ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/ev.pod
Revision: 1.223
Committed: Sun Dec 14 21:58:08 2008 UTC (15 years, 7 months ago) by root
Branch: MAIN
CVS Tags: rel-3_51, rel-3_52
Changes since 1.222: +16 -20 lines
Log Message:
*** empty log message ***

File Contents

# Content
1 =head1 NAME
2
3 libev - a high performance full-featured event loop written in C
4
5 =head1 SYNOPSIS
6
7 #include <ev.h>
8
9 =head2 EXAMPLE PROGRAM
10
11 // a single header file is required
12 #include <ev.h>
13
14 #include <stdio.h> // for puts
15
16 // every watcher type has its own typedef'd struct
17 // with the name ev_TYPE
18 ev_io stdin_watcher;
19 ev_timer timeout_watcher;
20
21 // all watcher callbacks have a similar signature
22 // this callback is called when data is readable on stdin
23 static void
24 stdin_cb (EV_P_ ev_io *w, int revents)
25 {
26 puts ("stdin ready");
27 // for one-shot events, one must manually stop the watcher
28 // with its corresponding stop function.
29 ev_io_stop (EV_A_ w);
30
31 // this causes all nested ev_loop's to stop iterating
32 ev_unloop (EV_A_ EVUNLOOP_ALL);
33 }
34
35 // another callback, this time for a time-out
36 static void
37 timeout_cb (EV_P_ ev_timer *w, int revents)
38 {
39 puts ("timeout");
40 // this causes the innermost ev_loop to stop iterating
41 ev_unloop (EV_A_ EVUNLOOP_ONE);
42 }
43
44 int
45 main (void)
46 {
47 // use the default event loop unless you have special needs
48 struct ev_loop *loop = ev_default_loop (0);
49
50 // initialise an io watcher, then start it
51 // this one will watch for stdin to become readable
52 ev_io_init (&stdin_watcher, stdin_cb, /*STDIN_FILENO*/ 0, EV_READ);
53 ev_io_start (loop, &stdin_watcher);
54
55 // initialise a timer watcher, then start it
56 // simple non-repeating 5.5 second timeout
57 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
58 ev_timer_start (loop, &timeout_watcher);
59
60 // now wait for events to arrive
61 ev_loop (loop, 0);
62
63 // unloop was called, so exit
64 return 0;
65 }
66
67 =head1 DESCRIPTION
68
69 The newest version of this document is also available as an html-formatted
70 web page you might find easier to navigate when reading it for the first
71 time: L<http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod>.
72
73 Libev is an event loop: you register interest in certain events (such as a
74 file descriptor being readable or a timeout occurring), and it will manage
75 these event sources and provide your program with events.
76
77 To do this, it must take more or less complete control over your process
78 (or thread) by executing the I<event loop> handler, and will then
79 communicate events via a callback mechanism.
80
81 You register interest in certain events by registering so-called I<event
82 watchers>, which are relatively small C structures you initialise with the
83 details of the event, and then hand it over to libev by I<starting> the
84 watcher.
85
86 =head2 FEATURES
87
88 Libev supports C<select>, C<poll>, the Linux-specific C<epoll>, the
89 BSD-specific C<kqueue> and the Solaris-specific event port mechanisms
90 for file descriptor events (C<ev_io>), the Linux C<inotify> interface
91 (for C<ev_stat>), relative timers (C<ev_timer>), absolute timers
92 with customised rescheduling (C<ev_periodic>), synchronous signals
93 (C<ev_signal>), process status change events (C<ev_child>), and event
94 watchers dealing with the event loop mechanism itself (C<ev_idle>,
95 C<ev_embed>, C<ev_prepare> and C<ev_check> watchers) as well as
96 file watchers (C<ev_stat>) and even limited support for fork events
97 (C<ev_fork>).
98
99 It also is quite fast (see this
100 L<benchmark|http://libev.schmorp.de/bench.html> comparing it to libevent
101 for example).
102
103 =head2 CONVENTIONS
104
105 Libev is very configurable. In this manual the default (and most common)
106 configuration will be described, which supports multiple event loops. For
107 more info about various configuration options please have a look at
108 B<EMBED> section in this manual. If libev was configured without support
109 for multiple event loops, then all functions taking an initial argument of
110 name C<loop> (which is always of type C<ev_loop *>) will not have
111 this argument.
