ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/ev.pod
Revision: 1.445
Committed: Fri Dec 21 06:54:30 2018 UTC (5 years, 4 months ago) by root
Branch: MAIN
CVS Tags: rel-4_25
Changes since 1.444: +6 -6 lines
Log Message:
*** empty log message ***

File Contents

# Content
1 =encoding utf-8
2
3 =head1 NAME
4
5 libev - a high performance full-featured event loop written in C
6
7 =head1 SYNOPSIS
8
9 #include <ev.h>
10
11 =head2 EXAMPLE PROGRAM
12
13 // a single header file is required
14 #include <ev.h>
15
16 #include <stdio.h> // for puts
17
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;
22
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);
32
33 // this causes all nested ev_run's to stop iterating
34 ev_break (EV_A_ EVBREAK_ALL);
35 }
36
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 }
45
46 int
47 main (void)
48 {
49 // use the default event loop unless you have special needs
50 struct ev_loop *loop = EV_DEFAULT;
51
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);
56
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);
61
62 // now wait for events to arrive
63 ev_run (loop, 0);
64
65 // break was called, so exit
66 return 0;
67 }
68
69 =head1 ABOUT THIS DOCUMENT
70
71 This document documents the libev software package.
72
73 The newest version of this document is also available as an html-formatted
74 web page you might find easier to navigate when reading it for the first
75 time: L<http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod>.
76
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.
81
82 Familiarity with event based programming techniques in general is assumed
83 throughout this document.
84
85 =head1 WHAT TO READ WHEN IN A HURRY
86
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>.
92
93 =head1 ABOUT LIBEV
94
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.
98
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.
102
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.
107
108 =head2 FEATURES
109
110 Libev supports C<select>, C<poll>, the Linux-specific C<epoll>, the
111 BSD-specific C<kqueue> and the Solaris-specific event port mechanisms
112 for file descriptor events (C<ev_io>), the Linux C<inotify> interface
113 (for C<ev_stat>), Linux eventfd/signalfd (for faster and cleaner
114 inter-thread wakeup (C<ev_async>)/signal handling (C<ev_signal>)) relative
115 timers (C<ev_timer>), absolute timers with customised rescheduling
116 (C<ev_periodic>), synchronous signals (C<ev_signal>), process status
117 change events (C<ev_child>), and event watchers dealing with the event
118 loop mechanism itself (C<ev_idle>, C<ev_embed>, C<ev_prepare> and
119 C<ev_check> watchers) as well as file watchers (C<ev_stat>) and even
120 limited support for fork events (C<ev_fork>).
121
122 It also is quite fast (see this
123 L<benchmark|http://libev.schmorp.de/bench.html> comparing it to libevent
124 for example).
125
126 =head2 CONVENTIONS
127
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.
135
136 =head2 TIME REPRESENTATION
137
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.
144
145 Unlike the name component C<stamp> might indicate, it is also used for
146 time differences (e.g. delays) throughout libev.
147
148 =head1 ERROR HANDLING
149
150 Libev knows three classes of errors: operating system errors, usage errors
151 and internal errors (bugs).
152
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 ()>.
158
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.
163
164 Libev also has a few internal error-checking C<assert>ions, and also has
165 extensive consistency checking code. These do not trigger under normal
166 circumstances, as they indicate either a bug in libev or worse.
167
168
169 =head1 GLOBAL FUNCTIONS
170
171 These functions can be called anytime, even before initialising the
172 library in any way.
173
174 =over 4
175
176 =item ev_tstamp ev_time ()
177
178 Returns the current time as libev would use it. Please note that the
179 C<ev_now> function is usually faster and also often returns the timestamp
180 you actually want to know. Also interesting is the combination of
181 C<ev_now_update> and C<ev_now>.
182
183 =item ev_sleep (ev_tstamp interval)
184
185 Sleep for the given interval: The current thread will be blocked
186 until either it is interrupted or the given time interval has
187 passed (approximately - it might return a bit earlier even if not
188 interrupted). Returns immediately if C<< interval <= 0 >>.
189
190 Basically this is a sub-second-resolution C<sleep ()>.
191
192 The range of the C<interval> is limited - libev only guarantees to work
193 with sleep times of up to one day (C<< interval <= 86400 >>).
194
195 =item int ev_version_major ()
196
197 =item int ev_version_minor ()
198
199 You can find out the major and minor ABI version numbers of the library
200 you linked against by calling the functions C<ev_version_major> and
201 C<ev_version_minor>. If you want, you can compare against the global
202 symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the
203 version of the library your program was compiled against.
204
205 These version numbers refer to the ABI version of the library, not the
206 release version.
207
208 Usually, it's a good idea to terminate if the major versions mismatch,
209 as this indicates an incompatible change. Minor versions are usually
210 compatible to older versions, so a larger minor version alone is usually
211 not a problem.
212
213 Example: Make sure we haven't accidentally been linked against the wrong
214 version (note, however, that this will not detect other ABI mismatches,
215 such as LFS or reentrancy).
216
217 assert (("libev version mismatch",
218 ev_version_major () == EV_VERSION_MAJOR
219 && ev_version_minor () >= EV_VERSION_MINOR));
220
221 =item unsigned int ev_supported_backends ()
222
223 Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
224 value) compiled into this binary of libev (independent of their
225 availability on the system you are running on). See C<ev_default_loop> for
226 a description of the set values.
227
228 Example: make sure we have the epoll method, because yeah this is cool and
229 a must have and can we have a torrent of it please!!!11
230
231 assert (("sorry, no epoll, no sex",
232 ev_supported_backends () & EVBACKEND_EPOLL));
233
234 =item unsigned int ev_recommended_backends ()
235
236 Return the set of all backends compiled into this binary of libev and
237 also recommended for this platform, meaning it will work for most file
238 descriptor types. This set is often smaller than the one returned by
239 C<ev_supported_backends>, as for example kqueue is broken on most BSDs
240 and will not be auto-detected unless you explicitly request it (assuming
241 you know what you are doing). This is the set of backends that libev will
242 probe for if you specify no backends explicitly.
243
244 =item unsigned int ev_embeddable_backends ()
245
246 Returns the set of backends that are embeddable in other event loops. This
247 value is platform-specific but can include backends not available on the
248 current system. To find which embeddable backends might be supported on
249 the current system, you would need to look at C<ev_embeddable_backends ()
250 & ev_supported_backends ()>, likewise for recommended ones.
251
252 See the description of C<ev_embed> watchers for more info.
253
254 =item ev_set_allocator (void *(*cb)(void *ptr, long size) throw ())
255
256 Sets the allocation function to use (the prototype is similar - the
257 semantics are identical to the C<realloc> C89/SuS/POSIX function). It is
258 used to allocate and free memory (no surprises here). If it returns zero
259 when memory needs to be allocated (C<size != 0>), the library might abort
260 or take some potentially destructive action.
261
262 Since some systems (at least OpenBSD and Darwin) fail to implement
263 correct C<realloc> semantics, libev will use a wrapper around the system
264 C<realloc> and C<free> functions by default.
265
266 You could override this function in high-availability programs to, say,
267 free some memory if it cannot allocate memory, to use a special allocator,
268 or even to sleep a while and retry until some memory is available.
269
270 Example: Replace the libev allocator with one that waits a bit and then
271 retries (example requires a standards-compliant C<realloc>).
272
273 static void *
274 persistent_realloc (void *ptr, size_t size)
275 {
276 for (;;)
277 {
278 void *newptr = realloc (ptr, size);
279
280 if (newptr)
281 return newptr;
282
283 sleep (60);
284 }
285 }
286
287 ...
288 ev_set_allocator (persistent_realloc);
289
290 =item ev_set_syserr_cb (void (*cb)(const char *msg) throw ())
291
292 Set the callback function to call on a retryable system call error (such
293 as failed select, poll, epoll_wait). The message is a printable string
294 indicating the system call or subsystem causing the problem. If this
295 callback is set, then libev will expect it to remedy the situation, no
296 matter what, when it returns. That is, libev will generally retry the
297 requested operation, or, if the condition doesn't go away, do bad stuff
298 (such as abort).
299
300 Example: This is basically the same thing that libev does internally, too.
301
302 static void
303 fatal_error (const char *msg)
304 {
305 perror (msg);
306 abort ();
307 }
308
309 ...
310 ev_set_syserr_cb (fatal_error);
311
312 =item ev_feed_signal (int signum)
313
314 This function can be used to "simulate" a signal receive. It is completely
315 safe to call this function at any time, from any context, including signal
316 handlers or random threads.
317
318 Its main use is to customise signal handling in your process, especially
319 in the presence of threads. For example, you could block signals
320 by default in all threads (and specifying C<EVFLAG_NOSIGMASK> when
321 creating any loops), and in one thread, use C<sigwait> or any other
322 mechanism to wait for signals, then "deliver" them to libev by calling
323 C<ev_feed_signal>.
324
325 =back
326
327 =head1 FUNCTIONS CONTROLLING EVENT LOOPS
328
329 An event loop is described by a C<struct ev_loop *> (the C<struct> is
330 I<not> optional in this case unless libev 3 compatibility is disabled, as
331 libev 3 had an C<ev_loop> function colliding with the struct name).
332
333 The library knows two types of such loops, the I<default> loop, which
334 supports child process events, and dynamically created event loops which
335 do not.
336
337 =over 4
338
339 =item struct ev_loop *ev_default_loop (unsigned int flags)
340
341 This returns the "default" event loop object, which is what you should
342 normally use when you just need "the event loop". Event loop objects and
343 the C<flags> parameter are described in more detail in the entry for
344 C<ev_loop_new>.
345
346 If the default loop is already initialised then this function simply
347 returns it (and ignores the flags. If that is troubling you, check
348 C<ev_backend ()> afterwards). Otherwise it will create it with the given
349 flags, which should almost always be C<0>, unless the caller is also the
350 one calling C<ev_run> or otherwise qualifies as "the main program".
351
352 If you don't know what event loop to use, use the one returned from this
353 function (or via the C<EV_DEFAULT> macro).
354
355 Note that this function is I<not> thread-safe, so if you want to use it
356 from multiple threads, you have to employ some kind of mutex (note also
357 that this case is unlikely, as loops cannot be shared easily between
358 threads anyway).
359
360 The default loop is the only loop that can handle C<ev_child> watchers,
361 and to do this, it always registers a handler for C<SIGCHLD>. If this is
362 a problem for your application you can either create a dynamic loop with
363 C<ev_loop_new> which doesn't do that, or you can simply overwrite the
364 C<SIGCHLD> signal handler I<after> calling C<ev_default_init>.
365
366 Example: This is the most typical usage.
367
368 if (!ev_default_loop (0))
369 fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
370
371 Example: Restrict libev to the select and poll backends, and do not allow
372 environment settings to be taken into account:
373
374 ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
375
376 =item struct ev_loop *ev_loop_new (unsigned int flags)
377
378 This will create and initialise a new event loop object. If the loop
379 could not be initialised, returns false.
380
381 This function is thread-safe, and one common way to use libev with
382 threads is indeed to create one loop per thread, and using the default
383 loop in the "main" or "initial" thread.
384
385 The flags argument can be used to specify special behaviour or specific
386 backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
387
388 The following flags are supported:
389
390 =over 4
391
392 =item C<EVFLAG_AUTO>
393
394 The default flags value. Use this if you have no clue (it's the right
395 thing, believe me).
396
397 =item C<EVFLAG_NOENV>
398
399 If this flag bit is or'ed into the flag value (or the program runs setuid
400 or setgid) then libev will I<not> look at the environment variable
401 C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
402 override the flags completely if it is found in the environment. This is
403 useful to try out specific backends to test their performance, to work
404 around bugs, or to make libev threadsafe (accessing environment variables
405 cannot be done in a threadsafe way, but usually it works if no other
406 thread modifies them).
407
408 =item C<EVFLAG_FORKCHECK>
409
410 Instead of calling C<ev_loop_fork> manually after a fork, you can also
411 make libev check for a fork in each iteration by enabling this flag.
412
413 This works by calling C<getpid ()> on every iteration of the loop,
414 and thus this might slow down your event loop if you do a lot of loop
415 iterations and little real work, but is usually not noticeable (on my
416 GNU/Linux system for example, C<getpid> is actually a simple 5-insn
417 sequence without a system call and thus I<very> fast, but my GNU/Linux
418 system also has C<pthread_atfork> which is even faster). (Update: glibc
419 versions 2.25 apparently removed the C<getpid> optimisation again).
420
421 The big advantage of this flag is that you can forget about fork (and
422 forget about forgetting to tell libev about forking, although you still
423 have to ignore C<SIGPIPE>) when you use this flag.
424
425 This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
426 environment variable.
427
428 =item C<EVFLAG_NOINOTIFY>
429
430 When this flag is specified, then libev will not attempt to use the
431 I<inotify> API for its C<ev_stat> watchers. Apart from debugging and
432 testing, this flag can be useful to conserve inotify file descriptors, as
433 otherwise each loop using C<ev_stat> watchers consumes one inotify handle.
434
435 =item C<EVFLAG_SIGNALFD>
436
437 When this flag is specified, then libev will attempt to use the
438 I<signalfd> API for its C<ev_signal> (and C<ev_child>) watchers. This API
439 delivers signals synchronously, which makes it both faster and might make
440 it possible to get the queued signal data. It can also simplify signal
441 handling with threads, as long as you properly block signals in your
442 threads that are not interested in handling them.
443
444 Signalfd will not be used by default as this changes your signal mask, and
445 there are a lot of shoddy libraries and programs (glib's threadpool for
446 example) that can't properly initialise their signal masks.
447
448 =item C<EVFLAG_NOSIGMASK>
449
450 When this flag is specified, then libev will avoid to modify the signal
451 mask. Specifically, this means you have to make sure signals are unblocked
452 when you want to receive them.
453
454 This behaviour is useful when you want to do your own signal handling, or
455 want to handle signals only in specific threads and want to avoid libev
456 unblocking the signals.
457
458 It's also required by POSIX in a threaded program, as libev calls
459 C<sigprocmask>, whose behaviour is officially unspecified.
460
461 This flag's behaviour will become the default in future versions of libev.
462
463 =item C<EVBACKEND_SELECT> (value 1, portable select backend)
464
465 This is your standard select(2) backend. Not I<completely> standard, as
466 libev tries to roll its own fd_set with no limits on the number of fds,
467 but if that fails, expect a fairly low limit on the number of fds when
468 using this backend. It doesn't scale too well (O(highest_fd)), but its
469 usually the fastest backend for a low number of (low-numbered :) fds.
470
471 To get good performance out of this backend you need a high amount of
472 parallelism (most of the file descriptors should be busy). If you are
473 writing a server, you should C<accept ()> in a loop to accept as many
474 connections as possible during one iteration. You might also want to have
475 a look at C<ev_set_io_collect_interval ()> to increase the amount of
476 readiness notifications you get per iteration.
477
478 This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
479 C<writefds> set (and to work around Microsoft Windows bugs, also onto the
480 C<exceptfds> set on that platform).
481
482 =item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
483
484 And this is your standard poll(2) backend. It's more complicated
485 than select, but handles sparse fds better and has no artificial
486 limit on the number of fds you can use (except it will slow down
487 considerably with a lot of inactive fds). It scales similarly to select,
488 i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
489 performance tips.
490
491 This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
492 C<EV_WRITE> to C<POLLOUT | POLLERR | POLLHUP>.
493
494 =item C<EVBACKEND_EPOLL> (value 4, Linux)
495
496 Use the linux-specific epoll(7) interface (for both pre- and post-2.6.9
497 kernels).
498
499 For few fds, this backend is a bit little slower than poll and select, but
500 it scales phenomenally better. While poll and select usually scale like
501 O(total_fds) where total_fds is the total number of fds (or the highest
502 fd), epoll scales either O(1) or O(active_fds).
503
504 The epoll mechanism deserves honorable mention as the most misdesigned
505 of the more advanced event mechanisms: mere annoyances include silently
506 dropping file descriptors, requiring a system call per change per file
507 descriptor (and unnecessary guessing of parameters), problems with dup,
508 returning before the timeout value, resulting in additional iterations
509 (and only giving 5ms accuracy while select on the same platform gives
510 0.1ms) and so on. The biggest issue is fork races, however - if a program
511 forks then I<both> parent and child process have to recreate the epoll
512 set, which can take considerable time (one syscall per file descriptor)
513 and is of course hard to detect.
514
515 Epoll is also notoriously buggy - embedding epoll fds I<should> work,
516 but of course I<doesn't>, and epoll just loves to report events for
517 totally I<different> file descriptors (even already closed ones, so
518 one cannot even remove them from the set) than registered in the set
519 (especially on SMP systems). Libev tries to counter these spurious
520 notifications by employing an additional generation counter and comparing
521 that against the events to filter out spurious ones, recreating the set
522 when required. Epoll also erroneously rounds down timeouts, but gives you
523 no way to know when and by how much, so sometimes you have to busy-wait
524 because epoll returns immediately despite a nonzero timeout. And last
525 not least, it also refuses to work with some file descriptors which work
526 perfectly fine with C<select> (files, many character devices...).
527
528 Epoll is truly the train wreck among event poll mechanisms, a frankenpoll,
529 cobbled together in a hurry, no thought to design or interaction with
530 others. Oh, the pain, will it ever stop...
531
532 While stopping, setting and starting an I/O watcher in the same iteration
533 will result in some caching, there is still a system call per such
534 incident (because the same I<file descriptor> could point to a different
535 I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
536 file descriptors might not work very well if you register events for both
537 file descriptors.
538
539 Best performance from this backend is achieved by not unregistering all
540 watchers for a file descriptor until it has been closed, if possible,
541 i.e. keep at least one watcher active per fd at all times. Stopping and
542 starting a watcher (without re-setting it) also usually doesn't cause
543 extra overhead. A fork can both result in spurious notifications as well
544 as in libev having to destroy and recreate the epoll object, which can
545 take considerable time and thus should be avoided.
546
547 All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
548 faster than epoll for maybe up to a hundred file descriptors, depending on
549 the usage. So sad.
550
551 While nominally embeddable in other event loops, this feature is broken in
552 all kernel versions tested so far.
553
554 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
555 C<EVBACKEND_POLL>.
556
557 =item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
558
559 Kqueue deserves special mention, as at the time of this writing, it
560 was broken on all BSDs except NetBSD (usually it doesn't work reliably
561 with anything but sockets and pipes, except on Darwin, where of course
562 it's completely useless). Unlike epoll, however, whose brokenness
563 is by design, these kqueue bugs can (and eventually will) be fixed
564 without API changes to existing programs. For this reason it's not being
565 "auto-detected" unless you explicitly specify it in the flags (i.e. using
566 C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
567 system like NetBSD.
568
569 You still can embed kqueue into a normal poll or select backend and use it
570 only for sockets (after having made sure that sockets work with kqueue on
571 the target platform). See C<ev_embed> watchers for more info.
572
573 It scales in the same way as the epoll backend, but the interface to the
574 kernel is more efficient (which says nothing about its actual speed, of
575 course). While stopping, setting and starting an I/O watcher does never
576 cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
577 two event changes per incident. Support for C<fork ()> is very bad (you
578 might have to leak fd's on fork, but it's more sane than epoll) and it
579 drops fds silently in similarly hard-to-detect cases.
580
581 This backend usually performs well under most conditions.
582
583 While nominally embeddable in other event loops, this doesn't work
584 everywhere, so you might need to test for this. And since it is broken
585 almost everywhere, you should only use it when you have a lot of sockets
586 (for which it usually works), by embedding it into another event loop
587 (e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
588 also broken on OS X)) and, did I mention it, using it only for sockets.
589
590 This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
591 C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
592 C<NOTE_EOF>.
593
594 =item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
595
596 This is not implemented yet (and might never be, unless you send me an
597 implementation). According to reports, C</dev/poll> only supports sockets
598 and is not embeddable, which would limit the usefulness of this backend
599 immensely.
600
601 =item C<EVBACKEND_PORT> (value 32, Solaris 10)
602
603 This uses the Solaris 10 event port mechanism. As with everything on Solaris,
604 it's really slow, but it still scales very well (O(active_fds)).
605
606 While this backend scales well, it requires one system call per active
607 file descriptor per loop iteration. For small and medium numbers of file
608 descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
609 might perform better.
610
611 On the positive side, this backend actually performed fully to
612 specification in all tests and is fully embeddable, which is a rare feat
613 among the OS-specific backends (I vastly prefer correctness over speed
614 hacks).
615
616 On the negative side, the interface is I<bizarre> - so bizarre that
617 even sun itself gets it wrong in their code examples: The event polling
618 function sometimes returns events to the caller even though an error
619 occurred, but with no indication whether it has done so or not (yes, it's
620 even documented that way) - deadly for edge-triggered interfaces where you
621 absolutely have to know whether an event occurred or not because you have
622 to re-arm the watcher.
623
624 Fortunately libev seems to be able to work around these idiocies.
625
626 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
627 C<EVBACKEND_POLL>.
628
629 =item C<EVBACKEND_ALL>
630
631 Try all backends (even potentially broken ones that wouldn't be tried
632 with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
633 C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.
634
635 It is definitely not recommended to use this flag, use whatever
636 C<ev_recommended_backends ()> returns, or simply do not specify a backend
637 at all.
638
639 =item C<EVBACKEND_MASK>
640
641 Not a backend at all, but a mask to select all backend bits from a
642 C<flags> value, in case you want to mask out any backends from a flags
643 value (e.g. when modifying the C<LIBEV_FLAGS> environment variable).
644
645 =back
646
647 If one or more of the backend flags are or'ed into the flags value,
648 then only these backends will be tried (in the reverse order as listed
649 here). If none are specified, all backends in C<ev_recommended_backends
650 ()> will be tried.
651
652 Example: Try to create a event loop that uses epoll and nothing else.
653
654 struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
655 if (!epoller)
656 fatal ("no epoll found here, maybe it hides under your chair");
657
658 Example: Use whatever libev has to offer, but make sure that kqueue is
659 used if available.
660
661 struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_KQUEUE);
662
663 =item ev_loop_destroy (loop)
664
665 Destroys an event loop object (frees all memory and kernel state
666 etc.). None of the active event watchers will be stopped in the normal
667 sense, so e.g. C<ev_is_active> might still return true. It is your
668 responsibility to either stop all watchers cleanly yourself I<before>
669 calling this function, or cope with the fact afterwards (which is usually
670 the easiest thing, you can just ignore the watchers and/or C<free ()> them
671 for example).
672
673 Note that certain global state, such as signal state (and installed signal
674 handlers), will not be freed by this function, and related watchers (such
675 as signal and child watchers) would need to be stopped manually.
676
677 This function is normally used on loop objects allocated by
678 C<ev_loop_new>, but it can also be used on the default loop returned by
679 C<ev_default_loop>, in which case it is not thread-safe.
680
681 Note that it is not advisable to call this function on the default loop
682 except in the rare occasion where you really need to free its resources.
683 If you need dynamically allocated loops it is better to use C<ev_loop_new>
684 and C<ev_loop_destroy>.
685
686 =item ev_loop_fork (loop)
687
688 This function sets a flag that causes subsequent C<ev_run> iterations
689 to reinitialise the kernel state for backends that have one. Despite
690 the name, you can call it anytime you are allowed to start or stop
691 watchers (except inside an C<ev_prepare> callback), but it makes most
692 sense after forking, in the child process. You I<must> call it (or use
693 C<EVFLAG_FORKCHECK>) in the child before resuming or calling C<ev_run>.
694
695 In addition, if you want to reuse a loop (via this function or
696 C<EVFLAG_FORKCHECK>), you I<also> have to ignore C<SIGPIPE>.
697
698 Again, you I<have> to call it on I<any> loop that you want to re-use after
699 a fork, I<even if you do not plan to use the loop in the parent>. This is
700 because some kernel interfaces *cough* I<kqueue> *cough* do funny things
701 during fork.
702
703 On the other hand, you only need to call this function in the child
704 process if and only if you want to use the event loop in the child. If
705 you just fork+exec or create a new loop in the child, you don't have to
706 call it at all (in fact, C<epoll> is so badly broken that it makes a
707 difference, but libev will usually detect this case on its own and do a
708 costly reset of the backend).
709
710 The function itself is quite fast and it's usually not a problem to call
711 it just in case after a fork.
712
713 Example: Automate calling C<ev_loop_fork> on the default loop when
714 using pthreads.
715
716 static void
717 post_fork_child (void)
718 {
719 ev_loop_fork (EV_DEFAULT);
720 }
721
722 ...
723 pthread_atfork (0, 0, post_fork_child);
724
725 =item int ev_is_default_loop (loop)
726
727 Returns true when the given loop is, in fact, the default loop, and false
728 otherwise.
729
730 =item unsigned int ev_iteration (loop)
731
732 Returns the current iteration count for the event loop, which is identical
733 to the number of times libev did poll for new events. It starts at C<0>
734 and happily wraps around with enough iterations.
735
736 This value can sometimes be useful as a generation counter of sorts (it
737 "ticks" the number of loop iterations), as it roughly corresponds with
738 C<ev_prepare> and C<ev_check> calls - and is incremented between the
739 prepare and check phases.
740
741 =item unsigned int ev_depth (loop)
742
743 Returns the number of times C<ev_run> was entered minus the number of
744 times C<ev_run> was exited normally, in other words, the recursion depth.
745
746 Outside C<ev_run>, this number is zero. In a callback, this number is
747 C<1>, unless C<ev_run> was invoked recursively (or from another thread),
748 in which case it is higher.
749
750 Leaving C<ev_run> abnormally (setjmp/longjmp, cancelling the thread,
751 throwing an exception etc.), doesn't count as "exit" - consider this
752 as a hint to avoid such ungentleman-like behaviour unless it's really
753 convenient, in which case it is fully supported.
754
755 =item unsigned int ev_backend (loop)
756
757 Returns one of the C<EVBACKEND_*> flags indicating the event backend in
758 use.
759
760 =item ev_tstamp ev_now (loop)
761
762 Returns the current "event loop time", which is the time the event loop
763 received events and started processing them. This timestamp does not
764 change as long as callbacks are being processed, and this is also the base
765 time used for relative timers. You can treat it as the timestamp of the
766 event occurring (or more correctly, libev finding out about it).
767
768 =item ev_now_update (loop)
769
770 Establishes the current time by querying the kernel, updating the time
771 returned by C<ev_now ()> in the progress. This is a costly operation and
772 is usually done automatically within C<ev_run ()>.
