ViewVC Help
View File | Revision Log | Show Annotations | Download File
Revision: 1.457
Committed: Fri Sep 6 22:17:52 2019 UTC (4 years, 10 months ago) by root
Branch: MAIN
CVS Tags: EV-rel-4_28, EV-rel-4_29, EV-rel-4_30
Changes since 1.456: +1 -1 lines
Log Message:
*** empty log message ***

File Contents

# Content
1 =encoding utf-8
3 =head1 NAME
5 libev - a high performance full-featured event loop written in C
7 =head1 SYNOPSIS
9 #include <ev.h>
13 // a single header file is required
14 #include <ev.h>
16 #include <stdio.h> // for puts
18 // every watcher type has its own typedef'd struct
19 // with the name ev_TYPE
20 ev_io stdin_watcher;
21 ev_timer timeout_watcher;
23 // all watcher callbacks have a similar signature
24 // this callback is called when data is readable on stdin
25 static void
26 stdin_cb (EV_P_ ev_io *w, int revents)
27 {
28 puts ("stdin ready");
29 // for one-shot events, one must manually stop the watcher
30 // with its corresponding stop function.
31 ev_io_stop (EV_A_ w);
33 // this causes all nested ev_run's to stop iterating
34 ev_break (EV_A_ EVBREAK_ALL);
35 }
37 // another callback, this time for a time-out
38 static void
39 timeout_cb (EV_P_ ev_timer *w, int revents)
40 {
41 puts ("timeout");
42 // this causes the innermost ev_run to stop iterating
43 ev_break (EV_A_ EVBREAK_ONE);
44 }
46 int
47 main (void)
48 {
49 // use the default event loop unless you have special needs
50 struct ev_loop *loop = EV_DEFAULT;
52 // initialise an io watcher, then start it
53 // this one will watch for stdin to become readable
54 ev_io_init (&stdin_watcher, stdin_cb, /*STDIN_FILENO*/ 0, EV_READ);
55 ev_io_start (loop, &stdin_watcher);
57 // initialise a timer watcher, then start it
58 // simple non-repeating 5.5 second timeout
59 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
60 ev_timer_start (loop, &timeout_watcher);
62 // now wait for events to arrive
63 ev_run (loop, 0);
65 // break was called, so exit
66 return 0;
67 }
71 This document documents the libev software package.
73 The newest version of this document is also available as an html-formatted
74 web page you might find easier to navigate when reading it for the first
75 time: L<>.
77 While this document tries to be as complete as possible in documenting
78 libev, its usage and the rationale behind its design, it is not a tutorial
79 on event-based programming, nor will it introduce event-based programming
80 with libev.
82 Familiarity with event based programming techniques in general is assumed
83 throughout this document.
87 This manual tries to be very detailed, but unfortunately, this also makes
88 it very long. If you just want to know the basics of libev, I suggest
89 reading L</ANATOMY OF A WATCHER>, then the L</EXAMPLE PROGRAM> above and
90 look up the missing functions in L</GLOBAL FUNCTIONS> and the C<ev_io> and
91 C<ev_timer> sections in L</WATCHER TYPES>.
93 =head1 ABOUT LIBEV
95 Libev is an event loop: you register interest in certain events (such as a
96 file descriptor being readable or a timeout occurring), and it will manage
97 these event sources and provide your program with events.
99 To do this, it must take more or less complete control over your process
100 (or thread) by executing the I<event loop> handler, and will then
101 communicate events via a callback mechanism.
103 You register interest in certain events by registering so-called I<event
104 watchers>, which are relatively small C structures you initialise with the
105 details of the event, and then hand it over to libev by I<starting> the
106 watcher.
108 =head2 FEATURES
110 Libev supports C<select>, C<poll>, the Linux-specific aio and C<epoll>
111 interfaces, the BSD-specific C<kqueue> and the Solaris-specific event port
112 mechanisms for file descriptor events (C<ev_io>), the Linux C<inotify>
113 interface (for C<ev_stat>), Linux eventfd/signalfd (for faster and cleaner
114 inter-thread wakeup (C<ev_async>)/signal handling (C<ev_signal>)) relative
115 timers (C<ev_timer>), absolute timers with customised rescheduling
116 (C<ev_periodic>), synchronous signals (C<ev_signal>), process status
117 change events (C<ev_child>), and event watchers dealing with the event
118 loop mechanism itself (C<ev_idle>, C<ev_embed>, C<ev_prepare> and
119 C<ev_check> watchers) as well as file watchers (C<ev_stat>) and even
120 limited support for fork events (C<ev_fork>).
122 It also is quite fast (see this
123 L<benchmark|> comparing it to libevent
124 for example).
126 =head2 CONVENTIONS
128 Libev is very configurable. In this manual the default (and most common)
129 configuration will be described, which supports multiple event loops. For
130 more info about various configuration options please have a look at
131 B<EMBED> section in this manual. If libev was configured without support
132 for multiple event loops, then all functions taking an initial argument of
133 name C<loop> (which is always of type C<struct ev_loop *>) will not have
134 this argument.
138 Libev represents time as a single floating point number, representing
139 the (fractional) number of seconds since the (POSIX) epoch (in practice
140 somewhere near the beginning of 1970, details are complicated, don't
141 ask). This type is called C<ev_tstamp>, which is what you should use
142 too. It usually aliases to the C<double> type in C. When you need to do
143 any calculations on it, you should treat it as some floating point value.
145 Unlike the name component C<stamp> might indicate, it is also used for
146 time differences (e.g. delays) throughout libev.
150 Libev knows three classes of errors: operating system errors, usage errors
151 and internal errors (bugs).
153 When libev catches an operating system error it cannot handle (for example
154 a system call indicating a condition libev cannot fix), it calls the callback
155 set via C<ev_set_syserr_cb>, which is supposed to fix the problem or
156 abort. The default is to print a diagnostic message and to call C<abort
157 ()>.
159 When libev detects a usage error such as a negative timer interval, then
160 it will print a diagnostic message and abort (via the C<assert> mechanism,
161 so C<NDEBUG> will disable this checking): these are programming errors in
162 the libev caller and need to be fixed there.
164 Via the C<EV_FREQUENT> macro you can compile in and/or enable extensive
165 consistency checking code inside libev that can be used to check for
166 internal inconsistencies, suually caused by application bugs.
168 Libev also has a few internal error-checking C<assert>ions. These do not
169 trigger under normal circumstances, as they indicate either a bug in libev
170 or worse.
175 These functions can be called anytime, even before initialising the
176 library in any way.
178 =over 4
180 =item ev_tstamp ev_time ()
182 Returns the current time as libev would use it. Please note that the
183 C<ev_now> function is usually faster and also often returns the timestamp
184 you actually want to know. Also interesting is the combination of
185 C<ev_now_update> and C<ev_now>.
187 =item ev_sleep (ev_tstamp interval)
189 Sleep for the given interval: The current thread will be blocked
190 until either it is interrupted or the given time interval has
191 passed (approximately - it might return a bit earlier even if not
192 interrupted). Returns immediately if C<< interval <= 0 >>.
194 Basically this is a sub-second-resolution C<sleep ()>.
196 The range of the C<interval> is limited - libev only guarantees to work
197 with sleep times of up to one day (C<< interval <= 86400 >>).
199 =item int ev_version_major ()
201 =item int ev_version_minor ()
203 You can find out the major and minor ABI version numbers of the library
204 you linked against by calling the functions C<ev_version_major> and
205 C<ev_version_minor>. If you want, you can compare against the global
206 symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the
207 version of the library your program was compiled against.
209 These version numbers refer to the ABI version of the library, not the
210 release version.
212 Usually, it's a good idea to terminate if the major versions mismatch,
213 as this indicates an incompatible change. Minor versions are usually
214 compatible to older versions, so a larger minor version alone is usually
215 not a problem.
217 Example: Make sure we haven't accidentally been linked against the wrong
218 version (note, however, that this will not detect other ABI mismatches,
219 such as LFS or reentrancy).
221 assert (("libev version mismatch",
222 ev_version_major () == EV_VERSION_MAJOR
223 && ev_version_minor () >= EV_VERSION_MINOR));
225 =item unsigned int ev_supported_backends ()
227 Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
228 value) compiled into this binary of libev (independent of their
229 availability on the system you are running on). See C<ev_default_loop> for
230 a description of the set values.
232 Example: make sure we have the epoll method, because yeah this is cool and
233 a must have and can we have a torrent of it please!!!11
235 assert (("sorry, no epoll, no sex",
236 ev_supported_backends () & EVBACKEND_EPOLL));
238 =item unsigned int ev_recommended_backends ()
240 Return the set of all backends compiled into this binary of libev and
241 also recommended for this platform, meaning it will work for most file
242 descriptor types. This set is often smaller than the one returned by
243 C<ev_supported_backends>, as for example kqueue is broken on most BSDs
244 and will not be auto-detected unless you explicitly request it (assuming
245 you know what you are doing). This is the set of backends that libev will
246 probe for if you specify no backends explicitly.
248 =item unsigned int ev_embeddable_backends ()
250 Returns the set of backends that are embeddable in other event loops. This
251 value is platform-specific but can include backends not available on the
252 current system. To find which embeddable backends might be supported on
253 the current system, you would need to look at C<ev_embeddable_backends ()
254 & ev_supported_backends ()>, likewise for recommended ones.
256 See the description of C<ev_embed> watchers for more info.
258 =item ev_set_allocator (void *(*cb)(void *ptr, long size) throw ())
260 Sets the allocation function to use (the prototype is similar - the
261 semantics are identical to the C<realloc> C89/SuS/POSIX function). It is
262 used to allocate and free memory (no surprises here). If it returns zero
263 when memory needs to be allocated (C<size != 0>), the library might abort
264 or take some potentially destructive action.
266 Since some systems (at least OpenBSD and Darwin) fail to implement
267 correct C<realloc> semantics, libev will use a wrapper around the system
268 C<realloc> and C<free> functions by default.
270 You could override this function in high-availability programs to, say,
271 free some memory if it cannot allocate memory, to use a special allocator,
272 or even to sleep a while and retry until some memory is available.
274 Example: The following is the C<realloc> function that libev itself uses
275 which should work with C<realloc> and C<free> functions of all kinds and
276 is probably a good basis for your own implementation.
278 static void *
279 ev_realloc_emul (void *ptr, long size) EV_NOEXCEPT
280 {
281 if (size)
282 return realloc (ptr, size);
284 free (ptr);
285 return 0;
286 }
288 Example: Replace the libev allocator with one that waits a bit and then
289 retries.
291 static void *
292 persistent_realloc (void *ptr, size_t size)
293 {
294 if (!size)
295 {
296 free (ptr);
297 return 0;
298 }
300 for (;;)
301 {
302 void *newptr = realloc (ptr, size);
304 if (newptr)
305 return newptr;
307 sleep (60);
308 }
309 }
311 ...
312 ev_set_allocator (persistent_realloc);
314 =item ev_set_syserr_cb (void (*cb)(const char *msg) throw ())
316 Set the callback function to call on a retryable system call error (such
317 as failed select, poll, epoll_wait). The message is a printable string
318 indicating the system call or subsystem causing the problem. If this
319 callback is set, then libev will expect it to remedy the situation, no
320 matter what, when it returns. That is, libev will generally retry the
321 requested operation, or, if the condition doesn't go away, do bad stuff
322 (such as abort).
324 Example: This is basically the same thing that libev does internally, too.
326 static void
327 fatal_error (const char *msg)
328 {
329 perror (msg);
330 abort ();
331 }
333 ...
334 ev_set_syserr_cb (fatal_error);
336 =item ev_feed_signal (int signum)
338 This function can be used to "simulate" a signal receive. It is completely
339 safe to call this function at any time, from any context, including signal
340 handlers or random threads.
342 Its main use is to customise signal handling in your process, especially
343 in the presence of threads. For example, you could block signals
344 by default in all threads (and specifying C<EVFLAG_NOSIGMASK> when
345 creating any loops), and in one thread, use C<sigwait> or any other
346 mechanism to wait for signals, then "deliver" them to libev by calling
347 C<ev_feed_signal>.
349 =back
353 An event loop is described by a C<struct ev_loop *> (the C<struct> is
354 I<not> optional in this case unless libev 3 compatibility is disabled, as
355 libev 3 had an C<ev_loop> function colliding with the struct name).
357 The library knows two types of such loops, the I<default> loop, which
358 supports child process events, and dynamically created event loops which
359 do not.
361 =over 4
363 =item struct ev_loop *ev_default_loop (unsigned int flags)
365 This returns the "default" event loop object, which is what you should
366 normally use when you just need "the event loop". Event loop objects and
367 the C<flags> parameter are described in more detail in the entry for
368 C<ev_loop_new>.
370 If the default loop is already initialised then this function simply
371 returns it (and ignores the flags. If that is troubling you, check
372 C<ev_backend ()> afterwards). Otherwise it will create it with the given
373 flags, which should almost always be C<0>, unless the caller is also the
374 one calling C<ev_run> or otherwise qualifies as "the main program".
376 If you don't know what event loop to use, use the one returned from this
377 function (or via the C<EV_DEFAULT> macro).
379 Note that this function is I<not> thread-safe, so if you want to use it
380 from multiple threads, you have to employ some kind of mutex (note also
381 that this case is unlikely, as loops cannot be shared easily between
382 threads anyway).
384 The default loop is the only loop that can handle C<ev_child> watchers,
385 and to do this, it always registers a handler for C<SIGCHLD>. If this is
386 a problem for your application you can either create a dynamic loop with
387 C<ev_loop_new> which doesn't do that, or you can simply overwrite the
388 C<SIGCHLD> signal handler I<after> calling C<ev_default_init>.
390 Example: This is the most typical usage.
392 if (!ev_default_loop (0))
393 fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
395 Example: Restrict libev to the select and poll backends, and do not allow
396 environment settings to be taken into account:
400 =item struct ev_loop *ev_loop_new (unsigned int flags)
402 This will create and initialise a new event loop object. If the loop
403 could not be initialised, returns false.
405 This function is thread-safe, and one common way to use libev with
406 threads is indeed to create one loop per thread, and using the default
407 loop in the "main" or "initial" thread.
409 The flags argument can be used to specify special behaviour or specific
410 backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
412 The following flags are supported:
414 =over 4
416 =item C<EVFLAG_AUTO>
418 The default flags value. Use this if you have no clue (it's the right
419 thing, believe me).
421 =item C<EVFLAG_NOENV>
423 If this flag bit is or'ed into the flag value (or the program runs setuid
424 or setgid) then libev will I<not> look at the environment variable
425 C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
426 override the flags completely if it is found in the environment. This is
427 useful to try out specific backends to test their performance, to work
428 around bugs, or to make libev threadsafe (accessing environment variables
429 cannot be done in a threadsafe way, but usually it works if no other
430 thread modifies them).
434 Instead of calling C<ev_loop_fork> manually after a fork, you can also
435 make libev check for a fork in each iteration by enabling this flag.
437 This works by calling C<getpid ()> on every iteration of the loop,
438 and thus this might slow down your event loop if you do a lot of loop
439 iterations and little real work, but is usually not noticeable (on my
440 GNU/Linux system for example, C<getpid> is actually a simple 5-insn
441 sequence without a system call and thus I<very> fast, but my GNU/Linux
442 system also has C<pthread_atfork> which is even faster). (Update: glibc
443 versions 2.25 apparently removed the C<getpid> optimisation again).
445 The big advantage of this flag is that you can forget about fork (and
446 forget about forgetting to tell libev about forking, although you still
447 have to ignore C<SIGPIPE>) when you use this flag.
449 This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
450 environment variable.
454 When this flag is specified, then libev will not attempt to use the
455 I<inotify> API for its C<ev_stat> watchers. Apart from debugging and
456 testing, this flag can be useful to conserve inotify file descriptors, as
457 otherwise each loop using C<ev_stat> watchers consumes one inotify handle.
461 When this flag is specified, then libev will attempt to use the
462 I<signalfd> API for its C<ev_signal> (and C<ev_child>) watchers. This API
463 delivers signals synchronously, which makes it both faster and might make
464 it possible to get the queued signal data. It can also simplify signal
465 handling with threads, as long as you properly block signals in your
466 threads that are not interested in handling them.
468 Signalfd will not be used by default as this changes your signal mask, and
469 there are a lot of shoddy libraries and programs (glib's threadpool for
470 example) that can't properly initialise their signal masks.
474 When this flag is specified, then libev will avoid to modify the signal
475 mask. Specifically, this means you have to make sure signals are unblocked
476 when you want to receive them.
478 This behaviour is useful when you want to do your own signal handling, or
479 want to handle signals only in specific threads and want to avoid libev
480 unblocking the signals.
482 It's also required by POSIX in a threaded program, as libev calls
483 C<sigprocmask>, whose behaviour is officially unspecified.
485 This flag's behaviour will become the default in future versions of libev.
487 =item C<EVBACKEND_SELECT> (value 1, portable select backend)
489 This is your standard select(2) backend. Not I<completely> standard, as
490 libev tries to roll its own fd_set with no limits on the number of fds,
491 but if that fails, expect a fairly low limit on the number of fds when
492 using this backend. It doesn't scale too well (O(highest_fd)), but its
493 usually the fastest backend for a low number of (low-numbered :) fds.
495 To get good performance out of this backend you need a high amount of
496 parallelism (most of the file descriptors should be busy). If you are
497 writing a server, you should C<accept ()> in a loop to accept as many
498 connections as possible during one iteration. You might also want to have
499 a look at C<ev_set_io_collect_interval ()> to increase the amount of
500 readiness notifications you get per iteration.
502 This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
503 C<writefds> set (and to work around Microsoft Windows bugs, also onto the
504 C<exceptfds> set on that platform).
506 =item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
508 And this is your standard poll(2) backend. It's more complicated
509 than select, but handles sparse fds better and has no artificial
510 limit on the number of fds you can use (except it will slow down
511 considerably with a lot of inactive fds). It scales similarly to select,
512 i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
513 performance tips.
515 This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
518 =item C<EVBACKEND_EPOLL> (value 4, Linux)
520 Use the Linux-specific epoll(7) interface (for both pre- and post-2.6.9
521 kernels).
523 For few fds, this backend is a bit little slower than poll and select, but
524 it scales phenomenally better. While poll and select usually scale like
525 O(total_fds) where total_fds is the total number of fds (or the highest
526 fd), epoll scales either O(1) or O(active_fds).
528 The epoll mechanism deserves honorable mention as the most misdesigned
529 of the more advanced event mechanisms: mere annoyances include silently
530 dropping file descriptors, requiring a system call per change per file
531 descriptor (and unnecessary guessing of parameters), problems with dup,
532 returning before the timeout value, resulting in additional iterations
533 (and only giving 5ms accuracy while select on the same platform gives
534 0.1ms) and so on. The biggest issue is fork races, however - if a program
535 forks then I<both> parent and child process have to recreate the epoll
536 set, which can take considerable time (one syscall per file descriptor)
537 and is of course hard to detect.
539 Epoll is also notoriously buggy - embedding epoll fds I<should> work,
540 but of course I<doesn't>, and epoll just loves to report events for
541 totally I<different> file descriptors (even already closed ones, so
542 one cannot even remove them from the set) than registered in the set
543 (especially on SMP systems). Libev tries to counter these spurious
544 notifications by employing an additional generation counter and comparing
545 that against the events to filter out spurious ones, recreating the set
546 when required. Epoll also erroneously rounds down timeouts, but gives you
547 no way to know when and by how much, so sometimes you have to busy-wait
548 because epoll returns immediately despite a nonzero timeout. And last
549 not least, it also refuses to work with some file descriptors which work
550 perfectly fine with C<select> (files, many character devices...).
552 Epoll is truly the train wreck among event poll mechanisms, a frankenpoll,
553 cobbled together in a hurry, no thought to design or interaction with
554 others. Oh, the pain, will it ever stop...
556 While stopping, setting and starting an I/O watcher in the same iteration
557 will result in some caching, there is still a system call per such
558 incident (because the same I<file descriptor> could point to a different
559 I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
560 file descriptors might not work very well if you register events for both
561 file descriptors.
563 Best performance from this backend is achieved by not unregistering all
564 watchers for a file descriptor until it has been closed, if possible,
565 i.e. keep at least one watcher active per fd at all times. Stopping and
566 starting a watcher (without re-setting it) also usually doesn't cause
567 extra overhead. A fork can both result in spurious notifications as well
568 as in libev having to destroy and recreate the epoll object, which can
569 take considerable time and thus should be avoided.
571 All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
572 faster than epoll for maybe up to a hundred file descriptors, depending on
573 the usage. So sad.
575 While nominally embeddable in other event loops, this feature is broken in
576 a lot of kernel revisions, but probably(!) works in current versions.
578 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
581 =item C<EVBACKEND_LINUXAIO> (value 64, Linux)
583 Use the Linux-specific Linux AIO (I<not> C<< aio(7) >> but C<<
584 io_submit(2) >>) event interface available in post-4.18 kernels (but libev
585 only tries to use it in 4.19+).
587 This is another Linux train wreck of an event interface.
589 If this backend works for you (as of this writing, it was very
590 experimental), it is the best event interface available on Linux and might
591 be well worth enabling it - if it isn't available in your kernel this will
592 be detected and this backend will be skipped.
594 This backend can batch oneshot requests and supports a user-space ring
595 buffer to receive events. It also doesn't suffer from most of the design
596 problems of epoll (such as not being able to remove event sources from
597 the epoll set), and generally sounds too good to be true. Because, this
598 being the Linux kernel, of course it suffers from a whole new set of
599 limitations, forcing you to fall back to epoll, inheriting all its design
600 issues.
602 For one, it is not easily embeddable (but probably could be done using
603 an event fd at some extra overhead). It also is subject to a system wide
604 limit that can be configured in F</proc/sys/fs/aio-max-nr>. If no AIO
605 requests are left, this backend will be skipped during initialisation, and
606 will switch to epoll when the loop is active.
608 Most problematic in practice, however, is that not all file descriptors
609 work with it. For example, in Linux 5.1, TCP sockets, pipes, event fds,
610 files, F</dev/null> and many others are supported, but ttys do not work
611 properly (a known bug that the kernel developers don't care about, see
612 L<>), so this is not
613 (yet?) a generic event polling interface.