112
113 =head2 TIME REPRESENTATION
114
115 Libev represents time as a single floating point number, representing the
116 (fractional) number of seconds since the (POSIX) epoch (somewhere near
117 the beginning of 1970, details are complicated, don't ask). This type is
118 called C<ev_tstamp>, which is what you should use too. It usually aliases
119 to the C<double> type in C, and when you need to do any calculations on
120 it, you should treat it as some floating point value. Unlike the name
121 component C<stamp> might indicate, it is also used for time differences
122 throughout libev.
123
124 =head1 ERROR HANDLING
125
126 Libev knows three classes of errors: operating system errors, usage errors
127 and internal errors (bugs).
128
129 When libev catches an operating system error it cannot handle (for example
130 a system call indicating a condition libev cannot fix), it calls the callback
131 set via C<ev_set_syserr_cb>, which is supposed to fix the problem or
132 abort. The default is to print a diagnostic message and to call C<abort
133 ()>.
134
135 When libev detects a usage error such as a negative timer interval, then
136 it will print a diagnostic message and abort (via the C<assert> mechanism,
137 so C<NDEBUG> will disable this checking): these are programming errors in
138 the libev caller and need to be fixed there.
139
140 Libev also has a few internal error-checking C<assert>ions, and also has
141 extensive consistency checking code. These do not trigger under normal
142 circumstances, as they indicate either a bug in libev or worse.
143
144
145 =head1 GLOBAL FUNCTIONS
146
147 These functions can be called anytime, even before initialising the
148 library in any way.
149
150 =over 4
151
152 =item ev_tstamp ev_time ()
153
154 Returns the current time as libev would use it. Please note that the
155 C<ev_now> function is usually faster and also often returns the timestamp
156 you actually want to know.
157
158 =item ev_sleep (ev_tstamp interval)
159
160 Sleep for the given interval: The current thread will be blocked until
161 either it is interrupted or the given time interval has passed. Basically
162 this is a sub-second-resolution C<sleep ()>.
163
164 =item int ev_version_major ()
165
166 =item int ev_version_minor ()
167
168 You can find out the major and minor ABI version numbers of the library
169 you linked against by calling the functions C<ev_version_major> and
170 C<ev_version_minor>. If you want, you can compare against the global
171 symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the
172 version of the library your program was compiled against.
173
174 These version numbers refer to the ABI version of the library, not the
175 release version.
176
177 Usually, it's a good idea to terminate if the major versions mismatch,
178 as this indicates an incompatible change. Minor versions are usually
179 compatible to older versions, so a larger minor version alone is usually
180 not a problem.
181
182 Example: Make sure we haven't accidentally been linked against the wrong
183 version.
184
185 assert (("libev version mismatch",
186 ev_version_major () == EV_VERSION_MAJOR
187 && ev_version_minor () >= EV_VERSION_MINOR));
188
189 =item unsigned int ev_supported_backends ()
190
191 Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
192 value) compiled into this binary of libev (independent of their
193 availability on the system you are running on). See C<ev_default_loop> for
194 a description of the set values.
195
196 Example: make sure we have the epoll method, because yeah this is cool and
197 a must have and can we have a torrent of it please!!!11
198
199 assert (("sorry, no epoll, no sex",
200 ev_supported_backends () & EVBACKEND_EPOLL));
201
202 =item unsigned int ev_recommended_backends ()
203
204 Return the set of all backends compiled into this binary of libev and also
205 recommended for this platform. This set is often smaller than the one
206 returned by C<ev_supported_backends>, as for example kqueue is broken on
207 most BSDs and will not be auto-detected unless you explicitly request it
208 (assuming you know what you are doing). This is the set of backends that
209 libev will probe for if you specify no backends explicitly.
210
211 =item unsigned int ev_embeddable_backends ()
212
213 Returns the set of backends that are embeddable in other event loops. This
214 is the theoretical, all-platform, value. To find which backends
215 might be supported on the current system, you would need to look at
216 C<ev_embeddable_backends () & ev_supported_backends ()>, likewise for
217 recommended ones.
218
219 See the description of C<ev_embed> watchers for more info.
220
221 =item ev_set_allocator (void *(*cb)(void *ptr, long size)) [NOT REENTRANT]
222
223 Sets the allocation function to use (the prototype is similar - the
224 semantics are identical to the C<realloc> C89/SuS/POSIX function). It is
225 used to allocate and free memory (no surprises here). If it returns zero
226 when memory needs to be allocated (C<size != 0>), the library might abort
227 or take some potentially destructive action.
