ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/ev.pod
Revision: 1.451
Committed: Mon Jun 24 00:19:26 2019 UTC (4 years, 10 months ago) by root
Branch: MAIN
Changes since 1.450: +7 -5 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 aio and C<epoll>
111 interfaces, the BSD-specific C<kqueue> and the Solaris-specific event port
112 mechanisms for file descriptor events (C<ev_io>), the Linux C<inotify>
113 interface (for C<ev_stat>), Linux eventfd/signalfd (for faster and cleaner
114 inter-thread wakeup (C<ev_async>)/signal handling (C<ev_signal>)) relative
115 timers (C<ev_timer>), absolute timers with customised rescheduling
116 (C<ev_periodic>), synchronous signals (C<ev_signal>), process status
117 change events (C<ev_child>), and event watchers dealing with the event
118 loop mechanism itself (C<ev_idle>, C<ev_embed>, C<ev_prepare> and
119 C<ev_check> watchers) as well as file watchers (C<ev_stat>) and even
120 limited support for fork events (C<ev_fork>).
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: The following is the C<realloc> function that libev itself uses
271 which should work with C<realloc> and C<free> functions of all kinds and
272 is probably a good basis for your own implementation.
273
274 static void *
275 ev_realloc_emul (void *ptr, long size) EV_NOEXCEPT
276 {
277 if (size)
278 return realloc (ptr, size);
279
280 free (ptr);
281 return 0;
282 }
283
284 Example: Replace the libev allocator with one that waits a bit and then
285 retries.
286
287 static void *
288 persistent_realloc (void *ptr, size_t size)
289 {
290 if (!size)
291 {
292 free (ptr);
293 return 0;
294 }
295
296 for (;;)
297 {
298 void *newptr = realloc (ptr, size);
299
300 if (newptr)
301 return newptr;
302
303 sleep (60);
304 }
305 }
306
307 ...
308 ev_set_allocator (persistent_realloc);
309
310 =item ev_set_syserr_cb (void (*cb)(const char *msg) throw ())
311
312 Set the callback function to call on a retryable system call error (such
313 as failed select, poll, epoll_wait). The message is a printable string
314 indicating the system call or subsystem causing the problem. If this
315 callback is set, then libev will expect it to remedy the situation, no
316 matter what, when it returns. That is, libev will generally retry the
317 requested operation, or, if the condition doesn't go away, do bad stuff
318 (such as abort).
319
320 Example: This is basically the same thing that libev does internally, too.
321
322 static void
323 fatal_error (const char *msg)
324 {
325 perror (msg);
326 abort ();
327 }
328
329 ...
330 ev_set_syserr_cb (fatal_error);
331
332 =item ev_feed_signal (int signum)
333
334 This function can be used to "simulate" a signal receive. It is completely
335 safe to call this function at any time, from any context, including signal
336 handlers or random threads.
337
338 Its main use is to customise signal handling in your process, especially
339 in the presence of threads. For example, you could block signals
340 by default in all threads (and specifying C<EVFLAG_NOSIGMASK> when
341 creating any loops), and in one thread, use C<sigwait> or any other
342 mechanism to wait for signals, then "deliver" them to libev by calling
343 C<ev_feed_signal>.
344
345 =back
346
347 =head1 FUNCTIONS CONTROLLING EVENT LOOPS
348
349 An event loop is described by a C<struct ev_loop *> (the C<struct> is
350 I<not> optional in this case unless libev 3 compatibility is disabled, as
351 libev 3 had an C<ev_loop> function colliding with the struct name).
352
353 The library knows two types of such loops, the I<default> loop, which
354 supports child process events, and dynamically created event loops which
355 do not.
356
357 =over 4
358
359 =item struct ev_loop *ev_default_loop (unsigned int flags)
360
361 This returns the "default" event loop object, which is what you should
362 normally use when you just need "the event loop". Event loop objects and
363 the C<flags> parameter are described in more detail in the entry for
364 C<ev_loop_new>.
365
366 If the default loop is already initialised then this function simply
367 returns it (and ignores the flags. If that is troubling you, check
368 C<ev_backend ()> afterwards). Otherwise it will create it with the given
369 flags, which should almost always be C<0>, unless the caller is also the
370 one calling C<ev_run> or otherwise qualifies as "the main program".
371
372 If you don't know what event loop to use, use the one returned from this
373 function (or via the C<EV_DEFAULT> macro).
374
375 Note that this function is I<not> thread-safe, so if you want to use it
376 from multiple threads, you have to employ some kind of mutex (note also
377 that this case is unlikely, as loops cannot be shared easily between
378 threads anyway).
379
380 The default loop is the only loop that can handle C<ev_child> watchers,
381 and to do this, it always registers a handler for C<SIGCHLD>. If this is
382 a problem for your application you can either create a dynamic loop with
383 C<ev_loop_new> which doesn't do that, or you can simply overwrite the
384 C<SIGCHLD> signal handler I<after> calling C<ev_default_init>.
385
386 Example: This is the most typical usage.
387
388 if (!ev_default_loop (0))
389 fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
390
391 Example: Restrict libev to the select and poll backends, and do not allow
392 environment settings to be taken into account:
393
394 ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
395
396 =item struct ev_loop *ev_loop_new (unsigned int flags)
397
398 This will create and initialise a new event loop object. If the loop
399 could not be initialised, returns false.
400
401 This function is thread-safe, and one common way to use libev with
402 threads is indeed to create one loop per thread, and using the default
403 loop in the "main" or "initial" thread.
404
405 The flags argument can be used to specify special behaviour or specific
406 backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
407
408 The following flags are supported:
409
410 =over 4
411
412 =item C<EVFLAG_AUTO>
413
414 The default flags value. Use this if you have no clue (it's the right
415 thing, believe me).
416
417 =item C<EVFLAG_NOENV>
418
419 If this flag bit is or'ed into the flag value (or the program runs setuid
420 or setgid) then libev will I<not> look at the environment variable
421 C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
422 override the flags completely if it is found in the environment. This is
423 useful to try out specific backends to test their performance, to work
424 around bugs, or to make libev threadsafe (accessing environment variables
425 cannot be done in a threadsafe way, but usually it works if no other
426 thread modifies them).
427
428 =item C<EVFLAG_FORKCHECK>
429
430 Instead of calling C<ev_loop_fork> manually after a fork, you can also
431 make libev check for a fork in each iteration by enabling this flag.
432
433 This works by calling C<getpid ()> on every iteration of the loop,
434 and thus this might slow down your event loop if you do a lot of loop
435 iterations and little real work, but is usually not noticeable (on my
436 GNU/Linux system for example, C<getpid> is actually a simple 5-insn
437 sequence without a system call and thus I<very> fast, but my GNU/Linux
438 system also has C<pthread_atfork> which is even faster). (Update: glibc
439 versions 2.25 apparently removed the C<getpid> optimisation again).
440
441 The big advantage of this flag is that you can forget about fork (and
442 forget about forgetting to tell libev about forking, although you still
443 have to ignore C<SIGPIPE>) when you use this flag.
444
445 This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
446 environment variable.
447
448 =item C<EVFLAG_NOINOTIFY>
449
450 When this flag is specified, then libev will not attempt to use the
451 I<inotify> API for its C<ev_stat> watchers. Apart from debugging and
452 testing, this flag can be useful to conserve inotify file descriptors, as
453 otherwise each loop using C<ev_stat> watchers consumes one inotify handle.
454
455 =item C<EVFLAG_SIGNALFD>
456
457 When this flag is specified, then libev will attempt to use the
458 I<signalfd> API for its C<ev_signal> (and C<ev_child>) watchers. This API
459 delivers signals synchronously, which makes it both faster and might make
460 it possible to get the queued signal data. It can also simplify signal
461 handling with threads, as long as you properly block signals in your
462 threads that are not interested in handling them.
463
464 Signalfd will not be used by default as this changes your signal mask, and
465 there are a lot of shoddy libraries and programs (glib's threadpool for
466 example) that can't properly initialise their signal masks.
467
468 =item C<EVFLAG_NOSIGMASK>
469
470 When this flag is specified, then libev will avoid to modify the signal
471 mask. Specifically, this means you have to make sure signals are unblocked
472 when you want to receive them.
473
474 This behaviour is useful when you want to do your own signal handling, or
475 want to handle signals only in specific threads and want to avoid libev
476 unblocking the signals.
477
478 It's also required by POSIX in a threaded program, as libev calls
479 C<sigprocmask>, whose behaviour is officially unspecified.
480
481 This flag's behaviour will become the default in future versions of libev.
482
483 =item C<EVBACKEND_SELECT> (value 1, portable select backend)
484
485 This is your standard select(2) backend. Not I<completely> standard, as
486 libev tries to roll its own fd_set with no limits on the number of fds,
487 but if that fails, expect a fairly low limit on the number of fds when
488 using this backend. It doesn't scale too well (O(highest_fd)), but its
489 usually the fastest backend for a low number of (low-numbered :) fds.
490
491 To get good performance out of this backend you need a high amount of
492 parallelism (most of the file descriptors should be busy). If you are
493 writing a server, you should C<accept ()> in a loop to accept as many
494 connections as possible during one iteration. You might also want to have
495 a look at C<ev_set_io_collect_interval ()> to increase the amount of
496 readiness notifications you get per iteration.
497
498 This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
499 C<writefds> set (and to work around Microsoft Windows bugs, also onto the
500 C<exceptfds> set on that platform).
501
502 =item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
503
504 And this is your standard poll(2) backend. It's more complicated
505 than select, but handles sparse fds better and has no artificial
506 limit on the number of fds you can use (except it will slow down
507 considerably with a lot of inactive fds). It scales similarly to select,
508 i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
509 performance tips.
510
511 This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
512 C<EV_WRITE> to C<POLLOUT | POLLERR | POLLHUP>.
513
514 =item C<EVBACKEND_EPOLL> (value 4, Linux)
515
516 Use the linux-specific epoll(7) interface (for both pre- and post-2.6.9
517 kernels).
518
519 For few fds, this backend is a bit little slower than poll and select, but
520 it scales phenomenally better. While poll and select usually scale like
521 O(total_fds) where total_fds is the total number of fds (or the highest
522 fd), epoll scales either O(1) or O(active_fds).
523
524 The epoll mechanism deserves honorable mention as the most misdesigned
525 of the more advanced event mechanisms: mere annoyances include silently
526 dropping file descriptors, requiring a system call per change per file
527 descriptor (and unnecessary guessing of parameters), problems with dup,
528 returning before the timeout value, resulting in additional iterations
529 (and only giving 5ms accuracy while select on the same platform gives
530 0.1ms) and so on. The biggest issue is fork races, however - if a program
531 forks then I<both> parent and child process have to recreate the epoll
532 set, which can take considerable time (one syscall per file descriptor)
533 and is of course hard to detect.
534
535 Epoll is also notoriously buggy - embedding epoll fds I<should> work,
536 but of course I<doesn't>, and epoll just loves to report events for
537 totally I<different> file descriptors (even already closed ones, so
538 one cannot even remove them from the set) than registered in the set
539 (especially on SMP systems). Libev tries to counter these spurious
540 notifications by employing an additional generation counter and comparing
541 that against the events to filter out spurious ones, recreating the set
542 when required. Epoll also erroneously rounds down timeouts, but gives you
543 no way to know when and by how much, so sometimes you have to busy-wait
544 because epoll returns immediately despite a nonzero timeout. And last
545 not least, it also refuses to work with some file descriptors which work
546 perfectly fine with C<select> (files, many character devices...).
547
548 Epoll is truly the train wreck among event poll mechanisms, a frankenpoll,
549 cobbled together in a hurry, no thought to design or interaction with
550 others. Oh, the pain, will it ever stop...
551
552 While stopping, setting and starting an I/O watcher in the same iteration
553 will result in some caching, there is still a system call per such
554 incident (because the same I<file descriptor> could point to a different
555 I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
556 file descriptors might not work very well if you register events for both
557 file descriptors.
558
559 Best performance from this backend is achieved by not unregistering all
560 watchers for a file descriptor until it has been closed, if possible,
561 i.e. keep at least one watcher active per fd at all times. Stopping and
562 starting a watcher (without re-setting it) also usually doesn't cause
563 extra overhead. A fork can both result in spurious notifications as well
564 as in libev having to destroy and recreate the epoll object, which can
565 take considerable time and thus should be avoided.
566
567 All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
568 faster than epoll for maybe up to a hundred file descriptors, depending on
569 the usage. So sad.
570
571 While nominally embeddable in other event loops, this feature is broken in
572 a lot of kernel revisions, but probably(!) works in current versions.
573
574 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
575 C<EVBACKEND_POLL>.
576
577 =item C<EVBACKEND_LINUXAIO> (value 64, Linux)
578
579 Use the linux-specific linux aio (I<not> C<< aio(7) >> but C<<
580 io_submit(2) >>) event interface available in post-4.18 kernels.
581
582 If this backend works for you (as of this writing, it was very
583 experimental), it is the best event interface available on linux and might
584 be well worth enabling it - if it isn't available in your kernel this will
585 be detected and this backend will be skipped.
586
587 This backend can batch oneshot requests and supports a user-space ring
588 buffer to receive events. It also doesn't suffer from most of the design
589 problems of epoll (such as not being able to remove event sources from the
590 epoll set), and generally sounds too good to be true. Because, this being
591 the linux kernel, of course it suffers from a whole new set of limitations.
592
593 For one, it is not easily embeddable (but probably could be done using
594 an event fd at some extra overhead). It also is subject to a system wide
595 limit that can be configured in F</proc/sys/fs/aio-max-nr> - each loop
596 currently requires C<61> of this number. If no aio requests are left, this
597 backend will be skipped during initialisation.
598
599 Most problematic in practise, however, is that not all file descriptors
600 work with it. For example, in linux 5.1, tcp sockets, pipes, event fds,
601 files, F</dev/null> and a few others are supported, but ttys do not work
602 properly (a known bug that the kernel developers don't care about, see
603 L<https://lore.kernel.org/patchwork/patch/1047453/>), so this is not
604 (yet?) a generic event polling interface.
605
606 Overall, it seems the linux developers just don't want it to have a
607 generic event handling mechanism other than C<select> or C<poll>.
608
609 To work around the fd type problem, the current version of libev uses
610 epoll as a fallback for file deescriptor types that do not work. Epoll
611 is used in, kind of, slow mode that hopefully avoids most of its design
612 problems and requires 1-3 extra syscalls per active fd every iteration.
613
614 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
615 C<EVBACKEND_POLL>.
616
617 =item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
618
619 Kqueue deserves special mention, as at the time of this writing, it
620 was broken on all BSDs except NetBSD (usually it doesn't work reliably
621 with anything but sockets and pipes, except on Darwin, where of course
622 it's completely useless). Unlike epoll, however, whose brokenness
623 is by design, these kqueue bugs can (and eventually will) be fixed
624 without API changes to existing programs. For this reason it's not being
625 "auto-detected" unless you explicitly specify it in the flags (i.e. using
626 C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
627 system like NetBSD.
628
629 You still can embed kqueue into a normal poll or select backend and use it
630 only for sockets (after having made sure that sockets work with kqueue on
631 the target platform). See C<ev_embed> watchers for more info.
632
633 It scales in the same way as the epoll backend, but the interface to the
634 kernel is more efficient (which says nothing about its actual speed, of
635 course). While stopping, setting and starting an I/O watcher does never
636 cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
637 two event changes per incident. Support for C<fork ()> is very bad (you
638 might have to leak fd's on fork, but it's more sane than epoll) and it
639 drops fds silently in similarly hard-to-detect cases.
640
641 This backend usually performs well under most conditions.
642
643 While nominally embeddable in other event loops, this doesn't work
644 everywhere, so you might need to test for this. And since it is broken
645 almost everywhere, you should only use it when you have a lot of sockets
646 (for which it usually works), by embedding it into another event loop
647 (e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
648 also broken on OS X)) and, did I mention it, using it only for sockets.
649
650 This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
651 C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
652 C<NOTE_EOF>.
653
654 =item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
655
656 This is not implemented yet (and might never be, unless you send me an
657 implementation). According to reports, C</dev/poll> only supports sockets
658 and is not embeddable, which would limit the usefulness of this backend
659 immensely.
660
661 =item C<EVBACKEND_PORT> (value 32, Solaris 10)
662
663 This uses the Solaris 10 event port mechanism. As with everything on Solaris,
664 it's really slow, but it still scales very well (O(active_fds)).
665
666 While this backend scales well, it requires one system call per active
667 file descriptor per loop iteration. For small and medium numbers of file
668 descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
669 might perform better.
670
671 On the positive side, this backend actually performed fully to
672 specification in all tests and is fully embeddable, which is a rare feat
673 among the OS-specific backends (I vastly prefer correctness over speed
674 hacks).
675
676 On the negative side, the interface is I<bizarre> - so bizarre that
677 even sun itself gets it wrong in their code examples: The event polling
678 function sometimes returns events to the caller even though an error
679 occurred, but with no indication whether it has done so or not (yes, it's
680 even documented that way) - deadly for edge-triggered interfaces where you
681 absolutely have to know whether an event occurred or not because you have
682 to re-arm the watcher.
683
684 Fortunately libev seems to be able to work around these idiocies.
685
686 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
687 C<EVBACKEND_POLL>.
688
689 =item C<EVBACKEND_ALL>
690
691 Try all backends (even potentially broken ones that wouldn't be tried
692 with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
693 C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.
694
695 It is definitely not recommended to use this flag, use whatever
696 C<ev_recommended_backends ()> returns, or simply do not specify a backend
697 at all.
698
699 =item C<EVBACKEND_MASK>
700
701 Not a backend at all, but a mask to select all backend bits from a
702 C<flags> value, in case you want to mask out any backends from a flags
703 value (e.g. when modifying the C<LIBEV_FLAGS> environment variable).
704
705 =back
706
707 If one or more of the backend flags are or'ed into the flags value,
708 then only these backends will be tried (in the reverse order as listed
709 here). If none are specified, all backends in C<ev_recommended_backends
710 ()> will be tried.
711
712 Example: Try to create a event loop that uses epoll and nothing else.
713
714 struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
715 if (!epoller)
716 fatal ("no epoll found here, maybe it hides under your chair");
717
718 Example: Use whatever libev has to offer, but make sure that kqueue is
719 used if available.
720
721 struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_KQUEUE);
722
723 Example: Similarly, on linux, you mgiht want to take advantage of the
724 linux aio backend if possible, but fall back to something else if that
725 isn't available.
726
727 struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_LINUXAIO);
728
729 =item ev_loop_destroy (loop)
730
731 Destroys an event loop object (frees all memory and kernel state
732 etc.). None of the active event watchers will be stopped in the normal
733 sense, so e.g. C<ev_is_active> might still return true. It is your
734 responsibility to either stop all watchers cleanly yourself I<before>
735 calling this function, or cope with the fact afterwards (which is usually
736 the easiest thing, you can just ignore the watchers and/or C<free ()> them
737 for example).
738
739 Note that certain global state, such as signal state (and installed signal
740 handlers), will not be freed by this function, and related watchers (such
741 as signal and child watchers) would need to be stopped manually.
742
743 This function is normally used on loop objects allocated by
744 C<ev_loop_new>, but it can also be used on the default loop returned by
745 C<ev_default_loop>, in which case it is not thread-safe.
746
747 Note that it is not advisable to call this function on the default loop
748 except in the rare occasion where you really need to free its resources.
749 If you need dynamically allocated loops it is better to use C<ev_loop_new>
750 and C<ev_loop_destroy>.
751
752 =item ev_loop_fork (loop)
753
754 This function sets a flag that causes subsequent C<ev_run> iterations
755 to reinitialise the kernel state for backends that have one. Despite
756 the name, you can call it anytime you are allowed to start or stop
757 watchers (except inside an C<ev_prepare> callback), but it makes most
758 sense after forking, in the child process. You I<must> call it (or use
759 C<EVFLAG_FORKCHECK>) in the child before resuming or calling C<ev_run>.
760
761 In addition, if you want to reuse a loop (via this function or
762 C<EVFLAG_FORKCHECK>), you I<also> have to ignore C<SIGPIPE>.
763
764 Again, you I<have> to call it on I<any> loop that you want to re-use after
765 a fork, I<even if you do not plan to use the loop in the parent>. This is
766 because some kernel interfaces *cough* I<kqueue> *cough* do funny things
767 during fork.
768
769 On the other hand, you only need to call this function in the child
770 process if and only if you want to use the event loop in the child. If
771 you just fork+exec or create a new loop in the child, you don't have to
772 call it at all (in fact, C<epoll> is so badly broken that it makes a
773 difference, but libev will usually detect this case on its own and do a
774 costly reset of the backend).
775
776 The function itself is quite fast and it's usually not a problem to call
777 it just in case after a fork.
778
779 Example: Automate calling C<ev_loop_fork> on the default loop when
780 using pthreads.
781
782 static void
783 post_fork_child (void)
784 {
785 ev_loop_fork (EV_DEFAULT);
786 }
787
788 ...
789 pthread_atfork (0, 0, post_fork_child);
790
791 =item int ev_is_default_loop (loop)
792
793 Returns true when the given loop is, in fact, the default loop, and false
794 otherwise.
795
796 =item unsigned int ev_iteration (loop)
797
798 Returns the current iteration count for the event loop, which is identical
799 to the number of times libev did poll for new events. It starts at C<0>
800 and happily wraps around with enough iterations.
801
802 This value can sometimes be useful as a generation counter of sorts (it
803 "ticks" the number of loop iterations), as it roughly corresponds with
804 C<ev_prepare> and C<ev_check> calls - and is incremented between the
805 prepare and check phases.
806
807 =item unsigned int ev_depth (loop)
808
809 Returns the number of times C<ev_run> was entered minus the number of
810 times C<ev_run> was exited normally, in other words, the recursion depth.
811
812 Outside C<ev_run>, this number is zero. In a callback, this number is
813 C<1>, unless C<ev_run> was invoked recursively (or from another thread),
814 in which case it is higher.
815
816 Leaving C<ev_run> abnormally (setjmp/longjmp, cancelling the thread,
817 throwing an exception etc.), doesn't count as "exit" - consider this
818 as a hint to avoid such ungentleman-like behaviour unless it's really
819 convenient, in which case it is fully supported.
820
821 =item unsigned int ev_backend (loop)
822
823 Returns one of the C<EVBACKEND_*> flags indicating the event backend in
824 use.
825
826 =item ev_tstamp ev_now (loop)
827
828 Returns the current "event loop time", which is the time the event loop
829 received events and started processing them. This timestamp does not
830 change as long as callbacks are being processed, and this is also the base
831 time used for relative timers. You can treat it as the timestamp of the
832 event occurring (or more correctly, libev finding out about it).
833
834 =item ev_now_update (loop)
835
836 Establishes the current time by querying the kernel, updating the time
837 returned by C<ev_now ()> in the progress. This is a costly operation and
838 is usually done automatically within C<ev_run ()>.
839
840 This function is rarely useful, but when some event callback runs for a
841 very long time without entering the event loop, updating libev's idea of
842 the current time is a good idea.
843
844 See also L</The special problem of time updates> in the C<ev_timer> section.
845
846 =item ev_suspend (loop)
847
848 =item ev_resume (loop)
849
850 These two functions suspend and resume an event loop, for use when the
851 loop is not used for a while and timeouts should not be processed.
852
853 A typical use case would be an interactive program such as a game: When
854 the user presses C<^Z> to suspend the game and resumes it an hour later it
855 would be best to handle timeouts as if no time had actually passed while
856 the program was suspended. This can be achieved by calling C<ev_suspend>
857 in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
858 C<ev_resume> directly afterwards to resume timer processing.
859
860 Effectively, all C<ev_timer> watchers will be delayed by the time spend
861 between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
862 will be rescheduled (that is, they will lose any events that would have
863 occurred while suspended).
864
865 After calling C<ev_suspend> you B<must not> call I<any> function on the
866 given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
867 without a previous call to C<ev_suspend>.
868
869 Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
870 event loop time (see C<ev_now_update>).
871
872 =item bool ev_run (loop, int flags)
873
874 Finally, this is it, the event handler. This function usually is called
875 after you have initialised all your watchers and you want to start
876 handling events. It will ask the operating system for any new events, call
877 the watcher callbacks, and then repeat the whole process indefinitely: This
878 is why event loops are called I<loops>.
879
880 If the flags argument is specified as C<0>, it will keep handling events
881 until either no event watchers are active anymore or C<ev_break> was
882 called.
883
884 The return value is false if there are no more active watchers (which
885 usually means "all jobs done" or "deadlock"), and true in all other cases
886 (which usually means " you should call C<ev_run> again").
887
888 Please note that an explicit C<ev_break> is usually better than
889 relying on all watchers to be stopped when deciding when a program has
890 finished (especially in interactive programs), but having a program
891 that automatically loops as long as it has to and no longer by virtue
892 of relying on its watchers stopping correctly, that is truly a thing of
893 beauty.
894
895 This function is I<mostly> exception-safe - you can break out of a
896 C<ev_run> call by calling C<longjmp> in a callback, throwing a C++
897 exception and so on. This does not decrement the C<ev_depth> value, nor
898 will it clear any outstanding C<EVBREAK_ONE> breaks.
899
900 A flags value of C<EVRUN_NOWAIT> will look for new events, will handle
901 those events and any already outstanding ones, but will not wait and
902 block your process in case there are no events and will return after one
903 iteration of the loop. This is sometimes useful to poll and handle new
904 events while doing lengthy calculations, to keep the program responsive.
905
906 A flags value of C<EVRUN_ONCE> will look for new events (waiting if
907 necessary) and will handle those and any already outstanding ones. It
908 will block your process until at least one new event arrives (which could
909 be an event internal to libev itself, so there is no guarantee that a
910 user-registered callback will be called), and will return after one
911 iteration of the loop.
912
913 This is useful if you are waiting for some external event in conjunction
914 with something not expressible using other libev watchers (i.e. "roll your
915 own C<ev_run>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
916 usually a better approach for this kind of thing.