615 Overall, it seems the Linux developers just don't want it to have a
616 generic event handling mechanism other than C<select> or C<poll>.
618 To work around all these problem, the current version of libev uses its
619 epoll backend as a fallback for file descriptor types that do not work. Or
620 falls back completely to epoll if the kernel acts up.
622 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
625 =item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
627 Kqueue deserves special mention, as at the time this backend was
628 implemented, it was broken on all BSDs except NetBSD (usually it doesn't
629 work reliably with anything but sockets and pipes, except on Darwin,
630 where of course it's completely useless). Unlike epoll, however, whose
631 brokenness is by design, these kqueue bugs can be (and mostly have been)
632 fixed without API changes to existing programs. For this reason it's not
633 being "auto-detected" on all platforms unless you explicitly specify it
634 in the flags (i.e. using C<EVBACKEND_KQUEUE>) or libev was compiled on a
635 known-to-be-good (-enough) system like NetBSD.
637 You still can embed kqueue into a normal poll or select backend and use it
638 only for sockets (after having made sure that sockets work with kqueue on
639 the target platform). See C<ev_embed> watchers for more info.
641 It scales in the same way as the epoll backend, but the interface to the
642 kernel is more efficient (which says nothing about its actual speed, of
643 course). While stopping, setting and starting an I/O watcher does never
644 cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
645 two event changes per incident. Support for C<fork ()> is very bad (you
646 might have to leak fds on fork, but it's more sane than epoll) and it
647 drops fds silently in similarly hard-to-detect cases.
649 This backend usually performs well under most conditions.
651 While nominally embeddable in other event loops, this doesn't work
652 everywhere, so you might need to test for this. And since it is broken
653 almost everywhere, you should only use it when you have a lot of sockets
654 (for which it usually works), by embedding it into another event loop
655 (e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
656 also broken on OS X)) and, did I mention it, using it only for sockets.
658 This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
659 C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
660 C<NOTE_EOF>.
662 =item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
664 This is not implemented yet (and might never be, unless you send me an
665 implementation). According to reports, C</dev/poll> only supports sockets
666 and is not embeddable, which would limit the usefulness of this backend
667 immensely.
669 =item C<EVBACKEND_PORT> (value 32, Solaris 10)
671 This uses the Solaris 10 event port mechanism. As with everything on Solaris,
672 it's really slow, but it still scales very well (O(active_fds)).
674 While this backend scales well, it requires one system call per active
675 file descriptor per loop iteration. For small and medium numbers of file
676 descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
677 might perform better.
679 On the positive side, this backend actually performed fully to
680 specification in all tests and is fully embeddable, which is a rare feat
681 among the OS-specific backends (I vastly prefer correctness over speed
682 hacks).
684 On the negative side, the interface is I<bizarre> - so bizarre that
685 even sun itself gets it wrong in their code examples: The event polling
686 function sometimes returns events to the caller even though an error
687 occurred, but with no indication whether it has done so or not (yes, it's
688 even documented that way) - deadly for edge-triggered interfaces where you
689 absolutely have to know whether an event occurred or not because you have
690 to re-arm the watcher.
692 Fortunately libev seems to be able to work around these idiocies.
694 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
697 =item C<EVBACKEND_ALL>
699 Try all backends (even potentially broken ones that wouldn't be tried
700 with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
703 It is definitely not recommended to use this flag, use whatever
704 C<ev_recommended_backends ()> returns, or simply do not specify a backend
705 at all.
709 Not a backend at all, but a mask to select all backend bits from a
710 C<flags> value, in case you want to mask out any backends from a flags
711 value (e.g. when modifying the C<LIBEV_FLAGS> environment variable).
713 =back
715 If one or more of the backend flags are or'ed into the flags value,
716 then only these backends will be tried (in the reverse order as listed
717 here). If none are specified, all backends in C<ev_recommended_backends
718 ()> will be tried.
720 Example: Try to create a event loop that uses epoll and nothing else.
722 struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
723 if (!epoller)
724 fatal ("no epoll found here, maybe it hides under your chair");
726 Example: Use whatever libev has to offer, but make sure that kqueue is
727 used if available.
729 struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_KQUEUE);
731 Example: Similarly, on linux, you mgiht want to take advantage of the
732 linux aio backend if possible, but fall back to something else if that
733 isn't available.
735 struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_LINUXAIO);
737 =item ev_loop_destroy (loop)
739 Destroys an event loop object (frees all memory and kernel state
740 etc.). None of the active event watchers will be stopped in the normal
741 sense, so e.g. C<ev_is_active> might still return true. It is your
742 responsibility to either stop all watchers cleanly yourself I<before>
743 calling this function, or cope with the fact afterwards (which is usually
744 the easiest thing, you can just ignore the watchers and/or C<free ()> them
745 for example).
747 Note that certain global state, such as signal state (and installed signal
748 handlers), will not be freed by this function, and related watchers (such
749 as signal and child watchers) would need to be stopped manually.
751 This function is normally used on loop objects allocated by
752 C<ev_loop_new>, but it can also be used on the default loop returned by
753 C<ev_default_loop>, in which case it is not thread-safe.
755 Note that it is not advisable to call this function on the default loop
756 except in the rare occasion where you really need to free its resources.
757 If you need dynamically allocated loops it is better to use C<ev_loop_new>
758 and C<ev_loop_destroy>.
760 =item ev_loop_fork (loop)
762 This function sets a flag that causes subsequent C<ev_run> iterations
763 to reinitialise the kernel state for backends that have one. Despite
764 the name, you can call it anytime you are allowed to start or stop
765 watchers (except inside an C<ev_prepare> callback), but it makes most
766 sense after forking, in the child process. You I<must> call it (or use
767 C<EVFLAG_FORKCHECK>) in the child before resuming or calling C<ev_run>.
769 In addition, if you want to reuse a loop (via this function or
770 C<EVFLAG_FORKCHECK>), you I<also> have to ignore C<SIGPIPE>.
772 Again, you I<have> to call it on I<any> loop that you want to re-use after
773 a fork, I<even if you do not plan to use the loop in the parent>. This is
774 because some kernel interfaces *cough* I<kqueue> *cough* do funny things
775 during fork.
777 On the other hand, you only need to call this function in the child
778 process if and only if you want to use the event loop in the child. If
779 you just fork+exec or create a new loop in the child, you don't have to
780 call it at all (in fact, C<epoll> is so badly broken that it makes a
781 difference, but libev will usually detect this case on its own and do a
782 costly reset of the backend).
784 The function itself is quite fast and it's usually not a problem to call
785 it just in case after a fork.
787 Example: Automate calling C<ev_loop_fork> on the default loop when
788 using pthreads.
790 static void
791 post_fork_child (void)
792 {
793 ev_loop_fork (EV_DEFAULT);
794 }
796 ...
797 pthread_atfork (0, 0, post_fork_child);
799 =item int ev_is_default_loop (loop)
801 Returns true when the given loop is, in fact, the default loop, and false
802 otherwise.
804 =item unsigned int ev_iteration (loop)
806 Returns the current iteration count for the event loop, which is identical
807 to the number of times libev did poll for new events. It starts at C<0>
808 and happily wraps around with enough iterations.
810 This value can sometimes be useful as a generation counter of sorts (it
811 "ticks" the number of loop iterations), as it roughly corresponds with
812 C<ev_prepare> and C<ev_check> calls - and is incremented between the
813 prepare and check phases.
815 =item unsigned int ev_depth (loop)
817 Returns the number of times C<ev_run> was entered minus the number of
818 times C<ev_run> was exited normally, in other words, the recursion depth.
820 Outside C<ev_run>, this number is zero. In a callback, this number is
821 C<1>, unless C<ev_run> was invoked recursively (or from another thread),
822 in which case it is higher.
824 Leaving C<ev_run> abnormally (setjmp/longjmp, cancelling the thread,
825 throwing an exception etc.), doesn't count as "exit" - consider this
826 as a hint to avoid such ungentleman-like behaviour unless it's really
827 convenient, in which case it is fully supported.
829 =item unsigned int ev_backend (loop)
831 Returns one of the C<EVBACKEND_*> flags indicating the event backend in
832 use.
834 =item ev_tstamp ev_now (loop)
836 Returns the current "event loop time", which is the time the event loop
837 received events and started processing them. This timestamp does not
838 change as long as callbacks are being processed, and this is also the base
839 time used for relative timers. You can treat it as the timestamp of the
840 event occurring (or more correctly, libev finding out about it).
842 =item ev_now_update (loop)
844 Establishes the current time by querying the kernel, updating the time
845 returned by C<ev_now ()> in the progress. This is a costly operation and
846 is usually done automatically within C<ev_run ()>.
848 This function is rarely useful, but when some event callback runs for a
849 very long time without entering the event loop, updating libev's idea of
850 the current time is a good idea.
852 See also L</The special problem of time updates> in the C<ev_timer> section.
854 =item ev_suspend (loop)
856 =item ev_resume (loop)
858 These two functions suspend and resume an event loop, for use when the
859 loop is not used for a while and timeouts should not be processed.
861 A typical use case would be an interactive program such as a game: When
862 the user presses C<^Z> to suspend the game and resumes it an hour later it
863 would be best to handle timeouts as if no time had actually passed while
864 the program was suspended. This can be achieved by calling C<ev_suspend>
865 in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
866 C<ev_resume> directly afterwards to resume timer processing.
868 Effectively, all C<ev_timer> watchers will be delayed by the time spend
869 between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
870 will be rescheduled (that is, they will lose any events that would have
871 occurred while suspended).
873 After calling C<ev_suspend> you B<must not> call I<any> function on the
874 given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
875 without a previous call to C<ev_suspend>.
877 Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
878 event loop time (see C<ev_now_update>).
880 =item bool ev_run (loop, int flags)
882 Finally, this is it, the event handler. This function usually is called
883 after you have initialised all your watchers and you want to start
884 handling events. It will ask the operating system for any new events, call
885 the watcher callbacks, and then repeat the whole process indefinitely: This
886 is why event loops are called I<loops>.
888 If the flags argument is specified as C<0>, it will keep handling events
889 until either no event watchers are active anymore or C<ev_break> was
890 called.
892 The return value is false if there are no more active watchers (which
893 usually means "all jobs done" or "deadlock"), and true in all other cases
894 (which usually means " you should call C<ev_run> again").
896 Please note that an explicit C<ev_break> is usually better than
897 relying on all watchers to be stopped when deciding when a program has
898 finished (especially in interactive programs), but having a program
899 that automatically loops as long as it has to and no longer by virtue
900 of relying on its watchers stopping correctly, that is truly a thing of
901 beauty.
903 This function is I<mostly> exception-safe - you can break out of a
904 C<ev_run> call by calling C<longjmp> in a callback, throwing a C++
905 exception and so on. This does not decrement the C<ev_depth> value, nor
906 will it clear any outstanding C<EVBREAK_ONE> breaks.
908 A flags value of C<EVRUN_NOWAIT> will look for new events, will handle
909 those events and any already outstanding ones, but will not wait and
910 block your process in case there are no events and will return after one
911 iteration of the loop. This is sometimes useful to poll and handle new
912 events while doing lengthy calculations, to keep the program responsive.
914 A flags value of C<EVRUN_ONCE> will look for new events (waiting if
915 necessary) and will handle those and any already outstanding ones. It
916 will block your process until at least one new event arrives (which could
917 be an event internal to libev itself, so there is no guarantee that a
918 user-registered callback will be called), and will return after one
919 iteration of the loop.
921 This is useful if you are waiting for some external event in conjunction
922 with something not expressible using other libev watchers (i.e. "roll your
923 own C<ev_run>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
924 usually a better approach for this kind of thing.
926 Here are the gory details of what C<ev_run> does (this is for your
927 understanding, not a guarantee that things will work exactly like this in
928 future versions):
930 - Increment loop depth.
931 - Reset the ev_break status.
932 - Before the first iteration, call any pending watchers.
933 LOOP:
934 - If EVFLAG_FORKCHECK was used, check for a fork.
935 - If a fork was detected (by any means), queue and call all fork watchers.
936 - Queue and call all prepare watchers.
937 - If ev_break was called, goto FINISH.
938 - If we have been forked, detach and recreate the kernel state
939 as to not disturb the other process.
940 - Update the kernel state with all outstanding changes.
941 - Update the "event loop time" (ev_now ()).
942 - Calculate for how long to sleep or block, if at all
943 (active idle watchers, EVRUN_NOWAIT or not having
944 any active watchers at all will result in not sleeping).
945 - Sleep if the I/O and timer collect interval say so.
946 - Increment loop iteration counter.
947 - Block the process, waiting for any events.
948 - Queue all outstanding I/O (fd) events.
949 - Update the "event loop time" (ev_now ()), and do time jump adjustments.
950 - Queue all expired timers.
951 - Queue all expired periodics.
952 - Queue all idle watchers with priority higher than that of pending events.
953 - Queue all check watchers.
954 - Call all queued watchers in reverse order (i.e. check watchers first).
955 Signals and child watchers are implemented as I/O watchers, and will
956 be handled here by queueing them when their watcher gets executed.
957 - If ev_break has been called, or EVRUN_ONCE or EVRUN_NOWAIT
958 were used, or there are no active watchers, goto FINISH, otherwise
959 continue with step LOOP.
961 - Reset the ev_break status iff it was EVBREAK_ONE.
962 - Decrement the loop depth.
963 - Return.
965 Example: Queue some jobs and then loop until no events are outstanding
966 anymore.
968 ... queue jobs here, make sure they register event watchers as long
969 ... as they still have work to do (even an idle watcher will do..)
970 ev_run (my_loop, 0);
971 ... jobs done or somebody called break. yeah!
973 =item ev_break (loop, how)
975 Can be used to make a call to C<ev_run> return early (but only after it
976 has processed all outstanding events). The C<how> argument must be either
977 C<EVBREAK_ONE>, which will make the innermost C<ev_run> call return, or
978 C<EVBREAK_ALL>, which will make all nested C<ev_run> calls return.
980 This "break state" will be cleared on the next call to C<ev_run>.
982 It is safe to call C<ev_break> from outside any C<ev_run> calls, too, in
983 which case it will have no effect.
985 =item ev_ref (loop)
987 =item ev_unref (loop)
989 Ref/unref can be used to add or remove a reference count on the event
990 loop: Every watcher keeps one reference, and as long as the reference
991 count is nonzero, C<ev_run> will not return on its own.
993 This is useful when you have a watcher that you never intend to
994 unregister, but that nevertheless should not keep C<ev_run> from
995 returning. In such a case, call C<ev_unref> after starting, and C<ev_ref>
996 before stopping it.
998 As an example, libev itself uses this for its internal signal pipe: It
999 is not visible to the libev user and should not keep C<ev_run> from
1000 exiting if no event watchers registered by it are active. It is also an
1001 excellent way to do this for generic recurring timers or from within
1002 third-party libraries. Just remember to I<unref after start> and I<ref
1003 before stop> (but only if the watcher wasn't active before, or was active
1004 before, respectively. Note also that libev might stop watchers itself
1005 (e.g. non-repeating timers) in which case you have to C<ev_ref>
1006 in the callback).
1008 Example: Create a signal watcher, but keep it from keeping C<ev_run>
1009 running when nothing else is active.
1011 ev_signal exitsig;
1012 ev_signal_init (&exitsig, sig_cb, SIGINT);
1013 ev_signal_start (loop, &exitsig);
1014 ev_unref (loop);
1016 Example: For some weird reason, unregister the above signal handler again.
1018 ev_ref (loop);
1019 ev_signal_stop (loop, &exitsig);
1021 =item ev_set_io_collect_interval (loop, ev_tstamp interval)
1023 =item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
1025 These advanced functions influence the time that libev will spend waiting
1026 for events. Both time intervals are by default C<0>, meaning that libev
1027 will try to invoke timer/periodic callbacks and I/O callbacks with minimum
1028 latency.
1030 Setting these to a higher value (the C<interval> I<must> be >= C<0>)
1031 allows libev to delay invocation of I/O and timer/periodic callbacks
1032 to increase efficiency of loop iterations (or to increase power-saving
1033 opportunities).
1035 The idea is that sometimes your program runs just fast enough to handle
1036 one (or very few) event(s) per loop iteration. While this makes the
1037 program responsive, it also wastes a lot of CPU time to poll for new
1038 events, especially with backends like C<select ()> which have a high
1039 overhead for the actual polling but can deliver many events at once.
1041 By setting a higher I<io collect interval> you allow libev to spend more
1042 time collecting I/O events, so you can handle more events per iteration,
1043 at the cost of increasing latency. Timeouts (both C<ev_periodic> and
1044 C<ev_timer>) will not be affected. Setting this to a non-null value will
1045 introduce an additional C<ev_sleep ()> call into most loop iterations. The
1046 sleep time ensures that libev will not poll for I/O events more often then
1047 once per this interval, on average (as long as the host time resolution is
1048 good enough).
1050 Likewise, by setting a higher I<timeout collect interval> you allow libev
1051 to spend more time collecting timeouts, at the expense of increased
1052 latency/jitter/inexactness (the watcher callback will be called
1053 later). C<ev_io> watchers will not be affected. Setting this to a non-null
1054 value will not introduce any overhead in libev.
1056 Many (busy) programs can usually benefit by setting the I/O collect
1057 interval to a value near C<0.1> or so, which is often enough for
1058 interactive servers (of course not for games), likewise for timeouts. It
1059 usually doesn't make much sense to set it to a lower value than C<0.01>,
1060 as this approaches the timing granularity of most systems. Note that if
1061 you do transactions with the outside world and you can't increase the
1062 parallelity, then this setting will limit your transaction rate (if you
1063 need to poll once per transaction and the I/O collect interval is 0.01,
1064 then you can't do more than 100 transactions per second).
1066 Setting the I<timeout collect interval> can improve the opportunity for
1067 saving power, as the program will "bundle" timer callback invocations that
1068 are "near" in time together, by delaying some, thus reducing the number of
1069 times the process sleeps and wakes up again. Another useful technique to
1070 reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
1071 they fire on, say, one-second boundaries only.
1073 Example: we only need 0.1s timeout granularity, and we wish not to poll
1074 more often than 100 times per second:
1076 ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
1077 ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
1079 =item ev_invoke_pending (loop)
1081 This call will simply invoke all pending watchers while resetting their
1082 pending state. Normally, C<ev_run> does this automatically when required,
1083 but when overriding the invoke callback this call comes handy. This
1084 function can be invoked from a watcher - this can be useful for example
1085 when you want to do some lengthy calculation and want to pass further
1086 event handling to another thread (you still have to make sure only one
1087 thread executes within C<ev_invoke_pending> or C<ev_run> of course).
1089 =item int ev_pending_count (loop)
1091 Returns the number of pending watchers - zero indicates that no watchers
1092 are pending.
1094 =item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
1096 This overrides the invoke pending functionality of the loop: Instead of
1097 invoking all pending watchers when there are any, C<ev_run> will call
1098 this callback instead. This is useful, for example, when you want to
1099 invoke the actual watchers inside another context (another thread etc.).
1101 If you want to reset the callback, use C<ev_invoke_pending> as new
1102 callback.
1104 =item ev_set_loop_release_cb (loop, void (*release)(EV_P) throw (), void (*acquire)(EV_P) throw ())
1106 Sometimes you want to share the same loop between multiple threads. This
1107 can be done relatively simply by putting mutex_lock/unlock calls around
1108 each call to a libev function.
1110 However, C<ev_run> can run an indefinite time, so it is not feasible
1111 to wait for it to return. One way around this is to wake up the event
1112 loop via C<ev_break> and C<ev_async_send>, another way is to set these
1113 I<release> and I<acquire> callbacks on the loop.
1115 When set, then C<release> will be called just before the thread is
1116 suspended waiting for new events, and C<acquire> is called just
1117 afterwards.
1119 Ideally, C<release> will just call your mutex_unlock function, and
1120 C<acquire> will just call the mutex_lock function again.
1122 While event loop modifications are allowed between invocations of
1123 C<release> and C<acquire> (that's their only purpose after all), no
1124 modifications done will affect the event loop, i.e. adding watchers will
1125 have no effect on the set of file descriptors being watched, or the time
1126 waited. Use an C<ev_async> watcher to wake up C<ev_run> when you want it
1127 to take note of any changes you made.
1129 In theory, threads executing C<ev_run> will be async-cancel safe between
1130 invocations of C<release> and C<acquire>.
1132 See also the locking example in the C<THREADS> section later in this
1133 document.
1135 =item ev_set_userdata (loop, void *data)
1137 =item void *ev_userdata (loop)
1139 Set and retrieve a single C<void *> associated with a loop. When
1140 C<ev_set_userdata> has never been called, then C<ev_userdata> returns
1141 C<0>.
1143 These two functions can be used to associate arbitrary data with a loop,
1144 and are intended solely for the C<invoke_pending_cb>, C<release> and
1145 C<acquire> callbacks described above, but of course can be (ab-)used for
1146 any other purpose as well.
1148 =item ev_verify (loop)
1150 This function only does something when C<EV_VERIFY> support has been
1151 compiled in, which is the default for non-minimal builds. It tries to go
1152 through all internal structures and checks them for validity. If anything
1153 is found to be inconsistent, it will print an error message to standard
1154 error and call C<abort ()>.
1156 This can be used to catch bugs inside libev itself: under normal
1157 circumstances, this function will never abort as of course libev keeps its
1158 data structures consistent.
1160 =back
1165 In the following description, uppercase C<TYPE> in names stands for the
1166 watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
1167 watchers and C<ev_io_start> for I/O watchers.
1169 A watcher is an opaque structure that you allocate and register to record
1170 your interest in some event. To make a concrete example, imagine you want
1171 to wait for STDIN to become readable, you would create an C<ev_io> watcher
1172 for that:
1174 static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
1175 {
1176 ev_io_stop (w);
1177 ev_break (loop, EVBREAK_ALL);
1178 }
1180 struct ev_loop *loop = ev_default_loop (0);
1182 ev_io stdin_watcher;
1184 ev_init (&stdin_watcher, my_cb);
1185 ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
1186 ev_io_start (loop, &stdin_watcher);
1188 ev_run (loop, 0);
1190 As you can see, you are responsible for allocating the memory for your
1191 watcher structures (and it is I<usually> a bad idea to do this on the
1192 stack).
