
2100 =head3 Watcher-Specific Functions and Data Members
2102 =over 4
2104 =item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
2106 =item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
2108 Configure the timer to trigger after C<after> seconds. If C<repeat>
2109 is C<0.>, then it will automatically be stopped once the timeout is
2110 reached. If it is positive, then the timer will automatically be
2111 configured to trigger again C<repeat> seconds later, again, and again,
2112 until stopped manually.
2114 The timer itself will do a best-effort at avoiding drift, that is, if
2115 you configure a timer to trigger every 10 seconds, then it will normally
2116 trigger at exactly 10 second intervals. If, however, your program cannot
2117 keep up with the timer (because it takes longer than those 10 seconds to
2118 do stuff) the timer will not fire more than once per event loop iteration.
2120 =item ev_timer_again (loop, ev_timer *)
2122 This will act as if the timer timed out, and restarts it again if it is
2123 repeating. It basically works like calling C<ev_timer_stop>, updating the
2124 timeout to the C<repeat> value and calling C<ev_timer_start>.
2126 The exact semantics are as in the following rules, all of which will be
2127 applied to the watcher:
2129 =over 4
2131 =item If the timer is pending, the pending status is always cleared.
2133 =item If the timer is started but non-repeating, stop it (as if it timed
2134 out, without invoking it).
2136 =item If the timer is repeating, make the C<repeat> value the new timeout
2137 and start the timer, if necessary.
2139 =back
2141 This sounds a bit complicated, see L</Be smart about timeouts>, above, for a
2142 usage example.
2144 =item ev_tstamp ev_timer_remaining (loop, ev_timer *)
2146 Returns the remaining time until a timer fires. If the timer is active,
2147 then this time is relative to the current event loop time, otherwise it's
2148 the timeout value currently configured.
2150 That is, after an C<ev_timer_set (w, 5, 7)>, C<ev_timer_remaining> returns
2151 C<5>. When the timer is started and one second passes, C<ev_timer_remaining>
2152 will return C<4>. When the timer expires and is restarted, it will return
2153 roughly C<7> (likely slightly less as callback invocation takes some time,
2154 too), and so on.
2156 =item ev_tstamp repeat [read-write]
2158 The current C<repeat> value. Will be used each time the watcher times out
2159 or C<ev_timer_again> is called, and determines the next timeout (if any),
2160 which is also when any modifications are taken into account.
2162 =back
2164 =head3 Examples
2166 Example: Create a timer that fires after 60 seconds.
2168 static void
2169 one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
2170 {
2171 .. one minute over, w is actually stopped right here
2172 }
2174 ev_timer mytimer;
2175 ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
2176 ev_timer_start (loop, &mytimer);
2178 Example: Create a timeout timer that times out after 10 seconds of
2179 inactivity.
2181 static void
2182 timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
2183 {
2184 .. ten seconds without any activity
2185 }
2187 ev_timer mytimer;
2188 ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
2189 ev_timer_again (&mytimer); /* start timer */
2190 ev_run (loop, 0);
2192 // and in some piece of code that gets executed on any "activity":
2193 // reset the timeout to start ticking again at 10 seconds
2194 ev_timer_again (&mytimer);
2197 =head2 C<ev_periodic> - to cron or not to cron?
2199 Periodic watchers are also timers of a kind, but they are very versatile
2200 (and unfortunately a bit complex).
2202 Unlike C<ev_timer>, periodic watchers are not based on real time (or
2203 relative time, the physical time that passes) but on wall clock time
2204 (absolute time, the thing you can read on your calender or clock). The
2205 difference is that wall clock time can run faster or slower than real
2206 time, and time jumps are not uncommon (e.g. when you adjust your
2207 wrist-watch).
2209 You can tell a periodic watcher to trigger after some specific point
2210 in time: for example, if you tell a periodic watcher to trigger "in 10
2211 seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
2212 not a delay) and then reset your system clock to January of the previous
2213 year, then it will take a year or more to trigger the event (unlike an
2214 C<ev_timer>, which would still trigger roughly 10 seconds after starting
2215 it, as it uses a relative timeout).
2217 C<ev_periodic> watchers can also be used to implement vastly more complex
2218 timers, such as triggering an event on each "midnight, local time", or
2219 other complicated rules. This cannot be done with C<ev_timer> watchers, as
2220 those cannot react to time jumps.
2222 As with timers, the callback is guaranteed to be invoked only when the
2223 point in time where it is supposed to trigger has passed. If multiple
2224 timers become ready during the same loop iteration then the ones with
2225 earlier time-out values are invoked before ones with later time-out values
2226 (but this is no longer true when a callback calls C<ev_run> recursively).
2228 =head3 Watcher-Specific Functions and Data Members
2230 =over 4
2232 =item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
2234 =item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
2236 Lots of arguments, let's sort it out... There are basically three modes of
2237 operation, and we will explain them from simplest to most complex:
2239 =over 4
2241 =item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
2243 In this configuration the watcher triggers an event after the wall clock
2244 time C<offset> has passed. It will not repeat and will not adjust when a
2245 time jump occurs, that is, if it is to be run at January 1st 2011 then it
2246 will be stopped and invoked when the system clock reaches or surpasses
2247 this point in time.
2249 =item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
2251 In this mode the watcher will always be scheduled to time out at the next
2252 C<offset + N * interval> time (for some integer N, which can also be
2253 negative) and then repeat, regardless of any time jumps. The C<offset>
2254 argument is merely an offset into the C<interval> periods.
2256 This can be used to create timers that do not drift with respect to the
2257 system clock, for example, here is an C<ev_periodic> that triggers each
2258 hour, on the hour (with respect to UTC):
2260 ev_periodic_set (&periodic, 0., 3600., 0);
2262 This doesn't mean there will always be 3600 seconds in between triggers,
2263 but only that the callback will be called when the system time shows a
2264 full hour (UTC), or more correctly, when the system time is evenly divisible
2265 by 3600.
2267 Another way to think about it (for the mathematically inclined) is that
2268 C<ev_periodic> will try to run the callback in this mode at the next possible
2269 time where C<time = offset (mod interval)>, regardless of any time jumps.
2271 The C<interval> I<MUST> be positive, and for numerical stability, the
2272 interval value should be higher than C<1/8192> (which is around 100
2273 microseconds) and C<offset> should be higher than C<0> and should have
2274 at most a similar magnitude as the current time (say, within a factor of
2275 ten). Typical values for offset are, in fact, C<0> or something between
2276 C<0> and C<interval>, which is also the recommended range.
2278 Note also that there is an upper limit to how often a timer can fire (CPU
2279 speed for example), so if C<interval> is very small then timing stability
2280 will of course deteriorate. Libev itself tries to be exact to be about one
2281 millisecond (if the OS supports it and the machine is fast enough).
2283 =item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
2285 In this mode the values for C<interval> and C<offset> are both being
2286 ignored. Instead, each time the periodic watcher gets scheduled, the
2287 reschedule callback will be called with the watcher as first, and the
2288 current time as second argument.
2290 NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, ever,
2291 or make ANY other event loop modifications whatsoever, unless explicitly
2292 allowed by documentation here>.
2294 If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
2295 it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
2296 only event loop modification you are allowed to do).
2298 The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
2299 *w, ev_tstamp now)>, e.g.:
2301 static ev_tstamp
2302 my_rescheduler (ev_periodic *w, ev_tstamp now)
2303 {
2304 return now + 60.;
2305 }
2307 It must return the next time to trigger, based on the passed time value
2308 (that is, the lowest time value larger than to the second argument). It
2309 will usually be called just before the callback will be triggered, but
2310 might be called at other times, too.
2312 NOTE: I<< This callback must always return a time that is higher than or
2313 equal to the passed C<now> value >>.
2315 This can be used to create very complex timers, such as a timer that
2316 triggers on "next midnight, local time". To do this, you would calculate the
2317 next midnight after C<now> and return the timestamp value for this. How
2318 you do this is, again, up to you (but it is not trivial, which is the main
2319 reason I omitted it as an example).
2321 =back
2323 =item ev_periodic_again (loop, ev_periodic *)
2325 Simply stops and restarts the periodic watcher again. This is only useful
2326 when you changed some parameters or the reschedule callback would return
2327 a different time than the last time it was called (e.g. in a crond like
2328 program when the crontabs have changed).
2330 =item ev_tstamp ev_periodic_at (ev_periodic *)
2332 When active, returns the absolute time that the watcher is supposed
2333 to trigger next. This is not the same as the C<offset> argument to
2334 C<ev_periodic_set>, but indeed works even in interval and manual
2335 rescheduling modes.
2337 =item ev_tstamp offset [read-write]
2339 When repeating, this contains the offset value, otherwise this is the
2340 absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
2341 although libev might modify this value for better numerical stability).
2343 Can be modified any time, but changes only take effect when the periodic
2344 timer fires or C<ev_periodic_again> is being called.
2346 =item ev_tstamp interval [read-write]
2348 The current interval value. Can be modified any time, but changes only
2349 take effect when the periodic timer fires or C<ev_periodic_again> is being
2350 called.
2352 =item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
2354 The current reschedule callback, or C<0>, if this functionality is
2355 switched off. Can be changed any time, but changes only take effect when
2356 the periodic timer fires or C<ev_periodic_again> is being called.
2358 =back
2360 =head3 Examples
2362 Example: Call a callback every hour, or, more precisely, whenever the
2363 system time is divisible by 3600. The callback invocation times have
2364 potentially a lot of jitter, but good long-term stability.
2366 static void
2367 clock_cb (struct ev_loop *loop, ev_periodic *w, int revents)
2368 {
2369 ... its now a full hour (UTC, or TAI or whatever your clock follows)
2370 }
2372 ev_periodic hourly_tick;
2373 ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
2374 ev_periodic_start (loop, &hourly_tick);
2376 Example: The same as above, but use a reschedule callback to do it:
2378 #include <math.h>
2380 static ev_tstamp
2381 my_scheduler_cb (ev_periodic *w, ev_tstamp now)
2382 {
2383 return now + (3600. - fmod (now, 3600.));
2384 }
2386 ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
2388 Example: Call a callback every hour, starting now:
2390 ev_periodic hourly_tick;
2391 ev_periodic_init (&hourly_tick, clock_cb,
2392 fmod (ev_now (loop), 3600.), 3600., 0);
2393 ev_periodic_start (loop, &hourly_tick);
2396 =head2 C<ev_signal> - signal me when a signal gets signalled!
2398 Signal watchers will trigger an event when the process receives a specific
2399 signal one or more times. Even though signals are very asynchronous, libev
2400 will try its best to deliver signals synchronously, i.e. as part of the
2401 normal event processing, like any other event.
2403 If you want signals to be delivered truly asynchronously, just use
2404 C<sigaction> as you would do without libev and forget about sharing
2405 the signal. You can even use C<ev_async> from a signal handler to
2406 synchronously wake up an event loop.
2408 You can configure as many watchers as you like for the same signal, but
2409 only within the same loop, i.e. you can watch for C<SIGINT> in your
2410 default loop and for C<SIGIO> in another loop, but you cannot watch for
2411 C<SIGINT> in both the default loop and another loop at the same time. At
2412 the moment, C<SIGCHLD> is permanently tied to the default loop.
2414 When the first watcher gets started will libev actually register something
2415 with the kernel (thus it coexists with your own signal handlers as long as
2416 you don't register any with libev for the same signal).
2418 If possible and supported, libev will install its handlers with
2419 C<SA_RESTART> (or equivalent) behaviour enabled, so system calls should
2420 not be unduly interrupted. If you have a problem with system calls getting
2421 interrupted by signals you can block all signals in an C<ev_check> watcher
2422 and unblock them in an C<ev_prepare> watcher.
2424 =head3 The special problem of inheritance over fork/execve/pthread_create
2426 Both the signal mask (C<sigprocmask>) and the signal disposition
2427 (C<sigaction>) are unspecified after starting a signal watcher (and after
2428 stopping it again), that is, libev might or might not block the signal,
2429 and might or might not set or restore the installed signal handler (but
2432 While this does not matter for the signal disposition (libev never
2433 sets signals to C<SIG_IGN>, so handlers will be reset to C<SIG_DFL> on
2434 C<execve>), this matters for the signal mask: many programs do not expect
2435 certain signals to be blocked.
2437 This means that before calling C<exec> (from the child) you should reset
2438 the signal mask to whatever "default" you expect (all clear is a good
2439 choice usually).