917
918 Here are the gory details of what C<ev_run> does (this is for your
919 understanding, not a guarantee that things will work exactly like this in
920 future versions):
921
922 - Increment loop depth.
923 - Reset the ev_break status.
924 - Before the first iteration, call any pending watchers.
925 LOOP:
926 - If EVFLAG_FORKCHECK was used, check for a fork.
927 - If a fork was detected (by any means), queue and call all fork watchers.
928 - Queue and call all prepare watchers.
929 - If ev_break was called, goto FINISH.
930 - If we have been forked, detach and recreate the kernel state
931 as to not disturb the other process.
932 - Update the kernel state with all outstanding changes.
933 - Update the "event loop time" (ev_now ()).
934 - Calculate for how long to sleep or block, if at all
935 (active idle watchers, EVRUN_NOWAIT or not having
936 any active watchers at all will result in not sleeping).
937 - Sleep if the I/O and timer collect interval say so.
938 - Increment loop iteration counter.
939 - Block the process, waiting for any events.
940 - Queue all outstanding I/O (fd) events.
941 - Update the "event loop time" (ev_now ()), and do time jump adjustments.
942 - Queue all expired timers.
943 - Queue all expired periodics.
944 - Queue all idle watchers with priority higher than that of pending events.
945 - Queue all check watchers.
946 - Call all queued watchers in reverse order (i.e. check watchers first).
947 Signals and child watchers are implemented as I/O watchers, and will
948 be handled here by queueing them when their watcher gets executed.
949 - If ev_break has been called, or EVRUN_ONCE or EVRUN_NOWAIT
950 were used, or there are no active watchers, goto FINISH, otherwise
951 continue with step LOOP.
952 FINISH:
953 - Reset the ev_break status iff it was EVBREAK_ONE.
954 - Decrement the loop depth.
955 - Return.
956
957 Example: Queue some jobs and then loop until no events are outstanding
958 anymore.
959
960 ... queue jobs here, make sure they register event watchers as long
961 ... as they still have work to do (even an idle watcher will do..)
962 ev_run (my_loop, 0);
963 ... jobs done or somebody called break. yeah!
964
965 =item ev_break (loop, how)
966
967 Can be used to make a call to C<ev_run> return early (but only after it
968 has processed all outstanding events). The C<how> argument must be either
969 C<EVBREAK_ONE>, which will make the innermost C<ev_run> call return, or
970 C<EVBREAK_ALL>, which will make all nested C<ev_run> calls return.
971
972 This "break state" will be cleared on the next call to C<ev_run>.
973
974 It is safe to call C<ev_break> from outside any C<ev_run> calls, too, in
975 which case it will have no effect.
976
977 =item ev_ref (loop)
978
979 =item ev_unref (loop)
980
981 Ref/unref can be used to add or remove a reference count on the event
982 loop: Every watcher keeps one reference, and as long as the reference
983 count is nonzero, C<ev_run> will not return on its own.
984
985 This is useful when you have a watcher that you never intend to
986 unregister, but that nevertheless should not keep C<ev_run> from
987 returning. In such a case, call C<ev_unref> after starting, and C<ev_ref>
988 before stopping it.
989
990 As an example, libev itself uses this for its internal signal pipe: It
991 is not visible to the libev user and should not keep C<ev_run> from
992 exiting if no event watchers registered by it are active. It is also an
993 excellent way to do this for generic recurring timers or from within
994 third-party libraries. Just remember to I<unref after start> and I<ref
995 before stop> (but only if the watcher wasn't active before, or was active
996 before, respectively. Note also that libev might stop watchers itself
997 (e.g. non-repeating timers) in which case you have to C<ev_ref>
998 in the callback).
999
1000 Example: Create a signal watcher, but keep it from keeping C<ev_run>
1001 running when nothing else is active.
1002
1003 ev_signal exitsig;
1004 ev_signal_init (&exitsig, sig_cb, SIGINT);
1005 ev_signal_start (loop, &exitsig);
1006 ev_unref (loop);
1007
1008 Example: For some weird reason, unregister the above signal handler again.
1009
1010 ev_ref (loop);
1011 ev_signal_stop (loop, &exitsig);
1012
1013 =item ev_set_io_collect_interval (loop, ev_tstamp interval)
1014
1015 =item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
1016
1017 These advanced functions influence the time that libev will spend waiting
1018 for events. Both time intervals are by default C<0>, meaning that libev
1019 will try to invoke timer/periodic callbacks and I/O callbacks with minimum
1020 latency.
1021
1022 Setting these to a higher value (the C<interval> I<must> be >= C<0>)
1023 allows libev to delay invocation of I/O and timer/periodic callbacks
1024 to increase efficiency of loop iterations (or to increase power-saving
1025 opportunities).
1026
1027 The idea is that sometimes your program runs just fast enough to handle
1028 one (or very few) event(s) per loop iteration. While this makes the
1029 program responsive, it also wastes a lot of CPU time to poll for new
1030 events, especially with backends like C<select ()> which have a high
1031 overhead for the actual polling but can deliver many events at once.
1032
1033 By setting a higher I<io collect interval> you allow libev to spend more
1034 time collecting I/O events, so you can handle more events per iteration,
1035 at the cost of increasing latency. Timeouts (both C<ev_periodic> and
1036 C<ev_timer>) will not be affected. Setting this to a non-null value will
1037 introduce an additional C<ev_sleep ()> call into most loop iterations. The
1038 sleep time ensures that libev will not poll for I/O events more often then
1039 once per this interval, on average (as long as the host time resolution is
1040 good enough).
1041
1042 Likewise, by setting a higher I<timeout collect interval> you allow libev
1043 to spend more time collecting timeouts, at the expense of increased
1044 latency/jitter/inexactness (the watcher callback will be called
1045 later). C<ev_io> watchers will not be affected. Setting this to a non-null
1046 value will not introduce any overhead in libev.
1047
1048 Many (busy) programs can usually benefit by setting the I/O collect
1049 interval to a value near C<0.1> or so, which is often enough for
1050 interactive servers (of course not for games), likewise for timeouts. It
1051 usually doesn't make much sense to set it to a lower value than C<0.01>,
1052 as this approaches the timing granularity of most systems. Note that if
1053 you do transactions with the outside world and you can't increase the
1054 parallelity, then this setting will limit your transaction rate (if you
1055 need to poll once per transaction and the I/O collect interval is 0.01,
1056 then you can't do more than 100 transactions per second).
1057
1058 Setting the I<timeout collect interval> can improve the opportunity for
1059 saving power, as the program will "bundle" timer callback invocations that
1060 are "near" in time together, by delaying some, thus reducing the number of
1061 times the process sleeps and wakes up again. Another useful technique to
1062 reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
1063 they fire on, say, one-second boundaries only.
1064
1065 Example: we only need 0.1s timeout granularity, and we wish not to poll
1066 more often than 100 times per second:
1067
1068 ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
1069 ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
1070
1071 =item ev_invoke_pending (loop)
1072
1073 This call will simply invoke all pending watchers while resetting their
1074 pending state. Normally, C<ev_run> does this automatically when required,
1075 but when overriding the invoke callback this call comes handy. This
1076 function can be invoked from a watcher - this can be useful for example
1077 when you want to do some lengthy calculation and want to pass further
1078 event handling to another thread (you still have to make sure only one
1079 thread executes within C<ev_invoke_pending> or C<ev_run> of course).
1080
1081 =item int ev_pending_count (loop)
1082
1083 Returns the number of pending watchers - zero indicates that no watchers
1084 are pending.
1085
1086 =item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
1087
1088 This overrides the invoke pending functionality of the loop: Instead of
1089 invoking all pending watchers when there are any, C<ev_run> will call
1090 this callback instead. This is useful, for example, when you want to
1091 invoke the actual watchers inside another context (another thread etc.).
1092
1093 If you want to reset the callback, use C<ev_invoke_pending> as new
1094 callback.
1095
1096 =item ev_set_loop_release_cb (loop, void (*release)(EV_P) throw (), void (*acquire)(EV_P) throw ())
1097
1098 Sometimes you want to share the same loop between multiple threads. This
1099 can be done relatively simply by putting mutex_lock/unlock calls around
1100 each call to a libev function.
1101
1102 However, C<ev_run> can run an indefinite time, so it is not feasible
1103 to wait for it to return. One way around this is to wake up the event
1104 loop via C<ev_break> and C<ev_async_send>, another way is to set these
1105 I<release> and I<acquire> callbacks on the loop.
1106
1107 When set, then C<release> will be called just before the thread is
1108 suspended waiting for new events, and C<acquire> is called just
1109 afterwards.
1110
1111 Ideally, C<release> will just call your mutex_unlock function, and
1112 C<acquire> will just call the mutex_lock function again.
1113
1114 While event loop modifications are allowed between invocations of
1115 C<release> and C<acquire> (that's their only purpose after all), no
1116 modifications done will affect the event loop, i.e. adding watchers will
1117 have no effect on the set of file descriptors being watched, or the time
1118 waited. Use an C<ev_async> watcher to wake up C<ev_run> when you want it
1119 to take note of any changes you made.
1120
1121 In theory, threads executing C<ev_run> will be async-cancel safe between
1122 invocations of C<release> and C<acquire>.
1123
1124 See also the locking example in the C<THREADS> section later in this
1125 document.
1126
1127 =item ev_set_userdata (loop, void *data)
1128
1129 =item void *ev_userdata (loop)
1130
1131 Set and retrieve a single C<void *> associated with a loop. When
1132 C<ev_set_userdata> has never been called, then C<ev_userdata> returns
1133 C<0>.
1134
1135 These two functions can be used to associate arbitrary data with a loop,
1136 and are intended solely for the C<invoke_pending_cb>, C<release> and
1137 C<acquire> callbacks described above, but of course can be (ab-)used for
1138 any other purpose as well.
1139
1140 =item ev_verify (loop)
1141
1142 This function only does something when C<EV_VERIFY> support has been
1143 compiled in, which is the default for non-minimal builds. It tries to go
1144 through all internal structures and checks them for validity. If anything
1145 is found to be inconsistent, it will print an error message to standard
1146 error and call C<abort ()>.
1147
1148 This can be used to catch bugs inside libev itself: under normal
1149 circumstances, this function will never abort as of course libev keeps its
1150 data structures consistent.
1151
1152 =back
1153
1154
1155 =head1 ANATOMY OF A WATCHER
1156
1157 In the following description, uppercase C<TYPE> in names stands for the
1158 watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
1159 watchers and C<ev_io_start> for I/O watchers.
1160
1161 A watcher is an opaque structure that you allocate and register to record
1162 your interest in some event. To make a concrete example, imagine you want
1163 to wait for STDIN to become readable, you would create an C<ev_io> watcher
1164 for that:
1165
1166 static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
1167 {
1168 ev_io_stop (w);
1169 ev_break (loop, EVBREAK_ALL);
1170 }
1171
1172 struct ev_loop *loop = ev_default_loop (0);
1173
1174 ev_io stdin_watcher;
1175
1176 ev_init (&stdin_watcher, my_cb);
1177 ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
1178 ev_io_start (loop, &stdin_watcher);
1179
1180 ev_run (loop, 0);
1181
1182 As you can see, you are responsible for allocating the memory for your
1183 watcher structures (and it is I<usually> a bad idea to do this on the
1184 stack).
1185
1186 Each watcher has an associated watcher structure (called C<struct ev_TYPE>
1187 or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
1188
1189 Each watcher structure must be initialised by a call to C<ev_init (watcher
1190 *, callback)>, which expects a callback to be provided. This callback is
1191 invoked each time the event occurs (or, in the case of I/O watchers, each
1192 time the event loop detects that the file descriptor given is readable
1193 and/or writable).
1194
1195 Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
1196 macro to configure it, with arguments specific to the watcher type. There
1197 is also a macro to combine initialisation and setting in one call: C<<
1198 ev_TYPE_init (watcher *, callback, ...) >>.
1199
1200 To make the watcher actually watch out for events, you have to start it
1201 with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
1202 *) >>), and you can stop watching for events at any time by calling the
1203 corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
1204
1205 As long as your watcher is active (has been started but not stopped) you
1206 must not touch the values stored in it. Most specifically you must never
1207 reinitialise it or call its C<ev_TYPE_set> macro.
1208
1209 Each and every callback receives the event loop pointer as first, the
1210 registered watcher structure as second, and a bitset of received events as
1211 third argument.
1212
1213 The received events usually include a single bit per event type received
1214 (you can receive multiple events at the same time). The possible bit masks
1215 are:
1216
1217 =over 4
1218
1219 =item C<EV_READ>
1220
1221 =item C<EV_WRITE>
1222
1223 The file descriptor in the C<ev_io> watcher has become readable and/or
1224 writable.
1225
1226 =item C<EV_TIMER>
1227
1228 The C<ev_timer> watcher has timed out.
1229
1230 =item C<EV_PERIODIC>
1231
1232 The C<ev_periodic> watcher has timed out.
1233
1234 =item C<EV_SIGNAL>
1235
1236 The signal specified in the C<ev_signal> watcher has been received by a thread.
1237
1238 =item C<EV_CHILD>
1239
1240 The pid specified in the C<ev_child> watcher has received a status change.
1241
1242 =item C<EV_STAT>
1243
1244 The path specified in the C<ev_stat> watcher changed its attributes somehow.
1245
1246 =item C<EV_IDLE>
1247
1248 The C<ev_idle> watcher has determined that you have nothing better to do.
1249
1250 =item C<EV_PREPARE>
1251
1252 =item C<EV_CHECK>
1253
1254 All C<ev_prepare> watchers are invoked just I<before> C<ev_run> starts to
1255 gather new events, and all C<ev_check> watchers are queued (not invoked)
1256 just after C<ev_run> has gathered them, but before it queues any callbacks
1257 for any received events. That means C<ev_prepare> watchers are the last
1258 watchers invoked before the event loop sleeps or polls for new events, and
1259 C<ev_check> watchers will be invoked before any other watchers of the same
1260 or lower priority within an event loop iteration.
1261
1262 Callbacks of both watcher types can start and stop as many watchers as
1263 they want, and all of them will be taken into account (for example, a
1264 C<ev_prepare> watcher might start an idle watcher to keep C<ev_run> from
1265 blocking).
1266
1267 =item C<EV_EMBED>
1268
1269 The embedded event loop specified in the C<ev_embed> watcher needs attention.
1270
1271 =item C<EV_FORK>
1272
1273 The event loop has been resumed in the child process after fork (see
1274 C<ev_fork>).
1275
1276 =item C<EV_CLEANUP>
1277
1278 The event loop is about to be destroyed (see C<ev_cleanup>).
1279
1280 =item C<EV_ASYNC>
1281
1282 The given async watcher has been asynchronously notified (see C<ev_async>).
1283
1284 =item C<EV_CUSTOM>
1285
1286 Not ever sent (or otherwise used) by libev itself, but can be freely used
1287 by libev users to signal watchers (e.g. via C<ev_feed_event>).
1288
1289 =item C<EV_ERROR>
1290
1291 An unspecified error has occurred, the watcher has been stopped. This might
1292 happen because the watcher could not be properly started because libev
1293 ran out of memory, a file descriptor was found to be closed or any other
1294 problem. Libev considers these application bugs.
1295
1296 You best act on it by reporting the problem and somehow coping with the
1297 watcher being stopped. Note that well-written programs should not receive
1298 an error ever, so when your watcher receives it, this usually indicates a
1299 bug in your program.
1300
1301 Libev will usually signal a few "dummy" events together with an error, for
1302 example it might indicate that a fd is readable or writable, and if your
1303 callbacks is well-written it can just attempt the operation and cope with
1304 the error from read() or write(). This will not work in multi-threaded
1305 programs, though, as the fd could already be closed and reused for another
1306 thing, so beware.
1307
1308 =back
1309
1310 =head2 GENERIC WATCHER FUNCTIONS
1311
1312 =over 4
1313
1314 =item C<ev_init> (ev_TYPE *watcher, callback)
1315
1316 This macro initialises the generic portion of a watcher. The contents
1317 of the watcher object can be arbitrary (so C<malloc> will do). Only
1318 the generic parts of the watcher are initialised, you I<need> to call
1319 the type-specific C<ev_TYPE_set> macro afterwards to initialise the
1320 type-specific parts. For each type there is also a C<ev_TYPE_init> macro
1321 which rolls both calls into one.
1322
1323 You can reinitialise a watcher at any time as long as it has been stopped
1324 (or never started) and there are no pending events outstanding.
1325
1326 The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
1327 int revents)>.
1328
1329 Example: Initialise an C<ev_io> watcher in two steps.
1330
1331 ev_io w;
1332 ev_init (&w, my_cb);
1333 ev_io_set (&w, STDIN_FILENO, EV_READ);
1334
1335 =item C<ev_TYPE_set> (ev_TYPE *watcher, [args])
1336
1337 This macro initialises the type-specific parts of a watcher. You need to
1338 call C<ev_init> at least once before you call this macro, but you can
1339 call C<ev_TYPE_set> any number of times. You must not, however, call this
1340 macro on a watcher that is active (it can be pending, however, which is a
1341 difference to the C<ev_init> macro).
1342
1343 Although some watcher types do not have type-specific arguments
1344 (e.g. C<ev_prepare>) you still need to call its C<set> macro.
1345
1346 See C<ev_init>, above, for an example.
1347
1348 =item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
1349
1350 This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
1351 calls into a single call. This is the most convenient method to initialise
1352 a watcher. The same limitations apply, of course.
1353
1354 Example: Initialise and set an C<ev_io> watcher in one step.
1355
1356 ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
1357
1358 =item C<ev_TYPE_start> (loop, ev_TYPE *watcher)
1359
1360 Starts (activates) the given watcher. Only active watchers will receive
1361 events. If the watcher is already active nothing will happen.
1362
1363 Example: Start the C<ev_io> watcher that is being abused as example in this
1364 whole section.
1365
1366 ev_io_start (EV_DEFAULT_UC, &w);
1367
1368 =item C<ev_TYPE_stop> (loop, ev_TYPE *watcher)
1369
1370 Stops the given watcher if active, and clears the pending status (whether
1371 the watcher was active or not).
1372
1373 It is possible that stopped watchers are pending - for example,
1374 non-repeating timers are being stopped when they become pending - but
1375 calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
1376 pending. If you want to free or reuse the memory used by the watcher it is
1377 therefore a good idea to always call its C<ev_TYPE_stop> function.
1378
1379 =item bool ev_is_active (ev_TYPE *watcher)
1380
1381 Returns a true value iff the watcher is active (i.e. it has been started
1382 and not yet been stopped). As long as a watcher is active you must not modify
1383 it.
1384
1385 =item bool ev_is_pending (ev_TYPE *watcher)
1386
1387 Returns a true value iff the watcher is pending, (i.e. it has outstanding
1388 events but its callback has not yet been invoked). As long as a watcher
1389 is pending (but not active) you must not call an init function on it (but
1390 C<ev_TYPE_set> is safe), you must not change its priority, and you must
1391 make sure the watcher is available to libev (e.g. you cannot C<free ()>
1392 it).
1393
1394 =item callback ev_cb (ev_TYPE *watcher)
1395
1396 Returns the callback currently set on the watcher.
1397
1398 =item ev_set_cb (ev_TYPE *watcher, callback)
1399
1400 Change the callback. You can change the callback at virtually any time
1401 (modulo threads).
1402
1403 =item ev_set_priority (ev_TYPE *watcher, int priority)
1404
1405 =item int ev_priority (ev_TYPE *watcher)
1406
1407 Set and query the priority of the watcher. The priority is a small
1408 integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
1409 (default: C<-2>). Pending watchers with higher priority will be invoked
1410 before watchers with lower priority, but priority will not keep watchers
1411 from being executed (except for C<ev_idle> watchers).
1412
1413 If you need to suppress invocation when higher priority events are pending
1414 you need to look at C<ev_idle> watchers, which provide this functionality.
1415
1416 You I<must not> change the priority of a watcher as long as it is active or
1417 pending.
1418
1419 Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
1420 fine, as long as you do not mind that the priority value you query might
1421 or might not have been clamped to the valid range.
1422
1423 The default priority used by watchers when no priority has been set is
1424 always C<0>, which is supposed to not be too high and not be too low :).
1425
1426 See L</WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
1427 priorities.
1428
1429 =item ev_invoke (loop, ev_TYPE *watcher, int revents)
1430
1431 Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
1432 C<loop> nor C<revents> need to be valid as long as the watcher callback
1433 can deal with that fact, as both are simply passed through to the
1434 callback.
1435
1436 =item int ev_clear_pending (loop, ev_TYPE *watcher)
1437
1438 If the watcher is pending, this function clears its pending status and
1439 returns its C<revents> bitset (as if its callback was invoked). If the
1440 watcher isn't pending it does nothing and returns C<0>.
1441
1442 Sometimes it can be useful to "poll" a watcher instead of waiting for its
1443 callback to be invoked, which can be accomplished with this function.
1444
1445 =item ev_feed_event (loop, ev_TYPE *watcher, int revents)
1446
1447 Feeds the given event set into the event loop, as if the specified event
1448 had happened for the specified watcher (which must be a pointer to an
1449 initialised but not necessarily started event watcher). Obviously you must
1450 not free the watcher as long as it has pending events.
1451
1452 Stopping the watcher, letting libev invoke it, or calling
1453 C<ev_clear_pending> will clear the pending event, even if the watcher was
1454 not started in the first place.
1455
1456 See also C<ev_feed_fd_event> and C<ev_feed_signal_event> for related
1457 functions that do not need a watcher.
1458
1459 =back
1460
1461 See also the L</ASSOCIATING CUSTOM DATA WITH A WATCHER> and L</BUILDING YOUR
1462 OWN COMPOSITE WATCHERS> idioms.
1463
1464 =head2 WATCHER STATES
1465
1466 There are various watcher states mentioned throughout this manual -
1467 active, pending and so on. In this section these states and the rules to
1468 transition between them will be described in more detail - and while these
1469 rules might look complicated, they usually do "the right thing".
1470
1471 =over 4
1472
1473 =item initialised
1474
1475 Before a watcher can be registered with the event loop it has to be
1476 initialised. This can be done with a call to C<ev_TYPE_init>, or calls to
1477 C<ev_init> followed by the watcher-specific C<ev_TYPE_set> function.
1478
1479 In this state it is simply some block of memory that is suitable for
1480 use in an event loop. It can be moved around, freed, reused etc. at
1481 will - as long as you either keep the memory contents intact, or call
1482 C<ev_TYPE_init> again.
1483
1484 =item started/running/active
1485
1486 Once a watcher has been started with a call to C<ev_TYPE_start> it becomes
1487 property of the event loop, and is actively waiting for events. While in
1488 this state it cannot be accessed (except in a few documented ways), moved,
1489 freed or anything else - the only legal thing is to keep a pointer to it,
1490 and call libev functions on it that are documented to work on active watchers.
1491
1492 =item pending
1493
1494 If a watcher is active and libev determines that an event it is interested
1495 in has occurred (such as a timer expiring), it will become pending. It will
1496 stay in this pending state until either it is stopped or its callback is
1497 about to be invoked, so it is not normally pending inside the watcher
1498 callback.
1499
1500 The watcher might or might not be active while it is pending (for example,
1501 an expired non-repeating timer can be pending but no longer active). If it
1502 is stopped, it can be freely accessed (e.g. by calling C<ev_TYPE_set>),
1503 but it is still property of the event loop at this time, so cannot be
1504 moved, freed or reused. And if it is active the rules described in the
1505 previous item still apply.
1506
1507 It is also possible to feed an event on a watcher that is not active (e.g.
1508 via C<ev_feed_event>), in which case it becomes pending without being
1509 active.
1510
1511 =item stopped
1512
1513 A watcher can be stopped implicitly by libev (in which case it might still
1514 be pending), or explicitly by calling its C<ev_TYPE_stop> function. The
1515 latter will clear any pending state the watcher might be in, regardless
1516 of whether it was active or not, so stopping a watcher explicitly before
1517 freeing it is often a good idea.
1518
1519 While stopped (and not pending) the watcher is essentially in the
1520 initialised state, that is, it can be reused, moved, modified in any way
1521 you wish (but when you trash the memory block, you need to C<ev_TYPE_init>
1522 it again).
1523
1524 =back
1525
1526 =head2 WATCHER PRIORITY MODELS
1527
1528 Many event loops support I<watcher priorities>, which are usually small
1529 integers that influence the ordering of event callback invocation
1530 between watchers in some way, all else being equal.
1531
1532 In libev, Watcher priorities can be set using C<ev_set_priority>. See its
1533 description for the more technical details such as the actual priority
1534 range.
1535
1536 There are two common ways how these these priorities are being interpreted
1537 by event loops:
1538
1539 In the more common lock-out model, higher priorities "lock out" invocation
1540 of lower priority watchers, which means as long as higher priority
1541 watchers receive events, lower priority watchers are not being invoked.
1542
1543 The less common only-for-ordering model uses priorities solely to order
1544 callback invocation within a single event loop iteration: Higher priority
1545 watchers are invoked before lower priority ones, but they all get invoked
1546 before polling for new events.
1547
1548 Libev uses the second (only-for-ordering) model for all its watchers
1549 except for idle watchers (which use the lock-out model).
1550
1551 The rationale behind this is that implementing the lock-out model for
1552 watchers is not well supported by most kernel interfaces, and most event
1553 libraries will just poll for the same events again and again as long as
1554 their callbacks have not been executed, which is very inefficient in the
1555 common case of one high-priority watcher locking out a mass of lower
1556 priority ones.
1557
1558 Static (ordering) priorities are most useful when you have two or more
1559 watchers handling the same resource: a typical usage example is having an
1560 C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
1561 timeouts. Under load, data might be received while the program handles
1562 other jobs, but since timers normally get invoked first, the timeout
1563 handler will be executed before checking for data. In that case, giving
1564 the timer a lower priority than the I/O watcher ensures that I/O will be
1565 handled first even under adverse conditions (which is usually, but not
1566 always, what you want).