1194 Each watcher has an associated watcher structure (called C<struct ev_TYPE>
1195 or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
1197 Each watcher structure must be initialised by a call to C<ev_init (watcher
1198 *, callback)>, which expects a callback to be provided. This callback is
1199 invoked each time the event occurs (or, in the case of I/O watchers, each
1200 time the event loop detects that the file descriptor given is readable
1201 and/or writable).
1203 Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
1204 macro to configure it, with arguments specific to the watcher type. There
1205 is also a macro to combine initialisation and setting in one call: C<<
1206 ev_TYPE_init (watcher *, callback, ...) >>.
1208 To make the watcher actually watch out for events, you have to start it
1209 with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
1210 *) >>), and you can stop watching for events at any time by calling the
1211 corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
1213 As long as your watcher is active (has been started but not stopped) you
1214 must not touch the values stored in it. Most specifically you must never
1215 reinitialise it or call its C<ev_TYPE_set> macro.
1217 Each and every callback receives the event loop pointer as first, the
1218 registered watcher structure as second, and a bitset of received events as
1219 third argument.
1221 The received events usually include a single bit per event type received
1222 (you can receive multiple events at the same time). The possible bit masks
1223 are:
1225 =over 4
1227 =item C<EV_READ>
1229 =item C<EV_WRITE>
1231 The file descriptor in the C<ev_io> watcher has become readable and/or
1232 writable.
1234 =item C<EV_TIMER>
1236 The C<ev_timer> watcher has timed out.
1238 =item C<EV_PERIODIC>
1240 The C<ev_periodic> watcher has timed out.
1242 =item C<EV_SIGNAL>
1244 The signal specified in the C<ev_signal> watcher has been received by a thread.
1246 =item C<EV_CHILD>
1248 The pid specified in the C<ev_child> watcher has received a status change.
1250 =item C<EV_STAT>
1252 The path specified in the C<ev_stat> watcher changed its attributes somehow.
1254 =item C<EV_IDLE>
1256 The C<ev_idle> watcher has determined that you have nothing better to do.
1258 =item C<EV_PREPARE>
1260 =item C<EV_CHECK>
1262 All C<ev_prepare> watchers are invoked just I<before> C<ev_run> starts to
1263 gather new events, and all C<ev_check> watchers are queued (not invoked)
1264 just after C<ev_run> has gathered them, but before it queues any callbacks
1265 for any received events. That means C<ev_prepare> watchers are the last
1266 watchers invoked before the event loop sleeps or polls for new events, and
1267 C<ev_check> watchers will be invoked before any other watchers of the same
1268 or lower priority within an event loop iteration.
1270 Callbacks of both watcher types can start and stop as many watchers as
1271 they want, and all of them will be taken into account (for example, a
1272 C<ev_prepare> watcher might start an idle watcher to keep C<ev_run> from
1273 blocking).
1275 =item C<EV_EMBED>
1277 The embedded event loop specified in the C<ev_embed> watcher needs attention.
1279 =item C<EV_FORK>
1281 The event loop has been resumed in the child process after fork (see
1282 C<ev_fork>).
1284 =item C<EV_CLEANUP>
1286 The event loop is about to be destroyed (see C<ev_cleanup>).
1288 =item C<EV_ASYNC>
1290 The given async watcher has been asynchronously notified (see C<ev_async>).
1292 =item C<EV_CUSTOM>
1294 Not ever sent (or otherwise used) by libev itself, but can be freely used
1295 by libev users to signal watchers (e.g. via C<ev_feed_event>).
1297 =item C<EV_ERROR>
1299 An unspecified error has occurred, the watcher has been stopped. This might
1300 happen because the watcher could not be properly started because libev
1301 ran out of memory, a file descriptor was found to be closed or any other
1302 problem. Libev considers these application bugs.
1304 You best act on it by reporting the problem and somehow coping with the
1305 watcher being stopped. Note that well-written programs should not receive
1306 an error ever, so when your watcher receives it, this usually indicates a
1307 bug in your program.
1309 Libev will usually signal a few "dummy" events together with an error, for
1310 example it might indicate that a fd is readable or writable, and if your
1311 callbacks is well-written it can just attempt the operation and cope with
1312 the error from read() or write(). This will not work in multi-threaded
1313 programs, though, as the fd could already be closed and reused for another
1314 thing, so beware.
1316 =back
1320 =over 4
1322 =item C<ev_init> (ev_TYPE *watcher, callback)
1324 This macro initialises the generic portion of a watcher. The contents
1325 of the watcher object can be arbitrary (so C<malloc> will do). Only
1326 the generic parts of the watcher are initialised, you I<need> to call
1327 the type-specific C<ev_TYPE_set> macro afterwards to initialise the
1328 type-specific parts. For each type there is also a C<ev_TYPE_init> macro
1329 which rolls both calls into one.
1331 You can reinitialise a watcher at any time as long as it has been stopped
1332 (or never started) and there are no pending events outstanding.
1334 The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
1335 int revents)>.
1337 Example: Initialise an C<ev_io> watcher in two steps.
1339 ev_io w;
1340 ev_init (&w, my_cb);
1341 ev_io_set (&w, STDIN_FILENO, EV_READ);
1343 =item C<ev_TYPE_set> (ev_TYPE *watcher, [args])
1345 This macro initialises the type-specific parts of a watcher. You need to
1346 call C<ev_init> at least once before you call this macro, but you can
1347 call C<ev_TYPE_set> any number of times. You must not, however, call this
1348 macro on a watcher that is active (it can be pending, however, which is a
1349 difference to the C<ev_init> macro).
1351 Although some watcher types do not have type-specific arguments
1352 (e.g. C<ev_prepare>) you still need to call its C<set> macro.
1354 See C<ev_init>, above, for an example.
1356 =item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
1358 This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
1359 calls into a single call. This is the most convenient method to initialise
1360 a watcher. The same limitations apply, of course.
1362 Example: Initialise and set an C<ev_io> watcher in one step.
1364 ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
1366 =item C<ev_TYPE_start> (loop, ev_TYPE *watcher)
1368 Starts (activates) the given watcher. Only active watchers will receive
1369 events. If the watcher is already active nothing will happen.
1371 Example: Start the C<ev_io> watcher that is being abused as example in this
1372 whole section.
1374 ev_io_start (EV_DEFAULT_UC, &w);
1376 =item C<ev_TYPE_stop> (loop, ev_TYPE *watcher)
1378 Stops the given watcher if active, and clears the pending status (whether
1379 the watcher was active or not).
1381 It is possible that stopped watchers are pending - for example,
1382 non-repeating timers are being stopped when they become pending - but
1383 calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
1384 pending. If you want to free or reuse the memory used by the watcher it is
1385 therefore a good idea to always call its C<ev_TYPE_stop> function.
1387 =item bool ev_is_active (ev_TYPE *watcher)
1389 Returns a true value iff the watcher is active (i.e. it has been started
1390 and not yet been stopped). As long as a watcher is active you must not modify
1391 it.
1393 =item bool ev_is_pending (ev_TYPE *watcher)
1395 Returns a true value iff the watcher is pending, (i.e. it has outstanding
1396 events but its callback has not yet been invoked). As long as a watcher
1397 is pending (but not active) you must not call an init function on it (but
1398 C<ev_TYPE_set> is safe), you must not change its priority, and you must
1399 make sure the watcher is available to libev (e.g. you cannot C<free ()>
1400 it).
1402 =item callback ev_cb (ev_TYPE *watcher)
1404 Returns the callback currently set on the watcher.
1406 =item ev_set_cb (ev_TYPE *watcher, callback)
1408 Change the callback. You can change the callback at virtually any time
1409 (modulo threads).
1411 =item ev_set_priority (ev_TYPE *watcher, int priority)
1413 =item int ev_priority (ev_TYPE *watcher)
1415 Set and query the priority of the watcher. The priority is a small
1416 integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
1417 (default: C<-2>). Pending watchers with higher priority will be invoked
1418 before watchers with lower priority, but priority will not keep watchers
1419 from being executed (except for C<ev_idle> watchers).
1421 If you need to suppress invocation when higher priority events are pending
1422 you need to look at C<ev_idle> watchers, which provide this functionality.
1424 You I<must not> change the priority of a watcher as long as it is active or
1425 pending.
1427 Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
1428 fine, as long as you do not mind that the priority value you query might
1429 or might not have been clamped to the valid range.
1431 The default priority used by watchers when no priority has been set is
1432 always C<0>, which is supposed to not be too high and not be too low :).
1434 See L</WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
1435 priorities.
1437 =item ev_invoke (loop, ev_TYPE *watcher, int revents)
1439 Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
1440 C<loop> nor C<revents> need to be valid as long as the watcher callback
1441 can deal with that fact, as both are simply passed through to the
1442 callback.
1444 =item int ev_clear_pending (loop, ev_TYPE *watcher)
1446 If the watcher is pending, this function clears its pending status and
1447 returns its C<revents> bitset (as if its callback was invoked). If the
1448 watcher isn't pending it does nothing and returns C<0>.
1450 Sometimes it can be useful to "poll" a watcher instead of waiting for its
1451 callback to be invoked, which can be accomplished with this function.
1453 =item ev_feed_event (loop, ev_TYPE *watcher, int revents)
1455 Feeds the given event set into the event loop, as if the specified event
1456 had happened for the specified watcher (which must be a pointer to an
1457 initialised but not necessarily started event watcher). Obviously you must
1458 not free the watcher as long as it has pending events.
1460 Stopping the watcher, letting libev invoke it, or calling
1461 C<ev_clear_pending> will clear the pending event, even if the watcher was
1462 not started in the first place.
1464 See also C<ev_feed_fd_event> and C<ev_feed_signal_event> for related
1465 functions that do not need a watcher.
1467 =back
1472 =head2 WATCHER STATES
1474 There are various watcher states mentioned throughout this manual -
1475 active, pending and so on. In this section these states and the rules to
1476 transition between them will be described in more detail - and while these
1477 rules might look complicated, they usually do "the right thing".
1479 =over 4
1481 =item initialised
1483 Before a watcher can be registered with the event loop it has to be
1484 initialised. This can be done with a call to C<ev_TYPE_init>, or calls to
1485 C<ev_init> followed by the watcher-specific C<ev_TYPE_set> function.
1487 In this state it is simply some block of memory that is suitable for
1488 use in an event loop. It can be moved around, freed, reused etc. at
1489 will - as long as you either keep the memory contents intact, or call
1490 C<ev_TYPE_init> again.
1492 =item started/running/active
1494 Once a watcher has been started with a call to C<ev_TYPE_start> it becomes
1495 property of the event loop, and is actively waiting for events. While in
1496 this state it cannot be accessed (except in a few documented ways), moved,
1497 freed or anything else - the only legal thing is to keep a pointer to it,
1498 and call libev functions on it that are documented to work on active watchers.
1500 =item pending
1502 If a watcher is active and libev determines that an event it is interested
1503 in has occurred (such as a timer expiring), it will become pending. It will
1504 stay in this pending state until either it is stopped or its callback is
1505 about to be invoked, so it is not normally pending inside the watcher
1506 callback.
1508 The watcher might or might not be active while it is pending (for example,
1509 an expired non-repeating timer can be pending but no longer active). If it
1510 is stopped, it can be freely accessed (e.g. by calling C<ev_TYPE_set>),
1511 but it is still property of the event loop at this time, so cannot be
1512 moved, freed or reused. And if it is active the rules described in the
1513 previous item still apply.
1515 It is also possible to feed an event on a watcher that is not active (e.g.
1516 via C<ev_feed_event>), in which case it becomes pending without being
1517 active.
1519 =item stopped
1521 A watcher can be stopped implicitly by libev (in which case it might still
1522 be pending), or explicitly by calling its C<ev_TYPE_stop> function. The
1523 latter will clear any pending state the watcher might be in, regardless
1524 of whether it was active or not, so stopping a watcher explicitly before
1525 freeing it is often a good idea.
1527 While stopped (and not pending) the watcher is essentially in the
1528 initialised state, that is, it can be reused, moved, modified in any way
1529 you wish (but when you trash the memory block, you need to C<ev_TYPE_init>
1530 it again).
1532 =back
1536 Many event loops support I<watcher priorities>, which are usually small
1537 integers that influence the ordering of event callback invocation
1538 between watchers in some way, all else being equal.
1540 In libev, watcher priorities can be set using C<ev_set_priority>. See its
1541 description for the more technical details such as the actual priority
1542 range.
1544 There are two common ways how these these priorities are being interpreted
1545 by event loops:
1547 In the more common lock-out model, higher priorities "lock out" invocation
1548 of lower priority watchers, which means as long as higher priority
1549 watchers receive events, lower priority watchers are not being invoked.
1551 The less common only-for-ordering model uses priorities solely to order
1552 callback invocation within a single event loop iteration: Higher priority
1553 watchers are invoked before lower priority ones, but they all get invoked
1554 before polling for new events.
1556 Libev uses the second (only-for-ordering) model for all its watchers
1557 except for idle watchers (which use the lock-out model).
1559 The rationale behind this is that implementing the lock-out model for
1560 watchers is not well supported by most kernel interfaces, and most event
1561 libraries will just poll for the same events again and again as long as
1562 their callbacks have not been executed, which is very inefficient in the
1563 common case of one high-priority watcher locking out a mass of lower
1564 priority ones.
1566 Static (ordering) priorities are most useful when you have two or more
1567 watchers handling the same resource: a typical usage example is having an
1568 C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
1569 timeouts. Under load, data might be received while the program handles
1570 other jobs, but since timers normally get invoked first, the timeout
1571 handler will be executed before checking for data. In that case, giving
1572 the timer a lower priority than the I/O watcher ensures that I/O will be
1573 handled first even under adverse conditions (which is usually, but not
1574 always, what you want).
1576 Since idle watchers use the "lock-out" model, meaning that idle watchers
1577 will only be executed when no same or higher priority watchers have
1578 received events, they can be used to implement the "lock-out" model when
1579 required.
1581 For example, to emulate how many other event libraries handle priorities,
1582 you can associate an C<ev_idle> watcher to each such watcher, and in
1583 the normal watcher callback, you just start the idle watcher. The real
1584 processing is done in the idle watcher callback. This causes libev to
1585 continuously poll and process kernel event data for the watcher, but when
1586 the lock-out case is known to be rare (which in turn is rare :), this is
1587 workable.
1589 Usually, however, the lock-out model implemented that way will perform
1590 miserably under the type of load it was designed to handle. In that case,
1591 it might be preferable to stop the real watcher before starting the
1592 idle watcher, so the kernel will not have to process the event in case
1593 the actual processing will be delayed for considerable time.
1595 Here is an example of an I/O watcher that should run at a strictly lower
1596 priority than the default, and which should only process data when no
1597 other events are pending:
1599 ev_idle idle; // actual processing watcher
1600 ev_io io; // actual event watcher
1602 static void
1603 io_cb (EV_P_ ev_io *w, int revents)
1604 {
1605 // stop the I/O watcher, we received the event, but
1606 // are not yet ready to handle it.
1607 ev_io_stop (EV_A_ w);
1609 // start the idle watcher to handle the actual event.
1610 // it will not be executed as long as other watchers
1611 // with the default priority are receiving events.
1612 ev_idle_start (EV_A_ &idle);
1613 }
1615 static void
1616 idle_cb (EV_P_ ev_idle *w, int revents)
1617 {
1618 // actual processing
1619 read (STDIN_FILENO, ...);
1621 // have to start the I/O watcher again, as
1622 // we have handled the event
1623 ev_io_start (EV_P_ &io);
1624 }
1626 // initialisation
1627 ev_idle_init (&idle, idle_cb);
1628 ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
1629 ev_io_start (EV_DEFAULT_ &io);
1631 In the "real" world, it might also be beneficial to start a timer, so that
1632 low-priority connections can not be locked out forever under load. This
1633 enables your program to keep a lower latency for important connections
1634 during short periods of high load, while not completely locking out less
1635 important ones.
1638 =head1 WATCHER TYPES
1640 This section describes each watcher in detail, but will not repeat
1641 information given in the last section. Any initialisation/set macros,
1642 functions and members specific to the watcher type are explained.
1644 Members are additionally marked with either I<[read-only]>, meaning that,
1645 while the watcher is active, you can look at the member and expect some
1646 sensible content, but you must not modify it (you can modify it while the
1647 watcher is stopped to your hearts content), or I<[read-write]>, which
1648 means you can expect it to have some sensible content while the watcher
1649 is active, but you can also modify it. Modifying it may not do something
1650 sensible or take immediate effect (or do anything at all), but libev will
1651 not crash or malfunction in any way.
1654 =head2 C<ev_io> - is this file descriptor readable or writable?
1656 I/O watchers check whether a file descriptor is readable or writable
1657 in each iteration of the event loop, or, more precisely, when reading
1658 would not block the process and writing would at least be able to write
1659 some data. This behaviour is called level-triggering because you keep
1660 receiving events as long as the condition persists. Remember you can stop
1661 the watcher if you don't want to act on the event and neither want to
1662 receive future events.
1664 In general you can register as many read and/or write event watchers per
1665 fd as you want (as long as you don't confuse yourself). Setting all file
1666 descriptors to non-blocking mode is also usually a good idea (but not
1667 required if you know what you are doing).
1669 Another thing you have to watch out for is that it is quite easy to
1670 receive "spurious" readiness notifications, that is, your callback might
1671 be called with C<EV_READ> but a subsequent C<read>(2) will actually block
1672 because there is no data. It is very easy to get into this situation even
1673 with a relatively standard program structure. Thus it is best to always
1674 use non-blocking I/O: An extra C<read>(2) returning C<EAGAIN> is far
1675 preferable to a program hanging until some data arrives.
1677 If you cannot run the fd in non-blocking mode (for example you should
1678 not play around with an Xlib connection), then you have to separately
1679 re-test whether a file descriptor is really ready with a known-to-be good
1680 interface such as poll (fortunately in the case of Xlib, it already does
1681 this on its own, so its quite safe to use). Some people additionally
1682 use C<SIGALRM> and an interval timer, just to be sure you won't block
1683 indefinitely.
1685 But really, best use non-blocking mode.
1687 =head3 The special problem of disappearing file descriptors
1689 Some backends (e.g. kqueue, epoll, linuxaio) need to be told about closing
1690 a file descriptor (either due to calling C<close> explicitly or any other
1691 means, such as C<dup2>). The reason is that you register interest in some
1692 file descriptor, but when it goes away, the operating system will silently
1693 drop this interest. If another file descriptor with the same number then
1694 is registered with libev, there is no efficient way to see that this is,
1695 in fact, a different file descriptor.
1697 To avoid having to explicitly tell libev about such cases, libev follows
1698 the following policy: Each time C<ev_io_set> is being called, libev
1699 will assume that this is potentially a new file descriptor, otherwise
1700 it is assumed that the file descriptor stays the same. That means that
1701 you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
1702 descriptor even if the file descriptor number itself did not change.
1704 This is how one would do it normally anyway, the important point is that
1705 the libev application should not optimise around libev but should leave
1706 optimisations to libev.
1708 =head3 The special problem of dup'ed file descriptors
1710 Some backends (e.g. epoll), cannot register events for file descriptors,
1711 but only events for the underlying file descriptions. That means when you
1712 have C<dup ()>'ed file descriptors or weirder constellations, and register
1713 events for them, only one file descriptor might actually receive events.
1715 There is no workaround possible except not registering events
1716 for potentially C<dup ()>'ed file descriptors, or to resort to
1719 =head3 The special problem of files
1721 Many people try to use C<select> (or libev) on file descriptors
1722 representing files, and expect it to become ready when their program
1723 doesn't block on disk accesses (which can take a long time on their own).
1725 However, this cannot ever work in the "expected" way - you get a readiness
1726 notification as soon as the kernel knows whether and how much data is
1727 there, and in the case of open files, that's always the case, so you
1728 always get a readiness notification instantly, and your read (or possibly
1729 write) will still block on the disk I/O.
1731 Another way to view it is that in the case of sockets, pipes, character
1732 devices and so on, there is another party (the sender) that delivers data
1733 on its own, but in the case of files, there is no such thing: the disk
1734 will not send data on its own, simply because it doesn't know what you
1735 wish to read - you would first have to request some data.
1737 Since files are typically not-so-well supported by advanced notification
1738 mechanism, libev tries hard to emulate POSIX behaviour with respect
1739 to files, even though you should not use it. The reason for this is
1740 convenience: sometimes you want to watch STDIN or STDOUT, which is
1741 usually a tty, often a pipe, but also sometimes files or special devices
1742 (for example, C<epoll> on Linux works with F</dev/random> but not with
1743 F</dev/urandom>), and even though the file might better be served with
1744 asynchronous I/O instead of with non-blocking I/O, it is still useful when
1745 it "just works" instead of freezing.
1747 So avoid file descriptors pointing to files when you know it (e.g. use
1748 libeio), but use them when it is convenient, e.g. for STDIN/STDOUT, or
1749 when you rarely read from a file instead of from a socket, and want to
1750 reuse the same code path.
1752 =head3 The special problem of fork
1754 Some backends (epoll, kqueue, linuxaio, iouring) do not support C<fork ()>
1755 at all or exhibit useless behaviour. Libev fully supports fork, but needs
1756 to be told about it in the child if you want to continue to use it in the
1757 child.
1759 To support fork in your child processes, you have to call C<ev_loop_fork
1760 ()> after a fork in the child, enable C<EVFLAG_FORKCHECK>, or resort to
1763 =head3 The special problem of SIGPIPE
1765 While not really specific to libev, it is easy to forget about C<SIGPIPE>:
1766 when writing to a pipe whose other end has been closed, your program gets
1767 sent a SIGPIPE, which, by default, aborts your program. For most programs
1768 this is sensible behaviour, for daemons, this is usually undesirable.
1770 So when you encounter spurious, unexplained daemon exits, make sure you
1771 ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
1772 somewhere, as that would have given you a big clue).
1774 =head3 The special problem of accept()ing when you can't
1776 Many implementations of the POSIX C<accept> function (for example,
1777 found in post-2004 Linux) have the peculiar behaviour of not removing a
1778 connection from the pending queue in all error cases.