228
229 Since some systems (at least OpenBSD and Darwin) fail to implement
230 correct C<realloc> semantics, libev will use a wrapper around the system
231 C<realloc> and C<free> functions by default.
232
233 You could override this function in high-availability programs to, say,
234 free some memory if it cannot allocate memory, to use a special allocator,
235 or even to sleep a while and retry until some memory is available.
236
237 Example: Replace the libev allocator with one that waits a bit and then
238 retries (example requires a standards-compliant C<realloc>).
239
240 static void *
241 persistent_realloc (void *ptr, size_t size)
242 {
243 for (;;)
244 {
245 void *newptr = realloc (ptr, size);
246
247 if (newptr)
248 return newptr;
249
250 sleep (60);
251 }
252 }
253
254 ...
255 ev_set_allocator (persistent_realloc);
256
257 =item ev_set_syserr_cb (void (*cb)(const char *msg)); [NOT REENTRANT]
258
259 Set the callback function to call on a retryable system call error (such
260 as failed select, poll, epoll_wait). The message is a printable string
261 indicating the system call or subsystem causing the problem. If this
262 callback is set, then libev will expect it to remedy the situation, no
263 matter what, when it returns. That is, libev will generally retry the
264 requested operation, or, if the condition doesn't go away, do bad stuff
265 (such as abort).
266
267 Example: This is basically the same thing that libev does internally, too.
268
269 static void
270 fatal_error (const char *msg)
271 {
272 perror (msg);
273 abort ();
274 }
275
276 ...
277 ev_set_syserr_cb (fatal_error);
278
279 =back
280
281 =head1 FUNCTIONS CONTROLLING THE EVENT LOOP
282
283 An event loop is described by a C<struct ev_loop *> (the C<struct>
284 is I<not> optional in this case, as there is also an C<ev_loop>
285 I<function>).
286
287 The library knows two types of such loops, the I<default> loop, which
288 supports signals and child events, and dynamically created loops which do
289 not.
290
291 =over 4
292
293 =item struct ev_loop *ev_default_loop (unsigned int flags)
294
295 This will initialise the default event loop if it hasn't been initialised
296 yet and return it. If the default loop could not be initialised, returns
297 false. If it already was initialised it simply returns it (and ignores the
298 flags. If that is troubling you, check C<ev_backend ()> afterwards).
299
300 If you don't know what event loop to use, use the one returned from this
301 function.
302
303 Note that this function is I<not> thread-safe, so if you want to use it
304 from multiple threads, you have to lock (note also that this is unlikely,
305 as loops cannot be shared easily between threads anyway).
306
307 The default loop is the only loop that can handle C<ev_signal> and
308 C<ev_child> watchers, and to do this, it always registers a handler
309 for C<SIGCHLD>. If this is a problem for your application you can either
310 create a dynamic loop with C<ev_loop_new> that doesn't do that, or you
311 can simply overwrite the C<SIGCHLD> signal handler I<after> calling
312 C<ev_default_init>.
313
314 The flags argument can be used to specify special behaviour or specific
315 backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
316
317 The following flags are supported:
318
319 =over 4
320
321 =item C<EVFLAG_AUTO>
322
323 The default flags value. Use this if you have no clue (it's the right
324 thing, believe me).
325
326 =item C<EVFLAG_NOENV>
327
328 If this flag bit is or'ed into the flag value (or the program runs setuid
329 or setgid) then libev will I<not> look at the environment variable
330 C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
331 override the flags completely if it is found in the environment. This is
332 useful to try out specific backends to test their performance, or to work
333 around bugs.
334
335 =item C<EVFLAG_FORKCHECK>
336
337 Instead of calling C<ev_default_fork> or C<ev_loop_fork> manually after
338 a fork, you can also make libev check for a fork in each iteration by
339 enabling this flag.
340
341 This works by calling C<getpid ()> on every iteration of the loop,
342 and thus this might slow down your event loop if you do a lot of loop
343 iterations and little real work, but is usually not noticeable (on my
344 GNU/Linux system for example, C<getpid> is actually a simple 5-insn sequence
345 without a system call and thus I<very> fast, but my GNU/Linux system also has
346 C<pthread_atfork> which is even faster).
347
348 The big advantage of this flag is that you can forget about fork (and
349 forget about forgetting to tell libev about forking) when you use this
350 flag.
351
352 This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
353 environment variable.