1567
1568 Since idle watchers use the "lock-out" model, meaning that idle watchers
1569 will only be executed when no same or higher priority watchers have
1570 received events, they can be used to implement the "lock-out" model when
1571 required.
1572
1573 For example, to emulate how many other event libraries handle priorities,
1574 you can associate an C<ev_idle> watcher to each such watcher, and in
1575 the normal watcher callback, you just start the idle watcher. The real
1576 processing is done in the idle watcher callback. This causes libev to
1577 continuously poll and process kernel event data for the watcher, but when
1578 the lock-out case is known to be rare (which in turn is rare :), this is
1579 workable.
1580
1581 Usually, however, the lock-out model implemented that way will perform
1582 miserably under the type of load it was designed to handle. In that case,
1583 it might be preferable to stop the real watcher before starting the
1584 idle watcher, so the kernel will not have to process the event in case
1585 the actual processing will be delayed for considerable time.
1586
1587 Here is an example of an I/O watcher that should run at a strictly lower
1588 priority than the default, and which should only process data when no
1589 other events are pending:
1590
1591 ev_idle idle; // actual processing watcher
1592 ev_io io; // actual event watcher
1593
1594 static void
1595 io_cb (EV_P_ ev_io *w, int revents)
1596 {
1597 // stop the I/O watcher, we received the event, but
1598 // are not yet ready to handle it.
1599 ev_io_stop (EV_A_ w);
1600
1601 // start the idle watcher to handle the actual event.
1602 // it will not be executed as long as other watchers
1603 // with the default priority are receiving events.
1604 ev_idle_start (EV_A_ &idle);
1605 }
1606
1607 static void
1608 idle_cb (EV_P_ ev_idle *w, int revents)
1609 {
1610 // actual processing
1611 read (STDIN_FILENO, ...);
1612
1613 // have to start the I/O watcher again, as
1614 // we have handled the event
1615 ev_io_start (EV_P_ &io);
1616 }
1617
1618 // initialisation
1619 ev_idle_init (&idle, idle_cb);
1620 ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
1621 ev_io_start (EV_DEFAULT_ &io);
1622
1623 In the "real" world, it might also be beneficial to start a timer, so that
1624 low-priority connections can not be locked out forever under load. This
1625 enables your program to keep a lower latency for important connections
1626 during short periods of high load, while not completely locking out less
1627 important ones.
1628
1629
1630 =head1 WATCHER TYPES
1631
1632 This section describes each watcher in detail, but will not repeat
1633 information given in the last section. Any initialisation/set macros,
1634 functions and members specific to the watcher type are explained.
1635
1636 Members are additionally marked with either I<[read-only]>, meaning that,
1637 while the watcher is active, you can look at the member and expect some
1638 sensible content, but you must not modify it (you can modify it while the
1639 watcher is stopped to your hearts content), or I<[read-write]>, which
1640 means you can expect it to have some sensible content while the watcher
1641 is active, but you can also modify it. Modifying it may not do something
1642 sensible or take immediate effect (or do anything at all), but libev will
1643 not crash or malfunction in any way.
1644
1645
1646 =head2 C<ev_io> - is this file descriptor readable or writable?
1647
1648 I/O watchers check whether a file descriptor is readable or writable
1649 in each iteration of the event loop, or, more precisely, when reading
1650 would not block the process and writing would at least be able to write
1651 some data. This behaviour is called level-triggering because you keep
1652 receiving events as long as the condition persists. Remember you can stop
1653 the watcher if you don't want to act on the event and neither want to
1654 receive future events.
1655
1656 In general you can register as many read and/or write event watchers per
1657 fd as you want (as long as you don't confuse yourself). Setting all file
1658 descriptors to non-blocking mode is also usually a good idea (but not
1659 required if you know what you are doing).
1660
1661 Another thing you have to watch out for is that it is quite easy to
1662 receive "spurious" readiness notifications, that is, your callback might
1663 be called with C<EV_READ> but a subsequent C<read>(2) will actually block
1664 because there is no data. It is very easy to get into this situation even
1665 with a relatively standard program structure. Thus it is best to always
1666 use non-blocking I/O: An extra C<read>(2) returning C<EAGAIN> is far
1667 preferable to a program hanging until some data arrives.
1668
1669 If you cannot run the fd in non-blocking mode (for example you should
1670 not play around with an Xlib connection), then you have to separately
1671 re-test whether a file descriptor is really ready with a known-to-be good
1672 interface such as poll (fortunately in the case of Xlib, it already does
1673 this on its own, so its quite safe to use). Some people additionally
1674 use C<SIGALRM> and an interval timer, just to be sure you won't block
1675 indefinitely.
1676
1677 But really, best use non-blocking mode.
1678
1679 =head3 The special problem of disappearing file descriptors
1680
1681 Some backends (e.g. kqueue, epoll, linuxaio) need to be told about closing
1682 a file descriptor (either due to calling C<close> explicitly or any other
1683 means, such as C<dup2>). The reason is that you register interest in some
1684 file descriptor, but when it goes away, the operating system will silently
1685 drop this interest. If another file descriptor with the same number then
1686 is registered with libev, there is no efficient way to see that this is,
1687 in fact, a different file descriptor.
1688
1689 To avoid having to explicitly tell libev about such cases, libev follows
1690 the following policy: Each time C<ev_io_set> is being called, libev
1691 will assume that this is potentially a new file descriptor, otherwise
1692 it is assumed that the file descriptor stays the same. That means that
1693 you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
1694 descriptor even if the file descriptor number itself did not change.
1695
1696 This is how one would do it normally anyway, the important point is that
1697 the libev application should not optimise around libev but should leave
1698 optimisations to libev.
1699
1700 =head3 The special problem of dup'ed file descriptors
1701
1702 Some backends (e.g. epoll), cannot register events for file descriptors,
1703 but only events for the underlying file descriptions. That means when you
1704 have C<dup ()>'ed file descriptors or weirder constellations, and register
1705 events for them, only one file descriptor might actually receive events.
1706
1707 There is no workaround possible except not registering events
1708 for potentially C<dup ()>'ed file descriptors, or to resort to
1709 C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
1710
1711 =head3 The special problem of files
1712
1713 Many people try to use C<select> (or libev) on file descriptors
1714 representing files, and expect it to become ready when their program
1715 doesn't block on disk accesses (which can take a long time on their own).
1716
1717 However, this cannot ever work in the "expected" way - you get a readiness
1718 notification as soon as the kernel knows whether and how much data is
1719 there, and in the case of open files, that's always the case, so you
1720 always get a readiness notification instantly, and your read (or possibly
1721 write) will still block on the disk I/O.
1722
1723 Another way to view it is that in the case of sockets, pipes, character
1724 devices and so on, there is another party (the sender) that delivers data
1725 on its own, but in the case of files, there is no such thing: the disk
1726 will not send data on its own, simply because it doesn't know what you
1727 wish to read - you would first have to request some data.
1728
1729 Since files are typically not-so-well supported by advanced notification
1730 mechanism, libev tries hard to emulate POSIX behaviour with respect
1731 to files, even though you should not use it. The reason for this is
1732 convenience: sometimes you want to watch STDIN or STDOUT, which is
1733 usually a tty, often a pipe, but also sometimes files or special devices
1734 (for example, C<epoll> on Linux works with F</dev/random> but not with
1735 F</dev/urandom>), and even though the file might better be served with
1736 asynchronous I/O instead of with non-blocking I/O, it is still useful when
1737 it "just works" instead of freezing.
1738
1739 So avoid file descriptors pointing to files when you know it (e.g. use
1740 libeio), but use them when it is convenient, e.g. for STDIN/STDOUT, or
1741 when you rarely read from a file instead of from a socket, and want to
1742 reuse the same code path.
1743
1744 =head3 The special problem of fork
1745
1746 Some backends (epoll, kqueue, probably linuxaio) do not support C<fork ()>
1747 at all or exhibit useless behaviour. Libev fully supports fork, but needs
1748 to be told about it in the child if you want to continue to use it in the
1749 child.
1750
1751 To support fork in your child processes, you have to call C<ev_loop_fork
1752 ()> after a fork in the child, enable C<EVFLAG_FORKCHECK>, or resort to
1753 C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
1754
1755 =head3 The special problem of SIGPIPE
1756
1757 While not really specific to libev, it is easy to forget about C<SIGPIPE>:
1758 when writing to a pipe whose other end has been closed, your program gets
1759 sent a SIGPIPE, which, by default, aborts your program. For most programs
1760 this is sensible behaviour, for daemons, this is usually undesirable.
1761
1762 So when you encounter spurious, unexplained daemon exits, make sure you
1763 ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
1764 somewhere, as that would have given you a big clue).
1765
1766 =head3 The special problem of accept()ing when you can't
1767
1768 Many implementations of the POSIX C<accept> function (for example,
1769 found in post-2004 Linux) have the peculiar behaviour of not removing a
1770 connection from the pending queue in all error cases.
1771
1772 For example, larger servers often run out of file descriptors (because
1773 of resource limits), causing C<accept> to fail with C<ENFILE> but not
1774 rejecting the connection, leading to libev signalling readiness on
1775 the next iteration again (the connection still exists after all), and
1776 typically causing the program to loop at 100% CPU usage.
1777
1778 Unfortunately, the set of errors that cause this issue differs between
1779 operating systems, there is usually little the app can do to remedy the
1780 situation, and no known thread-safe method of removing the connection to
1781 cope with overload is known (to me).
1782
1783 One of the easiest ways to handle this situation is to just ignore it
1784 - when the program encounters an overload, it will just loop until the
1785 situation is over. While this is a form of busy waiting, no OS offers an
1786 event-based way to handle this situation, so it's the best one can do.
1787
1788 A better way to handle the situation is to log any errors other than
1789 C<EAGAIN> and C<EWOULDBLOCK>, making sure not to flood the log with such
1790 messages, and continue as usual, which at least gives the user an idea of
1791 what could be wrong ("raise the ulimit!"). For extra points one could stop
1792 the C<ev_io> watcher on the listening fd "for a while", which reduces CPU
1793 usage.
1794
1795 If your program is single-threaded, then you could also keep a dummy file
1796 descriptor for overload situations (e.g. by opening F</dev/null>), and
1797 when you run into C<ENFILE> or C<EMFILE>, close it, run C<accept>,
1798 close that fd, and create a new dummy fd. This will gracefully refuse
1799 clients under typical overload conditions.
1800
1801 The last way to handle it is to simply log the error and C<exit>, as
1802 is often done with C<malloc> failures, but this results in an easy
1803 opportunity for a DoS attack.
1804
1805 =head3 Watcher-Specific Functions
1806
1807 =over 4
1808
1809 =item ev_io_init (ev_io *, callback, int fd, int events)
1810
1811 =item ev_io_set (ev_io *, int fd, int events)
1812
1813 Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
1814 receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
1815 C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
1816
1817 =item int fd [read-only]
1818
1819 The file descriptor being watched.
1820
1821 =item int events [read-only]
1822
1823 The events being watched.
1824
1825 =back
1826
1827 =head3 Examples
1828
1829 Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
1830 readable, but only once. Since it is likely line-buffered, you could
1831 attempt to read a whole line in the callback.
1832
1833 static void
1834 stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
1835 {
1836 ev_io_stop (loop, w);
1837 .. read from stdin here (or from w->fd) and handle any I/O errors
1838 }
1839
1840 ...
1841 struct ev_loop *loop = ev_default_init (0);
1842 ev_io stdin_readable;
1843 ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
1844 ev_io_start (loop, &stdin_readable);
1845 ev_run (loop, 0);
1846
1847
1848 =head2 C<ev_timer> - relative and optionally repeating timeouts
1849
1850 Timer watchers are simple relative timers that generate an event after a
1851 given time, and optionally repeating in regular intervals after that.
1852
1853 The timers are based on real time, that is, if you register an event that
1854 times out after an hour and you reset your system clock to January last
1855 year, it will still time out after (roughly) one hour. "Roughly" because
1856 detecting time jumps is hard, and some inaccuracies are unavoidable (the
1857 monotonic clock option helps a lot here).
1858
1859 The callback is guaranteed to be invoked only I<after> its timeout has
1860 passed (not I<at>, so on systems with very low-resolution clocks this
1861 might introduce a small delay, see "the special problem of being too
1862 early", below). If multiple timers become ready during the same loop
1863 iteration then the ones with earlier time-out values are invoked before
1864 ones of the same priority with later time-out values (but this is no
1865 longer true when a callback calls C<ev_run> recursively).
1866
1867 =head3 Be smart about timeouts
1868
1869 Many real-world problems involve some kind of timeout, usually for error
1870 recovery. A typical example is an HTTP request - if the other side hangs,
1871 you want to raise some error after a while.
1872
1873 What follows are some ways to handle this problem, from obvious and
1874 inefficient to smart and efficient.
1875
1876 In the following, a 60 second activity timeout is assumed - a timeout that
1877 gets reset to 60 seconds each time there is activity (e.g. each time some
1878 data or other life sign was received).
1879
1880 =over 4
1881
1882 =item 1. Use a timer and stop, reinitialise and start it on activity.
1883
1884 This is the most obvious, but not the most simple way: In the beginning,
1885 start the watcher:
1886
1887 ev_timer_init (timer, callback, 60., 0.);
1888 ev_timer_start (loop, timer);
1889
1890 Then, each time there is some activity, C<ev_timer_stop> it, initialise it
1891 and start it again:
1892
1893 ev_timer_stop (loop, timer);
1894 ev_timer_set (timer, 60., 0.);
1895 ev_timer_start (loop, timer);
1896
1897 This is relatively simple to implement, but means that each time there is
1898 some activity, libev will first have to remove the timer from its internal
1899 data structure and then add it again. Libev tries to be fast, but it's
1900 still not a constant-time operation.
1901
1902 =item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
1903
1904 This is the easiest way, and involves using C<ev_timer_again> instead of
1905 C<ev_timer_start>.
1906
1907 To implement this, configure an C<ev_timer> with a C<repeat> value
1908 of C<60> and then call C<ev_timer_again> at start and each time you
1909 successfully read or write some data. If you go into an idle state where
1910 you do not expect data to travel on the socket, you can C<ev_timer_stop>
1911 the timer, and C<ev_timer_again> will automatically restart it if need be.
1912
1913 That means you can ignore both the C<ev_timer_start> function and the
1914 C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
1915 member and C<ev_timer_again>.
1916
1917 At start:
1918
1919 ev_init (timer, callback);
1920 timer->repeat = 60.;
1921 ev_timer_again (loop, timer);
1922
1923 Each time there is some activity:
1924
1925 ev_timer_again (loop, timer);
1926
1927 It is even possible to change the time-out on the fly, regardless of
1928 whether the watcher is active or not:
1929
1930 timer->repeat = 30.;
1931 ev_timer_again (loop, timer);
1932
1933 This is slightly more efficient then stopping/starting the timer each time
1934 you want to modify its timeout value, as libev does not have to completely
1935 remove and re-insert the timer from/into its internal data structure.
1936
1937 It is, however, even simpler than the "obvious" way to do it.
1938
1939 =item 3. Let the timer time out, but then re-arm it as required.
1940
1941 This method is more tricky, but usually most efficient: Most timeouts are
1942 relatively long compared to the intervals between other activity - in
1943 our example, within 60 seconds, there are usually many I/O events with
1944 associated activity resets.
1945
1946 In this case, it would be more efficient to leave the C<ev_timer> alone,
1947 but remember the time of last activity, and check for a real timeout only
1948 within the callback:
1949
1950 ev_tstamp timeout = 60.;
1951 ev_tstamp last_activity; // time of last activity
1952 ev_timer timer;
1953
1954 static void
1955 callback (EV_P_ ev_timer *w, int revents)
1956 {
1957 // calculate when the timeout would happen
1958 ev_tstamp after = last_activity - ev_now (EV_A) + timeout;
1959
1960 // if negative, it means we the timeout already occurred
1961 if (after < 0.)
1962 {
1963 // timeout occurred, take action
1964 }
1965 else
1966 {
1967 // callback was invoked, but there was some recent
1968 // activity. simply restart the timer to time out
1969 // after "after" seconds, which is the earliest time
1970 // the timeout can occur.
1971 ev_timer_set (w, after, 0.);
1972 ev_timer_start (EV_A_ w);
1973 }
1974 }
1975
1976 To summarise the callback: first calculate in how many seconds the
1977 timeout will occur (by calculating the absolute time when it would occur,
1978 C<last_activity + timeout>, and subtracting the current time, C<ev_now
1979 (EV_A)> from that).
1980
1981 If this value is negative, then we are already past the timeout, i.e. we
1982 timed out, and need to do whatever is needed in this case.
1983
1984 Otherwise, we now the earliest time at which the timeout would trigger,
1985 and simply start the timer with this timeout value.
1986
1987 In other words, each time the callback is invoked it will check whether
1988 the timeout occurred. If not, it will simply reschedule itself to check
1989 again at the earliest time it could time out. Rinse. Repeat.
1990
1991 This scheme causes more callback invocations (about one every 60 seconds
1992 minus half the average time between activity), but virtually no calls to
1993 libev to change the timeout.
1994
1995 To start the machinery, simply initialise the watcher and set
1996 C<last_activity> to the current time (meaning there was some activity just
1997 now), then call the callback, which will "do the right thing" and start
1998 the timer:
1999
2000 last_activity = ev_now (EV_A);
2001 ev_init (&timer, callback);
2002 callback (EV_A_ &timer, 0);
2003
2004 When there is some activity, simply store the current time in
2005 C<last_activity>, no libev calls at all:
2006
2007 if (activity detected)
2008 last_activity = ev_now (EV_A);
2009
2010 When your timeout value changes, then the timeout can be changed by simply
2011 providing a new value, stopping the timer and calling the callback, which
2012 will again do the right thing (for example, time out immediately :).
2013
2014 timeout = new_value;
2015 ev_timer_stop (EV_A_ &timer);
2016 callback (EV_A_ &timer, 0);
2017
2018 This technique is slightly more complex, but in most cases where the
2019 time-out is unlikely to be triggered, much more efficient.
2020
2021 =item 4. Wee, just use a double-linked list for your timeouts.
2022
2023 If there is not one request, but many thousands (millions...), all
2024 employing some kind of timeout with the same timeout value, then one can
2025 do even better:
2026
2027 When starting the timeout, calculate the timeout value and put the timeout
2028 at the I<end> of the list.
2029
2030 Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
2031 the list is expected to fire (for example, using the technique #3).
2032
2033 When there is some activity, remove the timer from the list, recalculate
2034 the timeout, append it to the end of the list again, and make sure to
2035 update the C<ev_timer> if it was taken from the beginning of the list.
2036
2037 This way, one can manage an unlimited number of timeouts in O(1) time for
2038 starting, stopping and updating the timers, at the expense of a major
2039 complication, and having to use a constant timeout. The constant timeout
2040 ensures that the list stays sorted.
2041
2042 =back
2043
2044 So which method the best?
2045
2046 Method #2 is a simple no-brain-required solution that is adequate in most
2047 situations. Method #3 requires a bit more thinking, but handles many cases
2048 better, and isn't very complicated either. In most case, choosing either
2049 one is fine, with #3 being better in typical situations.
2050
2051 Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
2052 rather complicated, but extremely efficient, something that really pays
2053 off after the first million or so of active timers, i.e. it's usually
2054 overkill :)
2055
2056 =head3 The special problem of being too early
2057
2058 If you ask a timer to call your callback after three seconds, then
2059 you expect it to be invoked after three seconds - but of course, this
2060 cannot be guaranteed to infinite precision. Less obviously, it cannot be
2061 guaranteed to any precision by libev - imagine somebody suspending the
2062 process with a STOP signal for a few hours for example.
2063
2064 So, libev tries to invoke your callback as soon as possible I<after> the
2065 delay has occurred, but cannot guarantee this.
2066
2067 A less obvious failure mode is calling your callback too early: many event
2068 loops compare timestamps with a "elapsed delay >= requested delay", but
2069 this can cause your callback to be invoked much earlier than you would
2070 expect.
2071
2072 To see why, imagine a system with a clock that only offers full second
2073 resolution (think windows if you can't come up with a broken enough OS
2074 yourself). If you schedule a one-second timer at the time 500.9, then the
2075 event loop will schedule your timeout to elapse at a system time of 500
2076 (500.9 truncated to the resolution) + 1, or 501.
2077
2078 If an event library looks at the timeout 0.1s later, it will see "501 >=
2079 501" and invoke the callback 0.1s after it was started, even though a
2080 one-second delay was requested - this is being "too early", despite best
2081 intentions.
2082
2083 This is the reason why libev will never invoke the callback if the elapsed
2084 delay equals the requested delay, but only when the elapsed delay is
2085 larger than the requested delay. In the example above, libev would only invoke
2086 the callback at system time 502, or 1.1s after the timer was started.
2087
2088 So, while libev cannot guarantee that your callback will be invoked
2089 exactly when requested, it I<can> and I<does> guarantee that the requested
2090 delay has actually elapsed, or in other words, it always errs on the "too
2091 late" side of things.
2092
2093 =head3 The special problem of time updates
2094
2095 Establishing the current time is a costly operation (it usually takes
2096 at least one system call): EV therefore updates its idea of the current
2097 time only before and after C<ev_run> collects new events, which causes a
2098 growing difference between C<ev_now ()> and C<ev_time ()> when handling
2099 lots of events in one iteration.
2100
2101 The relative timeouts are calculated relative to the C<ev_now ()>
2102 time. This is usually the right thing as this timestamp refers to the time
2103 of the event triggering whatever timeout you are modifying/starting. If
2104 you suspect event processing to be delayed and you I<need> to base the
2105 timeout on the current time, use something like the following to adjust
2106 for it:
2107
2108 ev_timer_set (&timer, after + (ev_time () - ev_now ()), 0.);
2109
2110 If the event loop is suspended for a long time, you can also force an
2111 update of the time returned by C<ev_now ()> by calling C<ev_now_update
2112 ()>, although that will push the event time of all outstanding events
2113 further into the future.
2114
2115 =head3 The special problem of unsynchronised clocks
2116
2117 Modern systems have a variety of clocks - libev itself uses the normal
2118 "wall clock" clock and, if available, the monotonic clock (to avoid time
2119 jumps).
2120
2121 Neither of these clocks is synchronised with each other or any other clock
2122 on the system, so C<ev_time ()> might return a considerably different time
2123 than C<gettimeofday ()> or C<time ()>. On a GNU/Linux system, for example,
2124 a call to C<gettimeofday> might return a second count that is one higher
2125 than a directly following call to C<time>.
2126
2127 The moral of this is to only compare libev-related timestamps with
2128 C<ev_time ()> and C<ev_now ()>, at least if you want better precision than
2129 a second or so.
2130
2131 One more problem arises due to this lack of synchronisation: if libev uses
2132 the system monotonic clock and you compare timestamps from C<ev_time>
2133 or C<ev_now> from when you started your timer and when your callback is
2134 invoked, you will find that sometimes the callback is a bit "early".
2135
2136 This is because C<ev_timer>s work in real time, not wall clock time, so
2137 libev makes sure your callback is not invoked before the delay happened,
2138 I<measured according to the real time>, not the system clock.
2139
2140 If your timeouts are based on a physical timescale (e.g. "time out this
2141 connection after 100 seconds") then this shouldn't bother you as it is
2142 exactly the right behaviour.
2143
2144 If you want to compare wall clock/system timestamps to your timers, then
2145 you need to use C<ev_periodic>s, as these are based on the wall clock
2146 time, where your comparisons will always generate correct results.
2147
2148 =head3 The special problems of suspended animation
2149
2150 When you leave the server world it is quite customary to hit machines that
2151 can suspend/hibernate - what happens to the clocks during such a suspend?
2152
2153 Some quick tests made with a Linux 2.6.28 indicate that a suspend freezes
2154 all processes, while the clocks (C<times>, C<CLOCK_MONOTONIC>) continue
2155 to run until the system is suspended, but they will not advance while the
2156 system is suspended. That means, on resume, it will be as if the program
2157 was frozen for a few seconds, but the suspend time will not be counted
2158 towards C<ev_timer> when a monotonic clock source is used. The real time
2159 clock advanced as expected, but if it is used as sole clocksource, then a
2160 long suspend would be detected as a time jump by libev, and timers would
2161 be adjusted accordingly.
2162
2163 I would not be surprised to see different behaviour in different between
2164 operating systems, OS versions or even different hardware.
2165
2166 The other form of suspend (job control, or sending a SIGSTOP) will see a
2167 time jump in the monotonic clocks and the realtime clock. If the program
2168 is suspended for a very long time, and monotonic clock sources are in use,
2169 then you can expect C<ev_timer>s to expire as the full suspension time
2170 will be counted towards the timers. When no monotonic clock source is in
2171 use, then libev will again assume a timejump and adjust accordingly.
2172
2173 It might be beneficial for this latter case to call C<ev_suspend>
2174 and C<ev_resume> in code that handles C<SIGTSTP>, to at least get
2175 deterministic behaviour in this case (you can do nothing against
2176 C<SIGSTOP>).
2177
2178 =head3 Watcher-Specific Functions and Data Members
2179
2180 =over 4
2181
2182 =item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
2183
2184 =item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
2185
2186 Configure the timer to trigger after C<after> seconds (fractional and
2187 negative values are supported). If C<repeat> is C<0.>, then it will
2188 automatically be stopped once the timeout is reached. If it is positive,
2189 then the timer will automatically be configured to trigger again C<repeat>
2190 seconds later, again, and again, until stopped manually.
2191
2192 The timer itself will do a best-effort at avoiding drift, that is, if
2193 you configure a timer to trigger every 10 seconds, then it will normally
2194 trigger at exactly 10 second intervals. If, however, your program cannot
2195 keep up with the timer (because it takes longer than those 10 seconds to
2196 do stuff) the timer will not fire more than once per event loop iteration.
2197
2198 =item ev_timer_again (loop, ev_timer *)
2199
2200 This will act as if the timer timed out, and restarts it again if it is
2201 repeating. It basically works like calling C<ev_timer_stop>, updating the
2202 timeout to the C<repeat> value and calling C<ev_timer_start>.
