ViewVC Help
View File | Revision Log | Show Annotations | Download File
Revision: 1.241
Committed: Sat Apr 25 14:23:26 2009 UTC (15 years, 2 months ago) by root
Branch: MAIN
CVS Tags: rel-3_6
Changes since 1.240: +15 -14 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_loop's to stop iterating
32 ev_unloop (EV_A_ EVUNLOOP_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_loop to stop iterating
41 ev_unloop (EV_A_ EVUNLOOP_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_loop (0);
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_loop (loop, 0);
63 // unloop 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 Familarity with event based programming techniques in general is assumed
81 throughout this document.
83 =head1 ABOUT LIBEV
85 Libev is an event loop: you register interest in certain events (such as a
86 file descriptor being readable or a timeout occurring), and it will manage
87 these event sources and provide your program with events.
89 To do this, it must take more or less complete control over your process
90 (or thread) by executing the I<event loop> handler, and will then
91 communicate events via a callback mechanism.
93 You register interest in certain events by registering so-called I<event
94 watchers>, which are relatively small C structures you initialise with the
95 details of the event, and then hand it over to libev by I<starting> the
96 watcher.
98 =head2 FEATURES
100 Libev supports C<select>, C<poll>, the Linux-specific C<epoll>, the
101 BSD-specific C<kqueue> and the Solaris-specific event port mechanisms
102 for file descriptor events (C<ev_io>), the Linux C<inotify> interface
103 (for C<ev_stat>), relative timers (C<ev_timer>), absolute timers
104 with customised rescheduling (C<ev_periodic>), synchronous signals
105 (C<ev_signal>), process status change events (C<ev_child>), and event
106 watchers dealing with the event loop mechanism itself (C<ev_idle>,
107 C<ev_embed>, C<ev_prepare> and C<ev_check> watchers) as well as
108 file watchers (C<ev_stat>) and even limited support for fork events
109 (C<ev_fork>).
111 It also is quite fast (see this
112 L<benchmark|> comparing it to libevent
113 for example).
115 =head2 CONVENTIONS
117 Libev is very configurable. In this manual the default (and most common)
118 configuration will be described, which supports multiple event loops. For
119 more info about various configuration options please have a look at
120 B<EMBED> section in this manual. If libev was configured without support
121 for multiple event loops, then all functions taking an initial argument of
122 name C<loop> (which is always of type C<ev_loop *>) will not have
123 this argument.
127 Libev represents time as a single floating point number, representing
128 the (fractional) number of seconds since the (POSIX) epoch (somewhere
129 near the beginning of 1970, details are complicated, don't ask). This
130 type is called C<ev_tstamp>, which is what you should use too. It usually
131 aliases to the C<double> type in C. When you need to do any calculations
132 on it, you should treat it as some floating point value. Unlike the name
133 component C<stamp> might indicate, it is also used for time differences
134 throughout libev.
138 Libev knows three classes of errors: operating system errors, usage errors
139 and internal errors (bugs).
141 When libev catches an operating system error it cannot handle (for example
142 a system call indicating a condition libev cannot fix), it calls the callback
143 set via C<ev_set_syserr_cb>, which is supposed to fix the problem or
144 abort. The default is to print a diagnostic message and to call C<abort
145 ()>.
147 When libev detects a usage error such as a negative timer interval, then
148 it will print a diagnostic message and abort (via the C<assert> mechanism,
149 so C<NDEBUG> will disable this checking): these are programming errors in
150 the libev caller and need to be fixed there.
152 Libev also has a few internal error-checking C<assert>ions, and also has
153 extensive consistency checking code. These do not trigger under normal
154 circumstances, as they indicate either a bug in libev or worse.
159 These functions can be called anytime, even before initialising the
160 library in any way.
162 =over 4
164 =item ev_tstamp ev_time ()
166 Returns the current time as libev would use it. Please note that the
167 C<ev_now> function is usually faster and also often returns the timestamp
168 you actually want to know.
170 =item ev_sleep (ev_tstamp interval)
172 Sleep for the given interval: The current thread will be blocked until
173 either it is interrupted or the given time interval has passed. Basically
174 this is a sub-second-resolution C<sleep ()>.
176 =item int ev_version_major ()
178 =item int ev_version_minor ()
180 You can find out the major and minor ABI version numbers of the library
181 you linked against by calling the functions C<ev_version_major> and
182 C<ev_version_minor>. If you want, you can compare against the global
183 symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the
184 version of the library your program was compiled against.
186 These version numbers refer to the ABI version of the library, not the
187 release version.
189 Usually, it's a good idea to terminate if the major versions mismatch,
190 as this indicates an incompatible change. Minor versions are usually
191 compatible to older versions, so a larger minor version alone is usually
192 not a problem.
194 Example: Make sure we haven't accidentally been linked against the wrong
195 version.
197 assert (("libev version mismatch",
198 ev_version_major () == EV_VERSION_MAJOR
199 && ev_version_minor () >= EV_VERSION_MINOR));
201 =item unsigned int ev_supported_backends ()
203 Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
204 value) compiled into this binary of libev (independent of their
205 availability on the system you are running on). See C<ev_default_loop> for
206 a description of the set values.
208 Example: make sure we have the epoll method, because yeah this is cool and
209 a must have and can we have a torrent of it please!!!11
211 assert (("sorry, no epoll, no sex",
212 ev_supported_backends () & EVBACKEND_EPOLL));
214 =item unsigned int ev_recommended_backends ()
216 Return the set of all backends compiled into this binary of libev and also
217 recommended for this platform. This set is often smaller than the one
218 returned by C<ev_supported_backends>, as for example kqueue is broken on
219 most BSDs and will not be auto-detected unless you explicitly request it
220 (assuming you know what you are doing). This is the set of backends that
221 libev will probe for if you specify no backends explicitly.
223 =item unsigned int ev_embeddable_backends ()
225 Returns the set of backends that are embeddable in other event loops. This
226 is the theoretical, all-platform, value. To find which backends
227 might be supported on the current system, you would need to look at
228 C<ev_embeddable_backends () & ev_supported_backends ()>, likewise for
229 recommended ones.
231 See the description of C<ev_embed> watchers for more info.
233 =item ev_set_allocator (void *(*cb)(void *ptr, long size)) [NOT REENTRANT]
235 Sets the allocation function to use (the prototype is similar - the
236 semantics are identical to the C<realloc> C89/SuS/POSIX function). It is
237 used to allocate and free memory (no surprises here). If it returns zero
238 when memory needs to be allocated (C<size != 0>), the library might abort
239 or take some potentially destructive action.
241 Since some systems (at least OpenBSD and Darwin) fail to implement
242 correct C<realloc> semantics, libev will use a wrapper around the system
243 C<realloc> and C<free> functions by default.
245 You could override this function in high-availability programs to, say,
246 free some memory if it cannot allocate memory, to use a special allocator,
247 or even to sleep a while and retry until some memory is available.
249 Example: Replace the libev allocator with one that waits a bit and then
250 retries (example requires a standards-compliant C<realloc>).
252 static void *
253 persistent_realloc (void *ptr, size_t size)
254 {
255 for (;;)
256 {
257 void *newptr = realloc (ptr, size);
259 if (newptr)
260 return newptr;
262 sleep (60);
263 }
264 }
266 ...
267 ev_set_allocator (persistent_realloc);
269 =item ev_set_syserr_cb (void (*cb)(const char *msg)); [NOT REENTRANT]
271 Set the callback function to call on a retryable system call error (such
272 as failed select, poll, epoll_wait). The message is a printable string
273 indicating the system call or subsystem causing the problem. If this
274 callback is set, then libev will expect it to remedy the situation, no
275 matter what, when it returns. That is, libev will generally retry the
276 requested operation, or, if the condition doesn't go away, do bad stuff
277 (such as abort).
279 Example: This is basically the same thing that libev does internally, too.
281 static void
282 fatal_error (const char *msg)
283 {
284 perror (msg);
285 abort ();
286 }
288 ...
289 ev_set_syserr_cb (fatal_error);
291 =back
295 An event loop is described by a C<struct ev_loop *> (the C<struct>
296 is I<not> optional in this case, as there is also an C<ev_loop>
297 I<function>).
299 The library knows two types of such loops, the I<default> loop, which
300 supports signals and child events, and dynamically created loops which do
301 not.
303 =over 4
305 =item struct ev_loop *ev_default_loop (unsigned int flags)
307 This will initialise the default event loop if it hasn't been initialised
308 yet and return it. If the default loop could not be initialised, returns
309 false. If it already was initialised it simply returns it (and ignores the
310 flags. If that is troubling you, check C<ev_backend ()> afterwards).
312 If you don't know what event loop to use, use the one returned from this
313 function.
315 Note that this function is I<not> thread-safe, so if you want to use it
316 from multiple threads, you have to lock (note also that this is unlikely,
317 as loops cannot be shared easily between threads anyway).
319 The default loop is the only loop that can handle C<ev_signal> and
320 C<ev_child> watchers, and to do this, it always registers a handler
321 for C<SIGCHLD>. If this is a problem for your application you can either
322 create a dynamic loop with C<ev_loop_new> that doesn't do that, or you
323 can simply overwrite the C<SIGCHLD> signal handler I<after> calling
324 C<ev_default_init>.
326 The flags argument can be used to specify special behaviour or specific
327 backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
329 The following flags are supported:
331 =over 4
333 =item C<EVFLAG_AUTO>
335 The default flags value. Use this if you have no clue (it's the right
336 thing, believe me).
338 =item C<EVFLAG_NOENV>
340 If this flag bit is or'ed into the flag value (or the program runs setuid
341 or setgid) then libev will I<not> look at the environment variable
342 C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
343 override the flags completely if it is found in the environment. This is
344 useful to try out specific backends to test their performance, or to work
345 around bugs.
349 Instead of calling C<ev_default_fork> or C<ev_loop_fork> manually after
350 a fork, you can also make libev check for a fork in each iteration by
351 enabling this flag.
353 This works by calling C<getpid ()> on every iteration of the loop,
354 and thus this might slow down your event loop if you do a lot of loop
355 iterations and little real work, but is usually not noticeable (on my
356 GNU/Linux system for example, C<getpid> is actually a simple 5-insn sequence
357 without a system call and thus I<very> fast, but my GNU/Linux system also has
358 C<pthread_atfork> which is even faster).
360 The big advantage of this flag is that you can forget about fork (and
361 forget about forgetting to tell libev about forking) when you use this
362 flag.
364 This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
365 environment variable.
367 =item C<EVBACKEND_SELECT> (value 1, portable select backend)
369 This is your standard select(2) backend. Not I<completely> standard, as
370 libev tries to roll its own fd_set with no limits on the number of fds,
371 but if that fails, expect a fairly low limit on the number of fds when
372 using this backend. It doesn't scale too well (O(highest_fd)), but its
373 usually the fastest backend for a low number of (low-numbered :) fds.
375 To get good performance out of this backend you need a high amount of
376 parallelism (most of the file descriptors should be busy). If you are
377 writing a server, you should C<accept ()> in a loop to accept as many
378 connections as possible during one iteration. You might also want to have
379 a look at C<ev_set_io_collect_interval ()> to increase the amount of
380 readiness notifications you get per iteration.
382 This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
383 C<writefds> set (and to work around Microsoft Windows bugs, also onto the
384 C<exceptfds> set on that platform).
386 =item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
388 And this is your standard poll(2) backend. It's more complicated
389 than select, but handles sparse fds better and has no artificial
390 limit on the number of fds you can use (except it will slow down
391 considerably with a lot of inactive fds). It scales similarly to select,
392 i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
393 performance tips.
395 This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
398 =item C<EVBACKEND_EPOLL> (value 4, Linux)
400 For few fds, this backend is a bit little slower than poll and select,
401 but it scales phenomenally better. While poll and select usually scale
402 like O(total_fds) where n is the total number of fds (or the highest fd),
403 epoll scales either O(1) or O(active_fds).
405 The epoll mechanism deserves honorable mention as the most misdesigned
406 of the more advanced event mechanisms: mere annoyances include silently
407 dropping file descriptors, requiring a system call per change per file
408 descriptor (and unnecessary guessing of parameters), problems with dup and
409 so on. The biggest issue is fork races, however - if a program forks then
410 I<both> parent and child process have to recreate the epoll set, which can
411 take considerable time (one syscall per file descriptor) and is of course
412 hard to detect.
414 Epoll is also notoriously buggy - embedding epoll fds I<should> work, but
415 of course I<doesn't>, and epoll just loves to report events for totally
416 I<different> file descriptors (even already closed ones, so one cannot
417 even remove them from the set) than registered in the set (especially
418 on SMP systems). Libev tries to counter these spurious notifications by
419 employing an additional generation counter and comparing that against the
420 events to filter out spurious ones, recreating the set when required.
422 While stopping, setting and starting an I/O watcher in the same iteration
423 will result in some caching, there is still a system call per such
424 incident (because the same I<file descriptor> could point to a different
425 I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
426 file descriptors might not work very well if you register events for both
427 file descriptors.
429 Best performance from this backend is achieved by not unregistering all
430 watchers for a file descriptor until it has been closed, if possible,
431 i.e. keep at least one watcher active per fd at all times. Stopping and
432 starting a watcher (without re-setting it) also usually doesn't cause
433 extra overhead. A fork can both result in spurious notifications as well
434 as in libev having to destroy and recreate the epoll object, which can
435 take considerable time and thus should be avoided.
437 All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
438 faster than epoll for maybe up to a hundred file descriptors, depending on
439 the usage. So sad.
441 While nominally embeddable in other event loops, this feature is broken in
442 all kernel versions tested so far.
444 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
447 =item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
449 Kqueue deserves special mention, as at the time of this writing, it
450 was broken on all BSDs except NetBSD (usually it doesn't work reliably
451 with anything but sockets and pipes, except on Darwin, where of course
452 it's completely useless). Unlike epoll, however, whose brokenness
453 is by design, these kqueue bugs can (and eventually will) be fixed
454 without API changes to existing programs. For this reason it's not being
455 "auto-detected" unless you explicitly specify it in the flags (i.e. using
456 C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
457 system like NetBSD.
459 You still can embed kqueue into a normal poll or select backend and use it
460 only for sockets (after having made sure that sockets work with kqueue on
461 the target platform). See C<ev_embed> watchers for more info.
463 It scales in the same way as the epoll backend, but the interface to the
464 kernel is more efficient (which says nothing about its actual speed, of
465 course). While stopping, setting and starting an I/O watcher does never
466 cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
467 two event changes per incident. Support for C<fork ()> is very bad (but
468 sane, unlike epoll) and it drops fds silently in similarly hard-to-detect
469 cases
471 This backend usually performs well under most conditions.
473 While nominally embeddable in other event loops, this doesn't work
474 everywhere, so you might need to test for this. And since it is broken
475 almost everywhere, you should only use it when you have a lot of sockets
476 (for which it usually works), by embedding it into another event loop
477 (e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
478 also broken on OS X)) and, did I mention it, using it only for sockets.
480 This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
481 C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
482 C<NOTE_EOF>.
484 =item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
486 This is not implemented yet (and might never be, unless you send me an
487 implementation). According to reports, C</dev/poll> only supports sockets
488 and is not embeddable, which would limit the usefulness of this backend
489 immensely.
491 =item C<EVBACKEND_PORT> (value 32, Solaris 10)
493 This uses the Solaris 10 event port mechanism. As with everything on Solaris,
494 it's really slow, but it still scales very well (O(active_fds)).
496 Please note that Solaris event ports can deliver a lot of spurious
497 notifications, so you need to use non-blocking I/O or other means to avoid
498 blocking when no data (or space) is available.
500 While this backend scales well, it requires one system call per active
501 file descriptor per loop iteration. For small and medium numbers of file
502 descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
503 might perform better.
505 On the positive side, with the exception of the spurious readiness
506 notifications, this backend actually performed fully to specification
507 in all tests and is fully embeddable, which is a rare feat among the
508 OS-specific backends (I vastly prefer correctness over speed hacks).
510 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
513 =item C<EVBACKEND_ALL>
515 Try all backends (even potentially broken ones that wouldn't be tried
516 with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
519 It is definitely not recommended to use this flag.
521 =back
523 If one or more of these are or'ed into the flags value, then only these
524 backends will be tried (in the reverse order as listed here). If none are
525 specified, all backends in C<ev_recommended_backends ()> will be tried.
527 Example: This is the most typical usage.
529 if (!ev_default_loop (0))
530 fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
532 Example: Restrict libev to the select and poll backends, and do not allow
533 environment settings to be taken into account:
537 Example: Use whatever libev has to offer, but make sure that kqueue is
538 used if available (warning, breaks stuff, best use only with your own
539 private event loop and only if you know the OS supports your types of
540 fds):
542 ev_default_loop (ev_recommended_backends () | EVBACKEND_KQUEUE);
544 =item struct ev_loop *ev_loop_new (unsigned int flags)
546 Similar to C<ev_default_loop>, but always creates a new event loop that is
547 always distinct from the default loop. Unlike the default loop, it cannot
548 handle signal and child watchers, and attempts to do so will be greeted by
549 undefined behaviour (or a failed assertion if assertions are enabled).
551 Note that this function I<is> thread-safe, and the recommended way to use
552 libev with threads is indeed to create one loop per thread, and using the
553 default loop in the "main" or "initial" thread.
555 Example: Try to create a event loop that uses epoll and nothing else.
557 struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
558 if (!epoller)
559 fatal ("no epoll found here, maybe it hides under your chair");
561 =item ev_default_destroy ()
563 Destroys the default loop again (frees all memory and kernel state
564 etc.). None of the active event watchers will be stopped in the normal
565 sense, so e.g. C<ev_is_active> might still return true. It is your
566 responsibility to either stop all watchers cleanly yourself I<before>
567 calling this function, or cope with the fact afterwards (which is usually
568 the easiest thing, you can just ignore the watchers and/or C<free ()> them
569 for example).
571 Note that certain global state, such as signal state (and installed signal
572 handlers), will not be freed by this function, and related watchers (such
573 as signal and child watchers) would need to be stopped manually.