773
774 This function is rarely useful, but when some event callback runs for a
775 very long time without entering the event loop, updating libev's idea of
776 the current time is a good idea.
777
778 See also L</The special problem of time updates> in the C<ev_timer> section.
779
780 =item ev_suspend (loop)
781
782 =item ev_resume (loop)
783
784 These two functions suspend and resume an event loop, for use when the
785 loop is not used for a while and timeouts should not be processed.
786
787 A typical use case would be an interactive program such as a game: When
788 the user presses C<^Z> to suspend the game and resumes it an hour later it
789 would be best to handle timeouts as if no time had actually passed while
790 the program was suspended. This can be achieved by calling C<ev_suspend>
791 in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
792 C<ev_resume> directly afterwards to resume timer processing.
793
794 Effectively, all C<ev_timer> watchers will be delayed by the time spend
795 between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
796 will be rescheduled (that is, they will lose any events that would have
797 occurred while suspended).
798
799 After calling C<ev_suspend> you B<must not> call I<any> function on the
800 given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
801 without a previous call to C<ev_suspend>.
802
803 Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
804 event loop time (see C<ev_now_update>).
805
806 =item bool ev_run (loop, int flags)
807
808 Finally, this is it, the event handler. This function usually is called
809 after you have initialised all your watchers and you want to start
810 handling events. It will ask the operating system for any new events, call
811 the watcher callbacks, and then repeat the whole process indefinitely: This
812 is why event loops are called I<loops>.
813
814 If the flags argument is specified as C<0>, it will keep handling events
815 until either no event watchers are active anymore or C<ev_break> was
816 called.
817
818 The return value is false if there are no more active watchers (which
819 usually means "all jobs done" or "deadlock"), and true in all other cases
820 (which usually means " you should call C<ev_run> again").
821
822 Please note that an explicit C<ev_break> is usually better than
823 relying on all watchers to be stopped when deciding when a program has
824 finished (especially in interactive programs), but having a program
825 that automatically loops as long as it has to and no longer by virtue
826 of relying on its watchers stopping correctly, that is truly a thing of
827 beauty.
828
829 This function is I<mostly> exception-safe - you can break out of a
830 C<ev_run> call by calling C<longjmp> in a callback, throwing a C++
831 exception and so on. This does not decrement the C<ev_depth> value, nor
832 will it clear any outstanding C<EVBREAK_ONE> breaks.
833
834 A flags value of C<EVRUN_NOWAIT> will look for new events, will handle
835 those events and any already outstanding ones, but will not wait and
836 block your process in case there are no events and will return after one
837 iteration of the loop. This is sometimes useful to poll and handle new
838 events while doing lengthy calculations, to keep the program responsive.
839
840 A flags value of C<EVRUN_ONCE> will look for new events (waiting if
841 necessary) and will handle those and any already outstanding ones. It
842 will block your process until at least one new event arrives (which could
843 be an event internal to libev itself, so there is no guarantee that a
844 user-registered callback will be called), and will return after one
845 iteration of the loop.
846
847 This is useful if you are waiting for some external event in conjunction
848 with something not expressible using other libev watchers (i.e. "roll your
849 own C<ev_run>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
850 usually a better approach for this kind of thing.
851
852 Here are the gory details of what C<ev_run> does (this is for your
853 understanding, not a guarantee that things will work exactly like this in
854 future versions):
855
856 - Increment loop depth.
857 - Reset the ev_break status.
858 - Before the first iteration, call any pending watchers.
859 LOOP:
860 - If EVFLAG_FORKCHECK was used, check for a fork.
861 - If a fork was detected (by any means), queue and call all fork watchers.
862 - Queue and call all prepare watchers.
863 - If ev_break was called, goto FINISH.
864 - If we have been forked, detach and recreate the kernel state
865 as to not disturb the other process.
866 - Update the kernel state with all outstanding changes.
867 - Update the "event loop time" (ev_now ()).
868 - Calculate for how long to sleep or block, if at all
869 (active idle watchers, EVRUN_NOWAIT or not having
870 any active watchers at all will result in not sleeping).
871 - Sleep if the I/O and timer collect interval say so.
872 - Increment loop iteration counter.
873 - Block the process, waiting for any events.
874 - Queue all outstanding I/O (fd) events.
875 - Update the "event loop time" (ev_now ()), and do time jump adjustments.
876 - Queue all expired timers.
877 - Queue all expired periodics.
878 - Queue all idle watchers with priority higher than that of pending events.
879 - Queue all check watchers.
880 - Call all queued watchers in reverse order (i.e. check watchers first).
881 Signals and child watchers are implemented as I/O watchers, and will
882 be handled here by queueing them when their watcher gets executed.
883 - If ev_break has been called, or EVRUN_ONCE or EVRUN_NOWAIT
884 were used, or there are no active watchers, goto FINISH, otherwise
885 continue with step LOOP.
886 FINISH:
887 - Reset the ev_break status iff it was EVBREAK_ONE.
888 - Decrement the loop depth.
889 - Return.
890
891 Example: Queue some jobs and then loop until no events are outstanding
892 anymore.
893
894 ... queue jobs here, make sure they register event watchers as long
895 ... as they still have work to do (even an idle watcher will do..)
896 ev_run (my_loop, 0);
897 ... jobs done or somebody called break. yeah!
898
899 =item ev_break (loop, how)
900
901 Can be used to make a call to C<ev_run> return early (but only after it
902 has processed all outstanding events). The C<how> argument must be either
903 C<EVBREAK_ONE>, which will make the innermost C<ev_run> call return, or
904 C<EVBREAK_ALL>, which will make all nested C<ev_run> calls return.
905
906 This "break state" will be cleared on the next call to C<ev_run>.
907
908 It is safe to call C<ev_break> from outside any C<ev_run> calls, too, in
909 which case it will have no effect.
910
911 =item ev_ref (loop)
912
913 =item ev_unref (loop)
914
915 Ref/unref can be used to add or remove a reference count on the event
916 loop: Every watcher keeps one reference, and as long as the reference
917 count is nonzero, C<ev_run> will not return on its own.
918
919 This is useful when you have a watcher that you never intend to
920 unregister, but that nevertheless should not keep C<ev_run> from
921 returning. In such a case, call C<ev_unref> after starting, and C<ev_ref>
922 before stopping it.
923
924 As an example, libev itself uses this for its internal signal pipe: It
925 is not visible to the libev user and should not keep C<ev_run> from
926 exiting if no event watchers registered by it are active. It is also an
927 excellent way to do this for generic recurring timers or from within
928 third-party libraries. Just remember to I<unref after start> and I<ref
929 before stop> (but only if the watcher wasn't active before, or was active
930 before, respectively. Note also that libev might stop watchers itself
931 (e.g. non-repeating timers) in which case you have to C<ev_ref>
932 in the callback).
933
934 Example: Create a signal watcher, but keep it from keeping C<ev_run>
935 running when nothing else is active.
936
937 ev_signal exitsig;
938 ev_signal_init (&exitsig, sig_cb, SIGINT);
939 ev_signal_start (loop, &exitsig);
940 ev_unref (loop);
941
942 Example: For some weird reason, unregister the above signal handler again.
943
944 ev_ref (loop);
945 ev_signal_stop (loop, &exitsig);
946
947 =item ev_set_io_collect_interval (loop, ev_tstamp interval)
948
949 =item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
950
951 These advanced functions influence the time that libev will spend waiting
952 for events. Both time intervals are by default C<0>, meaning that libev
953 will try to invoke timer/periodic callbacks and I/O callbacks with minimum
954 latency.
955
956 Setting these to a higher value (the C<interval> I<must> be >= C<0>)
957 allows libev to delay invocation of I/O and timer/periodic callbacks
958 to increase efficiency of loop iterations (or to increase power-saving
959 opportunities).
960
961 The idea is that sometimes your program runs just fast enough to handle
962 one (or very few) event(s) per loop iteration. While this makes the
963 program responsive, it also wastes a lot of CPU time to poll for new
964 events, especially with backends like C<select ()> which have a high
965 overhead for the actual polling but can deliver many events at once.
966
967 By setting a higher I<io collect interval> you allow libev to spend more
968 time collecting I/O events, so you can handle more events per iteration,
969 at the cost of increasing latency. Timeouts (both C<ev_periodic> and
970 C<ev_timer>) will not be affected. Setting this to a non-null value will
971 introduce an additional C<ev_sleep ()> call into most loop iterations. The
972 sleep time ensures that libev will not poll for I/O events more often then
973 once per this interval, on average (as long as the host time resolution is
974 good enough).
975
976 Likewise, by setting a higher I<timeout collect interval> you allow libev
977 to spend more time collecting timeouts, at the expense of increased
978 latency/jitter/inexactness (the watcher callback will be called
979 later). C<ev_io> watchers will not be affected. Setting this to a non-null
980 value will not introduce any overhead in libev.
981
982 Many (busy) programs can usually benefit by setting the I/O collect
983 interval to a value near C<0.1> or so, which is often enough for
984 interactive servers (of course not for games), likewise for timeouts. It
985 usually doesn't make much sense to set it to a lower value than C<0.01>,
986 as this approaches the timing granularity of most systems. Note that if
987 you do transactions with the outside world and you can't increase the
988 parallelity, then this setting will limit your transaction rate (if you
989 need to poll once per transaction and the I/O collect interval is 0.01,
990 then you can't do more than 100 transactions per second).
991
992 Setting the I<timeout collect interval> can improve the opportunity for
993 saving power, as the program will "bundle" timer callback invocations that
994 are "near" in time together, by delaying some, thus reducing the number of
995 times the process sleeps and wakes up again. Another useful technique to
996 reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
997 they fire on, say, one-second boundaries only.
998
999 Example: we only need 0.1s timeout granularity, and we wish not to poll
1000 more often than 100 times per second:
1001
1002 ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
1003 ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
1004
1005 =item ev_invoke_pending (loop)
1006
1007 This call will simply invoke all pending watchers while resetting their
1008 pending state. Normally, C<ev_run> does this automatically when required,
1009 but when overriding the invoke callback this call comes handy. This
1010 function can be invoked from a watcher - this can be useful for example
1011 when you want to do some lengthy calculation and want to pass further
1012 event handling to another thread (you still have to make sure only one
1013 thread executes within C<ev_invoke_pending> or C<ev_run> of course).
1014
1015 =item int ev_pending_count (loop)
1016
1017 Returns the number of pending watchers - zero indicates that no watchers
1018 are pending.
1019
1020 =item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
1021
1022 This overrides the invoke pending functionality of the loop: Instead of
1023 invoking all pending watchers when there are any, C<ev_run> will call
1024 this callback instead. This is useful, for example, when you want to
1025 invoke the actual watchers inside another context (another thread etc.).
1026
1027 If you want to reset the callback, use C<ev_invoke_pending> as new
1028 callback.
1029
1030 =item ev_set_loop_release_cb (loop, void (*release)(EV_P) throw (), void (*acquire)(EV_P) throw ())
1031
1032 Sometimes you want to share the same loop between multiple threads. This
1033 can be done relatively simply by putting mutex_lock/unlock calls around
1034 each call to a libev function.
1035
1036 However, C<ev_run> can run an indefinite time, so it is not feasible
1037 to wait for it to return. One way around this is to wake up the event
1038 loop via C<ev_break> and C<ev_async_send>, another way is to set these
1039 I<release> and I<acquire> callbacks on the loop.
1040
1041 When set, then C<release> will be called just before the thread is
1042 suspended waiting for new events, and C<acquire> is called just
1043 afterwards.
1044
1045 Ideally, C<release> will just call your mutex_unlock function, and
1046 C<acquire> will just call the mutex_lock function again.
1047
1048 While event loop modifications are allowed between invocations of
1049 C<release> and C<acquire> (that's their only purpose after all), no
1050 modifications done will affect the event loop, i.e. adding watchers will
1051 have no effect on the set of file descriptors being watched, or the time
1052 waited. Use an C<ev_async> watcher to wake up C<ev_run> when you want it
1053 to take note of any changes you made.
1054
1055 In theory, threads executing C<ev_run> will be async-cancel safe between
1056 invocations of C<release> and C<acquire>.
1057
1058 See also the locking example in the C<THREADS> section later in this
1059 document.
1060
1061 =item ev_set_userdata (loop, void *data)
1062
1063 =item void *ev_userdata (loop)
1064
1065 Set and retrieve a single C<void *> associated with a loop. When
1066 C<ev_set_userdata> has never been called, then C<ev_userdata> returns
1067 C<0>.
1068
1069 These two functions can be used to associate arbitrary data with a loop,
1070 and are intended solely for the C<invoke_pending_cb>, C<release> and
1071 C<acquire> callbacks described above, but of course can be (ab-)used for
1072 any other purpose as well.
1073
1074 =item ev_verify (loop)
1075
1076 This function only does something when C<EV_VERIFY> support has been
1077 compiled in, which is the default for non-minimal builds. It tries to go
1078 through all internal structures and checks them for validity. If anything
1079 is found to be inconsistent, it will print an error message to standard
1080 error and call C<abort ()>.
1081
1082 This can be used to catch bugs inside libev itself: under normal
1083 circumstances, this function will never abort as of course libev keeps its
1084 data structures consistent.
1085
1086 =back
1087
1088
1089 =head1 ANATOMY OF A WATCHER
1090
1091 In the following description, uppercase C<TYPE> in names stands for the
1092 watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
1093 watchers and C<ev_io_start> for I/O watchers.
1094
1095 A watcher is an opaque structure that you allocate and register to record
1096 your interest in some event. To make a concrete example, imagine you want
1097 to wait for STDIN to become readable, you would create an C<ev_io> watcher
1098 for that:
1099
1100 static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
1101 {
1102 ev_io_stop (w);
1103 ev_break (loop, EVBREAK_ALL);
1104 }
1105
1106 struct ev_loop *loop = ev_default_loop (0);
1107
1108 ev_io stdin_watcher;
1109
1110 ev_init (&stdin_watcher, my_cb);
1111 ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
1112 ev_io_start (loop, &stdin_watcher);
1113
1114 ev_run (loop, 0);
1115
1116 As you can see, you are responsible for allocating the memory for your
1117 watcher structures (and it is I<usually> a bad idea to do this on the
1118 stack).
1119
1120 Each watcher has an associated watcher structure (called C<struct ev_TYPE>
1121 or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
1122
1123 Each watcher structure must be initialised by a call to C<ev_init (watcher
1124 *, callback)>, which expects a callback to be provided. This callback is
1125 invoked each time the event occurs (or, in the case of I/O watchers, each
1126 time the event loop detects that the file descriptor given is readable
1127 and/or writable).
1128
1129 Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
1130 macro to configure it, with arguments specific to the watcher type. There
1131 is also a macro to combine initialisation and setting in one call: C<<
1132 ev_TYPE_init (watcher *, callback, ...) >>.
1133
1134 To make the watcher actually watch out for events, you have to start it
1135 with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
1136 *) >>), and you can stop watching for events at any time by calling the
1137 corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
1138
1139 As long as your watcher is active (has been started but not stopped) you
1140 must not touch the values stored in it. Most specifically you must never
1141 reinitialise it or call its C<ev_TYPE_set> macro.
1142
1143 Each and every callback receives the event loop pointer as first, the
1144 registered watcher structure as second, and a bitset of received events as
1145 third argument.
1146
1147 The received events usually include a single bit per event type received
1148 (you can receive multiple events at the same time). The possible bit masks
1149 are:
1150
1151 =over 4
1152
1153 =item C<EV_READ>
1154
1155 =item C<EV_WRITE>
1156
1157 The file descriptor in the C<ev_io> watcher has become readable and/or
1158 writable.
1159
1160 =item C<EV_TIMER>
1161
1162 The C<ev_timer> watcher has timed out.
1163
1164 =item C<EV_PERIODIC>
1165
1166 The C<ev_periodic> watcher has timed out.
1167
1168 =item C<EV_SIGNAL>
1169
1170 The signal specified in the C<ev_signal> watcher has been received by a thread.
1171
1172 =item C<EV_CHILD>
1173
1174 The pid specified in the C<ev_child> watcher has received a status change.
1175
1176 =item C<EV_STAT>
1177
1178 The path specified in the C<ev_stat> watcher changed its attributes somehow.
1179
1180 =item C<EV_IDLE>
1181
1182 The C<ev_idle> watcher has determined that you have nothing better to do.
1183
1184 =item C<EV_PREPARE>
1185
1186 =item C<EV_CHECK>
1187
1188 All C<ev_prepare> watchers are invoked just I<before> C<ev_run> starts to
1189 gather new events, and all C<ev_check> watchers are queued (not invoked)
1190 just after C<ev_run> has gathered them, but before it queues any callbacks
1191 for any received events. That means C<ev_prepare> watchers are the last
1192 watchers invoked before the event loop sleeps or polls for new events, and
1193 C<ev_check> watchers will be invoked before any other watchers of the same
1194 or lower priority within an event loop iteration.
1195
1196 Callbacks of both watcher types can start and stop as many watchers as
1197 they want, and all of them will be taken into account (for example, a
1198 C<ev_prepare> watcher might start an idle watcher to keep C<ev_run> from
1199 blocking).
1200
1201 =item C<EV_EMBED>
1202
1203 The embedded event loop specified in the C<ev_embed> watcher needs attention.
1204
1205 =item C<EV_FORK>
1206
1207 The event loop has been resumed in the child process after fork (see
1208 C<ev_fork>).
1209
1210 =item C<EV_CLEANUP>
1211
1212 The event loop is about to be destroyed (see C<ev_cleanup>).
1213
1214 =item C<EV_ASYNC>
1215
1216 The given async watcher has been asynchronously notified (see C<ev_async>).
1217
1218 =item C<EV_CUSTOM>
1219
1220 Not ever sent (or otherwise used) by libev itself, but can be freely used
1221 by libev users to signal watchers (e.g. via C<ev_feed_event>).
1222
1223 =item C<EV_ERROR>
1224
1225 An unspecified error has occurred, the watcher has been stopped. This might
1226 happen because the watcher could not be properly started because libev
1227 ran out of memory, a file descriptor was found to be closed or any other
1228 problem. Libev considers these application bugs.
1229
1230 You best act on it by reporting the problem and somehow coping with the
1231 watcher being stopped. Note that well-written programs should not receive
1232 an error ever, so when your watcher receives it, this usually indicates a
1233 bug in your program.
1234
1235 Libev will usually signal a few "dummy" events together with an error, for
1236 example it might indicate that a fd is readable or writable, and if your
1237 callbacks is well-written it can just attempt the operation and cope with
1238 the error from read() or write(). This will not work in multi-threaded
1239 programs, though, as the fd could already be closed and reused for another
1240 thing, so beware.
1241
1242 =back
1243
1244 =head2 GENERIC WATCHER FUNCTIONS
1245
1246 =over 4
1247
1248 =item C<ev_init> (ev_TYPE *watcher, callback)
1249
1250 This macro initialises the generic portion of a watcher. The contents
1251 of the watcher object can be arbitrary (so C<malloc> will do). Only
1252 the generic parts of the watcher are initialised, you I<need> to call
1253 the type-specific C<ev_TYPE_set> macro afterwards to initialise the
1254 type-specific parts. For each type there is also a C<ev_TYPE_init> macro
1255 which rolls both calls into one.
1256
1257 You can reinitialise a watcher at any time as long as it has been stopped
1258 (or never started) and there are no pending events outstanding.
1259
1260 The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
1261 int revents)>.
1262
1263 Example: Initialise an C<ev_io> watcher in two steps.
1264
1265 ev_io w;
1266 ev_init (&w, my_cb);
1267 ev_io_set (&w, STDIN_FILENO, EV_READ);
1268
1269 =item C<ev_TYPE_set> (ev_TYPE *watcher, [args])
1270
1271 This macro initialises the type-specific parts of a watcher. You need to
1272 call C<ev_init> at least once before you call this macro, but you can
1273 call C<ev_TYPE_set> any number of times. You must not, however, call this
1274 macro on a watcher that is active (it can be pending, however, which is a
1275 difference to the C<ev_init> macro).
1276
1277 Although some watcher types do not have type-specific arguments
1278 (e.g. C<ev_prepare>) you still need to call its C<set> macro.
1279
1280 See C<ev_init>, above, for an example.
1281
1282 =item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
1283
1284 This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
1285 calls into a single call. This is the most convenient method to initialise
1286 a watcher. The same limitations apply, of course.
1287
1288 Example: Initialise and set an C<ev_io> watcher in one step.
1289
1290 ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
1291
1292 =item C<ev_TYPE_start> (loop, ev_TYPE *watcher)
1293
1294 Starts (activates) the given watcher. Only active watchers will receive
1295 events. If the watcher is already active nothing will happen.
1296
1297 Example: Start the C<ev_io> watcher that is being abused as example in this
1298 whole section.
1299
1300 ev_io_start (EV_DEFAULT_UC, &w);
1301
1302 =item C<ev_TYPE_stop> (loop, ev_TYPE *watcher)
1303
1304 Stops the given watcher if active, and clears the pending status (whether
1305 the watcher was active or not).
1306
1307 It is possible that stopped watchers are pending - for example,
1308 non-repeating timers are being stopped when they become pending - but
1309 calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
1310 pending. If you want to free or reuse the memory used by the watcher it is
1311 therefore a good idea to always call its C<ev_TYPE_stop> function.
1312
1313 =item bool ev_is_active (ev_TYPE *watcher)
1314
1315 Returns a true value iff the watcher is active (i.e. it has been started
1316 and not yet been stopped). As long as a watcher is active you must not modify
1317 it.
1318
1319 =item bool ev_is_pending (ev_TYPE *watcher)
1320
1321 Returns a true value iff the watcher is pending, (i.e. it has outstanding
1322 events but its callback has not yet been invoked). As long as a watcher
1323 is pending (but not active) you must not call an init function on it (but
1324 C<ev_TYPE_set> is safe), you must not change its priority, and you must
1325 make sure the watcher is available to libev (e.g. you cannot C<free ()>
1326 it).
1327
1328 =item callback ev_cb (ev_TYPE *watcher)
1329
1330 Returns the callback currently set on the watcher.
1331
1332 =item ev_set_cb (ev_TYPE *watcher, callback)
1333
1334 Change the callback. You can change the callback at virtually any time
1335 (modulo threads).
1336
1337 =item ev_set_priority (ev_TYPE *watcher, int priority)
1338
1339 =item int ev_priority (ev_TYPE *watcher)
1340
1341 Set and query the priority of the watcher. The priority is a small
1342 integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
1343 (default: C<-2>). Pending watchers with higher priority will be invoked
1344 before watchers with lower priority, but priority will not keep watchers
1345 from being executed (except for C<ev_idle> watchers).
1346
1347 If you need to suppress invocation when higher priority events are pending
1348 you need to look at C<ev_idle> watchers, which provide this functionality.
1349
1350 You I<must not> change the priority of a watcher as long as it is active or
1351 pending.
1352
1353 Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
1354 fine, as long as you do not mind that the priority value you query might
1355 or might not have been clamped to the valid range.
1356
1357 The default priority used by watchers when no priority has been set is
1358 always C<0>, which is supposed to not be too high and not be too low :).
1359
1360 See L</WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
1361 priorities.
1362
1363 =item ev_invoke (loop, ev_TYPE *watcher, int revents)
1364
1365 Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
1366 C<loop> nor C<revents> need to be valid as long as the watcher callback
1367 can deal with that fact, as both are simply passed through to the
1368 callback.
1369
1370 =item int ev_clear_pending (loop, ev_TYPE *watcher)
1371
1372 If the watcher is pending, this function clears its pending status and
1373 returns its C<revents> bitset (as if its callback was invoked). If the
1374 watcher isn't pending it does nothing and returns C<0>.
1375
1376 Sometimes it can be useful to "poll" a watcher instead of waiting for its
1377 callback to be invoked, which can be accomplished with this function.
1378
1379 =item ev_feed_event (loop, ev_TYPE *watcher, int revents)
1380
1381 Feeds the given event set into the event loop, as if the specified event
1382 had happened for the specified watcher (which must be a pointer to an
1383 initialised but not necessarily started event watcher). Obviously you must
1384 not free the watcher as long as it has pending events.
1385
1386 Stopping the watcher, letting libev invoke it, or calling
1387 C<ev_clear_pending> will clear the pending event, even if the watcher was
1388 not started in the first place.
1389
1390 See also C<ev_feed_fd_event> and C<ev_feed_signal_event> for related
1391 functions that do not need a watcher.
1392
1393 =back
1394
1395 See also the L</ASSOCIATING CUSTOM DATA WITH A WATCHER> and L</BUILDING YOUR
1396 OWN COMPOSITE WATCHERS> idioms.
1397
1398 =head2 WATCHER STATES
1399
1400 There are various watcher states mentioned throughout this manual -
1401 active, pending and so on. In this section these states and the rules to
1402 transition between them will be described in more detail - and while these
1403 rules might look complicated, they usually do "the right thing".
1404
1405 =over 4
1406
1407 =item initialised
1408
1409 Before a watcher can be registered with the event loop it has to be
1410 initialised. This can be done with a call to C<ev_TYPE_init>, or calls to
1411 C<ev_init> followed by the watcher-specific C<ev_TYPE_set> function.
1412
1413 In this state it is simply some block of memory that is suitable for
1414 use in an event loop. It can be moved around, freed, reused etc. at
1415 will - as long as you either keep the memory contents intact, or call
1416 C<ev_TYPE_init> again.
1417
1418 =item started/running/active
1419
1420 Once a watcher has been started with a call to C<ev_TYPE_start> it becomes
1421 property of the event loop, and is actively waiting for events. While in
1422 this state it cannot be accessed (except in a few documented ways), moved,
1423 freed or anything else - the only legal thing is to keep a pointer to it,
1424 and call libev functions on it that are documented to work on active watchers.
1425
1426 =item pending
1427
1428 If a watcher is active and libev determines that an event it is interested
1429 in has occurred (such as a timer expiring), it will become pending. It will
1430 stay in this pending state until either it is stopped or its callback is
1431 about to be invoked, so it is not normally pending inside the watcher
1432 callback.
1433
1434 The watcher might or might not be active while it is pending (for example,
1435 an expired non-repeating timer can be pending but no longer active). If it
1436 is stopped, it can be freely accessed (e.g. by calling C<ev_TYPE_set>),
1437 but it is still property of the event loop at this time, so cannot be
1438 moved, freed or reused. And if it is active the rules described in the
1439 previous item still apply.