2441 The simplest way to ensure that the signal mask is reset in the child is
2442 to install a fork handler with C<pthread_atfork> that resets it. That will
2443 catch fork calls done by libraries (such as the libc) as well.
2445 In current versions of libev, the signal will not be blocked indefinitely
2446 unless you use the C<signalfd> API (C<EV_SIGNALFD>). While this reduces
2447 the window of opportunity for problems, it will not go away, as libev
2448 I<has> to modify the signal mask, at least temporarily.
2450 So I can't stress this enough: I<If you do not reset your signal mask when
2451 you expect it to be empty, you have a race condition in your code>. This
2452 is not a libev-specific thing, this is true for most event libraries.
2454 =head3 The special problem of threads signal handling
2456 POSIX threads has problematic signal handling semantics, specifically,
2457 a lot of functionality (sigfd, sigwait etc.) only really works if all
2458 threads in a process block signals, which is hard to achieve.
2460 When you want to use sigwait (or mix libev signal handling with your own
2461 for the same signals), you can tackle this problem by globally blocking
2462 all signals before creating any threads (or creating them with a fully set
2463 sigprocmask) and also specifying the C<EVFLAG_NOSIGMASK> when creating
2464 loops. Then designate one thread as "signal receiver thread" which handles
2465 these signals. You can pass on any signals that libev might be interested
2466 in by calling C<ev_feed_signal>.
2468 =head3 Watcher-Specific Functions and Data Members
2470 =over 4
2472 =item ev_signal_init (ev_signal *, callback, int signum)
2474 =item ev_signal_set (ev_signal *, int signum)
2476 Configures the watcher to trigger on the given signal number (usually one
2477 of the C<SIGxxx> constants).
2479 =item int signum [read-only]
2481 The signal the watcher watches out for.
2483 =back
2485 =head3 Examples
2487 Example: Try to exit cleanly on SIGINT.
2489 static void
2490 sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
2491 {
2492 ev_break (loop, EVBREAK_ALL);
2493 }
2495 ev_signal signal_watcher;
2496 ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
2497 ev_signal_start (loop, &signal_watcher);
2500 =head2 C<ev_child> - watch out for process status changes
2502 Child watchers trigger when your process receives a SIGCHLD in response to
2503 some child status changes (most typically when a child of yours dies or
2504 exits). It is permissible to install a child watcher I<after> the child
2505 has been forked (which implies it might have already exited), as long
2506 as the event loop isn't entered (or is continued from a watcher), i.e.,
2507 forking and then immediately registering a watcher for the child is fine,
2508 but forking and registering a watcher a few event loop iterations later or
2509 in the next callback invocation is not.
2511 Only the default event loop is capable of handling signals, and therefore
2512 you can only register child watchers in the default event loop.
2514 Due to some design glitches inside libev, child watchers will always be
2515 handled at maximum priority (their priority is set to C<EV_MAXPRI> by
2516 libev)
2518 =head3 Process Interaction
2520 Libev grabs C<SIGCHLD> as soon as the default event loop is
2521 initialised. This is necessary to guarantee proper behaviour even if the
2522 first child watcher is started after the child exits. The occurrence
2523 of C<SIGCHLD> is recorded asynchronously, but child reaping is done
2524 synchronously as part of the event loop processing. Libev always reaps all
2525 children, even ones not watched.
2527 =head3 Overriding the Built-In Processing
2529 Libev offers no special support for overriding the built-in child
2530 processing, but if your application collides with libev's default child
2531 handler, you can override it easily by installing your own handler for
2532 C<SIGCHLD> after initialising the default loop, and making sure the
2533 default loop never gets destroyed. You are encouraged, however, to use an
2534 event-based approach to child reaping and thus use libev's support for
2535 that, so other libev users can use C<ev_child> watchers freely.
2537 =head3 Stopping the Child Watcher
2539 Currently, the child watcher never gets stopped, even when the
2540 child terminates, so normally one needs to stop the watcher in the
2541 callback. Future versions of libev might stop the watcher automatically
2542 when a child exit is detected (calling C<ev_child_stop> twice is not a
2543 problem).
2545 =head3 Watcher-Specific Functions and Data Members
2547 =over 4
2549 =item ev_child_init (ev_child *, callback, int pid, int trace)
2551 =item ev_child_set (ev_child *, int pid, int trace)
2553 Configures the watcher to wait for status changes of process C<pid> (or
2554 I<any> process if C<pid> is specified as C<0>). The callback can look
2555 at the C<rstatus> member of the C<ev_child> watcher structure to see
2556 the status word (use the macros from C<sys/wait.h> and see your systems
2557 C<waitpid> documentation). The C<rpid> member contains the pid of the
2558 process causing the status change. C<trace> must be either C<0> (only
2559 activate the watcher when the process terminates) or C<1> (additionally
2560 activate the watcher when the process is stopped or continued).
2562 =item int pid [read-only]
2564 The process id this watcher watches out for, or C<0>, meaning any process id.
2566 =item int rpid [read-write]
2568 The process id that detected a status change.
2570 =item int rstatus [read-write]
2572 The process exit/trace status caused by C<rpid> (see your systems
2573 C<waitpid> and C<sys/wait.h> documentation for details).
2575 =back
2577 =head3 Examples
2579 Example: C<fork()> a new process and install a child handler to wait for
2580 its completion.
2582 ev_child cw;
2584 static void
2585 child_cb (EV_P_ ev_child *w, int revents)
2586 {
2587 ev_child_stop (EV_A_ w);
2588 printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
2589 }
2591 pid_t pid = fork ();
2593 if (pid < 0)
2594 // error
2595 else if (pid == 0)
2596 {
2597 // the forked child executes here
2598 exit (1);
2599 }
2600 else
2601 {
2602 ev_child_init (&cw, child_cb, pid, 0);
2603 ev_child_start (EV_DEFAULT_ &cw);
2604 }
2607 =head2 C<ev_stat> - did the file attributes just change?
2609 This watches a file system path for attribute changes. That is, it calls
2610 C<stat> on that path in regular intervals (or when the OS says it changed)
2611 and sees if it changed compared to the last time, invoking the callback
2612 if it did. Starting the watcher C<stat>'s the file, so only changes that
2613 happen after the watcher has been started will be reported.
2615 The path does not need to exist: changing from "path exists" to "path does
2616 not exist" is a status change like any other. The condition "path does not
2617 exist" (or more correctly "path cannot be stat'ed") is signified by the
2618 C<st_nlink> field being zero (which is otherwise always forced to be at
2619 least one) and all the other fields of the stat buffer having unspecified
2620 contents.
2622 The path I<must not> end in a slash or contain special components such as
2623 C<.> or C<..>. The path I<should> be absolute: If it is relative and
2624 your working directory changes, then the behaviour is undefined.
2626 Since there is no portable change notification interface available, the
2627 portable implementation simply calls C<stat(2)> regularly on the path
2628 to see if it changed somehow. You can specify a recommended polling
2629 interval for this case. If you specify a polling interval of C<0> (highly
2630 recommended!) then a I<suitable, unspecified default> value will be used
2631 (which you can expect to be around five seconds, although this might
2632 change dynamically). Libev will also impose a minimum interval which is
2633 currently around C<0.1>, but that's usually overkill.
2635 This watcher type is not meant for massive numbers of stat watchers,
2636 as even with OS-supported change notifications, this can be
2637 resource-intensive.
2639 At the time of this writing, the only OS-specific interface implemented
2640 is the Linux inotify interface (implementing kqueue support is left as an
2641 exercise for the reader. Note, however, that the author sees no way of
2642 implementing C<ev_stat> semantics with kqueue, except as a hint).
2644 =head3 ABI Issues (Largefile Support)
2646 Libev by default (unless the user overrides this) uses the default
2647 compilation environment, which means that on systems with large file
2648 support disabled by default, you get the 32 bit version of the stat
2649 structure. When using the library from programs that change the ABI to
2650 use 64 bit file offsets the programs will fail. In that case you have to
2651 compile libev with the same flags to get binary compatibility. This is
2652 obviously the case with any flags that change the ABI, but the problem is
2653 most noticeably displayed with ev_stat and large file support.
2655 The solution for this is to lobby your distribution maker to make large
2656 file interfaces available by default (as e.g. FreeBSD does) and not
2657 optional. Libev cannot simply switch on large file support because it has
2658 to exchange stat structures with application programs compiled using the
2659 default compilation environment.
2661 =head3 Inotify and Kqueue
2663 When C<inotify (7)> support has been compiled into libev and present at
2664 runtime, it will be used to speed up change detection where possible. The
2665 inotify descriptor will be created lazily when the first C<ev_stat>
2666 watcher is being started.
2668 Inotify presence does not change the semantics of C<ev_stat> watchers
2669 except that changes might be detected earlier, and in some cases, to avoid
2670 making regular C<stat> calls. Even in the presence of inotify support
2671 there are many cases where libev has to resort to regular C<stat> polling,
2672 but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
2673 many bugs), the path exists (i.e. stat succeeds), and the path resides on
2674 a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
2675 xfs are fully working) libev usually gets away without polling.
2677 There is no support for kqueue, as apparently it cannot be used to
2678 implement this functionality, due to the requirement of having a file
2679 descriptor open on the object at all times, and detecting renames, unlinks
2680 etc. is difficult.
2682 =head3 C<stat ()> is a synchronous operation
2684 Libev doesn't normally do any kind of I/O itself, and so is not blocking
2685 the process. The exception are C<ev_stat> watchers - those call C<stat
2686 ()>, which is a synchronous operation.
2688 For local paths, this usually doesn't matter: unless the system is very
2689 busy or the intervals between stat's are large, a stat call will be fast,
2690 as the path data is usually in memory already (except when starting the
2691 watcher).
2693 For networked file systems, calling C<stat ()> can block an indefinite
2694 time due to network issues, and even under good conditions, a stat call
2695 often takes multiple milliseconds.
2697 Therefore, it is best to avoid using C<ev_stat> watchers on networked
2698 paths, although this is fully supported by libev.
2700 =head3 The special problem of stat time resolution
2702 The C<stat ()> system call only supports full-second resolution portably,
2703 and even on systems where the resolution is higher, most file systems
2704 still only support whole seconds.
2706 That means that, if the time is the only thing that changes, you can
2707 easily miss updates: on the first update, C<ev_stat> detects a change and
2708 calls your callback, which does something. When there is another update
2709 within the same second, C<ev_stat> will be unable to detect unless the
2710 stat data does change in other ways (e.g. file size).
2712 The solution to this is to delay acting on a change for slightly more
2713 than a second (or till slightly after the next full second boundary), using
2714 a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
2715 ev_timer_again (loop, w)>).
2717 The C<.02> offset is added to work around small timing inconsistencies
2718 of some operating systems (where the second counter of the current time
2719 might be be delayed. One such system is the Linux kernel, where a call to
2720 C<gettimeofday> might return a timestamp with a full second later than
2721 a subsequent C<time> call - if the equivalent of C<time ()> is used to
2722 update file times then there will be a small window where the kernel uses
2723 the previous second to update file times but libev might already execute
2724 the timer callback).
2726 =head3 Watcher-Specific Functions and Data Members
2728 =over 4
2730 =item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
2732 =item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
2734 Configures the watcher to wait for status changes of the given
2735 C<path>. The C<interval> is a hint on how quickly a change is expected to
2736 be detected and should normally be specified as C<0> to let libev choose
2737 a suitable value. The memory pointed to by C<path> must point to the same
2738 path for as long as the watcher is active.
2740 The callback will receive an C<EV_STAT> event when a change was detected,
2741 relative to the attributes at the time the watcher was started (or the
2742 last change was detected).
2744 =item ev_stat_stat (loop, ev_stat *)
2746 Updates the stat buffer immediately with new values. If you change the
2747 watched path in your callback, you could call this function to avoid
2748 detecting this change (while introducing a race condition if you are not
2749 the only one changing the path). Can also be useful simply to find out the
2750 new values.
2752 =item ev_statdata attr [read-only]
2754 The most-recently detected attributes of the file. Although the type is
2755 C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
2756 suitable for your system, but you can only rely on the POSIX-standardised
2757 members to be present. If the C<st_nlink> member is C<0>, then there was
2758 some error while C<stat>ing the file.