575 In general it is not advisable to call this function except in the
576 rare occasion where you really need to free e.g. the signal handling
577 pipe fds. If you need dynamically allocated loops it is better to use
578 C<ev_loop_new> and C<ev_loop_destroy>).
580 =item ev_loop_destroy (loop)
582 Like C<ev_default_destroy>, but destroys an event loop created by an
583 earlier call to C<ev_loop_new>.
585 =item ev_default_fork ()
587 This function sets a flag that causes subsequent C<ev_loop> iterations
588 to reinitialise the kernel state for backends that have one. Despite the
589 name, you can call it anytime, but it makes most sense after forking, in
590 the child process (or both child and parent, but that again makes little
591 sense). You I<must> call it in the child before using any of the libev
592 functions, and it will only take effect at the next C<ev_loop> iteration.
594 On the other hand, you only need to call this function in the child
595 process if and only if you want to use the event library in the child. If
596 you just fork+exec, you don't have to call it at all.
598 The function itself is quite fast and it's usually not a problem to call
599 it just in case after a fork. To make this easy, the function will fit in
600 quite nicely into a call to C<pthread_atfork>:
602 pthread_atfork (0, 0, ev_default_fork);
604 =item ev_loop_fork (loop)
606 Like C<ev_default_fork>, but acts on an event loop created by
607 C<ev_loop_new>. Yes, you have to call this on every allocated event loop
608 after fork that you want to re-use in the child, and how you do this is
609 entirely your own problem.
611 =item int ev_is_default_loop (loop)
613 Returns true when the given loop is, in fact, the default loop, and false
614 otherwise.
616 =item unsigned int ev_loop_count (loop)
618 Returns the count of loop iterations for the loop, which is identical to
619 the number of times libev did poll for new events. It starts at C<0> and
620 happily wraps around with enough iterations.
622 This value can sometimes be useful as a generation counter of sorts (it
623 "ticks" the number of loop iterations), as it roughly corresponds with
624 C<ev_prepare> and C<ev_check> calls.
626 =item unsigned int ev_backend (loop)
628 Returns one of the C<EVBACKEND_*> flags indicating the event backend in
629 use.
631 =item ev_tstamp ev_now (loop)
633 Returns the current "event loop time", which is the time the event loop
634 received events and started processing them. This timestamp does not
635 change as long as callbacks are being processed, and this is also the base
636 time used for relative timers. You can treat it as the timestamp of the
637 event occurring (or more correctly, libev finding out about it).
639 =item ev_now_update (loop)
641 Establishes the current time by querying the kernel, updating the time
642 returned by C<ev_now ()> in the progress. This is a costly operation and
643 is usually done automatically within C<ev_loop ()>.
645 This function is rarely useful, but when some event callback runs for a
646 very long time without entering the event loop, updating libev's idea of
647 the current time is a good idea.
649 See also L<The special problem of time updates> in the C<ev_timer> section.
651 =item ev_suspend (loop)
653 =item ev_resume (loop)
655 These two functions suspend and resume a loop, for use when the loop is
656 not used for a while and timeouts should not be processed.
658 A typical use case would be an interactive program such as a game: When
659 the user presses C<^Z> to suspend the game and resumes it an hour later it
660 would be best to handle timeouts as if no time had actually passed while
661 the program was suspended. This can be achieved by calling C<ev_suspend>
662 in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
663 C<ev_resume> directly afterwards to resume timer processing.
665 Effectively, all C<ev_timer> watchers will be delayed by the time spend
666 between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
667 will be rescheduled (that is, they will lose any events that would have
668 occured while suspended).
670 After calling C<ev_suspend> you B<must not> call I<any> function on the
671 given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
672 without a previous call to C<ev_suspend>.
674 Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
675 event loop time (see C<ev_now_update>).
677 =item ev_loop (loop, int flags)
679 Finally, this is it, the event handler. This function usually is called
680 after you initialised all your watchers and you want to start handling
681 events.
683 If the flags argument is specified as C<0>, it will not return until
684 either no event watchers are active anymore or C<ev_unloop> was called.
686 Please note that an explicit C<ev_unloop> is usually better than
687 relying on all watchers to be stopped when deciding when a program has
688 finished (especially in interactive programs), but having a program
689 that automatically loops as long as it has to and no longer by virtue
690 of relying on its watchers stopping correctly, that is truly a thing of
691 beauty.
693 A flags value of C<EVLOOP_NONBLOCK> will look for new events, will handle
694 those events and any already outstanding ones, but will not block your
695 process in case there are no events and will return after one iteration of
696 the loop.
698 A flags value of C<EVLOOP_ONESHOT> will look for new events (waiting if
699 necessary) and will handle those and any already outstanding ones. It
700 will block your process until at least one new event arrives (which could
701 be an event internal to libev itself, so there is no guarantee that a
702 user-registered callback will be called), and will return after one
703 iteration of the loop.
705 This is useful if you are waiting for some external event in conjunction
706 with something not expressible using other libev watchers (i.e. "roll your
707 own C<ev_loop>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
708 usually a better approach for this kind of thing.
710 Here are the gory details of what C<ev_loop> does:
712 - Before the first iteration, call any pending watchers.
713 * If EVFLAG_FORKCHECK was used, check for a fork.
714 - If a fork was detected (by any means), queue and call all fork watchers.
715 - Queue and call all prepare watchers.
716 - If we have been forked, detach and recreate the kernel state
717 as to not disturb the other process.
718 - Update the kernel state with all outstanding changes.
719 - Update the "event loop time" (ev_now ()).
720 - Calculate for how long to sleep or block, if at all
721 (active idle watchers, EVLOOP_NONBLOCK or not having
722 any active watchers at all will result in not sleeping).
723 - Sleep if the I/O and timer collect interval say so.
724 - Block the process, waiting for any events.
725 - Queue all outstanding I/O (fd) events.
726 - Update the "event loop time" (ev_now ()), and do time jump adjustments.
727 - Queue all expired timers.
728 - Queue all expired periodics.
729 - Unless any events are pending now, queue all idle watchers.
730 - Queue all check watchers.
731 - Call all queued watchers in reverse order (i.e. check watchers first).
732 Signals and child watchers are implemented as I/O watchers, and will
733 be handled here by queueing them when their watcher gets executed.
734 - If ev_unloop has been called, or EVLOOP_ONESHOT or EVLOOP_NONBLOCK
735 were used, or there are no active watchers, return, otherwise
736 continue with step *.
738 Example: Queue some jobs and then loop until no events are outstanding
739 anymore.
741 ... queue jobs here, make sure they register event watchers as long
742 ... as they still have work to do (even an idle watcher will do..)
743 ev_loop (my_loop, 0);
744 ... jobs done or somebody called unloop. yeah!
746 =item ev_unloop (loop, how)
748 Can be used to make a call to C<ev_loop> return early (but only after it
749 has processed all outstanding events). The C<how> argument must be either
750 C<EVUNLOOP_ONE>, which will make the innermost C<ev_loop> call return, or
751 C<EVUNLOOP_ALL>, which will make all nested C<ev_loop> calls return.
753 This "unloop state" will be cleared when entering C<ev_loop> again.
755 It is safe to call C<ev_unloop> from otuside any C<ev_loop> calls.
757 =item ev_ref (loop)
759 =item ev_unref (loop)
761 Ref/unref can be used to add or remove a reference count on the event
762 loop: Every watcher keeps one reference, and as long as the reference
763 count is nonzero, C<ev_loop> will not return on its own.
765 If you have a watcher you never unregister that should not keep C<ev_loop>
766 from returning, call ev_unref() after starting, and ev_ref() before
767 stopping it.
769 As an example, libev itself uses this for its internal signal pipe: It
770 is not visible to the libev user and should not keep C<ev_loop> from
771 exiting if no event watchers registered by it are active. It is also an
772 excellent way to do this for generic recurring timers or from within
773 third-party libraries. Just remember to I<unref after start> and I<ref
774 before stop> (but only if the watcher wasn't active before, or was active
775 before, respectively. Note also that libev might stop watchers itself
776 (e.g. non-repeating timers) in which case you have to C<ev_ref>
777 in the callback).
779 Example: Create a signal watcher, but keep it from keeping C<ev_loop>
780 running when nothing else is active.
782 ev_signal exitsig;
783 ev_signal_init (&exitsig, sig_cb, SIGINT);
784 ev_signal_start (loop, &exitsig);
785 evf_unref (loop);
787 Example: For some weird reason, unregister the above signal handler again.
789 ev_ref (loop);
790 ev_signal_stop (loop, &exitsig);
792 =item ev_set_io_collect_interval (loop, ev_tstamp interval)
794 =item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
796 These advanced functions influence the time that libev will spend waiting
797 for events. Both time intervals are by default C<0>, meaning that libev
798 will try to invoke timer/periodic callbacks and I/O callbacks with minimum
799 latency.
801 Setting these to a higher value (the C<interval> I<must> be >= C<0>)
802 allows libev to delay invocation of I/O and timer/periodic callbacks
803 to increase efficiency of loop iterations (or to increase power-saving
804 opportunities).
806 The idea is that sometimes your program runs just fast enough to handle
807 one (or very few) event(s) per loop iteration. While this makes the
808 program responsive, it also wastes a lot of CPU time to poll for new
809 events, especially with backends like C<select ()> which have a high
810 overhead for the actual polling but can deliver many events at once.
812 By setting a higher I<io collect interval> you allow libev to spend more
813 time collecting I/O events, so you can handle more events per iteration,
814 at the cost of increasing latency. Timeouts (both C<ev_periodic> and
815 C<ev_timer>) will be not affected. Setting this to a non-null value will
816 introduce an additional C<ev_sleep ()> call into most loop iterations.
818 Likewise, by setting a higher I<timeout collect interval> you allow libev
819 to spend more time collecting timeouts, at the expense of increased
820 latency/jitter/inexactness (the watcher callback will be called
821 later). C<ev_io> watchers will not be affected. Setting this to a non-null
822 value will not introduce any overhead in libev.
824 Many (busy) programs can usually benefit by setting the I/O collect
825 interval to a value near C<0.1> or so, which is often enough for
826 interactive servers (of course not for games), likewise for timeouts. It
827 usually doesn't make much sense to set it to a lower value than C<0.01>,
828 as this approaches the timing granularity of most systems.
830 Setting the I<timeout collect interval> can improve the opportunity for
831 saving power, as the program will "bundle" timer callback invocations that
832 are "near" in time together, by delaying some, thus reducing the number of
833 times the process sleeps and wakes up again. Another useful technique to
834 reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
835 they fire on, say, one-second boundaries only.
837 =item ev_loop_verify (loop)
839 This function only does something when C<EV_VERIFY> support has been
840 compiled in, which is the default for non-minimal builds. It tries to go
841 through all internal structures and checks them for validity. If anything
842 is found to be inconsistent, it will print an error message to standard
843 error and call C<abort ()>.
845 This can be used to catch bugs inside libev itself: under normal
846 circumstances, this function will never abort as of course libev keeps its
847 data structures consistent.
849 =back
854 In the following description, uppercase C<TYPE> in names stands for the
855 watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
856 watchers and C<ev_io_start> for I/O watchers.
858 A watcher is a structure that you create and register to record your
859 interest in some event. For instance, if you want to wait for STDIN to
860 become readable, you would create an C<ev_io> watcher for that:
862 static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
863 {
864 ev_io_stop (w);
865 ev_unloop (loop, EVUNLOOP_ALL);
866 }
868 struct ev_loop *loop = ev_default_loop (0);
870 ev_io stdin_watcher;
872 ev_init (&stdin_watcher, my_cb);
873 ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
874 ev_io_start (loop, &stdin_watcher);
876 ev_loop (loop, 0);
878 As you can see, you are responsible for allocating the memory for your
879 watcher structures (and it is I<usually> a bad idea to do this on the
880 stack).
882 Each watcher has an associated watcher structure (called C<struct ev_TYPE>
883 or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
885 Each watcher structure must be initialised by a call to C<ev_init
886 (watcher *, callback)>, which expects a callback to be provided. This
887 callback gets invoked each time the event occurs (or, in the case of I/O
888 watchers, each time the event loop detects that the file descriptor given
889 is readable and/or writable).
891 Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
892 macro to configure it, with arguments specific to the watcher type. There
893 is also a macro to combine initialisation and setting in one call: C<<
894 ev_TYPE_init (watcher *, callback, ...) >>.
896 To make the watcher actually watch out for events, you have to start it
897 with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
898 *) >>), and you can stop watching for events at any time by calling the
899 corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
901 As long as your watcher is active (has been started but not stopped) you
902 must not touch the values stored in it. Most specifically you must never
903 reinitialise it or call its C<ev_TYPE_set> macro.
905 Each and every callback receives the event loop pointer as first, the
906 registered watcher structure as second, and a bitset of received events as
907 third argument.
909 The received events usually include a single bit per event type received
910 (you can receive multiple events at the same time). The possible bit masks
911 are:
913 =over 4
915 =item C<EV_READ>
917 =item C<EV_WRITE>
919 The file descriptor in the C<ev_io> watcher has become readable and/or
920 writable.
922 =item C<EV_TIMEOUT>
924 The C<ev_timer> watcher has timed out.
926 =item C<EV_PERIODIC>
928 The C<ev_periodic> watcher has timed out.
930 =item C<EV_SIGNAL>
932 The signal specified in the C<ev_signal> watcher has been received by a thread.
934 =item C<EV_CHILD>
936 The pid specified in the C<ev_child> watcher has received a status change.
938 =item C<EV_STAT>
940 The path specified in the C<ev_stat> watcher changed its attributes somehow.
942 =item C<EV_IDLE>
944 The C<ev_idle> watcher has determined that you have nothing better to do.
946 =item C<EV_PREPARE>
948 =item C<EV_CHECK>
950 All C<ev_prepare> watchers are invoked just I<before> C<ev_loop> starts
951 to gather new events, and all C<ev_check> watchers are invoked just after
952 C<ev_loop> has gathered them, but before it invokes any callbacks for any
953 received events. Callbacks of both watcher types can start and stop as
954 many watchers as they want, and all of them will be taken into account
955 (for example, a C<ev_prepare> watcher might start an idle watcher to keep
956 C<ev_loop> from blocking).
958 =item C<EV_EMBED>
960 The embedded event loop specified in the C<ev_embed> watcher needs attention.
962 =item C<EV_FORK>
964 The event loop has been resumed in the child process after fork (see
965 C<ev_fork>).
967 =item C<EV_ASYNC>
969 The given async watcher has been asynchronously notified (see C<ev_async>).
971 =item C<EV_CUSTOM>
973 Not ever sent (or otherwise used) by libev itself, but can be freely used
974 by libev users to signal watchers (e.g. via C<ev_feed_event>).
976 =item C<EV_ERROR>
978 An unspecified error has occurred, the watcher has been stopped. This might
979 happen because the watcher could not be properly started because libev
980 ran out of memory, a file descriptor was found to be closed or any other
981 problem. Libev considers these application bugs.
983 You best act on it by reporting the problem and somehow coping with the
984 watcher being stopped. Note that well-written programs should not receive
985 an error ever, so when your watcher receives it, this usually indicates a
986 bug in your program.
988 Libev will usually signal a few "dummy" events together with an error, for
989 example it might indicate that a fd is readable or writable, and if your
990 callbacks is well-written it can just attempt the operation and cope with
991 the error from read() or write(). This will not work in multi-threaded
992 programs, though, as the fd could already be closed and reused for another
993 thing, so beware.
995 =back
999 =over 4
1001 =item C<ev_init> (ev_TYPE *watcher, callback)
1003 This macro initialises the generic portion of a watcher. The contents
1004 of the watcher object can be arbitrary (so C<malloc> will do). Only
1005 the generic parts of the watcher are initialised, you I<need> to call
1006 the type-specific C<ev_TYPE_set> macro afterwards to initialise the
1007 type-specific parts. For each type there is also a C<ev_TYPE_init> macro
1008 which rolls both calls into one.
1010 You can reinitialise a watcher at any time as long as it has been stopped
1011 (or never started) and there are no pending events outstanding.
1013 The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
1014 int revents)>.
1016 Example: Initialise an C<ev_io> watcher in two steps.
1018 ev_io w;
1019 ev_init (&w, my_cb);
1020 ev_io_set (&w, STDIN_FILENO, EV_READ);
1022 =item C<ev_TYPE_set> (ev_TYPE *, [args])
1024 This macro initialises the type-specific parts of a watcher. You need to
1025 call C<ev_init> at least once before you call this macro, but you can
1026 call C<ev_TYPE_set> any number of times. You must not, however, call this
1027 macro on a watcher that is active (it can be pending, however, which is a
1028 difference to the C<ev_init> macro).
1030 Although some watcher types do not have type-specific arguments
1031 (e.g. C<ev_prepare>) you still need to call its C<set> macro.
1033 See C<ev_init>, above, for an example.
1035 =item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
1037 This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
1038 calls into a single call. This is the most convenient method to initialise
1039 a watcher. The same limitations apply, of course.
1041 Example: Initialise and set an C<ev_io> watcher in one step.
1043 ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
1045 =item C<ev_TYPE_start> (loop *, ev_TYPE *watcher)
1047 Starts (activates) the given watcher. Only active watchers will receive
1048 events. If the watcher is already active nothing will happen.
1050 Example: Start the C<ev_io> watcher that is being abused as example in this
1051 whole section.
1053 ev_io_start (EV_DEFAULT_UC, &w);
1055 =item C<ev_TYPE_stop> (loop *, ev_TYPE *watcher)
1057 Stops the given watcher if active, and clears the pending status (whether
1058 the watcher was active or not).
1060 It is possible that stopped watchers are pending - for example,
1061 non-repeating timers are being stopped when they become pending - but
1062 calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
1063 pending. If you want to free or reuse the memory used by the watcher it is
1064 therefore a good idea to always call its C<ev_TYPE_stop> function.
1066 =item bool ev_is_active (ev_TYPE *watcher)
1068 Returns a true value iff the watcher is active (i.e. it has been started
1069 and not yet been stopped). As long as a watcher is active you must not modify
1070 it.