1440
1441 It is also possible to feed an event on a watcher that is not active (e.g.
1442 via C<ev_feed_event>), in which case it becomes pending without being
1443 active.
1444
1445 =item stopped
1446
1447 A watcher can be stopped implicitly by libev (in which case it might still
1448 be pending), or explicitly by calling its C<ev_TYPE_stop> function. The
1449 latter will clear any pending state the watcher might be in, regardless
1450 of whether it was active or not, so stopping a watcher explicitly before
1451 freeing it is often a good idea.
1452
1453 While stopped (and not pending) the watcher is essentially in the
1454 initialised state, that is, it can be reused, moved, modified in any way
1455 you wish (but when you trash the memory block, you need to C<ev_TYPE_init>
1456 it again).
1457
1458 =back
1459
1460 =head2 WATCHER PRIORITY MODELS
1461
1462 Many event loops support I<watcher priorities>, which are usually small
1463 integers that influence the ordering of event callback invocation
1464 between watchers in some way, all else being equal.
1465
1466 In libev, Watcher priorities can be set using C<ev_set_priority>. See its
1467 description for the more technical details such as the actual priority
1468 range.
1469
1470 There are two common ways how these these priorities are being interpreted
1471 by event loops:
1472
1473 In the more common lock-out model, higher priorities "lock out" invocation
1474 of lower priority watchers, which means as long as higher priority
1475 watchers receive events, lower priority watchers are not being invoked.
1476
1477 The less common only-for-ordering model uses priorities solely to order
1478 callback invocation within a single event loop iteration: Higher priority
1479 watchers are invoked before lower priority ones, but they all get invoked
1480 before polling for new events.
1481
1482 Libev uses the second (only-for-ordering) model for all its watchers
1483 except for idle watchers (which use the lock-out model).
1484
1485 The rationale behind this is that implementing the lock-out model for
1486 watchers is not well supported by most kernel interfaces, and most event
1487 libraries will just poll for the same events again and again as long as
1488 their callbacks have not been executed, which is very inefficient in the
1489 common case of one high-priority watcher locking out a mass of lower
1490 priority ones.
1491
1492 Static (ordering) priorities are most useful when you have two or more
1493 watchers handling the same resource: a typical usage example is having an
1494 C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
1495 timeouts. Under load, data might be received while the program handles
1496 other jobs, but since timers normally get invoked first, the timeout
1497 handler will be executed before checking for data. In that case, giving
1498 the timer a lower priority than the I/O watcher ensures that I/O will be
1499 handled first even under adverse conditions (which is usually, but not
1500 always, what you want).
1501
1502 Since idle watchers use the "lock-out" model, meaning that idle watchers
1503 will only be executed when no same or higher priority watchers have
1504 received events, they can be used to implement the "lock-out" model when
1505 required.
1506
1507 For example, to emulate how many other event libraries handle priorities,
1508 you can associate an C<ev_idle> watcher to each such watcher, and in
1509 the normal watcher callback, you just start the idle watcher. The real
1510 processing is done in the idle watcher callback. This causes libev to
1511 continuously poll and process kernel event data for the watcher, but when
1512 the lock-out case is known to be rare (which in turn is rare :), this is
1513 workable.
1514
1515 Usually, however, the lock-out model implemented that way will perform
1516 miserably under the type of load it was designed to handle. In that case,
1517 it might be preferable to stop the real watcher before starting the
1518 idle watcher, so the kernel will not have to process the event in case
1519 the actual processing will be delayed for considerable time.
1520
1521 Here is an example of an I/O watcher that should run at a strictly lower
1522 priority than the default, and which should only process data when no
1523 other events are pending:
1524
1525 ev_idle idle; // actual processing watcher
1526 ev_io io; // actual event watcher
1527
1528 static void
1529 io_cb (EV_P_ ev_io *w, int revents)
1530 {
1531 // stop the I/O watcher, we received the event, but
1532 // are not yet ready to handle it.
1533 ev_io_stop (EV_A_ w);
1534
1535 // start the idle watcher to handle the actual event.
1536 // it will not be executed as long as other watchers
1537 // with the default priority are receiving events.
1538 ev_idle_start (EV_A_ &idle);
1539 }
1540
1541 static void
1542 idle_cb (EV_P_ ev_idle *w, int revents)
1543 {
1544 // actual processing
1545 read (STDIN_FILENO, ...);
1546
1547 // have to start the I/O watcher again, as
1548 // we have handled the event
1549 ev_io_start (EV_P_ &io);
1550 }
1551
1552 // initialisation
1553 ev_idle_init (&idle, idle_cb);
1554 ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
1555 ev_io_start (EV_DEFAULT_ &io);
1556
1557 In the "real" world, it might also be beneficial to start a timer, so that
1558 low-priority connections can not be locked out forever under load. This
1559 enables your program to keep a lower latency for important connections
1560 during short periods of high load, while not completely locking out less
1561 important ones.
1562
1563
1564 =head1 WATCHER TYPES
1565
1566 This section describes each watcher in detail, but will not repeat
1567 information given in the last section. Any initialisation/set macros,
1568 functions and members specific to the watcher type are explained.
1569
1570 Members are additionally marked with either I<[read-only]>, meaning that,
1571 while the watcher is active, you can look at the member and expect some
1572 sensible content, but you must not modify it (you can modify it while the
1573 watcher is stopped to your hearts content), or I<[read-write]>, which
1574 means you can expect it to have some sensible content while the watcher
1575 is active, but you can also modify it. Modifying it may not do something
1576 sensible or take immediate effect (or do anything at all), but libev will
1577 not crash or malfunction in any way.
1578
1579
1580 =head2 C<ev_io> - is this file descriptor readable or writable?
1581
1582 I/O watchers check whether a file descriptor is readable or writable
1583 in each iteration of the event loop, or, more precisely, when reading
1584 would not block the process and writing would at least be able to write
1585 some data. This behaviour is called level-triggering because you keep
1586 receiving events as long as the condition persists. Remember you can stop
1587 the watcher if you don't want to act on the event and neither want to
1588 receive future events.
1589
1590 In general you can register as many read and/or write event watchers per
1591 fd as you want (as long as you don't confuse yourself). Setting all file
1592 descriptors to non-blocking mode is also usually a good idea (but not
1593 required if you know what you are doing).
1594
1595 Another thing you have to watch out for is that it is quite easy to
1596 receive "spurious" readiness notifications, that is, your callback might
1597 be called with C<EV_READ> but a subsequent C<read>(2) will actually block
1598 because there is no data. It is very easy to get into this situation even
1599 with a relatively standard program structure. Thus it is best to always
1600 use non-blocking I/O: An extra C<read>(2) returning C<EAGAIN> is far
1601 preferable to a program hanging until some data arrives.
1602
1603 If you cannot run the fd in non-blocking mode (for example you should
1604 not play around with an Xlib connection), then you have to separately
1605 re-test whether a file descriptor is really ready with a known-to-be good
1606 interface such as poll (fortunately in the case of Xlib, it already does
1607 this on its own, so its quite safe to use). Some people additionally
1608 use C<SIGALRM> and an interval timer, just to be sure you won't block
1609 indefinitely.
1610
1611 But really, best use non-blocking mode.
1612
1613 =head3 The special problem of disappearing file descriptors
1614
1615 Some backends (e.g. kqueue, epoll) need to be told about closing a file
1616 descriptor (either due to calling C<close> explicitly or any other means,
1617 such as C<dup2>). The reason is that you register interest in some file
1618 descriptor, but when it goes away, the operating system will silently drop
1619 this interest. If another file descriptor with the same number then is
1620 registered with libev, there is no efficient way to see that this is, in
1621 fact, a different file descriptor.
1622
1623 To avoid having to explicitly tell libev about such cases, libev follows
1624 the following policy: Each time C<ev_io_set> is being called, libev
1625 will assume that this is potentially a new file descriptor, otherwise
1626 it is assumed that the file descriptor stays the same. That means that
1627 you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
1628 descriptor even if the file descriptor number itself did not change.
1629
1630 This is how one would do it normally anyway, the important point is that
1631 the libev application should not optimise around libev but should leave
1632 optimisations to libev.
1633
1634 =head3 The special problem of dup'ed file descriptors
1635
1636 Some backends (e.g. epoll), cannot register events for file descriptors,
1637 but only events for the underlying file descriptions. That means when you
1638 have C<dup ()>'ed file descriptors or weirder constellations, and register
1639 events for them, only one file descriptor might actually receive events.
1640
1641 There is no workaround possible except not registering events
1642 for potentially C<dup ()>'ed file descriptors, or to resort to
1643 C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
1644
1645 =head3 The special problem of files
1646
1647 Many people try to use C<select> (or libev) on file descriptors
1648 representing files, and expect it to become ready when their program
1649 doesn't block on disk accesses (which can take a long time on their own).
1650
1651 However, this cannot ever work in the "expected" way - you get a readiness
1652 notification as soon as the kernel knows whether and how much data is
1653 there, and in the case of open files, that's always the case, so you
1654 always get a readiness notification instantly, and your read (or possibly
1655 write) will still block on the disk I/O.
1656
1657 Another way to view it is that in the case of sockets, pipes, character
1658 devices and so on, there is another party (the sender) that delivers data
1659 on its own, but in the case of files, there is no such thing: the disk
1660 will not send data on its own, simply because it doesn't know what you
1661 wish to read - you would first have to request some data.
1662
1663 Since files are typically not-so-well supported by advanced notification
1664 mechanism, libev tries hard to emulate POSIX behaviour with respect
1665 to files, even though you should not use it. The reason for this is
1666 convenience: sometimes you want to watch STDIN or STDOUT, which is
1667 usually a tty, often a pipe, but also sometimes files or special devices
1668 (for example, C<epoll> on Linux works with F</dev/random> but not with
1669 F</dev/urandom>), and even though the file might better be served with
1670 asynchronous I/O instead of with non-blocking I/O, it is still useful when
1671 it "just works" instead of freezing.
1672
1673 So avoid file descriptors pointing to files when you know it (e.g. use
1674 libeio), but use them when it is convenient, e.g. for STDIN/STDOUT, or
1675 when you rarely read from a file instead of from a socket, and want to
1676 reuse the same code path.
1677
1678 =head3 The special problem of fork
1679
1680 Some backends (epoll, kqueue) do not support C<fork ()> at all or exhibit
1681 useless behaviour. Libev fully supports fork, but needs to be told about
1682 it in the child if you want to continue to use it in the child.
1683
1684 To support fork in your child processes, you have to call C<ev_loop_fork
1685 ()> after a fork in the child, enable C<EVFLAG_FORKCHECK>, or resort to
1686 C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
1687
1688 =head3 The special problem of SIGPIPE
1689
1690 While not really specific to libev, it is easy to forget about C<SIGPIPE>:
1691 when writing to a pipe whose other end has been closed, your program gets
1692 sent a SIGPIPE, which, by default, aborts your program. For most programs
1693 this is sensible behaviour, for daemons, this is usually undesirable.
1694
1695 So when you encounter spurious, unexplained daemon exits, make sure you
1696 ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
1697 somewhere, as that would have given you a big clue).
1698
1699 =head3 The special problem of accept()ing when you can't
1700
1701 Many implementations of the POSIX C<accept> function (for example,
1702 found in post-2004 Linux) have the peculiar behaviour of not removing a
1703 connection from the pending queue in all error cases.
1704
1705 For example, larger servers often run out of file descriptors (because
1706 of resource limits), causing C<accept> to fail with C<ENFILE> but not
1707 rejecting the connection, leading to libev signalling readiness on
1708 the next iteration again (the connection still exists after all), and
1709 typically causing the program to loop at 100% CPU usage.
1710
1711 Unfortunately, the set of errors that cause this issue differs between
1712 operating systems, there is usually little the app can do to remedy the
1713 situation, and no known thread-safe method of removing the connection to
1714 cope with overload is known (to me).
1715
1716 One of the easiest ways to handle this situation is to just ignore it
1717 - when the program encounters an overload, it will just loop until the
1718 situation is over. While this is a form of busy waiting, no OS offers an
1719 event-based way to handle this situation, so it's the best one can do.
1720
1721 A better way to handle the situation is to log any errors other than
1722 C<EAGAIN> and C<EWOULDBLOCK>, making sure not to flood the log with such
1723 messages, and continue as usual, which at least gives the user an idea of
1724 what could be wrong ("raise the ulimit!"). For extra points one could stop
1725 the C<ev_io> watcher on the listening fd "for a while", which reduces CPU
1726 usage.
1727
1728 If your program is single-threaded, then you could also keep a dummy file
1729 descriptor for overload situations (e.g. by opening F</dev/null>), and
1730 when you run into C<ENFILE> or C<EMFILE>, close it, run C<accept>,
1731 close that fd, and create a new dummy fd. This will gracefully refuse
1732 clients under typical overload conditions.
1733
1734 The last way to handle it is to simply log the error and C<exit>, as
1735 is often done with C<malloc> failures, but this results in an easy
1736 opportunity for a DoS attack.
1737
1738 =head3 Watcher-Specific Functions
1739
1740 =over 4
1741
1742 =item ev_io_init (ev_io *, callback, int fd, int events)
1743
1744 =item ev_io_set (ev_io *, int fd, int events)
1745
1746 Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
1747 receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
1748 C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
1749
1750 =item int fd [read-only]
1751
1752 The file descriptor being watched.
1753
1754 =item int events [read-only]
1755
1756 The events being watched.
1757
1758 =back
1759
1760 =head3 Examples
1761
1762 Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
1763 readable, but only once. Since it is likely line-buffered, you could
1764 attempt to read a whole line in the callback.
1765
1766 static void
1767 stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
1768 {
1769 ev_io_stop (loop, w);
1770 .. read from stdin here (or from w->fd) and handle any I/O errors
1771 }
1772
1773 ...
1774 struct ev_loop *loop = ev_default_init (0);
1775 ev_io stdin_readable;
1776 ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
1777 ev_io_start (loop, &stdin_readable);
1778 ev_run (loop, 0);
1779
1780
1781 =head2 C<ev_timer> - relative and optionally repeating timeouts
1782
1783 Timer watchers are simple relative timers that generate an event after a
1784 given time, and optionally repeating in regular intervals after that.
1785
1786 The timers are based on real time, that is, if you register an event that
1787 times out after an hour and you reset your system clock to January last
1788 year, it will still time out after (roughly) one hour. "Roughly" because
1789 detecting time jumps is hard, and some inaccuracies are unavoidable (the
1790 monotonic clock option helps a lot here).
1791
1792 The callback is guaranteed to be invoked only I<after> its timeout has
1793 passed (not I<at>, so on systems with very low-resolution clocks this
1794 might introduce a small delay, see "the special problem of being too
1795 early", below). If multiple timers become ready during the same loop
1796 iteration then the ones with earlier time-out values are invoked before
1797 ones of the same priority with later time-out values (but this is no
1798 longer true when a callback calls C<ev_run> recursively).
1799
1800 =head3 Be smart about timeouts
1801
1802 Many real-world problems involve some kind of timeout, usually for error
1803 recovery. A typical example is an HTTP request - if the other side hangs,
1804 you want to raise some error after a while.
1805
1806 What follows are some ways to handle this problem, from obvious and
1807 inefficient to smart and efficient.
1808
1809 In the following, a 60 second activity timeout is assumed - a timeout that
1810 gets reset to 60 seconds each time there is activity (e.g. each time some
1811 data or other life sign was received).
1812
1813 =over 4
1814
1815 =item 1. Use a timer and stop, reinitialise and start it on activity.
1816
1817 This is the most obvious, but not the most simple way: In the beginning,
1818 start the watcher:
1819
1820 ev_timer_init (timer, callback, 60., 0.);
1821 ev_timer_start (loop, timer);
1822
1823 Then, each time there is some activity, C<ev_timer_stop> it, initialise it
1824 and start it again:
1825
1826 ev_timer_stop (loop, timer);
1827 ev_timer_set (timer, 60., 0.);
1828 ev_timer_start (loop, timer);
1829
1830 This is relatively simple to implement, but means that each time there is
1831 some activity, libev will first have to remove the timer from its internal
1832 data structure and then add it again. Libev tries to be fast, but it's
1833 still not a constant-time operation.
1834
1835 =item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
1836
1837 This is the easiest way, and involves using C<ev_timer_again> instead of
1838 C<ev_timer_start>.
1839
1840 To implement this, configure an C<ev_timer> with a C<repeat> value
1841 of C<60> and then call C<ev_timer_again> at start and each time you
1842 successfully read or write some data. If you go into an idle state where
1843 you do not expect data to travel on the socket, you can C<ev_timer_stop>
1844 the timer, and C<ev_timer_again> will automatically restart it if need be.
1845
1846 That means you can ignore both the C<ev_timer_start> function and the
1847 C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
1848 member and C<ev_timer_again>.
1849
1850 At start:
1851
1852 ev_init (timer, callback);
1853 timer->repeat = 60.;
1854 ev_timer_again (loop, timer);
1855
1856 Each time there is some activity:
1857
1858 ev_timer_again (loop, timer);
1859
1860 It is even possible to change the time-out on the fly, regardless of
1861 whether the watcher is active or not:
1862
1863 timer->repeat = 30.;
1864 ev_timer_again (loop, timer);
1865
1866 This is slightly more efficient then stopping/starting the timer each time
1867 you want to modify its timeout value, as libev does not have to completely
1868 remove and re-insert the timer from/into its internal data structure.
1869
1870 It is, however, even simpler than the "obvious" way to do it.
1871
1872 =item 3. Let the timer time out, but then re-arm it as required.
1873
1874 This method is more tricky, but usually most efficient: Most timeouts are
1875 relatively long compared to the intervals between other activity - in
1876 our example, within 60 seconds, there are usually many I/O events with
1877 associated activity resets.
1878
1879 In this case, it would be more efficient to leave the C<ev_timer> alone,
1880 but remember the time of last activity, and check for a real timeout only
1881 within the callback:
1882
1883 ev_tstamp timeout = 60.;
1884 ev_tstamp last_activity; // time of last activity
1885 ev_timer timer;
1886
1887 static void
1888 callback (EV_P_ ev_timer *w, int revents)
1889 {
1890 // calculate when the timeout would happen
1891 ev_tstamp after = last_activity - ev_now (EV_A) + timeout;
1892
1893 // if negative, it means we the timeout already occurred
1894 if (after < 0.)
1895 {
1896 // timeout occurred, take action
1897 }
1898 else
1899 {
1900 // callback was invoked, but there was some recent
1901 // activity. simply restart the timer to time out
1902 // after "after" seconds, which is the earliest time
1903 // the timeout can occur.
1904 ev_timer_set (w, after, 0.);
1905 ev_timer_start (EV_A_ w);
1906 }
1907 }
1908
1909 To summarise the callback: first calculate in how many seconds the
1910 timeout will occur (by calculating the absolute time when it would occur,
1911 C<last_activity + timeout>, and subtracting the current time, C<ev_now
1912 (EV_A)> from that).
1913
1914 If this value is negative, then we are already past the timeout, i.e. we
1915 timed out, and need to do whatever is needed in this case.
1916
1917 Otherwise, we now the earliest time at which the timeout would trigger,
1918 and simply start the timer with this timeout value.
1919
1920 In other words, each time the callback is invoked it will check whether
1921 the timeout occurred. If not, it will simply reschedule itself to check
1922 again at the earliest time it could time out. Rinse. Repeat.
1923
1924 This scheme causes more callback invocations (about one every 60 seconds
1925 minus half the average time between activity), but virtually no calls to
1926 libev to change the timeout.
1927
1928 To start the machinery, simply initialise the watcher and set
1929 C<last_activity> to the current time (meaning there was some activity just
1930 now), then call the callback, which will "do the right thing" and start
1931 the timer:
1932
1933 last_activity = ev_now (EV_A);
1934 ev_init (&timer, callback);
1935 callback (EV_A_ &timer, 0);
1936
1937 When there is some activity, simply store the current time in
1938 C<last_activity>, no libev calls at all:
1939
1940 if (activity detected)
1941 last_activity = ev_now (EV_A);
1942
1943 When your timeout value changes, then the timeout can be changed by simply
1944 providing a new value, stopping the timer and calling the callback, which
1945 will again do the right thing (for example, time out immediately :).
1946
1947 timeout = new_value;
1948 ev_timer_stop (EV_A_ &timer);
1949 callback (EV_A_ &timer, 0);
1950
1951 This technique is slightly more complex, but in most cases where the
1952 time-out is unlikely to be triggered, much more efficient.
1953
1954 =item 4. Wee, just use a double-linked list for your timeouts.
1955
1956 If there is not one request, but many thousands (millions...), all
1957 employing some kind of timeout with the same timeout value, then one can
1958 do even better:
1959
1960 When starting the timeout, calculate the timeout value and put the timeout
1961 at the I<end> of the list.
1962
1963 Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
1964 the list is expected to fire (for example, using the technique #3).
1965
1966 When there is some activity, remove the timer from the list, recalculate
1967 the timeout, append it to the end of the list again, and make sure to
1968 update the C<ev_timer> if it was taken from the beginning of the list.
1969
1970 This way, one can manage an unlimited number of timeouts in O(1) time for
1971 starting, stopping and updating the timers, at the expense of a major
1972 complication, and having to use a constant timeout. The constant timeout
1973 ensures that the list stays sorted.
1974
1975 =back
1976
1977 So which method the best?
1978
1979 Method #2 is a simple no-brain-required solution that is adequate in most
1980 situations. Method #3 requires a bit more thinking, but handles many cases
1981 better, and isn't very complicated either. In most case, choosing either
1982 one is fine, with #3 being better in typical situations.
1983
1984 Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
1985 rather complicated, but extremely efficient, something that really pays
1986 off after the first million or so of active timers, i.e. it's usually
1987 overkill :)
1988
1989 =head3 The special problem of being too early
1990
1991 If you ask a timer to call your callback after three seconds, then
1992 you expect it to be invoked after three seconds - but of course, this
1993 cannot be guaranteed to infinite precision. Less obviously, it cannot be
1994 guaranteed to any precision by libev - imagine somebody suspending the
1995 process with a STOP signal for a few hours for example.
1996
1997 So, libev tries to invoke your callback as soon as possible I<after> the
1998 delay has occurred, but cannot guarantee this.
1999
2000 A less obvious failure mode is calling your callback too early: many event
2001 loops compare timestamps with a "elapsed delay >= requested delay", but
2002 this can cause your callback to be invoked much earlier than you would
2003 expect.
2004
2005 To see why, imagine a system with a clock that only offers full second
2006 resolution (think windows if you can't come up with a broken enough OS
2007 yourself). If you schedule a one-second timer at the time 500.9, then the
2008 event loop will schedule your timeout to elapse at a system time of 500
2009 (500.9 truncated to the resolution) + 1, or 501.
2010
2011 If an event library looks at the timeout 0.1s later, it will see "501 >=
2012 501" and invoke the callback 0.1s after it was started, even though a
2013 one-second delay was requested - this is being "too early", despite best
2014 intentions.
2015
2016 This is the reason why libev will never invoke the callback if the elapsed
2017 delay equals the requested delay, but only when the elapsed delay is
2018 larger than the requested delay. In the example above, libev would only invoke
2019 the callback at system time 502, or 1.1s after the timer was started.
2020
2021 So, while libev cannot guarantee that your callback will be invoked
2022 exactly when requested, it I<can> and I<does> guarantee that the requested
2023 delay has actually elapsed, or in other words, it always errs on the "too
2024 late" side of things.
2025
2026 =head3 The special problem of time updates
2027
2028 Establishing the current time is a costly operation (it usually takes
2029 at least one system call): EV therefore updates its idea of the current
2030 time only before and after C<ev_run> collects new events, which causes a
2031 growing difference between C<ev_now ()> and C<ev_time ()> when handling
2032 lots of events in one iteration.
2033
2034 The relative timeouts are calculated relative to the C<ev_now ()>
2035 time. This is usually the right thing as this timestamp refers to the time
2036 of the event triggering whatever timeout you are modifying/starting. If
2037 you suspect event processing to be delayed and you I<need> to base the
2038 timeout on the current time, use something like the following to adjust
2039 for it:
2040
2041 ev_timer_set (&timer, after + (ev_time () - ev_now ()), 0.);
2042
2043 If the event loop is suspended for a long time, you can also force an
2044 update of the time returned by C<ev_now ()> by calling C<ev_now_update
2045 ()>, although that will push the event time of all outstanding events
2046 further into the future.
2047
2048 =head3 The special problem of unsynchronised clocks
2049
2050 Modern systems have a variety of clocks - libev itself uses the normal
2051 "wall clock" clock and, if available, the monotonic clock (to avoid time
2052 jumps).
2053
2054 Neither of these clocks is synchronised with each other or any other clock
2055 on the system, so C<ev_time ()> might return a considerably different time
2056 than C<gettimeofday ()> or C<time ()>. On a GNU/Linux system, for example,
2057 a call to C<gettimeofday> might return a second count that is one higher
2058 than a directly following call to C<time>.
2059
2060 The moral of this is to only compare libev-related timestamps with
2061 C<ev_time ()> and C<ev_now ()>, at least if you want better precision than
2062 a second or so.
2063
2064 One more problem arises due to this lack of synchronisation: if libev uses
2065 the system monotonic clock and you compare timestamps from C<ev_time>
2066 or C<ev_now> from when you started your timer and when your callback is
2067 invoked, you will find that sometimes the callback is a bit "early".
2068
2069 This is because C<ev_timer>s work in real time, not wall clock time, so
2070 libev makes sure your callback is not invoked before the delay happened,
2071 I<measured according to the real time>, not the system clock.
2072
2073 If your timeouts are based on a physical timescale (e.g. "time out this
2074 connection after 100 seconds") then this shouldn't bother you as it is
2075 exactly the right behaviour.
2076
2077 If you want to compare wall clock/system timestamps to your timers, then
2078 you need to use C<ev_periodic>s, as these are based on the wall clock
2079 time, where your comparisons will always generate correct results.
2080
2081 =head3 The special problems of suspended animation
2082
2083 When you leave the server world it is quite customary to hit machines that
2084 can suspend/hibernate - what happens to the clocks during such a suspend?