1780 For example, larger servers often run out of file descriptors (because
1781 of resource limits), causing C<accept> to fail with C<ENFILE> but not
1782 rejecting the connection, leading to libev signalling readiness on
1783 the next iteration again (the connection still exists after all), and
1784 typically causing the program to loop at 100% CPU usage.
1786 Unfortunately, the set of errors that cause this issue differs between
1787 operating systems, there is usually little the app can do to remedy the
1788 situation, and no known thread-safe method of removing the connection to
1789 cope with overload is known (to me).
1791 One of the easiest ways to handle this situation is to just ignore it
1792 - when the program encounters an overload, it will just loop until the
1793 situation is over. While this is a form of busy waiting, no OS offers an
1794 event-based way to handle this situation, so it's the best one can do.
1796 A better way to handle the situation is to log any errors other than
1797 C<EAGAIN> and C<EWOULDBLOCK>, making sure not to flood the log with such
1798 messages, and continue as usual, which at least gives the user an idea of
1799 what could be wrong ("raise the ulimit!"). For extra points one could stop
1800 the C<ev_io> watcher on the listening fd "for a while", which reduces CPU
1801 usage.
1803 If your program is single-threaded, then you could also keep a dummy file
1804 descriptor for overload situations (e.g. by opening F</dev/null>), and
1805 when you run into C<ENFILE> or C<EMFILE>, close it, run C<accept>,
1806 close that fd, and create a new dummy fd. This will gracefully refuse
1807 clients under typical overload conditions.
1809 The last way to handle it is to simply log the error and C<exit>, as
1810 is often done with C<malloc> failures, but this results in an easy
1811 opportunity for a DoS attack.
1813 =head3 Watcher-Specific Functions
1815 =over 4
1817 =item ev_io_init (ev_io *, callback, int fd, int events)
1819 =item ev_io_set (ev_io *, int fd, int events)
1821 Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
1822 receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
1823 C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
1825 =item int fd [read-only]
1827 The file descriptor being watched.
1829 =item int events [read-only]
1831 The events being watched.
1833 =back
1835 =head3 Examples
1837 Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
1838 readable, but only once. Since it is likely line-buffered, you could
1839 attempt to read a whole line in the callback.
1841 static void
1842 stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
1843 {
1844 ev_io_stop (loop, w);
1845 .. read from stdin here (or from w->fd) and handle any I/O errors
1846 }
1848 ...
1849 struct ev_loop *loop = ev_default_init (0);
1850 ev_io stdin_readable;
1851 ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
1852 ev_io_start (loop, &stdin_readable);
1853 ev_run (loop, 0);
1856 =head2 C<ev_timer> - relative and optionally repeating timeouts
1858 Timer watchers are simple relative timers that generate an event after a
1859 given time, and optionally repeating in regular intervals after that.
1861 The timers are based on real time, that is, if you register an event that
1862 times out after an hour and you reset your system clock to January last
1863 year, it will still time out after (roughly) one hour. "Roughly" because
1864 detecting time jumps is hard, and some inaccuracies are unavoidable (the
1865 monotonic clock option helps a lot here).
1867 The callback is guaranteed to be invoked only I<after> its timeout has
1868 passed (not I<at>, so on systems with very low-resolution clocks this
1869 might introduce a small delay, see "the special problem of being too
1870 early", below). If multiple timers become ready during the same loop
1871 iteration then the ones with earlier time-out values are invoked before
1872 ones of the same priority with later time-out values (but this is no
1873 longer true when a callback calls C<ev_run> recursively).
1875 =head3 Be smart about timeouts
1877 Many real-world problems involve some kind of timeout, usually for error
1878 recovery. A typical example is an HTTP request - if the other side hangs,
1879 you want to raise some error after a while.
1881 What follows are some ways to handle this problem, from obvious and
1882 inefficient to smart and efficient.
1884 In the following, a 60 second activity timeout is assumed - a timeout that
1885 gets reset to 60 seconds each time there is activity (e.g. each time some
1886 data or other life sign was received).
1888 =over 4
1890 =item 1. Use a timer and stop, reinitialise and start it on activity.
1892 This is the most obvious, but not the most simple way: In the beginning,
1893 start the watcher:
1895 ev_timer_init (timer, callback, 60., 0.);
1896 ev_timer_start (loop, timer);
1898 Then, each time there is some activity, C<ev_timer_stop> it, initialise it
1899 and start it again:
1901 ev_timer_stop (loop, timer);
1902 ev_timer_set (timer, 60., 0.);
1903 ev_timer_start (loop, timer);
1905 This is relatively simple to implement, but means that each time there is
1906 some activity, libev will first have to remove the timer from its internal
1907 data structure and then add it again. Libev tries to be fast, but it's
1908 still not a constant-time operation.
1910 =item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
1912 This is the easiest way, and involves using C<ev_timer_again> instead of
1913 C<ev_timer_start>.
1915 To implement this, configure an C<ev_timer> with a C<repeat> value
1916 of C<60> and then call C<ev_timer_again> at start and each time you
1917 successfully read or write some data. If you go into an idle state where
1918 you do not expect data to travel on the socket, you can C<ev_timer_stop>
1919 the timer, and C<ev_timer_again> will automatically restart it if need be.
1921 That means you can ignore both the C<ev_timer_start> function and the
1922 C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
1923 member and C<ev_timer_again>.
1925 At start:
1927 ev_init (timer, callback);
1928 timer->repeat = 60.;
1929 ev_timer_again (loop, timer);
1931 Each time there is some activity:
1933 ev_timer_again (loop, timer);
1935 It is even possible to change the time-out on the fly, regardless of
1936 whether the watcher is active or not:
1938 timer->repeat = 30.;
1939 ev_timer_again (loop, timer);
1941 This is slightly more efficient then stopping/starting the timer each time
1942 you want to modify its timeout value, as libev does not have to completely
1943 remove and re-insert the timer from/into its internal data structure.
1945 It is, however, even simpler than the "obvious" way to do it.
1947 =item 3. Let the timer time out, but then re-arm it as required.
1949 This method is more tricky, but usually most efficient: Most timeouts are
1950 relatively long compared to the intervals between other activity - in
1951 our example, within 60 seconds, there are usually many I/O events with
1952 associated activity resets.
1954 In this case, it would be more efficient to leave the C<ev_timer> alone,
1955 but remember the time of last activity, and check for a real timeout only
1956 within the callback:
1958 ev_tstamp timeout = 60.;
1959 ev_tstamp last_activity; // time of last activity
1960 ev_timer timer;
1962 static void
1963 callback (EV_P_ ev_timer *w, int revents)
1964 {
1965 // calculate when the timeout would happen
1966 ev_tstamp after = last_activity - ev_now (EV_A) + timeout;
1968 // if negative, it means we the timeout already occurred
1969 if (after < 0.)
1970 {
1971 // timeout occurred, take action
1972 }
1973 else
1974 {
1975 // callback was invoked, but there was some recent
1976 // activity. simply restart the timer to time out
1977 // after "after" seconds, which is the earliest time
1978 // the timeout can occur.
1979 ev_timer_set (w, after, 0.);
1980 ev_timer_start (EV_A_ w);
1981 }
1982 }
1984 To summarise the callback: first calculate in how many seconds the
1985 timeout will occur (by calculating the absolute time when it would occur,
1986 C<last_activity + timeout>, and subtracting the current time, C<ev_now
1987 (EV_A)> from that).
1989 If this value is negative, then we are already past the timeout, i.e. we
1990 timed out, and need to do whatever is needed in this case.
1992 Otherwise, we now the earliest time at which the timeout would trigger,
1993 and simply start the timer with this timeout value.
1995 In other words, each time the callback is invoked it will check whether
1996 the timeout occurred. If not, it will simply reschedule itself to check
1997 again at the earliest time it could time out. Rinse. Repeat.
1999 This scheme causes more callback invocations (about one every 60 seconds
2000 minus half the average time between activity), but virtually no calls to
2001 libev to change the timeout.
2003 To start the machinery, simply initialise the watcher and set
2004 C<last_activity> to the current time (meaning there was some activity just
2005 now), then call the callback, which will "do the right thing" and start
2006 the timer:
2008 last_activity = ev_now (EV_A);
2009 ev_init (&timer, callback);
2010 callback (EV_A_ &timer, 0);
2012 When there is some activity, simply store the current time in
2013 C<last_activity>, no libev calls at all:
2015 if (activity detected)
2016 last_activity = ev_now (EV_A);
2018 When your timeout value changes, then the timeout can be changed by simply
2019 providing a new value, stopping the timer and calling the callback, which
2020 will again do the right thing (for example, time out immediately :).
2022 timeout = new_value;
2023 ev_timer_stop (EV_A_ &timer);
2024 callback (EV_A_ &timer, 0);
2026 This technique is slightly more complex, but in most cases where the
2027 time-out is unlikely to be triggered, much more efficient.
2029 =item 4. Wee, just use a double-linked list for your timeouts.
2031 If there is not one request, but many thousands (millions...), all
2032 employing some kind of timeout with the same timeout value, then one can
2033 do even better:
2035 When starting the timeout, calculate the timeout value and put the timeout
2036 at the I<end> of the list.
2038 Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
2039 the list is expected to fire (for example, using the technique #3).
2041 When there is some activity, remove the timer from the list, recalculate
2042 the timeout, append it to the end of the list again, and make sure to
2043 update the C<ev_timer> if it was taken from the beginning of the list.
2045 This way, one can manage an unlimited number of timeouts in O(1) time for
2046 starting, stopping and updating the timers, at the expense of a major
2047 complication, and having to use a constant timeout. The constant timeout
2048 ensures that the list stays sorted.
2050 =back
2052 So which method the best?
2054 Method #2 is a simple no-brain-required solution that is adequate in most
2055 situations. Method #3 requires a bit more thinking, but handles many cases
2056 better, and isn't very complicated either. In most case, choosing either
2057 one is fine, with #3 being better in typical situations.
2059 Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
2060 rather complicated, but extremely efficient, something that really pays
2061 off after the first million or so of active timers, i.e. it's usually
2062 overkill :)
2064 =head3 The special problem of being too early
2066 If you ask a timer to call your callback after three seconds, then
2067 you expect it to be invoked after three seconds - but of course, this
2068 cannot be guaranteed to infinite precision. Less obviously, it cannot be
2069 guaranteed to any precision by libev - imagine somebody suspending the
2070 process with a STOP signal for a few hours for example.
2072 So, libev tries to invoke your callback as soon as possible I<after> the
2073 delay has occurred, but cannot guarantee this.
2075 A less obvious failure mode is calling your callback too early: many event
2076 loops compare timestamps with a "elapsed delay >= requested delay", but
2077 this can cause your callback to be invoked much earlier than you would
2078 expect.
2080 To see why, imagine a system with a clock that only offers full second
2081 resolution (think windows if you can't come up with a broken enough OS
2082 yourself). If you schedule a one-second timer at the time 500.9, then the
2083 event loop will schedule your timeout to elapse at a system time of 500
2084 (500.9 truncated to the resolution) + 1, or 501.
2086 If an event library looks at the timeout 0.1s later, it will see "501 >=
2087 501" and invoke the callback 0.1s after it was started, even though a
2088 one-second delay was requested - this is being "too early", despite best
2089 intentions.
2091 This is the reason why libev will never invoke the callback if the elapsed
2092 delay equals the requested delay, but only when the elapsed delay is
2093 larger than the requested delay. In the example above, libev would only invoke
2094 the callback at system time 502, or 1.1s after the timer was started.
2096 So, while libev cannot guarantee that your callback will be invoked
2097 exactly when requested, it I<can> and I<does> guarantee that the requested
2098 delay has actually elapsed, or in other words, it always errs on the "too
2099 late" side of things.
2101 =head3 The special problem of time updates
2103 Establishing the current time is a costly operation (it usually takes
2104 at least one system call): EV therefore updates its idea of the current
2105 time only before and after C<ev_run> collects new events, which causes a
2106 growing difference between C<ev_now ()> and C<ev_time ()> when handling
2107 lots of events in one iteration.
2109 The relative timeouts are calculated relative to the C<ev_now ()>
2110 time. This is usually the right thing as this timestamp refers to the time
2111 of the event triggering whatever timeout you are modifying/starting. If
2112 you suspect event processing to be delayed and you I<need> to base the
2113 timeout on the current time, use something like the following to adjust
2114 for it:
2116 ev_timer_set (&timer, after + (ev_time () - ev_now ()), 0.);
2118 If the event loop is suspended for a long time, you can also force an
2119 update of the time returned by C<ev_now ()> by calling C<ev_now_update
2120 ()>, although that will push the event time of all outstanding events
2121 further into the future.
2123 =head3 The special problem of unsynchronised clocks
2125 Modern systems have a variety of clocks - libev itself uses the normal
2126 "wall clock" clock and, if available, the monotonic clock (to avoid time
2127 jumps).
2129 Neither of these clocks is synchronised with each other or any other clock
2130 on the system, so C<ev_time ()> might return a considerably different time
2131 than C<gettimeofday ()> or C<time ()>. On a GNU/Linux system, for example,
2132 a call to C<gettimeofday> might return a second count that is one higher
2133 than a directly following call to C<time>.
2135 The moral of this is to only compare libev-related timestamps with
2136 C<ev_time ()> and C<ev_now ()>, at least if you want better precision than
2137 a second or so.
2139 One more problem arises due to this lack of synchronisation: if libev uses
2140 the system monotonic clock and you compare timestamps from C<ev_time>
2141 or C<ev_now> from when you started your timer and when your callback is
2142 invoked, you will find that sometimes the callback is a bit "early".
2144 This is because C<ev_timer>s work in real time, not wall clock time, so
2145 libev makes sure your callback is not invoked before the delay happened,
2146 I<measured according to the real time>, not the system clock.
2148 If your timeouts are based on a physical timescale (e.g. "time out this
2149 connection after 100 seconds") then this shouldn't bother you as it is
2150 exactly the right behaviour.
2152 If you want to compare wall clock/system timestamps to your timers, then
2153 you need to use C<ev_periodic>s, as these are based on the wall clock
2154 time, where your comparisons will always generate correct results.
2156 =head3 The special problems of suspended animation
2158 When you leave the server world it is quite customary to hit machines that
2159 can suspend/hibernate - what happens to the clocks during such a suspend?
2161 Some quick tests made with a Linux 2.6.28 indicate that a suspend freezes
2162 all processes, while the clocks (C<times>, C<CLOCK_MONOTONIC>) continue
2163 to run until the system is suspended, but they will not advance while the
2164 system is suspended. That means, on resume, it will be as if the program
2165 was frozen for a few seconds, but the suspend time will not be counted
2166 towards C<ev_timer> when a monotonic clock source is used. The real time
2167 clock advanced as expected, but if it is used as sole clocksource, then a
2168 long suspend would be detected as a time jump by libev, and timers would
2169 be adjusted accordingly.
2171 I would not be surprised to see different behaviour in different between
2172 operating systems, OS versions or even different hardware.
2174 The other form of suspend (job control, or sending a SIGSTOP) will see a
2175 time jump in the monotonic clocks and the realtime clock. If the program
2176 is suspended for a very long time, and monotonic clock sources are in use,
2177 then you can expect C<ev_timer>s to expire as the full suspension time
2178 will be counted towards the timers. When no monotonic clock source is in
2179 use, then libev will again assume a timejump and adjust accordingly.
2181 It might be beneficial for this latter case to call C<ev_suspend>
2182 and C<ev_resume> in code that handles C<SIGTSTP>, to at least get
2183 deterministic behaviour in this case (you can do nothing against
2184 C<SIGSTOP>).
2186 =head3 Watcher-Specific Functions and Data Members
2188 =over 4
2190 =item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
2192 =item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
2194 Configure the timer to trigger after C<after> seconds (fractional and
2195 negative values are supported). If C<repeat> is C<0.>, then it will
2196 automatically be stopped once the timeout is reached. If it is positive,
2197 then the timer will automatically be configured to trigger again C<repeat>
2198 seconds later, again, and again, until stopped manually.
2200 The timer itself will do a best-effort at avoiding drift, that is, if
2201 you configure a timer to trigger every 10 seconds, then it will normally
2202 trigger at exactly 10 second intervals. If, however, your program cannot
2203 keep up with the timer (because it takes longer than those 10 seconds to
2204 do stuff) the timer will not fire more than once per event loop iteration.
2206 =item ev_timer_again (loop, ev_timer *)
2208 This will act as if the timer timed out, and restarts it again if it is
2209 repeating. It basically works like calling C<ev_timer_stop>, updating the
2210 timeout to the C<repeat> value and calling C<ev_timer_start>.
2212 The exact semantics are as in the following rules, all of which will be
2213 applied to the watcher:
2215 =over 4
2217 =item If the timer is pending, the pending status is always cleared.
2219 =item If the timer is started but non-repeating, stop it (as if it timed
2220 out, without invoking it).
2222 =item If the timer is repeating, make the C<repeat> value the new timeout
2223 and start the timer, if necessary.
2225 =back
2227 This sounds a bit complicated, see L</Be smart about timeouts>, above, for a
2228 usage example.
2230 =item ev_tstamp ev_timer_remaining (loop, ev_timer *)
2232 Returns the remaining time until a timer fires. If the timer is active,
2233 then this time is relative to the current event loop time, otherwise it's
2234 the timeout value currently configured.
2236 That is, after an C<ev_timer_set (w, 5, 7)>, C<ev_timer_remaining> returns
2237 C<5>. When the timer is started and one second passes, C<ev_timer_remaining>
2238 will return C<4>. When the timer expires and is restarted, it will return
2239 roughly C<7> (likely slightly less as callback invocation takes some time,
2240 too), and so on.
2242 =item ev_tstamp repeat [read-write]
2244 The current C<repeat> value. Will be used each time the watcher times out
2245 or C<ev_timer_again> is called, and determines the next timeout (if any),
2246 which is also when any modifications are taken into account.
2248 =back
2250 =head3 Examples
2252 Example: Create a timer that fires after 60 seconds.
2254 static void
2255 one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
2256 {
2257 .. one minute over, w is actually stopped right here
2258 }
2260 ev_timer mytimer;
2261 ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
2262 ev_timer_start (loop, &mytimer);
2264 Example: Create a timeout timer that times out after 10 seconds of
2265 inactivity.
2267 static void
2268 timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
2269 {
2270 .. ten seconds without any activity
2271 }
2273 ev_timer mytimer;
2274 ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
2275 ev_timer_again (&mytimer); /* start timer */
2276 ev_run (loop, 0);
2278 // and in some piece of code that gets executed on any "activity":
2279 // reset the timeout to start ticking again at 10 seconds
2280 ev_timer_again (&mytimer);
2283 =head2 C<ev_periodic> - to cron or not to cron?
2285 Periodic watchers are also timers of a kind, but they are very versatile
2286 (and unfortunately a bit complex).
2288 Unlike C<ev_timer>, periodic watchers are not based on real time (or
2289 relative time, the physical time that passes) but on wall clock time
2290 (absolute time, the thing you can read on your calendar or clock). The
2291 difference is that wall clock time can run faster or slower than real
2292 time, and time jumps are not uncommon (e.g. when you adjust your
2293 wrist-watch).
2295 You can tell a periodic watcher to trigger after some specific point
2296 in time: for example, if you tell a periodic watcher to trigger "in 10
2297 seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
2298 not a delay) and then reset your system clock to January of the previous
2299 year, then it will take a year or more to trigger the event (unlike an
2300 C<ev_timer>, which would still trigger roughly 10 seconds after starting
2301 it, as it uses a relative timeout).
2303 C<ev_periodic> watchers can also be used to implement vastly more complex
2304 timers, such as triggering an event on each "midnight, local time", or
2305 other complicated rules. This cannot easily be done with C<ev_timer>
2306 watchers, as those cannot react to time jumps.
2308 As with timers, the callback is guaranteed to be invoked only when the
2309 point in time where it is supposed to trigger has passed. If multiple
2310 timers become ready during the same loop iteration then the ones with
2311 earlier time-out values are invoked before ones with later time-out values
2312 (but this is no longer true when a callback calls C<ev_run> recursively).
2314 =head3 Watcher-Specific Functions and Data Members
2316 =over 4
2318 =item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
2320 =item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
2322 Lots of arguments, let's sort it out... There are basically three modes of
2323 operation, and we will explain them from simplest to most complex:
2325 =over 4
2327 =item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
2329 In this configuration the watcher triggers an event after the wall clock
2330 time C<offset> has passed. It will not repeat and will not adjust when a
2331 time jump occurs, that is, if it is to be run at January 1st 2011 then it
2332 will be stopped and invoked when the system clock reaches or surpasses
2333 this point in time.
2335 =item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
2337 In this mode the watcher will always be scheduled to time out at the next
2338 C<offset + N * interval> time (for some integer N, which can also be
2339 negative) and then repeat, regardless of any time jumps. The C<offset>
2340 argument is merely an offset into the C<interval> periods.
2342 This can be used to create timers that do not drift with respect to the
2343 system clock, for example, here is an C<ev_periodic> that triggers each
2344 hour, on the hour (with respect to UTC):
2346 ev_periodic_set (&periodic, 0., 3600., 0);
2348 This doesn't mean there will always be 3600 seconds in between triggers,
2349 but only that the callback will be called when the system time shows a
2350 full hour (UTC), or more correctly, when the system time is evenly divisible
2351 by 3600.
2353 Another way to think about it (for the mathematically inclined) is that
2354 C<ev_periodic> will try to run the callback in this mode at the next possible
2355 time where C<time = offset (mod interval)>, regardless of any time jumps.
2357 The C<interval> I<MUST> be positive, and for numerical stability, the
2358 interval value should be higher than C<1/8192> (which is around 100
2359 microseconds) and C<offset> should be higher than C<0> and should have
2360 at most a similar magnitude as the current time (say, within a factor of
2361 ten). Typical values for offset are, in fact, C<0> or something between
2362 C<0> and C<interval>, which is also the recommended range.
2364 Note also that there is an upper limit to how often a timer can fire (CPU
2365 speed for example), so if C<interval> is very small then timing stability
2366 will of course deteriorate. Libev itself tries to be exact to be about one
2367 millisecond (if the OS supports it and the machine is fast enough).
2369 =item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
2371 In this mode the values for C<interval> and C<offset> are both being
2372 ignored. Instead, each time the periodic watcher gets scheduled, the
2373 reschedule callback will be called with the watcher as first, and the
2374 current time as second argument.