1072 =item bool ev_is_pending (ev_TYPE *watcher)
1074 Returns a true value iff the watcher is pending, (i.e. it has outstanding
1075 events but its callback has not yet been invoked). As long as a watcher
1076 is pending (but not active) you must not call an init function on it (but
1077 C<ev_TYPE_set> is safe), you must not change its priority, and you must
1078 make sure the watcher is available to libev (e.g. you cannot C<free ()>
1079 it).
1081 =item callback ev_cb (ev_TYPE *watcher)
1083 Returns the callback currently set on the watcher.
1085 =item ev_cb_set (ev_TYPE *watcher, callback)
1087 Change the callback. You can change the callback at virtually any time
1088 (modulo threads).
1090 =item ev_set_priority (ev_TYPE *watcher, priority)
1092 =item int ev_priority (ev_TYPE *watcher)
1094 Set and query the priority of the watcher. The priority is a small
1095 integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
1096 (default: C<-2>). Pending watchers with higher priority will be invoked
1097 before watchers with lower priority, but priority will not keep watchers
1098 from being executed (except for C<ev_idle> watchers).
1100 If you need to suppress invocation when higher priority events are pending
1101 you need to look at C<ev_idle> watchers, which provide this functionality.
1103 You I<must not> change the priority of a watcher as long as it is active or
1104 pending.
1106 Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
1107 fine, as long as you do not mind that the priority value you query might
1108 or might not have been clamped to the valid range.
1110 The default priority used by watchers when no priority has been set is
1111 always C<0>, which is supposed to not be too high and not be too low :).
1113 See L<WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
1114 priorities.
1116 =item ev_invoke (loop, ev_TYPE *watcher, int revents)
1118 Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
1119 C<loop> nor C<revents> need to be valid as long as the watcher callback
1120 can deal with that fact, as both are simply passed through to the
1121 callback.
1123 =item int ev_clear_pending (loop, ev_TYPE *watcher)
1125 If the watcher is pending, this function clears its pending status and
1126 returns its C<revents> bitset (as if its callback was invoked). If the
1127 watcher isn't pending it does nothing and returns C<0>.
1129 Sometimes it can be useful to "poll" a watcher instead of waiting for its
1130 callback to be invoked, which can be accomplished with this function.
1132 =back
1137 Each watcher has, by default, a member C<void *data> that you can change
1138 and read at any time: libev will completely ignore it. This can be used
1139 to associate arbitrary data with your watcher. If you need more data and
1140 don't want to allocate memory and store a pointer to it in that data
1141 member, you can also "subclass" the watcher type and provide your own
1142 data:
1144 struct my_io
1145 {
1146 ev_io io;
1147 int otherfd;
1148 void *somedata;
1149 struct whatever *mostinteresting;
1150 };
1152 ...
1153 struct my_io w;
1154 ev_io_init (&, my_cb, fd, EV_READ);
1156 And since your callback will be called with a pointer to the watcher, you
1157 can cast it back to your own type:
1159 static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
1160 {
1161 struct my_io *w = (struct my_io *)w_;
1162 ...
1163 }
1165 More interesting and less C-conformant ways of casting your callback type
1166 instead have been omitted.
1168 Another common scenario is to use some data structure with multiple
1169 embedded watchers:
1171 struct my_biggy
1172 {
1173 int some_data;
1174 ev_timer t1;
1175 ev_timer t2;
1176 }
1178 In this case getting the pointer to C<my_biggy> is a bit more
1179 complicated: Either you store the address of your C<my_biggy> struct
1180 in the C<data> member of the watcher (for woozies), or you need to use
1181 some pointer arithmetic using C<offsetof> inside your watchers (for real
1182 programmers):
1184 #include <stddef.h>
1186 static void
1187 t1_cb (EV_P_ ev_timer *w, int revents)
1188 {
1189 struct my_biggy big = (struct my_biggy *
1190 (((char *)w) - offsetof (struct my_biggy, t1));
1191 }
1193 static void
1194 t2_cb (EV_P_ ev_timer *w, int revents)
1195 {
1196 struct my_biggy big = (struct my_biggy *
1197 (((char *)w) - offsetof (struct my_biggy, t2));
1198 }
1202 Many event loops support I<watcher priorities>, which are usually small
1203 integers that influence the ordering of event callback invocation
1204 between watchers in some way, all else being equal.
1206 In libev, Watcher priorities can be set using C<ev_set_priority>. See its
1207 description for the more technical details such as the actual priority
1208 range.
1210 There are two common ways how these these priorities are being interpreted
1211 by event loops:
1213 In the more common lock-out model, higher priorities "lock out" invocation
1214 of lower priority watchers, which means as long as higher priority
1215 watchers receive events, lower priority watchers are not being invoked.
1217 The less common only-for-ordering model uses priorities solely to order
1218 callback invocation within a single event loop iteration: Higher priority
1219 watchers are invoked before lower priority ones, but they all get invoked
1220 before polling for new events.
1222 Libev uses the second (only-for-ordering) model for all its watchers
1223 except for idle watchers (which use the lock-out model).
1225 The rationale behind this is that implementing the lock-out model for
1226 watchers is not well supported by most kernel interfaces, and most event
1227 libraries will just poll for the same events again and again as long as
1228 their callbacks have not been executed, which is very inefficient in the
1229 common case of one high-priority watcher locking out a mass of lower
1230 priority ones.
1232 Static (ordering) priorities are most useful when you have two or more
1233 watchers handling the same resource: a typical usage example is having an
1234 C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
1235 timeouts. Under load, data might be received while the program handles
1236 other jobs, but since timers normally get invoked first, the timeout
1237 handler will be executed before checking for data. In that case, giving
1238 the timer a lower priority than the I/O watcher ensures that I/O will be
1239 handled first even under adverse conditions (which is usually, but not
1240 always, what you want).
1242 Since idle watchers use the "lock-out" model, meaning that idle watchers
1243 will only be executed when no same or higher priority watchers have
1244 received events, they can be used to implement the "lock-out" model when
1245 required.
1247 For example, to emulate how many other event libraries handle priorities,
1248 you can associate an C<ev_idle> watcher to each such watcher, and in
1249 the normal watcher callback, you just start the idle watcher. The real
1250 processing is done in the idle watcher callback. This causes libev to
1251 continously poll and process kernel event data for the watcher, but when
1252 the lock-out case is known to be rare (which in turn is rare :), this is
1253 workable.
1255 Usually, however, the lock-out model implemented that way will perform
1256 miserably under the type of load it was designed to handle. In that case,
1257 it might be preferable to stop the real watcher before starting the
1258 idle watcher, so the kernel will not have to process the event in case
1259 the actual processing will be delayed for considerable time.
1261 Here is an example of an I/O watcher that should run at a strictly lower
1262 priority than the default, and which should only process data when no
1263 other events are pending:
1265 ev_idle idle; // actual processing watcher
1266 ev_io io; // actual event watcher
1268 static void
1269 io_cb (EV_P_ ev_io *w, int revents)
1270 {
1271 // stop the I/O watcher, we received the event, but
1272 // are not yet ready to handle it.
1273 ev_io_stop (EV_A_ w);
1275 // start the idle watcher to ahndle the actual event.
1276 // it will not be executed as long as other watchers
1277 // with the default priority are receiving events.
1278 ev_idle_start (EV_A_ &idle);
1279 }
1281 static void
1282 idle-cb (EV_P_ ev_idle *w, int revents)
1283 {
1284 // actual processing
1285 read (STDIN_FILENO, ...);
1287 // have to start the I/O watcher again, as
1288 // we have handled the event
1289 ev_io_start (EV_P_ &io);
1290 }
1292 // initialisation
1293 ev_idle_init (&idle, idle_cb);
1294 ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
1295 ev_io_start (EV_DEFAULT_ &io);
1297 In the "real" world, it might also be beneficial to start a timer, so that
1298 low-priority connections can not be locked out forever under load. This
1299 enables your program to keep a lower latency for important connections
1300 during short periods of high load, while not completely locking out less
1301 important ones.
1304 =head1 WATCHER TYPES
1306 This section describes each watcher in detail, but will not repeat
1307 information given in the last section. Any initialisation/set macros,
1308 functions and members specific to the watcher type are explained.
1310 Members are additionally marked with either I<[read-only]>, meaning that,
1311 while the watcher is active, you can look at the member and expect some
1312 sensible content, but you must not modify it (you can modify it while the
1313 watcher is stopped to your hearts content), or I<[read-write]>, which
1314 means you can expect it to have some sensible content while the watcher
1315 is active, but you can also modify it. Modifying it may not do something
1316 sensible or take immediate effect (or do anything at all), but libev will
1317 not crash or malfunction in any way.
1320 =head2 C<ev_io> - is this file descriptor readable or writable?
1322 I/O watchers check whether a file descriptor is readable or writable
1323 in each iteration of the event loop, or, more precisely, when reading
1324 would not block the process and writing would at least be able to write
1325 some data. This behaviour is called level-triggering because you keep
1326 receiving events as long as the condition persists. Remember you can stop
1327 the watcher if you don't want to act on the event and neither want to
1328 receive future events.
1330 In general you can register as many read and/or write event watchers per
1331 fd as you want (as long as you don't confuse yourself). Setting all file
1332 descriptors to non-blocking mode is also usually a good idea (but not
1333 required if you know what you are doing).
1335 If you cannot use non-blocking mode, then force the use of a
1336 known-to-be-good backend (at the time of this writing, this includes only
1337 C<EVBACKEND_SELECT> and C<EVBACKEND_POLL>). The same applies to file
1338 descriptors for which non-blocking operation makes no sense (such as
1339 files) - libev doesn't guarentee any specific behaviour in that case.
1341 Another thing you have to watch out for is that it is quite easy to
1342 receive "spurious" readiness notifications, that is your callback might
1343 be called with C<EV_READ> but a subsequent C<read>(2) will actually block
1344 because there is no data. Not only are some backends known to create a
1345 lot of those (for example Solaris ports), it is very easy to get into
1346 this situation even with a relatively standard program structure. Thus
1347 it is best to always use non-blocking I/O: An extra C<read>(2) returning
1348 C<EAGAIN> is far preferable to a program hanging until some data arrives.
1350 If you cannot run the fd in non-blocking mode (for example you should
1351 not play around with an Xlib connection), then you have to separately
1352 re-test whether a file descriptor is really ready with a known-to-be good
1353 interface such as poll (fortunately in our Xlib example, Xlib already
1354 does this on its own, so its quite safe to use). Some people additionally
1355 use C<SIGALRM> and an interval timer, just to be sure you won't block
1356 indefinitely.
1358 But really, best use non-blocking mode.
1360 =head3 The special problem of disappearing file descriptors
1362 Some backends (e.g. kqueue, epoll) need to be told about closing a file
1363 descriptor (either due to calling C<close> explicitly or any other means,
1364 such as C<dup2>). The reason is that you register interest in some file
1365 descriptor, but when it goes away, the operating system will silently drop
1366 this interest. If another file descriptor with the same number then is
1367 registered with libev, there is no efficient way to see that this is, in
1368 fact, a different file descriptor.
1370 To avoid having to explicitly tell libev about such cases, libev follows
1371 the following policy: Each time C<ev_io_set> is being called, libev
1372 will assume that this is potentially a new file descriptor, otherwise
1373 it is assumed that the file descriptor stays the same. That means that
1374 you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
1375 descriptor even if the file descriptor number itself did not change.
1377 This is how one would do it normally anyway, the important point is that
1378 the libev application should not optimise around libev but should leave
1379 optimisations to libev.
1381 =head3 The special problem of dup'ed file descriptors
1383 Some backends (e.g. epoll), cannot register events for file descriptors,
1384 but only events for the underlying file descriptions. That means when you
1385 have C<dup ()>'ed file descriptors or weirder constellations, and register
1386 events for them, only one file descriptor might actually receive events.
1388 There is no workaround possible except not registering events
1389 for potentially C<dup ()>'ed file descriptors, or to resort to
1392 =head3 The special problem of fork
1394 Some backends (epoll, kqueue) do not support C<fork ()> at all or exhibit
1395 useless behaviour. Libev fully supports fork, but needs to be told about
1396 it in the child.
1398 To support fork in your programs, you either have to call
1399 C<ev_default_fork ()> or C<ev_loop_fork ()> after a fork in the child,
1400 enable C<EVFLAG_FORKCHECK>, or resort to C<EVBACKEND_SELECT> or
1403 =head3 The special problem of SIGPIPE
1405 While not really specific to libev, it is easy to forget about C<SIGPIPE>:
1406 when writing to a pipe whose other end has been closed, your program gets
1407 sent a SIGPIPE, which, by default, aborts your program. For most programs
1408 this is sensible behaviour, for daemons, this is usually undesirable.
1410 So when you encounter spurious, unexplained daemon exits, make sure you
1411 ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
1412 somewhere, as that would have given you a big clue).
1415 =head3 Watcher-Specific Functions
1417 =over 4
1419 =item ev_io_init (ev_io *, callback, int fd, int events)
1421 =item ev_io_set (ev_io *, int fd, int events)
1423 Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
1424 receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
1425 C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
1427 =item int fd [read-only]
1429 The file descriptor being watched.
1431 =item int events [read-only]
1433 The events being watched.
1435 =back
1437 =head3 Examples
1439 Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
1440 readable, but only once. Since it is likely line-buffered, you could
1441 attempt to read a whole line in the callback.
1443 static void
1444 stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
1445 {
1446 ev_io_stop (loop, w);
1447 .. read from stdin here (or from w->fd) and handle any I/O errors
1448 }
1450 ...
1451 struct ev_loop *loop = ev_default_init (0);
1452 ev_io stdin_readable;
1453 ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
1454 ev_io_start (loop, &stdin_readable);
1455 ev_loop (loop, 0);
1458 =head2 C<ev_timer> - relative and optionally repeating timeouts
1460 Timer watchers are simple relative timers that generate an event after a
1461 given time, and optionally repeating in regular intervals after that.
1463 The timers are based on real time, that is, if you register an event that
1464 times out after an hour and you reset your system clock to January last
1465 year, it will still time out after (roughly) one hour. "Roughly" because
1466 detecting time jumps is hard, and some inaccuracies are unavoidable (the
1467 monotonic clock option helps a lot here).
1469 The callback is guaranteed to be invoked only I<after> its timeout has
1470 passed (not I<at>, so on systems with very low-resolution clocks this
1471 might introduce a small delay). If multiple timers become ready during the
1472 same loop iteration then the ones with earlier time-out values are invoked
1473 before ones with later time-out values (but this is no longer true when a
1474 callback calls C<ev_loop> recursively).
1476 =head3 Be smart about timeouts
1478 Many real-world problems involve some kind of timeout, usually for error
1479 recovery. A typical example is an HTTP request - if the other side hangs,
1480 you want to raise some error after a while.
1482 What follows are some ways to handle this problem, from obvious and
1483 inefficient to smart and efficient.
1485 In the following, a 60 second activity timeout is assumed - a timeout that
1486 gets reset to 60 seconds each time there is activity (e.g. each time some
1487 data or other life sign was received).
1489 =over 4
1491 =item 1. Use a timer and stop, reinitialise and start it on activity.
1493 This is the most obvious, but not the most simple way: In the beginning,
1494 start the watcher:
1496 ev_timer_init (timer, callback, 60., 0.);
1497 ev_timer_start (loop, timer);
1499 Then, each time there is some activity, C<ev_timer_stop> it, initialise it
1500 and start it again:
1502 ev_timer_stop (loop, timer);
1503 ev_timer_set (timer, 60., 0.);
1504 ev_timer_start (loop, timer);
1506 This is relatively simple to implement, but means that each time there is
1507 some activity, libev will first have to remove the timer from its internal
1508 data structure and then add it again. Libev tries to be fast, but it's
1509 still not a constant-time operation.
1511 =item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
1513 This is the easiest way, and involves using C<ev_timer_again> instead of
1514 C<ev_timer_start>.
1516 To implement this, configure an C<ev_timer> with a C<repeat> value
1517 of C<60> and then call C<ev_timer_again> at start and each time you
1518 successfully read or write some data. If you go into an idle state where
1519 you do not expect data to travel on the socket, you can C<ev_timer_stop>
1520 the timer, and C<ev_timer_again> will automatically restart it if need be.
1522 That means you can ignore both the C<ev_timer_start> function and the
1523 C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
1524 member and C<ev_timer_again>.
1526 At start:
1528 ev_timer_init (timer, callback);
1529 timer->repeat = 60.;
1530 ev_timer_again (loop, timer);
1532 Each time there is some activity:
1534 ev_timer_again (loop, timer);
1536 It is even possible to change the time-out on the fly, regardless of
1537 whether the watcher is active or not:
1539 timer->repeat = 30.;
1540 ev_timer_again (loop, timer);
1542 This is slightly more efficient then stopping/starting the timer each time
1543 you want to modify its timeout value, as libev does not have to completely
1544 remove and re-insert the timer from/into its internal data structure.
1546 It is, however, even simpler than the "obvious" way to do it.
1548 =item 3. Let the timer time out, but then re-arm it as required.
1550 This method is more tricky, but usually most efficient: Most timeouts are
1551 relatively long compared to the intervals between other activity - in
1552 our example, within 60 seconds, there are usually many I/O events with
1553 associated activity resets.
1555 In this case, it would be more efficient to leave the C<ev_timer> alone,
1556 but remember the time of last activity, and check for a real timeout only
1557 within the callback:
1559 ev_tstamp last_activity; // time of last activity
1561 static void
1562 callback (EV_P_ ev_timer *w, int revents)
1563 {
1564 ev_tstamp now = ev_now (EV_A);
1565 ev_tstamp timeout = last_activity + 60.;
1567 // if last_activity + 60. is older than now, we did time out
1568 if (timeout < now)
1569 {
1570 // timeout occured, take action
1571 }
1572 else
1573 {
1574 // callback was invoked, but there was some activity, re-arm
1575 // the watcher to fire in last_activity + 60, which is
1576 // guaranteed to be in the future, so "again" is positive:
1577 w->repeat = timeout - now;
1578 ev_timer_again (EV_A_ w);
1579 }
1580 }
1582 To summarise the callback: first calculate the real timeout (defined
1583 as "60 seconds after the last activity"), then check if that time has
1584 been reached, which means something I<did>, in fact, time out. Otherwise
1585 the callback was invoked too early (C<timeout> is in the future), so
1586 re-schedule the timer to fire at that future time, to see if maybe we have
1587 a timeout then.