354
355 =item C<EVBACKEND_SELECT> (value 1, portable select backend)
356
357 This is your standard select(2) backend. Not I<completely> standard, as
358 libev tries to roll its own fd_set with no limits on the number of fds,
359 but if that fails, expect a fairly low limit on the number of fds when
360 using this backend. It doesn't scale too well (O(highest_fd)), but its
361 usually the fastest backend for a low number of (low-numbered :) fds.
362
363 To get good performance out of this backend you need a high amount of
364 parallelism (most of the file descriptors should be busy). If you are
365 writing a server, you should C<accept ()> in a loop to accept as many
366 connections as possible during one iteration. You might also want to have
367 a look at C<ev_set_io_collect_interval ()> to increase the amount of
368 readiness notifications you get per iteration.
369
370 This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
371 C<writefds> set (and to work around Microsoft Windows bugs, also onto the
372 C<exceptfds> set on that platform).
373
374 =item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
375
376 And this is your standard poll(2) backend. It's more complicated
377 than select, but handles sparse fds better and has no artificial
378 limit on the number of fds you can use (except it will slow down
379 considerably with a lot of inactive fds). It scales similarly to select,
380 i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
381 performance tips.
382
383 This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
384 C<EV_WRITE> to C<POLLOUT | POLLERR | POLLHUP>.
385
386 =item C<EVBACKEND_EPOLL> (value 4, Linux)
387
388 For few fds, this backend is a bit little slower than poll and select,
389 but it scales phenomenally better. While poll and select usually scale
390 like O(total_fds) where n is the total number of fds (or the highest fd),
391 epoll scales either O(1) or O(active_fds).
392
393 The epoll mechanism deserves honorable mention as the most misdesigned
394 of the more advanced event mechanisms: mere annoyances include silently
395 dropping file descriptors, requiring a system call per change per file
396 descriptor (and unnecessary guessing of parameters), problems with dup and
397 so on. The biggest issue is fork races, however - if a program forks then
398 I<both> parent and child process have to recreate the epoll set, which can
399 take considerable time (one syscall per file descriptor) and is of course
400 hard to detect.
401
402 Epoll is also notoriously buggy - embedding epoll fds I<should> work, but
403 of course I<doesn't>, and epoll just loves to report events for totally
404 I<different> file descriptors (even already closed ones, so one cannot
405 even remove them from the set) than registered in the set (especially
406 on SMP systems). Libev tries to counter these spurious notifications by
407 employing an additional generation counter and comparing that against the
408 events to filter out spurious ones, recreating the set when required.
409
410 While stopping, setting and starting an I/O watcher in the same iteration
411 will result in some caching, there is still a system call per such
412 incident (because the same I<file descriptor> could point to a different
413 I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
414 file descriptors might not work very well if you register events for both
415 file descriptors.
416
417 Best performance from this backend is achieved by not unregistering all
418 watchers for a file descriptor until it has been closed, if possible,
419 i.e. keep at least one watcher active per fd at all times. Stopping and
420 starting a watcher (without re-setting it) also usually doesn't cause
421 extra overhead. A fork can both result in spurious notifications as well
422 as in libev having to destroy and recreate the epoll object, which can
423 take considerable time and thus should be avoided.
424
425 All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
426 faster than epoll for maybe up to a hundred file descriptors, depending on
427 the usage. So sad.
428
429 While nominally embeddable in other event loops, this feature is broken in
430 all kernel versions tested so far.
431
432 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
433 C<EVBACKEND_POLL>.
434
435 =item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
436
437 Kqueue deserves special mention, as at the time of this writing, it
438 was broken on all BSDs except NetBSD (usually it doesn't work reliably
439 with anything but sockets and pipes, except on Darwin, where of course
440 it's completely useless). Unlike epoll, however, whose brokenness
441 is by design, these kqueue bugs can (and eventually will) be fixed
442 without API changes to existing programs. For this reason it's not being
443 "auto-detected" unless you explicitly specify it in the flags (i.e. using
444 C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
445 system like NetBSD.
446
447 You still can embed kqueue into a normal poll or select backend and use it
448 only for sockets (after having made sure that sockets work with kqueue on
449 the target platform). See C<ev_embed> watchers for more info.
450
451 It scales in the same way as the epoll backend, but the interface to the
452 kernel is more efficient (which says nothing about its actual speed, of
453 course). While stopping, setting and starting an I/O watcher does never
454 cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
455 two event changes per incident. Support for C<fork ()> is very bad (but
456 sane, unlike epoll) and it drops fds silently in similarly hard-to-detect
457 cases
458
459 This backend usually performs well under most conditions.
460
461 While nominally embeddable in other event loops, this doesn't work
462 everywhere, so you might need to test for this. And since it is broken
463 almost everywhere, you should only use it when you have a lot of sockets
464 (for which it usually works), by embedding it into another event loop
465 (e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
466 also broken on OS X)) and, did I mention it, using it only for sockets.