2376 NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, ever,
2377 or make ANY other event loop modifications whatsoever, unless explicitly
2378 allowed by documentation here>.
2380 If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
2381 it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
2382 only event loop modification you are allowed to do).
2384 The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
2385 *w, ev_tstamp now)>, e.g.:
2387 static ev_tstamp
2388 my_rescheduler (ev_periodic *w, ev_tstamp now)
2389 {
2390 return now + 60.;
2391 }
2393 It must return the next time to trigger, based on the passed time value
2394 (that is, the lowest time value larger than to the second argument). It
2395 will usually be called just before the callback will be triggered, but
2396 might be called at other times, too.
2398 NOTE: I<< This callback must always return a time that is higher than or
2399 equal to the passed C<now> value >>.
2401 This can be used to create very complex timers, such as a timer that
2402 triggers on "next midnight, local time". To do this, you would calculate
2403 the next midnight after C<now> and return the timestamp value for
2404 this. Here is a (completely untested, no error checking) example on how to
2405 do this:
2407 #include <time.h>
2409 static ev_tstamp
2410 my_rescheduler (ev_periodic *w, ev_tstamp now)
2411 {
2412 time_t tnow = (time_t)now;
2413 struct tm tm;
2414 localtime_r (&tnow, &tm);
2416 tm.tm_sec = tm.tm_min = tm.tm_hour = 0; // midnight current day
2417 ++tm.tm_mday; // midnight next day
2419 return mktime (&tm);
2420 }
2422 Note: this code might run into trouble on days that have more then two
2423 midnights (beginning and end).
2425 =back
2427 =item ev_periodic_again (loop, ev_periodic *)
2429 Simply stops and restarts the periodic watcher again. This is only useful
2430 when you changed some parameters or the reschedule callback would return
2431 a different time than the last time it was called (e.g. in a crond like
2432 program when the crontabs have changed).
2434 =item ev_tstamp ev_periodic_at (ev_periodic *)
2436 When active, returns the absolute time that the watcher is supposed
2437 to trigger next. This is not the same as the C<offset> argument to
2438 C<ev_periodic_set>, but indeed works even in interval and manual
2439 rescheduling modes.
2441 =item ev_tstamp offset [read-write]
2443 When repeating, this contains the offset value, otherwise this is the
2444 absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
2445 although libev might modify this value for better numerical stability).
2447 Can be modified any time, but changes only take effect when the periodic
2448 timer fires or C<ev_periodic_again> is being called.
2450 =item ev_tstamp interval [read-write]
2452 The current interval value. Can be modified any time, but changes only
2453 take effect when the periodic timer fires or C<ev_periodic_again> is being
2454 called.
2456 =item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
2458 The current reschedule callback, or C<0>, if this functionality is
2459 switched off. Can be changed any time, but changes only take effect when
2460 the periodic timer fires or C<ev_periodic_again> is being called.
2462 =back
2464 =head3 Examples
2466 Example: Call a callback every hour, or, more precisely, whenever the
2467 system time is divisible by 3600. The callback invocation times have
2468 potentially a lot of jitter, but good long-term stability.
2470 static void
2471 clock_cb (struct ev_loop *loop, ev_periodic *w, int revents)
2472 {
2473 ... its now a full hour (UTC, or TAI or whatever your clock follows)
2474 }
2476 ev_periodic hourly_tick;
2477 ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
2478 ev_periodic_start (loop, &hourly_tick);
2480 Example: The same as above, but use a reschedule callback to do it:
2482 #include <math.h>
2484 static ev_tstamp
2485 my_scheduler_cb (ev_periodic *w, ev_tstamp now)
2486 {
2487 return now + (3600. - fmod (now, 3600.));
2488 }
2490 ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
2492 Example: Call a callback every hour, starting now:
2494 ev_periodic hourly_tick;
2495 ev_periodic_init (&hourly_tick, clock_cb,
2496 fmod (ev_now (loop), 3600.), 3600., 0);
2497 ev_periodic_start (loop, &hourly_tick);
2500 =head2 C<ev_signal> - signal me when a signal gets signalled!
2502 Signal watchers will trigger an event when the process receives a specific
2503 signal one or more times. Even though signals are very asynchronous, libev
2504 will try its best to deliver signals synchronously, i.e. as part of the
2505 normal event processing, like any other event.
2507 If you want signals to be delivered truly asynchronously, just use
2508 C<sigaction> as you would do without libev and forget about sharing
2509 the signal. You can even use C<ev_async> from a signal handler to
2510 synchronously wake up an event loop.
2512 You can configure as many watchers as you like for the same signal, but
2513 only within the same loop, i.e. you can watch for C<SIGINT> in your
2514 default loop and for C<SIGIO> in another loop, but you cannot watch for
2515 C<SIGINT> in both the default loop and another loop at the same time. At
2516 the moment, C<SIGCHLD> is permanently tied to the default loop.
2518 Only after the first watcher for a signal is started will libev actually
2519 register something with the kernel. It thus coexists with your own signal
2520 handlers as long as you don't register any with libev for the same signal.
2522 If possible and supported, libev will install its handlers with
2523 C<SA_RESTART> (or equivalent) behaviour enabled, so system calls should
2524 not be unduly interrupted. If you have a problem with system calls getting
2525 interrupted by signals you can block all signals in an C<ev_check> watcher
2526 and unblock them in an C<ev_prepare> watcher.
2528 =head3 The special problem of inheritance over fork/execve/pthread_create
2530 Both the signal mask (C<sigprocmask>) and the signal disposition
2531 (C<sigaction>) are unspecified after starting a signal watcher (and after
2532 stopping it again), that is, libev might or might not block the signal,
2533 and might or might not set or restore the installed signal handler (but
2536 While this does not matter for the signal disposition (libev never
2537 sets signals to C<SIG_IGN>, so handlers will be reset to C<SIG_DFL> on
2538 C<execve>), this matters for the signal mask: many programs do not expect
2539 certain signals to be blocked.
2541 This means that before calling C<exec> (from the child) you should reset
2542 the signal mask to whatever "default" you expect (all clear is a good
2543 choice usually).
2545 The simplest way to ensure that the signal mask is reset in the child is
2546 to install a fork handler with C<pthread_atfork> that resets it. That will
2547 catch fork calls done by libraries (such as the libc) as well.
2549 In current versions of libev, the signal will not be blocked indefinitely
2550 unless you use the C<signalfd> API (C<EV_SIGNALFD>). While this reduces
2551 the window of opportunity for problems, it will not go away, as libev
2552 I<has> to modify the signal mask, at least temporarily.
2554 So I can't stress this enough: I<If you do not reset your signal mask when
2555 you expect it to be empty, you have a race condition in your code>. This
2556 is not a libev-specific thing, this is true for most event libraries.
2558 =head3 The special problem of threads signal handling
2560 POSIX threads has problematic signal handling semantics, specifically,
2561 a lot of functionality (sigfd, sigwait etc.) only really works if all
2562 threads in a process block signals, which is hard to achieve.
2564 When you want to use sigwait (or mix libev signal handling with your own
2565 for the same signals), you can tackle this problem by globally blocking
2566 all signals before creating any threads (or creating them with a fully set
2567 sigprocmask) and also specifying the C<EVFLAG_NOSIGMASK> when creating
2568 loops. Then designate one thread as "signal receiver thread" which handles
2569 these signals. You can pass on any signals that libev might be interested
2570 in by calling C<ev_feed_signal>.
2572 =head3 Watcher-Specific Functions and Data Members
2574 =over 4
2576 =item ev_signal_init (ev_signal *, callback, int signum)
2578 =item ev_signal_set (ev_signal *, int signum)
2580 Configures the watcher to trigger on the given signal number (usually one
2581 of the C<SIGxxx> constants).
2583 =item int signum [read-only]
2585 The signal the watcher watches out for.
2587 =back
2589 =head3 Examples
2591 Example: Try to exit cleanly on SIGINT.
2593 static void
2594 sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
2595 {
2596 ev_break (loop, EVBREAK_ALL);
2597 }
2599 ev_signal signal_watcher;
2600 ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
2601 ev_signal_start (loop, &signal_watcher);
2604 =head2 C<ev_child> - watch out for process status changes
2606 Child watchers trigger when your process receives a SIGCHLD in response to
2607 some child status changes (most typically when a child of yours dies or
2608 exits). It is permissible to install a child watcher I<after> the child
2609 has been forked (which implies it might have already exited), as long
2610 as the event loop isn't entered (or is continued from a watcher), i.e.,
2611 forking and then immediately registering a watcher for the child is fine,
2612 but forking and registering a watcher a few event loop iterations later or
2613 in the next callback invocation is not.
2615 Only the default event loop is capable of handling signals, and therefore
2616 you can only register child watchers in the default event loop.
2618 Due to some design glitches inside libev, child watchers will always be
2619 handled at maximum priority (their priority is set to C<EV_MAXPRI> by
2620 libev)
2622 =head3 Process Interaction
2624 Libev grabs C<SIGCHLD> as soon as the default event loop is
2625 initialised. This is necessary to guarantee proper behaviour even if the
2626 first child watcher is started after the child exits. The occurrence
2627 of C<SIGCHLD> is recorded asynchronously, but child reaping is done
2628 synchronously as part of the event loop processing. Libev always reaps all
2629 children, even ones not watched.
2631 =head3 Overriding the Built-In Processing
2633 Libev offers no special support for overriding the built-in child
2634 processing, but if your application collides with libev's default child
2635 handler, you can override it easily by installing your own handler for
2636 C<SIGCHLD> after initialising the default loop, and making sure the
2637 default loop never gets destroyed. You are encouraged, however, to use an
2638 event-based approach to child reaping and thus use libev's support for
2639 that, so other libev users can use C<ev_child> watchers freely.
2641 =head3 Stopping the Child Watcher
2643 Currently, the child watcher never gets stopped, even when the
2644 child terminates, so normally one needs to stop the watcher in the
2645 callback. Future versions of libev might stop the watcher automatically
2646 when a child exit is detected (calling C<ev_child_stop> twice is not a
2647 problem).
2649 =head3 Watcher-Specific Functions and Data Members
2651 =over 4
2653 =item ev_child_init (ev_child *, callback, int pid, int trace)
2655 =item ev_child_set (ev_child *, int pid, int trace)
2657 Configures the watcher to wait for status changes of process C<pid> (or
2658 I<any> process if C<pid> is specified as C<0>). The callback can look
2659 at the C<rstatus> member of the C<ev_child> watcher structure to see
2660 the status word (use the macros from C<sys/wait.h> and see your systems
2661 C<waitpid> documentation). The C<rpid> member contains the pid of the
2662 process causing the status change. C<trace> must be either C<0> (only
2663 activate the watcher when the process terminates) or C<1> (additionally
2664 activate the watcher when the process is stopped or continued).
2666 =item int pid [read-only]
2668 The process id this watcher watches out for, or C<0>, meaning any process id.
2670 =item int rpid [read-write]
2672 The process id that detected a status change.
2674 =item int rstatus [read-write]
2676 The process exit/trace status caused by C<rpid> (see your systems
2677 C<waitpid> and C<sys/wait.h> documentation for details).
2679 =back
2681 =head3 Examples
2683 Example: C<fork()> a new process and install a child handler to wait for
2684 its completion.
2686 ev_child cw;
2688 static void
2689 child_cb (EV_P_ ev_child *w, int revents)
2690 {
2691 ev_child_stop (EV_A_ w);
2692 printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
2693 }
2695 pid_t pid = fork ();
2697 if (pid < 0)
2698 // error
2699 else if (pid == 0)
2700 {
2701 // the forked child executes here
2702 exit (1);
2703 }
2704 else
2705 {
2706 ev_child_init (&cw, child_cb, pid, 0);
2707 ev_child_start (EV_DEFAULT_ &cw);
2708 }
2711 =head2 C<ev_stat> - did the file attributes just change?
2713 This watches a file system path for attribute changes. That is, it calls
2714 C<stat> on that path in regular intervals (or when the OS says it changed)
2715 and sees if it changed compared to the last time, invoking the callback
2716 if it did. Starting the watcher C<stat>'s the file, so only changes that
2717 happen after the watcher has been started will be reported.
2719 The path does not need to exist: changing from "path exists" to "path does
2720 not exist" is a status change like any other. The condition "path does not
2721 exist" (or more correctly "path cannot be stat'ed") is signified by the
2722 C<st_nlink> field being zero (which is otherwise always forced to be at
2723 least one) and all the other fields of the stat buffer having unspecified
2724 contents.
2726 The path I<must not> end in a slash or contain special components such as
2727 C<.> or C<..>. The path I<should> be absolute: If it is relative and
2728 your working directory changes, then the behaviour is undefined.
2730 Since there is no portable change notification interface available, the
2731 portable implementation simply calls C<stat(2)> regularly on the path
2732 to see if it changed somehow. You can specify a recommended polling
2733 interval for this case. If you specify a polling interval of C<0> (highly
2734 recommended!) then a I<suitable, unspecified default> value will be used
2735 (which you can expect to be around five seconds, although this might
2736 change dynamically). Libev will also impose a minimum interval which is
2737 currently around C<0.1>, but that's usually overkill.
2739 This watcher type is not meant for massive numbers of stat watchers,
2740 as even with OS-supported change notifications, this can be
2741 resource-intensive.
2743 At the time of this writing, the only OS-specific interface implemented
2744 is the Linux inotify interface (implementing kqueue support is left as an
2745 exercise for the reader. Note, however, that the author sees no way of
2746 implementing C<ev_stat> semantics with kqueue, except as a hint).
2748 =head3 ABI Issues (Largefile Support)
2750 Libev by default (unless the user overrides this) uses the default
2751 compilation environment, which means that on systems with large file
2752 support disabled by default, you get the 32 bit version of the stat
2753 structure. When using the library from programs that change the ABI to
2754 use 64 bit file offsets the programs will fail. In that case you have to
2755 compile libev with the same flags to get binary compatibility. This is
2756 obviously the case with any flags that change the ABI, but the problem is
2757 most noticeably displayed with ev_stat and large file support.
2759 The solution for this is to lobby your distribution maker to make large
2760 file interfaces available by default (as e.g. FreeBSD does) and not
2761 optional. Libev cannot simply switch on large file support because it has
2762 to exchange stat structures with application programs compiled using the
2763 default compilation environment.
2765 =head3 Inotify and Kqueue
2767 When C<inotify (7)> support has been compiled into libev and present at
2768 runtime, it will be used to speed up change detection where possible. The
2769 inotify descriptor will be created lazily when the first C<ev_stat>
2770 watcher is being started.
2772 Inotify presence does not change the semantics of C<ev_stat> watchers
2773 except that changes might be detected earlier, and in some cases, to avoid
2774 making regular C<stat> calls. Even in the presence of inotify support
2775 there are many cases where libev has to resort to regular C<stat> polling,
2776 but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
2777 many bugs), the path exists (i.e. stat succeeds), and the path resides on
2778 a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
2779 xfs are fully working) libev usually gets away without polling.
2781 There is no support for kqueue, as apparently it cannot be used to
2782 implement this functionality, due to the requirement of having a file
2783 descriptor open on the object at all times, and detecting renames, unlinks
2784 etc. is difficult.
2786 =head3 C<stat ()> is a synchronous operation
2788 Libev doesn't normally do any kind of I/O itself, and so is not blocking
2789 the process. The exception are C<ev_stat> watchers - those call C<stat
2790 ()>, which is a synchronous operation.
2792 For local paths, this usually doesn't matter: unless the system is very
2793 busy or the intervals between stat's are large, a stat call will be fast,
2794 as the path data is usually in memory already (except when starting the
2795 watcher).
2797 For networked file systems, calling C<stat ()> can block an indefinite
2798 time due to network issues, and even under good conditions, a stat call
2799 often takes multiple milliseconds.
2801 Therefore, it is best to avoid using C<ev_stat> watchers on networked
2802 paths, although this is fully supported by libev.
2804 =head3 The special problem of stat time resolution
2806 The C<stat ()> system call only supports full-second resolution portably,
2807 and even on systems where the resolution is higher, most file systems
2808 still only support whole seconds.
2810 That means that, if the time is the only thing that changes, you can
2811 easily miss updates: on the first update, C<ev_stat> detects a change and
2812 calls your callback, which does something. When there is another update
2813 within the same second, C<ev_stat> will be unable to detect unless the
2814 stat data does change in other ways (e.g. file size).
2816 The solution to this is to delay acting on a change for slightly more
2817 than a second (or till slightly after the next full second boundary), using
2818 a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
2819 ev_timer_again (loop, w)>).
2821 The C<.02> offset is added to work around small timing inconsistencies
2822 of some operating systems (where the second counter of the current time
2823 might be be delayed. One such system is the Linux kernel, where a call to
2824 C<gettimeofday> might return a timestamp with a full second later than
2825 a subsequent C<time> call - if the equivalent of C<time ()> is used to
2826 update file times then there will be a small window where the kernel uses
2827 the previous second to update file times but libev might already execute
2828 the timer callback).
2830 =head3 Watcher-Specific Functions and Data Members
2832 =over 4
2834 =item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
2836 =item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
2838 Configures the watcher to wait for status changes of the given
2839 C<path>. The C<interval> is a hint on how quickly a change is expected to
2840 be detected and should normally be specified as C<0> to let libev choose
2841 a suitable value. The memory pointed to by C<path> must point to the same
2842 path for as long as the watcher is active.
2844 The callback will receive an C<EV_STAT> event when a change was detected,
2845 relative to the attributes at the time the watcher was started (or the
2846 last change was detected).
2848 =item ev_stat_stat (loop, ev_stat *)
2850 Updates the stat buffer immediately with new values. If you change the
2851 watched path in your callback, you could call this function to avoid
2852 detecting this change (while introducing a race condition if you are not
2853 the only one changing the path). Can also be useful simply to find out the
2854 new values.
2856 =item ev_statdata attr [read-only]
2858 The most-recently detected attributes of the file. Although the type is
2859 C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
2860 suitable for your system, but you can only rely on the POSIX-standardised
2861 members to be present. If the C<st_nlink> member is C<0>, then there was
2862 some error while C<stat>ing the file.
2864 =item ev_statdata prev [read-only]
2866 The previous attributes of the file. The callback gets invoked whenever
2867 C<prev> != C<attr>, or, more precisely, one or more of these members
2868 differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
2869 C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
2871 =item ev_tstamp interval [read-only]
2873 The specified interval.
2875 =item const char *path [read-only]
2877 The file system path that is being watched.
2879 =back
2881 =head3 Examples
2883 Example: Watch C</etc/passwd> for attribute changes.
2885 static void
2886 passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
2887 {
2888 /* /etc/passwd changed in some way */
2889 if (w->attr.st_nlink)
2890 {
2891 printf ("passwd current size %ld\n", (long)w->attr.st_size);
2892 printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
2893 printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
2894 }
2895 else
2896 /* you shalt not abuse printf for puts */
2897 puts ("wow, /etc/passwd is not there, expect problems. "
2898 "if this is windows, they already arrived\n");
2899 }
2901 ...
2902 ev_stat passwd;
2904 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
2905 ev_stat_start (loop, &passwd);
2907 Example: Like above, but additionally use a one-second delay so we do not
2908 miss updates (however, frequent updates will delay processing, too, so
2909 one might do the work both on C<ev_stat> callback invocation I<and> on
2910 C<ev_timer> callback invocation).
2912 static ev_stat passwd;
2913 static ev_timer timer;
2915 static void
2916 timer_cb (EV_P_ ev_timer *w, int revents)
2917 {
2918 ev_timer_stop (EV_A_ w);
2920 /* now it's one second after the most recent passwd change */
2921 }
2923 static void
2924 stat_cb (EV_P_ ev_stat *w, int revents)
2925 {
2926 /* reset the one-second timer */
2927 ev_timer_again (EV_A_ &timer);
2928 }
2930 ...
2931 ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
2932 ev_stat_start (loop, &passwd);
2933 ev_timer_init (&timer, timer_cb, 0., 1.02);
2936 =head2 C<ev_idle> - when you've got nothing better to do...
2938 Idle watchers trigger events when no other events of the same or higher
2939 priority are pending (prepare, check and other idle watchers do not count
2940 as receiving "events").
2942 That is, as long as your process is busy handling sockets or timeouts
2943 (or even signals, imagine) of the same or higher priority it will not be
2944 triggered. But when your process is idle (or only lower-priority watchers
2945 are pending), the idle watchers are being called once per event loop
2946 iteration - until stopped, that is, or your process receives more events
2947 and becomes busy again with higher priority stuff.
2949 The most noteworthy effect is that as long as any idle watchers are
2950 active, the process will not block when waiting for new events.
2952 Apart from keeping your process non-blocking (which is a useful
2953 effect on its own sometimes), idle watchers are a good place to do
2954 "pseudo-background processing", or delay processing stuff to after the
2955 event loop has handled all outstanding events.
2957 =head3 Abusing an C<ev_idle> watcher for its side-effect
2959 As long as there is at least one active idle watcher, libev will never
2960 sleep unnecessarily. Or in other words, it will loop as fast as possible.
2961 For this to work, the idle watcher doesn't need to be invoked at all - the
2962 lowest priority will do.
2964 This mode of operation can be useful together with an C<ev_check> watcher,
2965 to do something on each event loop iteration - for example to balance load
2966 between different connections.
2968 See L</Abusing an ev_check watcher for its side-effect> for a longer
2969 example.
2971 =head3 Watcher-Specific Functions and Data Members
2973 =over 4
2975 =item ev_idle_init (ev_idle *, callback)
2977 Initialises and configures the idle watcher - it has no parameters of any
2978 kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
2979 believe me.
2981 =back
2983 =head3 Examples
2985 Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
2986 callback, free it. Also, use no error checking, as usual.
2988 static void
2989 idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
2990 {
2991 // stop the watcher
2992 ev_idle_stop (loop, w);
2994 // now we can free it
2995 free (w);
2997 // now do something you wanted to do when the program has
2998 // no longer anything immediate to do.
2999 }
3001 ev_idle *idle_watcher = malloc (sizeof (ev_idle));
3002 ev_idle_init (idle_watcher, idle_cb);
3003 ev_idle_start (loop, idle_watcher);
3006 =head2 C<ev_prepare> and C<ev_check> - customise your event loop!