2085
2086 Some quick tests made with a Linux 2.6.28 indicate that a suspend freezes
2087 all processes, while the clocks (C<times>, C<CLOCK_MONOTONIC>) continue
2088 to run until the system is suspended, but they will not advance while the
2089 system is suspended. That means, on resume, it will be as if the program
2090 was frozen for a few seconds, but the suspend time will not be counted
2091 towards C<ev_timer> when a monotonic clock source is used. The real time
2092 clock advanced as expected, but if it is used as sole clocksource, then a
2093 long suspend would be detected as a time jump by libev, and timers would
2094 be adjusted accordingly.
2095
2096 I would not be surprised to see different behaviour in different between
2097 operating systems, OS versions or even different hardware.
2098
2099 The other form of suspend (job control, or sending a SIGSTOP) will see a
2100 time jump in the monotonic clocks and the realtime clock. If the program
2101 is suspended for a very long time, and monotonic clock sources are in use,
2102 then you can expect C<ev_timer>s to expire as the full suspension time
2103 will be counted towards the timers. When no monotonic clock source is in
2104 use, then libev will again assume a timejump and adjust accordingly.
2105
2106 It might be beneficial for this latter case to call C<ev_suspend>
2107 and C<ev_resume> in code that handles C<SIGTSTP>, to at least get
2108 deterministic behaviour in this case (you can do nothing against
2109 C<SIGSTOP>).
2110
2111 =head3 Watcher-Specific Functions and Data Members
2112
2113 =over 4
2114
2115 =item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
2116
2117 =item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
2118
2119 Configure the timer to trigger after C<after> seconds (fractional and
2120 negative values are supported). If C<repeat> is C<0.>, then it will
2121 automatically be stopped once the timeout is reached. If it is positive,
2122 then the timer will automatically be configured to trigger again C<repeat>
2123 seconds later, again, and again, until stopped manually.
2124
2125 The timer itself will do a best-effort at avoiding drift, that is, if
2126 you configure a timer to trigger every 10 seconds, then it will normally
2127 trigger at exactly 10 second intervals. If, however, your program cannot
2128 keep up with the timer (because it takes longer than those 10 seconds to
2129 do stuff) the timer will not fire more than once per event loop iteration.
2130
2131 =item ev_timer_again (loop, ev_timer *)
2132
2133 This will act as if the timer timed out, and restarts it again if it is
2134 repeating. It basically works like calling C<ev_timer_stop>, updating the
2135 timeout to the C<repeat> value and calling C<ev_timer_start>.
2136
2137 The exact semantics are as in the following rules, all of which will be
2138 applied to the watcher:
2139
2140 =over 4
2141
2142 =item If the timer is pending, the pending status is always cleared.
2143
2144 =item If the timer is started but non-repeating, stop it (as if it timed
2145 out, without invoking it).
2146
2147 =item If the timer is repeating, make the C<repeat> value the new timeout
2148 and start the timer, if necessary.
2149
2150 =back
2151
2152 This sounds a bit complicated, see L</Be smart about timeouts>, above, for a
2153 usage example.
2154
2155 =item ev_tstamp ev_timer_remaining (loop, ev_timer *)
2156
2157 Returns the remaining time until a timer fires. If the timer is active,
2158 then this time is relative to the current event loop time, otherwise it's
2159 the timeout value currently configured.
2160
2161 That is, after an C<ev_timer_set (w, 5, 7)>, C<ev_timer_remaining> returns
2162 C<5>. When the timer is started and one second passes, C<ev_timer_remaining>
2163 will return C<4>. When the timer expires and is restarted, it will return
2164 roughly C<7> (likely slightly less as callback invocation takes some time,
2165 too), and so on.
2166
2167 =item ev_tstamp repeat [read-write]
2168
2169 The current C<repeat> value. Will be used each time the watcher times out
2170 or C<ev_timer_again> is called, and determines the next timeout (if any),
2171 which is also when any modifications are taken into account.
2172
2173 =back
2174
2175 =head3 Examples
2176
2177 Example: Create a timer that fires after 60 seconds.
2178
2179 static void
2180 one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
2181 {
2182 .. one minute over, w is actually stopped right here
2183 }
2184
2185 ev_timer mytimer;
2186 ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
2187 ev_timer_start (loop, &mytimer);
2188
2189 Example: Create a timeout timer that times out after 10 seconds of
2190 inactivity.
2191
2192 static void
2193 timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
2194 {
2195 .. ten seconds without any activity
2196 }
2197
2198 ev_timer mytimer;
2199 ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
2200 ev_timer_again (&mytimer); /* start timer */
2201 ev_run (loop, 0);
2202
2203 // and in some piece of code that gets executed on any "activity":
2204 // reset the timeout to start ticking again at 10 seconds
2205 ev_timer_again (&mytimer);
2206
2207
2208 =head2 C<ev_periodic> - to cron or not to cron?
2209
2210 Periodic watchers are also timers of a kind, but they are very versatile
2211 (and unfortunately a bit complex).
2212
2213 Unlike C<ev_timer>, periodic watchers are not based on real time (or
2214 relative time, the physical time that passes) but on wall clock time
2215 (absolute time, the thing you can read on your calendar or clock). The
2216 difference is that wall clock time can run faster or slower than real
2217 time, and time jumps are not uncommon (e.g. when you adjust your
2218 wrist-watch).
2219
2220 You can tell a periodic watcher to trigger after some specific point
2221 in time: for example, if you tell a periodic watcher to trigger "in 10
2222 seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
2223 not a delay) and then reset your system clock to January of the previous
2224 year, then it will take a year or more to trigger the event (unlike an
2225 C<ev_timer>, which would still trigger roughly 10 seconds after starting
2226 it, as it uses a relative timeout).
2227
2228 C<ev_periodic> watchers can also be used to implement vastly more complex
2229 timers, such as triggering an event on each "midnight, local time", or
2230 other complicated rules. This cannot easily be done with C<ev_timer>
2231 watchers, as those cannot react to time jumps.
2232
2233 As with timers, the callback is guaranteed to be invoked only when the
2234 point in time where it is supposed to trigger has passed. If multiple
2235 timers become ready during the same loop iteration then the ones with
2236 earlier time-out values are invoked before ones with later time-out values
2237 (but this is no longer true when a callback calls C<ev_run> recursively).
2238
2239 =head3 Watcher-Specific Functions and Data Members
2240
2241 =over 4
2242
2243 =item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
2244
2245 =item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
2246
2247 Lots of arguments, let's sort it out... There are basically three modes of
2248 operation, and we will explain them from simplest to most complex:
2249
2250 =over 4
2251
2252 =item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
2253
2254 In this configuration the watcher triggers an event after the wall clock
2255 time C<offset> has passed. It will not repeat and will not adjust when a
2256 time jump occurs, that is, if it is to be run at January 1st 2011 then it
2257 will be stopped and invoked when the system clock reaches or surpasses
2258 this point in time.
2259
2260 =item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
2261
2262 In this mode the watcher will always be scheduled to time out at the next
2263 C<offset + N * interval> time (for some integer N, which can also be
2264 negative) and then repeat, regardless of any time jumps. The C<offset>
2265 argument is merely an offset into the C<interval> periods.
2266
2267 This can be used to create timers that do not drift with respect to the
2268 system clock, for example, here is an C<ev_periodic> that triggers each
2269 hour, on the hour (with respect to UTC):
2270
2271 ev_periodic_set (&periodic, 0., 3600., 0);
2272
2273 This doesn't mean there will always be 3600 seconds in between triggers,
2274 but only that the callback will be called when the system time shows a
2275 full hour (UTC), or more correctly, when the system time is evenly divisible
2276 by 3600.
2277
2278 Another way to think about it (for the mathematically inclined) is that
2279 C<ev_periodic> will try to run the callback in this mode at the next possible
2280 time where C<time = offset (mod interval)>, regardless of any time jumps.
2281
2282 The C<interval> I<MUST> be positive, and for numerical stability, the
2283 interval value should be higher than C<1/8192> (which is around 100
2284 microseconds) and C<offset> should be higher than C<0> and should have
2285 at most a similar magnitude as the current time (say, within a factor of
2286 ten). Typical values for offset are, in fact, C<0> or something between
2287 C<0> and C<interval>, which is also the recommended range.
2288
2289 Note also that there is an upper limit to how often a timer can fire (CPU
2290 speed for example), so if C<interval> is very small then timing stability
2291 will of course deteriorate. Libev itself tries to be exact to be about one
2292 millisecond (if the OS supports it and the machine is fast enough).
2293
2294 =item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
2295
2296 In this mode the values for C<interval> and C<offset> are both being
2297 ignored. Instead, each time the periodic watcher gets scheduled, the
2298 reschedule callback will be called with the watcher as first, and the
2299 current time as second argument.
2300
2301 NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, ever,
2302 or make ANY other event loop modifications whatsoever, unless explicitly
2303 allowed by documentation here>.
2304
2305 If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
2306 it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
2307 only event loop modification you are allowed to do).
2308
2309 The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
2310 *w, ev_tstamp now)>, e.g.:
2311
2312 static ev_tstamp
2313 my_rescheduler (ev_periodic *w, ev_tstamp now)
2314 {
2315 return now + 60.;
2316 }
2317
2318 It must return the next time to trigger, based on the passed time value
2319 (that is, the lowest time value larger than to the second argument). It
2320 will usually be called just before the callback will be triggered, but
2321 might be called at other times, too.
2322
2323 NOTE: I<< This callback must always return a time that is higher than or
2324 equal to the passed C<now> value >>.
2325
2326 This can be used to create very complex timers, such as a timer that
2327 triggers on "next midnight, local time". To do this, you would calculate
2328 the next midnight after C<now> and return the timestamp value for
2329 this. Here is a (completely untested, no error checking) example on how to
2330 do this:
2331
2332 #include <time.h>
2333
2334 static ev_tstamp
2335 my_rescheduler (ev_periodic *w, ev_tstamp now)
2336 {
2337 time_t tnow = (time_t)now;
2338 struct tm tm;
2339 localtime_r (&tnow, &tm);
2340
2341 tm.tm_sec = tm.tm_min = tm.tm_hour = 0; // midnight current day
2342 ++tm.tm_mday; // midnight next day
2343
2344 return mktime (&tm);
2345 }
2346
2347 Note: this code might run into trouble on days that have more then two
2348 midnights (beginning and end).
2349
2350 =back
2351
2352 =item ev_periodic_again (loop, ev_periodic *)
2353
2354 Simply stops and restarts the periodic watcher again. This is only useful
2355 when you changed some parameters or the reschedule callback would return
2356 a different time than the last time it was called (e.g. in a crond like
2357 program when the crontabs have changed).
2358
2359 =item ev_tstamp ev_periodic_at (ev_periodic *)
2360
2361 When active, returns the absolute time that the watcher is supposed
2362 to trigger next. This is not the same as the C<offset> argument to
2363 C<ev_periodic_set>, but indeed works even in interval and manual
2364 rescheduling modes.
2365
2366 =item ev_tstamp offset [read-write]
2367
2368 When repeating, this contains the offset value, otherwise this is the
2369 absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
2370 although libev might modify this value for better numerical stability).
2371
2372 Can be modified any time, but changes only take effect when the periodic
2373 timer fires or C<ev_periodic_again> is being called.
2374
2375 =item ev_tstamp interval [read-write]
2376
2377 The current interval value. Can be modified any time, but changes only
2378 take effect when the periodic timer fires or C<ev_periodic_again> is being
2379 called.
2380
2381 =item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
2382
2383 The current reschedule callback, or C<0>, if this functionality is
2384 switched off. Can be changed any time, but changes only take effect when
2385 the periodic timer fires or C<ev_periodic_again> is being called.
2386
2387 =back
2388
2389 =head3 Examples
2390
2391 Example: Call a callback every hour, or, more precisely, whenever the
2392 system time is divisible by 3600. The callback invocation times have
2393 potentially a lot of jitter, but good long-term stability.
2394
2395 static void
2396 clock_cb (struct ev_loop *loop, ev_periodic *w, int revents)
2397 {
2398 ... its now a full hour (UTC, or TAI or whatever your clock follows)
2399 }
2400
2401 ev_periodic hourly_tick;
2402 ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
2403 ev_periodic_start (loop, &hourly_tick);
2404
2405 Example: The same as above, but use a reschedule callback to do it:
2406
2407 #include <math.h>
2408
2409 static ev_tstamp
2410 my_scheduler_cb (ev_periodic *w, ev_tstamp now)
2411 {
2412 return now + (3600. - fmod (now, 3600.));
2413 }
2414
2415 ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
2416
2417 Example: Call a callback every hour, starting now:
2418
2419 ev_periodic hourly_tick;
2420 ev_periodic_init (&hourly_tick, clock_cb,
2421 fmod (ev_now (loop), 3600.), 3600., 0);
2422 ev_periodic_start (loop, &hourly_tick);
2423
2424
2425 =head2 C<ev_signal> - signal me when a signal gets signalled!
2426
2427 Signal watchers will trigger an event when the process receives a specific
2428 signal one or more times. Even though signals are very asynchronous, libev
2429 will try its best to deliver signals synchronously, i.e. as part of the
2430 normal event processing, like any other event.
2431
2432 If you want signals to be delivered truly asynchronously, just use
2433 C<sigaction> as you would do without libev and forget about sharing
2434 the signal. You can even use C<ev_async> from a signal handler to
2435 synchronously wake up an event loop.
2436
2437 You can configure as many watchers as you like for the same signal, but
2438 only within the same loop, i.e. you can watch for C<SIGINT> in your
2439 default loop and for C<SIGIO> in another loop, but you cannot watch for
2440 C<SIGINT> in both the default loop and another loop at the same time. At
2441 the moment, C<SIGCHLD> is permanently tied to the default loop.
2442
2443 Only after the first watcher for a signal is started will libev actually
2444 register something with the kernel. It thus coexists with your own signal
2445 handlers as long as you don't register any with libev for the same signal.
2446
2447 If possible and supported, libev will install its handlers with
2448 C<SA_RESTART> (or equivalent) behaviour enabled, so system calls should
2449 not be unduly interrupted. If you have a problem with system calls getting
2450 interrupted by signals you can block all signals in an C<ev_check> watcher
2451 and unblock them in an C<ev_prepare> watcher.
2452
2453 =head3 The special problem of inheritance over fork/execve/pthread_create
2454
2455 Both the signal mask (C<sigprocmask>) and the signal disposition
2456 (C<sigaction>) are unspecified after starting a signal watcher (and after
2457 stopping it again), that is, libev might or might not block the signal,
2458 and might or might not set or restore the installed signal handler (but
2459 see C<EVFLAG_NOSIGMASK>).
2460
2461 While this does not matter for the signal disposition (libev never
2462 sets signals to C<SIG_IGN>, so handlers will be reset to C<SIG_DFL> on
2463 C<execve>), this matters for the signal mask: many programs do not expect
2464 certain signals to be blocked.
2465
2466 This means that before calling C<exec> (from the child) you should reset
2467 the signal mask to whatever "default" you expect (all clear is a good
2468 choice usually).
2469
2470 The simplest way to ensure that the signal mask is reset in the child is
2471 to install a fork handler with C<pthread_atfork> that resets it. That will
2472 catch fork calls done by libraries (such as the libc) as well.
2473
2474 In current versions of libev, the signal will not be blocked indefinitely
2475 unless you use the C<signalfd> API (C<EV_SIGNALFD>). While this reduces
2476 the window of opportunity for problems, it will not go away, as libev
2477 I<has> to modify the signal mask, at least temporarily.
2478
2479 So I can't stress this enough: I<If you do not reset your signal mask when
2480 you expect it to be empty, you have a race condition in your code>. This
2481 is not a libev-specific thing, this is true for most event libraries.
2482
2483 =head3 The special problem of threads signal handling
2484
2485 POSIX threads has problematic signal handling semantics, specifically,
2486 a lot of functionality (sigfd, sigwait etc.) only really works if all
2487 threads in a process block signals, which is hard to achieve.
2488
2489 When you want to use sigwait (or mix libev signal handling with your own
2490 for the same signals), you can tackle this problem by globally blocking
2491 all signals before creating any threads (or creating them with a fully set
2492 sigprocmask) and also specifying the C<EVFLAG_NOSIGMASK> when creating
2493 loops. Then designate one thread as "signal receiver thread" which handles
2494 these signals. You can pass on any signals that libev might be interested
2495 in by calling C<ev_feed_signal>.
2496
2497 =head3 Watcher-Specific Functions and Data Members
2498
2499 =over 4
2500
2501 =item ev_signal_init (ev_signal *, callback, int signum)
2502
2503 =item ev_signal_set (ev_signal *, int signum)
2504
2505 Configures the watcher to trigger on the given signal number (usually one
2506 of the C<SIGxxx> constants).
2507
2508 =item int signum [read-only]
2509
2510 The signal the watcher watches out for.
2511
2512 =back
2513
2514 =head3 Examples
2515
2516 Example: Try to exit cleanly on SIGINT.
2517
2518 static void
2519 sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
2520 {
2521 ev_break (loop, EVBREAK_ALL);
2522 }
2523
2524 ev_signal signal_watcher;
2525 ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
2526 ev_signal_start (loop, &signal_watcher);
2527
2528
2529 =head2 C<ev_child> - watch out for process status changes
2530
2531 Child watchers trigger when your process receives a SIGCHLD in response to
2532 some child status changes (most typically when a child of yours dies or
2533 exits). It is permissible to install a child watcher I<after> the child
2534 has been forked (which implies it might have already exited), as long
2535 as the event loop isn't entered (or is continued from a watcher), i.e.,
2536 forking and then immediately registering a watcher for the child is fine,
2537 but forking and registering a watcher a few event loop iterations later or
2538 in the next callback invocation is not.
2539
2540 Only the default event loop is capable of handling signals, and therefore
2541 you can only register child watchers in the default event loop.
2542
2543 Due to some design glitches inside libev, child watchers will always be
2544 handled at maximum priority (their priority is set to C<EV_MAXPRI> by
2545 libev)
2546
2547 =head3 Process Interaction
2548
2549 Libev grabs C<SIGCHLD> as soon as the default event loop is
2550 initialised. This is necessary to guarantee proper behaviour even if the
2551 first child watcher is started after the child exits. The occurrence
2552 of C<SIGCHLD> is recorded asynchronously, but child reaping is done
2553 synchronously as part of the event loop processing. Libev always reaps all
2554 children, even ones not watched.
2555
2556 =head3 Overriding the Built-In Processing
2557
2558 Libev offers no special support for overriding the built-in child
2559 processing, but if your application collides with libev's default child
2560 handler, you can override it easily by installing your own handler for
2561 C<SIGCHLD> after initialising the default loop, and making sure the
2562 default loop never gets destroyed. You are encouraged, however, to use an
2563 event-based approach to child reaping and thus use libev's support for
2564 that, so other libev users can use C<ev_child> watchers freely.
2565
2566 =head3 Stopping the Child Watcher
2567
2568 Currently, the child watcher never gets stopped, even when the
2569 child terminates, so normally one needs to stop the watcher in the
2570 callback. Future versions of libev might stop the watcher automatically
2571 when a child exit is detected (calling C<ev_child_stop> twice is not a
2572 problem).
2573
2574 =head3 Watcher-Specific Functions and Data Members
2575
2576 =over 4
2577
2578 =item ev_child_init (ev_child *, callback, int pid, int trace)
2579
2580 =item ev_child_set (ev_child *, int pid, int trace)
2581
2582 Configures the watcher to wait for status changes of process C<pid> (or
2583 I<any> process if C<pid> is specified as C<0>). The callback can look
2584 at the C<rstatus> member of the C<ev_child> watcher structure to see
2585 the status word (use the macros from C<sys/wait.h> and see your systems
2586 C<waitpid> documentation). The C<rpid> member contains the pid of the
2587 process causing the status change. C<trace> must be either C<0> (only
2588 activate the watcher when the process terminates) or C<1> (additionally
2589 activate the watcher when the process is stopped or continued).
2590
2591 =item int pid [read-only]
2592
2593 The process id this watcher watches out for, or C<0>, meaning any process id.
2594
2595 =item int rpid [read-write]
2596
2597 The process id that detected a status change.
2598
2599 =item int rstatus [read-write]
2600
2601 The process exit/trace status caused by C<rpid> (see your systems
2602 C<waitpid> and C<sys/wait.h> documentation for details).
2603
2604 =back
2605
2606 =head3 Examples
2607
2608 Example: C<fork()> a new process and install a child handler to wait for
2609 its completion.
2610
2611 ev_child cw;
2612
2613 static void
2614 child_cb (EV_P_ ev_child *w, int revents)
2615 {
2616 ev_child_stop (EV_A_ w);
2617 printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
2618 }
2619
2620 pid_t pid = fork ();
2621
2622 if (pid < 0)
2623 // error
2624 else if (pid == 0)
2625 {
2626 // the forked child executes here
2627 exit (1);
2628 }
2629 else
2630 {
2631 ev_child_init (&cw, child_cb, pid, 0);
2632 ev_child_start (EV_DEFAULT_ &cw);
2633 }
2634
2635
2636 =head2 C<ev_stat> - did the file attributes just change?
2637
2638 This watches a file system path for attribute changes. That is, it calls
2639 C<stat> on that path in regular intervals (or when the OS says it changed)
2640 and sees if it changed compared to the last time, invoking the callback
2641 if it did. Starting the watcher C<stat>'s the file, so only changes that
2642 happen after the watcher has been started will be reported.
2643
2644 The path does not need to exist: changing from "path exists" to "path does
2645 not exist" is a status change like any other. The condition "path does not
2646 exist" (or more correctly "path cannot be stat'ed") is signified by the
2647 C<st_nlink> field being zero (which is otherwise always forced to be at
2648 least one) and all the other fields of the stat buffer having unspecified
2649 contents.
2650
2651 The path I<must not> end in a slash or contain special components such as
2652 C<.> or C<..>. The path I<should> be absolute: If it is relative and
2653 your working directory changes, then the behaviour is undefined.
2654
2655 Since there is no portable change notification interface available, the
2656 portable implementation simply calls C<stat(2)> regularly on the path
2657 to see if it changed somehow. You can specify a recommended polling
2658 interval for this case. If you specify a polling interval of C<0> (highly
2659 recommended!) then a I<suitable, unspecified default> value will be used
2660 (which you can expect to be around five seconds, although this might
2661 change dynamically). Libev will also impose a minimum interval which is
2662 currently around C<0.1>, but that's usually overkill.
2663
2664 This watcher type is not meant for massive numbers of stat watchers,
2665 as even with OS-supported change notifications, this can be
2666 resource-intensive.
2667
2668 At the time of this writing, the only OS-specific interface implemented
2669 is the Linux inotify interface (implementing kqueue support is left as an
2670 exercise for the reader. Note, however, that the author sees no way of
2671 implementing C<ev_stat> semantics with kqueue, except as a hint).
2672
2673 =head3 ABI Issues (Largefile Support)
2674
2675 Libev by default (unless the user overrides this) uses the default
2676 compilation environment, which means that on systems with large file
2677 support disabled by default, you get the 32 bit version of the stat
2678 structure. When using the library from programs that change the ABI to
2679 use 64 bit file offsets the programs will fail. In that case you have to
2680 compile libev with the same flags to get binary compatibility. This is
2681 obviously the case with any flags that change the ABI, but the problem is
2682 most noticeably displayed with ev_stat and large file support.
2683
2684 The solution for this is to lobby your distribution maker to make large
2685 file interfaces available by default (as e.g. FreeBSD does) and not
2686 optional. Libev cannot simply switch on large file support because it has
2687 to exchange stat structures with application programs compiled using the
2688 default compilation environment.
2689
2690 =head3 Inotify and Kqueue
2691
2692 When C<inotify (7)> support has been compiled into libev and present at
2693 runtime, it will be used to speed up change detection where possible. The
2694 inotify descriptor will be created lazily when the first C<ev_stat>
2695 watcher is being started.
2696
2697 Inotify presence does not change the semantics of C<ev_stat> watchers
2698 except that changes might be detected earlier, and in some cases, to avoid
2699 making regular C<stat> calls. Even in the presence of inotify support
2700 there are many cases where libev has to resort to regular C<stat> polling,
2701 but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
2702 many bugs), the path exists (i.e. stat succeeds), and the path resides on
2703 a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
2704 xfs are fully working) libev usually gets away without polling.
2705
2706 There is no support for kqueue, as apparently it cannot be used to
2707 implement this functionality, due to the requirement of having a file
2708 descriptor open on the object at all times, and detecting renames, unlinks
2709 etc. is difficult.
2710
2711 =head3 C<stat ()> is a synchronous operation
2712
2713 Libev doesn't normally do any kind of I/O itself, and so is not blocking
2714 the process. The exception are C<ev_stat> watchers - those call C<stat
2715 ()>, which is a synchronous operation.
2716
2717 For local paths, this usually doesn't matter: unless the system is very
2718 busy or the intervals between stat's are large, a stat call will be fast,
2719 as the path data is usually in memory already (except when starting the
2720 watcher).
2721
2722 For networked file systems, calling C<stat ()> can block an indefinite
2723 time due to network issues, and even under good conditions, a stat call
2724 often takes multiple milliseconds.
2725
2726 Therefore, it is best to avoid using C<ev_stat> watchers on networked
2727 paths, although this is fully supported by libev.
2728
2729 =head3 The special problem of stat time resolution
2730
2731 The C<stat ()> system call only supports full-second resolution portably,
2732 and even on systems where the resolution is higher, most file systems
2733 still only support whole seconds.
2734
2735 That means that, if the time is the only thing that changes, you can
2736 easily miss updates: on the first update, C<ev_stat> detects a change and
2737 calls your callback, which does something. When there is another update
2738 within the same second, C<ev_stat> will be unable to detect unless the
2739 stat data does change in other ways (e.g. file size).
2740
2741 The solution to this is to delay acting on a change for slightly more
2742 than a second (or till slightly after the next full second boundary), using
2743 a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
2744 ev_timer_again (loop, w)>).
2745
2746 The C<.02> offset is added to work around small timing inconsistencies
2747 of some operating systems (where the second counter of the current time
2748 might be be delayed. One such system is the Linux kernel, where a call to
2749 C<gettimeofday> might return a timestamp with a full second later than
2750 a subsequent C<time> call - if the equivalent of C<time ()> is used to
2751 update file times then there will be a small window where the kernel uses
2752 the previous second to update file times but libev might already execute
2753 the timer callback).