2203
2204 The exact semantics are as in the following rules, all of which will be
2205 applied to the watcher:
2206
2207 =over 4
2208
2209 =item If the timer is pending, the pending status is always cleared.
2210
2211 =item If the timer is started but non-repeating, stop it (as if it timed
2212 out, without invoking it).
2213
2214 =item If the timer is repeating, make the C<repeat> value the new timeout
2215 and start the timer, if necessary.
2216
2217 =back
2218
2219 This sounds a bit complicated, see L</Be smart about timeouts>, above, for a
2220 usage example.
2221
2222 =item ev_tstamp ev_timer_remaining (loop, ev_timer *)
2223
2224 Returns the remaining time until a timer fires. If the timer is active,
2225 then this time is relative to the current event loop time, otherwise it's
2226 the timeout value currently configured.
2227
2228 That is, after an C<ev_timer_set (w, 5, 7)>, C<ev_timer_remaining> returns
2229 C<5>. When the timer is started and one second passes, C<ev_timer_remaining>
2230 will return C<4>. When the timer expires and is restarted, it will return
2231 roughly C<7> (likely slightly less as callback invocation takes some time,
2232 too), and so on.
2233
2234 =item ev_tstamp repeat [read-write]
2235
2236 The current C<repeat> value. Will be used each time the watcher times out
2237 or C<ev_timer_again> is called, and determines the next timeout (if any),
2238 which is also when any modifications are taken into account.
2239
2240 =back
2241
2242 =head3 Examples
2243
2244 Example: Create a timer that fires after 60 seconds.
2245
2246 static void
2247 one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
2248 {
2249 .. one minute over, w is actually stopped right here
2250 }
2251
2252 ev_timer mytimer;
2253 ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
2254 ev_timer_start (loop, &mytimer);
2255
2256 Example: Create a timeout timer that times out after 10 seconds of
2257 inactivity.
2258
2259 static void
2260 timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
2261 {
2262 .. ten seconds without any activity
2263 }
2264
2265 ev_timer mytimer;
2266 ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
2267 ev_timer_again (&mytimer); /* start timer */
2268 ev_run (loop, 0);
2269
2270 // and in some piece of code that gets executed on any "activity":
2271 // reset the timeout to start ticking again at 10 seconds
2272 ev_timer_again (&mytimer);
2273
2274
2275 =head2 C<ev_periodic> - to cron or not to cron?
2276
2277 Periodic watchers are also timers of a kind, but they are very versatile
2278 (and unfortunately a bit complex).
2279
2280 Unlike C<ev_timer>, periodic watchers are not based on real time (or
2281 relative time, the physical time that passes) but on wall clock time
2282 (absolute time, the thing you can read on your calendar or clock). The
2283 difference is that wall clock time can run faster or slower than real
2284 time, and time jumps are not uncommon (e.g. when you adjust your
2285 wrist-watch).
2286
2287 You can tell a periodic watcher to trigger after some specific point
2288 in time: for example, if you tell a periodic watcher to trigger "in 10
2289 seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
2290 not a delay) and then reset your system clock to January of the previous
2291 year, then it will take a year or more to trigger the event (unlike an
2292 C<ev_timer>, which would still trigger roughly 10 seconds after starting
2293 it, as it uses a relative timeout).
2294
2295 C<ev_periodic> watchers can also be used to implement vastly more complex
2296 timers, such as triggering an event on each "midnight, local time", or
2297 other complicated rules. This cannot easily be done with C<ev_timer>
2298 watchers, as those cannot react to time jumps.
2299
2300 As with timers, the callback is guaranteed to be invoked only when the
2301 point in time where it is supposed to trigger has passed. If multiple
2302 timers become ready during the same loop iteration then the ones with
2303 earlier time-out values are invoked before ones with later time-out values
2304 (but this is no longer true when a callback calls C<ev_run> recursively).
2305
2306 =head3 Watcher-Specific Functions and Data Members
2307
2308 =over 4
2309
2310 =item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
2311
2312 =item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
2313
2314 Lots of arguments, let's sort it out... There are basically three modes of
2315 operation, and we will explain them from simplest to most complex:
2316
2317 =over 4
2318
2319 =item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
2320
2321 In this configuration the watcher triggers an event after the wall clock
2322 time C<offset> has passed. It will not repeat and will not adjust when a
2323 time jump occurs, that is, if it is to be run at January 1st 2011 then it
2324 will be stopped and invoked when the system clock reaches or surpasses
2325 this point in time.
2326
2327 =item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
2328
2329 In this mode the watcher will always be scheduled to time out at the next
2330 C<offset + N * interval> time (for some integer N, which can also be
2331 negative) and then repeat, regardless of any time jumps. The C<offset>
2332 argument is merely an offset into the C<interval> periods.
2333
2334 This can be used to create timers that do not drift with respect to the
2335 system clock, for example, here is an C<ev_periodic> that triggers each
2336 hour, on the hour (with respect to UTC):
2337
2338 ev_periodic_set (&periodic, 0., 3600., 0);
2339
2340 This doesn't mean there will always be 3600 seconds in between triggers,
2341 but only that the callback will be called when the system time shows a
2342 full hour (UTC), or more correctly, when the system time is evenly divisible
2343 by 3600.
2344
2345 Another way to think about it (for the mathematically inclined) is that
2346 C<ev_periodic> will try to run the callback in this mode at the next possible
2347 time where C<time = offset (mod interval)>, regardless of any time jumps.
2348
2349 The C<interval> I<MUST> be positive, and for numerical stability, the
2350 interval value should be higher than C<1/8192> (which is around 100
2351 microseconds) and C<offset> should be higher than C<0> and should have
2352 at most a similar magnitude as the current time (say, within a factor of
2353 ten). Typical values for offset are, in fact, C<0> or something between
2354 C<0> and C<interval>, which is also the recommended range.
2355
2356 Note also that there is an upper limit to how often a timer can fire (CPU
2357 speed for example), so if C<interval> is very small then timing stability
2358 will of course deteriorate. Libev itself tries to be exact to be about one
2359 millisecond (if the OS supports it and the machine is fast enough).
2360
2361 =item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
2362
2363 In this mode the values for C<interval> and C<offset> are both being
2364 ignored. Instead, each time the periodic watcher gets scheduled, the
2365 reschedule callback will be called with the watcher as first, and the
2366 current time as second argument.
2367
2368 NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, ever,
2369 or make ANY other event loop modifications whatsoever, unless explicitly
2370 allowed by documentation here>.
2371
2372 If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
2373 it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
2374 only event loop modification you are allowed to do).
2375
2376 The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
2377 *w, ev_tstamp now)>, e.g.:
2378
2379 static ev_tstamp
2380 my_rescheduler (ev_periodic *w, ev_tstamp now)
2381 {
2382 return now + 60.;
2383 }
2384
2385 It must return the next time to trigger, based on the passed time value
2386 (that is, the lowest time value larger than to the second argument). It
2387 will usually be called just before the callback will be triggered, but
2388 might be called at other times, too.
2389
2390 NOTE: I<< This callback must always return a time that is higher than or
2391 equal to the passed C<now> value >>.
2392
2393 This can be used to create very complex timers, such as a timer that
2394 triggers on "next midnight, local time". To do this, you would calculate
2395 the next midnight after C<now> and return the timestamp value for
2396 this. Here is a (completely untested, no error checking) example on how to
2397 do this:
2398
2399 #include <time.h>
2400
2401 static ev_tstamp
2402 my_rescheduler (ev_periodic *w, ev_tstamp now)
2403 {
2404 time_t tnow = (time_t)now;
2405 struct tm tm;
2406 localtime_r (&tnow, &tm);
2407
2408 tm.tm_sec = tm.tm_min = tm.tm_hour = 0; // midnight current day
2409 ++tm.tm_mday; // midnight next day
2410
2411 return mktime (&tm);
2412 }
2413
2414 Note: this code might run into trouble on days that have more then two
2415 midnights (beginning and end).
2416
2417 =back
2418
2419 =item ev_periodic_again (loop, ev_periodic *)
2420
2421 Simply stops and restarts the periodic watcher again. This is only useful
2422 when you changed some parameters or the reschedule callback would return
2423 a different time than the last time it was called (e.g. in a crond like
2424 program when the crontabs have changed).
2425
2426 =item ev_tstamp ev_periodic_at (ev_periodic *)
2427
2428 When active, returns the absolute time that the watcher is supposed
2429 to trigger next. This is not the same as the C<offset> argument to
2430 C<ev_periodic_set>, but indeed works even in interval and manual
2431 rescheduling modes.
2432
2433 =item ev_tstamp offset [read-write]
2434
2435 When repeating, this contains the offset value, otherwise this is the
2436 absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
2437 although libev might modify this value for better numerical stability).
2438
2439 Can be modified any time, but changes only take effect when the periodic
2440 timer fires or C<ev_periodic_again> is being called.
2441
2442 =item ev_tstamp interval [read-write]
2443
2444 The current interval value. Can be modified any time, but changes only
2445 take effect when the periodic timer fires or C<ev_periodic_again> is being
2446 called.
2447
2448 =item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
2449
2450 The current reschedule callback, or C<0>, if this functionality is
2451 switched off. Can be changed any time, but changes only take effect when
2452 the periodic timer fires or C<ev_periodic_again> is being called.
2453
2454 =back
2455
2456 =head3 Examples
2457
2458 Example: Call a callback every hour, or, more precisely, whenever the
2459 system time is divisible by 3600. The callback invocation times have
2460 potentially a lot of jitter, but good long-term stability.
2461
2462 static void
2463 clock_cb (struct ev_loop *loop, ev_periodic *w, int revents)
2464 {
2465 ... its now a full hour (UTC, or TAI or whatever your clock follows)
2466 }
2467
2468 ev_periodic hourly_tick;
2469 ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
2470 ev_periodic_start (loop, &hourly_tick);
2471
2472 Example: The same as above, but use a reschedule callback to do it:
2473
2474 #include <math.h>
2475
2476 static ev_tstamp
2477 my_scheduler_cb (ev_periodic *w, ev_tstamp now)
2478 {
2479 return now + (3600. - fmod (now, 3600.));
2480 }
2481
2482 ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
2483
2484 Example: Call a callback every hour, starting now:
2485
2486 ev_periodic hourly_tick;
2487 ev_periodic_init (&hourly_tick, clock_cb,
2488 fmod (ev_now (loop), 3600.), 3600., 0);
2489 ev_periodic_start (loop, &hourly_tick);
2490
2491
2492 =head2 C<ev_signal> - signal me when a signal gets signalled!
2493
2494 Signal watchers will trigger an event when the process receives a specific
2495 signal one or more times. Even though signals are very asynchronous, libev
2496 will try its best to deliver signals synchronously, i.e. as part of the
2497 normal event processing, like any other event.
2498
2499 If you want signals to be delivered truly asynchronously, just use
2500 C<sigaction> as you would do without libev and forget about sharing
2501 the signal. You can even use C<ev_async> from a signal handler to
2502 synchronously wake up an event loop.
2503
2504 You can configure as many watchers as you like for the same signal, but
2505 only within the same loop, i.e. you can watch for C<SIGINT> in your
2506 default loop and for C<SIGIO> in another loop, but you cannot watch for
2507 C<SIGINT> in both the default loop and another loop at the same time. At
2508 the moment, C<SIGCHLD> is permanently tied to the default loop.
2509
2510 Only after the first watcher for a signal is started will libev actually
2511 register something with the kernel. It thus coexists with your own signal
2512 handlers as long as you don't register any with libev for the same signal.
2513
2514 If possible and supported, libev will install its handlers with
2515 C<SA_RESTART> (or equivalent) behaviour enabled, so system calls should
2516 not be unduly interrupted. If you have a problem with system calls getting
2517 interrupted by signals you can block all signals in an C<ev_check> watcher
2518 and unblock them in an C<ev_prepare> watcher.
2519
2520 =head3 The special problem of inheritance over fork/execve/pthread_create
2521
2522 Both the signal mask (C<sigprocmask>) and the signal disposition
2523 (C<sigaction>) are unspecified after starting a signal watcher (and after
2524 stopping it again), that is, libev might or might not block the signal,
2525 and might or might not set or restore the installed signal handler (but
2526 see C<EVFLAG_NOSIGMASK>).
2527
2528 While this does not matter for the signal disposition (libev never
2529 sets signals to C<SIG_IGN>, so handlers will be reset to C<SIG_DFL> on
2530 C<execve>), this matters for the signal mask: many programs do not expect
2531 certain signals to be blocked.
2532
2533 This means that before calling C<exec> (from the child) you should reset
2534 the signal mask to whatever "default" you expect (all clear is a good
2535 choice usually).
2536
2537 The simplest way to ensure that the signal mask is reset in the child is
2538 to install a fork handler with C<pthread_atfork> that resets it. That will
2539 catch fork calls done by libraries (such as the libc) as well.
2540
2541 In current versions of libev, the signal will not be blocked indefinitely
2542 unless you use the C<signalfd> API (C<EV_SIGNALFD>). While this reduces
2543 the window of opportunity for problems, it will not go away, as libev
2544 I<has> to modify the signal mask, at least temporarily.
2545
2546 So I can't stress this enough: I<If you do not reset your signal mask when
2547 you expect it to be empty, you have a race condition in your code>. This
2548 is not a libev-specific thing, this is true for most event libraries.
2549
2550 =head3 The special problem of threads signal handling
2551
2552 POSIX threads has problematic signal handling semantics, specifically,
2553 a lot of functionality (sigfd, sigwait etc.) only really works if all
2554 threads in a process block signals, which is hard to achieve.
2555
2556 When you want to use sigwait (or mix libev signal handling with your own
2557 for the same signals), you can tackle this problem by globally blocking
2558 all signals before creating any threads (or creating them with a fully set
2559 sigprocmask) and also specifying the C<EVFLAG_NOSIGMASK> when creating
2560 loops. Then designate one thread as "signal receiver thread" which handles
2561 these signals. You can pass on any signals that libev might be interested
2562 in by calling C<ev_feed_signal>.
2563
2564 =head3 Watcher-Specific Functions and Data Members
2565
2566 =over 4
2567
2568 =item ev_signal_init (ev_signal *, callback, int signum)
2569
2570 =item ev_signal_set (ev_signal *, int signum)
2571
2572 Configures the watcher to trigger on the given signal number (usually one
2573 of the C<SIGxxx> constants).
2574
2575 =item int signum [read-only]
2576
2577 The signal the watcher watches out for.
2578
2579 =back
2580
2581 =head3 Examples
2582
2583 Example: Try to exit cleanly on SIGINT.
2584
2585 static void
2586 sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
2587 {
2588 ev_break (loop, EVBREAK_ALL);
2589 }
2590
2591 ev_signal signal_watcher;
2592 ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
2593 ev_signal_start (loop, &signal_watcher);
2594
2595
2596 =head2 C<ev_child> - watch out for process status changes
2597
2598 Child watchers trigger when your process receives a SIGCHLD in response to
2599 some child status changes (most typically when a child of yours dies or
2600 exits). It is permissible to install a child watcher I<after> the child
2601 has been forked (which implies it might have already exited), as long
2602 as the event loop isn't entered (or is continued from a watcher), i.e.,
2603 forking and then immediately registering a watcher for the child is fine,
2604 but forking and registering a watcher a few event loop iterations later or
2605 in the next callback invocation is not.
2606
2607 Only the default event loop is capable of handling signals, and therefore
2608 you can only register child watchers in the default event loop.
2609
2610 Due to some design glitches inside libev, child watchers will always be
2611 handled at maximum priority (their priority is set to C<EV_MAXPRI> by
2612 libev)
2613
2614 =head3 Process Interaction
2615
2616 Libev grabs C<SIGCHLD> as soon as the default event loop is
2617 initialised. This is necessary to guarantee proper behaviour even if the
2618 first child watcher is started after the child exits. The occurrence
2619 of C<SIGCHLD> is recorded asynchronously, but child reaping is done
2620 synchronously as part of the event loop processing. Libev always reaps all
2621 children, even ones not watched.
2622
2623 =head3 Overriding the Built-In Processing
2624
2625 Libev offers no special support for overriding the built-in child
2626 processing, but if your application collides with libev's default child
2627 handler, you can override it easily by installing your own handler for
2628 C<SIGCHLD> after initialising the default loop, and making sure the
2629 default loop never gets destroyed. You are encouraged, however, to use an
2630 event-based approach to child reaping and thus use libev's support for
2631 that, so other libev users can use C<ev_child> watchers freely.
2632
2633 =head3 Stopping the Child Watcher
2634
2635 Currently, the child watcher never gets stopped, even when the
2636 child terminates, so normally one needs to stop the watcher in the
2637 callback. Future versions of libev might stop the watcher automatically
2638 when a child exit is detected (calling C<ev_child_stop> twice is not a
2639 problem).
2640
2641 =head3 Watcher-Specific Functions and Data Members
2642
2643 =over 4
2644
2645 =item ev_child_init (ev_child *, callback, int pid, int trace)
2646
2647 =item ev_child_set (ev_child *, int pid, int trace)
2648
2649 Configures the watcher to wait for status changes of process C<pid> (or
2650 I<any> process if C<pid> is specified as C<0>). The callback can look
2651 at the C<rstatus> member of the C<ev_child> watcher structure to see
2652 the status word (use the macros from C<sys/wait.h> and see your systems
2653 C<waitpid> documentation). The C<rpid> member contains the pid of the
2654 process causing the status change. C<trace> must be either C<0> (only
2655 activate the watcher when the process terminates) or C<1> (additionally
2656 activate the watcher when the process is stopped or continued).
2657
2658 =item int pid [read-only]
2659
2660 The process id this watcher watches out for, or C<0>, meaning any process id.
2661
2662 =item int rpid [read-write]
2663
2664 The process id that detected a status change.
2665
2666 =item int rstatus [read-write]
2667
2668 The process exit/trace status caused by C<rpid> (see your systems
2669 C<waitpid> and C<sys/wait.h> documentation for details).
2670
2671 =back
2672
2673 =head3 Examples
2674
2675 Example: C<fork()> a new process and install a child handler to wait for
2676 its completion.
2677
2678 ev_child cw;
2679
2680 static void
2681 child_cb (EV_P_ ev_child *w, int revents)
2682 {
2683 ev_child_stop (EV_A_ w);
2684 printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
2685 }
2686
2687 pid_t pid = fork ();
2688
2689 if (pid < 0)
2690 // error
2691 else if (pid == 0)
2692 {
2693 // the forked child executes here
2694 exit (1);
2695 }
2696 else
2697 {
2698 ev_child_init (&cw, child_cb, pid, 0);
2699 ev_child_start (EV_DEFAULT_ &cw);
2700 }
2701
2702
2703 =head2 C<ev_stat> - did the file attributes just change?
2704
2705 This watches a file system path for attribute changes. That is, it calls
2706 C<stat> on that path in regular intervals (or when the OS says it changed)
2707 and sees if it changed compared to the last time, invoking the callback
2708 if it did. Starting the watcher C<stat>'s the file, so only changes that
2709 happen after the watcher has been started will be reported.
2710
2711 The path does not need to exist: changing from "path exists" to "path does
2712 not exist" is a status change like any other. The condition "path does not
2713 exist" (or more correctly "path cannot be stat'ed") is signified by the
2714 C<st_nlink> field being zero (which is otherwise always forced to be at
2715 least one) and all the other fields of the stat buffer having unspecified
2716 contents.
2717
2718 The path I<must not> end in a slash or contain special components such as
2719 C<.> or C<..>. The path I<should> be absolute: If it is relative and
2720 your working directory changes, then the behaviour is undefined.
2721
2722 Since there is no portable change notification interface available, the
2723 portable implementation simply calls C<stat(2)> regularly on the path
2724 to see if it changed somehow. You can specify a recommended polling
2725 interval for this case. If you specify a polling interval of C<0> (highly
2726 recommended!) then a I<suitable, unspecified default> value will be used
2727 (which you can expect to be around five seconds, although this might
2728 change dynamically). Libev will also impose a minimum interval which is
2729 currently around C<0.1>, but that's usually overkill.
2730
2731 This watcher type is not meant for massive numbers of stat watchers,
2732 as even with OS-supported change notifications, this can be
2733 resource-intensive.
2734
2735 At the time of this writing, the only OS-specific interface implemented
2736 is the Linux inotify interface (implementing kqueue support is left as an
2737 exercise for the reader. Note, however, that the author sees no way of
2738 implementing C<ev_stat> semantics with kqueue, except as a hint).
2739
2740 =head3 ABI Issues (Largefile Support)
2741
2742 Libev by default (unless the user overrides this) uses the default
2743 compilation environment, which means that on systems with large file
2744 support disabled by default, you get the 32 bit version of the stat
2745 structure. When using the library from programs that change the ABI to
2746 use 64 bit file offsets the programs will fail. In that case you have to
2747 compile libev with the same flags to get binary compatibility. This is
2748 obviously the case with any flags that change the ABI, but the problem is
2749 most noticeably displayed with ev_stat and large file support.
2750
2751 The solution for this is to lobby your distribution maker to make large
2752 file interfaces available by default (as e.g. FreeBSD does) and not
2753 optional. Libev cannot simply switch on large file support because it has
2754 to exchange stat structures with application programs compiled using the
2755 default compilation environment.
2756
2757 =head3 Inotify and Kqueue
2758
2759 When C<inotify (7)> support has been compiled into libev and present at
2760 runtime, it will be used to speed up change detection where possible. The
2761 inotify descriptor will be created lazily when the first C<ev_stat>
2762 watcher is being started.
2763
2764 Inotify presence does not change the semantics of C<ev_stat> watchers
2765 except that changes might be detected earlier, and in some cases, to avoid
2766 making regular C<stat> calls. Even in the presence of inotify support
2767 there are many cases where libev has to resort to regular C<stat> polling,
2768 but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
2769 many bugs), the path exists (i.e. stat succeeds), and the path resides on
2770 a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
2771 xfs are fully working) libev usually gets away without polling.
2772
2773 There is no support for kqueue, as apparently it cannot be used to
2774 implement this functionality, due to the requirement of having a file
2775 descriptor open on the object at all times, and detecting renames, unlinks
2776 etc. is difficult.
2777
2778 =head3 C<stat ()> is a synchronous operation
2779
2780 Libev doesn't normally do any kind of I/O itself, and so is not blocking
2781 the process. The exception are C<ev_stat> watchers - those call C<stat
2782 ()>, which is a synchronous operation.
2783
2784 For local paths, this usually doesn't matter: unless the system is very
2785 busy or the intervals between stat's are large, a stat call will be fast,
2786 as the path data is usually in memory already (except when starting the
2787 watcher).
2788
2789 For networked file systems, calling C<stat ()> can block an indefinite
2790 time due to network issues, and even under good conditions, a stat call
2791 often takes multiple milliseconds.
2792
2793 Therefore, it is best to avoid using C<ev_stat> watchers on networked
2794 paths, although this is fully supported by libev.
2795
2796 =head3 The special problem of stat time resolution
2797
2798 The C<stat ()> system call only supports full-second resolution portably,
2799 and even on systems where the resolution is higher, most file systems
2800 still only support whole seconds.
2801
2802 That means that, if the time is the only thing that changes, you can
2803 easily miss updates: on the first update, C<ev_stat> detects a change and
2804 calls your callback, which does something. When there is another update
2805 within the same second, C<ev_stat> will be unable to detect unless the
2806 stat data does change in other ways (e.g. file size).
2807
2808 The solution to this is to delay acting on a change for slightly more
2809 than a second (or till slightly after the next full second boundary), using
2810 a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
2811 ev_timer_again (loop, w)>).
2812
2813 The C<.02> offset is added to work around small timing inconsistencies
2814 of some operating systems (where the second counter of the current time
2815 might be be delayed. One such system is the Linux kernel, where a call to
2816 C<gettimeofday> might return a timestamp with a full second later than
2817 a subsequent C<time> call - if the equivalent of C<time ()> is used to
2818 update file times then there will be a small window where the kernel uses
2819 the previous second to update file times but libev might already execute
2820 the timer callback).
2821
2822 =head3 Watcher-Specific Functions and Data Members
2823
2824 =over 4
2825
2826 =item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
2827
2828 =item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
2829
2830 Configures the watcher to wait for status changes of the given
2831 C<path>. The C<interval> is a hint on how quickly a change is expected to
2832 be detected and should normally be specified as C<0> to let libev choose
2833 a suitable value. The memory pointed to by C<path> must point to the same
2834 path for as long as the watcher is active.
2835
2836 The callback will receive an C<EV_STAT> event when a change was detected,
2837 relative to the attributes at the time the watcher was started (or the
2838 last change was detected).
2839
2840 =item ev_stat_stat (loop, ev_stat *)
2841
2842 Updates the stat buffer immediately with new values. If you change the
2843 watched path in your callback, you could call this function to avoid
2844 detecting this change (while introducing a race condition if you are not
2845 the only one changing the path). Can also be useful simply to find out the
2846 new values.
2847
2848 =item ev_statdata attr [read-only]
2849
2850 The most-recently detected attributes of the file. Although the type is
2851 C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
2852 suitable for your system, but you can only rely on the POSIX-standardised
2853 members to be present. If the C<st_nlink> member is C<0>, then there was
2854 some error while C<stat>ing the file.
2855
2856 =item ev_statdata prev [read-only]
2857
2858 The previous attributes of the file. The callback gets invoked whenever
2859 C<prev> != C<attr>, or, more precisely, one or more of these members
2860 differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
2861 C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
2862
2863 =item ev_tstamp interval [read-only]
2864
2865 The specified interval.
2866
2867 =item const char *path [read-only]
2868
2869 The file system path that is being watched.
2870
2871 =back
2872
2873 =head3 Examples
2874
2875 Example: Watch C</etc/passwd> for attribute changes.
2876
2877 static void
2878 passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
2879 {
2880 /* /etc/passwd changed in some way */
2881 if (w->attr.st_nlink)
2882 {
2883 printf ("passwd current size %ld\n", (long)w->attr.st_size);
2884 printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
2885 printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
2886 }
2887 else
2888 /* you shalt not abuse printf for puts */
2889 puts ("wow, /etc/passwd is not there, expect problems. "
2890 "if this is windows, they already arrived\n");
2891 }
2892
2893 ...