2760 =item ev_statdata prev [read-only]
2762 The previous attributes of the file. The callback gets invoked whenever
2763 C<prev> != C<attr>, or, more precisely, one or more of these members
2764 differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
2765 C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
2767 =item ev_tstamp interval [read-only]
2769 The specified interval.
2771 =item const char *path [read-only]
2773 The file system path that is being watched.
2775 =back
2777 =head3 Examples
2779 Example: Watch C</etc/passwd> for attribute changes.
2781 static void
2782 passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
2783 {
2784 /* /etc/passwd changed in some way */
2785 if (w->attr.st_nlink)
2786 {
2787 printf ("passwd current size %ld\n", (long)w->attr.st_size);
2788 printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
2789 printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
2790 }
2791 else
2792 /* you shalt not abuse printf for puts */
2793 puts ("wow, /etc/passwd is not there, expect problems. "
2794 "if this is windows, they already arrived\n");
2795 }
2797 ...
2798 ev_stat passwd;
2800 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
2801 ev_stat_start (loop, &passwd);
2803 Example: Like above, but additionally use a one-second delay so we do not
2804 miss updates (however, frequent updates will delay processing, too, so
2805 one might do the work both on C<ev_stat> callback invocation I<and> on
2806 C<ev_timer> callback invocation).
2808 static ev_stat passwd;
2809 static ev_timer timer;
2811 static void
2812 timer_cb (EV_P_ ev_timer *w, int revents)
2813 {
2814 ev_timer_stop (EV_A_ w);
2816 /* now it's one second after the most recent passwd change */
2817 }
2819 static void
2820 stat_cb (EV_P_ ev_stat *w, int revents)
2821 {
2822 /* reset the one-second timer */
2823 ev_timer_again (EV_A_ &timer);
2824 }
2826 ...
2827 ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
2828 ev_stat_start (loop, &passwd);
2829 ev_timer_init (&timer, timer_cb, 0., 1.02);
2832 =head2 C<ev_idle> - when you've got nothing better to do...
2834 Idle watchers trigger events when no other events of the same or higher
2835 priority are pending (prepare, check and other idle watchers do not count
2836 as receiving "events").
2838 That is, as long as your process is busy handling sockets or timeouts
2839 (or even signals, imagine) of the same or higher priority it will not be
2840 triggered. But when your process is idle (or only lower-priority watchers
2841 are pending), the idle watchers are being called once per event loop
2842 iteration - until stopped, that is, or your process receives more events
2843 and becomes busy again with higher priority stuff.
2845 The most noteworthy effect is that as long as any idle watchers are
2846 active, the process will not block when waiting for new events.
2848 Apart from keeping your process non-blocking (which is a useful
2849 effect on its own sometimes), idle watchers are a good place to do
2850 "pseudo-background processing", or delay processing stuff to after the
2851 event loop has handled all outstanding events.
2853 =head3 Abusing an C<ev_idle> watcher for its side-effect
2855 As long as there is at least one active idle watcher, libev will never
2856 sleep unnecessarily. Or in other words, it will loop as fast as possible.
2857 For this to work, the idle watcher doesn't need to be invoked at all - the
2858 lowest priority will do.
2860 This mode of operation can be useful together with an C<ev_check> watcher,
2861 to do something on each event loop iteration - for example to balance load
2862 between different connections.
2864 See L</Abusing an ev_check watcher for its side-effect> for a longer
2865 example.
2867 =head3 Watcher-Specific Functions and Data Members
2869 =over 4
2871 =item ev_idle_init (ev_idle *, callback)
2873 Initialises and configures the idle watcher - it has no parameters of any
2874 kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
2875 believe me.
2877 =back
2879 =head3 Examples
2881 Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
2882 callback, free it. Also, use no error checking, as usual.
2884 static void
2885 idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
2886 {
2887 // stop the watcher
2888 ev_idle_stop (loop, w);
2890 // now we can free it
2891 free (w);
2893 // now do something you wanted to do when the program has
2894 // no longer anything immediate to do.
2895 }
2897 ev_idle *idle_watcher = malloc (sizeof (ev_idle));
2898 ev_idle_init (idle_watcher, idle_cb);
2899 ev_idle_start (loop, idle_watcher);
2902 =head2 C<ev_prepare> and C<ev_check> - customise your event loop!
2904 Prepare and check watchers are often (but not always) used in pairs:
2905 prepare watchers get invoked before the process blocks and check watchers
2906 afterwards.
2908 You I<must not> call C<ev_run> or similar functions that enter
2909 the current event loop from either C<ev_prepare> or C<ev_check>
2910 watchers. Other loops than the current one are fine, however. The
2911 rationale behind this is that you do not need to check for recursion in
2912 those watchers, i.e. the sequence will always be C<ev_prepare>, blocking,
2913 C<ev_check> so if you have one watcher of each kind they will always be
2914 called in pairs bracketing the blocking call.
2916 Their main purpose is to integrate other event mechanisms into libev and
2917 their use is somewhat advanced. They could be used, for example, to track
2918 variable changes, implement your own watchers, integrate net-snmp or a
2919 coroutine library and lots more. They are also occasionally useful if
2920 you cache some data and want to flush it before blocking (for example,
2921 in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
2922 watcher).
2924 This is done by examining in each prepare call which file descriptors
2925 need to be watched by the other library, registering C<ev_io> watchers
2926 for them and starting an C<ev_timer> watcher for any timeouts (many
2927 libraries provide exactly this functionality). Then, in the check watcher,
2928 you check for any events that occurred (by checking the pending status
2929 of all watchers and stopping them) and call back into the library. The
2930 I/O and timer callbacks will never actually be called (but must be valid
2931 nevertheless, because you never know, you know?).
2933 As another example, the Perl Coro module uses these hooks to integrate
2934 coroutines into libev programs, by yielding to other active coroutines
2935 during each prepare and only letting the process block if no coroutines
2936 are ready to run (it's actually more complicated: it only runs coroutines
2937 with priority higher than or equal to the event loop and one coroutine
2938 of lower priority, but only once, using idle watchers to keep the event
2939 loop from blocking if lower-priority coroutines are active, thus mapping
2940 low-priority coroutines to idle/background tasks).
2942 When used for this purpose, it is recommended to give C<ev_check> watchers
2943 highest (C<EV_MAXPRI>) priority, to ensure that they are being run before
2944 any other watchers after the poll (this doesn't matter for C<ev_prepare>
2945 watchers).
2947 Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
2948 activate ("feed") events into libev. While libev fully supports this, they
2949 might get executed before other C<ev_check> watchers did their job. As
2950 C<ev_check> watchers are often used to embed other (non-libev) event
2951 loops those other event loops might be in an unusable state until their
2952 C<ev_check> watcher ran (always remind yourself to coexist peacefully with
2953 others).
2955 =head3 Abusing an C<ev_check> watcher for its side-effect
2957 C<ev_check> (and less often also C<ev_prepare>) watchers can also be
2958 useful because they are called once per event loop iteration. For
2959 example, if you want to handle a large number of connections fairly, you
2960 normally only do a bit of work for each active connection, and if there
2961 is more work to do, you wait for the next event loop iteration, so other
2962 connections have a chance of making progress.
2964 Using an C<ev_check> watcher is almost enough: it will be called on the
2965 next event loop iteration. However, that isn't as soon as possible -
2966 without external events, your C<ev_check> watcher will not be invoked.
2968 This is where C<ev_idle> watchers come in handy - all you need is a
2969 single global idle watcher that is active as long as you have one active
2970 C<ev_check> watcher. The C<ev_idle> watcher makes sure the event loop
2971 will not sleep, and the C<ev_check> watcher makes sure a callback gets
2972 invoked. Neither watcher alone can do that.
2974 =head3 Watcher-Specific Functions and Data Members
2976 =over 4
2978 =item ev_prepare_init (ev_prepare *, callback)
2980 =item ev_check_init (ev_check *, callback)
2982 Initialises and configures the prepare or check watcher - they have no
2983 parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
2984 macros, but using them is utterly, utterly, utterly and completely
2985 pointless.
2987 =back
2989 =head3 Examples
2991 There are a number of principal ways to embed other event loops or modules
2992 into libev. Here are some ideas on how to include libadns into libev
2993 (there is a Perl module named C<EV::ADNS> that does this, which you could
2994 use as a working example. Another Perl module named C<EV::Glib> embeds a
2995 Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
2996 Glib event loop).
2998 Method 1: Add IO watchers and a timeout watcher in a prepare handler,
2999 and in a check watcher, destroy them and call into libadns. What follows
3000 is pseudo-code only of course. This requires you to either use a low
3001 priority for the check watcher or use C<ev_clear_pending> explicitly, as
3002 the callbacks for the IO/timeout watchers might not have been called yet.
3004 static ev_io iow [nfd];
3005 static ev_timer tw;
3007 static void
3008 io_cb (struct ev_loop *loop, ev_io *w, int revents)
3009 {
3010 }
3012 // create io watchers for each fd and a timer before blocking
3013 static void
3014 adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
3015 {
3016 int timeout = 3600000;
3017 struct pollfd fds [nfd];
3018 // actual code will need to loop here and realloc etc.
3019 adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
3021 /* the callback is illegal, but won't be called as we stop during check */
3022 ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
3023 ev_timer_start (loop, &tw);
3025 // create one ev_io per pollfd
3026 for (int i = 0; i < nfd; ++i)
3027 {
3028 ev_io_init (iow + i, io_cb, fds [i].fd,
3029 ((fds [i].events & POLLIN ? EV_READ : 0)
3030 | (fds [i].events & POLLOUT ? EV_WRITE : 0)));
3032 fds [i].revents = 0;
3033 ev_io_start (loop, iow + i);
3034 }
3035 }
3037 // stop all watchers after blocking
3038 static void
3039 adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
3040 {
3041 ev_timer_stop (loop, &tw);
3043 for (int i = 0; i < nfd; ++i)
3044 {
3045 // set the relevant poll flags
3046 // could also call adns_processreadable etc. here
3047 struct pollfd *fd = fds + i;
3048 int revents = ev_clear_pending (iow + i);
3049 if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
3050 if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
3052 // now stop the watcher
3053 ev_io_stop (loop, iow + i);
3054 }
3056 adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
3057 }
3059 Method 2: This would be just like method 1, but you run C<adns_afterpoll>
3060 in the prepare watcher and would dispose of the check watcher.
3062 Method 3: If the module to be embedded supports explicit event
3063 notification (libadns does), you can also make use of the actual watcher
3064 callbacks, and only destroy/create the watchers in the prepare watcher.
3066 static void
3067 timer_cb (EV_P_ ev_timer *w, int revents)
3068 {
3069 adns_state ads = (adns_state)w->data;
3070 update_now (EV_A);
3072 adns_processtimeouts (ads, &tv_now);
3073 }
3075 static void
3076 io_cb (EV_P_ ev_io *w, int revents)
3077 {
3078 adns_state ads = (adns_state)w->data;
3079 update_now (EV_A);
3081 if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
3082 if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
3083 }
3085 // do not ever call adns_afterpoll
3087 Method 4: Do not use a prepare or check watcher because the module you
3088 want to embed is not flexible enough to support it. Instead, you can
3089 override their poll function. The drawback with this solution is that the
3090 main loop is now no longer controllable by EV. The C<Glib::EV> module uses
3091 this approach, effectively embedding EV as a client into the horrible
3092 libglib event loop.
3094 static gint
3095 event_poll_func (GPollFD *fds, guint nfds, gint timeout)
3096 {
3097 int got_events = 0;
3099 for (n = 0; n < nfds; ++n)
3100 // create/start io watcher that sets the relevant bits in fds[n] and increment got_events
3102 if (timeout >= 0)
3103 // create/start timer
3105 // poll
3106 ev_run (EV_A_ 0);
3108 // stop timer again
3109 if (timeout >= 0)
3110 ev_timer_stop (EV_A_ &to);
3112 // stop io watchers again - their callbacks should have set
3113 for (n = 0; n < nfds; ++n)
3114 ev_io_stop (EV_A_ iow [n]);
3116 return got_events;
3117 }
3120 =head2 C<ev_embed> - when one backend isn't enough...
3122 This is a rather advanced watcher type that lets you embed one event loop
3123 into another (currently only C<ev_io> events are supported in the embedded
3124 loop, other types of watchers might be handled in a delayed or incorrect
3125 fashion and must not be used).
3127 There are primarily two reasons you would want that: work around bugs and
3128 prioritise I/O.