2754
2755 =head3 Watcher-Specific Functions and Data Members
2756
2757 =over 4
2758
2759 =item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
2760
2761 =item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
2762
2763 Configures the watcher to wait for status changes of the given
2764 C<path>. The C<interval> is a hint on how quickly a change is expected to
2765 be detected and should normally be specified as C<0> to let libev choose
2766 a suitable value. The memory pointed to by C<path> must point to the same
2767 path for as long as the watcher is active.
2768
2769 The callback will receive an C<EV_STAT> event when a change was detected,
2770 relative to the attributes at the time the watcher was started (or the
2771 last change was detected).
2772
2773 =item ev_stat_stat (loop, ev_stat *)
2774
2775 Updates the stat buffer immediately with new values. If you change the
2776 watched path in your callback, you could call this function to avoid
2777 detecting this change (while introducing a race condition if you are not
2778 the only one changing the path). Can also be useful simply to find out the
2779 new values.
2780
2781 =item ev_statdata attr [read-only]
2782
2783 The most-recently detected attributes of the file. Although the type is
2784 C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
2785 suitable for your system, but you can only rely on the POSIX-standardised
2786 members to be present. If the C<st_nlink> member is C<0>, then there was
2787 some error while C<stat>ing the file.
2788
2789 =item ev_statdata prev [read-only]
2790
2791 The previous attributes of the file. The callback gets invoked whenever
2792 C<prev> != C<attr>, or, more precisely, one or more of these members
2793 differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
2794 C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
2795
2796 =item ev_tstamp interval [read-only]
2797
2798 The specified interval.
2799
2800 =item const char *path [read-only]
2801
2802 The file system path that is being watched.
2803
2804 =back
2805
2806 =head3 Examples
2807
2808 Example: Watch C</etc/passwd> for attribute changes.
2809
2810 static void
2811 passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
2812 {
2813 /* /etc/passwd changed in some way */
2814 if (w->attr.st_nlink)
2815 {
2816 printf ("passwd current size %ld\n", (long)w->attr.st_size);
2817 printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
2818 printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
2819 }
2820 else
2821 /* you shalt not abuse printf for puts */
2822 puts ("wow, /etc/passwd is not there, expect problems. "
2823 "if this is windows, they already arrived\n");
2824 }
2825
2826 ...
2827 ev_stat passwd;
2828
2829 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
2830 ev_stat_start (loop, &passwd);
2831
2832 Example: Like above, but additionally use a one-second delay so we do not
2833 miss updates (however, frequent updates will delay processing, too, so
2834 one might do the work both on C<ev_stat> callback invocation I<and> on
2835 C<ev_timer> callback invocation).
2836
2837 static ev_stat passwd;
2838 static ev_timer timer;
2839
2840 static void
2841 timer_cb (EV_P_ ev_timer *w, int revents)
2842 {
2843 ev_timer_stop (EV_A_ w);
2844
2845 /* now it's one second after the most recent passwd change */
2846 }
2847
2848 static void
2849 stat_cb (EV_P_ ev_stat *w, int revents)
2850 {
2851 /* reset the one-second timer */
2852 ev_timer_again (EV_A_ &timer);
2853 }
2854
2855 ...
2856 ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
2857 ev_stat_start (loop, &passwd);
2858 ev_timer_init (&timer, timer_cb, 0., 1.02);
2859
2860
2861 =head2 C<ev_idle> - when you've got nothing better to do...
2862
2863 Idle watchers trigger events when no other events of the same or higher
2864 priority are pending (prepare, check and other idle watchers do not count
2865 as receiving "events").
2866
2867 That is, as long as your process is busy handling sockets or timeouts
2868 (or even signals, imagine) of the same or higher priority it will not be
2869 triggered. But when your process is idle (or only lower-priority watchers
2870 are pending), the idle watchers are being called once per event loop
2871 iteration - until stopped, that is, or your process receives more events
2872 and becomes busy again with higher priority stuff.
2873
2874 The most noteworthy effect is that as long as any idle watchers are
2875 active, the process will not block when waiting for new events.
2876
2877 Apart from keeping your process non-blocking (which is a useful
2878 effect on its own sometimes), idle watchers are a good place to do
2879 "pseudo-background processing", or delay processing stuff to after the
2880 event loop has handled all outstanding events.
2881
2882 =head3 Abusing an C<ev_idle> watcher for its side-effect
2883
2884 As long as there is at least one active idle watcher, libev will never
2885 sleep unnecessarily. Or in other words, it will loop as fast as possible.
2886 For this to work, the idle watcher doesn't need to be invoked at all - the
2887 lowest priority will do.
2888
2889 This mode of operation can be useful together with an C<ev_check> watcher,
2890 to do something on each event loop iteration - for example to balance load
2891 between different connections.
2892
2893 See L</Abusing an ev_check watcher for its side-effect> for a longer
2894 example.
2895
2896 =head3 Watcher-Specific Functions and Data Members
2897
2898 =over 4
2899
2900 =item ev_idle_init (ev_idle *, callback)
2901
2902 Initialises and configures the idle watcher - it has no parameters of any
2903 kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
2904 believe me.
2905
2906 =back
2907
2908 =head3 Examples
2909
2910 Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
2911 callback, free it. Also, use no error checking, as usual.
2912
2913 static void
2914 idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
2915 {
2916 // stop the watcher
2917 ev_idle_stop (loop, w);
2918
2919 // now we can free it
2920 free (w);
2921
2922 // now do something you wanted to do when the program has
2923 // no longer anything immediate to do.
2924 }
2925
2926 ev_idle *idle_watcher = malloc (sizeof (ev_idle));
2927 ev_idle_init (idle_watcher, idle_cb);
2928 ev_idle_start (loop, idle_watcher);
2929
2930
2931 =head2 C<ev_prepare> and C<ev_check> - customise your event loop!
2932
2933 Prepare and check watchers are often (but not always) used in pairs:
2934 prepare watchers get invoked before the process blocks and check watchers
2935 afterwards.
2936
2937 You I<must not> call C<ev_run> (or similar functions that enter the
2938 current event loop) or C<ev_loop_fork> from either C<ev_prepare> or
2939 C<ev_check> watchers. Other loops than the current one are fine,
2940 however. The rationale behind this is that you do not need to check
2941 for recursion in those watchers, i.e. the sequence will always be
2942 C<ev_prepare>, blocking, C<ev_check> so if you have one watcher of each
2943 kind they will always be called in pairs bracketing the blocking call.
2944
2945 Their main purpose is to integrate other event mechanisms into libev and
2946 their use is somewhat advanced. They could be used, for example, to track
2947 variable changes, implement your own watchers, integrate net-snmp or a
2948 coroutine library and lots more. They are also occasionally useful if
2949 you cache some data and want to flush it before blocking (for example,
2950 in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
2951 watcher).
2952
2953 This is done by examining in each prepare call which file descriptors
2954 need to be watched by the other library, registering C<ev_io> watchers
2955 for them and starting an C<ev_timer> watcher for any timeouts (many
2956 libraries provide exactly this functionality). Then, in the check watcher,
2957 you check for any events that occurred (by checking the pending status
2958 of all watchers and stopping them) and call back into the library. The
2959 I/O and timer callbacks will never actually be called (but must be valid
2960 nevertheless, because you never know, you know?).
2961
2962 As another example, the Perl Coro module uses these hooks to integrate
2963 coroutines into libev programs, by yielding to other active coroutines
2964 during each prepare and only letting the process block if no coroutines
2965 are ready to run (it's actually more complicated: it only runs coroutines
2966 with priority higher than or equal to the event loop and one coroutine
2967 of lower priority, but only once, using idle watchers to keep the event
2968 loop from blocking if lower-priority coroutines are active, thus mapping
2969 low-priority coroutines to idle/background tasks).
2970
2971 When used for this purpose, it is recommended to give C<ev_check> watchers
2972 highest (C<EV_MAXPRI>) priority, to ensure that they are being run before
2973 any other watchers after the poll (this doesn't matter for C<ev_prepare>
2974 watchers).
2975
2976 Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
2977 activate ("feed") events into libev. While libev fully supports this, they
2978 might get executed before other C<ev_check> watchers did their job. As
2979 C<ev_check> watchers are often used to embed other (non-libev) event
2980 loops those other event loops might be in an unusable state until their
2981 C<ev_check> watcher ran (always remind yourself to coexist peacefully with
2982 others).
2983
2984 =head3 Abusing an C<ev_check> watcher for its side-effect
2985
2986 C<ev_check> (and less often also C<ev_prepare>) watchers can also be
2987 useful because they are called once per event loop iteration. For
2988 example, if you want to handle a large number of connections fairly, you
2989 normally only do a bit of work for each active connection, and if there
2990 is more work to do, you wait for the next event loop iteration, so other
2991 connections have a chance of making progress.
2992
2993 Using an C<ev_check> watcher is almost enough: it will be called on the
2994 next event loop iteration. However, that isn't as soon as possible -
2995 without external events, your C<ev_check> watcher will not be invoked.
2996
2997 This is where C<ev_idle> watchers come in handy - all you need is a
2998 single global idle watcher that is active as long as you have one active
2999 C<ev_check> watcher. The C<ev_idle> watcher makes sure the event loop
3000 will not sleep, and the C<ev_check> watcher makes sure a callback gets
3001 invoked. Neither watcher alone can do that.
3002
3003 =head3 Watcher-Specific Functions and Data Members
3004
3005 =over 4
3006
3007 =item ev_prepare_init (ev_prepare *, callback)
3008
3009 =item ev_check_init (ev_check *, callback)
3010
3011 Initialises and configures the prepare or check watcher - they have no
3012 parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
3013 macros, but using them is utterly, utterly, utterly and completely
3014 pointless.
3015
3016 =back
3017
3018 =head3 Examples
3019
3020 There are a number of principal ways to embed other event loops or modules
3021 into libev. Here are some ideas on how to include libadns into libev
3022 (there is a Perl module named C<EV::ADNS> that does this, which you could
3023 use as a working example. Another Perl module named C<EV::Glib> embeds a
3024 Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
3025 Glib event loop).
3026
3027 Method 1: Add IO watchers and a timeout watcher in a prepare handler,
3028 and in a check watcher, destroy them and call into libadns. What follows
3029 is pseudo-code only of course. This requires you to either use a low
3030 priority for the check watcher or use C<ev_clear_pending> explicitly, as
3031 the callbacks for the IO/timeout watchers might not have been called yet.
3032
3033 static ev_io iow [nfd];
3034 static ev_timer tw;
3035
3036 static void
3037 io_cb (struct ev_loop *loop, ev_io *w, int revents)
3038 {
3039 }
3040
3041 // create io watchers for each fd and a timer before blocking
3042 static void
3043 adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
3044 {
3045 int timeout = 3600000;
3046 struct pollfd fds [nfd];
3047 // actual code will need to loop here and realloc etc.
3048 adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
3049
3050 /* the callback is illegal, but won't be called as we stop during check */
3051 ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
3052 ev_timer_start (loop, &tw);
3053
3054 // create one ev_io per pollfd
3055 for (int i = 0; i < nfd; ++i)
3056 {
3057 ev_io_init (iow + i, io_cb, fds [i].fd,
3058 ((fds [i].events & POLLIN ? EV_READ : 0)
3059 | (fds [i].events & POLLOUT ? EV_WRITE : 0)));
3060
3061 fds [i].revents = 0;
3062 ev_io_start (loop, iow + i);
3063 }
3064 }
3065
3066 // stop all watchers after blocking
3067 static void
3068 adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
3069 {
3070 ev_timer_stop (loop, &tw);
3071
3072 for (int i = 0; i < nfd; ++i)
3073 {
3074 // set the relevant poll flags
3075 // could also call adns_processreadable etc. here
3076 struct pollfd *fd = fds + i;
3077 int revents = ev_clear_pending (iow + i);
3078 if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
3079 if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
3080
3081 // now stop the watcher
3082 ev_io_stop (loop, iow + i);
3083 }
3084
3085 adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
3086 }
3087
3088 Method 2: This would be just like method 1, but you run C<adns_afterpoll>
3089 in the prepare watcher and would dispose of the check watcher.
3090
3091 Method 3: If the module to be embedded supports explicit event
3092 notification (libadns does), you can also make use of the actual watcher
3093 callbacks, and only destroy/create the watchers in the prepare watcher.
3094
3095 static void
3096 timer_cb (EV_P_ ev_timer *w, int revents)
3097 {
3098 adns_state ads = (adns_state)w->data;
3099 update_now (EV_A);
3100
3101 adns_processtimeouts (ads, &tv_now);
3102 }
3103
3104 static void
3105 io_cb (EV_P_ ev_io *w, int revents)
3106 {
3107 adns_state ads = (adns_state)w->data;
3108 update_now (EV_A);
3109
3110 if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
3111 if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
3112 }
3113
3114 // do not ever call adns_afterpoll
3115
3116 Method 4: Do not use a prepare or check watcher because the module you
3117 want to embed is not flexible enough to support it. Instead, you can
3118 override their poll function. The drawback with this solution is that the
3119 main loop is now no longer controllable by EV. The C<Glib::EV> module uses
3120 this approach, effectively embedding EV as a client into the horrible
3121 libglib event loop.
3122
3123 static gint
3124 event_poll_func (GPollFD *fds, guint nfds, gint timeout)
3125 {
3126 int got_events = 0;
3127
3128 for (n = 0; n < nfds; ++n)
3129 // create/start io watcher that sets the relevant bits in fds[n] and increment got_events
3130
3131 if (timeout >= 0)
3132 // create/start timer
3133
3134 // poll
3135 ev_run (EV_A_ 0);
3136
3137 // stop timer again
3138 if (timeout >= 0)
3139 ev_timer_stop (EV_A_ &to);
3140
3141 // stop io watchers again - their callbacks should have set
3142 for (n = 0; n < nfds; ++n)
3143 ev_io_stop (EV_A_ iow [n]);
3144
3145 return got_events;
3146 }
3147
3148
3149 =head2 C<ev_embed> - when one backend isn't enough...
3150
3151 This is a rather advanced watcher type that lets you embed one event loop
3152 into another (currently only C<ev_io> events are supported in the embedded
3153 loop, other types of watchers might be handled in a delayed or incorrect
3154 fashion and must not be used).
3155
3156 There are primarily two reasons you would want that: work around bugs and
3157 prioritise I/O.
3158
3159 As an example for a bug workaround, the kqueue backend might only support
3160 sockets on some platform, so it is unusable as generic backend, but you
3161 still want to make use of it because you have many sockets and it scales
3162 so nicely. In this case, you would create a kqueue-based loop and embed
3163 it into your default loop (which might use e.g. poll). Overall operation
3164 will be a bit slower because first libev has to call C<poll> and then
3165 C<kevent>, but at least you can use both mechanisms for what they are
3166 best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
3167
3168 As for prioritising I/O: under rare circumstances you have the case where
3169 some fds have to be watched and handled very quickly (with low latency),
3170 and even priorities and idle watchers might have too much overhead. In
3171 this case you would put all the high priority stuff in one loop and all
3172 the rest in a second one, and embed the second one in the first.
3173
3174 As long as the watcher is active, the callback will be invoked every
3175 time there might be events pending in the embedded loop. The callback
3176 must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
3177 sweep and invoke their callbacks (the callback doesn't need to invoke the
3178 C<ev_embed_sweep> function directly, it could also start an idle watcher
3179 to give the embedded loop strictly lower priority for example).
3180
3181 You can also set the callback to C<0>, in which case the embed watcher
3182 will automatically execute the embedded loop sweep whenever necessary.
3183
3184 Fork detection will be handled transparently while the C<ev_embed> watcher
3185 is active, i.e., the embedded loop will automatically be forked when the
3186 embedding loop forks. In other cases, the user is responsible for calling
3187 C<ev_loop_fork> on the embedded loop.
3188
3189 Unfortunately, not all backends are embeddable: only the ones returned by
3190 C<ev_embeddable_backends> are, which, unfortunately, does not include any
3191 portable one.
3192
3193 So when you want to use this feature you will always have to be prepared
3194 that you cannot get an embeddable loop. The recommended way to get around
3195 this is to have a separate variables for your embeddable loop, try to
3196 create it, and if that fails, use the normal loop for everything.
3197
3198 =head3 C<ev_embed> and fork
3199
3200 While the C<ev_embed> watcher is running, forks in the embedding loop will
3201 automatically be applied to the embedded loop as well, so no special
3202 fork handling is required in that case. When the watcher is not running,
3203 however, it is still the task of the libev user to call C<ev_loop_fork ()>
3204 as applicable.
3205
3206 =head3 Watcher-Specific Functions and Data Members
3207
3208 =over 4
3209
3210 =item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
3211
3212 =item ev_embed_set (ev_embed *, struct ev_loop *embedded_loop)
3213
3214 Configures the watcher to embed the given loop, which must be
3215 embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
3216 invoked automatically, otherwise it is the responsibility of the callback
3217 to invoke it (it will continue to be called until the sweep has been done,
3218 if you do not want that, you need to temporarily stop the embed watcher).
3219
3220 =item ev_embed_sweep (loop, ev_embed *)
3221
3222 Make a single, non-blocking sweep over the embedded loop. This works
3223 similarly to C<ev_run (embedded_loop, EVRUN_NOWAIT)>, but in the most
3224 appropriate way for embedded loops.
3225
3226 =item struct ev_loop *other [read-only]
3227
3228 The embedded event loop.
3229
3230 =back
3231
3232 =head3 Examples
3233
3234 Example: Try to get an embeddable event loop and embed it into the default
3235 event loop. If that is not possible, use the default loop. The default
3236 loop is stored in C<loop_hi>, while the embeddable loop is stored in
3237 C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
3238 used).
3239
3240 struct ev_loop *loop_hi = ev_default_init (0);
3241 struct ev_loop *loop_lo = 0;
3242 ev_embed embed;
3243
3244 // see if there is a chance of getting one that works
3245 // (remember that a flags value of 0 means autodetection)
3246 loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
3247 ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
3248 : 0;
3249
3250 // if we got one, then embed it, otherwise default to loop_hi
3251 if (loop_lo)
3252 {
3253 ev_embed_init (&embed, 0, loop_lo);
3254 ev_embed_start (loop_hi, &embed);
3255 }
3256 else
3257 loop_lo = loop_hi;
3258
3259 Example: Check if kqueue is available but not recommended and create
3260 a kqueue backend for use with sockets (which usually work with any
3261 kqueue implementation). Store the kqueue/socket-only event loop in
3262 C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
3263
3264 struct ev_loop *loop = ev_default_init (0);
3265 struct ev_loop *loop_socket = 0;
3266 ev_embed embed;
3267
3268 if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
3269 if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
3270 {
3271 ev_embed_init (&embed, 0, loop_socket);
3272 ev_embed_start (loop, &embed);
3273 }
3274
3275 if (!loop_socket)
3276 loop_socket = loop;
3277
3278 // now use loop_socket for all sockets, and loop for everything else
3279
3280
3281 =head2 C<ev_fork> - the audacity to resume the event loop after a fork
3282
3283 Fork watchers are called when a C<fork ()> was detected (usually because
3284 whoever is a good citizen cared to tell libev about it by calling
3285 C<ev_loop_fork>). The invocation is done before the event loop blocks next
3286 and before C<ev_check> watchers are being called, and only in the child
3287 after the fork. If whoever good citizen calling C<ev_default_fork> cheats
3288 and calls it in the wrong process, the fork handlers will be invoked, too,
3289 of course.
3290
3291 =head3 The special problem of life after fork - how is it possible?
3292
3293 Most uses of C<fork ()> consist of forking, then some simple calls to set
3294 up/change the process environment, followed by a call to C<exec()>. This
3295 sequence should be handled by libev without any problems.
3296
3297 This changes when the application actually wants to do event handling
3298 in the child, or both parent in child, in effect "continuing" after the
3299 fork.
3300
3301 The default mode of operation (for libev, with application help to detect
3302 forks) is to duplicate all the state in the child, as would be expected
3303 when I<either> the parent I<or> the child process continues.
3304
3305 When both processes want to continue using libev, then this is usually the
3306 wrong result. In that case, usually one process (typically the parent) is
3307 supposed to continue with all watchers in place as before, while the other
3308 process typically wants to start fresh, i.e. without any active watchers.
3309
3310 The cleanest and most efficient way to achieve that with libev is to
3311 simply create a new event loop, which of course will be "empty", and
3312 use that for new watchers. This has the advantage of not touching more
3313 memory than necessary, and thus avoiding the copy-on-write, and the
3314 disadvantage of having to use multiple event loops (which do not support
3315 signal watchers).
3316
3317 When this is not possible, or you want to use the default loop for
3318 other reasons, then in the process that wants to start "fresh", call
3319 C<ev_loop_destroy (EV_DEFAULT)> followed by C<ev_default_loop (...)>.
3320 Destroying the default loop will "orphan" (not stop) all registered
3321 watchers, so you have to be careful not to execute code that modifies
3322 those watchers. Note also that in that case, you have to re-register any
3323 signal watchers.
3324
3325 =head3 Watcher-Specific Functions and Data Members
3326
3327 =over 4
3328
3329 =item ev_fork_init (ev_fork *, callback)
3330
3331 Initialises and configures the fork watcher - it has no parameters of any
3332 kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
3333 really.
3334
3335 =back
3336
3337
3338 =head2 C<ev_cleanup> - even the best things end
3339
3340 Cleanup watchers are called just before the event loop is being destroyed
3341 by a call to C<ev_loop_destroy>.
3342
3343 While there is no guarantee that the event loop gets destroyed, cleanup
3344 watchers provide a convenient method to install cleanup hooks for your
3345 program, worker threads and so on - you just to make sure to destroy the
3346 loop when you want them to be invoked.
3347
3348 Cleanup watchers are invoked in the same way as any other watcher. Unlike
3349 all other watchers, they do not keep a reference to the event loop (which
3350 makes a lot of sense if you think about it). Like all other watchers, you
3351 can call libev functions in the callback, except C<ev_cleanup_start>.
3352
3353 =head3 Watcher-Specific Functions and Data Members
3354
3355 =over 4
3356
3357 =item ev_cleanup_init (ev_cleanup *, callback)
3358
3359 Initialises and configures the cleanup watcher - it has no parameters of
3360 any kind. There is a C<ev_cleanup_set> macro, but using it is utterly
3361 pointless, I assure you.
3362
3363 =back
3364
3365 Example: Register an atexit handler to destroy the default loop, so any
3366 cleanup functions are called.
3367
3368 static void
3369 program_exits (void)
3370 {
3371 ev_loop_destroy (EV_DEFAULT_UC);
3372 }
3373
3374 ...
3375 atexit (program_exits);
3376
3377
3378 =head2 C<ev_async> - how to wake up an event loop
3379
3380 In general, you cannot use an C<ev_loop> from multiple threads or other
3381 asynchronous sources such as signal handlers (as opposed to multiple event
3382 loops - those are of course safe to use in different threads).
3383
3384 Sometimes, however, you need to wake up an event loop you do not control,
3385 for example because it belongs to another thread. This is what C<ev_async>
3386 watchers do: as long as the C<ev_async> watcher is active, you can signal
3387 it by calling C<ev_async_send>, which is thread- and signal safe.
3388
3389 This functionality is very similar to C<ev_signal> watchers, as signals,
3390 too, are asynchronous in nature, and signals, too, will be compressed
3391 (i.e. the number of callback invocations may be less than the number of
3392 C<ev_async_send> calls). In fact, you could use signal watchers as a kind
3393 of "global async watchers" by using a watcher on an otherwise unused
3394 signal, and C<ev_feed_signal> to signal this watcher from another thread,
3395 even without knowing which loop owns the signal.
3396
3397 =head3 Queueing
3398
3399 C<ev_async> does not support queueing of data in any way. The reason
3400 is that the author does not know of a simple (or any) algorithm for a
3401 multiple-writer-single-reader queue that works in all cases and doesn't
3402 need elaborate support such as pthreads or unportable memory access
3403 semantics.
3404
3405 That means that if you want to queue data, you have to provide your own
3406 queue. But at least I can tell you how to implement locking around your
3407 queue:
3408
3409 =over 4
3410
3411 =item queueing from a signal handler context
3412
3413 To implement race-free queueing, you simply add to the queue in the signal
3414 handler but you block the signal handler in the watcher callback. Here is
3415 an example that does that for some fictitious SIGUSR1 handler:
3416
3417 static ev_async mysig;
3418
3419 static void
3420 sigusr1_handler (void)
3421 {
3422 sometype data;
3423
3424 // no locking etc.
3425 queue_put (data);
3426 ev_async_send (EV_DEFAULT_ &mysig);
3427 }
3428
3429 static void
3430 mysig_cb (EV_P_ ev_async *w, int revents)
3431 {
3432 sometype data;
3433 sigset_t block, prev;
3434
3435 sigemptyset (&block);
3436 sigaddset (&block, SIGUSR1);
3437 sigprocmask (SIG_BLOCK, &block, &prev);
3438
3439 while (queue_get (&data))
3440 process (data);
3441
3442 if (sigismember (&prev, SIGUSR1)
3443 sigprocmask (SIG_UNBLOCK, &block, 0);
3444 }
3445
3446 (Note: pthreads in theory requires you to use C<pthread_setmask>
3447 instead of C<sigprocmask> when you use threads, but libev doesn't do it
3448 either...).
3449
3450 =item queueing from a thread context
3451
3452 The strategy for threads is different, as you cannot (easily) block
3453 threads but you can easily preempt them, so to queue safely you need to
3454 employ a traditional mutex lock, such as in this pthread example:
3455
3456 static ev_async mysig;
3457 static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
3458
3459 static void
3460 otherthread (void)
3461 {
3462 // only need to lock the actual queueing operation
3463 pthread_mutex_lock (&mymutex);
3464 queue_put (data);
3465 pthread_mutex_unlock (&mymutex);
3466
3467 ev_async_send (EV_DEFAULT_ &mysig);
3468 }
3469
3470 static void
3471 mysig_cb (EV_P_ ev_async *w, int revents)
3472 {
3473 pthread_mutex_lock (&mymutex);
3474
3475 while (queue_get (&data))
3476 process (data);
3477
3478 pthread_mutex_unlock (&mymutex);
3479 }
3480
3481 =back
3482
3483
3484 =head3 Watcher-Specific Functions and Data Members
3485
3486 =over 4
3487
3488 =item ev_async_init (ev_async *, callback)
3489
3490 Initialises and configures the async watcher - it has no parameters of any
3491 kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
3492 trust me.