467
468 This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
469 C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
470 C<NOTE_EOF>.
471
472 =item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
473
474 This is not implemented yet (and might never be, unless you send me an
475 implementation). According to reports, C</dev/poll> only supports sockets
476 and is not embeddable, which would limit the usefulness of this backend
477 immensely.
478
479 =item C<EVBACKEND_PORT> (value 32, Solaris 10)
480
481 This uses the Solaris 10 event port mechanism. As with everything on Solaris,
482 it's really slow, but it still scales very well (O(active_fds)).
483
484 Please note that Solaris event ports can deliver a lot of spurious
485 notifications, so you need to use non-blocking I/O or other means to avoid
486 blocking when no data (or space) is available.
487
488 While this backend scales well, it requires one system call per active
489 file descriptor per loop iteration. For small and medium numbers of file
490 descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
491 might perform better.
492
493 On the positive side, with the exception of the spurious readiness
494 notifications, this backend actually performed fully to specification
495 in all tests and is fully embeddable, which is a rare feat among the
496 OS-specific backends (I vastly prefer correctness over speed hacks).
497
498 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
499 C<EVBACKEND_POLL>.
500
501 =item C<EVBACKEND_ALL>
502
503 Try all backends (even potentially broken ones that wouldn't be tried
504 with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
505 C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.
506
507 It is definitely not recommended to use this flag.
508
509 =back
510
511 If one or more of these are or'ed into the flags value, then only these
512 backends will be tried (in the reverse order as listed here). If none are
513 specified, all backends in C<ev_recommended_backends ()> will be tried.
514
515 Example: This is the most typical usage.
516
517 if (!ev_default_loop (0))
518 fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
519
520 Example: Restrict libev to the select and poll backends, and do not allow
521 environment settings to be taken into account:
522
523 ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
524
525 Example: Use whatever libev has to offer, but make sure that kqueue is
526 used if available (warning, breaks stuff, best use only with your own
527 private event loop and only if you know the OS supports your types of
528 fds):
529
530 ev_default_loop (ev_recommended_backends () | EVBACKEND_KQUEUE);
531
532 =item struct ev_loop *ev_loop_new (unsigned int flags)
533
534 Similar to C<ev_default_loop>, but always creates a new event loop that is
535 always distinct from the default loop. Unlike the default loop, it cannot
536 handle signal and child watchers, and attempts to do so will be greeted by
537 undefined behaviour (or a failed assertion if assertions are enabled).
538
539 Note that this function I<is> thread-safe, and the recommended way to use
540 libev with threads is indeed to create one loop per thread, and using the
541 default loop in the "main" or "initial" thread.
542
543 Example: Try to create a event loop that uses epoll and nothing else.
544
545 struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
546 if (!epoller)
547 fatal ("no epoll found here, maybe it hides under your chair");
548
549 =item ev_default_destroy ()
550
551 Destroys the default loop again (frees all memory and kernel state
552 etc.). None of the active event watchers will be stopped in the normal
553 sense, so e.g. C<ev_is_active> might still return true. It is your
554 responsibility to either stop all watchers cleanly yourself I<before>
555 calling this function, or cope with the fact afterwards (which is usually
556 the easiest thing, you can just ignore the watchers and/or C<free ()> them
557 for example).
558
559 Note that certain global state, such as signal state (and installed signal
560 handlers), will not be freed by this function, and related watchers (such
561 as signal and child watchers) would need to be stopped manually.
562
563 In general it is not advisable to call this function except in the
564 rare occasion where you really need to free e.g. the signal handling
565 pipe fds. If you need dynamically allocated loops it is better to use
566 C<ev_loop_new> and C<ev_loop_destroy>).
567
568 =item ev_loop_destroy (loop)
569
570 Like C<ev_default_destroy>, but destroys an event loop created by an
571 earlier call to C<ev_loop_new>.
572
573 =item ev_default_fork ()
574
575 This function sets a flag that causes subsequent C<ev_loop> iterations
576 to reinitialise the kernel state for backends that have one. Despite the
577 name, you can call it anytime, but it makes most sense after forking, in
578 the child process (or both child and parent, but that again makes little
579 sense). You I<must> call it in the child before using any of the libev
580 functions, and it will only take effect at the next C<ev_loop> iteration.