2894 ev_stat passwd;
2895
2896 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
2897 ev_stat_start (loop, &passwd);
2898
2899 Example: Like above, but additionally use a one-second delay so we do not
2900 miss updates (however, frequent updates will delay processing, too, so
2901 one might do the work both on C<ev_stat> callback invocation I<and> on
2902 C<ev_timer> callback invocation).
2903
2904 static ev_stat passwd;
2905 static ev_timer timer;
2906
2907 static void
2908 timer_cb (EV_P_ ev_timer *w, int revents)
2909 {
2910 ev_timer_stop (EV_A_ w);
2911
2912 /* now it's one second after the most recent passwd change */
2913 }
2914
2915 static void
2916 stat_cb (EV_P_ ev_stat *w, int revents)
2917 {
2918 /* reset the one-second timer */
2919 ev_timer_again (EV_A_ &timer);
2920 }
2921
2922 ...
2923 ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
2924 ev_stat_start (loop, &passwd);
2925 ev_timer_init (&timer, timer_cb, 0., 1.02);
2926
2927
2928 =head2 C<ev_idle> - when you've got nothing better to do...
2929
2930 Idle watchers trigger events when no other events of the same or higher
2931 priority are pending (prepare, check and other idle watchers do not count
2932 as receiving "events").
2933
2934 That is, as long as your process is busy handling sockets or timeouts
2935 (or even signals, imagine) of the same or higher priority it will not be
2936 triggered. But when your process is idle (or only lower-priority watchers
2937 are pending), the idle watchers are being called once per event loop
2938 iteration - until stopped, that is, or your process receives more events
2939 and becomes busy again with higher priority stuff.
2940
2941 The most noteworthy effect is that as long as any idle watchers are
2942 active, the process will not block when waiting for new events.
2943
2944 Apart from keeping your process non-blocking (which is a useful
2945 effect on its own sometimes), idle watchers are a good place to do
2946 "pseudo-background processing", or delay processing stuff to after the
2947 event loop has handled all outstanding events.
2948
2949 =head3 Abusing an C<ev_idle> watcher for its side-effect
2950
2951 As long as there is at least one active idle watcher, libev will never
2952 sleep unnecessarily. Or in other words, it will loop as fast as possible.
2953 For this to work, the idle watcher doesn't need to be invoked at all - the
2954 lowest priority will do.
2955
2956 This mode of operation can be useful together with an C<ev_check> watcher,
2957 to do something on each event loop iteration - for example to balance load
2958 between different connections.
2959
2960 See L</Abusing an ev_check watcher for its side-effect> for a longer
2961 example.
2962
2963 =head3 Watcher-Specific Functions and Data Members
2964
2965 =over 4
2966
2967 =item ev_idle_init (ev_idle *, callback)
2968
2969 Initialises and configures the idle watcher - it has no parameters of any
2970 kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
2971 believe me.
2972
2973 =back
2974
2975 =head3 Examples
2976
2977 Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
2978 callback, free it. Also, use no error checking, as usual.
2979
2980 static void
2981 idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
2982 {
2983 // stop the watcher
2984 ev_idle_stop (loop, w);
2985
2986 // now we can free it
2987 free (w);
2988
2989 // now do something you wanted to do when the program has
2990 // no longer anything immediate to do.
2991 }
2992
2993 ev_idle *idle_watcher = malloc (sizeof (ev_idle));
2994 ev_idle_init (idle_watcher, idle_cb);
2995 ev_idle_start (loop, idle_watcher);
2996
2997
2998 =head2 C<ev_prepare> and C<ev_check> - customise your event loop!
2999
3000 Prepare and check watchers are often (but not always) used in pairs:
3001 prepare watchers get invoked before the process blocks and check watchers
3002 afterwards.
3003
3004 You I<must not> call C<ev_run> (or similar functions that enter the
3005 current event loop) or C<ev_loop_fork> from either C<ev_prepare> or
3006 C<ev_check> watchers. Other loops than the current one are fine,
3007 however. The rationale behind this is that you do not need to check
3008 for recursion in those watchers, i.e. the sequence will always be
3009 C<ev_prepare>, blocking, C<ev_check> so if you have one watcher of each
3010 kind they will always be called in pairs bracketing the blocking call.
3011
3012 Their main purpose is to integrate other event mechanisms into libev and
3013 their use is somewhat advanced. They could be used, for example, to track
3014 variable changes, implement your own watchers, integrate net-snmp or a
3015 coroutine library and lots more. They are also occasionally useful if
3016 you cache some data and want to flush it before blocking (for example,
3017 in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
3018 watcher).
3019
3020 This is done by examining in each prepare call which file descriptors
3021 need to be watched by the other library, registering C<ev_io> watchers
3022 for them and starting an C<ev_timer> watcher for any timeouts (many
3023 libraries provide exactly this functionality). Then, in the check watcher,
3024 you check for any events that occurred (by checking the pending status
3025 of all watchers and stopping them) and call back into the library. The
3026 I/O and timer callbacks will never actually be called (but must be valid
3027 nevertheless, because you never know, you know?).
3028
3029 As another example, the Perl Coro module uses these hooks to integrate
3030 coroutines into libev programs, by yielding to other active coroutines
3031 during each prepare and only letting the process block if no coroutines
3032 are ready to run (it's actually more complicated: it only runs coroutines
3033 with priority higher than or equal to the event loop and one coroutine
3034 of lower priority, but only once, using idle watchers to keep the event
3035 loop from blocking if lower-priority coroutines are active, thus mapping
3036 low-priority coroutines to idle/background tasks).
3037
3038 When used for this purpose, it is recommended to give C<ev_check> watchers
3039 highest (C<EV_MAXPRI>) priority, to ensure that they are being run before
3040 any other watchers after the poll (this doesn't matter for C<ev_prepare>
3041 watchers).
3042
3043 Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
3044 activate ("feed") events into libev. While libev fully supports this, they
3045 might get executed before other C<ev_check> watchers did their job. As
3046 C<ev_check> watchers are often used to embed other (non-libev) event
3047 loops those other event loops might be in an unusable state until their
3048 C<ev_check> watcher ran (always remind yourself to coexist peacefully with
3049 others).
3050
3051 =head3 Abusing an C<ev_check> watcher for its side-effect
3052
3053 C<ev_check> (and less often also C<ev_prepare>) watchers can also be
3054 useful because they are called once per event loop iteration. For
3055 example, if you want to handle a large number of connections fairly, you
3056 normally only do a bit of work for each active connection, and if there
3057 is more work to do, you wait for the next event loop iteration, so other
3058 connections have a chance of making progress.
3059
3060 Using an C<ev_check> watcher is almost enough: it will be called on the
3061 next event loop iteration. However, that isn't as soon as possible -
3062 without external events, your C<ev_check> watcher will not be invoked.
3063
3064 This is where C<ev_idle> watchers come in handy - all you need is a
3065 single global idle watcher that is active as long as you have one active
3066 C<ev_check> watcher. The C<ev_idle> watcher makes sure the event loop
3067 will not sleep, and the C<ev_check> watcher makes sure a callback gets
3068 invoked. Neither watcher alone can do that.
3069
3070 =head3 Watcher-Specific Functions and Data Members
3071
3072 =over 4
3073
3074 =item ev_prepare_init (ev_prepare *, callback)
3075
3076 =item ev_check_init (ev_check *, callback)
3077
3078 Initialises and configures the prepare or check watcher - they have no
3079 parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
3080 macros, but using them is utterly, utterly, utterly and completely
3081 pointless.
3082
3083 =back
3084
3085 =head3 Examples
3086
3087 There are a number of principal ways to embed other event loops or modules
3088 into libev. Here are some ideas on how to include libadns into libev
3089 (there is a Perl module named C<EV::ADNS> that does this, which you could
3090 use as a working example. Another Perl module named C<EV::Glib> embeds a
3091 Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
3092 Glib event loop).
3093
3094 Method 1: Add IO watchers and a timeout watcher in a prepare handler,
3095 and in a check watcher, destroy them and call into libadns. What follows
3096 is pseudo-code only of course. This requires you to either use a low
3097 priority for the check watcher or use C<ev_clear_pending> explicitly, as
3098 the callbacks for the IO/timeout watchers might not have been called yet.
3099
3100 static ev_io iow [nfd];
3101 static ev_timer tw;
3102
3103 static void
3104 io_cb (struct ev_loop *loop, ev_io *w, int revents)
3105 {
3106 }
3107
3108 // create io watchers for each fd and a timer before blocking
3109 static void
3110 adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
3111 {
3112 int timeout = 3600000;
3113 struct pollfd fds [nfd];
3114 // actual code will need to loop here and realloc etc.
3115 adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
3116
3117 /* the callback is illegal, but won't be called as we stop during check */
3118 ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
3119 ev_timer_start (loop, &tw);
3120
3121 // create one ev_io per pollfd
3122 for (int i = 0; i < nfd; ++i)
3123 {
3124 ev_io_init (iow + i, io_cb, fds [i].fd,
3125 ((fds [i].events & POLLIN ? EV_READ : 0)
3126 | (fds [i].events & POLLOUT ? EV_WRITE : 0)));
3127
3128 fds [i].revents = 0;
3129 ev_io_start (loop, iow + i);
3130 }
3131 }
3132
3133 // stop all watchers after blocking
3134 static void
3135 adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
3136 {
3137 ev_timer_stop (loop, &tw);
3138
3139 for (int i = 0; i < nfd; ++i)
3140 {
3141 // set the relevant poll flags
3142 // could also call adns_processreadable etc. here
3143 struct pollfd *fd = fds + i;
3144 int revents = ev_clear_pending (iow + i);
3145 if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
3146 if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
3147
3148 // now stop the watcher
3149 ev_io_stop (loop, iow + i);
3150 }
3151
3152 adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
3153 }
3154
3155 Method 2: This would be just like method 1, but you run C<adns_afterpoll>
3156 in the prepare watcher and would dispose of the check watcher.
3157
3158 Method 3: If the module to be embedded supports explicit event
3159 notification (libadns does), you can also make use of the actual watcher
3160 callbacks, and only destroy/create the watchers in the prepare watcher.
3161
3162 static void
3163 timer_cb (EV_P_ ev_timer *w, int revents)
3164 {
3165 adns_state ads = (adns_state)w->data;
3166 update_now (EV_A);
3167
3168 adns_processtimeouts (ads, &tv_now);
3169 }
3170
3171 static void
3172 io_cb (EV_P_ ev_io *w, int revents)
3173 {
3174 adns_state ads = (adns_state)w->data;
3175 update_now (EV_A);
3176
3177 if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
3178 if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
3179 }
3180
3181 // do not ever call adns_afterpoll
3182
3183 Method 4: Do not use a prepare or check watcher because the module you
3184 want to embed is not flexible enough to support it. Instead, you can
3185 override their poll function. The drawback with this solution is that the
3186 main loop is now no longer controllable by EV. The C<Glib::EV> module uses
3187 this approach, effectively embedding EV as a client into the horrible
3188 libglib event loop.
3189
3190 static gint
3191 event_poll_func (GPollFD *fds, guint nfds, gint timeout)
3192 {
3193 int got_events = 0;
3194
3195 for (n = 0; n < nfds; ++n)
3196 // create/start io watcher that sets the relevant bits in fds[n] and increment got_events
3197
3198 if (timeout >= 0)
3199 // create/start timer
3200
3201 // poll
3202 ev_run (EV_A_ 0);
3203
3204 // stop timer again
3205 if (timeout >= 0)
3206 ev_timer_stop (EV_A_ &to);
3207
3208 // stop io watchers again - their callbacks should have set
3209 for (n = 0; n < nfds; ++n)
3210 ev_io_stop (EV_A_ iow [n]);
3211
3212 return got_events;
3213 }
3214
3215
3216 =head2 C<ev_embed> - when one backend isn't enough...
3217
3218 This is a rather advanced watcher type that lets you embed one event loop
3219 into another (currently only C<ev_io> events are supported in the embedded
3220 loop, other types of watchers might be handled in a delayed or incorrect
3221 fashion and must not be used).
3222
3223 There are primarily two reasons you would want that: work around bugs and
3224 prioritise I/O.
3225
3226 As an example for a bug workaround, the kqueue backend might only support
3227 sockets on some platform, so it is unusable as generic backend, but you
3228 still want to make use of it because you have many sockets and it scales
3229 so nicely. In this case, you would create a kqueue-based loop and embed
3230 it into your default loop (which might use e.g. poll). Overall operation
3231 will be a bit slower because first libev has to call C<poll> and then
3232 C<kevent>, but at least you can use both mechanisms for what they are
3233 best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
3234
3235 As for prioritising I/O: under rare circumstances you have the case where
3236 some fds have to be watched and handled very quickly (with low latency),
3237 and even priorities and idle watchers might have too much overhead. In
3238 this case you would put all the high priority stuff in one loop and all
3239 the rest in a second one, and embed the second one in the first.
3240
3241 As long as the watcher is active, the callback will be invoked every
3242 time there might be events pending in the embedded loop. The callback
3243 must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
3244 sweep and invoke their callbacks (the callback doesn't need to invoke the
3245 C<ev_embed_sweep> function directly, it could also start an idle watcher
3246 to give the embedded loop strictly lower priority for example).
3247
3248 You can also set the callback to C<0>, in which case the embed watcher
3249 will automatically execute the embedded loop sweep whenever necessary.
3250
3251 Fork detection will be handled transparently while the C<ev_embed> watcher
3252 is active, i.e., the embedded loop will automatically be forked when the
3253 embedding loop forks. In other cases, the user is responsible for calling
3254 C<ev_loop_fork> on the embedded loop.
3255
3256 Unfortunately, not all backends are embeddable: only the ones returned by
3257 C<ev_embeddable_backends> are, which, unfortunately, does not include any
3258 portable one.
3259
3260 So when you want to use this feature you will always have to be prepared
3261 that you cannot get an embeddable loop. The recommended way to get around
3262 this is to have a separate variables for your embeddable loop, try to
3263 create it, and if that fails, use the normal loop for everything.
3264
3265 =head3 C<ev_embed> and fork
3266
3267 While the C<ev_embed> watcher is running, forks in the embedding loop will
3268 automatically be applied to the embedded loop as well, so no special
3269 fork handling is required in that case. When the watcher is not running,
3270 however, it is still the task of the libev user to call C<ev_loop_fork ()>
3271 as applicable.
3272
3273 =head3 Watcher-Specific Functions and Data Members
3274
3275 =over 4
3276
3277 =item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
3278
3279 =item ev_embed_set (ev_embed *, struct ev_loop *embedded_loop)
3280
3281 Configures the watcher to embed the given loop, which must be
3282 embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
3283 invoked automatically, otherwise it is the responsibility of the callback
3284 to invoke it (it will continue to be called until the sweep has been done,
3285 if you do not want that, you need to temporarily stop the embed watcher).
3286
3287 =item ev_embed_sweep (loop, ev_embed *)
3288
3289 Make a single, non-blocking sweep over the embedded loop. This works
3290 similarly to C<ev_run (embedded_loop, EVRUN_NOWAIT)>, but in the most
3291 appropriate way for embedded loops.
3292
3293 =item struct ev_loop *other [read-only]
3294
3295 The embedded event loop.
3296
3297 =back
3298
3299 =head3 Examples
3300
3301 Example: Try to get an embeddable event loop and embed it into the default
3302 event loop. If that is not possible, use the default loop. The default
3303 loop is stored in C<loop_hi>, while the embeddable loop is stored in
3304 C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
3305 used).
3306
3307 struct ev_loop *loop_hi = ev_default_init (0);
3308 struct ev_loop *loop_lo = 0;
3309 ev_embed embed;
3310
3311 // see if there is a chance of getting one that works
3312 // (remember that a flags value of 0 means autodetection)
3313 loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
3314 ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
3315 : 0;
3316
3317 // if we got one, then embed it, otherwise default to loop_hi
3318 if (loop_lo)
3319 {
3320 ev_embed_init (&embed, 0, loop_lo);
3321 ev_embed_start (loop_hi, &embed);
3322 }
3323 else
3324 loop_lo = loop_hi;
3325
3326 Example: Check if kqueue is available but not recommended and create
3327 a kqueue backend for use with sockets (which usually work with any
3328 kqueue implementation). Store the kqueue/socket-only event loop in
3329 C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
3330
3331 struct ev_loop *loop = ev_default_init (0);
3332 struct ev_loop *loop_socket = 0;
3333 ev_embed embed;
3334
3335 if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
3336 if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
3337 {
3338 ev_embed_init (&embed, 0, loop_socket);
3339 ev_embed_start (loop, &embed);
3340 }
3341
3342 if (!loop_socket)
3343 loop_socket = loop;
3344
3345 // now use loop_socket for all sockets, and loop for everything else
3346
3347
3348 =head2 C<ev_fork> - the audacity to resume the event loop after a fork
3349
3350 Fork watchers are called when a C<fork ()> was detected (usually because
3351 whoever is a good citizen cared to tell libev about it by calling
3352 C<ev_loop_fork>). The invocation is done before the event loop blocks next
3353 and before C<ev_check> watchers are being called, and only in the child
3354 after the fork. If whoever good citizen calling C<ev_default_fork> cheats
3355 and calls it in the wrong process, the fork handlers will be invoked, too,
3356 of course.
3357
3358 =head3 The special problem of life after fork - how is it possible?
3359
3360 Most uses of C<fork ()> consist of forking, then some simple calls to set
3361 up/change the process environment, followed by a call to C<exec()>. This
3362 sequence should be handled by libev without any problems.
3363
3364 This changes when the application actually wants to do event handling
3365 in the child, or both parent in child, in effect "continuing" after the
3366 fork.
3367
3368 The default mode of operation (for libev, with application help to detect
3369 forks) is to duplicate all the state in the child, as would be expected
3370 when I<either> the parent I<or> the child process continues.
3371
3372 When both processes want to continue using libev, then this is usually the
3373 wrong result. In that case, usually one process (typically the parent) is
3374 supposed to continue with all watchers in place as before, while the other
3375 process typically wants to start fresh, i.e. without any active watchers.
3376
3377 The cleanest and most efficient way to achieve that with libev is to
3378 simply create a new event loop, which of course will be "empty", and
3379 use that for new watchers. This has the advantage of not touching more
3380 memory than necessary, and thus avoiding the copy-on-write, and the
3381 disadvantage of having to use multiple event loops (which do not support
3382 signal watchers).
3383
3384 When this is not possible, or you want to use the default loop for
3385 other reasons, then in the process that wants to start "fresh", call
3386 C<ev_loop_destroy (EV_DEFAULT)> followed by C<ev_default_loop (...)>.
3387 Destroying the default loop will "orphan" (not stop) all registered
3388 watchers, so you have to be careful not to execute code that modifies
3389 those watchers. Note also that in that case, you have to re-register any
3390 signal watchers.
3391
3392 =head3 Watcher-Specific Functions and Data Members
3393
3394 =over 4
3395
3396 =item ev_fork_init (ev_fork *, callback)
3397
3398 Initialises and configures the fork watcher - it has no parameters of any
3399 kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
3400 really.
3401
3402 =back
3403
3404
3405 =head2 C<ev_cleanup> - even the best things end
3406
3407 Cleanup watchers are called just before the event loop is being destroyed
3408 by a call to C<ev_loop_destroy>.
3409
3410 While there is no guarantee that the event loop gets destroyed, cleanup
3411 watchers provide a convenient method to install cleanup hooks for your
3412 program, worker threads and so on - you just to make sure to destroy the
3413 loop when you want them to be invoked.
3414
3415 Cleanup watchers are invoked in the same way as any other watcher. Unlike
3416 all other watchers, they do not keep a reference to the event loop (which
3417 makes a lot of sense if you think about it). Like all other watchers, you
3418 can call libev functions in the callback, except C<ev_cleanup_start>.
3419
3420 =head3 Watcher-Specific Functions and Data Members
3421
3422 =over 4
3423
3424 =item ev_cleanup_init (ev_cleanup *, callback)
3425
3426 Initialises and configures the cleanup watcher - it has no parameters of
3427 any kind. There is a C<ev_cleanup_set> macro, but using it is utterly
3428 pointless, I assure you.
3429
3430 =back
3431
3432 Example: Register an atexit handler to destroy the default loop, so any
3433 cleanup functions are called.
3434
3435 static void
3436 program_exits (void)
3437 {
3438 ev_loop_destroy (EV_DEFAULT_UC);
3439 }
3440
3441 ...
3442 atexit (program_exits);
3443
3444
3445 =head2 C<ev_async> - how to wake up an event loop
3446
3447 In general, you cannot use an C<ev_loop> from multiple threads or other
3448 asynchronous sources such as signal handlers (as opposed to multiple event
3449 loops - those are of course safe to use in different threads).
3450
3451 Sometimes, however, you need to wake up an event loop you do not control,
3452 for example because it belongs to another thread. This is what C<ev_async>
3453 watchers do: as long as the C<ev_async> watcher is active, you can signal
3454 it by calling C<ev_async_send>, which is thread- and signal safe.
3455
3456 This functionality is very similar to C<ev_signal> watchers, as signals,
3457 too, are asynchronous in nature, and signals, too, will be compressed
3458 (i.e. the number of callback invocations may be less than the number of
3459 C<ev_async_send> calls). In fact, you could use signal watchers as a kind
3460 of "global async watchers" by using a watcher on an otherwise unused
3461 signal, and C<ev_feed_signal> to signal this watcher from another thread,
3462 even without knowing which loop owns the signal.
3463
3464 =head3 Queueing
3465
3466 C<ev_async> does not support queueing of data in any way. The reason
3467 is that the author does not know of a simple (or any) algorithm for a
3468 multiple-writer-single-reader queue that works in all cases and doesn't
3469 need elaborate support such as pthreads or unportable memory access
3470 semantics.
3471
3472 That means that if you want to queue data, you have to provide your own
3473 queue. But at least I can tell you how to implement locking around your
3474 queue:
3475
3476 =over 4
3477
3478 =item queueing from a signal handler context
3479
3480 To implement race-free queueing, you simply add to the queue in the signal
3481 handler but you block the signal handler in the watcher callback. Here is
3482 an example that does that for some fictitious SIGUSR1 handler:
3483
3484 static ev_async mysig;
3485
3486 static void
3487 sigusr1_handler (void)
3488 {
3489 sometype data;
3490
3491 // no locking etc.
3492 queue_put (data);
3493 ev_async_send (EV_DEFAULT_ &mysig);
3494 }
3495
3496 static void
3497 mysig_cb (EV_P_ ev_async *w, int revents)
3498 {
3499 sometype data;
3500 sigset_t block, prev;
3501
3502 sigemptyset (&block);
3503 sigaddset (&block, SIGUSR1);
3504 sigprocmask (SIG_BLOCK, &block, &prev);
3505
3506 while (queue_get (&data))
3507 process (data);
3508
3509 if (sigismember (&prev, SIGUSR1)
3510 sigprocmask (SIG_UNBLOCK, &block, 0);
3511 }
3512
3513 (Note: pthreads in theory requires you to use C<pthread_setmask>
3514 instead of C<sigprocmask> when you use threads, but libev doesn't do it
3515 either...).
3516
3517 =item queueing from a thread context
3518
3519 The strategy for threads is different, as you cannot (easily) block
3520 threads but you can easily preempt them, so to queue safely you need to
3521 employ a traditional mutex lock, such as in this pthread example:
3522
3523 static ev_async mysig;
3524 static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
3525
3526 static void
3527 otherthread (void)
3528 {
3529 // only need to lock the actual queueing operation
3530 pthread_mutex_lock (&mymutex);
3531 queue_put (data);
3532 pthread_mutex_unlock (&mymutex);
3533
3534 ev_async_send (EV_DEFAULT_ &mysig);
3535 }
3536
3537 static void
3538 mysig_cb (EV_P_ ev_async *w, int revents)
3539 {
3540 pthread_mutex_lock (&mymutex);
3541
3542 while (queue_get (&data))
3543 process (data);
3544
3545 pthread_mutex_unlock (&mymutex);
3546 }
3547
3548 =back
3549
3550
3551 =head3 Watcher-Specific Functions and Data Members
3552
3553 =over 4
3554
3555 =item ev_async_init (ev_async *, callback)
3556
3557 Initialises and configures the async watcher - it has no parameters of any
3558 kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
3559 trust me.
3560
3561 =item ev_async_send (loop, ev_async *)
3562
3563 Sends/signals/activates the given C<ev_async> watcher, that is, feeds
3564 an C<EV_ASYNC> event on the watcher into the event loop, and instantly
3565 returns.
3566
3567 Unlike C<ev_feed_event>, this call is safe to do from other threads,
3568 signal or similar contexts (see the discussion of C<EV_ATOMIC_T> in the
3569 embedding section below on what exactly this means).
3570
3571 Note that, as with other watchers in libev, multiple events might get
3572 compressed into a single callback invocation (another way to look at
3573 this is that C<ev_async> watchers are level-triggered: they are set on
3574 C<ev_async_send>, reset when the event loop detects that).
3575
3576 This call incurs the overhead of at most one extra system call per event
3577 loop iteration, if the event loop is blocked, and no syscall at all if
3578 the event loop (or your program) is processing events. That means that
3579 repeated calls are basically free (there is no need to avoid calls for
3580 performance reasons) and that the overhead becomes smaller (typically
3581 zero) under load.
3582
3583 =item bool = ev_async_pending (ev_async *)
3584
3585 Returns a non-zero value when C<ev_async_send> has been called on the
3586 watcher but the event has not yet been processed (or even noted) by the
3587 event loop.
3588
3589 C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
3590 the loop iterates next and checks for the watcher to have become active,
3591 it will reset the flag again. C<ev_async_pending> can be used to very
3592 quickly check whether invoking the loop might be a good idea.