3008 Prepare and check watchers are often (but not always) used in pairs:
3009 prepare watchers get invoked before the process blocks and check watchers
3010 afterwards.
3012 You I<must not> call C<ev_run> (or similar functions that enter the
3013 current event loop) or C<ev_loop_fork> from either C<ev_prepare> or
3014 C<ev_check> watchers. Other loops than the current one are fine,
3015 however. The rationale behind this is that you do not need to check
3016 for recursion in those watchers, i.e. the sequence will always be
3017 C<ev_prepare>, blocking, C<ev_check> so if you have one watcher of each
3018 kind they will always be called in pairs bracketing the blocking call.
3020 Their main purpose is to integrate other event mechanisms into libev and
3021 their use is somewhat advanced. They could be used, for example, to track
3022 variable changes, implement your own watchers, integrate net-snmp or a
3023 coroutine library and lots more. They are also occasionally useful if
3024 you cache some data and want to flush it before blocking (for example,
3025 in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
3026 watcher).
3028 This is done by examining in each prepare call which file descriptors
3029 need to be watched by the other library, registering C<ev_io> watchers
3030 for them and starting an C<ev_timer> watcher for any timeouts (many
3031 libraries provide exactly this functionality). Then, in the check watcher,
3032 you check for any events that occurred (by checking the pending status
3033 of all watchers and stopping them) and call back into the library. The
3034 I/O and timer callbacks will never actually be called (but must be valid
3035 nevertheless, because you never know, you know?).
3037 As another example, the Perl Coro module uses these hooks to integrate
3038 coroutines into libev programs, by yielding to other active coroutines
3039 during each prepare and only letting the process block if no coroutines
3040 are ready to run (it's actually more complicated: it only runs coroutines
3041 with priority higher than or equal to the event loop and one coroutine
3042 of lower priority, but only once, using idle watchers to keep the event
3043 loop from blocking if lower-priority coroutines are active, thus mapping
3044 low-priority coroutines to idle/background tasks).
3046 When used for this purpose, it is recommended to give C<ev_check> watchers
3047 highest (C<EV_MAXPRI>) priority, to ensure that they are being run before
3048 any other watchers after the poll (this doesn't matter for C<ev_prepare>
3049 watchers).
3051 Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
3052 activate ("feed") events into libev. While libev fully supports this, they
3053 might get executed before other C<ev_check> watchers did their job. As
3054 C<ev_check> watchers are often used to embed other (non-libev) event
3055 loops those other event loops might be in an unusable state until their
3056 C<ev_check> watcher ran (always remind yourself to coexist peacefully with
3057 others).
3059 =head3 Abusing an C<ev_check> watcher for its side-effect
3061 C<ev_check> (and less often also C<ev_prepare>) watchers can also be
3062 useful because they are called once per event loop iteration. For
3063 example, if you want to handle a large number of connections fairly, you
3064 normally only do a bit of work for each active connection, and if there
3065 is more work to do, you wait for the next event loop iteration, so other
3066 connections have a chance of making progress.
3068 Using an C<ev_check> watcher is almost enough: it will be called on the
3069 next event loop iteration. However, that isn't as soon as possible -
3070 without external events, your C<ev_check> watcher will not be invoked.
3072 This is where C<ev_idle> watchers come in handy - all you need is a
3073 single global idle watcher that is active as long as you have one active
3074 C<ev_check> watcher. The C<ev_idle> watcher makes sure the event loop
3075 will not sleep, and the C<ev_check> watcher makes sure a callback gets
3076 invoked. Neither watcher alone can do that.
3078 =head3 Watcher-Specific Functions and Data Members
3080 =over 4
3082 =item ev_prepare_init (ev_prepare *, callback)
3084 =item ev_check_init (ev_check *, callback)
3086 Initialises and configures the prepare or check watcher - they have no
3087 parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
3088 macros, but using them is utterly, utterly, utterly and completely
3089 pointless.
3091 =back
3093 =head3 Examples
3095 There are a number of principal ways to embed other event loops or modules
3096 into libev. Here are some ideas on how to include libadns into libev
3097 (there is a Perl module named C<EV::ADNS> that does this, which you could
3098 use as a working example. Another Perl module named C<EV::Glib> embeds a
3099 Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
3100 Glib event loop).
3102 Method 1: Add IO watchers and a timeout watcher in a prepare handler,
3103 and in a check watcher, destroy them and call into libadns. What follows
3104 is pseudo-code only of course. This requires you to either use a low
3105 priority for the check watcher or use C<ev_clear_pending> explicitly, as
3106 the callbacks for the IO/timeout watchers might not have been called yet.
3108 static ev_io iow [nfd];
3109 static ev_timer tw;
3111 static void
3112 io_cb (struct ev_loop *loop, ev_io *w, int revents)
3113 {
3114 }
3116 // create io watchers for each fd and a timer before blocking
3117 static void
3118 adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
3119 {
3120 int timeout = 3600000;
3121 struct pollfd fds [nfd];
3122 // actual code will need to loop here and realloc etc.
3123 adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
3125 /* the callback is illegal, but won't be called as we stop during check */
3126 ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
3127 ev_timer_start (loop, &tw);
3129 // create one ev_io per pollfd
3130 for (int i = 0; i < nfd; ++i)
3131 {
3132 ev_io_init (iow + i, io_cb, fds [i].fd,
3133 ((fds [i].events & POLLIN ? EV_READ : 0)
3134 | (fds [i].events & POLLOUT ? EV_WRITE : 0)));
3136 fds [i].revents = 0;
3137 ev_io_start (loop, iow + i);
3138 }
3139 }
3141 // stop all watchers after blocking
3142 static void
3143 adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
3144 {
3145 ev_timer_stop (loop, &tw);
3147 for (int i = 0; i < nfd; ++i)
3148 {
3149 // set the relevant poll flags
3150 // could also call adns_processreadable etc. here
3151 struct pollfd *fd = fds + i;
3152 int revents = ev_clear_pending (iow + i);
3153 if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
3154 if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
3156 // now stop the watcher
3157 ev_io_stop (loop, iow + i);
3158 }
3160 adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
3161 }
3163 Method 2: This would be just like method 1, but you run C<adns_afterpoll>
3164 in the prepare watcher and would dispose of the check watcher.
3166 Method 3: If the module to be embedded supports explicit event
3167 notification (libadns does), you can also make use of the actual watcher
3168 callbacks, and only destroy/create the watchers in the prepare watcher.
3170 static void
3171 timer_cb (EV_P_ ev_timer *w, int revents)
3172 {
3173 adns_state ads = (adns_state)w->data;
3174 update_now (EV_A);
3176 adns_processtimeouts (ads, &tv_now);
3177 }
3179 static void
3180 io_cb (EV_P_ ev_io *w, int revents)
3181 {
3182 adns_state ads = (adns_state)w->data;
3183 update_now (EV_A);
3185 if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
3186 if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
3187 }
3189 // do not ever call adns_afterpoll
3191 Method 4: Do not use a prepare or check watcher because the module you
3192 want to embed is not flexible enough to support it. Instead, you can
3193 override their poll function. The drawback with this solution is that the
3194 main loop is now no longer controllable by EV. The C<Glib::EV> module uses
3195 this approach, effectively embedding EV as a client into the horrible
3196 libglib event loop.
3198 static gint
3199 event_poll_func (GPollFD *fds, guint nfds, gint timeout)
3200 {
3201 int got_events = 0;
3203 for (n = 0; n < nfds; ++n)
3204 // create/start io watcher that sets the relevant bits in fds[n] and increment got_events
3206 if (timeout >= 0)
3207 // create/start timer
3209 // poll
3210 ev_run (EV_A_ 0);
3212 // stop timer again
3213 if (timeout >= 0)
3214 ev_timer_stop (EV_A_ &to);
3216 // stop io watchers again - their callbacks should have set
3217 for (n = 0; n < nfds; ++n)
3218 ev_io_stop (EV_A_ iow [n]);
3220 return got_events;
3221 }
3224 =head2 C<ev_embed> - when one backend isn't enough...
3226 This is a rather advanced watcher type that lets you embed one event loop
3227 into another (currently only C<ev_io> events are supported in the embedded
3228 loop, other types of watchers might be handled in a delayed or incorrect
3229 fashion and must not be used).
3231 There are primarily two reasons you would want that: work around bugs and
3232 prioritise I/O.
3234 As an example for a bug workaround, the kqueue backend might only support
3235 sockets on some platform, so it is unusable as generic backend, but you
3236 still want to make use of it because you have many sockets and it scales
3237 so nicely. In this case, you would create a kqueue-based loop and embed
3238 it into your default loop (which might use e.g. poll). Overall operation
3239 will be a bit slower because first libev has to call C<poll> and then
3240 C<kevent>, but at least you can use both mechanisms for what they are
3241 best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
3243 As for prioritising I/O: under rare circumstances you have the case where
3244 some fds have to be watched and handled very quickly (with low latency),
3245 and even priorities and idle watchers might have too much overhead. In
3246 this case you would put all the high priority stuff in one loop and all
3247 the rest in a second one, and embed the second one in the first.
3249 As long as the watcher is active, the callback will be invoked every
3250 time there might be events pending in the embedded loop. The callback
3251 must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
3252 sweep and invoke their callbacks (the callback doesn't need to invoke the
3253 C<ev_embed_sweep> function directly, it could also start an idle watcher
3254 to give the embedded loop strictly lower priority for example).
3256 You can also set the callback to C<0>, in which case the embed watcher
3257 will automatically execute the embedded loop sweep whenever necessary.
3259 Fork detection will be handled transparently while the C<ev_embed> watcher
3260 is active, i.e., the embedded loop will automatically be forked when the
3261 embedding loop forks. In other cases, the user is responsible for calling
3262 C<ev_loop_fork> on the embedded loop.
3264 Unfortunately, not all backends are embeddable: only the ones returned by
3265 C<ev_embeddable_backends> are, which, unfortunately, does not include any
3266 portable one.
3268 So when you want to use this feature you will always have to be prepared
3269 that you cannot get an embeddable loop. The recommended way to get around
3270 this is to have a separate variables for your embeddable loop, try to
3271 create it, and if that fails, use the normal loop for everything.
3273 =head3 C<ev_embed> and fork
3275 While the C<ev_embed> watcher is running, forks in the embedding loop will
3276 automatically be applied to the embedded loop as well, so no special
3277 fork handling is required in that case. When the watcher is not running,
3278 however, it is still the task of the libev user to call C<ev_loop_fork ()>
3279 as applicable.
3281 =head3 Watcher-Specific Functions and Data Members
3283 =over 4
3285 =item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
3287 =item ev_embed_set (ev_embed *, struct ev_loop *embedded_loop)
3289 Configures the watcher to embed the given loop, which must be
3290 embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
3291 invoked automatically, otherwise it is the responsibility of the callback
3292 to invoke it (it will continue to be called until the sweep has been done,
3293 if you do not want that, you need to temporarily stop the embed watcher).
3295 =item ev_embed_sweep (loop, ev_embed *)
3297 Make a single, non-blocking sweep over the embedded loop. This works
3298 similarly to C<ev_run (embedded_loop, EVRUN_NOWAIT)>, but in the most
3299 appropriate way for embedded loops.
3301 =item struct ev_loop *other [read-only]
3303 The embedded event loop.
3305 =back
3307 =head3 Examples
3309 Example: Try to get an embeddable event loop and embed it into the default
3310 event loop. If that is not possible, use the default loop. The default
3311 loop is stored in C<loop_hi>, while the embeddable loop is stored in
3312 C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
3313 used).
3315 struct ev_loop *loop_hi = ev_default_init (0);
3316 struct ev_loop *loop_lo = 0;
3317 ev_embed embed;
3319 // see if there is a chance of getting one that works
3320 // (remember that a flags value of 0 means autodetection)
3321 loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
3322 ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
3323 : 0;
3325 // if we got one, then embed it, otherwise default to loop_hi
3326 if (loop_lo)
3327 {
3328 ev_embed_init (&embed, 0, loop_lo);
3329 ev_embed_start (loop_hi, &embed);
3330 }
3331 else
3332 loop_lo = loop_hi;
3334 Example: Check if kqueue is available but not recommended and create
3335 a kqueue backend for use with sockets (which usually work with any
3336 kqueue implementation). Store the kqueue/socket-only event loop in
3337 C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
3339 struct ev_loop *loop = ev_default_init (0);
3340 struct ev_loop *loop_socket = 0;
3341 ev_embed embed;
3343 if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
3344 if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
3345 {
3346 ev_embed_init (&embed, 0, loop_socket);
3347 ev_embed_start (loop, &embed);
3348 }
3350 if (!loop_socket)
3351 loop_socket = loop;
3353 // now use loop_socket for all sockets, and loop for everything else
3356 =head2 C<ev_fork> - the audacity to resume the event loop after a fork
3358 Fork watchers are called when a C<fork ()> was detected (usually because
3359 whoever is a good citizen cared to tell libev about it by calling
3360 C<ev_loop_fork>). The invocation is done before the event loop blocks next
3361 and before C<ev_check> watchers are being called, and only in the child
3362 after the fork. If whoever good citizen calling C<ev_default_fork> cheats
3363 and calls it in the wrong process, the fork handlers will be invoked, too,
3364 of course.
3366 =head3 The special problem of life after fork - how is it possible?
3368 Most uses of C<fork ()> consist of forking, then some simple calls to set
3369 up/change the process environment, followed by a call to C<exec()>. This
3370 sequence should be handled by libev without any problems.
3372 This changes when the application actually wants to do event handling
3373 in the child, or both parent in child, in effect "continuing" after the
3374 fork.
3376 The default mode of operation (for libev, with application help to detect
3377 forks) is to duplicate all the state in the child, as would be expected
3378 when I<either> the parent I<or> the child process continues.
3380 When both processes want to continue using libev, then this is usually the
3381 wrong result. In that case, usually one process (typically the parent) is
3382 supposed to continue with all watchers in place as before, while the other
3383 process typically wants to start fresh, i.e. without any active watchers.
3385 The cleanest and most efficient way to achieve that with libev is to
3386 simply create a new event loop, which of course will be "empty", and
3387 use that for new watchers. This has the advantage of not touching more
3388 memory than necessary, and thus avoiding the copy-on-write, and the
3389 disadvantage of having to use multiple event loops (which do not support
3390 signal watchers).
3392 When this is not possible, or you want to use the default loop for
3393 other reasons, then in the process that wants to start "fresh", call
3394 C<ev_loop_destroy (EV_DEFAULT)> followed by C<ev_default_loop (...)>.
3395 Destroying the default loop will "orphan" (not stop) all registered
3396 watchers, so you have to be careful not to execute code that modifies
3397 those watchers. Note also that in that case, you have to re-register any
3398 signal watchers.
3400 =head3 Watcher-Specific Functions and Data Members
3402 =over 4
3404 =item ev_fork_init (ev_fork *, callback)
3406 Initialises and configures the fork watcher - it has no parameters of any
3407 kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
3408 really.
3410 =back
3413 =head2 C<ev_cleanup> - even the best things end
3415 Cleanup watchers are called just before the event loop is being destroyed
3416 by a call to C<ev_loop_destroy>.
3418 While there is no guarantee that the event loop gets destroyed, cleanup
3419 watchers provide a convenient method to install cleanup hooks for your
3420 program, worker threads and so on - you just to make sure to destroy the
3421 loop when you want them to be invoked.
3423 Cleanup watchers are invoked in the same way as any other watcher. Unlike
3424 all other watchers, they do not keep a reference to the event loop (which
3425 makes a lot of sense if you think about it). Like all other watchers, you
3426 can call libev functions in the callback, except C<ev_cleanup_start>.
3428 =head3 Watcher-Specific Functions and Data Members
3430 =over 4
3432 =item ev_cleanup_init (ev_cleanup *, callback)
3434 Initialises and configures the cleanup watcher - it has no parameters of
3435 any kind. There is a C<ev_cleanup_set> macro, but using it is utterly
3436 pointless, I assure you.
3438 =back
3440 Example: Register an atexit handler to destroy the default loop, so any
3441 cleanup functions are called.
3443 static void
3444 program_exits (void)
3445 {
3446 ev_loop_destroy (EV_DEFAULT_UC);
3447 }
3449 ...
3450 atexit (program_exits);
3453 =head2 C<ev_async> - how to wake up an event loop
3455 In general, you cannot use an C<ev_loop> from multiple threads or other
3456 asynchronous sources such as signal handlers (as opposed to multiple event
3457 loops - those are of course safe to use in different threads).
3459 Sometimes, however, you need to wake up an event loop you do not control,
3460 for example because it belongs to another thread. This is what C<ev_async>
3461 watchers do: as long as the C<ev_async> watcher is active, you can signal
3462 it by calling C<ev_async_send>, which is thread- and signal safe.
3464 This functionality is very similar to C<ev_signal> watchers, as signals,
3465 too, are asynchronous in nature, and signals, too, will be compressed
3466 (i.e. the number of callback invocations may be less than the number of
3467 C<ev_async_send> calls). In fact, you could use signal watchers as a kind
3468 of "global async watchers" by using a watcher on an otherwise unused
3469 signal, and C<ev_feed_signal> to signal this watcher from another thread,
3470 even without knowing which loop owns the signal.
3472 =head3 Queueing
3474 C<ev_async> does not support queueing of data in any way. The reason
3475 is that the author does not know of a simple (or any) algorithm for a
3476 multiple-writer-single-reader queue that works in all cases and doesn't
3477 need elaborate support such as pthreads or unportable memory access
3478 semantics.
3480 That means that if you want to queue data, you have to provide your own
3481 queue. But at least I can tell you how to implement locking around your
3482 queue:
3484 =over 4
3486 =item queueing from a signal handler context
3488 To implement race-free queueing, you simply add to the queue in the signal
3489 handler but you block the signal handler in the watcher callback. Here is
3490 an example that does that for some fictitious SIGUSR1 handler:
3492 static ev_async mysig;
3494 static void
3495 sigusr1_handler (void)
3496 {
3497 sometype data;
3499 // no locking etc.
3500 queue_put (data);
3501 ev_async_send (EV_DEFAULT_ &mysig);
3502 }
3504 static void
3505 mysig_cb (EV_P_ ev_async *w, int revents)
3506 {
3507 sometype data;
3508 sigset_t block, prev;
3510 sigemptyset (&block);
3511 sigaddset (&block, SIGUSR1);
3512 sigprocmask (SIG_BLOCK, &block, &prev);
3514 while (queue_get (&data))
3515 process (data);
3517 if (sigismember (&prev, SIGUSR1)
3518 sigprocmask (SIG_UNBLOCK, &block, 0);
3519 }
3521 (Note: pthreads in theory requires you to use C<pthread_setmask>
3522 instead of C<sigprocmask> when you use threads, but libev doesn't do it
3523 either...).
3525 =item queueing from a thread context
3527 The strategy for threads is different, as you cannot (easily) block
3528 threads but you can easily preempt them, so to queue safely you need to
3529 employ a traditional mutex lock, such as in this pthread example:
3531 static ev_async mysig;
3532 static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
3534 static void
3535 otherthread (void)
3536 {
3537 // only need to lock the actual queueing operation
3538 pthread_mutex_lock (&mymutex);
3539 queue_put (data);
3540 pthread_mutex_unlock (&mymutex);
3542 ev_async_send (EV_DEFAULT_ &mysig);
3543 }
3545 static void
3546 mysig_cb (EV_P_ ev_async *w, int revents)
3547 {
3548 pthread_mutex_lock (&mymutex);
3550 while (queue_get (&data))
3551 process (data);
3553 pthread_mutex_unlock (&mymutex);
3554 }
3556 =back
3559 =head3 Watcher-Specific Functions and Data Members
3561 =over 4
3563 =item ev_async_init (ev_async *, callback)
3565 Initialises and configures the async watcher - it has no parameters of any
3566 kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
3567 trust me.
3569 =item ev_async_send (loop, ev_async *)
3571 Sends/signals/activates the given C<ev_async> watcher, that is, feeds
3572 an C<EV_ASYNC> event on the watcher into the event loop, and instantly
3573 returns.
3575 Unlike C<ev_feed_event>, this call is safe to do from other threads,
3576 signal or similar contexts (see the discussion of C<EV_ATOMIC_T> in the
3577 embedding section below on what exactly this means).
3579 Note that, as with other watchers in libev, multiple events might get
3580 compressed into a single callback invocation (another way to look at
3581 this is that C<ev_async> watchers are level-triggered: they are set on
3582 C<ev_async_send>, reset when the event loop detects that).
3584 This call incurs the overhead of at most one extra system call per event
3585 loop iteration, if the event loop is blocked, and no syscall at all if
3586 the event loop (or your program) is processing events. That means that
3587 repeated calls are basically free (there is no need to avoid calls for
3588 performance reasons) and that the overhead becomes smaller (typically
3589 zero) under load.
3591 =item bool = ev_async_pending (ev_async *)
3593 Returns a non-zero value when C<ev_async_send> has been called on the
3594 watcher but the event has not yet been processed (or even noted) by the
3595 event loop.
3597 C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
3598 the loop iterates next and checks for the watcher to have become active,
3599 it will reset the flag again. C<ev_async_pending> can be used to very
3600 quickly check whether invoking the loop might be a good idea.
3602 Not that this does I<not> check whether the watcher itself is pending,
3603 only whether it has been requested to make this watcher pending: there
3604 is a time window between the event loop checking and resetting the async
3605 notification, and the callback being invoked.
3607 =back
3612 There are some other functions of possible interest. Described. Here. Now.
3614 =over 4
3616 =item ev_once (loop, int fd, int events, ev_tstamp timeout, callback, arg)
3618 This function combines a simple timer and an I/O watcher, calls your
3619 callback on whichever event happens first and automatically stops both
3620 watchers. This is useful if you want to wait for a single event on an fd
3621 or timeout without having to allocate/configure/start/stop/free one or
3622 more watchers yourself.
3624 If C<fd> is less than 0, then no I/O watcher will be started and the
3625 C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
3626 the given C<fd> and C<events> set will be created and started.
3628 If C<timeout> is less than 0, then no timeout watcher will be
3629 started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
3630 repeat = 0) will be started. C<0> is a valid timeout.