581
582 On the other hand, you only need to call this function in the child
583 process if and only if you want to use the event library in the child. If
584 you just fork+exec, you don't have to call it at all.
585
586 The function itself is quite fast and it's usually not a problem to call
587 it just in case after a fork. To make this easy, the function will fit in
588 quite nicely into a call to C<pthread_atfork>:
589
590 pthread_atfork (0, 0, ev_default_fork);
591
592 =item ev_loop_fork (loop)
593
594 Like C<ev_default_fork>, but acts on an event loop created by
595 C<ev_loop_new>. Yes, you have to call this on every allocated event loop
596 after fork that you want to re-use in the child, and how you do this is
597 entirely your own problem.
598
599 =item int ev_is_default_loop (loop)
600
601 Returns true when the given loop is, in fact, the default loop, and false
602 otherwise.
603
604 =item unsigned int ev_loop_count (loop)
605
606 Returns the count of loop iterations for the loop, which is identical to
607 the number of times libev did poll for new events. It starts at C<0> and
608 happily wraps around with enough iterations.
609
610 This value can sometimes be useful as a generation counter of sorts (it
611 "ticks" the number of loop iterations), as it roughly corresponds with
612 C<ev_prepare> and C<ev_check> calls.
613
614 =item unsigned int ev_backend (loop)
615
616 Returns one of the C<EVBACKEND_*> flags indicating the event backend in
617 use.
618
619 =item ev_tstamp ev_now (loop)
620
621 Returns the current "event loop time", which is the time the event loop
622 received events and started processing them. This timestamp does not
623 change as long as callbacks are being processed, and this is also the base
624 time used for relative timers. You can treat it as the timestamp of the
625 event occurring (or more correctly, libev finding out about it).
626
627 =item ev_now_update (loop)
628
629 Establishes the current time by querying the kernel, updating the time
630 returned by C<ev_now ()> in the progress. This is a costly operation and
631 is usually done automatically within C<ev_loop ()>.
632
633 This function is rarely useful, but when some event callback runs for a
634 very long time without entering the event loop, updating libev's idea of
635 the current time is a good idea.
636
637 See also "The special problem of time updates" in the C<ev_timer> section.
638
639 =item ev_loop (loop, int flags)
640
641 Finally, this is it, the event handler. This function usually is called
642 after you initialised all your watchers and you want to start handling
643 events.
644
645 If the flags argument is specified as C<0>, it will not return until
646 either no event watchers are active anymore or C<ev_unloop> was called.
647
648 Please note that an explicit C<ev_unloop> is usually better than
649 relying on all watchers to be stopped when deciding when a program has
650 finished (especially in interactive programs), but having a program
651 that automatically loops as long as it has to and no longer by virtue
652 of relying on its watchers stopping correctly, that is truly a thing of
653 beauty.
654
655 A flags value of C<EVLOOP_NONBLOCK> will look for new events, will handle
656 those events and any already outstanding ones, but will not block your
657 process in case there are no events and will return after one iteration of
658 the loop.
659
660 A flags value of C<EVLOOP_ONESHOT> will look for new events (waiting if
661 necessary) and will handle those and any already outstanding ones. It
662 will block your process until at least one new event arrives (which could
663 be an event internal to libev itself, so there is no guarantee that a
664 user-registered callback will be called), and will return after one
665 iteration of the loop.
666
667 This is useful if you are waiting for some external event in conjunction
668 with something not expressible using other libev watchers (i.e. "roll your
669 own C<ev_loop>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
670 usually a better approach for this kind of thing.
671
672 Here are the gory details of what C<ev_loop> does:
673
674 - Before the first iteration, call any pending watchers.
675 * If EVFLAG_FORKCHECK was used, check for a fork.
676 - If a fork was detected (by any means), queue and call all fork watchers.
677 - Queue and call all prepare watchers.
678 - If we have been forked, detach and recreate the kernel state
679 as to not disturb the other process.
680 - Update the kernel state with all outstanding changes.
681 - Update the "event loop time" (ev_now ()).
682 - Calculate for how long to sleep or block, if at all
683 (active idle watchers, EVLOOP_NONBLOCK or not having
684 any active watchers at all will result in not sleeping).
685 - Sleep if the I/O and timer collect interval say so.
686 - Block the process, waiting for any events.
687 - Queue all outstanding I/O (fd) events.