3593
3594 Not that this does I<not> check whether the watcher itself is pending,
3595 only whether it has been requested to make this watcher pending: there
3596 is a time window between the event loop checking and resetting the async
3597 notification, and the callback being invoked.
3598
3599 =back
3600
3601
3602 =head1 OTHER FUNCTIONS
3603
3604 There are some other functions of possible interest. Described. Here. Now.
3605
3606 =over 4
3607
3608 =item ev_once (loop, int fd, int events, ev_tstamp timeout, callback, arg)
3609
3610 This function combines a simple timer and an I/O watcher, calls your
3611 callback on whichever event happens first and automatically stops both
3612 watchers. This is useful if you want to wait for a single event on an fd
3613 or timeout without having to allocate/configure/start/stop/free one or
3614 more watchers yourself.
3615
3616 If C<fd> is less than 0, then no I/O watcher will be started and the
3617 C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
3618 the given C<fd> and C<events> set will be created and started.
3619
3620 If C<timeout> is less than 0, then no timeout watcher will be
3621 started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
3622 repeat = 0) will be started. C<0> is a valid timeout.
3623
3624 The callback has the type C<void (*cb)(int revents, void *arg)> and is
3625 passed an C<revents> set like normal event callbacks (a combination of
3626 C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMER>) and the C<arg>
3627 value passed to C<ev_once>. Note that it is possible to receive I<both>
3628 a timeout and an io event at the same time - you probably should give io
3629 events precedence.
3630
3631 Example: wait up to ten seconds for data to appear on STDIN_FILENO.
3632
3633 static void stdin_ready (int revents, void *arg)
3634 {
3635 if (revents & EV_READ)
3636 /* stdin might have data for us, joy! */;
3637 else if (revents & EV_TIMER)
3638 /* doh, nothing entered */;
3639 }
3640
3641 ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
3642
3643 =item ev_feed_fd_event (loop, int fd, int revents)
3644
3645 Feed an event on the given fd, as if a file descriptor backend detected
3646 the given events.
3647
3648 =item ev_feed_signal_event (loop, int signum)
3649
3650 Feed an event as if the given signal occurred. See also C<ev_feed_signal>,
3651 which is async-safe.
3652
3653 =back
3654
3655
3656 =head1 COMMON OR USEFUL IDIOMS (OR BOTH)
3657
3658 This section explains some common idioms that are not immediately
3659 obvious. Note that examples are sprinkled over the whole manual, and this
3660 section only contains stuff that wouldn't fit anywhere else.
3661
3662 =head2 ASSOCIATING CUSTOM DATA WITH A WATCHER
3663
3664 Each watcher has, by default, a C<void *data> member that you can read
3665 or modify at any time: libev will completely ignore it. This can be used
3666 to associate arbitrary data with your watcher. If you need more data and
3667 don't want to allocate memory separately and store a pointer to it in that
3668 data member, you can also "subclass" the watcher type and provide your own
3669 data:
3670
3671 struct my_io
3672 {
3673 ev_io io;
3674 int otherfd;
3675 void *somedata;
3676 struct whatever *mostinteresting;
3677 };
3678
3679 ...
3680 struct my_io w;
3681 ev_io_init (&w.io, my_cb, fd, EV_READ);
3682
3683 And since your callback will be called with a pointer to the watcher, you
3684 can cast it back to your own type:
3685
3686 static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
3687 {
3688 struct my_io *w = (struct my_io *)w_;
3689 ...
3690 }
3691
3692 More interesting and less C-conformant ways of casting your callback
3693 function type instead have been omitted.
3694
3695 =head2 BUILDING YOUR OWN COMPOSITE WATCHERS
3696
3697 Another common scenario is to use some data structure with multiple
3698 embedded watchers, in effect creating your own watcher that combines
3699 multiple libev event sources into one "super-watcher":
3700
3701 struct my_biggy
3702 {
3703 int some_data;
3704 ev_timer t1;
3705 ev_timer t2;
3706 }
3707
3708 In this case getting the pointer to C<my_biggy> is a bit more
3709 complicated: Either you store the address of your C<my_biggy> struct in
3710 the C<data> member of the watcher (for woozies or C++ coders), or you need
3711 to use some pointer arithmetic using C<offsetof> inside your watchers (for
3712 real programmers):
3713
3714 #include <stddef.h>
3715
3716 static void
3717 t1_cb (EV_P_ ev_timer *w, int revents)
3718 {
3719 struct my_biggy big = (struct my_biggy *)
3720 (((char *)w) - offsetof (struct my_biggy, t1));
3721 }
3722
3723 static void
3724 t2_cb (EV_P_ ev_timer *w, int revents)
3725 {
3726 struct my_biggy big = (struct my_biggy *)
3727 (((char *)w) - offsetof (struct my_biggy, t2));
3728 }
3729
3730 =head2 AVOIDING FINISHING BEFORE RETURNING
3731
3732 Often you have structures like this in event-based programs:
3733
3734 callback ()
3735 {
3736 free (request);
3737 }
3738
3739 request = start_new_request (..., callback);
3740
3741 The intent is to start some "lengthy" operation. The C<request> could be
3742 used to cancel the operation, or do other things with it.
3743
3744 It's not uncommon to have code paths in C<start_new_request> that
3745 immediately invoke the callback, for example, to report errors. Or you add
3746 some caching layer that finds that it can skip the lengthy aspects of the
3747 operation and simply invoke the callback with the result.
3748
3749 The problem here is that this will happen I<before> C<start_new_request>
3750 has returned, so C<request> is not set.
3751
3752 Even if you pass the request by some safer means to the callback, you
3753 might want to do something to the request after starting it, such as
3754 canceling it, which probably isn't working so well when the callback has
3755 already been invoked.
3756
3757 A common way around all these issues is to make sure that
3758 C<start_new_request> I<always> returns before the callback is invoked. If
3759 C<start_new_request> immediately knows the result, it can artificially
3760 delay invoking the callback by using a C<prepare> or C<idle> watcher for
3761 example, or more sneakily, by reusing an existing (stopped) watcher and
3762 pushing it into the pending queue:
3763
3764 ev_set_cb (watcher, callback);
3765 ev_feed_event (EV_A_ watcher, 0);
3766
3767 This way, C<start_new_request> can safely return before the callback is
3768 invoked, while not delaying callback invocation too much.
3769
3770 =head2 MODEL/NESTED EVENT LOOP INVOCATIONS AND EXIT CONDITIONS
3771
3772 Often (especially in GUI toolkits) there are places where you have
3773 I<modal> interaction, which is most easily implemented by recursively
3774 invoking C<ev_run>.
3775
3776 This brings the problem of exiting - a callback might want to finish the
3777 main C<ev_run> call, but not the nested one (e.g. user clicked "Quit", but
3778 a modal "Are you sure?" dialog is still waiting), or just the nested one
3779 and not the main one (e.g. user clocked "Ok" in a modal dialog), or some
3780 other combination: In these cases, a simple C<ev_break> will not work.
3781
3782 The solution is to maintain "break this loop" variable for each C<ev_run>
3783 invocation, and use a loop around C<ev_run> until the condition is
3784 triggered, using C<EVRUN_ONCE>:
3785
3786 // main loop
3787 int exit_main_loop = 0;
3788
3789 while (!exit_main_loop)
3790 ev_run (EV_DEFAULT_ EVRUN_ONCE);
3791
3792 // in a modal watcher
3793 int exit_nested_loop = 0;
3794
3795 while (!exit_nested_loop)
3796 ev_run (EV_A_ EVRUN_ONCE);
3797
3798 To exit from any of these loops, just set the corresponding exit variable:
3799
3800 // exit modal loop
3801 exit_nested_loop = 1;
3802
3803 // exit main program, after modal loop is finished
3804 exit_main_loop = 1;
3805
3806 // exit both
3807 exit_main_loop = exit_nested_loop = 1;
3808
3809 =head2 THREAD LOCKING EXAMPLE
3810
3811 Here is a fictitious example of how to run an event loop in a different
3812 thread from where callbacks are being invoked and watchers are
3813 created/added/removed.
3814
3815 For a real-world example, see the C<EV::Loop::Async> perl module,
3816 which uses exactly this technique (which is suited for many high-level
3817 languages).
3818
3819 The example uses a pthread mutex to protect the loop data, a condition
3820 variable to wait for callback invocations, an async watcher to notify the
3821 event loop thread and an unspecified mechanism to wake up the main thread.
3822
3823 First, you need to associate some data with the event loop:
3824
3825 typedef struct {
3826 mutex_t lock; /* global loop lock */
3827 ev_async async_w;
3828 thread_t tid;
3829 cond_t invoke_cv;
3830 } userdata;
3831
3832 void prepare_loop (EV_P)
3833 {
3834 // for simplicity, we use a static userdata struct.
3835 static userdata u;
3836
3837 ev_async_init (&u->async_w, async_cb);
3838 ev_async_start (EV_A_ &u->async_w);
3839
3840 pthread_mutex_init (&u->lock, 0);
3841 pthread_cond_init (&u->invoke_cv, 0);
3842
3843 // now associate this with the loop
3844 ev_set_userdata (EV_A_ u);
3845 ev_set_invoke_pending_cb (EV_A_ l_invoke);
3846 ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
3847
3848 // then create the thread running ev_run
3849 pthread_create (&u->tid, 0, l_run, EV_A);
3850 }
3851
3852 The callback for the C<ev_async> watcher does nothing: the watcher is used
3853 solely to wake up the event loop so it takes notice of any new watchers
3854 that might have been added:
3855
3856 static void
3857 async_cb (EV_P_ ev_async *w, int revents)
3858 {
3859 // just used for the side effects
3860 }
3861
3862 The C<l_release> and C<l_acquire> callbacks simply unlock/lock the mutex
3863 protecting the loop data, respectively.
3864
3865 static void
3866 l_release (EV_P)
3867 {
3868 userdata *u = ev_userdata (EV_A);
3869 pthread_mutex_unlock (&u->lock);
3870 }
3871
3872 static void
3873 l_acquire (EV_P)
3874 {
3875 userdata *u = ev_userdata (EV_A);
3876 pthread_mutex_lock (&u->lock);
3877 }
3878
3879 The event loop thread first acquires the mutex, and then jumps straight
3880 into C<ev_run>:
3881
3882 void *
3883 l_run (void *thr_arg)
3884 {
3885 struct ev_loop *loop = (struct ev_loop *)thr_arg;
3886
3887 l_acquire (EV_A);
3888 pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
3889 ev_run (EV_A_ 0);
3890 l_release (EV_A);
3891
3892 return 0;
3893 }
3894
3895 Instead of invoking all pending watchers, the C<l_invoke> callback will
3896 signal the main thread via some unspecified mechanism (signals? pipe
3897 writes? C<Async::Interrupt>?) and then waits until all pending watchers
3898 have been called (in a while loop because a) spurious wakeups are possible
3899 and b) skipping inter-thread-communication when there are no pending
3900 watchers is very beneficial):
3901
3902 static void
3903 l_invoke (EV_P)
3904 {
3905 userdata *u = ev_userdata (EV_A);
3906
3907 while (ev_pending_count (EV_A))
3908 {
3909 wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
3910 pthread_cond_wait (&u->invoke_cv, &u->lock);
3911 }
3912 }
3913
3914 Now, whenever the main thread gets told to invoke pending watchers, it
3915 will grab the lock, call C<ev_invoke_pending> and then signal the loop
3916 thread to continue:
3917
3918 static void
3919 real_invoke_pending (EV_P)
3920 {
3921 userdata *u = ev_userdata (EV_A);
3922
3923 pthread_mutex_lock (&u->lock);
3924 ev_invoke_pending (EV_A);
3925 pthread_cond_signal (&u->invoke_cv);
3926 pthread_mutex_unlock (&u->lock);
3927 }
3928
3929 Whenever you want to start/stop a watcher or do other modifications to an
3930 event loop, you will now have to lock:
3931
3932 ev_timer timeout_watcher;
3933 userdata *u = ev_userdata (EV_A);
3934
3935 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
3936
3937 pthread_mutex_lock (&u->lock);
3938 ev_timer_start (EV_A_ &timeout_watcher);
3939 ev_async_send (EV_A_ &u->async_w);
3940 pthread_mutex_unlock (&u->lock);
3941
3942 Note that sending the C<ev_async> watcher is required because otherwise
3943 an event loop currently blocking in the kernel will have no knowledge
3944 about the newly added timer. By waking up the loop it will pick up any new
3945 watchers in the next event loop iteration.
3946
3947 =head2 THREADS, COROUTINES, CONTINUATIONS, QUEUES... INSTEAD OF CALLBACKS
3948
3949 While the overhead of a callback that e.g. schedules a thread is small, it
3950 is still an overhead. If you embed libev, and your main usage is with some
3951 kind of threads or coroutines, you might want to customise libev so that
3952 doesn't need callbacks anymore.
3953
3954 Imagine you have coroutines that you can switch to using a function
3955 C<switch_to (coro)>, that libev runs in a coroutine called C<libev_coro>
3956 and that due to some magic, the currently active coroutine is stored in a
3957 global called C<current_coro>. Then you can build your own "wait for libev
3958 event" primitive by changing C<EV_CB_DECLARE> and C<EV_CB_INVOKE> (note
3959 the differing C<;> conventions):
3960
3961 #define EV_CB_DECLARE(type) struct my_coro *cb;
3962 #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
3963
3964 That means instead of having a C callback function, you store the
3965 coroutine to switch to in each watcher, and instead of having libev call
3966 your callback, you instead have it switch to that coroutine.
3967
3968 A coroutine might now wait for an event with a function called
3969 C<wait_for_event>. (the watcher needs to be started, as always, but it doesn't
3970 matter when, or whether the watcher is active or not when this function is
3971 called):
3972
3973 void
3974 wait_for_event (ev_watcher *w)
3975 {
3976 ev_set_cb (w, current_coro);
3977 switch_to (libev_coro);
3978 }
3979
3980 That basically suspends the coroutine inside C<wait_for_event> and
3981 continues the libev coroutine, which, when appropriate, switches back to
3982 this or any other coroutine.
3983
3984 You can do similar tricks if you have, say, threads with an event queue -
3985 instead of storing a coroutine, you store the queue object and instead of
3986 switching to a coroutine, you push the watcher onto the queue and notify
3987 any waiters.
3988
3989 To embed libev, see L</EMBEDDING>, but in short, it's easiest to create two
3990 files, F<my_ev.h> and F<my_ev.c> that include the respective libev files:
3991
3992 // my_ev.h
3993 #define EV_CB_DECLARE(type) struct my_coro *cb;
3994 #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
3995 #include "../libev/ev.h"
3996
3997 // my_ev.c
3998 #define EV_H "my_ev.h"
3999 #include "../libev/ev.c"
4000
4001 And then use F<my_ev.h> when you would normally use F<ev.h>, and compile
4002 F<my_ev.c> into your project. When properly specifying include paths, you
4003 can even use F<ev.h> as header file name directly.
4004
4005
4006 =head1 LIBEVENT EMULATION
4007
4008 Libev offers a compatibility emulation layer for libevent. It cannot
4009 emulate the internals of libevent, so here are some usage hints:
4010
4011 =over 4
4012
4013 =item * Only the libevent-1.4.1-beta API is being emulated.
4014
4015 This was the newest libevent version available when libev was implemented,
4016 and is still mostly unchanged in 2010.
4017
4018 =item * Use it by including <event.h>, as usual.
4019
4020 =item * The following members are fully supported: ev_base, ev_callback,
4021 ev_arg, ev_fd, ev_res, ev_events.
4022
4023 =item * Avoid using ev_flags and the EVLIST_*-macros, while it is
4024 maintained by libev, it does not work exactly the same way as in libevent (consider
4025 it a private API).
4026
4027 =item * Priorities are not currently supported. Initialising priorities
4028 will fail and all watchers will have the same priority, even though there
4029 is an ev_pri field.
4030
4031 =item * In libevent, the last base created gets the signals, in libev, the
4032 base that registered the signal gets the signals.
4033
4034 =item * Other members are not supported.
4035
4036 =item * The libev emulation is I<not> ABI compatible to libevent, you need
4037 to use the libev header file and library.
4038
4039 =back
4040
4041 =head1 C++ SUPPORT
4042
4043 =head2 C API
4044
4045 The normal C API should work fine when used from C++: both ev.h and the
4046 libev sources can be compiled as C++. Therefore, code that uses the C API
4047 will work fine.
4048
4049 Proper exception specifications might have to be added to callbacks passed
4050 to libev: exceptions may be thrown only from watcher callbacks, all other
4051 callbacks (allocator, syserr, loop acquire/release and periodic reschedule
4052 callbacks) must not throw exceptions, and might need a C<noexcept>
4053 specification. If you have code that needs to be compiled as both C and
4054 C++ you can use the C<EV_NOEXCEPT> macro for this:
4055
4056 static void
4057 fatal_error (const char *msg) EV_NOEXCEPT
4058 {
4059 perror (msg);
4060 abort ();
4061 }
4062
4063 ...
4064 ev_set_syserr_cb (fatal_error);
4065
4066 The only API functions that can currently throw exceptions are C<ev_run>,
4067 C<ev_invoke>, C<ev_invoke_pending> and C<ev_loop_destroy> (the latter
4068 because it runs cleanup watchers).
4069
4070 Throwing exceptions in watcher callbacks is only supported if libev itself
4071 is compiled with a C++ compiler or your C and C++ environments allow
4072 throwing exceptions through C libraries (most do).
4073
4074 =head2 C++ API
4075
4076 Libev comes with some simplistic wrapper classes for C++ that mainly allow
4077 you to use some convenience methods to start/stop watchers and also change
4078 the callback model to a model using method callbacks on objects.
4079
4080 To use it,
4081
4082 #include <ev++.h>
4083
4084 This automatically includes F<ev.h> and puts all of its definitions (many
4085 of them macros) into the global namespace. All C++ specific things are
4086 put into the C<ev> namespace. It should support all the same embedding
4087 options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
4088
4089 Care has been taken to keep the overhead low. The only data member the C++
4090 classes add (compared to plain C-style watchers) is the event loop pointer
4091 that the watcher is associated with (or no additional members at all if
4092 you disable C<EV_MULTIPLICITY> when embedding libev).
4093
4094 Currently, functions, static and non-static member functions and classes
4095 with C<operator ()> can be used as callbacks. Other types should be easy
4096 to add as long as they only need one additional pointer for context. If
4097 you need support for other types of functors please contact the author
4098 (preferably after implementing it).
4099
4100 For all this to work, your C++ compiler either has to use the same calling
4101 conventions as your C compiler (for static member functions), or you have
4102 to embed libev and compile libev itself as C++.
4103
4104 Here is a list of things available in the C<ev> namespace:
4105
4106 =over 4
4107
4108 =item C<ev::READ>, C<ev::WRITE> etc.
4109
4110 These are just enum values with the same values as the C<EV_READ> etc.
4111 macros from F<ev.h>.
4112
4113 =item C<ev::tstamp>, C<ev::now>
4114
4115 Aliases to the same types/functions as with the C<ev_> prefix.
4116
4117 =item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
4118
4119 For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
4120 the same name in the C<ev> namespace, with the exception of C<ev_signal>
4121 which is called C<ev::sig> to avoid clashes with the C<signal> macro
4122 defined by many implementations.
4123
4124 All of those classes have these methods:
4125
4126 =over 4
4127
4128 =item ev::TYPE::TYPE ()
4129
4130 =item ev::TYPE::TYPE (loop)
4131
4132 =item ev::TYPE::~TYPE
4133
4134 The constructor (optionally) takes an event loop to associate the watcher
4135 with. If it is omitted, it will use C<EV_DEFAULT>.
4136
4137 The constructor calls C<ev_init> for you, which means you have to call the
4138 C<set> method before starting it.
4139
4140 It will not set a callback, however: You have to call the templated C<set>
4141 method to set a callback before you can start the watcher.
4142
4143 (The reason why you have to use a method is a limitation in C++ which does
4144 not allow explicit template arguments for constructors).
4145
4146 The destructor automatically stops the watcher if it is active.
4147
4148 =item w->set<class, &class::method> (object *)
4149
4150 This method sets the callback method to call. The method has to have a
4151 signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
4152 first argument and the C<revents> as second. The object must be given as
4153 parameter and is stored in the C<data> member of the watcher.
4154
4155 This method synthesizes efficient thunking code to call your method from
4156 the C callback that libev requires. If your compiler can inline your
4157 callback (i.e. it is visible to it at the place of the C<set> call and
4158 your compiler is good :), then the method will be fully inlined into the
4159 thunking function, making it as fast as a direct C callback.
4160
4161 Example: simple class declaration and watcher initialisation
4162
4163 struct myclass
4164 {
4165 void io_cb (ev::io &w, int revents) { }
4166 }
4167
4168 myclass obj;
4169 ev::io iow;
4170 iow.set <myclass, &myclass::io_cb> (&obj);
4171
4172 =item w->set (object *)
4173
4174 This is a variation of a method callback - leaving out the method to call
4175 will default the method to C<operator ()>, which makes it possible to use
4176 functor objects without having to manually specify the C<operator ()> all
4177 the time. Incidentally, you can then also leave out the template argument
4178 list.
4179
4180 The C<operator ()> method prototype must be C<void operator ()(watcher &w,
4181 int revents)>.
4182
4183 See the method-C<set> above for more details.
4184
4185 Example: use a functor object as callback.
4186
4187 struct myfunctor
4188 {
4189 void operator() (ev::io &w, int revents)
4190 {
4191 ...
4192 }
4193 }
4194
4195 myfunctor f;
4196
4197 ev::io w;
4198 w.set (&f);
4199
4200 =item w->set<function> (void *data = 0)
4201
4202 Also sets a callback, but uses a static method or plain function as
4203 callback. The optional C<data> argument will be stored in the watcher's
4204 C<data> member and is free for you to use.
4205
4206 The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
4207
4208 See the method-C<set> above for more details.
4209
4210 Example: Use a plain function as callback.
4211
4212 static void io_cb (ev::io &w, int revents) { }
4213 iow.set <io_cb> ();
4214
4215 =item w->set (loop)
4216
4217 Associates a different C<struct ev_loop> with this watcher. You can only
4218 do this when the watcher is inactive (and not pending either).
4219
4220 =item w->set ([arguments])
4221
4222 Basically the same as C<ev_TYPE_set> (except for C<ev::embed> watchers>),
4223 with the same arguments. Either this method or a suitable start method
4224 must be called at least once. Unlike the C counterpart, an active watcher
4225 gets automatically stopped and restarted when reconfiguring it with this
4226 method.
4227
4228 For C<ev::embed> watchers this method is called C<set_embed>, to avoid
4229 clashing with the C<set (loop)> method.
4230
4231 =item w->start ()
4232
4233 Starts the watcher. Note that there is no C<loop> argument, as the
4234 constructor already stores the event loop.
4235
4236 =item w->start ([arguments])
4237
4238 Instead of calling C<set> and C<start> methods separately, it is often
4239 convenient to wrap them in one call. Uses the same type of arguments as
4240 the configure C<set> method of the watcher.
4241
4242 =item w->stop ()
4243
4244 Stops the watcher if it is active. Again, no C<loop> argument.
4245
4246 =item w->again () (C<ev::timer>, C<ev::periodic> only)
4247
4248 For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
4249 C<ev_TYPE_again> function.
4250
4251 =item w->sweep () (C<ev::embed> only)
4252
4253 Invokes C<ev_embed_sweep>.
4254
4255 =item w->update () (C<ev::stat> only)
4256
4257 Invokes C<ev_stat_stat>.
4258
4259 =back
4260
4261 =back
4262
4263 Example: Define a class with two I/O and idle watchers, start the I/O
4264 watchers in the constructor.
4265
4266 class myclass
4267 {
4268 ev::io io ; void io_cb (ev::io &w, int revents);
4269 ev::io io2 ; void io2_cb (ev::io &w, int revents);
4270 ev::idle idle; void idle_cb (ev::idle &w, int revents);
4271
4272 myclass (int fd)
4273 {
4274 io .set <myclass, &myclass::io_cb > (this);
4275 io2 .set <myclass, &myclass::io2_cb > (this);
4276 idle.set <myclass, &myclass::idle_cb> (this);
4277
4278 io.set (fd, ev::WRITE); // configure the watcher
4279 io.start (); // start it whenever convenient
4280
4281 io2.start (fd, ev::READ); // set + start in one call
4282 }
4283 };
4284
4285
4286 =head1 OTHER LANGUAGE BINDINGS
4287
4288 Libev does not offer other language bindings itself, but bindings for a
4289 number of languages exist in the form of third-party packages. If you know
4290 any interesting language binding in addition to the ones listed here, drop
4291 me a note.
4292
4293 =over 4
4294
4295 =item Perl
4296
4297 The EV module implements the full libev API and is actually used to test
4298 libev. EV is developed together with libev. Apart from the EV core module,
4299 there are additional modules that implement libev-compatible interfaces
4300 to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
4301 C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
4302 and C<EV::Glib>).
4303
4304 It can be found and installed via CPAN, its homepage is at
4305 L<http://software.schmorp.de/pkg/EV>.
4306
4307 =item Python
4308
4309 Python bindings can be found at L<http://code.google.com/p/pyev/>. It
4310 seems to be quite complete and well-documented.
4311
4312 =item Ruby
4313
4314 Tony Arcieri has written a ruby extension that offers access to a subset
4315 of the libev API and adds file handle abstractions, asynchronous DNS and
4316 more on top of it. It can be found via gem servers. Its homepage is at
4317 L<http://rev.rubyforge.org/>.
4318
4319 Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
4320 makes rev work even on mingw.
4321
4322 =item Haskell
4323
4324 A haskell binding to libev is available at
4325 L<http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
4326
4327 =item D
4328
4329 Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
4330 be found at L<http://www.llucax.com.ar/proj/ev.d/index.html>.