3632 The callback has the type C<void (*cb)(int revents, void *arg)> and is
3633 passed an C<revents> set like normal event callbacks (a combination of
3634 C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMER>) and the C<arg>
3635 value passed to C<ev_once>. Note that it is possible to receive I<both>
3636 a timeout and an io event at the same time - you probably should give io
3637 events precedence.
3639 Example: wait up to ten seconds for data to appear on STDIN_FILENO.
3641 static void stdin_ready (int revents, void *arg)
3642 {
3643 if (revents & EV_READ)
3644 /* stdin might have data for us, joy! */;
3645 else if (revents & EV_TIMER)
3646 /* doh, nothing entered */;
3647 }
3649 ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
3651 =item ev_feed_fd_event (loop, int fd, int revents)
3653 Feed an event on the given fd, as if a file descriptor backend detected
3654 the given events.
3656 =item ev_feed_signal_event (loop, int signum)
3658 Feed an event as if the given signal occurred. See also C<ev_feed_signal>,
3659 which is async-safe.
3661 =back
3666 This section explains some common idioms that are not immediately
3667 obvious. Note that examples are sprinkled over the whole manual, and this
3668 section only contains stuff that wouldn't fit anywhere else.
3672 Each watcher has, by default, a C<void *data> member that you can read
3673 or modify at any time: libev will completely ignore it. This can be used
3674 to associate arbitrary data with your watcher. If you need more data and
3675 don't want to allocate memory separately and store a pointer to it in that
3676 data member, you can also "subclass" the watcher type and provide your own
3677 data:
3679 struct my_io
3680 {
3681 ev_io io;
3682 int otherfd;
3683 void *somedata;
3684 struct whatever *mostinteresting;
3685 };
3687 ...
3688 struct my_io w;
3689 ev_io_init (&, my_cb, fd, EV_READ);
3691 And since your callback will be called with a pointer to the watcher, you
3692 can cast it back to your own type:
3694 static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
3695 {
3696 struct my_io *w = (struct my_io *)w_;
3697 ...
3698 }
3700 More interesting and less C-conformant ways of casting your callback
3701 function type instead have been omitted.
3705 Another common scenario is to use some data structure with multiple
3706 embedded watchers, in effect creating your own watcher that combines
3707 multiple libev event sources into one "super-watcher":
3709 struct my_biggy
3710 {
3711 int some_data;
3712 ev_timer t1;
3713 ev_timer t2;
3714 }
3716 In this case getting the pointer to C<my_biggy> is a bit more
3717 complicated: Either you store the address of your C<my_biggy> struct in
3718 the C<data> member of the watcher (for woozies or C++ coders), or you need
3719 to use some pointer arithmetic using C<offsetof> inside your watchers (for
3720 real programmers):
3722 #include <stddef.h>
3724 static void
3725 t1_cb (EV_P_ ev_timer *w, int revents)
3726 {
3727 struct my_biggy big = (struct my_biggy *)
3728 (((char *)w) - offsetof (struct my_biggy, t1));
3729 }
3731 static void
3732 t2_cb (EV_P_ ev_timer *w, int revents)
3733 {
3734 struct my_biggy big = (struct my_biggy *)
3735 (((char *)w) - offsetof (struct my_biggy, t2));
3736 }
3740 Often you have structures like this in event-based programs:
3742 callback ()
3743 {
3744 free (request);
3745 }
3747 request = start_new_request (..., callback);
3749 The intent is to start some "lengthy" operation. The C<request> could be
3750 used to cancel the operation, or do other things with it.
3752 It's not uncommon to have code paths in C<start_new_request> that
3753 immediately invoke the callback, for example, to report errors. Or you add
3754 some caching layer that finds that it can skip the lengthy aspects of the
3755 operation and simply invoke the callback with the result.
3757 The problem here is that this will happen I<before> C<start_new_request>
3758 has returned, so C<request> is not set.
3760 Even if you pass the request by some safer means to the callback, you
3761 might want to do something to the request after starting it, such as
3762 canceling it, which probably isn't working so well when the callback has
3763 already been invoked.
3765 A common way around all these issues is to make sure that
3766 C<start_new_request> I<always> returns before the callback is invoked. If
3767 C<start_new_request> immediately knows the result, it can artificially
3768 delay invoking the callback by using a C<prepare> or C<idle> watcher for
3769 example, or more sneakily, by reusing an existing (stopped) watcher and
3770 pushing it into the pending queue:
3772 ev_set_cb (watcher, callback);
3773 ev_feed_event (EV_A_ watcher, 0);
3775 This way, C<start_new_request> can safely return before the callback is
3776 invoked, while not delaying callback invocation too much.
3780 Often (especially in GUI toolkits) there are places where you have
3781 I<modal> interaction, which is most easily implemented by recursively
3782 invoking C<ev_run>.
3784 This brings the problem of exiting - a callback might want to finish the
3785 main C<ev_run> call, but not the nested one (e.g. user clicked "Quit", but
3786 a modal "Are you sure?" dialog is still waiting), or just the nested one
3787 and not the main one (e.g. user clocked "Ok" in a modal dialog), or some
3788 other combination: In these cases, a simple C<ev_break> will not work.
3790 The solution is to maintain "break this loop" variable for each C<ev_run>
3791 invocation, and use a loop around C<ev_run> until the condition is
3792 triggered, using C<EVRUN_ONCE>:
3794 // main loop
3795 int exit_main_loop = 0;
3797 while (!exit_main_loop)
3798 ev_run (EV_DEFAULT_ EVRUN_ONCE);
3800 // in a modal watcher
3801 int exit_nested_loop = 0;
3803 while (!exit_nested_loop)
3804 ev_run (EV_A_ EVRUN_ONCE);
3806 To exit from any of these loops, just set the corresponding exit variable:
3808 // exit modal loop
3809 exit_nested_loop = 1;
3811 // exit main program, after modal loop is finished
3812 exit_main_loop = 1;
3814 // exit both
3815 exit_main_loop = exit_nested_loop = 1;
3819 Here is a fictitious example of how to run an event loop in a different
3820 thread from where callbacks are being invoked and watchers are
3821 created/added/removed.
3823 For a real-world example, see the C<EV::Loop::Async> perl module,
3824 which uses exactly this technique (which is suited for many high-level
3825 languages).
3827 The example uses a pthread mutex to protect the loop data, a condition
3828 variable to wait for callback invocations, an async watcher to notify the
3829 event loop thread and an unspecified mechanism to wake up the main thread.
3831 First, you need to associate some data with the event loop:
3833 typedef struct {
3834 mutex_t lock; /* global loop lock */
3835 ev_async async_w;
3836 thread_t tid;
3837 cond_t invoke_cv;
3838 } userdata;
3840 void prepare_loop (EV_P)
3841 {
3842 // for simplicity, we use a static userdata struct.
3843 static userdata u;
3845 ev_async_init (&u->async_w, async_cb);
3846 ev_async_start (EV_A_ &u->async_w);
3848 pthread_mutex_init (&u->lock, 0);
3849 pthread_cond_init (&u->invoke_cv, 0);
3851 // now associate this with the loop
3852 ev_set_userdata (EV_A_ u);
3853 ev_set_invoke_pending_cb (EV_A_ l_invoke);
3854 ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
3856 // then create the thread running ev_run
3857 pthread_create (&u->tid, 0, l_run, EV_A);
3858 }
3860 The callback for the C<ev_async> watcher does nothing: the watcher is used
3861 solely to wake up the event loop so it takes notice of any new watchers
3862 that might have been added:
3864 static void
3865 async_cb (EV_P_ ev_async *w, int revents)
3866 {
3867 // just used for the side effects
3868 }
3870 The C<l_release> and C<l_acquire> callbacks simply unlock/lock the mutex
3871 protecting the loop data, respectively.
3873 static void
3874 l_release (EV_P)
3875 {
3876 userdata *u = ev_userdata (EV_A);
3877 pthread_mutex_unlock (&u->lock);
3878 }
3880 static void
3881 l_acquire (EV_P)
3882 {
3883 userdata *u = ev_userdata (EV_A);
3884 pthread_mutex_lock (&u->lock);
3885 }
3887 The event loop thread first acquires the mutex, and then jumps straight
3888 into C<ev_run>:
3890 void *
3891 l_run (void *thr_arg)
3892 {
3893 struct ev_loop *loop = (struct ev_loop *)thr_arg;
3895 l_acquire (EV_A);
3896 pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
3897 ev_run (EV_A_ 0);
3898 l_release (EV_A);
3900 return 0;
3901 }
3903 Instead of invoking all pending watchers, the C<l_invoke> callback will
3904 signal the main thread via some unspecified mechanism (signals? pipe
3905 writes? C<Async::Interrupt>?) and then waits until all pending watchers
3906 have been called (in a while loop because a) spurious wakeups are possible
3907 and b) skipping inter-thread-communication when there are no pending
3908 watchers is very beneficial):
3910 static void
3911 l_invoke (EV_P)
3912 {
3913 userdata *u = ev_userdata (EV_A);
3915 while (ev_pending_count (EV_A))
3916 {
3917 wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
3918 pthread_cond_wait (&u->invoke_cv, &u->lock);
3919 }
3920 }
3922 Now, whenever the main thread gets told to invoke pending watchers, it
3923 will grab the lock, call C<ev_invoke_pending> and then signal the loop
3924 thread to continue:
3926 static void
3927 real_invoke_pending (EV_P)
3928 {
3929 userdata *u = ev_userdata (EV_A);
3931 pthread_mutex_lock (&u->lock);
3932 ev_invoke_pending (EV_A);
3933 pthread_cond_signal (&u->invoke_cv);
3934 pthread_mutex_unlock (&u->lock);
3935 }
3937 Whenever you want to start/stop a watcher or do other modifications to an
3938 event loop, you will now have to lock:
3940 ev_timer timeout_watcher;
3941 userdata *u = ev_userdata (EV_A);
3943 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
3945 pthread_mutex_lock (&u->lock);
3946 ev_timer_start (EV_A_ &timeout_watcher);
3947 ev_async_send (EV_A_ &u->async_w);
3948 pthread_mutex_unlock (&u->lock);
3950 Note that sending the C<ev_async> watcher is required because otherwise
3951 an event loop currently blocking in the kernel will have no knowledge
3952 about the newly added timer. By waking up the loop it will pick up any new
3953 watchers in the next event loop iteration.
3957 While the overhead of a callback that e.g. schedules a thread is small, it
3958 is still an overhead. If you embed libev, and your main usage is with some
3959 kind of threads or coroutines, you might want to customise libev so that
3960 doesn't need callbacks anymore.
3962 Imagine you have coroutines that you can switch to using a function
3963 C<switch_to (coro)>, that libev runs in a coroutine called C<libev_coro>
3964 and that due to some magic, the currently active coroutine is stored in a
3965 global called C<current_coro>. Then you can build your own "wait for libev
3966 event" primitive by changing C<EV_CB_DECLARE> and C<EV_CB_INVOKE> (note
3967 the differing C<;> conventions):
3969 #define EV_CB_DECLARE(type) struct my_coro *cb;
3970 #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
3972 That means instead of having a C callback function, you store the
3973 coroutine to switch to in each watcher, and instead of having libev call
3974 your callback, you instead have it switch to that coroutine.
3976 A coroutine might now wait for an event with a function called
3977 C<wait_for_event>. (the watcher needs to be started, as always, but it doesn't
3978 matter when, or whether the watcher is active or not when this function is
3979 called):
3981 void
3982 wait_for_event (ev_watcher *w)
3983 {
3984 ev_set_cb (w, current_coro);
3985 switch_to (libev_coro);
3986 }
3988 That basically suspends the coroutine inside C<wait_for_event> and
3989 continues the libev coroutine, which, when appropriate, switches back to
3990 this or any other coroutine.
3992 You can do similar tricks if you have, say, threads with an event queue -
3993 instead of storing a coroutine, you store the queue object and instead of
3994 switching to a coroutine, you push the watcher onto the queue and notify
3995 any waiters.
3997 To embed libev, see L</EMBEDDING>, but in short, it's easiest to create two
3998 files, F<my_ev.h> and F<my_ev.c> that include the respective libev files:
4000 // my_ev.h
4001 #define EV_CB_DECLARE(type) struct my_coro *cb;
4002 #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
4003 #include "../libev/ev.h"
4005 // my_ev.c
4006 #define EV_H "my_ev.h"
4007 #include "../libev/ev.c"
4009 And then use F<my_ev.h> when you would normally use F<ev.h>, and compile
4010 F<my_ev.c> into your project. When properly specifying include paths, you
4011 can even use F<ev.h> as header file name directly.
4016 Libev offers a compatibility emulation layer for libevent. It cannot
4017 emulate the internals of libevent, so here are some usage hints:
4019 =over 4
4021 =item * Only the libevent-1.4.1-beta API is being emulated.
4023 This was the newest libevent version available when libev was implemented,
4024 and is still mostly unchanged in 2010.
4026 =item * Use it by including <event.h>, as usual.
4028 =item * The following members are fully supported: ev_base, ev_callback,
4029 ev_arg, ev_fd, ev_res, ev_events.
4031 =item * Avoid using ev_flags and the EVLIST_*-macros, while it is
4032 maintained by libev, it does not work exactly the same way as in libevent (consider
4033 it a private API).
4035 =item * Priorities are not currently supported. Initialising priorities
4036 will fail and all watchers will have the same priority, even though there
4037 is an ev_pri field.
4039 =item * In libevent, the last base created gets the signals, in libev, the
4040 base that registered the signal gets the signals.
4042 =item * Other members are not supported.
4044 =item * The libev emulation is I<not> ABI compatible to libevent, you need
4045 to use the libev header file and library.
4047 =back
4049 =head1 C++ SUPPORT
4051 =head2 C API
4053 The normal C API should work fine when used from C++: both ev.h and the
4054 libev sources can be compiled as C++. Therefore, code that uses the C API
4055 will work fine.
4057 Proper exception specifications might have to be added to callbacks passed
4058 to libev: exceptions may be thrown only from watcher callbacks, all other
4059 callbacks (allocator, syserr, loop acquire/release and periodic reschedule
4060 callbacks) must not throw exceptions, and might need a C<noexcept>
4061 specification. If you have code that needs to be compiled as both C and
4062 C++ you can use the C<EV_NOEXCEPT> macro for this:
4064 static void
4065 fatal_error (const char *msg) EV_NOEXCEPT
4066 {
4067 perror (msg);
4068 abort ();
4069 }
4071 ...
4072 ev_set_syserr_cb (fatal_error);
4074 The only API functions that can currently throw exceptions are C<ev_run>,
4075 C<ev_invoke>, C<ev_invoke_pending> and C<ev_loop_destroy> (the latter
4076 because it runs cleanup watchers).
4078 Throwing exceptions in watcher callbacks is only supported if libev itself
4079 is compiled with a C++ compiler or your C and C++ environments allow
4080 throwing exceptions through C libraries (most do).
4082 =head2 C++ API
4084 Libev comes with some simplistic wrapper classes for C++ that mainly allow
4085 you to use some convenience methods to start/stop watchers and also change
4086 the callback model to a model using method callbacks on objects.
4088 To use it,
4090 #include <ev++.h>
4092 This automatically includes F<ev.h> and puts all of its definitions (many
4093 of them macros) into the global namespace. All C++ specific things are
4094 put into the C<ev> namespace. It should support all the same embedding
4095 options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
4097 Care has been taken to keep the overhead low. The only data member the C++
4098 classes add (compared to plain C-style watchers) is the event loop pointer
4099 that the watcher is associated with (or no additional members at all if
4100 you disable C<EV_MULTIPLICITY> when embedding libev).
4102 Currently, functions, static and non-static member functions and classes
4103 with C<operator ()> can be used as callbacks. Other types should be easy
4104 to add as long as they only need one additional pointer for context. If
4105 you need support for other types of functors please contact the author
4106 (preferably after implementing it).
4108 For all this to work, your C++ compiler either has to use the same calling
4109 conventions as your C compiler (for static member functions), or you have
4110 to embed libev and compile libev itself as C++.
4112 Here is a list of things available in the C<ev> namespace:
4114 =over 4
4116 =item C<ev::READ>, C<ev::WRITE> etc.
4118 These are just enum values with the same values as the C<EV_READ> etc.
4119 macros from F<ev.h>.
4121 =item C<ev::tstamp>, C<ev::now>
4123 Aliases to the same types/functions as with the C<ev_> prefix.
4125 =item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
4127 For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
4128 the same name in the C<ev> namespace, with the exception of C<ev_signal>
4129 which is called C<ev::sig> to avoid clashes with the C<signal> macro
4130 defined by many implementations.
4132 All of those classes have these methods:
4134 =over 4
4136 =item ev::TYPE::TYPE ()
4138 =item ev::TYPE::TYPE (loop)
4140 =item ev::TYPE::~TYPE
4142 The constructor (optionally) takes an event loop to associate the watcher
4143 with. If it is omitted, it will use C<EV_DEFAULT>.
4145 The constructor calls C<ev_init> for you, which means you have to call the
4146 C<set> method before starting it.
4148 It will not set a callback, however: You have to call the templated C<set>
4149 method to set a callback before you can start the watcher.
4151 (The reason why you have to use a method is a limitation in C++ which does
4152 not allow explicit template arguments for constructors).
4154 The destructor automatically stops the watcher if it is active.
4156 =item w->set<class, &class::method> (object *)
4158 This method sets the callback method to call. The method has to have a
4159 signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
4160 first argument and the C<revents> as second. The object must be given as
4161 parameter and is stored in the C<data> member of the watcher.
4163 This method synthesizes efficient thunking code to call your method from
4164 the C callback that libev requires. If your compiler can inline your
4165 callback (i.e. it is visible to it at the place of the C<set> call and
4166 your compiler is good :), then the method will be fully inlined into the
4167 thunking function, making it as fast as a direct C callback.
4169 Example: simple class declaration and watcher initialisation
4171 struct myclass
4172 {
4173 void io_cb (ev::io &w, int revents) { }
4174 }
4176 myclass obj;
4177 ev::io iow;
4178 iow.set <myclass, &myclass::io_cb> (&obj);
4180 =item w->set (object *)
4182 This is a variation of a method callback - leaving out the method to call
4183 will default the method to C<operator ()>, which makes it possible to use
4184 functor objects without having to manually specify the C<operator ()> all
4185 the time. Incidentally, you can then also leave out the template argument
4186 list.
4188 The C<operator ()> method prototype must be C<void operator ()(watcher &w,
4189 int revents)>.
4191 See the method-C<set> above for more details.
4193 Example: use a functor object as callback.
4195 struct myfunctor
4196 {
4197 void operator() (ev::io &w, int revents)
4198 {
4199 ...
4200 }
4201 }
4203 myfunctor f;
4205 ev::io w;
4206 w.set (&f);
4208 =item w->set<function> (void *data = 0)
4210 Also sets a callback, but uses a static method or plain function as
4211 callback. The optional C<data> argument will be stored in the watcher's
4212 C<data> member and is free for you to use.
4214 The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
4216 See the method-C<set> above for more details.
4218 Example: Use a plain function as callback.
4220 static void io_cb (ev::io &w, int revents) { }
4221 iow.set <io_cb> ();
4223 =item w->set (loop)
4225 Associates a different C<struct ev_loop> with this watcher. You can only
4226 do this when the watcher is inactive (and not pending either).
4228 =item w->set ([arguments])
4230 Basically the same as C<ev_TYPE_set> (except for C<ev::embed> watchers>),
4231 with the same arguments. Either this method or a suitable start method
4232 must be called at least once. Unlike the C counterpart, an active watcher
4233 gets automatically stopped and restarted when reconfiguring it with this
4234 method.
4236 For C<ev::embed> watchers this method is called C<set_embed>, to avoid
4237 clashing with the C<set (loop)> method.
4239 =item w->start ()
4241 Starts the watcher. Note that there is no C<loop> argument, as the
4242 constructor already stores the event loop.
4244 =item w->start ([arguments])
4246 Instead of calling C<set> and C<start> methods separately, it is often
4247 convenient to wrap them in one call. Uses the same type of arguments as
4248 the configure C<set> method of the watcher.
4250 =item w->stop ()
4252 Stops the watcher if it is active. Again, no C<loop> argument.
4254 =item w->again () (C<ev::timer>, C<ev::periodic> only)
4256 For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
4257 C<ev_TYPE_again> function.
4259 =item w->sweep () (C<ev::embed> only)
4261 Invokes C<ev_embed_sweep>.
4263 =item w->update () (C<ev::stat> only)
4265 Invokes C<ev_stat_stat>.
4267 =back
4269 =back
4271 Example: Define a class with two I/O and idle watchers, start the I/O
4272 watchers in the constructor.
4274 class myclass
4275 {
4276 ev::io io ; void io_cb (ev::io &w, int revents);
4277 ev::io io2 ; void io2_cb (ev::io &w, int revents);
4278 ev::idle idle; void idle_cb (ev::idle &w, int revents);
4280 myclass (int fd)
4281 {
4282 io .set <myclass, &myclass::io_cb > (this);
4283 io2 .set <myclass, &myclass::io2_cb > (this);
4284 idle.set <myclass, &myclass::idle_cb> (this);
4286 io.set (fd, ev::WRITE); // configure the watcher
4287 io.start (); // start it whenever convenient
4289 io2.start (fd, ev::READ); // set + start in one call
4290 }
4291 };
4296 Libev does not offer other language bindings itself, but bindings for a
4297 number of languages exist in the form of third-party packages. If you know
4298 any interesting language binding in addition to the ones listed here, drop
4299 me a note.
4301 =over 4
4303 =item Perl
4305 The EV module implements the full libev API and is actually used to test
4306 libev. EV is developed together with libev. Apart from the EV core module,
4307 there are additional modules that implement libev-compatible interfaces
4308 to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
4309 C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
4310 and C<EV::Glib>).
4312 It can be found and installed via CPAN, its homepage is at
4313 L<>.
4315 =item Python
4317 Python bindings can be found at L<>. It
4318 seems to be quite complete and well-documented.
4320 =item Ruby
4322 Tony Arcieri has written a ruby extension that offers access to a subset
4323 of the libev API and adds file handle abstractions, asynchronous DNS and
4324 more on top of it. It can be found via gem servers. Its homepage is at
4325 L<>.
4327 Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
4328 makes rev work even on mingw.