1589 Note how C<ev_timer_again> is used, taking advantage of the
1590 C<ev_timer_again> optimisation when the timer is already running.
1592 This scheme causes more callback invocations (about one every 60 seconds
1593 minus half the average time between activity), but virtually no calls to
1594 libev to change the timeout.
1596 To start the timer, simply initialise the watcher and set C<last_activity>
1597 to the current time (meaning we just have some activity :), then call the
1598 callback, which will "do the right thing" and start the timer:
1600 ev_timer_init (timer, callback);
1601 last_activity = ev_now (loop);
1602 callback (loop, timer, EV_TIMEOUT);
1604 And when there is some activity, simply store the current time in
1605 C<last_activity>, no libev calls at all:
1607 last_actiivty = ev_now (loop);
1609 This technique is slightly more complex, but in most cases where the
1610 time-out is unlikely to be triggered, much more efficient.
1612 Changing the timeout is trivial as well (if it isn't hard-coded in the
1613 callback :) - just change the timeout and invoke the callback, which will
1614 fix things for you.
1616 =item 4. Wee, just use a double-linked list for your timeouts.
1618 If there is not one request, but many thousands (millions...), all
1619 employing some kind of timeout with the same timeout value, then one can
1620 do even better:
1622 When starting the timeout, calculate the timeout value and put the timeout
1623 at the I<end> of the list.
1625 Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
1626 the list is expected to fire (for example, using the technique #3).
1628 When there is some activity, remove the timer from the list, recalculate
1629 the timeout, append it to the end of the list again, and make sure to
1630 update the C<ev_timer> if it was taken from the beginning of the list.
1632 This way, one can manage an unlimited number of timeouts in O(1) time for
1633 starting, stopping and updating the timers, at the expense of a major
1634 complication, and having to use a constant timeout. The constant timeout
1635 ensures that the list stays sorted.
1637 =back
1639 So which method the best?
1641 Method #2 is a simple no-brain-required solution that is adequate in most
1642 situations. Method #3 requires a bit more thinking, but handles many cases
1643 better, and isn't very complicated either. In most case, choosing either
1644 one is fine, with #3 being better in typical situations.
1646 Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
1647 rather complicated, but extremely efficient, something that really pays
1648 off after the first million or so of active timers, i.e. it's usually
1649 overkill :)
1651 =head3 The special problem of time updates
1653 Establishing the current time is a costly operation (it usually takes at
1654 least two system calls): EV therefore updates its idea of the current
1655 time only before and after C<ev_loop> collects new events, which causes a
1656 growing difference between C<ev_now ()> and C<ev_time ()> when handling
1657 lots of events in one iteration.
1659 The relative timeouts are calculated relative to the C<ev_now ()>
1660 time. This is usually the right thing as this timestamp refers to the time
1661 of the event triggering whatever timeout you are modifying/starting. If
1662 you suspect event processing to be delayed and you I<need> to base the
1663 timeout on the current time, use something like this to adjust for this:
1665 ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);
1667 If the event loop is suspended for a long time, you can also force an
1668 update of the time returned by C<ev_now ()> by calling C<ev_now_update
1669 ()>.
1671 =head3 Watcher-Specific Functions and Data Members
1673 =over 4
1675 =item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
1677 =item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
1679 Configure the timer to trigger after C<after> seconds. If C<repeat>
1680 is C<0.>, then it will automatically be stopped once the timeout is
1681 reached. If it is positive, then the timer will automatically be
1682 configured to trigger again C<repeat> seconds later, again, and again,
1683 until stopped manually.
1685 The timer itself will do a best-effort at avoiding drift, that is, if
1686 you configure a timer to trigger every 10 seconds, then it will normally
1687 trigger at exactly 10 second intervals. If, however, your program cannot
1688 keep up with the timer (because it takes longer than those 10 seconds to
1689 do stuff) the timer will not fire more than once per event loop iteration.
1691 =item ev_timer_again (loop, ev_timer *)
1693 This will act as if the timer timed out and restart it again if it is
1694 repeating. The exact semantics are:
1696 If the timer is pending, its pending status is cleared.
1698 If the timer is started but non-repeating, stop it (as if it timed out).
1700 If the timer is repeating, either start it if necessary (with the
1701 C<repeat> value), or reset the running timer to the C<repeat> value.
1703 This sounds a bit complicated, see L<Be smart about timeouts>, above, for a
1704 usage example.
1706 =item ev_tstamp repeat [read-write]
1708 The current C<repeat> value. Will be used each time the watcher times out
1709 or C<ev_timer_again> is called, and determines the next timeout (if any),
1710 which is also when any modifications are taken into account.
1712 =back
1714 =head3 Examples
1716 Example: Create a timer that fires after 60 seconds.
1718 static void
1719 one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
1720 {
1721 .. one minute over, w is actually stopped right here
1722 }
1724 ev_timer mytimer;
1725 ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
1726 ev_timer_start (loop, &mytimer);
1728 Example: Create a timeout timer that times out after 10 seconds of
1729 inactivity.
1731 static void
1732 timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
1733 {
1734 .. ten seconds without any activity
1735 }
1737 ev_timer mytimer;
1738 ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
1739 ev_timer_again (&mytimer); /* start timer */
1740 ev_loop (loop, 0);
1742 // and in some piece of code that gets executed on any "activity":
1743 // reset the timeout to start ticking again at 10 seconds
1744 ev_timer_again (&mytimer);
1747 =head2 C<ev_periodic> - to cron or not to cron?
1749 Periodic watchers are also timers of a kind, but they are very versatile
1750 (and unfortunately a bit complex).
1752 Unlike C<ev_timer>, periodic watchers are not based on real time (or
1753 relative time, the physical time that passes) but on wall clock time
1754 (absolute time, the thing you can read on your calender or clock). The
1755 difference is that wall clock time can run faster or slower than real
1756 time, and time jumps are not uncommon (e.g. when you adjust your
1757 wrist-watch).
1759 You can tell a periodic watcher to trigger after some specific point
1760 in time: for example, if you tell a periodic watcher to trigger "in 10
1761 seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
1762 not a delay) and then reset your system clock to January of the previous
1763 year, then it will take a year or more to trigger the event (unlike an
1764 C<ev_timer>, which would still trigger roughly 10 seconds after starting
1765 it, as it uses a relative timeout).
1767 C<ev_periodic> watchers can also be used to implement vastly more complex
1768 timers, such as triggering an event on each "midnight, local time", or
1769 other complicated rules. This cannot be done with C<ev_timer> watchers, as
1770 those cannot react to time jumps.
1772 As with timers, the callback is guaranteed to be invoked only when the
1773 point in time where it is supposed to trigger has passed. If multiple
1774 timers become ready during the same loop iteration then the ones with
1775 earlier time-out values are invoked before ones with later time-out values
1776 (but this is no longer true when a callback calls C<ev_loop> recursively).
1778 =head3 Watcher-Specific Functions and Data Members
1780 =over 4
1782 =item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
1784 =item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
1786 Lots of arguments, let's sort it out... There are basically three modes of
1787 operation, and we will explain them from simplest to most complex:
1789 =over 4
1791 =item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
1793 In this configuration the watcher triggers an event after the wall clock
1794 time C<offset> has passed. It will not repeat and will not adjust when a
1795 time jump occurs, that is, if it is to be run at January 1st 2011 then it
1796 will be stopped and invoked when the system clock reaches or surpasses
1797 this point in time.
1799 =item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
1801 In this mode the watcher will always be scheduled to time out at the next
1802 C<offset + N * interval> time (for some integer N, which can also be
1803 negative) and then repeat, regardless of any time jumps. The C<offset>
1804 argument is merely an offset into the C<interval> periods.
1806 This can be used to create timers that do not drift with respect to the
1807 system clock, for example, here is an C<ev_periodic> that triggers each
1808 hour, on the hour (with respect to UTC):
1810 ev_periodic_set (&periodic, 0., 3600., 0);
1812 This doesn't mean there will always be 3600 seconds in between triggers,
1813 but only that the callback will be called when the system time shows a
1814 full hour (UTC), or more correctly, when the system time is evenly divisible
1815 by 3600.
1817 Another way to think about it (for the mathematically inclined) is that
1818 C<ev_periodic> will try to run the callback in this mode at the next possible
1819 time where C<time = offset (mod interval)>, regardless of any time jumps.
1821 For numerical stability it is preferable that the C<offset> value is near
1822 C<ev_now ()> (the current time), but there is no range requirement for
1823 this value, and in fact is often specified as zero.
1825 Note also that there is an upper limit to how often a timer can fire (CPU
1826 speed for example), so if C<interval> is very small then timing stability
1827 will of course deteriorate. Libev itself tries to be exact to be about one
1828 millisecond (if the OS supports it and the machine is fast enough).
1830 =item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
1832 In this mode the values for C<interval> and C<offset> are both being
1833 ignored. Instead, each time the periodic watcher gets scheduled, the
1834 reschedule callback will be called with the watcher as first, and the
1835 current time as second argument.
1837 NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, ever,
1838 or make ANY other event loop modifications whatsoever, unless explicitly
1839 allowed by documentation here>.
1841 If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
1842 it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
1843 only event loop modification you are allowed to do).
1845 The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
1846 *w, ev_tstamp now)>, e.g.:
1848 static ev_tstamp
1849 my_rescheduler (ev_periodic *w, ev_tstamp now)
1850 {
1851 return now + 60.;
1852 }
1854 It must return the next time to trigger, based on the passed time value
1855 (that is, the lowest time value larger than to the second argument). It
1856 will usually be called just before the callback will be triggered, but
1857 might be called at other times, too.
1859 NOTE: I<< This callback must always return a time that is higher than or
1860 equal to the passed C<now> value >>.
1862 This can be used to create very complex timers, such as a timer that
1863 triggers on "next midnight, local time". To do this, you would calculate the
1864 next midnight after C<now> and return the timestamp value for this. How
1865 you do this is, again, up to you (but it is not trivial, which is the main
1866 reason I omitted it as an example).
1868 =back
1870 =item ev_periodic_again (loop, ev_periodic *)
1872 Simply stops and restarts the periodic watcher again. This is only useful
1873 when you changed some parameters or the reschedule callback would return
1874 a different time than the last time it was called (e.g. in a crond like
1875 program when the crontabs have changed).
1877 =item ev_tstamp ev_periodic_at (ev_periodic *)
1879 When active, returns the absolute time that the watcher is supposed
1880 to trigger next. This is not the same as the C<offset> argument to
1881 C<ev_periodic_set>, but indeed works even in interval and manual
1882 rescheduling modes.
1884 =item ev_tstamp offset [read-write]
1886 When repeating, this contains the offset value, otherwise this is the
1887 absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
1888 although libev might modify this value for better numerical stability).
1890 Can be modified any time, but changes only take effect when the periodic
1891 timer fires or C<ev_periodic_again> is being called.
1893 =item ev_tstamp interval [read-write]
1895 The current interval value. Can be modified any time, but changes only
1896 take effect when the periodic timer fires or C<ev_periodic_again> is being
1897 called.
1899 =item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
1901 The current reschedule callback, or C<0>, if this functionality is
1902 switched off. Can be changed any time, but changes only take effect when
1903 the periodic timer fires or C<ev_periodic_again> is being called.
1905 =back
1907 =head3 Examples
1909 Example: Call a callback every hour, or, more precisely, whenever the
1910 system time is divisible by 3600. The callback invocation times have
1911 potentially a lot of jitter, but good long-term stability.
1913 static void
1914 clock_cb (struct ev_loop *loop, ev_io *w, int revents)
1915 {
1916 ... its now a full hour (UTC, or TAI or whatever your clock follows)
1917 }
1919 ev_periodic hourly_tick;
1920 ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
1921 ev_periodic_start (loop, &hourly_tick);
1923 Example: The same as above, but use a reschedule callback to do it:
1925 #include <math.h>
1927 static ev_tstamp
1928 my_scheduler_cb (ev_periodic *w, ev_tstamp now)
1929 {
1930 return now + (3600. - fmod (now, 3600.));
1931 }
1933 ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
1935 Example: Call a callback every hour, starting now:
1937 ev_periodic hourly_tick;
1938 ev_periodic_init (&hourly_tick, clock_cb,
1939 fmod (ev_now (loop), 3600.), 3600., 0);
1940 ev_periodic_start (loop, &hourly_tick);
1943 =head2 C<ev_signal> - signal me when a signal gets signalled!
1945 Signal watchers will trigger an event when the process receives a specific
1946 signal one or more times. Even though signals are very asynchronous, libev
1947 will try it's best to deliver signals synchronously, i.e. as part of the
1948 normal event processing, like any other event.
1950 If you want signals asynchronously, just use C<sigaction> as you would
1951 do without libev and forget about sharing the signal. You can even use
1952 C<ev_async> from a signal handler to synchronously wake up an event loop.
1954 You can configure as many watchers as you like per signal. Only when the
1955 first watcher gets started will libev actually register a signal handler
1956 with the kernel (thus it coexists with your own signal handlers as long as
1957 you don't register any with libev for the same signal). Similarly, when
1958 the last signal watcher for a signal is stopped, libev will reset the
1959 signal handler to SIG_DFL (regardless of what it was set to before).
1961 If possible and supported, libev will install its handlers with
1962 C<SA_RESTART> behaviour enabled, so system calls should not be unduly
1963 interrupted. If you have a problem with system calls getting interrupted by
1964 signals you can block all signals in an C<ev_check> watcher and unblock
1965 them in an C<ev_prepare> watcher.
1967 =head3 Watcher-Specific Functions and Data Members
1969 =over 4
1971 =item ev_signal_init (ev_signal *, callback, int signum)
1973 =item ev_signal_set (ev_signal *, int signum)
1975 Configures the watcher to trigger on the given signal number (usually one
1976 of the C<SIGxxx> constants).
1978 =item int signum [read-only]
1980 The signal the watcher watches out for.
1982 =back
1984 =head3 Examples
1986 Example: Try to exit cleanly on SIGINT.
1988 static void
1989 sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
1990 {
1991 ev_unloop (loop, EVUNLOOP_ALL);
1992 }
1994 ev_signal signal_watcher;
1995 ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
1996 ev_signal_start (loop, &signal_watcher);
1999 =head2 C<ev_child> - watch out for process status changes
2001 Child watchers trigger when your process receives a SIGCHLD in response to
2002 some child status changes (most typically when a child of yours dies or
2003 exits). It is permissible to install a child watcher I<after> the child
2004 has been forked (which implies it might have already exited), as long
2005 as the event loop isn't entered (or is continued from a watcher), i.e.,
2006 forking and then immediately registering a watcher for the child is fine,
2007 but forking and registering a watcher a few event loop iterations later is
2008 not.
2010 Only the default event loop is capable of handling signals, and therefore
2011 you can only register child watchers in the default event loop.
2013 =head3 Process Interaction
2015 Libev grabs C<SIGCHLD> as soon as the default event loop is
2016 initialised. This is necessary to guarantee proper behaviour even if
2017 the first child watcher is started after the child exits. The occurrence
2018 of C<SIGCHLD> is recorded asynchronously, but child reaping is done
2019 synchronously as part of the event loop processing. Libev always reaps all
2020 children, even ones not watched.
2022 =head3 Overriding the Built-In Processing
2024 Libev offers no special support for overriding the built-in child
2025 processing, but if your application collides with libev's default child
2026 handler, you can override it easily by installing your own handler for
2027 C<SIGCHLD> after initialising the default loop, and making sure the
2028 default loop never gets destroyed. You are encouraged, however, to use an
2029 event-based approach to child reaping and thus use libev's support for
2030 that, so other libev users can use C<ev_child> watchers freely.
2032 =head3 Stopping the Child Watcher
2034 Currently, the child watcher never gets stopped, even when the
2035 child terminates, so normally one needs to stop the watcher in the
2036 callback. Future versions of libev might stop the watcher automatically
2037 when a child exit is detected.
2039 =head3 Watcher-Specific Functions and Data Members
2041 =over 4
2043 =item ev_child_init (ev_child *, callback, int pid, int trace)
2045 =item ev_child_set (ev_child *, int pid, int trace)
2047 Configures the watcher to wait for status changes of process C<pid> (or
2048 I<any> process if C<pid> is specified as C<0>). The callback can look
2049 at the C<rstatus> member of the C<ev_child> watcher structure to see
2050 the status word (use the macros from C<sys/wait.h> and see your systems
2051 C<waitpid> documentation). The C<rpid> member contains the pid of the
2052 process causing the status change. C<trace> must be either C<0> (only
2053 activate the watcher when the process terminates) or C<1> (additionally
2054 activate the watcher when the process is stopped or continued).
2056 =item int pid [read-only]
2058 The process id this watcher watches out for, or C<0>, meaning any process id.
2060 =item int rpid [read-write]
2062 The process id that detected a status change.
2064 =item int rstatus [read-write]
2066 The process exit/trace status caused by C<rpid> (see your systems
2067 C<waitpid> and C<sys/wait.h> documentation for details).
2069 =back
2071 =head3 Examples
2073 Example: C<fork()> a new process and install a child handler to wait for
2074 its completion.
2076 ev_child cw;
2078 static void
2079 child_cb (EV_P_ ev_child *w, int revents)
2080 {
2081 ev_child_stop (EV_A_ w);
2082 printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
2083 }
2085 pid_t pid = fork ();
2087 if (pid < 0)
2088 // error
2089 else if (pid == 0)
2090 {
2091 // the forked child executes here
2092 exit (1);
2093 }
2094 else
2095 {
2096 ev_child_init (&cw, child_cb, pid, 0);
2097 ev_child_start (EV_DEFAULT_ &cw);
2098 }
2101 =head2 C<ev_stat> - did the file attributes just change?
2103 This watches a file system path for attribute changes. That is, it calls
2104 C<stat> on that path in regular intervals (or when the OS says it changed)
2105 and sees if it changed compared to the last time, invoking the callback if
2106 it did.