3130 As an example for a bug workaround, the kqueue backend might only support
3131 sockets on some platform, so it is unusable as generic backend, but you
3132 still want to make use of it because you have many sockets and it scales
3133 so nicely. In this case, you would create a kqueue-based loop and embed
3134 it into your default loop (which might use e.g. poll). Overall operation
3135 will be a bit slower because first libev has to call C<poll> and then
3136 C<kevent>, but at least you can use both mechanisms for what they are
3137 best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
3139 As for prioritising I/O: under rare circumstances you have the case where
3140 some fds have to be watched and handled very quickly (with low latency),
3141 and even priorities and idle watchers might have too much overhead. In
3142 this case you would put all the high priority stuff in one loop and all
3143 the rest in a second one, and embed the second one in the first.
3145 As long as the watcher is active, the callback will be invoked every
3146 time there might be events pending in the embedded loop. The callback
3147 must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
3148 sweep and invoke their callbacks (the callback doesn't need to invoke the
3149 C<ev_embed_sweep> function directly, it could also start an idle watcher
3150 to give the embedded loop strictly lower priority for example).
3152 You can also set the callback to C<0>, in which case the embed watcher
3153 will automatically execute the embedded loop sweep whenever necessary.
3155 Fork detection will be handled transparently while the C<ev_embed> watcher
3156 is active, i.e., the embedded loop will automatically be forked when the
3157 embedding loop forks. In other cases, the user is responsible for calling
3158 C<ev_loop_fork> on the embedded loop.
3160 Unfortunately, not all backends are embeddable: only the ones returned by
3161 C<ev_embeddable_backends> are, which, unfortunately, does not include any
3162 portable one.
3164 So when you want to use this feature you will always have to be prepared
3165 that you cannot get an embeddable loop. The recommended way to get around
3166 this is to have a separate variables for your embeddable loop, try to
3167 create it, and if that fails, use the normal loop for everything.
3169 =head3 C<ev_embed> and fork
3171 While the C<ev_embed> watcher is running, forks in the embedding loop will
3172 automatically be applied to the embedded loop as well, so no special
3173 fork handling is required in that case. When the watcher is not running,
3174 however, it is still the task of the libev user to call C<ev_loop_fork ()>
3175 as applicable.
3177 =head3 Watcher-Specific Functions and Data Members
3179 =over 4
3181 =item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
3183 =item ev_embed_set (ev_embed *, struct ev_loop *embedded_loop)
3185 Configures the watcher to embed the given loop, which must be
3186 embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
3187 invoked automatically, otherwise it is the responsibility of the callback
3188 to invoke it (it will continue to be called until the sweep has been done,
3189 if you do not want that, you need to temporarily stop the embed watcher).
3191 =item ev_embed_sweep (loop, ev_embed *)
3193 Make a single, non-blocking sweep over the embedded loop. This works
3194 similarly to C<ev_run (embedded_loop, EVRUN_NOWAIT)>, but in the most
3195 appropriate way for embedded loops.
3197 =item struct ev_loop *other [read-only]
3199 The embedded event loop.
3201 =back
3203 =head3 Examples
3205 Example: Try to get an embeddable event loop and embed it into the default
3206 event loop. If that is not possible, use the default loop. The default
3207 loop is stored in C<loop_hi>, while the embeddable loop is stored in
3208 C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
3209 used).
3211 struct ev_loop *loop_hi = ev_default_init (0);
3212 struct ev_loop *loop_lo = 0;
3213 ev_embed embed;
3215 // see if there is a chance of getting one that works
3216 // (remember that a flags value of 0 means autodetection)
3217 loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
3218 ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
3219 : 0;
3221 // if we got one, then embed it, otherwise default to loop_hi
3222 if (loop_lo)
3223 {
3224 ev_embed_init (&embed, 0, loop_lo);
3225 ev_embed_start (loop_hi, &embed);
3226 }
3227 else
3228 loop_lo = loop_hi;
3230 Example: Check if kqueue is available but not recommended and create
3231 a kqueue backend for use with sockets (which usually work with any
3232 kqueue implementation). Store the kqueue/socket-only event loop in
3233 C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
3235 struct ev_loop *loop = ev_default_init (0);
3236 struct ev_loop *loop_socket = 0;
3237 ev_embed embed;
3239 if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
3240 if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
3241 {
3242 ev_embed_init (&embed, 0, loop_socket);
3243 ev_embed_start (loop, &embed);
3244 }
3246 if (!loop_socket)
3247 loop_socket = loop;
3249 // now use loop_socket for all sockets, and loop for everything else
3252 =head2 C<ev_fork> - the audacity to resume the event loop after a fork
3254 Fork watchers are called when a C<fork ()> was detected (usually because
3255 whoever is a good citizen cared to tell libev about it by calling
3256 C<ev_loop_fork>). The invocation is done before the event loop blocks next
3257 and before C<ev_check> watchers are being called, and only in the child
3258 after the fork. If whoever good citizen calling C<ev_default_fork> cheats
3259 and calls it in the wrong process, the fork handlers will be invoked, too,
3260 of course.
3262 =head3 The special problem of life after fork - how is it possible?
3264 Most uses of C<fork()> consist of forking, then some simple calls to set
3265 up/change the process environment, followed by a call to C<exec()>. This
3266 sequence should be handled by libev without any problems.
3268 This changes when the application actually wants to do event handling
3269 in the child, or both parent in child, in effect "continuing" after the
3270 fork.
3272 The default mode of operation (for libev, with application help to detect
3273 forks) is to duplicate all the state in the child, as would be expected
3274 when I<either> the parent I<or> the child process continues.
3276 When both processes want to continue using libev, then this is usually the
3277 wrong result. In that case, usually one process (typically the parent) is
3278 supposed to continue with all watchers in place as before, while the other
3279 process typically wants to start fresh, i.e. without any active watchers.
3281 The cleanest and most efficient way to achieve that with libev is to
3282 simply create a new event loop, which of course will be "empty", and
3283 use that for new watchers. This has the advantage of not touching more
3284 memory than necessary, and thus avoiding the copy-on-write, and the
3285 disadvantage of having to use multiple event loops (which do not support
3286 signal watchers).
3288 When this is not possible, or you want to use the default loop for
3289 other reasons, then in the process that wants to start "fresh", call
3290 C<ev_loop_destroy (EV_DEFAULT)> followed by C<ev_default_loop (...)>.
3291 Destroying the default loop will "orphan" (not stop) all registered
3292 watchers, so you have to be careful not to execute code that modifies
3293 those watchers. Note also that in that case, you have to re-register any
3294 signal watchers.
3296 =head3 Watcher-Specific Functions and Data Members
3298 =over 4
3300 =item ev_fork_init (ev_fork *, callback)
3302 Initialises and configures the fork watcher - it has no parameters of any
3303 kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
3304 really.
3306 =back
3309 =head2 C<ev_cleanup> - even the best things end
3311 Cleanup watchers are called just before the event loop is being destroyed
3312 by a call to C<ev_loop_destroy>.
3314 While there is no guarantee that the event loop gets destroyed, cleanup
3315 watchers provide a convenient method to install cleanup hooks for your
3316 program, worker threads and so on - you just to make sure to destroy the
3317 loop when you want them to be invoked.
3319 Cleanup watchers are invoked in the same way as any other watcher. Unlike
3320 all other watchers, they do not keep a reference to the event loop (which
3321 makes a lot of sense if you think about it). Like all other watchers, you
3322 can call libev functions in the callback, except C<ev_cleanup_start>.
3324 =head3 Watcher-Specific Functions and Data Members
3326 =over 4
3328 =item ev_cleanup_init (ev_cleanup *, callback)
3330 Initialises and configures the cleanup watcher - it has no parameters of
3331 any kind. There is a C<ev_cleanup_set> macro, but using it is utterly
3332 pointless, I assure you.
3334 =back
3336 Example: Register an atexit handler to destroy the default loop, so any
3337 cleanup functions are called.
3339 static void
3340 program_exits (void)
3341 {
3342 ev_loop_destroy (EV_DEFAULT_UC);
3343 }
3345 ...
3346 atexit (program_exits);
3349 =head2 C<ev_async> - how to wake up an event loop
3351 In general, you cannot use an C<ev_loop> from multiple threads or other
3352 asynchronous sources such as signal handlers (as opposed to multiple event
3353 loops - those are of course safe to use in different threads).
3355 Sometimes, however, you need to wake up an event loop you do not control,
3356 for example because it belongs to another thread. This is what C<ev_async>
3357 watchers do: as long as the C<ev_async> watcher is active, you can signal
3358 it by calling C<ev_async_send>, which is thread- and signal safe.
3360 This functionality is very similar to C<ev_signal> watchers, as signals,
3361 too, are asynchronous in nature, and signals, too, will be compressed
3362 (i.e. the number of callback invocations may be less than the number of
3363 C<ev_async_send> calls). In fact, you could use signal watchers as a kind
3364 of "global async watchers" by using a watcher on an otherwise unused
3365 signal, and C<ev_feed_signal> to signal this watcher from another thread,
3366 even without knowing which loop owns the signal.
3368 =head3 Queueing
3370 C<ev_async> does not support queueing of data in any way. The reason
3371 is that the author does not know of a simple (or any) algorithm for a
3372 multiple-writer-single-reader queue that works in all cases and doesn't
3373 need elaborate support such as pthreads or unportable memory access
3374 semantics.
3376 That means that if you want to queue data, you have to provide your own
3377 queue. But at least I can tell you how to implement locking around your
3378 queue:
3380 =over 4
3382 =item queueing from a signal handler context
3384 To implement race-free queueing, you simply add to the queue in the signal
3385 handler but you block the signal handler in the watcher callback. Here is
3386 an example that does that for some fictitious SIGUSR1 handler:
3388 static ev_async mysig;
3390 static void
3391 sigusr1_handler (void)
3392 {
3393 sometype data;
3395 // no locking etc.
3396 queue_put (data);
3397 ev_async_send (EV_DEFAULT_ &mysig);
3398 }
3400 static void
3401 mysig_cb (EV_P_ ev_async *w, int revents)
3402 {
3403 sometype data;
3404 sigset_t block, prev;
3406 sigemptyset (&block);
3407 sigaddset (&block, SIGUSR1);
3408 sigprocmask (SIG_BLOCK, &block, &prev);
3410 while (queue_get (&data))
3411 process (data);
3413 if (sigismember (&prev, SIGUSR1)
3414 sigprocmask (SIG_UNBLOCK, &block, 0);
3415 }
3417 (Note: pthreads in theory requires you to use C<pthread_setmask>
3418 instead of C<sigprocmask> when you use threads, but libev doesn't do it
3419 either...).
3421 =item queueing from a thread context
3423 The strategy for threads is different, as you cannot (easily) block
3424 threads but you can easily preempt them, so to queue safely you need to
3425 employ a traditional mutex lock, such as in this pthread example:
3427 static ev_async mysig;
3428 static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
3430 static void
3431 otherthread (void)
3432 {
3433 // only need to lock the actual queueing operation
3434 pthread_mutex_lock (&mymutex);
3435 queue_put (data);
3436 pthread_mutex_unlock (&mymutex);
3438 ev_async_send (EV_DEFAULT_ &mysig);
3439 }
3441 static void
3442 mysig_cb (EV_P_ ev_async *w, int revents)
3443 {
3444 pthread_mutex_lock (&mymutex);
3446 while (queue_get (&data))
3447 process (data);
3449 pthread_mutex_unlock (&mymutex);
3450 }
3452 =back
3455 =head3 Watcher-Specific Functions and Data Members
3457 =over 4
3459 =item ev_async_init (ev_async *, callback)
3461 Initialises and configures the async watcher - it has no parameters of any
3462 kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
3463 trust me.
3465 =item ev_async_send (loop, ev_async *)
3467 Sends/signals/activates the given C<ev_async> watcher, that is, feeds
3468 an C<EV_ASYNC> event on the watcher into the event loop, and instantly
3469 returns.
3471 Unlike C<ev_feed_event>, this call is safe to do from other threads,
3472 signal or similar contexts (see the discussion of C<EV_ATOMIC_T> in the
3473 embedding section below on what exactly this means).
3475 Note that, as with other watchers in libev, multiple events might get
3476 compressed into a single callback invocation (another way to look at
3477 this is that C<ev_async> watchers are level-triggered: they are set on
3478 C<ev_async_send>, reset when the event loop detects that).
3480 This call incurs the overhead of at most one extra system call per event
3481 loop iteration, if the event loop is blocked, and no syscall at all if
3482 the event loop (or your program) is processing events. That means that
3483 repeated calls are basically free (there is no need to avoid calls for
3484 performance reasons) and that the overhead becomes smaller (typically
3485 zero) under load.