4331
4332 =item Ocaml
4333
4334 Erkki Seppala has written Ocaml bindings for libev, to be found at
4335 L<http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
4336
4337 =item Lua
4338
4339 Brian Maher has written a partial interface to libev for lua (at the
4340 time of this writing, only C<ev_io> and C<ev_timer>), to be found at
4341 L<http://github.com/brimworks/lua-ev>.
4342
4343 =item Javascript
4344
4345 Node.js (L<http://nodejs.org>) uses libev as the underlying event library.
4346
4347 =item Others
4348
4349 There are others, and I stopped counting.
4350
4351 =back
4352
4353
4354 =head1 MACRO MAGIC
4355
4356 Libev can be compiled with a variety of options, the most fundamental
4357 of which is C<EV_MULTIPLICITY>. This option determines whether (most)
4358 functions and callbacks have an initial C<struct ev_loop *> argument.
4359
4360 To make it easier to write programs that cope with either variant, the
4361 following macros are defined:
4362
4363 =over 4
4364
4365 =item C<EV_A>, C<EV_A_>
4366
4367 This provides the loop I<argument> for functions, if one is required ("ev
4368 loop argument"). The C<EV_A> form is used when this is the sole argument,
4369 C<EV_A_> is used when other arguments are following. Example:
4370
4371 ev_unref (EV_A);
4372 ev_timer_add (EV_A_ watcher);
4373 ev_run (EV_A_ 0);
4374
4375 It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
4376 which is often provided by the following macro.
4377
4378 =item C<EV_P>, C<EV_P_>
4379
4380 This provides the loop I<parameter> for functions, if one is required ("ev
4381 loop parameter"). The C<EV_P> form is used when this is the sole parameter,
4382 C<EV_P_> is used when other parameters are following. Example:
4383
4384 // this is how ev_unref is being declared
4385 static void ev_unref (EV_P);
4386
4387 // this is how you can declare your typical callback
4388 static void cb (EV_P_ ev_timer *w, int revents)
4389
4390 It declares a parameter C<loop> of type C<struct ev_loop *>, quite
4391 suitable for use with C<EV_A>.
4392
4393 =item C<EV_DEFAULT>, C<EV_DEFAULT_>
4394
4395 Similar to the other two macros, this gives you the value of the default
4396 loop, if multiple loops are supported ("ev loop default"). The default loop
4397 will be initialised if it isn't already initialised.
4398
4399 For non-multiplicity builds, these macros do nothing, so you always have
4400 to initialise the loop somewhere.
4401
4402 =item C<EV_DEFAULT_UC>, C<EV_DEFAULT_UC_>
4403
4404 Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
4405 default loop has been initialised (C<UC> == unchecked). Their behaviour
4406 is undefined when the default loop has not been initialised by a previous
4407 execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
4408
4409 It is often prudent to use C<EV_DEFAULT> when initialising the first
4410 watcher in a function but use C<EV_DEFAULT_UC> afterwards.
4411
4412 =back
4413
4414 Example: Declare and initialise a check watcher, utilising the above
4415 macros so it will work regardless of whether multiple loops are supported
4416 or not.
4417
4418 static void
4419 check_cb (EV_P_ ev_timer *w, int revents)
4420 {
4421 ev_check_stop (EV_A_ w);
4422 }
4423
4424 ev_check check;
4425 ev_check_init (&check, check_cb);
4426 ev_check_start (EV_DEFAULT_ &check);
4427 ev_run (EV_DEFAULT_ 0);
4428
4429 =head1 EMBEDDING
4430
4431 Libev can (and often is) directly embedded into host
4432 applications. Examples of applications that embed it include the Deliantra
4433 Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
4434 and rxvt-unicode.
4435
4436 The goal is to enable you to just copy the necessary files into your
4437 source directory without having to change even a single line in them, so
4438 you can easily upgrade by simply copying (or having a checked-out copy of
4439 libev somewhere in your source tree).
4440
4441 =head2 FILESETS
4442
4443 Depending on what features you need you need to include one or more sets of files
4444 in your application.
4445
4446 =head3 CORE EVENT LOOP
4447
4448 To include only the libev core (all the C<ev_*> functions), with manual
4449 configuration (no autoconf):
4450
4451 #define EV_STANDALONE 1
4452 #include "ev.c"
4453
4454 This will automatically include F<ev.h>, too, and should be done in a
4455 single C source file only to provide the function implementations. To use
4456 it, do the same for F<ev.h> in all files wishing to use this API (best
4457 done by writing a wrapper around F<ev.h> that you can include instead and
4458 where you can put other configuration options):
4459
4460 #define EV_STANDALONE 1
4461 #include "ev.h"
4462
4463 Both header files and implementation files can be compiled with a C++
4464 compiler (at least, that's a stated goal, and breakage will be treated
4465 as a bug).
4466
4467 You need the following files in your source tree, or in a directory
4468 in your include path (e.g. in libev/ when using -Ilibev):
4469
4470 ev.h
4471 ev.c
4472 ev_vars.h
4473 ev_wrap.h
4474
4475 ev_win32.c required on win32 platforms only
4476
4477 ev_select.c only when select backend is enabled
4478 ev_poll.c only when poll backend is enabled
4479 ev_epoll.c only when the epoll backend is enabled
4480 ev_linuxaio.c only when the linux aio backend is enabled
4481 ev_kqueue.c only when the kqueue backend is enabled
4482 ev_port.c only when the solaris port backend is enabled
4483
4484 F<ev.c> includes the backend files directly when enabled, so you only need
4485 to compile this single file.
4486
4487 =head3 LIBEVENT COMPATIBILITY API
4488
4489 To include the libevent compatibility API, also include:
4490
4491 #include "event.c"
4492
4493 in the file including F<ev.c>, and:
4494
4495 #include "event.h"
4496
4497 in the files that want to use the libevent API. This also includes F<ev.h>.
4498
4499 You need the following additional files for this:
4500
4501 event.h
4502 event.c
4503
4504 =head3 AUTOCONF SUPPORT
4505
4506 Instead of using C<EV_STANDALONE=1> and providing your configuration in
4507 whatever way you want, you can also C<m4_include([libev.m4])> in your
4508 F<configure.ac> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
4509 include F<config.h> and configure itself accordingly.
4510
4511 For this of course you need the m4 file:
4512
4513 libev.m4
4514
4515 =head2 PREPROCESSOR SYMBOLS/MACROS
4516
4517 Libev can be configured via a variety of preprocessor symbols you have to
4518 define before including (or compiling) any of its files. The default in
4519 the absence of autoconf is documented for every option.
4520
4521 Symbols marked with "(h)" do not change the ABI, and can have different
4522 values when compiling libev vs. including F<ev.h>, so it is permissible
4523 to redefine them before including F<ev.h> without breaking compatibility
4524 to a compiled library. All other symbols change the ABI, which means all
4525 users of libev and the libev code itself must be compiled with compatible
4526 settings.
4527
4528 =over 4
4529
4530 =item EV_COMPAT3 (h)
4531
4532 Backwards compatibility is a major concern for libev. This is why this
4533 release of libev comes with wrappers for the functions and symbols that
4534 have been renamed between libev version 3 and 4.
4535
4536 You can disable these wrappers (to test compatibility with future
4537 versions) by defining C<EV_COMPAT3> to C<0> when compiling your
4538 sources. This has the additional advantage that you can drop the C<struct>
4539 from C<struct ev_loop> declarations, as libev will provide an C<ev_loop>
4540 typedef in that case.
4541
4542 In some future version, the default for C<EV_COMPAT3> will become C<0>,
4543 and in some even more future version the compatibility code will be
4544 removed completely.
4545
4546 =item EV_STANDALONE (h)
4547
4548 Must always be C<1> if you do not use autoconf configuration, which
4549 keeps libev from including F<config.h>, and it also defines dummy
4550 implementations for some libevent functions (such as logging, which is not
4551 supported). It will also not define any of the structs usually found in
4552 F<event.h> that are not directly supported by the libev core alone.
4553
4554 In standalone mode, libev will still try to automatically deduce the
4555 configuration, but has to be more conservative.
4556
4557 =item EV_USE_FLOOR
4558
4559 If defined to be C<1>, libev will use the C<floor ()> function for its
4560 periodic reschedule calculations, otherwise libev will fall back on a
4561 portable (slower) implementation. If you enable this, you usually have to
4562 link against libm or something equivalent. Enabling this when the C<floor>
4563 function is not available will fail, so the safe default is to not enable
4564 this.
4565
4566 =item EV_USE_MONOTONIC
4567
4568 If defined to be C<1>, libev will try to detect the availability of the
4569 monotonic clock option at both compile time and runtime. Otherwise no
4570 use of the monotonic clock option will be attempted. If you enable this,
4571 you usually have to link against librt or something similar. Enabling it
4572 when the functionality isn't available is safe, though, although you have
4573 to make sure you link against any libraries where the C<clock_gettime>
4574 function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
4575
4576 =item EV_USE_REALTIME
4577
4578 If defined to be C<1>, libev will try to detect the availability of the
4579 real-time clock option at compile time (and assume its availability
4580 at runtime if successful). Otherwise no use of the real-time clock
4581 option will be attempted. This effectively replaces C<gettimeofday>
4582 by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
4583 correctness. See the note about libraries in the description of
4584 C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
4585 C<EV_USE_CLOCK_SYSCALL>.
4586
4587 =item EV_USE_CLOCK_SYSCALL
4588
4589 If defined to be C<1>, libev will try to use a direct syscall instead
4590 of calling the system-provided C<clock_gettime> function. This option
4591 exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
4592 unconditionally pulls in C<libpthread>, slowing down single-threaded
4593 programs needlessly. Using a direct syscall is slightly slower (in
4594 theory), because no optimised vdso implementation can be used, but avoids
4595 the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
4596 higher, as it simplifies linking (no need for C<-lrt>).
4597
4598 =item EV_USE_NANOSLEEP
4599
4600 If defined to be C<1>, libev will assume that C<nanosleep ()> is available
4601 and will use it for delays. Otherwise it will use C<select ()>.
4602
4603 =item EV_USE_EVENTFD
4604
4605 If defined to be C<1>, then libev will assume that C<eventfd ()> is
4606 available and will probe for kernel support at runtime. This will improve
4607 C<ev_signal> and C<ev_async> performance and reduce resource consumption.
4608 If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
4609 2.7 or newer, otherwise disabled.
4610
4611 =item EV_USE_SELECT
4612
4613 If undefined or defined to be C<1>, libev will compile in support for the
4614 C<select>(2) backend. No attempt at auto-detection will be done: if no
4615 other method takes over, select will be it. Otherwise the select backend
4616 will not be compiled in.
4617
4618 =item EV_SELECT_USE_FD_SET
4619
4620 If defined to C<1>, then the select backend will use the system C<fd_set>
4621 structure. This is useful if libev doesn't compile due to a missing
4622 C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
4623 on exotic systems. This usually limits the range of file descriptors to
4624 some low limit such as 1024 or might have other limitations (winsocket
4625 only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
4626 configures the maximum size of the C<fd_set>.
4627
4628 =item EV_SELECT_IS_WINSOCKET
4629
4630 When defined to C<1>, the select backend will assume that
4631 select/socket/connect etc. don't understand file descriptors but
4632 wants osf handles on win32 (this is the case when the select to
4633 be used is the winsock select). This means that it will call
4634 C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
4635 it is assumed that all these functions actually work on fds, even
4636 on win32. Should not be defined on non-win32 platforms.
4637
4638 =item EV_FD_TO_WIN32_HANDLE(fd)
4639
4640 If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
4641 file descriptors to socket handles. When not defining this symbol (the
4642 default), then libev will call C<_get_osfhandle>, which is usually
4643 correct. In some cases, programs use their own file descriptor management,
4644 in which case they can provide this function to map fds to socket handles.
4645
4646 =item EV_WIN32_HANDLE_TO_FD(handle)
4647
4648 If C<EV_SELECT_IS_WINSOCKET> then libev maps handles to file descriptors
4649 using the standard C<_open_osfhandle> function. For programs implementing
4650 their own fd to handle mapping, overwriting this function makes it easier
4651 to do so. This can be done by defining this macro to an appropriate value.
4652
4653 =item EV_WIN32_CLOSE_FD(fd)
4654
4655 If programs implement their own fd to handle mapping on win32, then this
4656 macro can be used to override the C<close> function, useful to unregister
4657 file descriptors again. Note that the replacement function has to close
4658 the underlying OS handle.
4659
4660 =item EV_USE_WSASOCKET
4661
4662 If defined to be C<1>, libev will use C<WSASocket> to create its internal
4663 communication socket, which works better in some environments. Otherwise,
4664 the normal C<socket> function will be used, which works better in other
4665 environments.
4666
4667 =item EV_USE_POLL
4668
4669 If defined to be C<1>, libev will compile in support for the C<poll>(2)
4670 backend. Otherwise it will be enabled on non-win32 platforms. It
4671 takes precedence over select.
4672
4673 =item EV_USE_EPOLL
4674
4675 If defined to be C<1>, libev will compile in support for the Linux
4676 C<epoll>(7) backend. Its availability will be detected at runtime,
4677 otherwise another method will be used as fallback. This is the preferred
4678 backend for GNU/Linux systems. If undefined, it will be enabled if the
4679 headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
4680
4681 =item EV_USE_LINUXAIO
4682
4683 If defined to be C<1>, libev will compile in support for the Linux
4684 aio backend. Due to it's currenbt limitations it has to be requested
4685 explicitly. If undefined, it will be enabled on linux, otherwise
4686 disabled.
4687
4688 =item EV_USE_KQUEUE
4689
4690 If defined to be C<1>, libev will compile in support for the BSD style
4691 C<kqueue>(2) backend. Its actual availability will be detected at runtime,
4692 otherwise another method will be used as fallback. This is the preferred
4693 backend for BSD and BSD-like systems, although on most BSDs kqueue only
4694 supports some types of fds correctly (the only platform we found that
4695 supports ptys for example was NetBSD), so kqueue might be compiled in, but
4696 not be used unless explicitly requested. The best way to use it is to find
4697 out whether kqueue supports your type of fd properly and use an embedded
4698 kqueue loop.
4699
4700 =item EV_USE_PORT
4701
4702 If defined to be C<1>, libev will compile in support for the Solaris
4703 10 port style backend. Its availability will be detected at runtime,
4704 otherwise another method will be used as fallback. This is the preferred
4705 backend for Solaris 10 systems.
4706
4707 =item EV_USE_DEVPOLL
4708
4709 Reserved for future expansion, works like the USE symbols above.
4710
4711 =item EV_USE_INOTIFY
4712
4713 If defined to be C<1>, libev will compile in support for the Linux inotify
4714 interface to speed up C<ev_stat> watchers. Its actual availability will
4715 be detected at runtime. If undefined, it will be enabled if the headers
4716 indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
4717
4718 =item EV_NO_SMP
4719
4720 If defined to be C<1>, libev will assume that memory is always coherent
4721 between threads, that is, threads can be used, but threads never run on
4722 different cpus (or different cpu cores). This reduces dependencies
4723 and makes libev faster.
4724
4725 =item EV_NO_THREADS
4726
4727 If defined to be C<1>, libev will assume that it will never be called from
4728 different threads (that includes signal handlers), which is a stronger
4729 assumption than C<EV_NO_SMP>, above. This reduces dependencies and makes
4730 libev faster.
4731
4732 =item EV_ATOMIC_T
4733
4734 Libev requires an integer type (suitable for storing C<0> or C<1>) whose
4735 access is atomic with respect to other threads or signal contexts. No
4736 such type is easily found in the C language, so you can provide your own
4737 type that you know is safe for your purposes. It is used both for signal
4738 handler "locking" as well as for signal and thread safety in C<ev_async>
4739 watchers.
4740
4741 In the absence of this define, libev will use C<sig_atomic_t volatile>
4742 (from F<signal.h>), which is usually good enough on most platforms.
4743
4744 =item EV_H (h)
4745
4746 The name of the F<ev.h> header file used to include it. The default if
4747 undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
4748 used to virtually rename the F<ev.h> header file in case of conflicts.
4749
4750 =item EV_CONFIG_H (h)
4751
4752 If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
4753 F<ev.c>'s idea of where to find the F<config.h> file, similarly to
4754 C<EV_H>, above.
4755
4756 =item EV_EVENT_H (h)
4757
4758 Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
4759 of how the F<event.h> header can be found, the default is C<"event.h">.
4760
4761 =item EV_PROTOTYPES (h)
4762
4763 If defined to be C<0>, then F<ev.h> will not define any function
4764 prototypes, but still define all the structs and other symbols. This is
4765 occasionally useful if you want to provide your own wrapper functions
4766 around libev functions.
4767
4768 =item EV_MULTIPLICITY
4769
4770 If undefined or defined to C<1>, then all event-loop-specific functions
4771 will have the C<struct ev_loop *> as first argument, and you can create
4772 additional independent event loops. Otherwise there will be no support
4773 for multiple event loops and there is no first event loop pointer
4774 argument. Instead, all functions act on the single default loop.
4775
4776 Note that C<EV_DEFAULT> and C<EV_DEFAULT_> will no longer provide a
4777 default loop when multiplicity is switched off - you always have to
4778 initialise the loop manually in this case.
4779
4780 =item EV_MINPRI
4781
4782 =item EV_MAXPRI
4783
4784 The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
4785 C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
4786 provide for more priorities by overriding those symbols (usually defined
4787 to be C<-2> and C<2>, respectively).
4788
4789 When doing priority-based operations, libev usually has to linearly search
4790 all the priorities, so having many of them (hundreds) uses a lot of space
4791 and time, so using the defaults of five priorities (-2 .. +2) is usually
4792 fine.
4793
4794 If your embedding application does not need any priorities, defining these
4795 both to C<0> will save some memory and CPU.
4796
4797 =item EV_PERIODIC_ENABLE, EV_IDLE_ENABLE, EV_EMBED_ENABLE, EV_STAT_ENABLE,
4798 EV_PREPARE_ENABLE, EV_CHECK_ENABLE, EV_FORK_ENABLE, EV_SIGNAL_ENABLE,
4799 EV_ASYNC_ENABLE, EV_CHILD_ENABLE.
4800
4801 If undefined or defined to be C<1> (and the platform supports it), then
4802 the respective watcher type is supported. If defined to be C<0>, then it
4803 is not. Disabling watcher types mainly saves code size.
4804
4805 =item EV_FEATURES
4806
4807 If you need to shave off some kilobytes of code at the expense of some
4808 speed (but with the full API), you can define this symbol to request
4809 certain subsets of functionality. The default is to enable all features
4810 that can be enabled on the platform.
4811
4812 A typical way to use this symbol is to define it to C<0> (or to a bitset
4813 with some broad features you want) and then selectively re-enable
4814 additional parts you want, for example if you want everything minimal,
4815 but multiple event loop support, async and child watchers and the poll
4816 backend, use this:
4817
4818 #define EV_FEATURES 0
4819 #define EV_MULTIPLICITY 1
4820 #define EV_USE_POLL 1
4821 #define EV_CHILD_ENABLE 1
4822 #define EV_ASYNC_ENABLE 1
4823
4824 The actual value is a bitset, it can be a combination of the following
4825 values (by default, all of these are enabled):
4826
4827 =over 4
4828
4829 =item C<1> - faster/larger code
4830
4831 Use larger code to speed up some operations.
4832
4833 Currently this is used to override some inlining decisions (enlarging the
4834 code size by roughly 30% on amd64).
4835
4836 When optimising for size, use of compiler flags such as C<-Os> with
4837 gcc is recommended, as well as C<-DNDEBUG>, as libev contains a number of
4838 assertions.
4839
4840 The default is off when C<__OPTIMIZE_SIZE__> is defined by your compiler
4841 (e.g. gcc with C<-Os>).
4842
4843 =item C<2> - faster/larger data structures
4844
4845 Replaces the small 2-heap for timer management by a faster 4-heap, larger
4846 hash table sizes and so on. This will usually further increase code size
4847 and can additionally have an effect on the size of data structures at
4848 runtime.
4849
4850 The default is off when C<__OPTIMIZE_SIZE__> is defined by your compiler
4851 (e.g. gcc with C<-Os>).
4852
4853 =item C<4> - full API configuration
4854
4855 This enables priorities (sets C<EV_MAXPRI>=2 and C<EV_MINPRI>=-2), and
4856 enables multiplicity (C<EV_MULTIPLICITY>=1).
4857
4858 =item C<8> - full API
4859
4860 This enables a lot of the "lesser used" API functions. See C<ev.h> for
4861 details on which parts of the API are still available without this
4862 feature, and do not complain if this subset changes over time.
4863
4864 =item C<16> - enable all optional watcher types
4865
4866 Enables all optional watcher types. If you want to selectively enable
4867 only some watcher types other than I/O and timers (e.g. prepare,
4868 embed, async, child...) you can enable them manually by defining
4869 C<EV_watchertype_ENABLE> to C<1> instead.
4870
4871 =item C<32> - enable all backends
4872
4873 This enables all backends - without this feature, you need to enable at
4874 least one backend manually (C<EV_USE_SELECT> is a good choice).
4875
4876 =item C<64> - enable OS-specific "helper" APIs
4877
4878 Enable inotify, eventfd, signalfd and similar OS-specific helper APIs by
4879 default.
4880
4881 =back
4882
4883 Compiling with C<gcc -Os -DEV_STANDALONE -DEV_USE_EPOLL=1 -DEV_FEATURES=0>
4884 reduces the compiled size of libev from 24.7Kb code/2.8Kb data to 6.5Kb
4885 code/0.3Kb data on my GNU/Linux amd64 system, while still giving you I/O
4886 watchers, timers and monotonic clock support.
4887
4888 With an intelligent-enough linker (gcc+binutils are intelligent enough
4889 when you use C<-Wl,--gc-sections -ffunction-sections>) functions unused by
4890 your program might be left out as well - a binary starting a timer and an
4891 I/O watcher then might come out at only 5Kb.
4892
4893 =item EV_API_STATIC
4894
4895 If this symbol is defined (by default it is not), then all identifiers
4896 will have static linkage. This means that libev will not export any
4897 identifiers, and you cannot link against libev anymore. This can be useful
4898 when you embed libev, only want to use libev functions in a single file,
4899 and do not want its identifiers to be visible.
4900
4901 To use this, define C<EV_API_STATIC> and include F<ev.c> in the file that
4902 wants to use libev.
4903
4904 This option only works when libev is compiled with a C compiler, as C++
4905 doesn't support the required declaration syntax.
4906
4907 =item EV_AVOID_STDIO
4908
4909 If this is set to C<1> at compiletime, then libev will avoid using stdio
4910 functions (printf, scanf, perror etc.). This will increase the code size
4911 somewhat, but if your program doesn't otherwise depend on stdio and your
4912 libc allows it, this avoids linking in the stdio library which is quite
4913 big.
4914
4915 Note that error messages might become less precise when this option is
4916 enabled.
4917
4918 =item EV_NSIG
4919
4920 The highest supported signal number, +1 (or, the number of
4921 signals): Normally, libev tries to deduce the maximum number of signals
4922 automatically, but sometimes this fails, in which case it can be
4923 specified. Also, using a lower number than detected (C<32> should be
4924 good for about any system in existence) can save some memory, as libev
4925 statically allocates some 12-24 bytes per signal number.
4926
4927 =item EV_PID_HASHSIZE
4928
4929 C<ev_child> watchers use a small hash table to distribute workload by
4930 pid. The default size is C<16> (or C<1> with C<EV_FEATURES> disabled),
4931 usually more than enough. If you need to manage thousands of children you
4932 might want to increase this value (I<must> be a power of two).
4933
4934 =item EV_INOTIFY_HASHSIZE
4935
4936 C<ev_stat> watchers use a small hash table to distribute workload by
4937 inotify watch id. The default size is C<16> (or C<1> with C<EV_FEATURES>
4938 disabled), usually more than enough. If you need to manage thousands of
4939 C<ev_stat> watchers you might want to increase this value (I<must> be a
4940 power of two).
4941
4942 =item EV_USE_4HEAP
4943
4944 Heaps are not very cache-efficient. To improve the cache-efficiency of the
4945 timer and periodics heaps, libev uses a 4-heap when this symbol is defined
4946 to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
4947 faster performance with many (thousands) of watchers.
4948
4949 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4950 will be C<0>.
4951
4952 =item EV_HEAP_CACHE_AT
4953
4954 Heaps are not very cache-efficient. To improve the cache-efficiency of the
4955 timer and periodics heaps, libev can cache the timestamp (I<at>) within
4956 the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
4957 which uses 8-12 bytes more per watcher and a few hundred bytes more code,
4958 but avoids random read accesses on heap changes. This improves performance
4959 noticeably with many (hundreds) of watchers.
4960
4961 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4962 will be C<0>.
4963
4964 =item EV_VERIFY
4965
4966 Controls how much internal verification (see C<ev_verify ()>) will
4967 be done: If set to C<0>, no internal verification code will be compiled
4968 in. If set to C<1>, then verification code will be compiled in, but not
4969 called. If set to C<2>, then the internal verification code will be
4970 called once per loop, which can slow down libev. If set to C<3>, then the
4971 verification code will be called very frequently, which will slow down
4972 libev considerably.
4973
4974 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4975 will be C<0>.
4976
4977 =item EV_COMMON
4978
4979 By default, all watchers have a C<void *data> member. By redefining
4980 this macro to something else you can include more and other types of
4981 members. You have to define it each time you include one of the files,
4982 though, and it must be identical each time.
4983
4984 For example, the perl EV module uses something like this:
4985
4986 #define EV_COMMON \
4987 SV *self; /* contains this struct */ \
4988 SV *cb_sv, *fh /* note no trailing ";" */
4989
4990 =item EV_CB_DECLARE (type)
4991
4992 =item EV_CB_INVOKE (watcher, revents)
4993
4994 =item ev_set_cb (ev, cb)
4995
4996 Can be used to change the callback member declaration in each watcher,
4997 and the way callbacks are invoked and set. Must expand to a struct member
4998 definition and a statement, respectively. See the F<ev.h> header file for
4999 their default definitions. One possible use for overriding these is to
5000 avoid the C<struct ev_loop *> as first argument in all cases, or to use
5001 method calls instead of plain function calls in C++.