2108 The path does not need to exist: changing from "path exists" to "path does
2109 not exist" is a status change like any other. The condition "path does not
2110 exist" (or more correctly "path cannot be stat'ed") is signified by the
2111 C<st_nlink> field being zero (which is otherwise always forced to be at
2112 least one) and all the other fields of the stat buffer having unspecified
2113 contents.
2115 The path I<must not> end in a slash or contain special components such as
2116 C<.> or C<..>. The path I<should> be absolute: If it is relative and
2117 your working directory changes, then the behaviour is undefined.
2119 Since there is no portable change notification interface available, the
2120 portable implementation simply calls C<stat(2)> regularly on the path
2121 to see if it changed somehow. You can specify a recommended polling
2122 interval for this case. If you specify a polling interval of C<0> (highly
2123 recommended!) then a I<suitable, unspecified default> value will be used
2124 (which you can expect to be around five seconds, although this might
2125 change dynamically). Libev will also impose a minimum interval which is
2126 currently around C<0.1>, but that's usually overkill.
2128 This watcher type is not meant for massive numbers of stat watchers,
2129 as even with OS-supported change notifications, this can be
2130 resource-intensive.
2132 At the time of this writing, the only OS-specific interface implemented
2133 is the Linux inotify interface (implementing kqueue support is left as an
2134 exercise for the reader. Note, however, that the author sees no way of
2135 implementing C<ev_stat> semantics with kqueue, except as a hint).
2137 =head3 ABI Issues (Largefile Support)
2139 Libev by default (unless the user overrides this) uses the default
2140 compilation environment, which means that on systems with large file
2141 support disabled by default, you get the 32 bit version of the stat
2142 structure. When using the library from programs that change the ABI to
2143 use 64 bit file offsets the programs will fail. In that case you have to
2144 compile libev with the same flags to get binary compatibility. This is
2145 obviously the case with any flags that change the ABI, but the problem is
2146 most noticeably displayed with ev_stat and large file support.
2148 The solution for this is to lobby your distribution maker to make large
2149 file interfaces available by default (as e.g. FreeBSD does) and not
2150 optional. Libev cannot simply switch on large file support because it has
2151 to exchange stat structures with application programs compiled using the
2152 default compilation environment.
2154 =head3 Inotify and Kqueue
2156 When C<inotify (7)> support has been compiled into libev and present at
2157 runtime, it will be used to speed up change detection where possible. The
2158 inotify descriptor will be created lazily when the first C<ev_stat>
2159 watcher is being started.
2161 Inotify presence does not change the semantics of C<ev_stat> watchers
2162 except that changes might be detected earlier, and in some cases, to avoid
2163 making regular C<stat> calls. Even in the presence of inotify support
2164 there are many cases where libev has to resort to regular C<stat> polling,
2165 but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
2166 many bugs), the path exists (i.e. stat succeeds), and the path resides on
2167 a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
2168 xfs are fully working) libev usually gets away without polling.
2170 There is no support for kqueue, as apparently it cannot be used to
2171 implement this functionality, due to the requirement of having a file
2172 descriptor open on the object at all times, and detecting renames, unlinks
2173 etc. is difficult.
2175 =head3 C<stat ()> is a synchronous operation
2177 Libev doesn't normally do any kind of I/O itself, and so is not blocking
2178 the process. The exception are C<ev_stat> watchers - those call C<stat
2179 ()>, which is a synchronous operation.
2181 For local paths, this usually doesn't matter: unless the system is very
2182 busy or the intervals between stat's are large, a stat call will be fast,
2183 as the path data is usually in memory already (except when starting the
2184 watcher).
2186 For networked file systems, calling C<stat ()> can block an indefinite
2187 time due to network issues, and even under good conditions, a stat call
2188 often takes multiple milliseconds.
2190 Therefore, it is best to avoid using C<ev_stat> watchers on networked
2191 paths, although this is fully supported by libev.
2193 =head3 The special problem of stat time resolution
2195 The C<stat ()> system call only supports full-second resolution portably,
2196 and even on systems where the resolution is higher, most file systems
2197 still only support whole seconds.
2199 That means that, if the time is the only thing that changes, you can
2200 easily miss updates: on the first update, C<ev_stat> detects a change and
2201 calls your callback, which does something. When there is another update
2202 within the same second, C<ev_stat> will be unable to detect unless the
2203 stat data does change in other ways (e.g. file size).
2205 The solution to this is to delay acting on a change for slightly more
2206 than a second (or till slightly after the next full second boundary), using
2207 a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
2208 ev_timer_again (loop, w)>).
2210 The C<.02> offset is added to work around small timing inconsistencies
2211 of some operating systems (where the second counter of the current time
2212 might be be delayed. One such system is the Linux kernel, where a call to
2213 C<gettimeofday> might return a timestamp with a full second later than
2214 a subsequent C<time> call - if the equivalent of C<time ()> is used to
2215 update file times then there will be a small window where the kernel uses
2216 the previous second to update file times but libev might already execute
2217 the timer callback).
2219 =head3 Watcher-Specific Functions and Data Members
2221 =over 4
2223 =item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
2225 =item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
2227 Configures the watcher to wait for status changes of the given
2228 C<path>. The C<interval> is a hint on how quickly a change is expected to
2229 be detected and should normally be specified as C<0> to let libev choose
2230 a suitable value. The memory pointed to by C<path> must point to the same
2231 path for as long as the watcher is active.
2233 The callback will receive an C<EV_STAT> event when a change was detected,
2234 relative to the attributes at the time the watcher was started (or the
2235 last change was detected).
2237 =item ev_stat_stat (loop, ev_stat *)
2239 Updates the stat buffer immediately with new values. If you change the
2240 watched path in your callback, you could call this function to avoid
2241 detecting this change (while introducing a race condition if you are not
2242 the only one changing the path). Can also be useful simply to find out the
2243 new values.
2245 =item ev_statdata attr [read-only]
2247 The most-recently detected attributes of the file. Although the type is
2248 C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
2249 suitable for your system, but you can only rely on the POSIX-standardised
2250 members to be present. If the C<st_nlink> member is C<0>, then there was
2251 some error while C<stat>ing the file.
2253 =item ev_statdata prev [read-only]
2255 The previous attributes of the file. The callback gets invoked whenever
2256 C<prev> != C<attr>, or, more precisely, one or more of these members
2257 differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
2258 C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
2260 =item ev_tstamp interval [read-only]
2262 The specified interval.
2264 =item const char *path [read-only]
2266 The file system path that is being watched.
2268 =back
2270 =head3 Examples
2272 Example: Watch C</etc/passwd> for attribute changes.
2274 static void
2275 passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
2276 {
2277 /* /etc/passwd changed in some way */
2278 if (w->attr.st_nlink)
2279 {
2280 printf ("passwd current size %ld\n", (long)w->attr.st_size);
2281 printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
2282 printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
2283 }
2284 else
2285 /* you shalt not abuse printf for puts */
2286 puts ("wow, /etc/passwd is not there, expect problems. "
2287 "if this is windows, they already arrived\n");
2288 }
2290 ...
2291 ev_stat passwd;
2293 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
2294 ev_stat_start (loop, &passwd);
2296 Example: Like above, but additionally use a one-second delay so we do not
2297 miss updates (however, frequent updates will delay processing, too, so
2298 one might do the work both on C<ev_stat> callback invocation I<and> on
2299 C<ev_timer> callback invocation).
2301 static ev_stat passwd;
2302 static ev_timer timer;
2304 static void
2305 timer_cb (EV_P_ ev_timer *w, int revents)
2306 {
2307 ev_timer_stop (EV_A_ w);
2309 /* now it's one second after the most recent passwd change */
2310 }
2312 static void
2313 stat_cb (EV_P_ ev_stat *w, int revents)
2314 {
2315 /* reset the one-second timer */
2316 ev_timer_again (EV_A_ &timer);
2317 }
2319 ...
2320 ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
2321 ev_stat_start (loop, &passwd);
2322 ev_timer_init (&timer, timer_cb, 0., 1.02);
2325 =head2 C<ev_idle> - when you've got nothing better to do...
2327 Idle watchers trigger events when no other events of the same or higher
2328 priority are pending (prepare, check and other idle watchers do not count
2329 as receiving "events").
2331 That is, as long as your process is busy handling sockets or timeouts
2332 (or even signals, imagine) of the same or higher priority it will not be
2333 triggered. But when your process is idle (or only lower-priority watchers
2334 are pending), the idle watchers are being called once per event loop
2335 iteration - until stopped, that is, or your process receives more events
2336 and becomes busy again with higher priority stuff.
2338 The most noteworthy effect is that as long as any idle watchers are
2339 active, the process will not block when waiting for new events.
2341 Apart from keeping your process non-blocking (which is a useful
2342 effect on its own sometimes), idle watchers are a good place to do
2343 "pseudo-background processing", or delay processing stuff to after the
2344 event loop has handled all outstanding events.
2346 =head3 Watcher-Specific Functions and Data Members
2348 =over 4
2350 =item ev_idle_init (ev_idle *, callback)
2352 Initialises and configures the idle watcher - it has no parameters of any
2353 kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
2354 believe me.
2356 =back
2358 =head3 Examples
2360 Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
2361 callback, free it. Also, use no error checking, as usual.
2363 static void
2364 idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
2365 {
2366 free (w);
2367 // now do something you wanted to do when the program has
2368 // no longer anything immediate to do.
2369 }
2371 ev_idle *idle_watcher = malloc (sizeof (ev_idle));
2372 ev_idle_init (idle_watcher, idle_cb);
2373 ev_idle_start (loop, idle_cb);
2376 =head2 C<ev_prepare> and C<ev_check> - customise your event loop!
2378 Prepare and check watchers are usually (but not always) used in pairs:
2379 prepare watchers get invoked before the process blocks and check watchers
2380 afterwards.
2382 You I<must not> call C<ev_loop> or similar functions that enter
2383 the current event loop from either C<ev_prepare> or C<ev_check>
2384 watchers. Other loops than the current one are fine, however. The
2385 rationale behind this is that you do not need to check for recursion in
2386 those watchers, i.e. the sequence will always be C<ev_prepare>, blocking,
2387 C<ev_check> so if you have one watcher of each kind they will always be
2388 called in pairs bracketing the blocking call.
2390 Their main purpose is to integrate other event mechanisms into libev and
2391 their use is somewhat advanced. They could be used, for example, to track
2392 variable changes, implement your own watchers, integrate net-snmp or a
2393 coroutine library and lots more. They are also occasionally useful if
2394 you cache some data and want to flush it before blocking (for example,
2395 in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
2396 watcher).
2398 This is done by examining in each prepare call which file descriptors
2399 need to be watched by the other library, registering C<ev_io> watchers
2400 for them and starting an C<ev_timer> watcher for any timeouts (many
2401 libraries provide exactly this functionality). Then, in the check watcher,
2402 you check for any events that occurred (by checking the pending status
2403 of all watchers and stopping them) and call back into the library. The
2404 I/O and timer callbacks will never actually be called (but must be valid
2405 nevertheless, because you never know, you know?).
2407 As another example, the Perl Coro module uses these hooks to integrate
2408 coroutines into libev programs, by yielding to other active coroutines
2409 during each prepare and only letting the process block if no coroutines
2410 are ready to run (it's actually more complicated: it only runs coroutines
2411 with priority higher than or equal to the event loop and one coroutine
2412 of lower priority, but only once, using idle watchers to keep the event
2413 loop from blocking if lower-priority coroutines are active, thus mapping
2414 low-priority coroutines to idle/background tasks).
2416 It is recommended to give C<ev_check> watchers highest (C<EV_MAXPRI>)
2417 priority, to ensure that they are being run before any other watchers
2418 after the poll (this doesn't matter for C<ev_prepare> watchers).
2420 Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
2421 activate ("feed") events into libev. While libev fully supports this, they
2422 might get executed before other C<ev_check> watchers did their job. As
2423 C<ev_check> watchers are often used to embed other (non-libev) event
2424 loops those other event loops might be in an unusable state until their
2425 C<ev_check> watcher ran (always remind yourself to coexist peacefully with
2426 others).
2428 =head3 Watcher-Specific Functions and Data Members
2430 =over 4
2432 =item ev_prepare_init (ev_prepare *, callback)
2434 =item ev_check_init (ev_check *, callback)
2436 Initialises and configures the prepare or check watcher - they have no
2437 parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
2438 macros, but using them is utterly, utterly, utterly and completely
2439 pointless.
2441 =back
2443 =head3 Examples
2445 There are a number of principal ways to embed other event loops or modules
2446 into libev. Here are some ideas on how to include libadns into libev
2447 (there is a Perl module named C<EV::ADNS> that does this, which you could
2448 use as a working example. Another Perl module named C<EV::Glib> embeds a
2449 Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
2450 Glib event loop).
2452 Method 1: Add IO watchers and a timeout watcher in a prepare handler,
2453 and in a check watcher, destroy them and call into libadns. What follows
2454 is pseudo-code only of course. This requires you to either use a low
2455 priority for the check watcher or use C<ev_clear_pending> explicitly, as
2456 the callbacks for the IO/timeout watchers might not have been called yet.
2458 static ev_io iow [nfd];
2459 static ev_timer tw;
2461 static void
2462 io_cb (struct ev_loop *loop, ev_io *w, int revents)
2463 {
2464 }
2466 // create io watchers for each fd and a timer before blocking
2467 static void
2468 adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
2469 {
2470 int timeout = 3600000;
2471 struct pollfd fds [nfd];
2472 // actual code will need to loop here and realloc etc.
2473 adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
2475 /* the callback is illegal, but won't be called as we stop during check */
2476 ev_timer_init (&tw, 0, timeout * 1e-3);
2477 ev_timer_start (loop, &tw);
2479 // create one ev_io per pollfd
2480 for (int i = 0; i < nfd; ++i)
2481 {
2482 ev_io_init (iow + i, io_cb, fds [i].fd,
2483 ((fds [i].events & POLLIN ? EV_READ : 0)
2484 | (fds [i].events & POLLOUT ? EV_WRITE : 0)));
2486 fds [i].revents = 0;
2487 ev_io_start (loop, iow + i);
2488 }
2489 }
2491 // stop all watchers after blocking
2492 static void
2493 adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
2494 {
2495 ev_timer_stop (loop, &tw);
2497 for (int i = 0; i < nfd; ++i)
2498 {
2499 // set the relevant poll flags
2500 // could also call adns_processreadable etc. here
2501 struct pollfd *fd = fds + i;
2502 int revents = ev_clear_pending (iow + i);
2503 if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
2504 if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
2506 // now stop the watcher
2507 ev_io_stop (loop, iow + i);
2508 }
2510 adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
2511 }
2513 Method 2: This would be just like method 1, but you run C<adns_afterpoll>
2514 in the prepare watcher and would dispose of the check watcher.
2516 Method 3: If the module to be embedded supports explicit event
2517 notification (libadns does), you can also make use of the actual watcher
2518 callbacks, and only destroy/create the watchers in the prepare watcher.
2520 static void
2521 timer_cb (EV_P_ ev_timer *w, int revents)
2522 {
2523 adns_state ads = (adns_state)w->data;
2524 update_now (EV_A);
2526 adns_processtimeouts (ads, &tv_now);
2527 }
2529 static void
2530 io_cb (EV_P_ ev_io *w, int revents)
2531 {
2532 adns_state ads = (adns_state)w->data;
2533 update_now (EV_A);
2535 if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
2536 if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
2537 }
2539 // do not ever call adns_afterpoll
2541 Method 4: Do not use a prepare or check watcher because the module you
2542 want to embed is not flexible enough to support it. Instead, you can
2543 override their poll function. The drawback with this solution is that the
2544 main loop is now no longer controllable by EV. The C<Glib::EV> module uses
2545 this approach, effectively embedding EV as a client into the horrible
2546 libglib event loop.
2548 static gint
2549 event_poll_func (GPollFD *fds, guint nfds, gint timeout)
2550 {
2551 int got_events = 0;
2553 for (n = 0; n < nfds; ++n)
2554 // create/start io watcher that sets the relevant bits in fds[n] and increment got_events
2556 if (timeout >= 0)
2557 // create/start timer
2559 // poll
2560 ev_loop (EV_A_ 0);
2562 // stop timer again
2563 if (timeout >= 0)
2564 ev_timer_stop (EV_A_ &to);
2566 // stop io watchers again - their callbacks should have set
2567 for (n = 0; n < nfds; ++n)
2568 ev_io_stop (EV_A_ iow [n]);
2570 return got_events;
2571 }
2574 =head2 C<ev_embed> - when one backend isn't enough...
2576 This is a rather advanced watcher type that lets you embed one event loop
2577 into another (currently only C<ev_io> events are supported in the embedded
2578 loop, other types of watchers might be handled in a delayed or incorrect
2579 fashion and must not be used).
2581 There are primarily two reasons you would want that: work around bugs and
2582 prioritise I/O.
2584 As an example for a bug workaround, the kqueue backend might only support
2585 sockets on some platform, so it is unusable as generic backend, but you
2586 still want to make use of it because you have many sockets and it scales
2587 so nicely. In this case, you would create a kqueue-based loop and embed
2588 it into your default loop (which might use e.g. poll). Overall operation
2589 will be a bit slower because first libev has to call C<poll> and then
2590 C<kevent>, but at least you can use both mechanisms for what they are
2591 best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
2593 As for prioritising I/O: under rare circumstances you have the case where
2594 some fds have to be watched and handled very quickly (with low latency),
2595 and even priorities and idle watchers might have too much overhead. In
2596 this case you would put all the high priority stuff in one loop and all
2597 the rest in a second one, and embed the second one in the first.
2599 As long as the watcher is active, the callback will be invoked every
2600 time there might be events pending in the embedded loop. The callback
2601 must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
2602 sweep and invoke their callbacks (the callback doesn't need to invoke the
2603 C<ev_embed_sweep> function directly, it could also start an idle watcher
2604 to give the embedded loop strictly lower priority for example).
2606 You can also set the callback to C<0>, in which case the embed watcher
2607 will automatically execute the embedded loop sweep whenever necessary.