3487 =item bool = ev_async_pending (ev_async *)
3489 Returns a non-zero value when C<ev_async_send> has been called on the
3490 watcher but the event has not yet been processed (or even noted) by the
3491 event loop.
3493 C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
3494 the loop iterates next and checks for the watcher to have become active,
3495 it will reset the flag again. C<ev_async_pending> can be used to very
3496 quickly check whether invoking the loop might be a good idea.
3498 Not that this does I<not> check whether the watcher itself is pending,
3499 only whether it has been requested to make this watcher pending: there
3500 is a time window between the event loop checking and resetting the async
3501 notification, and the callback being invoked.
3503 =back
3508 There are some other functions of possible interest. Described. Here. Now.
3510 =over 4
3512 =item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)
3514 This function combines a simple timer and an I/O watcher, calls your
3515 callback on whichever event happens first and automatically stops both
3516 watchers. This is useful if you want to wait for a single event on an fd
3517 or timeout without having to allocate/configure/start/stop/free one or
3518 more watchers yourself.
3520 If C<fd> is less than 0, then no I/O watcher will be started and the
3521 C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
3522 the given C<fd> and C<events> set will be created and started.
3524 If C<timeout> is less than 0, then no timeout watcher will be
3525 started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
3526 repeat = 0) will be started. C<0> is a valid timeout.
3528 The callback has the type C<void (*cb)(int revents, void *arg)> and is
3529 passed an C<revents> set like normal event callbacks (a combination of
3530 C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMER>) and the C<arg>
3531 value passed to C<ev_once>. Note that it is possible to receive I<both>
3532 a timeout and an io event at the same time - you probably should give io
3533 events precedence.
3535 Example: wait up to ten seconds for data to appear on STDIN_FILENO.
3537 static void stdin_ready (int revents, void *arg)
3538 {
3539 if (revents & EV_READ)
3540 /* stdin might have data for us, joy! */;
3541 else if (revents & EV_TIMER)
3542 /* doh, nothing entered */;
3543 }
3545 ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
3547 =item ev_feed_fd_event (loop, int fd, int revents)
3549 Feed an event on the given fd, as if a file descriptor backend detected
3550 the given events.
3552 =item ev_feed_signal_event (loop, int signum)
3554 Feed an event as if the given signal occurred. See also C<ev_feed_signal>,
3555 which is async-safe.
3557 =back
3562 This section explains some common idioms that are not immediately
3563 obvious. Note that examples are sprinkled over the whole manual, and this
3564 section only contains stuff that wouldn't fit anywhere else.
3568 Each watcher has, by default, a C<void *data> member that you can read
3569 or modify at any time: libev will completely ignore it. This can be used
3570 to associate arbitrary data with your watcher. If you need more data and
3571 don't want to allocate memory separately and store a pointer to it in that
3572 data member, you can also "subclass" the watcher type and provide your own
3573 data:
3575 struct my_io
3576 {
3577 ev_io io;
3578 int otherfd;
3579 void *somedata;
3580 struct whatever *mostinteresting;
3581 };
3583 ...
3584 struct my_io w;
3585 ev_io_init (&, my_cb, fd, EV_READ);
3587 And since your callback will be called with a pointer to the watcher, you
3588 can cast it back to your own type:
3590 static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
3591 {
3592 struct my_io *w = (struct my_io *)w_;
3593 ...
3594 }
3596 More interesting and less C-conformant ways of casting your callback
3597 function type instead have been omitted.
3601 Another common scenario is to use some data structure with multiple
3602 embedded watchers, in effect creating your own watcher that combines
3603 multiple libev event sources into one "super-watcher":
3605 struct my_biggy
3606 {
3607 int some_data;
3608 ev_timer t1;
3609 ev_timer t2;
3610 }
3612 In this case getting the pointer to C<my_biggy> is a bit more
3613 complicated: Either you store the address of your C<my_biggy> struct in
3614 the C<data> member of the watcher (for woozies or C++ coders), or you need
3615 to use some pointer arithmetic using C<offsetof> inside your watchers (for
3616 real programmers):
3618 #include <stddef.h>
3620 static void
3621 t1_cb (EV_P_ ev_timer *w, int revents)
3622 {
3623 struct my_biggy big = (struct my_biggy *)
3624 (((char *)w) - offsetof (struct my_biggy, t1));
3625 }
3627 static void
3628 t2_cb (EV_P_ ev_timer *w, int revents)
3629 {
3630 struct my_biggy big = (struct my_biggy *)
3631 (((char *)w) - offsetof (struct my_biggy, t2));
3632 }
3636 Often you have structures like this in event-based programs:
3638 callback ()
3639 {
3640 free (request);
3641 }
3643 request = start_new_request (..., callback);
3645 The intent is to start some "lengthy" operation. The C<request> could be
3646 used to cancel the operation, or do other things with it.
3648 It's not uncommon to have code paths in C<start_new_request> that
3649 immediately invoke the callback, for example, to report errors. Or you add
3650 some caching layer that finds that it can skip the lengthy aspects of the
3651 operation and simply invoke the callback with the result.
3653 The problem here is that this will happen I<before> C<start_new_request>
3654 has returned, so C<request> is not set.
3656 Even if you pass the request by some safer means to the callback, you
3657 might want to do something to the request after starting it, such as
3658 canceling it, which probably isn't working so well when the callback has
3659 already been invoked.
3661 A common way around all these issues is to make sure that
3662 C<start_new_request> I<always> returns before the callback is invoked. If
3663 C<start_new_request> immediately knows the result, it can artificially
3664 delay invoking the callback by using a C<prepare> or C<idle> watcher for
3665 example, or more sneakily, by reusing an existing (stopped) watcher and
3666 pushing it into the pending queue:
3668 ev_set_cb (watcher, callback);
3669 ev_feed_event (EV_A_ watcher, 0);
3671 This way, C<start_new_request> can safely return before the callback is
3672 invoked, while not delaying callback invocation too much.
3676 Often (especially in GUI toolkits) there are places where you have
3677 I<modal> interaction, which is most easily implemented by recursively
3678 invoking C<ev_run>.
3680 This brings the problem of exiting - a callback might want to finish the
3681 main C<ev_run> call, but not the nested one (e.g. user clicked "Quit", but
3682 a modal "Are you sure?" dialog is still waiting), or just the nested one
3683 and not the main one (e.g. user clocked "Ok" in a modal dialog), or some
3684 other combination: In these cases, a simple C<ev_break> will not work.
3686 The solution is to maintain "break this loop" variable for each C<ev_run>
3687 invocation, and use a loop around C<ev_run> until the condition is
3688 triggered, using C<EVRUN_ONCE>:
3690 // main loop
3691 int exit_main_loop = 0;
3693 while (!exit_main_loop)
3694 ev_run (EV_DEFAULT_ EVRUN_ONCE);
3696 // in a modal watcher
3697 int exit_nested_loop = 0;
3699 while (!exit_nested_loop)
3700 ev_run (EV_A_ EVRUN_ONCE);
3702 To exit from any of these loops, just set the corresponding exit variable:
3704 // exit modal loop
3705 exit_nested_loop = 1;
3707 // exit main program, after modal loop is finished
3708 exit_main_loop = 1;
3710 // exit both
3711 exit_main_loop = exit_nested_loop = 1;
3715 Here is a fictitious example of how to run an event loop in a different
3716 thread from where callbacks are being invoked and watchers are
3717 created/added/removed.
3719 For a real-world example, see the C<EV::Loop::Async> perl module,
3720 which uses exactly this technique (which is suited for many high-level
3721 languages).
3723 The example uses a pthread mutex to protect the loop data, a condition
3724 variable to wait for callback invocations, an async watcher to notify the
3725 event loop thread and an unspecified mechanism to wake up the main thread.
3727 First, you need to associate some data with the event loop:
3729 typedef struct {
3730 mutex_t lock; /* global loop lock */
3731 ev_async async_w;
3732 thread_t tid;
3733 cond_t invoke_cv;
3734 } userdata;
3736 void prepare_loop (EV_P)
3737 {
3738 // for simplicity, we use a static userdata struct.
3739 static userdata u;
3741 ev_async_init (&u->async_w, async_cb);
3742 ev_async_start (EV_A_ &u->async_w);
3744 pthread_mutex_init (&u->lock, 0);
3745 pthread_cond_init (&u->invoke_cv, 0);
3747 // now associate this with the loop
3748 ev_set_userdata (EV_A_ u);
3749 ev_set_invoke_pending_cb (EV_A_ l_invoke);
3750 ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
3752 // then create the thread running ev_run
3753 pthread_create (&u->tid, 0, l_run, EV_A);
3754 }
3756 The callback for the C<ev_async> watcher does nothing: the watcher is used
3757 solely to wake up the event loop so it takes notice of any new watchers
3758 that might have been added:
3760 static void
3761 async_cb (EV_P_ ev_async *w, int revents)
3762 {
3763 // just used for the side effects
3764 }
3766 The C<l_release> and C<l_acquire> callbacks simply unlock/lock the mutex
3767 protecting the loop data, respectively.
3769 static void
3770 l_release (EV_P)
3771 {
3772 userdata *u = ev_userdata (EV_A);
3773 pthread_mutex_unlock (&u->lock);
3774 }
3776 static void
3777 l_acquire (EV_P)
3778 {
3779 userdata *u = ev_userdata (EV_A);
3780 pthread_mutex_lock (&u->lock);
3781 }
3783 The event loop thread first acquires the mutex, and then jumps straight
3784 into C<ev_run>:
3786 void *
3787 l_run (void *thr_arg)
3788 {
3789 struct ev_loop *loop = (struct ev_loop *)thr_arg;
3791 l_acquire (EV_A);
3792 pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
3793 ev_run (EV_A_ 0);
3794 l_release (EV_A);
3796 return 0;
3797 }
3799 Instead of invoking all pending watchers, the C<l_invoke> callback will
3800 signal the main thread via some unspecified mechanism (signals? pipe
3801 writes? C<Async::Interrupt>?) and then waits until all pending watchers
3802 have been called (in a while loop because a) spurious wakeups are possible
3803 and b) skipping inter-thread-communication when there are no pending
3804 watchers is very beneficial):
3806 static void
3807 l_invoke (EV_P)
3808 {
3809 userdata *u = ev_userdata (EV_A);
3811 while (ev_pending_count (EV_A))
3812 {
3813 wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
3814 pthread_cond_wait (&u->invoke_cv, &u->lock);
3815 }
3816 }
3818 Now, whenever the main thread gets told to invoke pending watchers, it
3819 will grab the lock, call C<ev_invoke_pending> and then signal the loop
3820 thread to continue:
3822 static void
3823 real_invoke_pending (EV_P)
3824 {
3825 userdata *u = ev_userdata (EV_A);
3827 pthread_mutex_lock (&u->lock);
3828 ev_invoke_pending (EV_A);
3829 pthread_cond_signal (&u->invoke_cv);
3830 pthread_mutex_unlock (&u->lock);
3831 }
3833 Whenever you want to start/stop a watcher or do other modifications to an
3834 event loop, you will now have to lock:
3836 ev_timer timeout_watcher;
3837 userdata *u = ev_userdata (EV_A);
3839 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
3841 pthread_mutex_lock (&u->lock);
3842 ev_timer_start (EV_A_ &timeout_watcher);
3843 ev_async_send (EV_A_ &u->async_w);
3844 pthread_mutex_unlock (&u->lock);
3846 Note that sending the C<ev_async> watcher is required because otherwise
3847 an event loop currently blocking in the kernel will have no knowledge
3848 about the newly added timer. By waking up the loop it will pick up any new
3849 watchers in the next event loop iteration.
3853 While the overhead of a callback that e.g. schedules a thread is small, it
3854 is still an overhead. If you embed libev, and your main usage is with some
3855 kind of threads or coroutines, you might want to customise libev so that
3856 doesn't need callbacks anymore.
3858 Imagine you have coroutines that you can switch to using a function
3859 C<switch_to (coro)>, that libev runs in a coroutine called C<libev_coro>
3860 and that due to some magic, the currently active coroutine is stored in a
3861 global called C<current_coro>. Then you can build your own "wait for libev
3862 event" primitive by changing C<EV_CB_DECLARE> and C<EV_CB_INVOKE> (note
3863 the differing C<;> conventions):
3865 #define EV_CB_DECLARE(type) struct my_coro *cb;
3866 #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
3868 That means instead of having a C callback function, you store the
3869 coroutine to switch to in each watcher, and instead of having libev call
3870 your callback, you instead have it switch to that coroutine.