3493
3494 =item ev_async_send (loop, ev_async *)
3495
3496 Sends/signals/activates the given C<ev_async> watcher, that is, feeds
3497 an C<EV_ASYNC> event on the watcher into the event loop, and instantly
3498 returns.
3499
3500 Unlike C<ev_feed_event>, this call is safe to do from other threads,
3501 signal or similar contexts (see the discussion of C<EV_ATOMIC_T> in the
3502 embedding section below on what exactly this means).
3503
3504 Note that, as with other watchers in libev, multiple events might get
3505 compressed into a single callback invocation (another way to look at
3506 this is that C<ev_async> watchers are level-triggered: they are set on
3507 C<ev_async_send>, reset when the event loop detects that).
3508
3509 This call incurs the overhead of at most one extra system call per event
3510 loop iteration, if the event loop is blocked, and no syscall at all if
3511 the event loop (or your program) is processing events. That means that
3512 repeated calls are basically free (there is no need to avoid calls for
3513 performance reasons) and that the overhead becomes smaller (typically
3514 zero) under load.
3515
3516 =item bool = ev_async_pending (ev_async *)
3517
3518 Returns a non-zero value when C<ev_async_send> has been called on the
3519 watcher but the event has not yet been processed (or even noted) by the
3520 event loop.
3521
3522 C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
3523 the loop iterates next and checks for the watcher to have become active,
3524 it will reset the flag again. C<ev_async_pending> can be used to very
3525 quickly check whether invoking the loop might be a good idea.
3526
3527 Not that this does I<not> check whether the watcher itself is pending,
3528 only whether it has been requested to make this watcher pending: there
3529 is a time window between the event loop checking and resetting the async
3530 notification, and the callback being invoked.
3531
3532 =back
3533
3534
3535 =head1 OTHER FUNCTIONS
3536
3537 There are some other functions of possible interest. Described. Here. Now.
3538
3539 =over 4
3540
3541 =item ev_once (loop, int fd, int events, ev_tstamp timeout, callback, arg)
3542
3543 This function combines a simple timer and an I/O watcher, calls your
3544 callback on whichever event happens first and automatically stops both
3545 watchers. This is useful if you want to wait for a single event on an fd
3546 or timeout without having to allocate/configure/start/stop/free one or
3547 more watchers yourself.
3548
3549 If C<fd> is less than 0, then no I/O watcher will be started and the
3550 C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
3551 the given C<fd> and C<events> set will be created and started.
3552
3553 If C<timeout> is less than 0, then no timeout watcher will be
3554 started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
3555 repeat = 0) will be started. C<0> is a valid timeout.
3556
3557 The callback has the type C<void (*cb)(int revents, void *arg)> and is
3558 passed an C<revents> set like normal event callbacks (a combination of
3559 C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMER>) and the C<arg>
3560 value passed to C<ev_once>. Note that it is possible to receive I<both>
3561 a timeout and an io event at the same time - you probably should give io
3562 events precedence.
3563
3564 Example: wait up to ten seconds for data to appear on STDIN_FILENO.
3565
3566 static void stdin_ready (int revents, void *arg)
3567 {
3568 if (revents & EV_READ)
3569 /* stdin might have data for us, joy! */;
3570 else if (revents & EV_TIMER)
3571 /* doh, nothing entered */;
3572 }
3573
3574 ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
3575
3576 =item ev_feed_fd_event (loop, int fd, int revents)
3577
3578 Feed an event on the given fd, as if a file descriptor backend detected
3579 the given events.
3580
3581 =item ev_feed_signal_event (loop, int signum)
3582
3583 Feed an event as if the given signal occurred. See also C<ev_feed_signal>,
3584 which is async-safe.
3585
3586 =back
3587
3588
3589 =head1 COMMON OR USEFUL IDIOMS (OR BOTH)
3590
3591 This section explains some common idioms that are not immediately
3592 obvious. Note that examples are sprinkled over the whole manual, and this
3593 section only contains stuff that wouldn't fit anywhere else.
3594
3595 =head2 ASSOCIATING CUSTOM DATA WITH A WATCHER
3596
3597 Each watcher has, by default, a C<void *data> member that you can read
3598 or modify at any time: libev will completely ignore it. This can be used
3599 to associate arbitrary data with your watcher. If you need more data and
3600 don't want to allocate memory separately and store a pointer to it in that
3601 data member, you can also "subclass" the watcher type and provide your own
3602 data:
3603
3604 struct my_io
3605 {
3606 ev_io io;
3607 int otherfd;
3608 void *somedata;
3609 struct whatever *mostinteresting;
3610 };
3611
3612 ...
3613 struct my_io w;
3614 ev_io_init (&w.io, my_cb, fd, EV_READ);
3615
3616 And since your callback will be called with a pointer to the watcher, you
3617 can cast it back to your own type:
3618
3619 static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
3620 {
3621 struct my_io *w = (struct my_io *)w_;
3622 ...
3623 }
3624
3625 More interesting and less C-conformant ways of casting your callback
3626 function type instead have been omitted.
3627
3628 =head2 BUILDING YOUR OWN COMPOSITE WATCHERS
3629
3630 Another common scenario is to use some data structure with multiple
3631 embedded watchers, in effect creating your own watcher that combines
3632 multiple libev event sources into one "super-watcher":
3633
3634 struct my_biggy
3635 {
3636 int some_data;
3637 ev_timer t1;
3638 ev_timer t2;
3639 }
3640
3641 In this case getting the pointer to C<my_biggy> is a bit more
3642 complicated: Either you store the address of your C<my_biggy> struct in
3643 the C<data> member of the watcher (for woozies or C++ coders), or you need
3644 to use some pointer arithmetic using C<offsetof> inside your watchers (for
3645 real programmers):
3646
3647 #include <stddef.h>
3648
3649 static void
3650 t1_cb (EV_P_ ev_timer *w, int revents)
3651 {
3652 struct my_biggy big = (struct my_biggy *)
3653 (((char *)w) - offsetof (struct my_biggy, t1));
3654 }
3655
3656 static void
3657 t2_cb (EV_P_ ev_timer *w, int revents)
3658 {
3659 struct my_biggy big = (struct my_biggy *)
3660 (((char *)w) - offsetof (struct my_biggy, t2));
3661 }
3662
3663 =head2 AVOIDING FINISHING BEFORE RETURNING
3664
3665 Often you have structures like this in event-based programs:
3666
3667 callback ()
3668 {
3669 free (request);
3670 }
3671
3672 request = start_new_request (..., callback);
3673
3674 The intent is to start some "lengthy" operation. The C<request> could be
3675 used to cancel the operation, or do other things with it.
3676
3677 It's not uncommon to have code paths in C<start_new_request> that
3678 immediately invoke the callback, for example, to report errors. Or you add
3679 some caching layer that finds that it can skip the lengthy aspects of the
3680 operation and simply invoke the callback with the result.
3681
3682 The problem here is that this will happen I<before> C<start_new_request>
3683 has returned, so C<request> is not set.
3684
3685 Even if you pass the request by some safer means to the callback, you
3686 might want to do something to the request after starting it, such as
3687 canceling it, which probably isn't working so well when the callback has
3688 already been invoked.
3689
3690 A common way around all these issues is to make sure that
3691 C<start_new_request> I<always> returns before the callback is invoked. If
3692 C<start_new_request> immediately knows the result, it can artificially
3693 delay invoking the callback by using a C<prepare> or C<idle> watcher for
3694 example, or more sneakily, by reusing an existing (stopped) watcher and
3695 pushing it into the pending queue:
3696
3697 ev_set_cb (watcher, callback);
3698 ev_feed_event (EV_A_ watcher, 0);
3699
3700 This way, C<start_new_request> can safely return before the callback is
3701 invoked, while not delaying callback invocation too much.
3702
3703 =head2 MODEL/NESTED EVENT LOOP INVOCATIONS AND EXIT CONDITIONS
3704
3705 Often (especially in GUI toolkits) there are places where you have
3706 I<modal> interaction, which is most easily implemented by recursively
3707 invoking C<ev_run>.
3708
3709 This brings the problem of exiting - a callback might want to finish the
3710 main C<ev_run> call, but not the nested one (e.g. user clicked "Quit", but
3711 a modal "Are you sure?" dialog is still waiting), or just the nested one
3712 and not the main one (e.g. user clocked "Ok" in a modal dialog), or some
3713 other combination: In these cases, a simple C<ev_break> will not work.
3714
3715 The solution is to maintain "break this loop" variable for each C<ev_run>
3716 invocation, and use a loop around C<ev_run> until the condition is
3717 triggered, using C<EVRUN_ONCE>:
3718
3719 // main loop
3720 int exit_main_loop = 0;
3721
3722 while (!exit_main_loop)
3723 ev_run (EV_DEFAULT_ EVRUN_ONCE);
3724
3725 // in a modal watcher
3726 int exit_nested_loop = 0;
3727
3728 while (!exit_nested_loop)
3729 ev_run (EV_A_ EVRUN_ONCE);
3730
3731 To exit from any of these loops, just set the corresponding exit variable:
3732
3733 // exit modal loop
3734 exit_nested_loop = 1;
3735
3736 // exit main program, after modal loop is finished
3737 exit_main_loop = 1;
3738
3739 // exit both
3740 exit_main_loop = exit_nested_loop = 1;
3741
3742 =head2 THREAD LOCKING EXAMPLE
3743
3744 Here is a fictitious example of how to run an event loop in a different
3745 thread from where callbacks are being invoked and watchers are
3746 created/added/removed.
3747
3748 For a real-world example, see the C<EV::Loop::Async> perl module,
3749 which uses exactly this technique (which is suited for many high-level
3750 languages).
3751
3752 The example uses a pthread mutex to protect the loop data, a condition
3753 variable to wait for callback invocations, an async watcher to notify the
3754 event loop thread and an unspecified mechanism to wake up the main thread.
3755
3756 First, you need to associate some data with the event loop:
3757
3758 typedef struct {
3759 mutex_t lock; /* global loop lock */
3760 ev_async async_w;
3761 thread_t tid;
3762 cond_t invoke_cv;
3763 } userdata;
3764
3765 void prepare_loop (EV_P)
3766 {
3767 // for simplicity, we use a static userdata struct.
3768 static userdata u;
3769
3770 ev_async_init (&u->async_w, async_cb);
3771 ev_async_start (EV_A_ &u->async_w);
3772
3773 pthread_mutex_init (&u->lock, 0);
3774 pthread_cond_init (&u->invoke_cv, 0);
3775
3776 // now associate this with the loop
3777 ev_set_userdata (EV_A_ u);
3778 ev_set_invoke_pending_cb (EV_A_ l_invoke);
3779 ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
3780
3781 // then create the thread running ev_run
3782 pthread_create (&u->tid, 0, l_run, EV_A);
3783 }
3784
3785 The callback for the C<ev_async> watcher does nothing: the watcher is used
3786 solely to wake up the event loop so it takes notice of any new watchers
3787 that might have been added:
3788
3789 static void
3790 async_cb (EV_P_ ev_async *w, int revents)
3791 {
3792 // just used for the side effects
3793 }
3794
3795 The C<l_release> and C<l_acquire> callbacks simply unlock/lock the mutex
3796 protecting the loop data, respectively.
3797
3798 static void
3799 l_release (EV_P)
3800 {
3801 userdata *u = ev_userdata (EV_A);
3802 pthread_mutex_unlock (&u->lock);
3803 }
3804
3805 static void
3806 l_acquire (EV_P)
3807 {
3808 userdata *u = ev_userdata (EV_A);
3809 pthread_mutex_lock (&u->lock);
3810 }
3811
3812 The event loop thread first acquires the mutex, and then jumps straight
3813 into C<ev_run>:
3814
3815 void *
3816 l_run (void *thr_arg)
3817 {
3818 struct ev_loop *loop = (struct ev_loop *)thr_arg;
3819
3820 l_acquire (EV_A);
3821 pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
3822 ev_run (EV_A_ 0);
3823 l_release (EV_A);
3824
3825 return 0;
3826 }
3827
3828 Instead of invoking all pending watchers, the C<l_invoke> callback will
3829 signal the main thread via some unspecified mechanism (signals? pipe
3830 writes? C<Async::Interrupt>?) and then waits until all pending watchers
3831 have been called (in a while loop because a) spurious wakeups are possible
3832 and b) skipping inter-thread-communication when there are no pending
3833 watchers is very beneficial):
3834
3835 static void
3836 l_invoke (EV_P)
3837 {
3838 userdata *u = ev_userdata (EV_A);
3839
3840 while (ev_pending_count (EV_A))
3841 {
3842 wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
3843 pthread_cond_wait (&u->invoke_cv, &u->lock);
3844 }
3845 }
3846
3847 Now, whenever the main thread gets told to invoke pending watchers, it
3848 will grab the lock, call C<ev_invoke_pending> and then signal the loop
3849 thread to continue:
3850
3851 static void
3852 real_invoke_pending (EV_P)
3853 {
3854 userdata *u = ev_userdata (EV_A);
3855
3856 pthread_mutex_lock (&u->lock);
3857 ev_invoke_pending (EV_A);
3858 pthread_cond_signal (&u->invoke_cv);
3859 pthread_mutex_unlock (&u->lock);
3860 }
3861
3862 Whenever you want to start/stop a watcher or do other modifications to an
3863 event loop, you will now have to lock:
3864
3865 ev_timer timeout_watcher;
3866 userdata *u = ev_userdata (EV_A);
3867
3868 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
3869
3870 pthread_mutex_lock (&u->lock);
3871 ev_timer_start (EV_A_ &timeout_watcher);
3872 ev_async_send (EV_A_ &u->async_w);
3873 pthread_mutex_unlock (&u->lock);
3874
3875 Note that sending the C<ev_async> watcher is required because otherwise
3876 an event loop currently blocking in the kernel will have no knowledge
3877 about the newly added timer. By waking up the loop it will pick up any new
3878 watchers in the next event loop iteration.
3879
3880 =head2 THREADS, COROUTINES, CONTINUATIONS, QUEUES... INSTEAD OF CALLBACKS
3881
3882 While the overhead of a callback that e.g. schedules a thread is small, it
3883 is still an overhead. If you embed libev, and your main usage is with some
3884 kind of threads or coroutines, you might want to customise libev so that
3885 doesn't need callbacks anymore.
3886
3887 Imagine you have coroutines that you can switch to using a function
3888 C<switch_to (coro)>, that libev runs in a coroutine called C<libev_coro>
3889 and that due to some magic, the currently active coroutine is stored in a
3890 global called C<current_coro>. Then you can build your own "wait for libev
3891 event" primitive by changing C<EV_CB_DECLARE> and C<EV_CB_INVOKE> (note
3892 the differing C<;> conventions):
3893
3894 #define EV_CB_DECLARE(type) struct my_coro *cb;
3895 #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
3896
3897 That means instead of having a C callback function, you store the
3898 coroutine to switch to in each watcher, and instead of having libev call
3899 your callback, you instead have it switch to that coroutine.
3900
3901 A coroutine might now wait for an event with a function called
3902 C<wait_for_event>. (the watcher needs to be started, as always, but it doesn't
3903 matter when, or whether the watcher is active or not when this function is
3904 called):
3905
3906 void
3907 wait_for_event (ev_watcher *w)
3908 {
3909 ev_set_cb (w, current_coro);
3910 switch_to (libev_coro);
3911 }
3912
3913 That basically suspends the coroutine inside C<wait_for_event> and
3914 continues the libev coroutine, which, when appropriate, switches back to
3915 this or any other coroutine.
3916
3917 You can do similar tricks if you have, say, threads with an event queue -
3918 instead of storing a coroutine, you store the queue object and instead of
3919 switching to a coroutine, you push the watcher onto the queue and notify
3920 any waiters.
3921
3922 To embed libev, see L</EMBEDDING>, but in short, it's easiest to create two
3923 files, F<my_ev.h> and F<my_ev.c> that include the respective libev files:
3924
3925 // my_ev.h
3926 #define EV_CB_DECLARE(type) struct my_coro *cb;
3927 #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
3928 #include "../libev/ev.h"
3929
3930 // my_ev.c
3931 #define EV_H "my_ev.h"
3932 #include "../libev/ev.c"
3933
3934 And then use F<my_ev.h> when you would normally use F<ev.h>, and compile
3935 F<my_ev.c> into your project. When properly specifying include paths, you
3936 can even use F<ev.h> as header file name directly.
3937
3938
3939 =head1 LIBEVENT EMULATION
3940
3941 Libev offers a compatibility emulation layer for libevent. It cannot
3942 emulate the internals of libevent, so here are some usage hints:
3943
3944 =over 4
3945
3946 =item * Only the libevent-1.4.1-beta API is being emulated.
3947
3948 This was the newest libevent version available when libev was implemented,
3949 and is still mostly unchanged in 2010.
3950
3951 =item * Use it by including <event.h>, as usual.
3952
3953 =item * The following members are fully supported: ev_base, ev_callback,
3954 ev_arg, ev_fd, ev_res, ev_events.
3955
3956 =item * Avoid using ev_flags and the EVLIST_*-macros, while it is
3957 maintained by libev, it does not work exactly the same way as in libevent (consider
3958 it a private API).
3959
3960 =item * Priorities are not currently supported. Initialising priorities
3961 will fail and all watchers will have the same priority, even though there
3962 is an ev_pri field.
3963
3964 =item * In libevent, the last base created gets the signals, in libev, the
3965 base that registered the signal gets the signals.
3966
3967 =item * Other members are not supported.
3968
3969 =item * The libev emulation is I<not> ABI compatible to libevent, you need
3970 to use the libev header file and library.
3971
3972 =back
3973
3974 =head1 C++ SUPPORT
3975
3976 =head2 C API
3977
3978 The normal C API should work fine when used from C++: both ev.h and the
3979 libev sources can be compiled as C++. Therefore, code that uses the C API
3980 will work fine.
3981
3982 Proper exception specifications might have to be added to callbacks passed
3983 to libev: exceptions may be thrown only from watcher callbacks, all other
3984 callbacks (allocator, syserr, loop acquire/release and periodic reschedule
3985 callbacks) must not throw exceptions, and might need a C<noexcept>
3986 specification. If you have code that needs to be compiled as both C and
3987 C++ you can use the C<EV_NOEXCEPT> macro for this:
3988
3989 static void
3990 fatal_error (const char *msg) EV_NOEXCEPT
3991 {
3992 perror (msg);
3993 abort ();
3994 }
3995
3996 ...
3997 ev_set_syserr_cb (fatal_error);
3998
3999 The only API functions that can currently throw exceptions are C<ev_run>,
4000 C<ev_invoke>, C<ev_invoke_pending> and C<ev_loop_destroy> (the latter
4001 because it runs cleanup watchers).
4002
4003 Throwing exceptions in watcher callbacks is only supported if libev itself
4004 is compiled with a C++ compiler or your C and C++ environments allow
4005 throwing exceptions through C libraries (most do).
4006
4007 =head2 C++ API
4008
4009 Libev comes with some simplistic wrapper classes for C++ that mainly allow
4010 you to use some convenience methods to start/stop watchers and also change
4011 the callback model to a model using method callbacks on objects.
4012
4013 To use it,
4014
4015 #include <ev++.h>
4016
4017 This automatically includes F<ev.h> and puts all of its definitions (many
4018 of them macros) into the global namespace. All C++ specific things are
4019 put into the C<ev> namespace. It should support all the same embedding
4020 options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
4021
4022 Care has been taken to keep the overhead low. The only data member the C++
4023 classes add (compared to plain C-style watchers) is the event loop pointer
4024 that the watcher is associated with (or no additional members at all if
4025 you disable C<EV_MULTIPLICITY> when embedding libev).
4026
4027 Currently, functions, static and non-static member functions and classes
4028 with C<operator ()> can be used as callbacks. Other types should be easy
4029 to add as long as they only need one additional pointer for context. If
4030 you need support for other types of functors please contact the author
4031 (preferably after implementing it).
4032
4033 For all this to work, your C++ compiler either has to use the same calling
4034 conventions as your C compiler (for static member functions), or you have
4035 to embed libev and compile libev itself as C++.
4036
4037 Here is a list of things available in the C<ev> namespace:
4038
4039 =over 4
4040
4041 =item C<ev::READ>, C<ev::WRITE> etc.
4042
4043 These are just enum values with the same values as the C<EV_READ> etc.
4044 macros from F<ev.h>.
4045
4046 =item C<ev::tstamp>, C<ev::now>
4047
4048 Aliases to the same types/functions as with the C<ev_> prefix.
4049
4050 =item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
4051
4052 For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
4053 the same name in the C<ev> namespace, with the exception of C<ev_signal>
4054 which is called C<ev::sig> to avoid clashes with the C<signal> macro
4055 defined by many implementations.
4056
4057 All of those classes have these methods:
4058
4059 =over 4
4060
4061 =item ev::TYPE::TYPE ()
4062
4063 =item ev::TYPE::TYPE (loop)
4064
4065 =item ev::TYPE::~TYPE
4066
4067 The constructor (optionally) takes an event loop to associate the watcher
4068 with. If it is omitted, it will use C<EV_DEFAULT>.
4069
4070 The constructor calls C<ev_init> for you, which means you have to call the
4071 C<set> method before starting it.
4072
4073 It will not set a callback, however: You have to call the templated C<set>
4074 method to set a callback before you can start the watcher.
4075
4076 (The reason why you have to use a method is a limitation in C++ which does
4077 not allow explicit template arguments for constructors).
4078
4079 The destructor automatically stops the watcher if it is active.
4080
4081 =item w->set<class, &class::method> (object *)
4082
4083 This method sets the callback method to call. The method has to have a
4084 signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
4085 first argument and the C<revents> as second. The object must be given as
4086 parameter and is stored in the C<data> member of the watcher.
4087
4088 This method synthesizes efficient thunking code to call your method from
4089 the C callback that libev requires. If your compiler can inline your
4090 callback (i.e. it is visible to it at the place of the C<set> call and
4091 your compiler is good :), then the method will be fully inlined into the
4092 thunking function, making it as fast as a direct C callback.
4093
4094 Example: simple class declaration and watcher initialisation
4095
4096 struct myclass
4097 {
4098 void io_cb (ev::io &w, int revents) { }
4099 }
4100
4101 myclass obj;
4102 ev::io iow;
4103 iow.set <myclass, &myclass::io_cb> (&obj);
4104
4105 =item w->set (object *)
4106
4107 This is a variation of a method callback - leaving out the method to call
4108 will default the method to C<operator ()>, which makes it possible to use
4109 functor objects without having to manually specify the C<operator ()> all
4110 the time. Incidentally, you can then also leave out the template argument
4111 list.
4112
4113 The C<operator ()> method prototype must be C<void operator ()(watcher &w,
4114 int revents)>.
4115
4116 See the method-C<set> above for more details.
4117
4118 Example: use a functor object as callback.
4119
4120 struct myfunctor
4121 {
4122 void operator() (ev::io &w, int revents)
4123 {
4124 ...
4125 }
4126 }
4127
4128 myfunctor f;
4129
4130 ev::io w;
4131 w.set (&f);
4132
4133 =item w->set<function> (void *data = 0)
4134
4135 Also sets a callback, but uses a static method or plain function as
4136 callback. The optional C<data> argument will be stored in the watcher's
4137 C<data> member and is free for you to use.
4138
4139 The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
4140
4141 See the method-C<set> above for more details.
4142
4143 Example: Use a plain function as callback.
4144
4145 static void io_cb (ev::io &w, int revents) { }
4146 iow.set <io_cb> ();
4147
4148 =item w->set (loop)
4149
4150 Associates a different C<struct ev_loop> with this watcher. You can only
4151 do this when the watcher is inactive (and not pending either).
4152
4153 =item w->set ([arguments])
4154
4155 Basically the same as C<ev_TYPE_set> (except for C<ev::embed> watchers>),
4156 with the same arguments. Either this method or a suitable start method
4157 must be called at least once. Unlike the C counterpart, an active watcher
4158 gets automatically stopped and restarted when reconfiguring it with this
4159 method.
4160
4161 For C<ev::embed> watchers this method is called C<set_embed>, to avoid
4162 clashing with the C<set (loop)> method.
4163
4164 =item w->start ()
4165
4166 Starts the watcher. Note that there is no C<loop> argument, as the
4167 constructor already stores the event loop.
4168
4169 =item w->start ([arguments])
4170
4171 Instead of calling C<set> and C<start> methods separately, it is often
4172 convenient to wrap them in one call. Uses the same type of arguments as
4173 the configure C<set> method of the watcher.
4174
4175 =item w->stop ()
4176
4177 Stops the watcher if it is active. Again, no C<loop> argument.
4178
4179 =item w->again () (C<ev::timer>, C<ev::periodic> only)
4180
4181 For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
4182 C<ev_TYPE_again> function.
4183
4184 =item w->sweep () (C<ev::embed> only)
4185
4186 Invokes C<ev_embed_sweep>.
4187
4188 =item w->update () (C<ev::stat> only)
4189
4190 Invokes C<ev_stat_stat>.
4191
4192 =back
4193
4194 =back
4195
4196 Example: Define a class with two I/O and idle watchers, start the I/O
4197 watchers in the constructor.
4198
4199 class myclass
4200 {
4201 ev::io io ; void io_cb (ev::io &w, int revents);
4202 ev::io io2 ; void io2_cb (ev::io &w, int revents);
4203 ev::idle idle; void idle_cb (ev::idle &w, int revents);
4204
4205 myclass (int fd)
4206 {
4207 io .set <myclass, &myclass::io_cb > (this);
4208 io2 .set <myclass, &myclass::io2_cb > (this);
4209 idle.set <myclass, &myclass::idle_cb> (this);
4210
4211 io.set (fd, ev::WRITE); // configure the watcher
4212 io.start (); // start it whenever convenient
4213
4214 io2.start (fd, ev::READ); // set + start in one call
4215 }
4216 };
4217
4218
4219 =head1 OTHER LANGUAGE BINDINGS
4220
4221 Libev does not offer other language bindings itself, but bindings for a
4222 number of languages exist in the form of third-party packages. If you know
4223 any interesting language binding in addition to the ones listed here, drop
4224 me a note.
4225
4226 =over 4
4227
4228 =item Perl
4229
4230 The EV module implements the full libev API and is actually used to test
4231 libev. EV is developed together with libev. Apart from the EV core module,
4232 there are additional modules that implement libev-compatible interfaces
4233 to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
4234 C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
4235 and C<EV::Glib>).