4330 =item Haskell
4332 A haskell binding to libev is available at
4333 L<>.
4335 =item D
4337 Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
4338 be found at L<>.
4340 =item Ocaml
4342 Erkki Seppala has written Ocaml bindings for libev, to be found at
4343 L<>.
4345 =item Lua
4347 Brian Maher has written a partial interface to libev for lua (at the
4348 time of this writing, only C<ev_io> and C<ev_timer>), to be found at
4349 L<>.
4351 =item Javascript
4353 Node.js (L<>) uses libev as the underlying event library.
4355 =item Others
4357 There are others, and I stopped counting.
4359 =back
4362 =head1 MACRO MAGIC
4364 Libev can be compiled with a variety of options, the most fundamental
4365 of which is C<EV_MULTIPLICITY>. This option determines whether (most)
4366 functions and callbacks have an initial C<struct ev_loop *> argument.
4368 To make it easier to write programs that cope with either variant, the
4369 following macros are defined:
4371 =over 4
4373 =item C<EV_A>, C<EV_A_>
4375 This provides the loop I<argument> for functions, if one is required ("ev
4376 loop argument"). The C<EV_A> form is used when this is the sole argument,
4377 C<EV_A_> is used when other arguments are following. Example:
4379 ev_unref (EV_A);
4380 ev_timer_add (EV_A_ watcher);
4381 ev_run (EV_A_ 0);
4383 It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
4384 which is often provided by the following macro.
4386 =item C<EV_P>, C<EV_P_>
4388 This provides the loop I<parameter> for functions, if one is required ("ev
4389 loop parameter"). The C<EV_P> form is used when this is the sole parameter,
4390 C<EV_P_> is used when other parameters are following. Example:
4392 // this is how ev_unref is being declared
4393 static void ev_unref (EV_P);
4395 // this is how you can declare your typical callback
4396 static void cb (EV_P_ ev_timer *w, int revents)
4398 It declares a parameter C<loop> of type C<struct ev_loop *>, quite
4399 suitable for use with C<EV_A>.
4401 =item C<EV_DEFAULT>, C<EV_DEFAULT_>
4403 Similar to the other two macros, this gives you the value of the default
4404 loop, if multiple loops are supported ("ev loop default"). The default loop
4405 will be initialised if it isn't already initialised.
4407 For non-multiplicity builds, these macros do nothing, so you always have
4408 to initialise the loop somewhere.
4412 Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
4413 default loop has been initialised (C<UC> == unchecked). Their behaviour
4414 is undefined when the default loop has not been initialised by a previous
4415 execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
4417 It is often prudent to use C<EV_DEFAULT> when initialising the first
4418 watcher in a function but use C<EV_DEFAULT_UC> afterwards.
4420 =back
4422 Example: Declare and initialise a check watcher, utilising the above
4423 macros so it will work regardless of whether multiple loops are supported
4424 or not.
4426 static void
4427 check_cb (EV_P_ ev_timer *w, int revents)
4428 {
4429 ev_check_stop (EV_A_ w);
4430 }
4432 ev_check check;
4433 ev_check_init (&check, check_cb);
4434 ev_check_start (EV_DEFAULT_ &check);
4435 ev_run (EV_DEFAULT_ 0);
4437 =head1 EMBEDDING
4439 Libev can (and often is) directly embedded into host
4440 applications. Examples of applications that embed it include the Deliantra
4441 Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
4442 and rxvt-unicode.
4444 The goal is to enable you to just copy the necessary files into your
4445 source directory without having to change even a single line in them, so
4446 you can easily upgrade by simply copying (or having a checked-out copy of
4447 libev somewhere in your source tree).
4449 =head2 FILESETS
4451 Depending on what features you need you need to include one or more sets of files
4452 in your application.
4454 =head3 CORE EVENT LOOP
4456 To include only the libev core (all the C<ev_*> functions), with manual
4457 configuration (no autoconf):
4459 #define EV_STANDALONE 1
4460 #include "ev.c"
4462 This will automatically include F<ev.h>, too, and should be done in a
4463 single C source file only to provide the function implementations. To use
4464 it, do the same for F<ev.h> in all files wishing to use this API (best
4465 done by writing a wrapper around F<ev.h> that you can include instead and
4466 where you can put other configuration options):
4468 #define EV_STANDALONE 1
4469 #include "ev.h"
4471 Both header files and implementation files can be compiled with a C++
4472 compiler (at least, that's a stated goal, and breakage will be treated
4473 as a bug).
4475 You need the following files in your source tree, or in a directory
4476 in your include path (e.g. in libev/ when using -Ilibev):
4478 ev.h
4479 ev.c
4480 ev_vars.h
4481 ev_wrap.h
4483 ev_win32.c required on win32 platforms only
4485 ev_select.c only when select backend is enabled
4486 ev_poll.c only when poll backend is enabled
4487 ev_epoll.c only when the epoll backend is enabled
4488 ev_linuxaio.c only when the linux aio backend is enabled
4489 ev_iouring.c only when the linux io_uring backend is enabled
4490 ev_kqueue.c only when the kqueue backend is enabled
4491 ev_port.c only when the solaris port backend is enabled
4493 F<ev.c> includes the backend files directly when enabled, so you only need
4494 to compile this single file.
4498 To include the libevent compatibility API, also include:
4500 #include "event.c"
4502 in the file including F<ev.c>, and:
4504 #include "event.h"
4506 in the files that want to use the libevent API. This also includes F<ev.h>.
4508 You need the following additional files for this:
4510 event.h
4511 event.c
4515 Instead of using C<EV_STANDALONE=1> and providing your configuration in
4516 whatever way you want, you can also C<m4_include([libev.m4])> in your
4517 F<> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
4518 include F<config.h> and configure itself accordingly.
4520 For this of course you need the m4 file:
4522 libev.m4
4526 Libev can be configured via a variety of preprocessor symbols you have to
4527 define before including (or compiling) any of its files. The default in
4528 the absence of autoconf is documented for every option.
4530 Symbols marked with "(h)" do not change the ABI, and can have different
4531 values when compiling libev vs. including F<ev.h>, so it is permissible
4532 to redefine them before including F<ev.h> without breaking compatibility
4533 to a compiled library. All other symbols change the ABI, which means all
4534 users of libev and the libev code itself must be compiled with compatible
4535 settings.
4537 =over 4
4539 =item EV_COMPAT3 (h)
4541 Backwards compatibility is a major concern for libev. This is why this
4542 release of libev comes with wrappers for the functions and symbols that
4543 have been renamed between libev version 3 and 4.
4545 You can disable these wrappers (to test compatibility with future
4546 versions) by defining C<EV_COMPAT3> to C<0> when compiling your
4547 sources. This has the additional advantage that you can drop the C<struct>
4548 from C<struct ev_loop> declarations, as libev will provide an C<ev_loop>
4549 typedef in that case.
4551 In some future version, the default for C<EV_COMPAT3> will become C<0>,
4552 and in some even more future version the compatibility code will be
4553 removed completely.
4555 =item EV_STANDALONE (h)
4557 Must always be C<1> if you do not use autoconf configuration, which
4558 keeps libev from including F<config.h>, and it also defines dummy
4559 implementations for some libevent functions (such as logging, which is not
4560 supported). It will also not define any of the structs usually found in
4561 F<event.h> that are not directly supported by the libev core alone.
4563 In standalone mode, libev will still try to automatically deduce the
4564 configuration, but has to be more conservative.
4566 =item EV_USE_FLOOR
4568 If defined to be C<1>, libev will use the C<floor ()> function for its
4569 periodic reschedule calculations, otherwise libev will fall back on a
4570 portable (slower) implementation. If you enable this, you usually have to
4571 link against libm or something equivalent. Enabling this when the C<floor>
4572 function is not available will fail, so the safe default is to not enable
4573 this.
4577 If defined to be C<1>, libev will try to detect the availability of the
4578 monotonic clock option at both compile time and runtime. Otherwise no
4579 use of the monotonic clock option will be attempted. If you enable this,
4580 you usually have to link against librt or something similar. Enabling it
4581 when the functionality isn't available is safe, though, although you have
4582 to make sure you link against any libraries where the C<clock_gettime>
4583 function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
4585 =item EV_USE_REALTIME
4587 If defined to be C<1>, libev will try to detect the availability of the
4588 real-time clock option at compile time (and assume its availability
4589 at runtime if successful). Otherwise no use of the real-time clock
4590 option will be attempted. This effectively replaces C<gettimeofday>
4591 by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
4592 correctness. See the note about libraries in the description of
4593 C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
4598 If defined to be C<1>, libev will try to use a direct syscall instead
4599 of calling the system-provided C<clock_gettime> function. This option
4600 exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
4601 unconditionally pulls in C<libpthread>, slowing down single-threaded
4602 programs needlessly. Using a direct syscall is slightly slower (in
4603 theory), because no optimised vdso implementation can be used, but avoids
4604 the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
4605 higher, as it simplifies linking (no need for C<-lrt>).
4609 If defined to be C<1>, libev will assume that C<nanosleep ()> is available
4610 and will use it for delays. Otherwise it will use C<select ()>.
4612 =item EV_USE_EVENTFD
4614 If defined to be C<1>, then libev will assume that C<eventfd ()> is
4615 available and will probe for kernel support at runtime. This will improve
4616 C<ev_signal> and C<ev_async> performance and reduce resource consumption.
4617 If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
4618 2.7 or newer, otherwise disabled.
4620 =item EV_USE_SELECT
4622 If undefined or defined to be C<1>, libev will compile in support for the
4623 C<select>(2) backend. No attempt at auto-detection will be done: if no
4624 other method takes over, select will be it. Otherwise the select backend
4625 will not be compiled in.
4629 If defined to C<1>, then the select backend will use the system C<fd_set>
4630 structure. This is useful if libev doesn't compile due to a missing
4631 C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
4632 on exotic systems. This usually limits the range of file descriptors to
4633 some low limit such as 1024 or might have other limitations (winsocket
4634 only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
4635 configures the maximum size of the C<fd_set>.
4639 When defined to C<1>, the select backend will assume that
4640 select/socket/connect etc. don't understand file descriptors but
4641 wants osf handles on win32 (this is the case when the select to
4642 be used is the winsock select). This means that it will call
4643 C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
4644 it is assumed that all these functions actually work on fds, even
4645 on win32. Should not be defined on non-win32 platforms.
4647 =item EV_FD_TO_WIN32_HANDLE(fd)
4649 If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
4650 file descriptors to socket handles. When not defining this symbol (the
4651 default), then libev will call C<_get_osfhandle>, which is usually
4652 correct. In some cases, programs use their own file descriptor management,
4653 in which case they can provide this function to map fds to socket handles.
4655 =item EV_WIN32_HANDLE_TO_FD(handle)
4657 If C<EV_SELECT_IS_WINSOCKET> then libev maps handles to file descriptors
4658 using the standard C<_open_osfhandle> function. For programs implementing
4659 their own fd to handle mapping, overwriting this function makes it easier
4660 to do so. This can be done by defining this macro to an appropriate value.
4662 =item EV_WIN32_CLOSE_FD(fd)
4664 If programs implement their own fd to handle mapping on win32, then this
4665 macro can be used to override the C<close> function, useful to unregister
4666 file descriptors again. Note that the replacement function has to close
4667 the underlying OS handle.
4671 If defined to be C<1>, libev will use C<WSASocket> to create its internal
4672 communication socket, which works better in some environments. Otherwise,
4673 the normal C<socket> function will be used, which works better in other
4674 environments.
4676 =item EV_USE_POLL
4678 If defined to be C<1>, libev will compile in support for the C<poll>(2)
4679 backend. Otherwise it will be enabled on non-win32 platforms. It
4680 takes precedence over select.
4682 =item EV_USE_EPOLL
4684 If defined to be C<1>, libev will compile in support for the Linux
4685 C<epoll>(7) backend. Its availability will be detected at runtime,
4686 otherwise another method will be used as fallback. This is the preferred
4687 backend for GNU/Linux systems. If undefined, it will be enabled if the
4688 headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
4690 =item EV_USE_LINUXAIO
4692 If defined to be C<1>, libev will compile in support for the Linux aio
4693 backend (C<EV_USE_EPOLL> must also be enabled). If undefined, it will be
4694 enabled on linux, otherwise disabled.
4696 =item EV_USE_IOURING
4698 If defined to be C<1>, libev will compile in support for the Linux
4699 io_uring backend (C<EV_USE_EPOLL> must also be enabled). Due to it's
4700 current limitations it has to be requested explicitly. If undefined, it
4701 will be enabled on linux, otherwise disabled.
4703 =item EV_USE_KQUEUE
4705 If defined to be C<1>, libev will compile in support for the BSD style
4706 C<kqueue>(2) backend. Its actual availability will be detected at runtime,
4707 otherwise another method will be used as fallback. This is the preferred
4708 backend for BSD and BSD-like systems, although on most BSDs kqueue only
4709 supports some types of fds correctly (the only platform we found that
4710 supports ptys for example was NetBSD), so kqueue might be compiled in, but
4711 not be used unless explicitly requested. The best way to use it is to find
4712 out whether kqueue supports your type of fd properly and use an embedded
4713 kqueue loop.
4715 =item EV_USE_PORT
4717 If defined to be C<1>, libev will compile in support for the Solaris
4718 10 port style backend. Its availability will be detected at runtime,
4719 otherwise another method will be used as fallback. This is the preferred
4720 backend for Solaris 10 systems.
4722 =item EV_USE_DEVPOLL
4724 Reserved for future expansion, works like the USE symbols above.
4726 =item EV_USE_INOTIFY
4728 If defined to be C<1>, libev will compile in support for the Linux inotify
4729 interface to speed up C<ev_stat> watchers. Its actual availability will
4730 be detected at runtime. If undefined, it will be enabled if the headers
4731 indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
4733 =item EV_NO_SMP
4735 If defined to be C<1>, libev will assume that memory is always coherent
4736 between threads, that is, threads can be used, but threads never run on
4737 different cpus (or different cpu cores). This reduces dependencies
4738 and makes libev faster.
4740 =item EV_NO_THREADS
4742 If defined to be C<1>, libev will assume that it will never be called from
4743 different threads (that includes signal handlers), which is a stronger
4744 assumption than C<EV_NO_SMP>, above. This reduces dependencies and makes
4745 libev faster.
4747 =item EV_ATOMIC_T
4749 Libev requires an integer type (suitable for storing C<0> or C<1>) whose
4750 access is atomic with respect to other threads or signal contexts. No
4751 such type is easily found in the C language, so you can provide your own
4752 type that you know is safe for your purposes. It is used both for signal
4753 handler "locking" as well as for signal and thread safety in C<ev_async>
4754 watchers.
4756 In the absence of this define, libev will use C<sig_atomic_t volatile>
4757 (from F<signal.h>), which is usually good enough on most platforms.
4759 =item EV_H (h)
4761 The name of the F<ev.h> header file used to include it. The default if
4762 undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
4763 used to virtually rename the F<ev.h> header file in case of conflicts.
4765 =item EV_CONFIG_H (h)
4767 If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
4768 F<ev.c>'s idea of where to find the F<config.h> file, similarly to
4769 C<EV_H>, above.
4771 =item EV_EVENT_H (h)
4773 Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
4774 of how the F<event.h> header can be found, the default is C<"event.h">.
4776 =item EV_PROTOTYPES (h)
4778 If defined to be C<0>, then F<ev.h> will not define any function
4779 prototypes, but still define all the structs and other symbols. This is
4780 occasionally useful if you want to provide your own wrapper functions
4781 around libev functions.
4785 If undefined or defined to C<1>, then all event-loop-specific functions
4786 will have the C<struct ev_loop *> as first argument, and you can create
4787 additional independent event loops. Otherwise there will be no support
4788 for multiple event loops and there is no first event loop pointer
4789 argument. Instead, all functions act on the single default loop.
4791 Note that C<EV_DEFAULT> and C<EV_DEFAULT_> will no longer provide a
4792 default loop when multiplicity is switched off - you always have to
4793 initialise the loop manually in this case.
4795 =item EV_MINPRI
4797 =item EV_MAXPRI
4799 The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
4800 C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
4801 provide for more priorities by overriding those symbols (usually defined
4802 to be C<-2> and C<2>, respectively).
4804 When doing priority-based operations, libev usually has to linearly search
4805 all the priorities, so having many of them (hundreds) uses a lot of space
4806 and time, so using the defaults of five priorities (-2 .. +2) is usually
4807 fine.
4809 If your embedding application does not need any priorities, defining these
4810 both to C<0> will save some memory and CPU.
4816 If undefined or defined to be C<1> (and the platform supports it), then
4817 the respective watcher type is supported. If defined to be C<0>, then it
4818 is not. Disabling watcher types mainly saves code size.
4820 =item EV_FEATURES
4822 If you need to shave off some kilobytes of code at the expense of some
4823 speed (but with the full API), you can define this symbol to request
4824 certain subsets of functionality. The default is to enable all features
4825 that can be enabled on the platform.
4827 A typical way to use this symbol is to define it to C<0> (or to a bitset
4828 with some broad features you want) and then selectively re-enable
4829 additional parts you want, for example if you want everything minimal,
4830 but multiple event loop support, async and child watchers and the poll
4831 backend, use this:
4833 #define EV_FEATURES 0
4834 #define EV_MULTIPLICITY 1
4835 #define EV_USE_POLL 1
4836 #define EV_CHILD_ENABLE 1
4837 #define EV_ASYNC_ENABLE 1
4839 The actual value is a bitset, it can be a combination of the following
4840 values (by default, all of these are enabled):
4842 =over 4
4844 =item C<1> - faster/larger code
4846 Use larger code to speed up some operations.
4848 Currently this is used to override some inlining decisions (enlarging the
4849 code size by roughly 30% on amd64).
4851 When optimising for size, use of compiler flags such as C<-Os> with
4852 gcc is recommended, as well as C<-DNDEBUG>, as libev contains a number of
4853 assertions.
4855 The default is off when C<__OPTIMIZE_SIZE__> is defined by your compiler
4856 (e.g. gcc with C<-Os>).
4858 =item C<2> - faster/larger data structures
4860 Replaces the small 2-heap for timer management by a faster 4-heap, larger
4861 hash table sizes and so on. This will usually further increase code size
4862 and can additionally have an effect on the size of data structures at
4863 runtime.
4865 The default is off when C<__OPTIMIZE_SIZE__> is defined by your compiler
4866 (e.g. gcc with C<-Os>).
4868 =item C<4> - full API configuration
4870 This enables priorities (sets C<EV_MAXPRI>=2 and C<EV_MINPRI>=-2), and
4871 enables multiplicity (C<EV_MULTIPLICITY>=1).
4873 =item C<8> - full API
4875 This enables a lot of the "lesser used" API functions. See C<ev.h> for
4876 details on which parts of the API are still available without this
4877 feature, and do not complain if this subset changes over time.
4879 =item C<16> - enable all optional watcher types
4881 Enables all optional watcher types. If you want to selectively enable
4882 only some watcher types other than I/O and timers (e.g. prepare,
4883 embed, async, child...) you can enable them manually by defining
4884 C<EV_watchertype_ENABLE> to C<1> instead.
4886 =item C<32> - enable all backends
4888 This enables all backends - without this feature, you need to enable at
4889 least one backend manually (C<EV_USE_SELECT> is a good choice).
4891 =item C<64> - enable OS-specific "helper" APIs
4893 Enable inotify, eventfd, signalfd and similar OS-specific helper APIs by
4894 default.
4896 =back
4898 Compiling with C<gcc -Os -DEV_STANDALONE -DEV_USE_EPOLL=1 -DEV_FEATURES=0>
4899 reduces the compiled size of libev from 24.7Kb code/2.8Kb data to 6.5Kb
4900 code/0.3Kb data on my GNU/Linux amd64 system, while still giving you I/O
4901 watchers, timers and monotonic clock support.
4903 With an intelligent-enough linker (gcc+binutils are intelligent enough
4904 when you use C<-Wl,--gc-sections -ffunction-sections>) functions unused by
4905 your program might be left out as well - a binary starting a timer and an
4906 I/O watcher then might come out at only 5Kb.
4908 =item EV_API_STATIC
4910 If this symbol is defined (by default it is not), then all identifiers
4911 will have static linkage. This means that libev will not export any
4912 identifiers, and you cannot link against libev anymore. This can be useful
4913 when you embed libev, only want to use libev functions in a single file,
4914 and do not want its identifiers to be visible.
4916 To use this, define C<EV_API_STATIC> and include F<ev.c> in the file that
4917 wants to use libev.
4919 This option only works when libev is compiled with a C compiler, as C++
4920 doesn't support the required declaration syntax.
4922 =item EV_AVOID_STDIO
4924 If this is set to C<1> at compiletime, then libev will avoid using stdio
4925 functions (printf, scanf, perror etc.). This will increase the code size
4926 somewhat, but if your program doesn't otherwise depend on stdio and your
4927 libc allows it, this avoids linking in the stdio library which is quite
4928 big.
4930 Note that error messages might become less precise when this option is
4931 enabled.
4933 =item EV_NSIG
4935 The highest supported signal number, +1 (or, the number of
4936 signals): Normally, libev tries to deduce the maximum number of signals
4937 automatically, but sometimes this fails, in which case it can be
4938 specified. Also, using a lower number than detected (C<32> should be
4939 good for about any system in existence) can save some memory, as libev
4940 statically allocates some 12-24 bytes per signal number.
4942 =item EV_PID_HASHSIZE
4944 C<ev_child> watchers use a small hash table to distribute workload by
4945 pid. The default size is C<16> (or C<1> with C<EV_FEATURES> disabled),
4946 usually more than enough. If you need to manage thousands of children you
4947 might want to increase this value (I<must> be a power of two).
4951 C<ev_stat> watchers use a small hash table to distribute workload by
4952 inotify watch id. The default size is C<16> (or C<1> with C<EV_FEATURES>
4953 disabled), usually more than enough. If you need to manage thousands of
4954 C<ev_stat> watchers you might want to increase this value (I<must> be a
4955 power of two).
4957 =item EV_USE_4HEAP
4959 Heaps are not very cache-efficient. To improve the cache-efficiency of the
4960 timer and periodics heaps, libev uses a 4-heap when this symbol is defined