3872 A coroutine might now wait for an event with a function called
3873 C<wait_for_event>. (the watcher needs to be started, as always, but it doesn't
3874 matter when, or whether the watcher is active or not when this function is
3875 called):
3877 void
3878 wait_for_event (ev_watcher *w)
3879 {
3880 ev_set_cb (w, current_coro);
3881 switch_to (libev_coro);
3882 }
3884 That basically suspends the coroutine inside C<wait_for_event> and
3885 continues the libev coroutine, which, when appropriate, switches back to
3886 this or any other coroutine.
3888 You can do similar tricks if you have, say, threads with an event queue -
3889 instead of storing a coroutine, you store the queue object and instead of
3890 switching to a coroutine, you push the watcher onto the queue and notify
3891 any waiters.
3893 To embed libev, see L</EMBEDDING>, but in short, it's easiest to create two
3894 files, F<my_ev.h> and F<my_ev.c> that include the respective libev files:
3896 // my_ev.h
3897 #define EV_CB_DECLARE(type) struct my_coro *cb;
3898 #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb);
3899 #include "../libev/ev.h"
3901 // my_ev.c
3902 #define EV_H "my_ev.h"
3903 #include "../libev/ev.c"
3905 And then use F<my_ev.h> when you would normally use F<ev.h>, and compile
3906 F<my_ev.c> into your project. When properly specifying include paths, you
3907 can even use F<ev.h> as header file name directly.
3912 Libev offers a compatibility emulation layer for libevent. It cannot
3913 emulate the internals of libevent, so here are some usage hints:
3915 =over 4
3917 =item * Only the libevent-1.4.1-beta API is being emulated.
3919 This was the newest libevent version available when libev was implemented,
3920 and is still mostly unchanged in 2010.
3922 =item * Use it by including <event.h>, as usual.
3924 =item * The following members are fully supported: ev_base, ev_callback,
3925 ev_arg, ev_fd, ev_res, ev_events.
3927 =item * Avoid using ev_flags and the EVLIST_*-macros, while it is
3928 maintained by libev, it does not work exactly the same way as in libevent (consider
3929 it a private API).
3931 =item * Priorities are not currently supported. Initialising priorities
3932 will fail and all watchers will have the same priority, even though there
3933 is an ev_pri field.
3935 =item * In libevent, the last base created gets the signals, in libev, the
3936 base that registered the signal gets the signals.
3938 =item * Other members are not supported.
3940 =item * The libev emulation is I<not> ABI compatible to libevent, you need
3941 to use the libev header file and library.
3943 =back
3945 =head1 C++ SUPPORT
3947 =head2 C API
3949 The normal C API should work fine when used from C++: both ev.h and the
3950 libev sources can be compiled as C++. Therefore, code that uses the C API
3951 will work fine.
3953 Proper exception specifications might have to be added to callbacks passed
3954 to libev: exceptions may be thrown only from watcher callbacks, all
3955 other callbacks (allocator, syserr, loop acquire/release and periodic
3956 reschedule callbacks) must not throw exceptions, and might need a C<throw
3957 ()> specification. If you have code that needs to be compiled as both C
3958 and C++ you can use the C<EV_THROW> macro for this:
3960 static void
3961 fatal_error (const char *msg) EV_THROW
3962 {
3963 perror (msg);
3964 abort ();
3965 }
3967 ...
3968 ev_set_syserr_cb (fatal_error);
3970 The only API functions that can currently throw exceptions are C<ev_run>,
3971 C<ev_invoke>, C<ev_invoke_pending> and C<ev_loop_destroy> (the latter
3972 because it runs cleanup watchers).
3974 Throwing exceptions in watcher callbacks is only supported if libev itself
3975 is compiled with a C++ compiler or your C and C++ environments allow
3976 throwing exceptions through C libraries (most do).
3978 =head2 C++ API
3980 Libev comes with some simplistic wrapper classes for C++ that mainly allow
3981 you to use some convenience methods to start/stop watchers and also change
3982 the callback model to a model using method callbacks on objects.
3984 To use it,
3986 #include <ev++.h>
3988 This automatically includes F<ev.h> and puts all of its definitions (many
3989 of them macros) into the global namespace. All C++ specific things are
3990 put into the C<ev> namespace. It should support all the same embedding
3991 options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
3993 Care has been taken to keep the overhead low. The only data member the C++
3994 classes add (compared to plain C-style watchers) is the event loop pointer
3995 that the watcher is associated with (or no additional members at all if
3996 you disable C<EV_MULTIPLICITY> when embedding libev).
3998 Currently, functions, static and non-static member functions and classes
3999 with C<operator ()> can be used as callbacks. Other types should be easy
4000 to add as long as they only need one additional pointer for context. If
4001 you need support for other types of functors please contact the author
4002 (preferably after implementing it).
4004 For all this to work, your C++ compiler either has to use the same calling
4005 conventions as your C compiler (for static member functions), or you have
4006 to embed libev and compile libev itself as C++.
4008 Here is a list of things available in the C<ev> namespace:
4010 =over 4
4012 =item C<ev::READ>, C<ev::WRITE> etc.
4014 These are just enum values with the same values as the C<EV_READ> etc.
4015 macros from F<ev.h>.
4017 =item C<ev::tstamp>, C<ev::now>
4019 Aliases to the same types/functions as with the C<ev_> prefix.
4021 =item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
4023 For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
4024 the same name in the C<ev> namespace, with the exception of C<ev_signal>
4025 which is called C<ev::sig> to avoid clashes with the C<signal> macro
4026 defined by many implementations.
4028 All of those classes have these methods:
4030 =over 4
4032 =item ev::TYPE::TYPE ()
4034 =item ev::TYPE::TYPE (loop)
4036 =item ev::TYPE::~TYPE
4038 The constructor (optionally) takes an event loop to associate the watcher
4039 with. If it is omitted, it will use C<EV_DEFAULT>.
4041 The constructor calls C<ev_init> for you, which means you have to call the
4042 C<set> method before starting it.
4044 It will not set a callback, however: You have to call the templated C<set>
4045 method to set a callback before you can start the watcher.
4047 (The reason why you have to use a method is a limitation in C++ which does
4048 not allow explicit template arguments for constructors).
4050 The destructor automatically stops the watcher if it is active.
4052 =item w->set<class, &class::method> (object *)
4054 This method sets the callback method to call. The method has to have a
4055 signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
4056 first argument and the C<revents> as second. The object must be given as
4057 parameter and is stored in the C<data> member of the watcher.
4059 This method synthesizes efficient thunking code to call your method from
4060 the C callback that libev requires. If your compiler can inline your
4061 callback (i.e. it is visible to it at the place of the C<set> call and
4062 your compiler is good :), then the method will be fully inlined into the
4063 thunking function, making it as fast as a direct C callback.
4065 Example: simple class declaration and watcher initialisation
4067 struct myclass
4068 {
4069 void io_cb (ev::io &w, int revents) { }
4070 }
4072 myclass obj;
4073 ev::io iow;
4074 iow.set <myclass, &myclass::io_cb> (&obj);
4076 =item w->set (object *)
4078 This is a variation of a method callback - leaving out the method to call
4079 will default the method to C<operator ()>, which makes it possible to use
4080 functor objects without having to manually specify the C<operator ()> all
4081 the time. Incidentally, you can then also leave out the template argument
4082 list.
4084 The C<operator ()> method prototype must be C<void operator ()(watcher &w,
4085 int revents)>.
4087 See the method-C<set> above for more details.
4089 Example: use a functor object as callback.
4091 struct myfunctor
4092 {
4093 void operator() (ev::io &w, int revents)
4094 {
4095 ...
4096 }
4097 }
4099 myfunctor f;
4101 ev::io w;
4102 w.set (&f);
4104 =item w->set<function> (void *data = 0)
4106 Also sets a callback, but uses a static method or plain function as
4107 callback. The optional C<data> argument will be stored in the watcher's
4108 C<data> member and is free for you to use.
4110 The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
4112 See the method-C<set> above for more details.
4114 Example: Use a plain function as callback.
4116 static void io_cb (ev::io &w, int revents) { }
4117 iow.set <io_cb> ();
4119 =item w->set (loop)
4121 Associates a different C<struct ev_loop> with this watcher. You can only
4122 do this when the watcher is inactive (and not pending either).
4124 =item w->set ([arguments])
4126 Basically the same as C<ev_TYPE_set> (except for C<ev::embed> watchers>),
4127 with the same arguments. Either this method or a suitable start method
4128 must be called at least once. Unlike the C counterpart, an active watcher
4129 gets automatically stopped and restarted when reconfiguring it with this
4130 method.
4132 For C<ev::embed> watchers this method is called C<set_embed>, to avoid
4133 clashing with the C<set (loop)> method.
4135 =item w->start ()
4137 Starts the watcher. Note that there is no C<loop> argument, as the
4138 constructor already stores the event loop.
4140 =item w->start ([arguments])
4142 Instead of calling C<set> and C<start> methods separately, it is often
4143 convenient to wrap them in one call. Uses the same type of arguments as
4144 the configure C<set> method of the watcher.
4146 =item w->stop ()
4148 Stops the watcher if it is active. Again, no C<loop> argument.
4150 =item w->again () (C<ev::timer>, C<ev::periodic> only)
4152 For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
4153 C<ev_TYPE_again> function.
4155 =item w->sweep () (C<ev::embed> only)
4157 Invokes C<ev_embed_sweep>.
4159 =item w->update () (C<ev::stat> only)
4161 Invokes C<ev_stat_stat>.
4163 =back
4165 =back
4167 Example: Define a class with two I/O and idle watchers, start the I/O
4168 watchers in the constructor.
4170 class myclass
4171 {
4172 ev::io io ; void io_cb (ev::io &w, int revents);
4173 ev::io io2 ; void io2_cb (ev::io &w, int revents);
4174 ev::idle idle; void idle_cb (ev::idle &w, int revents);
4176 myclass (int fd)
4177 {
4178 io .set <myclass, &myclass::io_cb > (this);
4179 io2 .set <myclass, &myclass::io2_cb > (this);
4180 idle.set <myclass, &myclass::idle_cb> (this);
4182 io.set (fd, ev::WRITE); // configure the watcher
4183 io.start (); // start it whenever convenient
4185 io2.start (fd, ev::READ); // set + start in one call
4186 }
4187 };
4192 Libev does not offer other language bindings itself, but bindings for a
4193 number of languages exist in the form of third-party packages. If you know
4194 any interesting language binding in addition to the ones listed here, drop
4195 me a note.
4197 =over 4
4199 =item Perl
4201 The EV module implements the full libev API and is actually used to test
4202 libev. EV is developed together with libev. Apart from the EV core module,
4203 there are additional modules that implement libev-compatible interfaces
4204 to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
4205 C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
4206 and C<EV::Glib>).
4208 It can be found and installed via CPAN, its homepage is at
4209 L<>.
4211 =item Python
4213 Python bindings can be found at L<>. It
4214 seems to be quite complete and well-documented.
4216 =item Ruby
4218 Tony Arcieri has written a ruby extension that offers access to a subset
4219 of the libev API and adds file handle abstractions, asynchronous DNS and
4220 more on top of it. It can be found via gem servers. Its homepage is at
4221 L<>.
4223 Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
4224 makes rev work even on mingw.
4226 =item Haskell
4228 A haskell binding to libev is available at
4229 L<>.
4231 =item D
4233 Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
4234 be found at L<>.
4236 =item Ocaml
4238 Erkki Seppala has written Ocaml bindings for libev, to be found at
4239 L<>.
4241 =item Lua
4243 Brian Maher has written a partial interface to libev for lua (at the
4244 time of this writing, only C<ev_io> and C<ev_timer>), to be found at
4245 L<>.
4247 =item Javascript
4249 Node.js (L<>) uses libev as the underlying event library.
4251 =item Others
4253 There are others, and I stopped counting.
4255 =back
4258 =head1 MACRO MAGIC
4260 Libev can be compiled with a variety of options, the most fundamental
4261 of which is C<EV_MULTIPLICITY>. This option determines whether (most)
4262 functions and callbacks have an initial C<struct ev_loop *> argument.