4236
4237 It can be found and installed via CPAN, its homepage is at
4238 L<http://software.schmorp.de/pkg/EV>.
4239
4240 =item Python
4241
4242 Python bindings can be found at L<http://code.google.com/p/pyev/>. It
4243 seems to be quite complete and well-documented.
4244
4245 =item Ruby
4246
4247 Tony Arcieri has written a ruby extension that offers access to a subset
4248 of the libev API and adds file handle abstractions, asynchronous DNS and
4249 more on top of it. It can be found via gem servers. Its homepage is at
4250 L<http://rev.rubyforge.org/>.
4251
4252 Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
4253 makes rev work even on mingw.
4254
4255 =item Haskell
4256
4257 A haskell binding to libev is available at
4258 L<http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
4259
4260 =item D
4261
4262 Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
4263 be found at L<http://www.llucax.com.ar/proj/ev.d/index.html>.
4264
4265 =item Ocaml
4266
4267 Erkki Seppala has written Ocaml bindings for libev, to be found at
4268 L<http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
4269
4270 =item Lua
4271
4272 Brian Maher has written a partial interface to libev for lua (at the
4273 time of this writing, only C<ev_io> and C<ev_timer>), to be found at
4274 L<http://github.com/brimworks/lua-ev>.
4275
4276 =item Javascript
4277
4278 Node.js (L<http://nodejs.org>) uses libev as the underlying event library.
4279
4280 =item Others
4281
4282 There are others, and I stopped counting.
4283
4284 =back
4285
4286
4287 =head1 MACRO MAGIC
4288
4289 Libev can be compiled with a variety of options, the most fundamental
4290 of which is C<EV_MULTIPLICITY>. This option determines whether (most)
4291 functions and callbacks have an initial C<struct ev_loop *> argument.
4292
4293 To make it easier to write programs that cope with either variant, the
4294 following macros are defined:
4295
4296 =over 4
4297
4298 =item C<EV_A>, C<EV_A_>
4299
4300 This provides the loop I<argument> for functions, if one is required ("ev
4301 loop argument"). The C<EV_A> form is used when this is the sole argument,
4302 C<EV_A_> is used when other arguments are following. Example:
4303
4304 ev_unref (EV_A);
4305 ev_timer_add (EV_A_ watcher);
4306 ev_run (EV_A_ 0);
4307
4308 It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
4309 which is often provided by the following macro.
4310
4311 =item C<EV_P>, C<EV_P_>
4312
4313 This provides the loop I<parameter> for functions, if one is required ("ev
4314 loop parameter"). The C<EV_P> form is used when this is the sole parameter,
4315 C<EV_P_> is used when other parameters are following. Example:
4316
4317 // this is how ev_unref is being declared
4318 static void ev_unref (EV_P);
4319
4320 // this is how you can declare your typical callback
4321 static void cb (EV_P_ ev_timer *w, int revents)
4322
4323 It declares a parameter C<loop> of type C<struct ev_loop *>, quite
4324 suitable for use with C<EV_A>.
4325
4326 =item C<EV_DEFAULT>, C<EV_DEFAULT_>
4327
4328 Similar to the other two macros, this gives you the value of the default
4329 loop, if multiple loops are supported ("ev loop default"). The default loop
4330 will be initialised if it isn't already initialised.
4331
4332 For non-multiplicity builds, these macros do nothing, so you always have
4333 to initialise the loop somewhere.
4334
4335 =item C<EV_DEFAULT_UC>, C<EV_DEFAULT_UC_>
4336
4337 Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
4338 default loop has been initialised (C<UC> == unchecked). Their behaviour
4339 is undefined when the default loop has not been initialised by a previous
4340 execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
4341
4342 It is often prudent to use C<EV_DEFAULT> when initialising the first
4343 watcher in a function but use C<EV_DEFAULT_UC> afterwards.
4344
4345 =back
4346
4347 Example: Declare and initialise a check watcher, utilising the above
4348 macros so it will work regardless of whether multiple loops are supported
4349 or not.
4350
4351 static void
4352 check_cb (EV_P_ ev_timer *w, int revents)
4353 {
4354 ev_check_stop (EV_A_ w);
4355 }
4356
4357 ev_check check;
4358 ev_check_init (&check, check_cb);
4359 ev_check_start (EV_DEFAULT_ &check);
4360 ev_run (EV_DEFAULT_ 0);
4361
4362 =head1 EMBEDDING
4363
4364 Libev can (and often is) directly embedded into host
4365 applications. Examples of applications that embed it include the Deliantra
4366 Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
4367 and rxvt-unicode.
4368
4369 The goal is to enable you to just copy the necessary files into your
4370 source directory without having to change even a single line in them, so
4371 you can easily upgrade by simply copying (or having a checked-out copy of
4372 libev somewhere in your source tree).
4373
4374 =head2 FILESETS
4375
4376 Depending on what features you need you need to include one or more sets of files
4377 in your application.
4378
4379 =head3 CORE EVENT LOOP
4380
4381 To include only the libev core (all the C<ev_*> functions), with manual
4382 configuration (no autoconf):
4383
4384 #define EV_STANDALONE 1
4385 #include "ev.c"
4386
4387 This will automatically include F<ev.h>, too, and should be done in a
4388 single C source file only to provide the function implementations. To use
4389 it, do the same for F<ev.h> in all files wishing to use this API (best
4390 done by writing a wrapper around F<ev.h> that you can include instead and
4391 where you can put other configuration options):
4392
4393 #define EV_STANDALONE 1
4394 #include "ev.h"
4395
4396 Both header files and implementation files can be compiled with a C++
4397 compiler (at least, that's a stated goal, and breakage will be treated
4398 as a bug).
4399
4400 You need the following files in your source tree, or in a directory
4401 in your include path (e.g. in libev/ when using -Ilibev):
4402
4403 ev.h
4404 ev.c
4405 ev_vars.h
4406 ev_wrap.h
4407
4408 ev_win32.c required on win32 platforms only
4409
4410 ev_select.c only when select backend is enabled
4411 ev_poll.c only when poll backend is enabled
4412 ev_epoll.c only when the epoll backend is enabled
4413 ev_kqueue.c only when the kqueue backend is enabled
4414 ev_port.c only when the solaris port backend is enabled
4415
4416 F<ev.c> includes the backend files directly when enabled, so you only need
4417 to compile this single file.
4418
4419 =head3 LIBEVENT COMPATIBILITY API
4420
4421 To include the libevent compatibility API, also include:
4422
4423 #include "event.c"
4424
4425 in the file including F<ev.c>, and:
4426
4427 #include "event.h"
4428
4429 in the files that want to use the libevent API. This also includes F<ev.h>.
4430
4431 You need the following additional files for this:
4432
4433 event.h
4434 event.c
4435
4436 =head3 AUTOCONF SUPPORT
4437
4438 Instead of using C<EV_STANDALONE=1> and providing your configuration in
4439 whatever way you want, you can also C<m4_include([libev.m4])> in your
4440 F<configure.ac> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
4441 include F<config.h> and configure itself accordingly.
4442
4443 For this of course you need the m4 file:
4444
4445 libev.m4
4446
4447 =head2 PREPROCESSOR SYMBOLS/MACROS
4448
4449 Libev can be configured via a variety of preprocessor symbols you have to
4450 define before including (or compiling) any of its files. The default in
4451 the absence of autoconf is documented for every option.
4452
4453 Symbols marked with "(h)" do not change the ABI, and can have different
4454 values when compiling libev vs. including F<ev.h>, so it is permissible
4455 to redefine them before including F<ev.h> without breaking compatibility
4456 to a compiled library. All other symbols change the ABI, which means all
4457 users of libev and the libev code itself must be compiled with compatible
4458 settings.
4459
4460 =over 4
4461
4462 =item EV_COMPAT3 (h)
4463
4464 Backwards compatibility is a major concern for libev. This is why this
4465 release of libev comes with wrappers for the functions and symbols that
4466 have been renamed between libev version 3 and 4.
4467
4468 You can disable these wrappers (to test compatibility with future
4469 versions) by defining C<EV_COMPAT3> to C<0> when compiling your
4470 sources. This has the additional advantage that you can drop the C<struct>
4471 from C<struct ev_loop> declarations, as libev will provide an C<ev_loop>
4472 typedef in that case.
4473
4474 In some future version, the default for C<EV_COMPAT3> will become C<0>,
4475 and in some even more future version the compatibility code will be
4476 removed completely.
4477
4478 =item EV_STANDALONE (h)
4479
4480 Must always be C<1> if you do not use autoconf configuration, which
4481 keeps libev from including F<config.h>, and it also defines dummy
4482 implementations for some libevent functions (such as logging, which is not
4483 supported). It will also not define any of the structs usually found in
4484 F<event.h> that are not directly supported by the libev core alone.
4485
4486 In standalone mode, libev will still try to automatically deduce the
4487 configuration, but has to be more conservative.
4488
4489 =item EV_USE_FLOOR
4490
4491 If defined to be C<1>, libev will use the C<floor ()> function for its
4492 periodic reschedule calculations, otherwise libev will fall back on a
4493 portable (slower) implementation. If you enable this, you usually have to
4494 link against libm or something equivalent. Enabling this when the C<floor>
4495 function is not available will fail, so the safe default is to not enable
4496 this.
4497
4498 =item EV_USE_MONOTONIC
4499
4500 If defined to be C<1>, libev will try to detect the availability of the
4501 monotonic clock option at both compile time and runtime. Otherwise no
4502 use of the monotonic clock option will be attempted. If you enable this,
4503 you usually have to link against librt or something similar. Enabling it
4504 when the functionality isn't available is safe, though, although you have
4505 to make sure you link against any libraries where the C<clock_gettime>
4506 function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
4507
4508 =item EV_USE_REALTIME
4509
4510 If defined to be C<1>, libev will try to detect the availability of the
4511 real-time clock option at compile time (and assume its availability
4512 at runtime if successful). Otherwise no use of the real-time clock
4513 option will be attempted. This effectively replaces C<gettimeofday>
4514 by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
4515 correctness. See the note about libraries in the description of
4516 C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
4517 C<EV_USE_CLOCK_SYSCALL>.
4518
4519 =item EV_USE_CLOCK_SYSCALL
4520
4521 If defined to be C<1>, libev will try to use a direct syscall instead
4522 of calling the system-provided C<clock_gettime> function. This option
4523 exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
4524 unconditionally pulls in C<libpthread>, slowing down single-threaded
4525 programs needlessly. Using a direct syscall is slightly slower (in
4526 theory), because no optimised vdso implementation can be used, but avoids
4527 the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
4528 higher, as it simplifies linking (no need for C<-lrt>).
4529
4530 =item EV_USE_NANOSLEEP
4531
4532 If defined to be C<1>, libev will assume that C<nanosleep ()> is available
4533 and will use it for delays. Otherwise it will use C<select ()>.
4534
4535 =item EV_USE_EVENTFD
4536
4537 If defined to be C<1>, then libev will assume that C<eventfd ()> is
4538 available and will probe for kernel support at runtime. This will improve
4539 C<ev_signal> and C<ev_async> performance and reduce resource consumption.
4540 If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
4541 2.7 or newer, otherwise disabled.
4542
4543 =item EV_USE_SELECT
4544
4545 If undefined or defined to be C<1>, libev will compile in support for the
4546 C<select>(2) backend. No attempt at auto-detection will be done: if no
4547 other method takes over, select will be it. Otherwise the select backend
4548 will not be compiled in.
4549
4550 =item EV_SELECT_USE_FD_SET
4551
4552 If defined to C<1>, then the select backend will use the system C<fd_set>
4553 structure. This is useful if libev doesn't compile due to a missing
4554 C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
4555 on exotic systems. This usually limits the range of file descriptors to
4556 some low limit such as 1024 or might have other limitations (winsocket
4557 only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
4558 configures the maximum size of the C<fd_set>.
4559
4560 =item EV_SELECT_IS_WINSOCKET
4561
4562 When defined to C<1>, the select backend will assume that
4563 select/socket/connect etc. don't understand file descriptors but
4564 wants osf handles on win32 (this is the case when the select to
4565 be used is the winsock select). This means that it will call
4566 C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
4567 it is assumed that all these functions actually work on fds, even
4568 on win32. Should not be defined on non-win32 platforms.
4569
4570 =item EV_FD_TO_WIN32_HANDLE(fd)
4571
4572 If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
4573 file descriptors to socket handles. When not defining this symbol (the
4574 default), then libev will call C<_get_osfhandle>, which is usually
4575 correct. In some cases, programs use their own file descriptor management,
4576 in which case they can provide this function to map fds to socket handles.
4577
4578 =item EV_WIN32_HANDLE_TO_FD(handle)
4579
4580 If C<EV_SELECT_IS_WINSOCKET> then libev maps handles to file descriptors
4581 using the standard C<_open_osfhandle> function. For programs implementing
4582 their own fd to handle mapping, overwriting this function makes it easier
4583 to do so. This can be done by defining this macro to an appropriate value.
4584
4585 =item EV_WIN32_CLOSE_FD(fd)
4586
4587 If programs implement their own fd to handle mapping on win32, then this
4588 macro can be used to override the C<close> function, useful to unregister
4589 file descriptors again. Note that the replacement function has to close
4590 the underlying OS handle.
4591
4592 =item EV_USE_WSASOCKET
4593
4594 If defined to be C<1>, libev will use C<WSASocket> to create its internal
4595 communication socket, which works better in some environments. Otherwise,
4596 the normal C<socket> function will be used, which works better in other
4597 environments.
4598
4599 =item EV_USE_POLL
4600
4601 If defined to be C<1>, libev will compile in support for the C<poll>(2)
4602 backend. Otherwise it will be enabled on non-win32 platforms. It
4603 takes precedence over select.
4604
4605 =item EV_USE_EPOLL
4606
4607 If defined to be C<1>, libev will compile in support for the Linux
4608 C<epoll>(7) backend. Its availability will be detected at runtime,
4609 otherwise another method will be used as fallback. This is the preferred
4610 backend for GNU/Linux systems. If undefined, it will be enabled if the
4611 headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
4612
4613 =item EV_USE_KQUEUE
4614
4615 If defined to be C<1>, libev will compile in support for the BSD style
4616 C<kqueue>(2) backend. Its actual availability will be detected at runtime,
4617 otherwise another method will be used as fallback. This is the preferred
4618 backend for BSD and BSD-like systems, although on most BSDs kqueue only
4619 supports some types of fds correctly (the only platform we found that
4620 supports ptys for example was NetBSD), so kqueue might be compiled in, but
4621 not be used unless explicitly requested. The best way to use it is to find
4622 out whether kqueue supports your type of fd properly and use an embedded
4623 kqueue loop.
4624
4625 =item EV_USE_PORT
4626
4627 If defined to be C<1>, libev will compile in support for the Solaris
4628 10 port style backend. Its availability will be detected at runtime,
4629 otherwise another method will be used as fallback. This is the preferred
4630 backend for Solaris 10 systems.
4631
4632 =item EV_USE_DEVPOLL
4633
4634 Reserved for future expansion, works like the USE symbols above.
4635
4636 =item EV_USE_INOTIFY
4637
4638 If defined to be C<1>, libev will compile in support for the Linux inotify
4639 interface to speed up C<ev_stat> watchers. Its actual availability will
4640 be detected at runtime. If undefined, it will be enabled if the headers
4641 indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
4642
4643 =item EV_NO_SMP
4644
4645 If defined to be C<1>, libev will assume that memory is always coherent
4646 between threads, that is, threads can be used, but threads never run on
4647 different cpus (or different cpu cores). This reduces dependencies
4648 and makes libev faster.
4649
4650 =item EV_NO_THREADS
4651
4652 If defined to be C<1>, libev will assume that it will never be called from
4653 different threads (that includes signal handlers), which is a stronger
4654 assumption than C<EV_NO_SMP>, above. This reduces dependencies and makes
4655 libev faster.
4656
4657 =item EV_ATOMIC_T
4658
4659 Libev requires an integer type (suitable for storing C<0> or C<1>) whose
4660 access is atomic with respect to other threads or signal contexts. No
4661 such type is easily found in the C language, so you can provide your own
4662 type that you know is safe for your purposes. It is used both for signal
4663 handler "locking" as well as for signal and thread safety in C<ev_async>
4664 watchers.
4665
4666 In the absence of this define, libev will use C<sig_atomic_t volatile>
4667 (from F<signal.h>), which is usually good enough on most platforms.
4668
4669 =item EV_H (h)
4670
4671 The name of the F<ev.h> header file used to include it. The default if
4672 undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
4673 used to virtually rename the F<ev.h> header file in case of conflicts.
4674
4675 =item EV_CONFIG_H (h)
4676
4677 If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
4678 F<ev.c>'s idea of where to find the F<config.h> file, similarly to
4679 C<EV_H>, above.
4680
4681 =item EV_EVENT_H (h)
4682
4683 Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
4684 of how the F<event.h> header can be found, the default is C<"event.h">.
4685
4686 =item EV_PROTOTYPES (h)
4687
4688 If defined to be C<0>, then F<ev.h> will not define any function
4689 prototypes, but still define all the structs and other symbols. This is
4690 occasionally useful if you want to provide your own wrapper functions
4691 around libev functions.
4692
4693 =item EV_MULTIPLICITY
4694
4695 If undefined or defined to C<1>, then all event-loop-specific functions
4696 will have the C<struct ev_loop *> as first argument, and you can create
4697 additional independent event loops. Otherwise there will be no support
4698 for multiple event loops and there is no first event loop pointer
4699 argument. Instead, all functions act on the single default loop.
4700
4701 Note that C<EV_DEFAULT> and C<EV_DEFAULT_> will no longer provide a
4702 default loop when multiplicity is switched off - you always have to
4703 initialise the loop manually in this case.
4704
4705 =item EV_MINPRI
4706
4707 =item EV_MAXPRI
4708
4709 The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
4710 C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
4711 provide for more priorities by overriding those symbols (usually defined
4712 to be C<-2> and C<2>, respectively).
4713
4714 When doing priority-based operations, libev usually has to linearly search
4715 all the priorities, so having many of them (hundreds) uses a lot of space
4716 and time, so using the defaults of five priorities (-2 .. +2) is usually
4717 fine.
4718
4719 If your embedding application does not need any priorities, defining these
4720 both to C<0> will save some memory and CPU.
4721
4722 =item EV_PERIODIC_ENABLE, EV_IDLE_ENABLE, EV_EMBED_ENABLE, EV_STAT_ENABLE,
4723 EV_PREPARE_ENABLE, EV_CHECK_ENABLE, EV_FORK_ENABLE, EV_SIGNAL_ENABLE,
4724 EV_ASYNC_ENABLE, EV_CHILD_ENABLE.
4725
4726 If undefined or defined to be C<1> (and the platform supports it), then
4727 the respective watcher type is supported. If defined to be C<0>, then it
4728 is not. Disabling watcher types mainly saves code size.
4729
4730 =item EV_FEATURES
4731
4732 If you need to shave off some kilobytes of code at the expense of some
4733 speed (but with the full API), you can define this symbol to request
4734 certain subsets of functionality. The default is to enable all features
4735 that can be enabled on the platform.
4736
4737 A typical way to use this symbol is to define it to C<0> (or to a bitset
4738 with some broad features you want) and then selectively re-enable
4739 additional parts you want, for example if you want everything minimal,
4740 but multiple event loop support, async and child watchers and the poll
4741 backend, use this:
4742
4743 #define EV_FEATURES 0
4744 #define EV_MULTIPLICITY 1
4745 #define EV_USE_POLL 1
4746 #define EV_CHILD_ENABLE 1
4747 #define EV_ASYNC_ENABLE 1
4748
4749 The actual value is a bitset, it can be a combination of the following
4750 values (by default, all of these are enabled):
4751
4752 =over 4
4753
4754 =item C<1> - faster/larger code
4755
4756 Use larger code to speed up some operations.
4757
4758 Currently this is used to override some inlining decisions (enlarging the
4759 code size by roughly 30% on amd64).
4760
4761 When optimising for size, use of compiler flags such as C<-Os> with
4762 gcc is recommended, as well as C<-DNDEBUG>, as libev contains a number of
4763 assertions.
4764
4765 The default is off when C<__OPTIMIZE_SIZE__> is defined by your compiler
4766 (e.g. gcc with C<-Os>).
4767
4768 =item C<2> - faster/larger data structures
4769
4770 Replaces the small 2-heap for timer management by a faster 4-heap, larger
4771 hash table sizes and so on. This will usually further increase code size
4772 and can additionally have an effect on the size of data structures at
4773 runtime.
4774
4775 The default is off when C<__OPTIMIZE_SIZE__> is defined by your compiler
4776 (e.g. gcc with C<-Os>).
4777
4778 =item C<4> - full API configuration
4779
4780 This enables priorities (sets C<EV_MAXPRI>=2 and C<EV_MINPRI>=-2), and
4781 enables multiplicity (C<EV_MULTIPLICITY>=1).
4782
4783 =item C<8> - full API
4784
4785 This enables a lot of the "lesser used" API functions. See C<ev.h> for
4786 details on which parts of the API are still available without this
4787 feature, and do not complain if this subset changes over time.
4788
4789 =item C<16> - enable all optional watcher types
4790
4791 Enables all optional watcher types. If you want to selectively enable
4792 only some watcher types other than I/O and timers (e.g. prepare,
4793 embed, async, child...) you can enable them manually by defining
4794 C<EV_watchertype_ENABLE> to C<1> instead.
4795
4796 =item C<32> - enable all backends
4797
4798 This enables all backends - without this feature, you need to enable at
4799 least one backend manually (C<EV_USE_SELECT> is a good choice).
4800
4801 =item C<64> - enable OS-specific "helper" APIs
4802
4803 Enable inotify, eventfd, signalfd and similar OS-specific helper APIs by
4804 default.
4805
4806 =back
4807
4808 Compiling with C<gcc -Os -DEV_STANDALONE -DEV_USE_EPOLL=1 -DEV_FEATURES=0>
4809 reduces the compiled size of libev from 24.7Kb code/2.8Kb data to 6.5Kb
4810 code/0.3Kb data on my GNU/Linux amd64 system, while still giving you I/O
4811 watchers, timers and monotonic clock support.
4812
4813 With an intelligent-enough linker (gcc+binutils are intelligent enough
4814 when you use C<-Wl,--gc-sections -ffunction-sections>) functions unused by
4815 your program might be left out as well - a binary starting a timer and an
4816 I/O watcher then might come out at only 5Kb.
4817
4818 =item EV_API_STATIC
4819
4820 If this symbol is defined (by default it is not), then all identifiers
4821 will have static linkage. This means that libev will not export any
4822 identifiers, and you cannot link against libev anymore. This can be useful
4823 when you embed libev, only want to use libev functions in a single file,
4824 and do not want its identifiers to be visible.
4825
4826 To use this, define C<EV_API_STATIC> and include F<ev.c> in the file that
4827 wants to use libev.
4828
4829 This option only works when libev is compiled with a C compiler, as C++
4830 doesn't support the required declaration syntax.
4831
4832 =item EV_AVOID_STDIO
4833
4834 If this is set to C<1> at compiletime, then libev will avoid using stdio
4835 functions (printf, scanf, perror etc.). This will increase the code size
4836 somewhat, but if your program doesn't otherwise depend on stdio and your
4837 libc allows it, this avoids linking in the stdio library which is quite
4838 big.
4839
4840 Note that error messages might become less precise when this option is
4841 enabled.
4842
4843 =item EV_NSIG
4844
4845 The highest supported signal number, +1 (or, the number of
4846 signals): Normally, libev tries to deduce the maximum number of signals
4847 automatically, but sometimes this fails, in which case it can be
4848 specified. Also, using a lower number than detected (C<32> should be
4849 good for about any system in existence) can save some memory, as libev
4850 statically allocates some 12-24 bytes per signal number.
4851
4852 =item EV_PID_HASHSIZE
4853
4854 C<ev_child> watchers use a small hash table to distribute workload by
4855 pid. The default size is C<16> (or C<1> with C<EV_FEATURES> disabled),
4856 usually more than enough. If you need to manage thousands of children you
4857 might want to increase this value (I<must> be a power of two).
4858
4859 =item EV_INOTIFY_HASHSIZE
4860
4861 C<ev_stat> watchers use a small hash table to distribute workload by
4862 inotify watch id. The default size is C<16> (or C<1> with C<EV_FEATURES>
4863 disabled), usually more than enough. If you need to manage thousands of
4864 C<ev_stat> watchers you might want to increase this value (I<must> be a
4865 power of two).
4866
4867 =item EV_USE_4HEAP
4868
4869 Heaps are not very cache-efficient. To improve the cache-efficiency of the
4870 timer and periodics heaps, libev uses a 4-heap when this symbol is defined
4871 to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
4872 faster performance with many (thousands) of watchers.
4873
4874 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4875 will be C<0>.
4876
4877 =item EV_HEAP_CACHE_AT
4878
4879 Heaps are not very cache-efficient. To improve the cache-efficiency of the
4880 timer and periodics heaps, libev can cache the timestamp (I<at>) within
4881 the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
4882 which uses 8-12 bytes more per watcher and a few hundred bytes more code,
4883 but avoids random read accesses on heap changes. This improves performance
4884 noticeably with many (hundreds) of watchers.
4885
4886 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4887 will be C<0>.
4888
4889 =item EV_VERIFY
4890
4891 Controls how much internal verification (see C<ev_verify ()>) will
4892 be done: If set to C<0>, no internal verification code will be compiled
4893 in. If set to C<1>, then verification code will be compiled in, but not
4894 called. If set to C<2>, then the internal verification code will be
4895 called once per loop, which can slow down libev. If set to C<3>, then the
4896 verification code will be called very frequently, which will slow down
4897 libev considerably.
4898
4899 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4900 will be C<0>.
4901
4902 =item EV_COMMON
4903
4904 By default, all watchers have a C<void *data> member. By redefining
4905 this macro to something else you can include more and other types of
4906 members. You have to define it each time you include one of the files,
4907 though, and it must be identical each time.
4908
4909 For example, the perl EV module uses something like this:
4910
4911 #define EV_COMMON \
4912 SV *self; /* contains this struct */ \
4913 SV *cb_sv, *fh /* note no trailing ";" */
4914
4915 =item EV_CB_DECLARE (type)
4916
4917 =item EV_CB_INVOKE (watcher, revents)
4918
4919 =item ev_set_cb (ev, cb)
4920
4921 Can be used to change the callback member declaration in each watcher,
4922 and the way callbacks are invoked and set. Must expand to a struct member
4923 definition and a statement, respectively. See the F<ev.h> header file for
4924 their default definitions. One possible use for overriding these is to
4925 avoid the C<struct ev_loop *> as first argument in all cases, or to use
4926 method calls instead of plain function calls in C++.