688 - Update the "event loop time" (ev_now ()), and do time jump adjustments.
689 - Queue all expired timers.
690 - Queue all expired periodics.
691 - Unless any events are pending now, queue all idle watchers.
692 - Queue all check watchers.
693 - Call all queued watchers in reverse order (i.e. check watchers first).
694 Signals and child watchers are implemented as I/O watchers, and will
695 be handled here by queueing them when their watcher gets executed.
696 - If ev_unloop has been called, or EVLOOP_ONESHOT or EVLOOP_NONBLOCK
697 were used, or there are no active watchers, return, otherwise
698 continue with step *.
699
700 Example: Queue some jobs and then loop until no events are outstanding
701 anymore.
702
703 ... queue jobs here, make sure they register event watchers as long
704 ... as they still have work to do (even an idle watcher will do..)
705 ev_loop (my_loop, 0);
706 ... jobs done or somebody called unloop. yeah!
707
708 =item ev_unloop (loop, how)
709
710 Can be used to make a call to C<ev_loop> return early (but only after it
711 has processed all outstanding events). The C<how> argument must be either
712 C<EVUNLOOP_ONE>, which will make the innermost C<ev_loop> call return, or
713 C<EVUNLOOP_ALL>, which will make all nested C<ev_loop> calls return.
714
715 This "unloop state" will be cleared when entering C<ev_loop> again.
716
717 It is safe to call C<ev_unloop> from otuside any C<ev_loop> calls.
718
719 =item ev_ref (loop)
720
721 =item ev_unref (loop)
722
723 Ref/unref can be used to add or remove a reference count on the event
724 loop: Every watcher keeps one reference, and as long as the reference
725 count is nonzero, C<ev_loop> will not return on its own.
726
727 If you have a watcher you never unregister that should not keep C<ev_loop>
728 from returning, call ev_unref() after starting, and ev_ref() before
729 stopping it.
730
731 As an example, libev itself uses this for its internal signal pipe: It is
732 not visible to the libev user and should not keep C<ev_loop> from exiting
733 if no event watchers registered by it are active. It is also an excellent
734 way to do this for generic recurring timers or from within third-party
735 libraries. Just remember to I<unref after start> and I<ref before stop>
736 (but only if the watcher wasn't active before, or was active before,
737 respectively).
738
739 Example: Create a signal watcher, but keep it from keeping C<ev_loop>
740 running when nothing else is active.
741
742 ev_signal exitsig;
743 ev_signal_init (&exitsig, sig_cb, SIGINT);
744 ev_signal_start (loop, &exitsig);
745 evf_unref (loop);
746
747 Example: For some weird reason, unregister the above signal handler again.
748
749 ev_ref (loop);
750 ev_signal_stop (loop, &exitsig);
751
752 =item ev_set_io_collect_interval (loop, ev_tstamp interval)
753
754 =item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
755
756 These advanced functions influence the time that libev will spend waiting
757 for events. Both time intervals are by default C<0>, meaning that libev
758 will try to invoke timer/periodic callbacks and I/O callbacks with minimum
759 latency.
760
761 Setting these to a higher value (the C<interval> I<must> be >= C<0>)
762 allows libev to delay invocation of I/O and timer/periodic callbacks
763 to increase efficiency of loop iterations (or to increase power-saving
764 opportunities).
765
766 The idea is that sometimes your program runs just fast enough to handle
767 one (or very few) event(s) per loop iteration. While this makes the
768 program responsive, it also wastes a lot of CPU time to poll for new
769 events, especially with backends like C<select ()> which have a high
770 overhead for the actual polling but can deliver many events at once.
771
772 By setting a higher I<io collect interval> you allow libev to spend more
773 time collecting I/O events, so you can handle more events per iteration,
774 at the cost of increasing latency. Timeouts (both C<ev_periodic> and
775 C<ev_timer>) will be not affected. Setting this to a non-null value will
776 introduce an additional C<ev_sleep ()> call into most loop iterations.
777
778 Likewise, by setting a higher I<timeout collect interval> you allow libev
779 to spend more time collecting timeouts, at the expense of increased
780 latency/jitter/inexactness (the watcher callback will be called
781 later). C<ev_io> watchers will not be affected. Setting this to a non-null
782 value will not introduce any overhead in libev.
783
784 Many (busy) programs can usually benefit by setting the I/O collect
785 interval to a value near C<0.1> or so, which is often enough for
786 interactive servers (of course not for games), likewise for timeouts. It
787 usually doesn't make much sense to set it to a lower value than C<0.01>,
788 as this approaches the timing granularity of most systems.