4264 To make it easier to write programs that cope with either variant, the
4265 following macros are defined:
4267 =over 4
4269 =item C<EV_A>, C<EV_A_>
4271 This provides the loop I<argument> for functions, if one is required ("ev
4272 loop argument"). The C<EV_A> form is used when this is the sole argument,
4273 C<EV_A_> is used when other arguments are following. Example:
4275 ev_unref (EV_A);
4276 ev_timer_add (EV_A_ watcher);
4277 ev_run (EV_A_ 0);
4279 It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
4280 which is often provided by the following macro.
4282 =item C<EV_P>, C<EV_P_>
4284 This provides the loop I<parameter> for functions, if one is required ("ev
4285 loop parameter"). The C<EV_P> form is used when this is the sole parameter,
4286 C<EV_P_> is used when other parameters are following. Example:
4288 // this is how ev_unref is being declared
4289 static void ev_unref (EV_P);
4291 // this is how you can declare your typical callback
4292 static void cb (EV_P_ ev_timer *w, int revents)
4294 It declares a parameter C<loop> of type C<struct ev_loop *>, quite
4295 suitable for use with C<EV_A>.
4297 =item C<EV_DEFAULT>, C<EV_DEFAULT_>
4299 Similar to the other two macros, this gives you the value of the default
4300 loop, if multiple loops are supported ("ev loop default"). The default loop
4301 will be initialised if it isn't already initialised.
4303 For non-multiplicity builds, these macros do nothing, so you always have
4304 to initialise the loop somewhere.
4308 Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
4309 default loop has been initialised (C<UC> == unchecked). Their behaviour
4310 is undefined when the default loop has not been initialised by a previous
4311 execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
4313 It is often prudent to use C<EV_DEFAULT> when initialising the first
4314 watcher in a function but use C<EV_DEFAULT_UC> afterwards.
4316 =back
4318 Example: Declare and initialise a check watcher, utilising the above
4319 macros so it will work regardless of whether multiple loops are supported
4320 or not.
4322 static void
4323 check_cb (EV_P_ ev_timer *w, int revents)
4324 {
4325 ev_check_stop (EV_A_ w);
4326 }
4328 ev_check check;
4329 ev_check_init (&check, check_cb);
4330 ev_check_start (EV_DEFAULT_ &check);
4331 ev_run (EV_DEFAULT_ 0);
4333 =head1 EMBEDDING
4335 Libev can (and often is) directly embedded into host
4336 applications. Examples of applications that embed it include the Deliantra
4337 Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
4338 and rxvt-unicode.
4340 The goal is to enable you to just copy the necessary files into your
4341 source directory without having to change even a single line in them, so
4342 you can easily upgrade by simply copying (or having a checked-out copy of
4343 libev somewhere in your source tree).
4345 =head2 FILESETS
4347 Depending on what features you need you need to include one or more sets of files
4348 in your application.
4350 =head3 CORE EVENT LOOP
4352 To include only the libev core (all the C<ev_*> functions), with manual
4353 configuration (no autoconf):
4355 #define EV_STANDALONE 1
4356 #include "ev.c"
4358 This will automatically include F<ev.h>, too, and should be done in a
4359 single C source file only to provide the function implementations. To use
4360 it, do the same for F<ev.h> in all files wishing to use this API (best
4361 done by writing a wrapper around F<ev.h> that you can include instead and
4362 where you can put other configuration options):
4364 #define EV_STANDALONE 1
4365 #include "ev.h"
4367 Both header files and implementation files can be compiled with a C++
4368 compiler (at least, that's a stated goal, and breakage will be treated
4369 as a bug).
4371 You need the following files in your source tree, or in a directory
4372 in your include path (e.g. in libev/ when using -Ilibev):
4374 ev.h
4375 ev.c
4376 ev_vars.h
4377 ev_wrap.h
4379 ev_win32.c required on win32 platforms only
4381 ev_select.c only when select backend is enabled (which is enabled by default)
4382 ev_poll.c only when poll backend is enabled (disabled by default)
4383 ev_epoll.c only when the epoll backend is enabled (disabled by default)
4384 ev_kqueue.c only when the kqueue backend is enabled (disabled by default)
4385 ev_port.c only when the solaris port backend is enabled (disabled by default)
4387 F<ev.c> includes the backend files directly when enabled, so you only need
4388 to compile this single file.
4392 To include the libevent compatibility API, also include:
4394 #include "event.c"
4396 in the file including F<ev.c>, and:
4398 #include "event.h"
4400 in the files that want to use the libevent API. This also includes F<ev.h>.
4402 You need the following additional files for this:
4404 event.h
4405 event.c
4409 Instead of using C<EV_STANDALONE=1> and providing your configuration in
4410 whatever way you want, you can also C<m4_include([libev.m4])> in your
4411 F<> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
4412 include F<config.h> and configure itself accordingly.
4414 For this of course you need the m4 file:
4416 libev.m4
4420 Libev can be configured via a variety of preprocessor symbols you have to
4421 define before including (or compiling) any of its files. The default in
4422 the absence of autoconf is documented for every option.
4424 Symbols marked with "(h)" do not change the ABI, and can have different
4425 values when compiling libev vs. including F<ev.h>, so it is permissible
4426 to redefine them before including F<ev.h> without breaking compatibility
4427 to a compiled library. All other symbols change the ABI, which means all
4428 users of libev and the libev code itself must be compiled with compatible
4429 settings.
4431 =over 4
4433 =item EV_COMPAT3 (h)
4435 Backwards compatibility is a major concern for libev. This is why this
4436 release of libev comes with wrappers for the functions and symbols that
4437 have been renamed between libev version 3 and 4.
4439 You can disable these wrappers (to test compatibility with future
4440 versions) by defining C<EV_COMPAT3> to C<0> when compiling your
4441 sources. This has the additional advantage that you can drop the C<struct>
4442 from C<struct ev_loop> declarations, as libev will provide an C<ev_loop>
4443 typedef in that case.
4445 In some future version, the default for C<EV_COMPAT3> will become C<0>,
4446 and in some even more future version the compatibility code will be
4447 removed completely.
4449 =item EV_STANDALONE (h)
4451 Must always be C<1> if you do not use autoconf configuration, which
4452 keeps libev from including F<config.h>, and it also defines dummy
4453 implementations for some libevent functions (such as logging, which is not
4454 supported). It will also not define any of the structs usually found in
4455 F<event.h> that are not directly supported by the libev core alone.
4457 In standalone mode, libev will still try to automatically deduce the
4458 configuration, but has to be more conservative.
4460 =item EV_USE_FLOOR
4462 If defined to be C<1>, libev will use the C<floor ()> function for its
4463 periodic reschedule calculations, otherwise libev will fall back on a
4464 portable (slower) implementation. If you enable this, you usually have to
4465 link against libm or something equivalent. Enabling this when the C<floor>
4466 function is not available will fail, so the safe default is to not enable
4467 this.
4471 If defined to be C<1>, libev will try to detect the availability of the
4472 monotonic clock option at both compile time and runtime. Otherwise no
4473 use of the monotonic clock option will be attempted. If you enable this,
4474 you usually have to link against librt or something similar. Enabling it
4475 when the functionality isn't available is safe, though, although you have
4476 to make sure you link against any libraries where the C<clock_gettime>
4477 function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
4479 =item EV_USE_REALTIME
4481 If defined to be C<1>, libev will try to detect the availability of the
4482 real-time clock option at compile time (and assume its availability
4483 at runtime if successful). Otherwise no use of the real-time clock
4484 option will be attempted. This effectively replaces C<gettimeofday>
4485 by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
4486 correctness. See the note about libraries in the description of
4487 C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
4492 If defined to be C<1>, libev will try to use a direct syscall instead
4493 of calling the system-provided C<clock_gettime> function. This option
4494 exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
4495 unconditionally pulls in C<libpthread>, slowing down single-threaded
4496 programs needlessly. Using a direct syscall is slightly slower (in
4497 theory), because no optimised vdso implementation can be used, but avoids
4498 the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
4499 higher, as it simplifies linking (no need for C<-lrt>).
4503 If defined to be C<1>, libev will assume that C<nanosleep ()> is available
4504 and will use it for delays. Otherwise it will use C<select ()>.
4506 =item EV_USE_EVENTFD
4508 If defined to be C<1>, then libev will assume that C<eventfd ()> is
4509 available and will probe for kernel support at runtime. This will improve
4510 C<ev_signal> and C<ev_async> performance and reduce resource consumption.
4511 If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
4512 2.7 or newer, otherwise disabled.
4514 =item EV_USE_SELECT
4516 If undefined or defined to be C<1>, libev will compile in support for the
4517 C<select>(2) backend. No attempt at auto-detection will be done: if no
4518 other method takes over, select will be it. Otherwise the select backend
4519 will not be compiled in.
4523 If defined to C<1>, then the select backend will use the system C<fd_set>
4524 structure. This is useful if libev doesn't compile due to a missing
4525 C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
4526 on exotic systems. This usually limits the range of file descriptors to
4527 some low limit such as 1024 or might have other limitations (winsocket
4528 only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
4529 configures the maximum size of the C<fd_set>.
4533 When defined to C<1>, the select backend will assume that
4534 select/socket/connect etc. don't understand file descriptors but
4535 wants osf handles on win32 (this is the case when the select to
4536 be used is the winsock select). This means that it will call
4537 C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
4538 it is assumed that all these functions actually work on fds, even
4539 on win32. Should not be defined on non-win32 platforms.
4541 =item EV_FD_TO_WIN32_HANDLE(fd)
4543 If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
4544 file descriptors to socket handles. When not defining this symbol (the
4545 default), then libev will call C<_get_osfhandle>, which is usually
4546 correct. In some cases, programs use their own file descriptor management,
4547 in which case they can provide this function to map fds to socket handles.
4549 =item EV_WIN32_HANDLE_TO_FD(handle)
4551 If C<EV_SELECT_IS_WINSOCKET> then libev maps handles to file descriptors
4552 using the standard C<_open_osfhandle> function. For programs implementing
4553 their own fd to handle mapping, overwriting this function makes it easier
4554 to do so. This can be done by defining this macro to an appropriate value.
4556 =item EV_WIN32_CLOSE_FD(fd)
4558 If programs implement their own fd to handle mapping on win32, then this
4559 macro can be used to override the C<close> function, useful to unregister
4560 file descriptors again. Note that the replacement function has to close
4561 the underlying OS handle.
4565 If defined to be C<1>, libev will use C<WSASocket> to create its internal
4566 communication socket, which works better in some environments. Otherwise,
4567 the normal C<socket> function will be used, which works better in other
4568 environments.
4570 =item EV_USE_POLL
4572 If defined to be C<1>, libev will compile in support for the C<poll>(2)
4573 backend. Otherwise it will be enabled on non-win32 platforms. It
4574 takes precedence over select.
4576 =item EV_USE_EPOLL
4578 If defined to be C<1>, libev will compile in support for the Linux
4579 C<epoll>(7) backend. Its availability will be detected at runtime,
4580 otherwise another method will be used as fallback. This is the preferred
4581 backend for GNU/Linux systems. If undefined, it will be enabled if the
4582 headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
4584 =item EV_USE_KQUEUE
4586 If defined to be C<1>, libev will compile in support for the BSD style
4587 C<kqueue>(2) backend. Its actual availability will be detected at runtime,
4588 otherwise another method will be used as fallback. This is the preferred
4589 backend for BSD and BSD-like systems, although on most BSDs kqueue only
4590 supports some types of fds correctly (the only platform we found that
4591 supports ptys for example was NetBSD), so kqueue might be compiled in, but
4592 not be used unless explicitly requested. The best way to use it is to find
4593 out whether kqueue supports your type of fd properly and use an embedded
4594 kqueue loop.
4596 =item EV_USE_PORT
4598 If defined to be C<1>, libev will compile in support for the Solaris
4599 10 port style backend. Its availability will be detected at runtime,
4600 otherwise another method will be used as fallback. This is the preferred
4601 backend for Solaris 10 systems.
4603 =item EV_USE_DEVPOLL
4605 Reserved for future expansion, works like the USE symbols above.
4607 =item EV_USE_INOTIFY
4609 If defined to be C<1>, libev will compile in support for the Linux inotify
4610 interface to speed up C<ev_stat> watchers. Its actual availability will
4611 be detected at runtime. If undefined, it will be enabled if the headers
4612 indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
4614 =item EV_NO_SMP
4616 If defined to be C<1>, libev will assume that memory is always coherent
4617 between threads, that is, threads can be used, but threads never run on
4618 different cpus (or different cpu cores). This reduces dependencies
4619 and makes libev faster.