2609 Fork detection will be handled transparently while the C<ev_embed> watcher
2610 is active, i.e., the embedded loop will automatically be forked when the
2611 embedding loop forks. In other cases, the user is responsible for calling
2612 C<ev_loop_fork> on the embedded loop.
2614 Unfortunately, not all backends are embeddable: only the ones returned by
2615 C<ev_embeddable_backends> are, which, unfortunately, does not include any
2616 portable one.
2618 So when you want to use this feature you will always have to be prepared
2619 that you cannot get an embeddable loop. The recommended way to get around
2620 this is to have a separate variables for your embeddable loop, try to
2621 create it, and if that fails, use the normal loop for everything.
2623 =head3 C<ev_embed> and fork
2625 While the C<ev_embed> watcher is running, forks in the embedding loop will
2626 automatically be applied to the embedded loop as well, so no special
2627 fork handling is required in that case. When the watcher is not running,
2628 however, it is still the task of the libev user to call C<ev_loop_fork ()>
2629 as applicable.
2631 =head3 Watcher-Specific Functions and Data Members
2633 =over 4
2635 =item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
2637 =item ev_embed_set (ev_embed *, callback, struct ev_loop *embedded_loop)
2639 Configures the watcher to embed the given loop, which must be
2640 embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
2641 invoked automatically, otherwise it is the responsibility of the callback
2642 to invoke it (it will continue to be called until the sweep has been done,
2643 if you do not want that, you need to temporarily stop the embed watcher).
2645 =item ev_embed_sweep (loop, ev_embed *)
2647 Make a single, non-blocking sweep over the embedded loop. This works
2648 similarly to C<ev_loop (embedded_loop, EVLOOP_NONBLOCK)>, but in the most
2649 appropriate way for embedded loops.
2651 =item struct ev_loop *other [read-only]
2653 The embedded event loop.
2655 =back
2657 =head3 Examples
2659 Example: Try to get an embeddable event loop and embed it into the default
2660 event loop. If that is not possible, use the default loop. The default
2661 loop is stored in C<loop_hi>, while the embeddable loop is stored in
2662 C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
2663 used).
2665 struct ev_loop *loop_hi = ev_default_init (0);
2666 struct ev_loop *loop_lo = 0;
2667 ev_embed embed;
2669 // see if there is a chance of getting one that works
2670 // (remember that a flags value of 0 means autodetection)
2671 loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
2672 ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
2673 : 0;
2675 // if we got one, then embed it, otherwise default to loop_hi
2676 if (loop_lo)
2677 {
2678 ev_embed_init (&embed, 0, loop_lo);
2679 ev_embed_start (loop_hi, &embed);
2680 }
2681 else
2682 loop_lo = loop_hi;
2684 Example: Check if kqueue is available but not recommended and create
2685 a kqueue backend for use with sockets (which usually work with any
2686 kqueue implementation). Store the kqueue/socket-only event loop in
2687 C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
2689 struct ev_loop *loop = ev_default_init (0);
2690 struct ev_loop *loop_socket = 0;
2691 ev_embed embed;
2693 if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
2694 if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
2695 {
2696 ev_embed_init (&embed, 0, loop_socket);
2697 ev_embed_start (loop, &embed);
2698 }
2700 if (!loop_socket)
2701 loop_socket = loop;
2703 // now use loop_socket for all sockets, and loop for everything else
2706 =head2 C<ev_fork> - the audacity to resume the event loop after a fork
2708 Fork watchers are called when a C<fork ()> was detected (usually because
2709 whoever is a good citizen cared to tell libev about it by calling
2710 C<ev_default_fork> or C<ev_loop_fork>). The invocation is done before the
2711 event loop blocks next and before C<ev_check> watchers are being called,
2712 and only in the child after the fork. If whoever good citizen calling
2713 C<ev_default_fork> cheats and calls it in the wrong process, the fork
2714 handlers will be invoked, too, of course.
2716 =head3 The special problem of life after fork - how is it possible?
2718 Most uses of C<fork()> consist of forking, then some simple calls to ste
2719 up/change the process environment, followed by a call to C<exec()>. This
2720 sequence should be handled by libev without any problems.
2722 This changes when the application actually wants to do event handling
2723 in the child, or both parent in child, in effect "continuing" after the
2724 fork.
2726 The default mode of operation (for libev, with application help to detect
2727 forks) is to duplicate all the state in the child, as would be expected
2728 when I<either> the parent I<or> the child process continues.
2730 When both processes want to continue using libev, then this is usually the
2731 wrong result. In that case, usually one process (typically the parent) is
2732 supposed to continue with all watchers in place as before, while the other
2733 process typically wants to start fresh, i.e. without any active watchers.
2735 The cleanest and most efficient way to achieve that with libev is to
2736 simply create a new event loop, which of course will be "empty", and
2737 use that for new watchers. This has the advantage of not touching more
2738 memory than necessary, and thus avoiding the copy-on-write, and the
2739 disadvantage of having to use multiple event loops (which do not support
2740 signal watchers).
2742 When this is not possible, or you want to use the default loop for
2743 other reasons, then in the process that wants to start "fresh", call
2744 C<ev_default_destroy ()> followed by C<ev_default_loop (...)>. Destroying
2745 the default loop will "orphan" (not stop) all registered watchers, so you
2746 have to be careful not to execute code that modifies those watchers. Note
2747 also that in that case, you have to re-register any signal watchers.
2749 =head3 Watcher-Specific Functions and Data Members
2751 =over 4
2753 =item ev_fork_init (ev_signal *, callback)
2755 Initialises and configures the fork watcher - it has no parameters of any
2756 kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
2757 believe me.
2759 =back
2762 =head2 C<ev_async> - how to wake up another event loop
2764 In general, you cannot use an C<ev_loop> from multiple threads or other
2765 asynchronous sources such as signal handlers (as opposed to multiple event
2766 loops - those are of course safe to use in different threads).
2768 Sometimes, however, you need to wake up another event loop you do not
2769 control, for example because it belongs to another thread. This is what
2770 C<ev_async> watchers do: as long as the C<ev_async> watcher is active, you
2771 can signal it by calling C<ev_async_send>, which is thread- and signal
2772 safe.
2774 This functionality is very similar to C<ev_signal> watchers, as signals,
2775 too, are asynchronous in nature, and signals, too, will be compressed
2776 (i.e. the number of callback invocations may be less than the number of
2777 C<ev_async_sent> calls).
2779 Unlike C<ev_signal> watchers, C<ev_async> works with any event loop, not
2780 just the default loop.
2782 =head3 Queueing
2784 C<ev_async> does not support queueing of data in any way. The reason
2785 is that the author does not know of a simple (or any) algorithm for a
2786 multiple-writer-single-reader queue that works in all cases and doesn't
2787 need elaborate support such as pthreads.
2789 That means that if you want to queue data, you have to provide your own
2790 queue. But at least I can tell you how to implement locking around your
2791 queue:
2793 =over 4
2795 =item queueing from a signal handler context
2797 To implement race-free queueing, you simply add to the queue in the signal
2798 handler but you block the signal handler in the watcher callback. Here is
2799 an example that does that for some fictitious SIGUSR1 handler:
2801 static ev_async mysig;
2803 static void
2804 sigusr1_handler (void)
2805 {
2806 sometype data;
2808 // no locking etc.
2809 queue_put (data);
2810 ev_async_send (EV_DEFAULT_ &mysig);
2811 }
2813 static void
2814 mysig_cb (EV_P_ ev_async *w, int revents)
2815 {
2816 sometype data;
2817 sigset_t block, prev;
2819 sigemptyset (&block);
2820 sigaddset (&block, SIGUSR1);
2821 sigprocmask (SIG_BLOCK, &block, &prev);
2823 while (queue_get (&data))
2824 process (data);
2826 if (sigismember (&prev, SIGUSR1)
2827 sigprocmask (SIG_UNBLOCK, &block, 0);
2828 }
2830 (Note: pthreads in theory requires you to use C<pthread_setmask>
2831 instead of C<sigprocmask> when you use threads, but libev doesn't do it
2832 either...).
2834 =item queueing from a thread context
2836 The strategy for threads is different, as you cannot (easily) block
2837 threads but you can easily preempt them, so to queue safely you need to
2838 employ a traditional mutex lock, such as in this pthread example:
2840 static ev_async mysig;
2841 static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
2843 static void
2844 otherthread (void)
2845 {
2846 // only need to lock the actual queueing operation
2847 pthread_mutex_lock (&mymutex);
2848 queue_put (data);
2849 pthread_mutex_unlock (&mymutex);
2851 ev_async_send (EV_DEFAULT_ &mysig);
2852 }
2854 static void
2855 mysig_cb (EV_P_ ev_async *w, int revents)
2856 {
2857 pthread_mutex_lock (&mymutex);
2859 while (queue_get (&data))
2860 process (data);
2862 pthread_mutex_unlock (&mymutex);
2863 }
2865 =back
2868 =head3 Watcher-Specific Functions and Data Members
2870 =over 4
2872 =item ev_async_init (ev_async *, callback)
2874 Initialises and configures the async watcher - it has no parameters of any
2875 kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
2876 trust me.
2878 =item ev_async_send (loop, ev_async *)
2880 Sends/signals/activates the given C<ev_async> watcher, that is, feeds
2881 an C<EV_ASYNC> event on the watcher into the event loop. Unlike
2882 C<ev_feed_event>, this call is safe to do from other threads, signal or
2883 similar contexts (see the discussion of C<EV_ATOMIC_T> in the embedding
2884 section below on what exactly this means).
2886 Note that, as with other watchers in libev, multiple events might get
2887 compressed into a single callback invocation (another way to look at this
2888 is that C<ev_async> watchers are level-triggered, set on C<ev_async_send>,
2889 reset when the event loop detects that).
2891 This call incurs the overhead of a system call only once per event loop
2892 iteration, so while the overhead might be noticeable, it doesn't apply to
2893 repeated calls to C<ev_async_send> for the same event loop.
2895 =item bool = ev_async_pending (ev_async *)
2897 Returns a non-zero value when C<ev_async_send> has been called on the
2898 watcher but the event has not yet been processed (or even noted) by the
2899 event loop.
2901 C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
2902 the loop iterates next and checks for the watcher to have become active,
2903 it will reset the flag again. C<ev_async_pending> can be used to very
2904 quickly check whether invoking the loop might be a good idea.
2906 Not that this does I<not> check whether the watcher itself is pending,
2907 only whether it has been requested to make this watcher pending: there
2908 is a time window between the event loop checking and resetting the async
2909 notification, and the callback being invoked.
2911 =back
2916 There are some other functions of possible interest. Described. Here. Now.
2918 =over 4
2920 =item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)
2922 This function combines a simple timer and an I/O watcher, calls your
2923 callback on whichever event happens first and automatically stops both
2924 watchers. This is useful if you want to wait for a single event on an fd
2925 or timeout without having to allocate/configure/start/stop/free one or
2926 more watchers yourself.
2928 If C<fd> is less than 0, then no I/O watcher will be started and the
2929 C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
2930 the given C<fd> and C<events> set will be created and started.
2932 If C<timeout> is less than 0, then no timeout watcher will be
2933 started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
2934 repeat = 0) will be started. C<0> is a valid timeout.
2936 The callback has the type C<void (*cb)(int revents, void *arg)> and gets
2937 passed an C<revents> set like normal event callbacks (a combination of
2938 C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMEOUT>) and the C<arg>
2939 value passed to C<ev_once>. Note that it is possible to receive I<both>
2940 a timeout and an io event at the same time - you probably should give io
2941 events precedence.
2943 Example: wait up to ten seconds for data to appear on STDIN_FILENO.
2945 static void stdin_ready (int revents, void *arg)
2946 {
2947 if (revents & EV_READ)
2948 /* stdin might have data for us, joy! */;
2949 else if (revents & EV_TIMEOUT)
2950 /* doh, nothing entered */;
2951 }
2953 ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
2955 =item ev_feed_event (struct ev_loop *, watcher *, int revents)
2957 Feeds the given event set into the event loop, as if the specified event
2958 had happened for the specified watcher (which must be a pointer to an
2959 initialised but not necessarily started event watcher).
2961 =item ev_feed_fd_event (struct ev_loop *, int fd, int revents)
2963 Feed an event on the given fd, as if a file descriptor backend detected
2964 the given events it.
2966 =item ev_feed_signal_event (struct ev_loop *loop, int signum)
2968 Feed an event as if the given signal occurred (C<loop> must be the default
2969 loop!).
2971 =back
2976 Libev offers a compatibility emulation layer for libevent. It cannot
2977 emulate the internals of libevent, so here are some usage hints:
2979 =over 4
2981 =item * Use it by including <event.h>, as usual.
2983 =item * The following members are fully supported: ev_base, ev_callback,
2984 ev_arg, ev_fd, ev_res, ev_events.
2986 =item * Avoid using ev_flags and the EVLIST_*-macros, while it is
2987 maintained by libev, it does not work exactly the same way as in libevent (consider
2988 it a private API).
2990 =item * Priorities are not currently supported. Initialising priorities
2991 will fail and all watchers will have the same priority, even though there
2992 is an ev_pri field.
2994 =item * In libevent, the last base created gets the signals, in libev, the
2995 first base created (== the default loop) gets the signals.
2997 =item * Other members are not supported.
2999 =item * The libev emulation is I<not> ABI compatible to libevent, you need
3000 to use the libev header file and library.
3002 =back
3004 =head1 C++ SUPPORT
3006 Libev comes with some simplistic wrapper classes for C++ that mainly allow
3007 you to use some convenience methods to start/stop watchers and also change
3008 the callback model to a model using method callbacks on objects.
3010 To use it,
3012 #include <ev++.h>
3014 This automatically includes F<ev.h> and puts all of its definitions (many
3015 of them macros) into the global namespace. All C++ specific things are
3016 put into the C<ev> namespace. It should support all the same embedding
3017 options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
3019 Care has been taken to keep the overhead low. The only data member the C++
3020 classes add (compared to plain C-style watchers) is the event loop pointer
3021 that the watcher is associated with (or no additional members at all if
3022 you disable C<EV_MULTIPLICITY> when embedding libev).
3024 Currently, functions, and static and non-static member functions can be
3025 used as callbacks. Other types should be easy to add as long as they only
3026 need one additional pointer for context. If you need support for other
3027 types of functors please contact the author (preferably after implementing
3028 it).
3030 Here is a list of things available in the C<ev> namespace:
3032 =over 4
3034 =item C<ev::READ>, C<ev::WRITE> etc.
3036 These are just enum values with the same values as the C<EV_READ> etc.
3037 macros from F<ev.h>.
3039 =item C<ev::tstamp>, C<ev::now>
3041 Aliases to the same types/functions as with the C<ev_> prefix.
3043 =item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
3045 For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
3046 the same name in the C<ev> namespace, with the exception of C<ev_signal>
3047 which is called C<ev::sig> to avoid clashes with the C<signal> macro
3048 defines by many implementations.
3050 All of those classes have these methods:
3052 =over 4
3054 =item ev::TYPE::TYPE ()
3056 =item ev::TYPE::TYPE (struct ev_loop *)
3058 =item ev::TYPE::~TYPE
3060 The constructor (optionally) takes an event loop to associate the watcher
3061 with. If it is omitted, it will use C<EV_DEFAULT>.
3063 The constructor calls C<ev_init> for you, which means you have to call the
3064 C<set> method before starting it.
3066 It will not set a callback, however: You have to call the templated C<set>
3067 method to set a callback before you can start the watcher.
3069 (The reason why you have to use a method is a limitation in C++ which does
3070 not allow explicit template arguments for constructors).
3072 The destructor automatically stops the watcher if it is active.
3074 =item w->set<class, &class::method> (object *)
3076 This method sets the callback method to call. The method has to have a
3077 signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
3078 first argument and the C<revents> as second. The object must be given as
3079 parameter and is stored in the C<data> member of the watcher.
3081 This method synthesizes efficient thunking code to call your method from
3082 the C callback that libev requires. If your compiler can inline your
3083 callback (i.e. it is visible to it at the place of the C<set> call and
3084 your compiler is good :), then the method will be fully inlined into the
3085 thunking function, making it as fast as a direct C callback.
3087 Example: simple class declaration and watcher initialisation
3089 struct myclass
3090 {
3091 void io_cb (ev::io &w, int revents) { }
3092 }
3094 myclass obj;
3095 ev::io iow;
3096 iow.set <myclass, &myclass::io_cb> (&obj);
3098 =item w->set (object *)
3100 This is an B<experimental> feature that might go away in a future version.
3102 This is a variation of a method callback - leaving out the method to call
3103 will default the method to C<operator ()>, which makes it possible to use
3104 functor objects without having to manually specify the C<operator ()> all
3105 the time. Incidentally, you can then also leave out the template argument
3106 list.
3108 The C<operator ()> method prototype must be C<void operator ()(watcher &w,
3109 int revents)>.
3111 See the method-C<set> above for more details.
3113 Example: use a functor object as callback.
3115 struct myfunctor
3116 {
3117 void operator() (ev::io &w, int revents)
3118 {
3119 ...
3120 }
3121 }
3123 myfunctor f;
3125 ev::io w;
3126 w.set (&f);
3128 =item w->set<function> (void *data = 0)
3130 Also sets a callback, but uses a static method or plain function as
3131 callback. The optional C<data> argument will be stored in the watcher's
3132 C<data> member and is free for you to use.
3134 The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
3136 See the method-C<set> above for more details.
3138 Example: Use a plain function as callback.
3140 static void io_cb (ev::io &w, int revents) { }
3141 iow.set <io_cb> ();
3143 =item w->set (struct ev_loop *)
3145 Associates a different C<struct ev_loop> with this watcher. You can only
3146 do this when the watcher is inactive (and not pending either).
3148 =item w->set ([arguments])
3150 Basically the same as C<ev_TYPE_set>, with the same arguments. Must be
3151 called at least once. Unlike the C counterpart, an active watcher gets
3152 automatically stopped and restarted when reconfiguring it with this
3153 method.
3155 =item w->start ()
3157 Starts the watcher. Note that there is no C<loop> argument, as the
3158 constructor already stores the event loop.