5002
5003 =back
5004
5005 =head2 EXPORTED API SYMBOLS
5006
5007 If you need to re-export the API (e.g. via a DLL) and you need a list of
5008 exported symbols, you can use the provided F<Symbol.*> files which list
5009 all public symbols, one per line:
5010
5011 Symbols.ev for libev proper
5012 Symbols.event for the libevent emulation
5013
5014 This can also be used to rename all public symbols to avoid clashes with
5015 multiple versions of libev linked together (which is obviously bad in
5016 itself, but sometimes it is inconvenient to avoid this).
5017
5018 A sed command like this will create wrapper C<#define>'s that you need to
5019 include before including F<ev.h>:
5020
5021 <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
5022
5023 This would create a file F<wrap.h> which essentially looks like this:
5024
5025 #define ev_backend myprefix_ev_backend
5026 #define ev_check_start myprefix_ev_check_start
5027 #define ev_check_stop myprefix_ev_check_stop
5028 ...
5029
5030 =head2 EXAMPLES
5031
5032 For a real-world example of a program the includes libev
5033 verbatim, you can have a look at the EV perl module
5034 (L<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
5035 the F<libev/> subdirectory and includes them in the F<EV/EVAPI.h> (public
5036 interface) and F<EV.xs> (implementation) files. Only the F<EV.xs> file
5037 will be compiled. It is pretty complex because it provides its own header
5038 file.
5039
5040 The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
5041 that everybody includes and which overrides some configure choices:
5042
5043 #define EV_FEATURES 8
5044 #define EV_USE_SELECT 1
5045 #define EV_PREPARE_ENABLE 1
5046 #define EV_IDLE_ENABLE 1
5047 #define EV_SIGNAL_ENABLE 1
5048 #define EV_CHILD_ENABLE 1
5049 #define EV_USE_STDEXCEPT 0
5050 #define EV_CONFIG_H <config.h>
5051
5052 #include "ev++.h"
5053
5054 And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
5055
5056 #include "ev_cpp.h"
5057 #include "ev.c"
5058
5059 =head1 INTERACTION WITH OTHER PROGRAMS, LIBRARIES OR THE ENVIRONMENT
5060
5061 =head2 THREADS AND COROUTINES
5062
5063 =head3 THREADS
5064
5065 All libev functions are reentrant and thread-safe unless explicitly
5066 documented otherwise, but libev implements no locking itself. This means
5067 that you can use as many loops as you want in parallel, as long as there
5068 are no concurrent calls into any libev function with the same loop
5069 parameter (C<ev_default_*> calls have an implicit default loop parameter,
5070 of course): libev guarantees that different event loops share no data
5071 structures that need any locking.
5072
5073 Or to put it differently: calls with different loop parameters can be done
5074 concurrently from multiple threads, calls with the same loop parameter
5075 must be done serially (but can be done from different threads, as long as
5076 only one thread ever is inside a call at any point in time, e.g. by using
5077 a mutex per loop).
5078
5079 Specifically to support threads (and signal handlers), libev implements
5080 so-called C<ev_async> watchers, which allow some limited form of
5081 concurrency on the same event loop, namely waking it up "from the
5082 outside".
5083
5084 If you want to know which design (one loop, locking, or multiple loops
5085 without or something else still) is best for your problem, then I cannot
5086 help you, but here is some generic advice:
5087
5088 =over 4
5089
5090 =item * most applications have a main thread: use the default libev loop
5091 in that thread, or create a separate thread running only the default loop.
5092
5093 This helps integrating other libraries or software modules that use libev
5094 themselves and don't care/know about threading.
5095
5096 =item * one loop per thread is usually a good model.
5097
5098 Doing this is almost never wrong, sometimes a better-performance model
5099 exists, but it is always a good start.
5100
5101 =item * other models exist, such as the leader/follower pattern, where one
5102 loop is handed through multiple threads in a kind of round-robin fashion.
5103
5104 Choosing a model is hard - look around, learn, know that usually you can do
5105 better than you currently do :-)
5106
5107 =item * often you need to talk to some other thread which blocks in the
5108 event loop.
5109
5110 C<ev_async> watchers can be used to wake them up from other threads safely
5111 (or from signal contexts...).
5112
5113 An example use would be to communicate signals or other events that only
5114 work in the default loop by registering the signal watcher with the
5115 default loop and triggering an C<ev_async> watcher from the default loop
5116 watcher callback into the event loop interested in the signal.
5117
5118 =back
5119
5120 See also L</THREAD LOCKING EXAMPLE>.
5121
5122 =head3 COROUTINES
5123
5124 Libev is very accommodating to coroutines ("cooperative threads"):
5125 libev fully supports nesting calls to its functions from different
5126 coroutines (e.g. you can call C<ev_run> on the same loop from two
5127 different coroutines, and switch freely between both coroutines running
5128 the loop, as long as you don't confuse yourself). The only exception is
5129 that you must not do this from C<ev_periodic> reschedule callbacks.
5130
5131 Care has been taken to ensure that libev does not keep local state inside
5132 C<ev_run>, and other calls do not usually allow for coroutine switches as
5133 they do not call any callbacks.
5134
5135 =head2 COMPILER WARNINGS
5136
5137 Depending on your compiler and compiler settings, you might get no or a
5138 lot of warnings when compiling libev code. Some people are apparently
5139 scared by this.
5140
5141 However, these are unavoidable for many reasons. For one, each compiler
5142 has different warnings, and each user has different tastes regarding
5143 warning options. "Warn-free" code therefore cannot be a goal except when
5144 targeting a specific compiler and compiler-version.
5145
5146 Another reason is that some compiler warnings require elaborate
5147 workarounds, or other changes to the code that make it less clear and less
5148 maintainable.
5149
5150 And of course, some compiler warnings are just plain stupid, or simply
5151 wrong (because they don't actually warn about the condition their message
5152 seems to warn about). For example, certain older gcc versions had some
5153 warnings that resulted in an extreme number of false positives. These have
5154 been fixed, but some people still insist on making code warn-free with
5155 such buggy versions.
5156
5157 While libev is written to generate as few warnings as possible,
5158 "warn-free" code is not a goal, and it is recommended not to build libev
5159 with any compiler warnings enabled unless you are prepared to cope with
5160 them (e.g. by ignoring them). Remember that warnings are just that:
5161 warnings, not errors, or proof of bugs.
5162
5163
5164 =head2 VALGRIND
5165
5166 Valgrind has a special section here because it is a popular tool that is
5167 highly useful. Unfortunately, valgrind reports are very hard to interpret.
5168
5169 If you think you found a bug (memory leak, uninitialised data access etc.)
5170 in libev, then check twice: If valgrind reports something like:
5171
5172 ==2274== definitely lost: 0 bytes in 0 blocks.
5173 ==2274== possibly lost: 0 bytes in 0 blocks.
5174 ==2274== still reachable: 256 bytes in 1 blocks.
5175
5176 Then there is no memory leak, just as memory accounted to global variables
5177 is not a memleak - the memory is still being referenced, and didn't leak.
5178
5179 Similarly, under some circumstances, valgrind might report kernel bugs
5180 as if it were a bug in libev (e.g. in realloc or in the poll backend,
5181 although an acceptable workaround has been found here), or it might be
5182 confused.
5183
5184 Keep in mind that valgrind is a very good tool, but only a tool. Don't
5185 make it into some kind of religion.
5186
5187 If you are unsure about something, feel free to contact the mailing list
5188 with the full valgrind report and an explanation on why you think this
5189 is a bug in libev (best check the archives, too :). However, don't be
5190 annoyed when you get a brisk "this is no bug" answer and take the chance
5191 of learning how to interpret valgrind properly.
5192
5193 If you need, for some reason, empty reports from valgrind for your project
5194 I suggest using suppression lists.
5195
5196
5197 =head1 PORTABILITY NOTES
5198
5199 =head2 GNU/LINUX 32 BIT LIMITATIONS
5200
5201 GNU/Linux is the only common platform that supports 64 bit file/large file
5202 interfaces but I<disables> them by default.
5203
5204 That means that libev compiled in the default environment doesn't support
5205 files larger than 2GiB or so, which mainly affects C<ev_stat> watchers.
5206
5207 Unfortunately, many programs try to work around this GNU/Linux issue
5208 by enabling the large file API, which makes them incompatible with the
5209 standard libev compiled for their system.
5210
5211 Likewise, libev cannot enable the large file API itself as this would
5212 suddenly make it incompatible to the default compile time environment,
5213 i.e. all programs not using special compile switches.
5214
5215 =head2 OS/X AND DARWIN BUGS
5216
5217 The whole thing is a bug if you ask me - basically any system interface
5218 you touch is broken, whether it is locales, poll, kqueue or even the
5219 OpenGL drivers.
5220
5221 =head3 C<kqueue> is buggy
5222
5223 The kqueue syscall is broken in all known versions - most versions support
5224 only sockets, many support pipes.
5225
5226 Libev tries to work around this by not using C<kqueue> by default on this
5227 rotten platform, but of course you can still ask for it when creating a
5228 loop - embedding a socket-only kqueue loop into a select-based one is
5229 probably going to work well.
5230
5231 =head3 C<poll> is buggy
5232
5233 Instead of fixing C<kqueue>, Apple replaced their (working) C<poll>
5234 implementation by something calling C<kqueue> internally around the 10.5.6
5235 release, so now C<kqueue> I<and> C<poll> are broken.
5236
5237 Libev tries to work around this by not using C<poll> by default on
5238 this rotten platform, but of course you can still ask for it when creating
5239 a loop.
5240
5241 =head3 C<select> is buggy
5242
5243 All that's left is C<select>, and of course Apple found a way to fuck this
5244 one up as well: On OS/X, C<select> actively limits the number of file
5245 descriptors you can pass in to 1024 - your program suddenly crashes when
5246 you use more.
5247
5248 There is an undocumented "workaround" for this - defining
5249 C<_DARWIN_UNLIMITED_SELECT>, which libev tries to use, so select I<should>
5250 work on OS/X.
5251
5252 =head2 SOLARIS PROBLEMS AND WORKAROUNDS
5253
5254 =head3 C<errno> reentrancy
5255
5256 The default compile environment on Solaris is unfortunately so
5257 thread-unsafe that you can't even use components/libraries compiled
5258 without C<-D_REENTRANT> in a threaded program, which, of course, isn't
5259 defined by default. A valid, if stupid, implementation choice.
5260
5261 If you want to use libev in threaded environments you have to make sure
5262 it's compiled with C<_REENTRANT> defined.
5263
5264 =head3 Event port backend
5265
5266 The scalable event interface for Solaris is called "event
5267 ports". Unfortunately, this mechanism is very buggy in all major
5268 releases. If you run into high CPU usage, your program freezes or you get
5269 a large number of spurious wakeups, make sure you have all the relevant
5270 and latest kernel patches applied. No, I don't know which ones, but there
5271 are multiple ones to apply, and afterwards, event ports actually work
5272 great.
5273
5274 If you can't get it to work, you can try running the program by setting
5275 the environment variable C<LIBEV_FLAGS=3> to only allow C<poll> and
5276 C<select> backends.
5277
5278 =head2 AIX POLL BUG
5279
5280 AIX unfortunately has a broken C<poll.h> header. Libev works around
5281 this by trying to avoid the poll backend altogether (i.e. it's not even
5282 compiled in), which normally isn't a big problem as C<select> works fine
5283 with large bitsets on AIX, and AIX is dead anyway.
5284
5285 =head2 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
5286
5287 =head3 General issues
5288
5289 Win32 doesn't support any of the standards (e.g. POSIX) that libev
5290 requires, and its I/O model is fundamentally incompatible with the POSIX
5291 model. Libev still offers limited functionality on this platform in
5292 the form of the C<EVBACKEND_SELECT> backend, and only supports socket
5293 descriptors. This only applies when using Win32 natively, not when using
5294 e.g. cygwin. Actually, it only applies to the microsofts own compilers,
5295 as every compiler comes with a slightly differently broken/incompatible
5296 environment.
5297
5298 Lifting these limitations would basically require the full
5299 re-implementation of the I/O system. If you are into this kind of thing,
5300 then note that glib does exactly that for you in a very portable way (note
5301 also that glib is the slowest event library known to man).
5302
5303 There is no supported compilation method available on windows except
5304 embedding it into other applications.
5305
5306 Sensible signal handling is officially unsupported by Microsoft - libev
5307 tries its best, but under most conditions, signals will simply not work.
5308
5309 Not a libev limitation but worth mentioning: windows apparently doesn't
5310 accept large writes: instead of resulting in a partial write, windows will
5311 either accept everything or return C<ENOBUFS> if the buffer is too large,
5312 so make sure you only write small amounts into your sockets (less than a
5313 megabyte seems safe, but this apparently depends on the amount of memory
5314 available).
5315
5316 Due to the many, low, and arbitrary limits on the win32 platform and
5317 the abysmal performance of winsockets, using a large number of sockets
5318 is not recommended (and not reasonable). If your program needs to use
5319 more than a hundred or so sockets, then likely it needs to use a totally
5320 different implementation for windows, as libev offers the POSIX readiness
5321 notification model, which cannot be implemented efficiently on windows
5322 (due to Microsoft monopoly games).
5323
5324 A typical way to use libev under windows is to embed it (see the embedding
5325 section for details) and use the following F<evwrap.h> header file instead
5326 of F<ev.h>:
5327
5328 #define EV_STANDALONE /* keeps ev from requiring config.h */
5329 #define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
5330
5331 #include "ev.h"
5332
5333 And compile the following F<evwrap.c> file into your project (make sure
5334 you do I<not> compile the F<ev.c> or any other embedded source files!):
5335
5336 #include "evwrap.h"
5337 #include "ev.c"
5338
5339 =head3 The winsocket C<select> function
5340
5341 The winsocket C<select> function doesn't follow POSIX in that it
5342 requires socket I<handles> and not socket I<file descriptors> (it is
5343 also extremely buggy). This makes select very inefficient, and also
5344 requires a mapping from file descriptors to socket handles (the Microsoft
5345 C runtime provides the function C<_open_osfhandle> for this). See the
5346 discussion of the C<EV_SELECT_USE_FD_SET>, C<EV_SELECT_IS_WINSOCKET> and
5347 C<EV_FD_TO_WIN32_HANDLE> preprocessor symbols for more info.
5348
5349 The configuration for a "naked" win32 using the Microsoft runtime
5350 libraries and raw winsocket select is:
5351
5352 #define EV_USE_SELECT 1
5353 #define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
5354
5355 Note that winsockets handling of fd sets is O(n), so you can easily get a
5356 complexity in the O(n²) range when using win32.
5357
5358 =head3 Limited number of file descriptors
5359
5360 Windows has numerous arbitrary (and low) limits on things.
5361
5362 Early versions of winsocket's select only supported waiting for a maximum
5363 of C<64> handles (probably owning to the fact that all windows kernels
5364 can only wait for C<64> things at the same time internally; Microsoft
5365 recommends spawning a chain of threads and wait for 63 handles and the
5366 previous thread in each. Sounds great!).
5367
5368 Newer versions support more handles, but you need to define C<FD_SETSIZE>
5369 to some high number (e.g. C<2048>) before compiling the winsocket select
5370 call (which might be in libev or elsewhere, for example, perl and many
5371 other interpreters do their own select emulation on windows).
5372
5373 Another limit is the number of file descriptors in the Microsoft runtime
5374 libraries, which by default is C<64> (there must be a hidden I<64>
5375 fetish or something like this inside Microsoft). You can increase this
5376 by calling C<_setmaxstdio>, which can increase this limit to C<2048>
5377 (another arbitrary limit), but is broken in many versions of the Microsoft
5378 runtime libraries. This might get you to about C<512> or C<2048> sockets
5379 (depending on windows version and/or the phase of the moon). To get more,
5380 you need to wrap all I/O functions and provide your own fd management, but
5381 the cost of calling select (O(n²)) will likely make this unworkable.
5382
5383 =head2 PORTABILITY REQUIREMENTS
5384
5385 In addition to a working ISO-C implementation and of course the
5386 backend-specific APIs, libev relies on a few additional extensions:
5387
5388 =over 4
5389
5390 =item C<void (*)(ev_watcher_type *, int revents)> must have compatible
5391 calling conventions regardless of C<ev_watcher_type *>.
5392
5393 Libev assumes not only that all watcher pointers have the same internal
5394 structure (guaranteed by POSIX but not by ISO C for example), but it also
5395 assumes that the same (machine) code can be used to call any watcher
5396 callback: The watcher callbacks have different type signatures, but libev
5397 calls them using an C<ev_watcher *> internally.
5398
5399 =item null pointers and integer zero are represented by 0 bytes
5400
5401 Libev uses C<memset> to initialise structs and arrays to C<0> bytes, and
5402 relies on this setting pointers and integers to null.
5403
5404 =item pointer accesses must be thread-atomic
5405
5406 Accessing a pointer value must be atomic, it must both be readable and
5407 writable in one piece - this is the case on all current architectures.
5408
5409 =item C<sig_atomic_t volatile> must be thread-atomic as well
5410
5411 The type C<sig_atomic_t volatile> (or whatever is defined as
5412 C<EV_ATOMIC_T>) must be atomic with respect to accesses from different
5413 threads. This is not part of the specification for C<sig_atomic_t>, but is
5414 believed to be sufficiently portable.
5415
5416 =item C<sigprocmask> must work in a threaded environment
5417
5418 Libev uses C<sigprocmask> to temporarily block signals. This is not
5419 allowed in a threaded program (C<pthread_sigmask> has to be used). Typical
5420 pthread implementations will either allow C<sigprocmask> in the "main
5421 thread" or will block signals process-wide, both behaviours would
5422 be compatible with libev. Interaction between C<sigprocmask> and
5423 C<pthread_sigmask> could complicate things, however.
5424
5425 The most portable way to handle signals is to block signals in all threads
5426 except the initial one, and run the signal handling loop in the initial
5427 thread as well.
5428
5429 =item C<long> must be large enough for common memory allocation sizes
5430
5431 To improve portability and simplify its API, libev uses C<long> internally
5432 instead of C<size_t> when allocating its data structures. On non-POSIX
5433 systems (Microsoft...) this might be unexpectedly low, but is still at
5434 least 31 bits everywhere, which is enough for hundreds of millions of
5435 watchers.
5436
5437 =item C<double> must hold a time value in seconds with enough accuracy
5438
5439 The type C<double> is used to represent timestamps. It is required to
5440 have at least 51 bits of mantissa (and 9 bits of exponent), which is
5441 good enough for at least into the year 4000 with millisecond accuracy
5442 (the design goal for libev). This requirement is overfulfilled by
5443 implementations using IEEE 754, which is basically all existing ones.
5444
5445 With IEEE 754 doubles, you get microsecond accuracy until at least the
5446 year 2255 (and millisecond accuracy till the year 287396 - by then, libev
5447 is either obsolete or somebody patched it to use C<long double> or
5448 something like that, just kidding).
5449
5450 =back
5451
5452 If you know of other additional requirements drop me a note.
5453
5454
5455 =head1 ALGORITHMIC COMPLEXITIES
5456
5457 In this section the complexities of (many of) the algorithms used inside
5458 libev will be documented. For complexity discussions about backends see
5459 the documentation for C<ev_default_init>.
5460
5461 All of the following are about amortised time: If an array needs to be
5462 extended, libev needs to realloc and move the whole array, but this
5463 happens asymptotically rarer with higher number of elements, so O(1) might
5464 mean that libev does a lengthy realloc operation in rare cases, but on
5465 average it is much faster and asymptotically approaches constant time.
5466
5467 =over 4
5468
5469 =item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
5470
5471 This means that, when you have a watcher that triggers in one hour and
5472 there are 100 watchers that would trigger before that, then inserting will
5473 have to skip roughly seven (C<ld 100>) of these watchers.
5474
5475 =item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
5476
5477 That means that changing a timer costs less than removing/adding them,
5478 as only the relative motion in the event queue has to be paid for.
5479
5480 =item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
5481
5482 These just add the watcher into an array or at the head of a list.
5483
5484 =item Stopping check/prepare/idle/fork/async watchers: O(1)
5485
5486 =item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
5487
5488 These watchers are stored in lists, so they need to be walked to find the
5489 correct watcher to remove. The lists are usually short (you don't usually
5490 have many watchers waiting for the same fd or signal: one is typical, two
5491 is rare).
5492
5493 =item Finding the next timer in each loop iteration: O(1)
5494
5495 By virtue of using a binary or 4-heap, the next timer is always found at a
5496 fixed position in the storage array.
5497
5498 =item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
5499
5500 A change means an I/O watcher gets started or stopped, which requires
5501 libev to recalculate its status (and possibly tell the kernel, depending
5502 on backend and whether C<ev_io_set> was used).
5503
5504 =item Activating one watcher (putting it into the pending state): O(1)
5505
5506 =item Priority handling: O(number_of_priorities)
5507
5508 Priorities are implemented by allocating some space for each
5509 priority. When doing priority-based operations, libev usually has to
5510 linearly search all the priorities, but starting/stopping and activating
5511 watchers becomes O(1) with respect to priority handling.
5512
5513 =item Sending an ev_async: O(1)
5514
5515 =item Processing ev_async_send: O(number_of_async_watchers)
5516
5517 =item Processing signals: O(max_signal_number)
5518
5519 Sending involves a system call I<iff> there were no other C<ev_async_send>
5520 calls in the current loop iteration and the loop is currently
5521 blocked. Checking for async and signal events involves iterating over all
5522 running async watchers or all signal numbers.
5523
5524 =back
5525
5526
5527 =head1 PORTING FROM LIBEV 3.X TO 4.X
5528
5529 The major version 4 introduced some incompatible changes to the API.
5530
5531 At the moment, the C<ev.h> header file provides compatibility definitions
5532 for all changes, so most programs should still compile. The compatibility
5533 layer might be removed in later versions of libev, so better update to the
5534 new API early than late.
5535
5536 =over 4
5537
5538 =item C<EV_COMPAT3> backwards compatibility mechanism
5539
5540 The backward compatibility mechanism can be controlled by
5541 C<EV_COMPAT3>. See L</"PREPROCESSOR SYMBOLS/MACROS"> in the L</EMBEDDING>
5542 section.
5543
5544 =item C<ev_default_destroy> and C<ev_default_fork> have been removed
5545
5546 These calls can be replaced easily by their C<ev_loop_xxx> counterparts:
5547
5548 ev_loop_destroy (EV_DEFAULT_UC);
5549 ev_loop_fork (EV_DEFAULT);
5550
5551 =item function/symbol renames
5552
5553 A number of functions and symbols have been renamed:
5554
5555 ev_loop => ev_run
5556 EVLOOP_NONBLOCK => EVRUN_NOWAIT
5557 EVLOOP_ONESHOT => EVRUN_ONCE
5558
5559 ev_unloop => ev_break
5560 EVUNLOOP_CANCEL => EVBREAK_CANCEL
5561 EVUNLOOP_ONE => EVBREAK_ONE
5562 EVUNLOOP_ALL => EVBREAK_ALL
5563
5564 EV_TIMEOUT => EV_TIMER
5565
5566 ev_loop_count => ev_iteration
5567 ev_loop_depth => ev_depth
5568 ev_loop_verify => ev_verify
5569
5570 Most functions working on C<struct ev_loop> objects don't have an
5571 C<ev_loop_> prefix, so it was removed; C<ev_loop>, C<ev_unloop> and
5572 associated constants have been renamed to not collide with the C<struct
5573 ev_loop> anymore and C<EV_TIMER> now follows the same naming scheme
5574 as all other watcher types. Note that C<ev_loop_fork> is still called
5575 C<ev_loop_fork> because it would otherwise clash with the C<ev_fork>
5576 typedef.
5577
5578 =item C<EV_MINIMAL> mechanism replaced by C<EV_FEATURES>
5579
5580 The preprocessor symbol C<EV_MINIMAL> has been replaced by a different
5581 mechanism, C<EV_FEATURES>. Programs using C<EV_MINIMAL> usually compile
5582 and work, but the library code will of course be larger.
5583
5584 =back
5585
5586
5587 =head1 GLOSSARY
5588
5589 =over 4
5590
5591 =item active
5592
5593 A watcher is active as long as it has been started and not yet stopped.
5594 See L</WATCHER STATES> for details.
5595
5596 =item application
5597
5598 In this document, an application is whatever is using libev.
5599
5600 =item backend
5601
5602 The part of the code dealing with the operating system interfaces.
5603
5604 =item callback
5605
5606 The address of a function that is called when some event has been
5607 detected. Callbacks are being passed the event loop, the watcher that
5608 received the event, and the actual event bitset.
5609
5610 =item callback/watcher invocation
5611
5612 The act of calling the callback associated with a watcher.
5613
5614 =item event
5615
5616 A change of state of some external event, such as data now being available
5617 for reading on a file descriptor, time having passed or simply not having
5618 any other events happening anymore.
5619
5620 In libev, events are represented as single bits (such as C<EV_READ> or
5621 C<EV_TIMER>).
5622
5623 =item event library
5624
5625 A software package implementing an event model and loop.
5626
5627 =item event loop
5628
5629 An entity that handles and processes external events and converts them
5630 into callback invocations.
5631
5632 =item event model
5633
5634 The model used to describe how an event loop handles and processes
5635 watchers and events.
5636
5637 =item pending
5638
5639 A watcher is pending as soon as the corresponding event has been
5640 detected. See L</WATCHER STATES> for details.
5641
5642 =item real time
5643
5644 The physical time that is observed. It is apparently strictly monotonic :)
5645
5646 =item wall-clock time
5647
5648 The time and date as shown on clocks. Unlike real time, it can actually
5649 be wrong and jump forwards and backwards, e.g. when you adjust your
5650 clock.
5651
5652 =item watcher
5653
5654 A data structure that describes interest in certain events. Watchers need
5655 to be started (attached to an event loop) before they can receive events.
5656
5657 =back
5658
5659 =head1 AUTHOR
5660
5661 Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael
5662 Magnusson and Emanuele Giaquinta, and minor corrections by many others.
5663