4621 =item EV_NO_THREADS
4623 If defined to be C<1>, libev will assume that it will never be called from
4624 different threads (that includes signal handlers), which is a stronger
4625 assumption than C<EV_NO_SMP>, above. This reduces dependencies and makes
4626 libev faster.
4628 =item EV_ATOMIC_T
4630 Libev requires an integer type (suitable for storing C<0> or C<1>) whose
4631 access is atomic with respect to other threads or signal contexts. No
4632 such type is easily found in the C language, so you can provide your own
4633 type that you know is safe for your purposes. It is used both for signal
4634 handler "locking" as well as for signal and thread safety in C<ev_async>
4635 watchers.
4637 In the absence of this define, libev will use C<sig_atomic_t volatile>
4638 (from F<signal.h>), which is usually good enough on most platforms.
4640 =item EV_H (h)
4642 The name of the F<ev.h> header file used to include it. The default if
4643 undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
4644 used to virtually rename the F<ev.h> header file in case of conflicts.
4646 =item EV_CONFIG_H (h)
4648 If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
4649 F<ev.c>'s idea of where to find the F<config.h> file, similarly to
4650 C<EV_H>, above.
4652 =item EV_EVENT_H (h)
4654 Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
4655 of how the F<event.h> header can be found, the default is C<"event.h">.
4657 =item EV_PROTOTYPES (h)
4659 If defined to be C<0>, then F<ev.h> will not define any function
4660 prototypes, but still define all the structs and other symbols. This is
4661 occasionally useful if you want to provide your own wrapper functions
4662 around libev functions.
4666 If undefined or defined to C<1>, then all event-loop-specific functions
4667 will have the C<struct ev_loop *> as first argument, and you can create
4668 additional independent event loops. Otherwise there will be no support
4669 for multiple event loops and there is no first event loop pointer
4670 argument. Instead, all functions act on the single default loop.
4672 Note that C<EV_DEFAULT> and C<EV_DEFAULT_> will no longer provide a
4673 default loop when multiplicity is switched off - you always have to
4674 initialise the loop manually in this case.
4676 =item EV_MINPRI
4678 =item EV_MAXPRI
4680 The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
4681 C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
4682 provide for more priorities by overriding those symbols (usually defined
4683 to be C<-2> and C<2>, respectively).
4685 When doing priority-based operations, libev usually has to linearly search
4686 all the priorities, so having many of them (hundreds) uses a lot of space
4687 and time, so using the defaults of five priorities (-2 .. +2) is usually
4688 fine.
4690 If your embedding application does not need any priorities, defining these
4691 both to C<0> will save some memory and CPU.
4697 If undefined or defined to be C<1> (and the platform supports it), then
4698 the respective watcher type is supported. If defined to be C<0>, then it
4699 is not. Disabling watcher types mainly saves code size.
4701 =item EV_FEATURES
4703 If you need to shave off some kilobytes of code at the expense of some
4704 speed (but with the full API), you can define this symbol to request
4705 certain subsets of functionality. The default is to enable all features
4706 that can be enabled on the platform.
4708 A typical way to use this symbol is to define it to C<0> (or to a bitset
4709 with some broad features you want) and then selectively re-enable
4710 additional parts you want, for example if you want everything minimal,
4711 but multiple event loop support, async and child watchers and the poll
4712 backend, use this:
4714 #define EV_FEATURES 0
4715 #define EV_MULTIPLICITY 1
4716 #define EV_USE_POLL 1
4717 #define EV_CHILD_ENABLE 1
4718 #define EV_ASYNC_ENABLE 1
4720 The actual value is a bitset, it can be a combination of the following
4721 values (by default, all of these are enabled):
4723 =over 4
4725 =item C<1> - faster/larger code
4727 Use larger code to speed up some operations.
4729 Currently this is used to override some inlining decisions (enlarging the
4730 code size by roughly 30% on amd64).
4732 When optimising for size, use of compiler flags such as C<-Os> with
4733 gcc is recommended, as well as C<-DNDEBUG>, as libev contains a number of
4734 assertions.
4736 The default is off when C<__OPTIMIZE_SIZE__> is defined by your compiler
4737 (e.g. gcc with C<-Os>).
4739 =item C<2> - faster/larger data structures
4741 Replaces the small 2-heap for timer management by a faster 4-heap, larger
4742 hash table sizes and so on. This will usually further increase code size
4743 and can additionally have an effect on the size of data structures at
4744 runtime.
4746 The default is off when C<__OPTIMIZE_SIZE__> is defined by your compiler
4747 (e.g. gcc with C<-Os>).
4749 =item C<4> - full API configuration
4751 This enables priorities (sets C<EV_MAXPRI>=2 and C<EV_MINPRI>=-2), and
4752 enables multiplicity (C<EV_MULTIPLICITY>=1).
4754 =item C<8> - full API
4756 This enables a lot of the "lesser used" API functions. See C<ev.h> for
4757 details on which parts of the API are still available without this
4758 feature, and do not complain if this subset changes over time.
4760 =item C<16> - enable all optional watcher types
4762 Enables all optional watcher types. If you want to selectively enable
4763 only some watcher types other than I/O and timers (e.g. prepare,
4764 embed, async, child...) you can enable them manually by defining
4765 C<EV_watchertype_ENABLE> to C<1> instead.
4767 =item C<32> - enable all backends
4769 This enables all backends - without this feature, you need to enable at
4770 least one backend manually (C<EV_USE_SELECT> is a good choice).
4772 =item C<64> - enable OS-specific "helper" APIs
4774 Enable inotify, eventfd, signalfd and similar OS-specific helper APIs by
4775 default.
4777 =back
4779 Compiling with C<gcc -Os -DEV_STANDALONE -DEV_USE_EPOLL=1 -DEV_FEATURES=0>
4780 reduces the compiled size of libev from 24.7Kb code/2.8Kb data to 6.5Kb
4781 code/0.3Kb data on my GNU/Linux amd64 system, while still giving you I/O
4782 watchers, timers and monotonic clock support.
4784 With an intelligent-enough linker (gcc+binutils are intelligent enough
4785 when you use C<-Wl,--gc-sections -ffunction-sections>) functions unused by
4786 your program might be left out as well - a binary starting a timer and an
4787 I/O watcher then might come out at only 5Kb.
4789 =item EV_API_STATIC
4791 If this symbol is defined (by default it is not), then all identifiers
4792 will have static linkage. This means that libev will not export any
4793 identifiers, and you cannot link against libev anymore. This can be useful
4794 when you embed libev, only want to use libev functions in a single file,
4795 and do not want its identifiers to be visible.
4797 To use this, define C<EV_API_STATIC> and include F<ev.c> in the file that
4798 wants to use libev.
4800 This option only works when libev is compiled with a C compiler, as C++
4801 doesn't support the required declaration syntax.
4803 =item EV_AVOID_STDIO
4805 If this is set to C<1> at compiletime, then libev will avoid using stdio
4806 functions (printf, scanf, perror etc.). This will increase the code size
4807 somewhat, but if your program doesn't otherwise depend on stdio and your
4808 libc allows it, this avoids linking in the stdio library which is quite
4809 big.
4811 Note that error messages might become less precise when this option is
4812 enabled.
4814 =item EV_NSIG
4816 The highest supported signal number, +1 (or, the number of
4817 signals): Normally, libev tries to deduce the maximum number of signals
4818 automatically, but sometimes this fails, in which case it can be
4819 specified. Also, using a lower number than detected (C<32> should be
4820 good for about any system in existence) can save some memory, as libev
4821 statically allocates some 12-24 bytes per signal number.
4823 =item EV_PID_HASHSIZE
4825 C<ev_child> watchers use a small hash table to distribute workload by
4826 pid. The default size is C<16> (or C<1> with C<EV_FEATURES> disabled),
4827 usually more than enough. If you need to manage thousands of children you
4828 might want to increase this value (I<must> be a power of two).
4832 C<ev_stat> watchers use a small hash table to distribute workload by
4833 inotify watch id. The default size is C<16> (or C<1> with C<EV_FEATURES>
4834 disabled), usually more than enough. If you need to manage thousands of
4835 C<ev_stat> watchers you might want to increase this value (I<must> be a
4836 power of two).
4838 =item EV_USE_4HEAP
4840 Heaps are not very cache-efficient. To improve the cache-efficiency of the
4841 timer and periodics heaps, libev uses a 4-heap when this symbol is defined
4842 to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
4843 faster performance with many (thousands) of watchers.
4845 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4846 will be C<0>.
4848 =item EV_HEAP_CACHE_AT
4850 Heaps are not very cache-efficient. To improve the cache-efficiency of the
4851 timer and periodics heaps, libev can cache the timestamp (I<at>) within
4852 the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
4853 which uses 8-12 bytes more per watcher and a few hundred bytes more code,
4854 but avoids random read accesses on heap changes. This improves performance
4855 noticeably with many (hundreds) of watchers.
4857 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4858 will be C<0>.
4860 =item EV_VERIFY
4862 Controls how much internal verification (see C<ev_verify ()>) will
4863 be done: If set to C<0>, no internal verification code will be compiled
4864 in. If set to C<1>, then verification code will be compiled in, but not
4865 called. If set to C<2>, then the internal verification code will be
4866 called once per loop, which can slow down libev. If set to C<3>, then the
4867 verification code will be called very frequently, which will slow down
4868 libev considerably.
4870 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4871 will be C<0>.
4873 =item EV_COMMON
4875 By default, all watchers have a C<void *data> member. By redefining
4876 this macro to something else you can include more and other types of
4877 members. You have to define it each time you include one of the files,
4878 though, and it must be identical each time.
4880 For example, the perl EV module uses something like this:
4882 #define EV_COMMON \
4883 SV *self; /* contains this struct */ \
4884 SV *cb_sv, *fh /* note no trailing ";" */
4886 =item EV_CB_DECLARE (type)
4888 =item EV_CB_INVOKE (watcher, revents)
4890 =item ev_set_cb (ev, cb)
4892 Can be used to change the callback member declaration in each watcher,
4893 and the way callbacks are invoked and set. Must expand to a struct member
4894 definition and a statement, respectively. See the F<ev.h> header file for
4895 their default definitions. One possible use for overriding these is to
4896 avoid the C<struct ev_loop *> as first argument in all cases, or to use
4897 method calls instead of plain function calls in C++.
4899 =back
4903 If you need to re-export the API (e.g. via a DLL) and you need a list of
4904 exported symbols, you can use the provided F<Symbol.*> files which list
4905 all public symbols, one per line:
4907 Symbols.ev for libev proper
4908 Symbols.event for the libevent emulation
4910 This can also be used to rename all public symbols to avoid clashes with
4911 multiple versions of libev linked together (which is obviously bad in
4912 itself, but sometimes it is inconvenient to avoid this).
4914 A sed command like this will create wrapper C<#define>'s that you need to
4915 include before including F<ev.h>:
4917 <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
4919 This would create a file F<wrap.h> which essentially looks like this:
4921 #define ev_backend myprefix_ev_backend
4922 #define ev_check_start myprefix_ev_check_start
4923 #define ev_check_stop myprefix_ev_check_stop
4924 ...
4926 =head2 EXAMPLES
4928 For a real-world example of a program the includes libev
4929 verbatim, you can have a look at the EV perl module
4930 (L<>). It has the libev files in
4931 the F<libev/> subdirectory and includes them in the F<EV/EVAPI.h> (public
4932 interface) and F<EV.xs> (implementation) files. Only the F<EV.xs> file
4933 will be compiled. It is pretty complex because it provides its own header
4934 file.
4936 The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
4937 that everybody includes and which overrides some configure choices:
4939 #define EV_FEATURES 8
4940 #define EV_USE_SELECT 1
4941 #define EV_PREPARE_ENABLE 1
4942 #define EV_IDLE_ENABLE 1
4943 #define EV_SIGNAL_ENABLE 1
4944 #define EV_CHILD_ENABLE 1
4945 #define EV_USE_STDEXCEPT 0
4946 #define EV_CONFIG_H <config.h>
4948 #include "ev++.h"
4950 And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
4952 #include "ev_cpp.h"
4953 #include "ev.c"
4959 =head3 THREADS
4961 All libev functions are reentrant and thread-safe unless explicitly
4962 documented otherwise, but lib