3160 =item w->stop ()
3162 Stops the watcher if it is active. Again, no C<loop> argument.
3164 =item w->again () (C<ev::timer>, C<ev::periodic> only)
3166 For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
3167 C<ev_TYPE_again> function.
3169 =item w->sweep () (C<ev::embed> only)
3171 Invokes C<ev_embed_sweep>.
3173 =item w->update () (C<ev::stat> only)
3175 Invokes C<ev_stat_stat>.
3177 =back
3179 =back
3181 Example: Define a class with an IO and idle watcher, start one of them in
3182 the constructor.
3184 class myclass
3185 {
3186 ev::io io ; void io_cb (ev::io &w, int revents);
3187 ev::idle idle; void idle_cb (ev::idle &w, int revents);
3189 myclass (int fd)
3190 {
3191 io .set <myclass, &myclass::io_cb > (this);
3192 idle.set <myclass, &myclass::idle_cb> (this);
3194 io.start (fd, ev::READ);
3195 }
3196 };
3201 Libev does not offer other language bindings itself, but bindings for a
3202 number of languages exist in the form of third-party packages. If you know
3203 any interesting language binding in addition to the ones listed here, drop
3204 me a note.
3206 =over 4
3208 =item Perl
3210 The EV module implements the full libev API and is actually used to test
3211 libev. EV is developed together with libev. Apart from the EV core module,
3212 there are additional modules that implement libev-compatible interfaces
3213 to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
3214 C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
3215 and C<EV::Glib>).
3217 It can be found and installed via CPAN, its homepage is at
3218 L<>.
3220 =item Python
3222 Python bindings can be found at L<>. It
3223 seems to be quite complete and well-documented.
3225 =item Ruby
3227 Tony Arcieri has written a ruby extension that offers access to a subset
3228 of the libev API and adds file handle abstractions, asynchronous DNS and
3229 more on top of it. It can be found via gem servers. Its homepage is at
3230 L<>.
3232 Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
3233 makes rev work even on mingw.
3235 =item Haskell
3237 A haskell binding to libev is available at
3238 L<>.
3240 =item D
3242 Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
3243 be found at L<>.
3245 =item Ocaml
3247 Erkki Seppala has written Ocaml bindings for libev, to be found at
3248 L<>.
3250 =back
3253 =head1 MACRO MAGIC
3255 Libev can be compiled with a variety of options, the most fundamental
3256 of which is C<EV_MULTIPLICITY>. This option determines whether (most)
3257 functions and callbacks have an initial C<struct ev_loop *> argument.
3259 To make it easier to write programs that cope with either variant, the
3260 following macros are defined:
3262 =over 4
3264 =item C<EV_A>, C<EV_A_>
3266 This provides the loop I<argument> for functions, if one is required ("ev
3267 loop argument"). The C<EV_A> form is used when this is the sole argument,
3268 C<EV_A_> is used when other arguments are following. Example:
3270 ev_unref (EV_A);
3271 ev_timer_add (EV_A_ watcher);
3272 ev_loop (EV_A_ 0);
3274 It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
3275 which is often provided by the following macro.
3277 =item C<EV_P>, C<EV_P_>
3279 This provides the loop I<parameter> for functions, if one is required ("ev
3280 loop parameter"). The C<EV_P> form is used when this is the sole parameter,
3281 C<EV_P_> is used when other parameters are following. Example:
3283 // this is how ev_unref is being declared
3284 static void ev_unref (EV_P);
3286 // this is how you can declare your typical callback
3287 static void cb (EV_P_ ev_timer *w, int revents)
3289 It declares a parameter C<loop> of type C<struct ev_loop *>, quite
3290 suitable for use with C<EV_A>.
3292 =item C<EV_DEFAULT>, C<EV_DEFAULT_>
3294 Similar to the other two macros, this gives you the value of the default
3295 loop, if multiple loops are supported ("ev loop default").
3299 Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
3300 default loop has been initialised (C<UC> == unchecked). Their behaviour
3301 is undefined when the default loop has not been initialised by a previous
3302 execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
3304 It is often prudent to use C<EV_DEFAULT> when initialising the first
3305 watcher in a function but use C<EV_DEFAULT_UC> afterwards.
3307 =back
3309 Example: Declare and initialise a check watcher, utilising the above
3310 macros so it will work regardless of whether multiple loops are supported
3311 or not.
3313 static void
3314 check_cb (EV_P_ ev_timer *w, int revents)
3315 {
3316 ev_check_stop (EV_A_ w);
3317 }
3319 ev_check check;
3320 ev_check_init (&check, check_cb);
3321 ev_check_start (EV_DEFAULT_ &check);
3322 ev_loop (EV_DEFAULT_ 0);
3324 =head1 EMBEDDING
3326 Libev can (and often is) directly embedded into host
3327 applications. Examples of applications that embed it include the Deliantra
3328 Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
3329 and rxvt-unicode.
3331 The goal is to enable you to just copy the necessary files into your
3332 source directory without having to change even a single line in them, so
3333 you can easily upgrade by simply copying (or having a checked-out copy of
3334 libev somewhere in your source tree).
3336 =head2 FILESETS
3338 Depending on what features you need you need to include one or more sets of files
3339 in your application.
3341 =head3 CORE EVENT LOOP
3343 To include only the libev core (all the C<ev_*> functions), with manual
3344 configuration (no autoconf):
3346 #define EV_STANDALONE 1
3347 #include "ev.c"
3349 This will automatically include F<ev.h>, too, and should be done in a
3350 single C source file only to provide the function implementations. To use
3351 it, do the same for F<ev.h> in all files wishing to use this API (best
3352 done by writing a wrapper around F<ev.h> that you can include instead and
3353 where you can put other configuration options):
3355 #define EV_STANDALONE 1
3356 #include "ev.h"
3358 Both header files and implementation files can be compiled with a C++
3359 compiler (at least, that's a stated goal, and breakage will be treated
3360 as a bug).
3362 You need the following files in your source tree, or in a directory
3363 in your include path (e.g. in libev/ when using -Ilibev):
3365 ev.h
3366 ev.c
3367 ev_vars.h
3368 ev_wrap.h
3370 ev_win32.c required on win32 platforms only
3372 ev_select.c only when select backend is enabled (which is enabled by default)
3373 ev_poll.c only when poll backend is enabled (disabled by default)
3374 ev_epoll.c only when the epoll backend is enabled (disabled by default)
3375 ev_kqueue.c only when the kqueue backend is enabled (disabled by default)
3376 ev_port.c only when the solaris port backend is enabled (disabled by default)
3378 F<ev.c> includes the backend files directly when enabled, so you only need
3379 to compile this single file.
3383 To include the libevent compatibility API, also include:
3385 #include "event.c"
3387 in the file including F<ev.c>, and:
3389 #include "event.h"
3391 in the files that want to use the libevent API. This also includes F<ev.h>.
3393 You need the following additional files for this:
3395 event.h
3396 event.c
3400 Instead of using C<EV_STANDALONE=1> and providing your configuration in
3401 whatever way you want, you can also C<m4_include([libev.m4])> in your
3402 F<> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
3403 include F<config.h> and configure itself accordingly.
3405 For this of course you need the m4 file:
3407 libev.m4
3411 Libev can be configured via a variety of preprocessor symbols you have to
3412 define before including any of its files. The default in the absence of
3413 autoconf is documented for every option.
3415 =over 4
3417 =item EV_STANDALONE
3419 Must always be C<1> if you do not use autoconf configuration, which
3420 keeps libev from including F<config.h>, and it also defines dummy
3421 implementations for some libevent functions (such as logging, which is not
3422 supported). It will also not define any of the structs usually found in
3423 F<event.h> that are not directly supported by the libev core alone.
3425 In stanbdalone mode, libev will still try to automatically deduce the
3426 configuration, but has to be more conservative.
3430 If defined to be C<1>, libev will try to detect the availability of the
3431 monotonic clock option at both compile time and runtime. Otherwise no
3432 use of the monotonic clock option will be attempted. If you enable this,
3433 you usually have to link against librt or something similar. Enabling it
3434 when the functionality isn't available is safe, though, although you have
3435 to make sure you link against any libraries where the C<clock_gettime>
3436 function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
3438 =item EV_USE_REALTIME
3440 If defined to be C<1>, libev will try to detect the availability of the
3441 real-time clock option at compile time (and assume its availability
3442 at runtime if successful). Otherwise no use of the real-time clock
3443 option will be attempted. This effectively replaces C<gettimeofday>
3444 by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
3445 correctness. See the note about libraries in the description of
3446 C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
3451 If defined to be C<1>, libev will try to use a direct syscall instead
3452 of calling the system-provided C<clock_gettime> function. This option
3453 exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
3454 unconditionally pulls in C<libpthread>, slowing down single-threaded
3455 programs needlessly. Using a direct syscall is slightly slower (in
3456 theory), because no optimised vdso implementation can be used, but avoids
3457 the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
3458 higher, as it simplifies linking (no need for C<-lrt>).
3462 If defined to be C<1>, libev will assume that C<nanosleep ()> is available
3463 and will use it for delays. Otherwise it will use C<select ()>.
3465 =item EV_USE_EVENTFD
3467 If defined to be C<1>, then libev will assume that C<eventfd ()> is
3468 available and will probe for kernel support at runtime. This will improve
3469 C<ev_signal> and C<ev_async> performance and reduce resource consumption.
3470 If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
3471 2.7 or newer, otherwise disabled.
3473 =item EV_USE_SELECT
3475 If undefined or defined to be C<1>, libev will compile in support for the
3476 C<select>(2) backend. No attempt at auto-detection will be done: if no
3477 other method takes over, select will be it. Otherwise the select backend
3478 will not be compiled in.
3482 If defined to C<1>, then the select backend will use the system C<fd_set>
3483 structure. This is useful if libev doesn't compile due to a missing
3484 C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
3485 on exotic systems. This usually limits the range of file descriptors to
3486 some low limit such as 1024 or might have other limitations (winsocket
3487 only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
3488 configures the maximum size of the C<fd_set>.
3492 When defined to C<1>, the select backend will assume that
3493 select/socket/connect etc. don't understand file descriptors but
3494 wants osf handles on win32 (this is the case when the select to
3495 be used is the winsock select). This means that it will call
3496 C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
3497 it is assumed that all these functions actually work on fds, even
3498 on win32. Should not be defined on non-win32 platforms.
3500 =item EV_FD_TO_WIN32_HANDLE
3502 If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
3503 file descriptors to socket handles. When not defining this symbol (the
3504 default), then libev will call C<_get_osfhandle>, which is usually
3505 correct. In some cases, programs use their own file descriptor management,
3506 in which case they can provide this function to map fds to socket handles.
3508 =item EV_USE_POLL
3510 If defined to be C<1>, libev will compile in support for the C<poll>(2)
3511 backend. Otherwise it will be enabled on non-win32 platforms. It
3512 takes precedence over select.
3514 =item EV_USE_EPOLL
3516 If defined to be C<1>, libev will compile in support for the Linux
3517 C<epoll>(7) backend. Its availability will be detected at runtime,
3518 otherwise another method will be used as fallback. This is the preferred
3519 backend for GNU/Linux systems. If undefined, it will be enabled if the
3520 headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
3522 =item EV_USE_KQUEUE
3524 If defined to be C<1>, libev will compile in support for the BSD style
3525 C<kqueue>(2) backend. Its actual availability will be detected at runtime,
3526 otherwise another method will be used as fallback. This is the preferred
3527 backend for BSD and BSD-like systems, although on most BSDs kqueue only
3528 supports some types of fds correctly (the only platform we found that
3529 supports ptys for example was NetBSD), so kqueue might be compiled in, but
3530 not be used unless explicitly requested. The best way to use it is to find
3531 out whether kqueue supports your type of fd properly and use an embedded
3532 kqueue loop.
3534 =item EV_USE_PORT
3536 If defined to be C<1>, libev will compile in support for the Solaris
3537 10 port style backend. Its availability will be detected at runtime,
3538 otherwise another method will be used as fallback. This is the preferred
3539 backend for Solaris 10 systems.
3541 =item EV_USE_DEVPOLL
3543 Reserved for future expansion, works like the USE symbols above.
3545 =item EV_USE_INOTIFY
3547 If defined to be C<1>, libev will compile in support for the Linux inotify
3548 interface to speed up C<ev_stat> watchers. Its actual availability will
3549 be detected at runtime. If undefined, it will be enabled if the headers
3550 indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
3552 =item EV_ATOMIC_T
3554 Libev requires an integer type (suitable for storing C<0> or C<1>) whose
3555 access is atomic with respect to other threads or signal contexts. No such
3556 type is easily found in the C language, so you can provide your own type
3557 that you know is safe for your purposes. It is used both for signal handler "locking"
3558 as well as for signal and thread safety in C<ev_async> watchers.
3560 In the absence of this define, libev will use C<sig_atomic_t volatile>
3561 (from F<signal.h>), which is usually good enough on most platforms.
3563 =item EV_H
3565 The name of the F<ev.h> header file used to include it. The default if
3566 undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
3567 used to virtually rename the F<ev.h> header file in case of conflicts.
3569 =item EV_CONFIG_H
3571 If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
3572 F<ev.c>'s idea of where to find the F<config.h> file, similarly to
3573 C<EV_H>, above.
3575 =item EV_EVENT_H
3577 Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
3578 of how the F<event.h> header can be found, the default is C<"event.h">.
3580 =item EV_PROTOTYPES
3582 If defined to be C<0>, then F<ev.h> will not define any function
3583 prototypes, but still define all the structs and other symbols. This is
3584 occasionally useful if you want to provide your own wrapper functions
3585 around libev functions.
3589 If undefined or defined to C<1>, then all event-loop-specific functions
3590 will have the C<struct ev_loop *> as first argument, and you can create
3591 additional independent event loops. Otherwise there will be no support
3592 for multiple event loops and there is no first event loop pointer
3593 argument. Instead, all functions act on the single default loop.
3595 =item EV_MINPRI
3597 =item EV_MAXPRI
3599 The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
3600 C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
3601 provide for more priorities by overriding those symbols (usually defined
3602 to be C<-2> and C<2>, respectively).
3604 When doing priority-based operations, libev usually has to linearly search
3605 all the priorities, so having many of them (hundreds) uses a lot of space
3606 and time, so using the defaults of five priorities (-2 .. +2) is usually
3607 fine.
3609 If your embedding application does not need any priorities, defining these
3610 both to C<0> will save some memory and CPU.
3614 If undefined or defined to be C<1>, then periodic timers are supported. If
3615 defined to be C<0>, then they are not. Disabling them saves a few kB of
3616 code.
3618 =item EV_IDLE_ENABLE
3620 If undefined or defined to be C<1>, then idle watchers are supported. If
3621 defined to be C<0>, then they are not. Disabling them saves a few kB of
3622 code.
3624 =item EV_EMBED_ENABLE
3626 If undefined or defined to be C<1>, then embed watchers are supported. If
3627 defined to be C<0>, then they are not. Embed watchers rely on most other
3628 watcher types, which therefore must not be disabled.
3630 =item EV_STAT_ENABLE
3632 If undefined or defined to be C<1>, then stat watchers are supported. If
3633 defined to be C<0>, then they are not.
3635 =item EV_FORK_ENABLE
3637 If undefined or defined to be C<1>, then fork watchers are supported. If
3638 defined to be C<0>, then they are not.
3640 =item EV_ASYNC_ENABLE
3642 If undefined or defined to be C<1>, then async watchers are supported. If
3643 defined to be C<0>, then they are not.
3645 =item EV_MINIMAL
3647 If you need to shave off some kilobytes of code at the expense of some
3648 speed, define this symbol to C<1>. Currently this is used to override some
3649 inlining decisions, saves roughly 30% code size on amd64. It also selects a
3650 much smaller 2-heap for timer management over the default 4-heap.
3652 =item EV_PID_HASHSIZE
3654 C<ev_child> watchers use a small hash table to distribute workload by
3655 pid. The default size is C<16> (or C<1> with C<EV_MINIMAL>), usually more
3656 than enough. If you need to manage thousands of children you might want to
3657 increase this value (I<must> be a power of two).
3661 C<ev_stat> watchers use a small hash table to distribute workload by
3662 inotify watch id. The default size is C<16> (or C<1> with C<EV_MINIMAL>),
3663 usually more than enough. If you need to manage thousands of C<ev_stat>
3664 watchers you might want to increase this value (I<must> be a power of
3665 two).
3667 =item EV_USE_4HEAP
3669 Heaps are not very cache-efficient. To improve the cache-efficiency of the
3670 timer and periodics heaps, libev uses a 4-heap when this symbol is defined
3671 to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
3672 faster performance with many (thousands) of watchers.
3674 The default is C<1> unless C<EV_MINIMAL> is set in which case it is C<0>
3675 (disabled).
3677 =item EV_HEAP_CACHE_AT
3679 Heaps are not very cache-efficient. To improve the cache-efficiency of the
3680 timer and periodics heaps, libev can cache the timestamp (I<at>) within
3681 the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
3682 which uses 8-12 bytes more per watcher and a few hundred bytes more code,
3683 but avoids random read accesses on heap changes. This improves performance
3684 noticeably with many (hundreds) of watchers.
3686 The default is C<1> unless C<EV_MINIMAL> is set in which case it is C<0>
3687 (disabled).
3689 =item EV_VERIFY
3691 Controls how much internal verification (see C<ev_loop_verify ()>) will
3692 be done: If set to C<0>, no internal verification code will be compiled
3693 in. If set to C<1>, then verification code will be compiled in, but not
3694 called. If set to C<2>, then the internal verification code will be
3695 called once per loop, which can slow down libev. If set to C<3>, then the
3696 verification code will be called very frequently, which will slow down
3697 libev considerably.
3699 The default is C<1>, unless C<EV_MINIMAL> is set, in which case it will be
3700 C<0>.
3702 =item EV_COMMON
3704 By default, all watchers have a C<void *data> member. By redefining
3705 this macro to a something else you can include more and other types of
3706 members. You have to define it each time you include one of the files,
3707 though, and it must be identical each time.