4927
4928 =back
4929
4930 =head2 EXPORTED API SYMBOLS
4931
4932 If you need to re-export the API (e.g. via a DLL) and you need a list of
4933 exported symbols, you can use the provided F<Symbol.*> files which list
4934 all public symbols, one per line:
4935
4936 Symbols.ev for libev proper
4937 Symbols.event for the libevent emulation
4938
4939 This can also be used to rename all public symbols to avoid clashes with
4940 multiple versions of libev linked together (which is obviously bad in
4941 itself, but sometimes it is inconvenient to avoid this).
4942
4943 A sed command like this will create wrapper C<#define>'s that you need to
4944 include before including F<ev.h>:
4945
4946 <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
4947
4948 This would create a file F<wrap.h> which essentially looks like this:
4949
4950 #define ev_backend myprefix_ev_backend
4951 #define ev_check_start myprefix_ev_check_start
4952 #define ev_check_stop myprefix_ev_check_stop
4953 ...
4954
4955 =head2 EXAMPLES
4956
4957 For a real-world example of a program the includes libev
4958 verbatim, you can have a look at the EV perl module
4959 (L<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
4960 the F<libev/> subdirectory and includes them in the F<EV/EVAPI.h> (public
4961 interface) and F<EV.xs> (implementation) files. Only the F<EV.xs> file
4962 will be compiled. It is pretty complex because it provides its own header
4963 file.
4964
4965 The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
4966 that everybody includes and which overrides some configure choices:
4967
4968 #define EV_FEATURES 8
4969 #define EV_USE_SELECT 1
4970 #define EV_PREPARE_ENABLE 1
4971 #define EV_IDLE_ENABLE 1
4972 #define EV_SIGNAL_ENABLE 1
4973 #define EV_CHILD_ENABLE 1
4974 #define EV_USE_STDEXCEPT 0
4975 #define EV_CONFIG_H <config.h>
4976
4977 #include "ev++.h"
4978
4979 And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
4980
4981 #include "ev_cpp.h"
4982 #include "ev.c"
4983
4984 =head1 INTERACTION WITH OTHER PROGRAMS, LIBRARIES OR THE ENVIRONMENT
4985
4986 =head2 THREADS AND COROUTINES
4987
4988 =head3 THREADS
4989
4990 All libev functions are reentrant and thread-safe unless explicitly
4991 documented otherwise, but libev implements no locking itself. This means
4992 that you can use as many loops as you want in parallel, as long as there
4993 are no concurrent calls into any libev function with the same loop
4994 parameter (C<ev_default_*> calls have an implicit default loop parameter,
4995 of course): libev guarantees that different event loops share no data
4996 structures that need any locking.
4997
4998 Or to put it differently: calls with different loop parameters can be done
4999 concurrently from multiple threads, calls with the same loop parameter
5000 must be done serially (but can be done from different threads, as long as
5001 only one thread ever is inside a call at any point in time, e.g. by using
5002 a mutex per loop).
5003
5004 Specifically to support threads (and signal handlers), libev implements
5005 so-called C<ev_async> watchers, which allow some limited form of
5006 concurrency on the same event loop, namely waking it up "from the
5007 outside".
5008
5009 If you want to know which design (one loop, locking, or multiple loops
5010 without or something else still) is best for your problem, then I cannot
5011 help you, but here is some generic advice:
5012
5013 =over 4
5014
5015 =item * most applications have a main thread: use the default libev loop
5016 in that thread, or create a separate thread running only the default loop.
5017
5018 This helps integrating other libraries or software modules that use libev
5019 themselves and don't care/know about threading.
5020
5021 =item * one loop per thread is usually a good model.
5022
5023 Doing this is almost never wrong, sometimes a better-performance model
5024 exists, but it is always a good start.
5025
5026 =item * other models exist, such as the leader/follower pattern, where one
5027 loop is handed through multiple threads in a kind of round-robin fashion.
5028
5029 Choosing a model is hard - look around, learn, know that usually you can do
5030 better than you currently do :-)
5031
5032 =item * often you need to talk to some other thread which blocks in the
5033 event loop.
5034
5035 C<ev_async> watchers can be used to wake them up from other threads safely
5036 (or from signal contexts...).
5037
5038 An example use would be to communicate signals or other events that only
5039 work in the default loop by registering the signal watcher with the
5040 default loop and triggering an C<ev_async> watcher from the default loop
5041 watcher callback into the event loop interested in the signal.
5042
5043 =back
5044
5045 See also L</THREAD LOCKING EXAMPLE>.
5046
5047 =head3 COROUTINES
5048
5049 Libev is very accommodating to coroutines ("cooperative threads"):
5050 libev fully supports nesting calls to its functions from different
5051 coroutines (e.g. you can call C<ev_run> on the same loop from two
5052 different coroutines, and switch freely between both coroutines running
5053 the loop, as long as you don't confuse yourself). The only exception is
5054 that you must not do this from C<ev_periodic> reschedule callbacks.
5055
5056 Care has been taken to ensure that libev does not keep local state inside
5057 C<ev_run>, and other calls do not usually allow for coroutine switches as
5058 they do not call any callbacks.
5059
5060 =head2 COMPILER WARNINGS
5061
5062 Depending on your compiler and compiler settings, you might get no or a
5063 lot of warnings when compiling libev code. Some people are apparently
5064 scared by this.
5065
5066 However, these are unavoidable for many reasons. For one, each compiler
5067 has different warnings, and each user has different tastes regarding
5068 warning options. "Warn-free" code therefore cannot be a goal except when
5069 targeting a specific compiler and compiler-version.
5070
5071 Another reason is that some compiler warnings require elaborate
5072 workarounds, or other changes to the code that make it less clear and less
5073 maintainable.
5074
5075 And of course, some compiler warnings are just plain stupid, or simply
5076 wrong (because they don't actually warn about the condition their message
5077 seems to warn about). For example, certain older gcc versions had some
5078 warnings that resulted in an extreme number of false positives. These have
5079 been fixed, but some people still insist on making code warn-free with
5080 such buggy versions.
5081
5082 While libev is written to generate as few warnings as possible,
5083 "warn-free" code is not a goal, and it is recommended not to build libev
5084 with any compiler warnings enabled unless you are prepared to cope with
5085 them (e.g. by ignoring them). Remember that warnings are just that:
5086 warnings, not errors, or proof of bugs.
5087
5088
5089 =head2 VALGRIND
5090
5091 Valgrind has a special section here because it is a popular tool that is
5092 highly useful. Unfortunately, valgrind reports are very hard to interpret.
5093
5094 If you think you found a bug (memory leak, uninitialised data access etc.)
5095 in libev, then check twice: If valgrind reports something like:
5096
5097 ==2274== definitely lost: 0 bytes in 0 blocks.
5098 ==2274== possibly lost: 0 bytes in 0 blocks.
5099 ==2274== still reachable: 256 bytes in 1 blocks.
5100
5101 Then there is no memory leak, just as memory accounted to global variables
5102 is not a memleak - the memory is still being referenced, and didn't leak.
5103
5104 Similarly, under some circumstances, valgrind might report kernel bugs
5105 as if it were a bug in libev (e.g. in realloc or in the poll backend,
5106 although an acceptable workaround has been found here), or it might be
5107 confused.
5108
5109 Keep in mind that valgrind is a very good tool, but only a tool. Don't
5110 make it into some kind of religion.
5111
5112 If you are unsure about something, feel free to contact the mailing list
5113 with the full valgrind report and an explanation on why you think this
5114 is a bug in libev (best check the archives, too :). However, don't be
5115 annoyed when you get a brisk "this is no bug" answer and take the chance
5116 of learning how to interpret valgrind properly.
5117
5118 If you need, for some reason, empty reports from valgrind for your project
5119 I suggest using suppression lists.
5120
5121
5122 =head1 PORTABILITY NOTES
5123
5124 =head2 GNU/LINUX 32 BIT LIMITATIONS
5125
5126 GNU/Linux is the only common platform that supports 64 bit file/large file
5127 interfaces but I<disables> them by default.
5128
5129 That means that libev compiled in the default environment doesn't support
5130 files larger than 2GiB or so, which mainly affects C<ev_stat> watchers.
5131
5132 Unfortunately, many programs try to work around this GNU/Linux issue
5133 by enabling the large file API, which makes them incompatible with the
5134 standard libev compiled for their system.
5135
5136 Likewise, libev cannot enable the large file API itself as this would
5137 suddenly make it incompatible to the default compile time environment,
5138 i.e. all programs not using special compile switches.
5139
5140 =head2 OS/X AND DARWIN BUGS
5141
5142 The whole thing is a bug if you ask me - basically any system interface
5143 you touch is broken, whether it is locales, poll, kqueue or even the
5144 OpenGL drivers.
5145
5146 =head3 C<kqueue> is buggy
5147
5148 The kqueue syscall is broken in all known versions - most versions support
5149 only sockets, many support pipes.
5150
5151 Libev tries to work around this by not using C<kqueue> by default on this
5152 rotten platform, but of course you can still ask for it when creating a
5153 loop - embedding a socket-only kqueue loop into a select-based one is
5154 probably going to work well.
5155
5156 =head3 C<poll> is buggy
5157
5158 Instead of fixing C<kqueue>, Apple replaced their (working) C<poll>
5159 implementation by something calling C<kqueue> internally around the 10.5.6
5160 release, so now C<kqueue> I<and> C<poll> are broken.
5161
5162 Libev tries to work around this by not using C<poll> by default on
5163 this rotten platform, but of course you can still ask for it when creating
5164 a loop.
5165
5166 =head3 C<select> is buggy
5167
5168 All that's left is C<select>, and of course Apple found a way to fuck this
5169 one up as well: On OS/X, C<select> actively limits the number of file
5170 descriptors you can pass in to 1024 - your program suddenly crashes when
5171 you use more.
5172
5173 There is an undocumented "workaround" for this - defining
5174 C<_DARWIN_UNLIMITED_SELECT>, which libev tries to use, so select I<should>
5175 work on OS/X.
5176
5177 =head2 SOLARIS PROBLEMS AND WORKAROUNDS
5178
5179 =head3 C<errno> reentrancy
5180
5181 The default compile environment on Solaris is unfortunately so
5182 thread-unsafe that you can't even use components/libraries compiled
5183 without C<-D_REENTRANT> in a threaded program, which, of course, isn't
5184 defined by default. A valid, if stupid, implementation choice.
5185
5186 If you want to use libev in threaded environments you have to make sure
5187 it's compiled with C<_REENTRANT> defined.
5188
5189 =head3 Event port backend
5190
5191 The scalable event interface for Solaris is called "event
5192 ports". Unfortunately, this mechanism is very buggy in all major
5193 releases. If you run into high CPU usage, your program freezes or you get
5194 a large number of spurious wakeups, make sure you have all the relevant
5195 and latest kernel patches applied. No, I don't know which ones, but there
5196 are multiple ones to apply, and afterwards, event ports actually work
5197 great.
5198
5199 If you can't get it to work, you can try running the program by setting
5200 the environment variable C<LIBEV_FLAGS=3> to only allow C<poll> and
5201 C<select> backends.
5202
5203 =head2 AIX POLL BUG
5204
5205 AIX unfortunately has a broken C<poll.h> header. Libev works around
5206 this by trying to avoid the poll backend altogether (i.e. it's not even
5207 compiled in), which normally isn't a big problem as C<select> works fine
5208 with large bitsets on AIX, and AIX is dead anyway.
5209
5210 =head2 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
5211
5212 =head3 General issues
5213
5214 Win32 doesn't support any of the standards (e.g. POSIX) that libev
5215 requires, and its I/O model is fundamentally incompatible with the POSIX
5216 model. Libev still offers limited functionality on this platform in
5217 the form of the C<EVBACKEND_SELECT> backend, and only supports socket
5218 descriptors. This only applies when using Win32 natively, not when using
5219 e.g. cygwin. Actually, it only applies to the microsofts own compilers,
5220 as every compiler comes with a slightly differently broken/incompatible
5221 environment.
5222
5223 Lifting these limitations would basically require the full
5224 re-implementation of the I/O system. If you are into this kind of thing,
5225 then note that glib does exactly that for you in a very portable way (note
5226 also that glib is the slowest event library known to man).
5227
5228 There is no supported compilation method available on windows except
5229 embedding it into other applications.
5230
5231 Sensible signal handling is officially unsupported by Microsoft - libev
5232 tries its best, but under most conditions, signals will simply not work.
5233
5234 Not a libev limitation but worth mentioning: windows apparently doesn't
5235 accept large writes: instead of resulting in a partial write, windows will
5236 either accept everything or return C<ENOBUFS> if the buffer is too large,
5237 so make sure you only write small amounts into your sockets (less than a
5238 megabyte seems safe, but this apparently depends on the amount of memory
5239 available).
5240
5241 Due to the many, low, and arbitrary limits on the win32 platform and
5242 the abysmal performance of winsockets, using a large number of sockets
5243 is not recommended (and not reasonable). If your program needs to use
5244 more than a hundred or so sockets, then likely it needs to use a totally
5245 different implementation for windows, as libev offers the POSIX readiness
5246 notification model, which cannot be implemented efficiently on windows
5247 (due to Microsoft monopoly games).
5248
5249 A typical way to use libev under windows is to embed it (see the embedding
5250 section for details) and use the following F<evwrap.h> header file instead
5251 of F<ev.h>:
5252
5253 #define EV_STANDALONE /* keeps ev from requiring config.h */
5254 #define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
5255
5256 #include "ev.h"
5257
5258 And compile the following F<evwrap.c> file into your project (make sure
5259 you do I<not> compile the F<ev.c> or any other embedded source files!):
5260
5261 #include "evwrap.h"
5262 #include "ev.c"
5263
5264 =head3 The winsocket C<select> function
5265
5266 The winsocket C<select> function doesn't follow POSIX in that it
5267 requires socket I<handles> and not socket I<file descriptors> (it is
5268 also extremely buggy). This makes select very inefficient, and also
5269 requires a mapping from file descriptors to socket handles (the Microsoft
5270 C runtime provides the function C<_open_osfhandle> for this). See the
5271 discussion of the C<EV_SELECT_USE_FD_SET>, C<EV_SELECT_IS_WINSOCKET> and
5272 C<EV_FD_TO_WIN32_HANDLE> preprocessor symbols for more info.
5273
5274 The configuration for a "naked" win32 using the Microsoft runtime
5275 libraries and raw winsocket select is:
5276
5277 #define EV_USE_SELECT 1
5278 #define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
5279
5280 Note that winsockets handling of fd sets is O(n), so you can easily get a
5281 complexity in the O(n²) range when using win32.
5282
5283 =head3 Limited number of file descriptors
5284
5285 Windows has numerous arbitrary (and low) limits on things.
5286
5287 Early versions of winsocket's select only supported waiting for a maximum
5288 of C<64> handles (probably owning to the fact that all windows kernels
5289 can only wait for C<64> things at the same time internally; Microsoft
5290 recommends spawning a chain of threads and wait for 63 handles and the
5291 previous thread in each. Sounds great!).
5292
5293 Newer versions support more handles, but you need to define C<FD_SETSIZE>
5294 to some high number (e.g. C<2048>) before compiling the winsocket select
5295 call (which might be in libev or elsewhere, for example, perl and many
5296 other interpreters do their own select emulation on windows).
5297
5298 Another limit is the number of file descriptors in the Microsoft runtime
5299 libraries, which by default is C<64> (there must be a hidden I<64>
5300 fetish or something like this inside Microsoft). You can increase this
5301 by calling C<_setmaxstdio>, which can increase this limit to C<2048>
5302 (another arbitrary limit), but is broken in many versions of the Microsoft
5303 runtime libraries. This might get you to about C<512> or C<2048> sockets
5304 (depending on windows version and/or the phase of the moon). To get more,
5305 you need to wrap all I/O functions and provide your own fd management, but
5306 the cost of calling select (O(n²)) will likely make this unworkable.
5307
5308 =head2 PORTABILITY REQUIREMENTS
5309
5310 In addition to a working ISO-C implementation and of course the
5311 backend-specific APIs, libev relies on a few additional extensions:
5312
5313 =over 4
5314
5315 =item C<void (*)(ev_watcher_type *, int revents)> must have compatible
5316 calling conventions regardless of C<ev_watcher_type *>.
5317
5318 Libev assumes not only that all watcher pointers have the same internal
5319 structure (guaranteed by POSIX but not by ISO C for example), but it also
5320 assumes that the same (machine) code can be used to call any watcher
5321 callback: The watcher callbacks have different type signatures, but libev
5322 calls them using an C<ev_watcher *> internally.
5323
5324 =item null pointers and integer zero are represented by 0 bytes
5325
5326 Libev uses C<memset> to initialise structs and arrays to C<0> bytes, and
5327 relies on this setting pointers and integers to null.
5328
5329 =item pointer accesses must be thread-atomic
5330
5331 Accessing a pointer value must be atomic, it must both be readable and
5332 writable in one piece - this is the case on all current architectures.
5333
5334 =item C<sig_atomic_t volatile> must be thread-atomic as well
5335
5336 The type C<sig_atomic_t volatile> (or whatever is defined as
5337 C<EV_ATOMIC_T>) must be atomic with respect to accesses from different
5338 threads. This is not part of the specification for C<sig_atomic_t>, but is
5339 believed to be sufficiently portable.
5340
5341 =item C<sigprocmask> must work in a threaded environment
5342
5343 Libev uses C<sigprocmask> to temporarily block signals. This is not
5344 allowed in a threaded program (C<pthread_sigmask> has to be used). Typical
5345 pthread implementations will either allow C<sigprocmask> in the "main
5346 thread" or will block signals process-wide, both behaviours would
5347 be compatible with libev. Interaction between C<sigprocmask> and
5348 C<pthread_sigmask> could complicate things, however.
5349
5350 The most portable way to handle signals is to block signals in all threads
5351 except the initial one, and run the signal handling loop in the initial
5352 thread as well.
5353
5354 =item C<long> must be large enough for common memory allocation sizes
5355
5356 To improve portability and simplify its API, libev uses C<long> internally
5357 instead of C<size_t> when allocating its data structures. On non-POSIX
5358 systems (Microsoft...) this might be unexpectedly low, but is still at
5359 least 31 bits everywhere, which is enough for hundreds of millions of
5360 watchers.
5361
5362 =item C<double> must hold a time value in seconds with enough accuracy
5363
5364 The type C<double> is used to represent timestamps. It is required to
5365 have at least 51 bits of mantissa (and 9 bits of exponent), which is
5366 good enough for at least into the year 4000 with millisecond accuracy
5367 (the design goal for libev). This requirement is overfulfilled by
5368 implementations using IEEE 754, which is basically all existing ones.
5369
5370 With IEEE 754 doubles, you get microsecond accuracy until at least the
5371 year 2255 (and millisecond accuracy till the year 287396 - by then, libev
5372 is either obsolete or somebody patched it to use C<long double> or
5373 something like that, just kidding).
5374
5375 =back
5376
5377 If you know of other additional requirements drop me a note.
5378
5379
5380 =head1 ALGORITHMIC COMPLEXITIES
5381
5382 In this section the complexities of (many of) the algorithms used inside
5383 libev will be documented. For complexity discussions about backends see
5384 the documentation for C<ev_default_init>.
5385
5386 All of the following are about amortised time: If an array needs to be
5387 extended, libev needs to realloc and move the whole array, but this
5388 happens asymptotically rarer with higher number of elements, so O(1) might
5389 mean that libev does a lengthy realloc operation in rare cases, but on
5390 average it is much faster and asymptotically approaches constant time.
5391
5392 =over 4
5393
5394 =item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
5395
5396 This means that, when you have a watcher that triggers in one hour and
5397 there are 100 watchers that would trigger before that, then inserting will
5398 have to skip roughly seven (C<ld 100>) of these watchers.
5399
5400 =item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
5401
5402 That means that changing a timer costs less than removing/adding them,
5403 as only the relative motion in the event queue has to be paid for.
5404
5405 =item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
5406
5407 These just add the watcher into an array or at the head of a list.
5408
5409 =item Stopping check/prepare/idle/fork/async watchers: O(1)
5410
5411 =item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
5412
5413 These watchers are stored in lists, so they need to be walked to find the
5414 correct watcher to remove. The lists are usually short (you don't usually
5415 have many watchers waiting for the same fd or signal: one is typical, two
5416 is rare).
5417
5418 =item Finding the next timer in each loop iteration: O(1)
5419
5420 By virtue of using a binary or 4-heap, the next timer is always found at a
5421 fixed position in the storage array.
5422
5423 =item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
5424
5425 A change means an I/O watcher gets started or stopped, which requires
5426 libev to recalculate its status (and possibly tell the kernel, depending
5427 on backend and whether C<ev_io_set> was used).
5428
5429 =item Activating one watcher (putting it into the pending state): O(1)
5430
5431 =item Priority handling: O(number_of_priorities)
5432
5433 Priorities are implemented by allocating some space for each
5434 priority. When doing priority-based operations, libev usually has to
5435 linearly search all the priorities, but starting/stopping and activating
5436 watchers becomes O(1) with respect to priority handling.
5437
5438 =item Sending an ev_async: O(1)
5439
5440 =item Processing ev_async_send: O(number_of_async_watchers)
5441
5442 =item Processing signals: O(max_signal_number)
5443
5444 Sending involves a system call I<iff> there were no other C<ev_async_send>
5445 calls in the current loop iteration and the loop is currently
5446 blocked. Checking for async and signal events involves iterating over all
5447 running async watchers or all signal numbers.
5448
5449 =back
5450
5451
5452 =head1 PORTING FROM LIBEV 3.X TO 4.X
5453
5454 The major version 4 introduced some incompatible changes to the API.
5455
5456 At the moment, the C<ev.h> header file provides compatibility definitions
5457 for all changes, so most programs should still compile. The compatibility
5458 layer might be removed in later versions of libev, so better update to the
5459 new API early than late.
5460
5461 =over 4
5462
5463 =item C<EV_COMPAT3> backwards compatibility mechanism
5464
5465 The backward compatibility mechanism can be controlled by
5466 C<EV_COMPAT3>. See L</"PREPROCESSOR SYMBOLS/MACROS"> in the L</EMBEDDING>
5467 section.
5468
5469 =item C<ev_default_destroy> and C<ev_default_fork> have been removed
5470
5471 These calls can be replaced easily by their C<ev_loop_xxx> counterparts:
5472
5473 ev_loop_destroy (EV_DEFAULT_UC);
5474 ev_loop_fork (EV_DEFAULT);
5475
5476 =item function/symbol renames
5477
5478 A number of functions and symbols have been renamed:
5479
5480 ev_loop => ev_run
5481 EVLOOP_NONBLOCK => EVRUN_NOWAIT
5482 EVLOOP_ONESHOT => EVRUN_ONCE
5483
5484 ev_unloop => ev_break
5485 EVUNLOOP_CANCEL => EVBREAK_CANCEL
5486 EVUNLOOP_ONE => EVBREAK_ONE
5487 EVUNLOOP_ALL => EVBREAK_ALL
5488
5489 EV_TIMEOUT => EV_TIMER
5490
5491 ev_loop_count => ev_iteration
5492 ev_loop_depth => ev_depth
5493 ev_loop_verify => ev_verify
5494
5495 Most functions working on C<struct ev_loop> objects don't have an
5496 C<ev_loop_> prefix, so it was removed; C<ev_loop>, C<ev_unloop> and
5497 associated constants have been renamed to not collide with the C<struct
5498 ev_loop> anymore and C<EV_TIMER> now follows the same naming scheme
5499 as all other watcher types. Note that C<ev_loop_fork> is still called
5500 C<ev_loop_fork> because it would otherwise clash with the C<ev_fork>
5501 typedef.
5502
5503 =item C<EV_MINIMAL> mechanism replaced by C<EV_FEATURES>
5504
5505 The preprocessor symbol C<EV_MINIMAL> has been replaced by a different
5506 mechanism, C<EV_FEATURES>. Programs using C<EV_MINIMAL> usually compile
5507 and work, but the library code will of course be larger.
5508
5509 =back
5510
5511
5512 =head1 GLOSSARY
5513
5514 =over 4
5515
5516 =item active
5517
5518 A watcher is active as long as it has been started and not yet stopped.
5519 See L</WATCHER STATES> for details.
5520
5521 =item application
5522
5523 In this document, an application is whatever is using libev.
5524
5525 =item backend
5526
5527 The part of the code dealing with the operating system interfaces.
5528
5529 =item callback
5530
5531 The address of a function that is called when some event has been
5532 detected. Callbacks are being passed the event loop, the watcher that
5533 received the event, and the actual event bitset.
5534
5535 =item callback/watcher invocation
5536
5537 The act of calling the callback associated with a watcher.
5538
5539 =item event
5540
5541 A change of state of some external event, such as data now being available
5542 for reading on a file descriptor, time having passed or simply not having
5543 any other events happening anymore.
5544
5545 In libev, events are represented as single bits (such as C<EV_READ> or
5546 C<EV_TIMER>).
5547
5548 =item event library
5549
5550 A software package implementing an event model and loop.
5551
5552 =item event loop
5553
5554 An entity that handles and processes external events and converts them
5555 into callback invocations.
5556
5557 =item event model
5558
5559 The model used to describe how an event loop handles and processes
5560 watchers and events.
5561
5562 =item pending
5563
5564 A watcher is pending as soon as the corresponding event has been
5565 detected. See L</WATCHER STATES> for details.
5566
5567 =item real time
5568
5569 The physical time that is observed. It is apparently strictly monotonic :)
5570
5571 =item wall-clock time
5572
5573 The time and date as shown on clocks. Unlike real time, it can actually
5574 be wrong and jump forwards and backwards, e.g. when you adjust your
5575 clock.
5576
5577 =item watcher
5578
5579 A data structure that describes interest in certain events. Watchers need
5580 to be started (attached to an event loop) before they can receive events.
5581
5582 =back
5583
5584 =head1 AUTHOR
5585
5586 Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael
5587 Magnusson and Emanuele Giaquinta, and minor corrections by many others.
5588