3709 For example, the perl EV module uses something like this:
3711 #define EV_COMMON \
3712 SV *self; /* contains this struct */ \
3713 SV *cb_sv, *fh /* note no trailing ";" */
3715 =item EV_CB_DECLARE (type)
3717 =item EV_CB_INVOKE (watcher, revents)
3719 =item ev_set_cb (ev, cb)
3721 Can be used to change the callback member declaration in each watcher,
3722 and the way callbacks are invoked and set. Must expand to a struct member
3723 definition and a statement, respectively. See the F<ev.h> header file for
3724 their default definitions. One possible use for overriding these is to
3725 avoid the C<struct ev_loop *> as first argument in all cases, or to use
3726 method calls instead of plain function calls in C++.
3728 =back
3732 If you need to re-export the API (e.g. via a DLL) and you need a list of
3733 exported symbols, you can use the provided F<Symbol.*> files which list
3734 all public symbols, one per line:
3736 Symbols.ev for libev proper
3737 Symbols.event for the libevent emulation
3739 This can also be used to rename all public symbols to avoid clashes with
3740 multiple versions of libev linked together (which is obviously bad in
3741 itself, but sometimes it is inconvenient to avoid this).
3743 A sed command like this will create wrapper C<#define>'s that you need to
3744 include before including F<ev.h>:
3746 <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
3748 This would create a file F<wrap.h> which essentially looks like this:
3750 #define ev_backend myprefix_ev_backend
3751 #define ev_check_start myprefix_ev_check_start
3752 #define ev_check_stop myprefix_ev_check_stop
3753 ...
3755 =head2 EXAMPLES
3757 For a real-world example of a program the includes libev
3758 verbatim, you can have a look at the EV perl module
3759 (L<>). It has the libev files in
3760 the F<libev/> subdirectory and includes them in the F<EV/EVAPI.h> (public
3761 interface) and F<EV.xs> (implementation) files. Only the F<EV.xs> file
3762 will be compiled. It is pretty complex because it provides its own header
3763 file.
3765 The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
3766 that everybody includes and which overrides some configure choices:
3768 #define EV_MINIMAL 1
3769 #define EV_USE_POLL 0
3770 #define EV_MULTIPLICITY 0
3771 #define EV_PERIODIC_ENABLE 0
3772 #define EV_STAT_ENABLE 0
3773 #define EV_FORK_ENABLE 0
3774 #define EV_CONFIG_H <config.h>
3775 #define EV_MINPRI 0
3776 #define EV_MAXPRI 0
3778 #include "ev++.h"
3780 And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
3782 #include "ev_cpp.h"
3783 #include "ev.c"
3789 =head3 THREADS
3791 All libev functions are reentrant and thread-safe unless explicitly
3792 documented otherwise, but libev implements no locking itself. This means
3793 that you can use as many loops as you want in parallel, as long as there
3794 are no concurrent calls into any libev function with the same loop
3795 parameter (C<ev_default_*> calls have an implicit default loop parameter,
3796 of course): libev guarantees that different event loops share no data
3797 structures that need any locking.
3799 Or to put it differently: calls with different loop parameters can be done
3800 concurrently from multiple threads, calls with the same loop parameter
3801 must be done serially (but can be done from different threads, as long as
3802 only one thread ever is inside a call at any point in time, e.g. by using
3803 a mutex per loop).
3805 Specifically to support threads (and signal handlers), libev implements
3806 so-called C<ev_async> watchers, which allow some limited form of
3807 concurrency on the same event loop, namely waking it up "from the
3808 outside".
3810 If you want to know which design (one loop, locking, or multiple loops
3811 without or something else still) is best for your problem, then I cannot
3812 help you, but here is some generic advice:
3814 =over 4
3816 =item * most applications have a main thread: use the default libev loop
3817 in that thread, or create a separate thread running only the default loop.
3819 This helps integrating other libraries or software modules that use libev
3820 themselves and don't care/know about threading.
3822 =item * one loop per thread is usually a good model.
3824 Doing this is almost never wrong, sometimes a better-performance model
3825 exists, but it is always a good start.
3827 =item * other models exist, such as the leader/follower pattern, where one
3828 loop is handed through multiple threads in a kind of round-robin fashion.
3830 Choosing a model is hard - look around, learn, know that usually you can do
3831 better than you currently do :-)
3833 =item * often you need to talk to some other thread which blocks in the
3834 event loop.
3836 C<ev_async> watchers can be used to wake them up from other threads safely
3837 (or from signal contexts...).
3839 An example use would be to communicate signals or other events that only
3840 work in the default loop by registering the signal watcher with the
3841 default loop and triggering an C<ev_async> watcher from the default loop
3842 watcher callback into the event loop interested in the signal.
3844 =back
3846 =head3 COROUTINES
3848 Libev is very accommodating to coroutines ("cooperative threads"):
3849 libev fully supports nesting calls to its functions from different
3850 coroutines (e.g. you can call C<ev_loop> on the same loop from two
3851 different coroutines, and switch freely between both coroutines running the
3852 loop, as long as you don't confuse yourself). The only exception is that
3853 you must not do this from C<ev_periodic> reschedule callbacks.
3855 Care has been taken to ensure that libev does not keep local state inside
3856 C<ev_loop>, and other calls do not usually allow for coroutine switches as
3857 they do not call any callbacks.
3861 Depending on your compiler and compiler settings, you might get no or a
3862 lot of warnings when compiling libev code. Some people are apparently
3863 scared by this.
3865 However, these are unavoidable for many reasons. For one, each compiler
3866 has different warnings, and each user has different tastes regarding
3867 warning options. "Warn-free" code therefore cannot be a goal except when
3868 targeting a specific compiler and compiler-version.
3870 Another reason is that some compiler warnings require elaborate
3871 workarounds, or other changes to the code that make it less clear and less
3872 maintainable.
3874 And of course, some compiler warnings are just plain stupid, or simply
3875 wrong (because they don't actually warn about the condition their message
3876 seems to warn about). For example, certain older gcc versions had some
3877 warnings that resulted an extreme number of false positives. These have
3878 been fixed, but some people still insist on making code warn-free with
3879 such buggy versions.
3881 While libev is written to generate as few warnings as possible,
3882 "warn-free" code is not a goal, and it is recommended not to build libev
3883 with any compiler warnings enabled unless you are prepared to cope with
3884 them (e.g. by ignoring them). Remember that warnings are just that:
3885 warnings, not errors, or proof of bugs.
3888 =head2 VALGRIND
3890 Valgrind has a special section here because it is a popular tool that is
3891 highly useful. Unfortunately, valgrind reports are very hard to interpret.
3893 If you think you found a bug (memory leak, uninitialised data access etc.)
3894 in libev, then check twice: If valgrind reports something like:
3896 ==2274== definitely lost: 0 bytes in 0 blocks.
3897 ==2274== possibly lost: 0 bytes in 0 blocks.
3898 ==2274== still reachable: 256 bytes in 1 blocks.
3900 Then there is no memory leak, just as memory accounted to global variables
3901 is not a memleak - the memory is still being referenced, and didn't leak.
3903 Similarly, under some circumstances, valgrind might report kernel bugs
3904 as if it were a bug in libev (e.g. in realloc or in the poll backend,
3905 although an acceptable workaround has been found here), or it might be
3906 confused.
3908 Keep in mind that valgrind is a very good tool, but only a tool. Don't
3909 make it into some kind of religion.
3911 If you are unsure about something, feel free to contact the mailing list
3912 with the full valgrind report and an explanation on why you think this
3913 is a bug in libev (best check the archives, too :). However, don't be
3914 annoyed when you get a brisk "this is no bug" answer and take the chance
3915 of learning how to interpret valgrind properly.
3917 If you need, for some reason, empty reports from valgrind for your project
3918 I suggest using suppression lists.
3925 Win32 doesn't support any of the standards (e.g. POSIX) that libev
3926 requires, and its I/O model is fundamentally incompatible with the POSIX
3927 model. Libev still offers limited functionality on this platform in
3928 the form of the C<EVBACKEND_SELECT> backend, and only supports socket
3929 descriptors. This only applies when using Win32 natively, not when using
3930 e.g. cygwin.
3932 Lifting these limitations would basically require the full
3933 re-implementation of the I/O system. If you are into these kinds of
3934 things, then note that glib does exactly that for you in a very portable
3935 way (note also that glib is the slowest event library known to man).
3937 There is no supported compilation method available on windows except
3938 embedding it into other applications.
3940 Sensible signal handling is officially unsupported by Microsoft - libev
3941 tries its best, but under most conditions, signals will simply not work.
3943 Not a libev limitation but worth mentioning: windows apparently doesn't
3944 accept large writes: instead of resulting in a partial write, windows will
3945 either accept everything or return C<ENOBUFS> if the buffer is too large,
3946 so make sure you only write small amounts into your sockets (less than a
3947 megabyte seems safe, but this apparently depends on the amount of memory
3948 available).
3950 Due to the many, low, and arbitrary limits on the win32 platform and
3951 the abysmal performance of winsockets, using a large number of sockets
3952 is not recommended (and not reasonable). If your program needs to use
3953 more than a hundred or so sockets, then likely it needs to use a totally
3954 different implementation for windows, as libev offers the POSIX readiness
3955 notification model, which cannot be implemented efficiently on windows
3956 (due to Microsoft monopoly games).
3958 A typical way to use libev under windows is to embed it (see the embedding
3959 section for details) and use the following F<evwrap.h> header file instead
3960 of F<ev.h>:
3962 #define EV_STANDALONE /* keeps ev from requiring config.h */
3963 #define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
3965 #include "ev.h"
3967 And compile the following F<evwrap.c> file into your project (make sure
3968 you do I<not> compile the F<ev.c> or any other embedded source files!):
3970 #include "evwrap.h"
3971 #include "ev.c"
3973 =over 4
3975 =item The winsocket select function
3977 The winsocket C<select> function doesn't follow POSIX in that it
3978 requires socket I<handles> and not socket I<file descriptors> (it is
3979 also extremely buggy). This makes select very inefficient, and also
3980 requires a mapping from file descriptors to socket handles (the Microsoft
3981 C runtime provides the function C<_open_osfhandle> for this). See the
3982 discussion of the C<EV_SELECT_USE_FD_SET>, C<EV_SELECT_IS_WINSOCKET> and
3983 C<EV_FD_TO_WIN32_HANDLE> preprocessor symbols for more info.
3985 The configuration for a "naked" win32 using the Microsoft runtime
3986 libraries and raw winsocket select is:
3988 #define EV_USE_SELECT 1
3989 #define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
3991 Note that winsockets handling of fd sets is O(n), so you can easily get a
3992 complexity in the O(n²) range when using win32.
3994 =item Limited number of file descriptors
3996 Windows has numerous arbitrary (and low) limits on things.
3998 Early versions of winsocket's select only supported waiting for a maximum
3999 of C<64> handles (probably owning to the fact that all windows kernels
4000 can only wait for C<64> things at the same time internally; Microsoft
4001 recommends spawning a chain of threads and wait for 63 handles and the
4002 previous thread in each. Sounds great!).
4004 Newer versions support more handles, but you need to define C<FD_SETSIZE>
4005 to some high number (e.g. C<2048>) before compiling the winsocket select
4006 call (which might be in libev or elsewhere, for example, perl and many
4007 other interpreters do their own select emulation on windows).
4009 Another limit is the number of file descriptors in the Microsoft runtime
4010 libraries, which by default is C<64> (there must be a hidden I<64>
4011 fetish or something like this inside Microsoft). You can increase this
4012 by calling C<_setmaxstdio>, which can increase this limit to C<2048>
4013 (another arbitrary limit), but is broken in many versions of the Microsoft
4014 runtime libraries. This might get you to about C<512> or C<2048> sockets
4015 (depending on windows version and/or the phase of the moon). To get more,
4016 you need to wrap all I/O functions and provide your own fd management, but
4017 the cost of calling select (O(n²)) will likely make this unworkable.
4019 =back
4023 In addition to a working ISO-C implementation and of course the
4024 backend-specific APIs, libev relies on a few additional extensions:
4026 =over 4
4028 =item C<void (*)(ev_watcher_type *, int revents)> must have compatible
4029 calling conventions regardless of C<ev_watcher_type *>.
4031 Libev assumes not only that all watcher pointers have the same internal
4032 structure (guaranteed by POSIX but not by ISO C for example), but it also
4033 assumes that the same (machine) code can be used to call any watcher
4034 callback: The watcher callbacks have different type signatures, but libev
4035 calls them using an C<ev_watcher *> internally.
4037 =item C<sig_atomic_t volatile> must be thread-atomic as well
4039 The type C<sig_atomic_t volatile> (or whatever is defined as
4040 C<EV_ATOMIC_T>) must be atomic with respect to accesses from different
4041 threads. This is not part of the specification for C<sig_atomic_t>, but is
4042 believed to be sufficiently portable.
4044 =item C<sigprocmask> must work in a threaded environment
4046 Libev uses C<sigprocmask> to temporarily block signals. This is not
4047 allowed in a threaded program (C<pthread_sigmask> has to be used). Typical
4048 pthread implementations will either allow C<sigprocmask> in the "main
4049 thread" or will block signals process-wide, both behaviours would
4050 be compatible with libev. Interaction between C<sigprocmask> and
4051 C<pthread_sigmask> could complicate things, however.
4053 The most portable way to handle signals is to block signals in all threads
4054 except the initial one, and run the default loop in the initial thread as
4055 well.
4057 =item C<long> must be large enough for common memory allocation sizes
4059 To improve portability and simplify its API, libev uses C<long> internally
4060 instead of C<size_t> when allocating its data structures. On non-POSIX
4061 systems (Microsoft...) this might be unexpectedly low, but is still at
4062 least 31 bits everywhere, which is enough for hundreds of millions of
4063 watchers.
4065 =item C<double> must hold a time value in seconds with enough accuracy
4067 The type C<double> is used to represent timestamps. It is required to
4068 have at least 51 bits of mantissa (and 9 bits of exponent), which is good
4069 enough for at least into the year 4000. This requirement is fulfilled by
4070 implementations implementing IEEE 754 (basically all existing ones).
4072 =back
4074 If you know of other additional requirements drop me a note.
4079 In this section the complexities of (many of) the algorithms used inside
4080 libev will be documented. For complexity discussions about backends see
4081 the documentation for C<ev_default_init>.
4083 All of the following are about amortised time: If an array needs to be
4084 extended, libev needs to realloc and move the whole array, but this
4085 happens asymptotically rarer with higher number of elements, so O(1) might
4086 mean that libev does a lengthy realloc operation in rare cases, but on
4087 average it is much faster and asymptotically approaches constant time.
4089 =over 4
4091 =item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
4093 This means that, when you have a watcher that triggers in one hour and
4094 there are 100 watchers that would trigger before that, then inserting will
4095 have to skip roughly seven (C<ld 100>) of these watchers.
4097 =item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
4099 That means that changing a timer costs less than removing/adding them,
4100 as only the relative motion in the event queue has to be paid for.
4102 =item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
4104 These just add the watcher into an array or at the head of a list.
4106 =item Stopping check/prepare/idle/fork/async watchers: O(1)
4108 =item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
4110 These watchers are stored in lists, so they need to be walked to find the
4111 correct watcher to remove. The lists are usually short (you don't usually
4112 have many watchers waiting for the same fd or signal: one is typical, two
4113 is rare).
4115 =item Finding the next timer in each loop iteration: O(1)
4117 By virtue of using a binary or 4-heap, the next timer is always found at a
4118 fixed position in the storage array.
4120 =item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
4122 A change means an I/O watcher gets started or stopped, which requires
4123 libev to recalculate its status (and possibly tell the kernel, depending
4124 on backend and whether C<ev_io_set> was used).
4126 =item Activating one watcher (putting it into the pending state): O(1)
4128 =item Priority handling: O(number_of_priorities)
4130 Priorities are implemented by allocating some space for each
4131 priority. When doing priority-based operations, libev usually has to
4132 linearly search all the priorities, but starting/stopping and activating
4133 watchers becomes O(1) with respect to priority handling.
4135 =item Sending an ev_async: O(1)
4137 =item Processing ev_async_send: O(number_of_async_watchers)
4139 =item Processing signals: O(max_signal_number)
4141 Sending involves a system call I<iff> there were no other C<ev_async_send>
4142 calls in the current loop iteration. Checking for async and signal events
4143 involves iterating over all running async watchers or all signal numbers.
4145 =back
4148 =head1 GLOSSARY
4150 =over 4
4152 =item active
4154 A watcher is active as long as it has been started (has been attached to
4155 an event loop) but not yet stopped (disassociated from the event loop).
4157 =item application
4159 In this document, an application is whatever is using libev.
4161 =item callback
4163 The address of a function that is called when some event has been
4164 detected. Callbacks are being passed the event loop, the watcher that
4165 received the event, and the actual event bitset.
4167 =item callback invocation
4169 The act of calling the callback associated with a watcher.
4171 =item event
4173 A change of state of some external event, such as data now being available
4174 for reading on a file descriptor, time having passed or simply not having
4175 any other events happening anymore.
4177 In libev, events are represented as single bits (such as C<EV_READ> or
4178 C<EV_TIMEOUT>).
4180 =item event library
4182 A software package implementing an event model and loop.
4184 =item event loop
4186 An entity that handles and processes external events and converts them
4187 into callback invocations.
4189 =item event model
4191 The model used to describe how an event loop handles and processes
4192 watchers and events.
4194 =item pending
4196 A watcher is pending as soon as the corresponding event has been detected,
4197 and stops being pending as soon as the watcher will be invoked or its
4198 pending status is explicitly cleared by the application.
4200 A watcher can be pending, but not active. Stopping a watcher also clears
4201 its pending status.
4203 =item real time
4205 The physical time that is observed. It is apparently strictly monotonic :)
4207 =item wall-clock time
4209 The time and date as shown on clocks. Unlike real time, it can actually
4210 be wrong and jump forwards and backwards, e.g. when the you adjust your
4211 clock.
4213 =item watcher
4215 A data structure that describes interest in certain events. Watchers need
4216 to be started (attached to an event loop) before they can receive events.
4218 =item watcher invocation
4220 The act of calling the callback associated with a watcher.
4222 =back
4224 =head1 AUTHOR
4226 Marc Lehmann <>, with repeated corrections by Mikael Magnusson.