789
790 Setting the I<timeout collect interval> can improve the opportunity for
791 saving power, as the program will "bundle" timer callback invocations that
792 are "near" in time together, by delaying some, thus reducing the number of
793 times the process sleeps and wakes up again. Another useful technique to
794 reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
795 they fire on, say, one-second boundaries only.
796
797 =item ev_loop_verify (loop)
798
799 This function only does something when C<EV_VERIFY> support has been
800 compiled in, which is the default for non-minimal builds. It tries to go
801 through all internal structures and checks them for validity. If anything
802 is found to be inconsistent, it will print an error message to standard
803 error and call C<abort ()>.
804
805 This can be used to catch bugs inside libev itself: under normal
806 circumstances, this function will never abort as of course libev keeps its
807 data structures consistent.
808
809 =back
810
811
812 =head1 ANATOMY OF A WATCHER
813
814 In the following description, uppercase C<TYPE> in names stands for the
815 watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
816 watchers and C<ev_io_start> for I/O watchers.
817
818 A watcher is a structure that you create and register to record your
819 interest in some event. For instance, if you want to wait for STDIN to
820 become readable, you would create an C<ev_io> watcher for that:
821
822 static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
823 {
824 ev_io_stop (w);
825 ev_unloop (loop, EVUNLOOP_ALL);
826 }
827
828 struct ev_loop *loop = ev_default_loop (0);
829
830 ev_io stdin_watcher;
831
832 ev_init (&stdin_watcher, my_cb);
833 ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
834 ev_io_start (loop, &stdin_watcher);
835
836 ev_loop (loop, 0);
837
838 As you can see, you are responsible for allocating the memory for your
839 watcher structures (and it is I<usually> a bad idea to do this on the
840 stack).
841
842 Each watcher has an associated watcher structure (called C<struct ev_TYPE>
843 or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
844
845 Each watcher structure must be initialised by a call to C<ev_init
846 (watcher *, callback)>, which expects a callback to be provided. This
847 callback gets invoked each time the event occurs (or, in the case of I/O
848 watchers, each time the event loop detects that the file descriptor given
849 is readable and/or writable).
850
851 Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
852 macro to configure it, with arguments specific to the watcher type. There
853 is also a macro to combine initialisation and setting in one call: C<<
854 ev_TYPE_init (watcher *, callback, ...) >>.
855
856 To make the watcher actually watch out for events, you have to start it
857 with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
858 *) >>), and you can stop watching for events at any time by calling the
859 corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
860
861 As long as your watcher is active (has been started but not stopped) you
862 must not touch the values stored in it. Most specifically you must never
863 reinitialise it or call its C<ev_TYPE_set> macro.
864
865 Each and every callback receives the event loop pointer as first, the
866 registered watcher structure as second, and a bitset of received events as
867 third argument.
868
869 The received events usually include a single bit per event type received
870 (you can receive multiple events at the same time). The possible bit masks
871 are:
872
873 =over 4
874
875 =item C<EV_READ>
876
877 =item C<EV_WRITE>
878
879 The file descriptor in the C<ev_io> watcher has become readable and/or
880 writable.
881
882 =item C<EV_TIMEOUT>
883
884 The C<ev_timer> watcher has timed out.
885
886 =item C<EV_PERIODIC>
887
888 The C<ev_periodic> watcher has timed out.
889
890 =item C<EV_SIGNAL>
891
892 The signal specified in the C<ev_signal> watcher has been received by a thread.
893
894 =item C<EV_CHILD>
895
896 The pid specified in the C<ev_child> watcher has received a status change.
897
898 =item C<EV_STAT>
899
900 The path specified in the C<ev_stat> watcher changed its attributes somehow.
901
902 =item C<EV_IDLE>
903
904 The C<ev_idle> watcher has determined that you have nothing better to do.
905
906 =item C<EV_PREPARE>
907
908 =item C<EV_CHECK>
909
910 All C<ev_prepare> watchers are invoked just I<before> C<ev_loop> starts
911 to gather new events, and all C<ev_check> watchers are invoked just after
912 C<ev_loop> has gathered them, but before it invokes any callbacks for any
913 received events. Callbacks of both watcher types can start and stop as
914 many watchers as they want, and all of them will be taken into account
915 (for example, a C<ev_prepare> watcher might start an idle watcher to keep
916 C<ev_loop> from blocking).
917
918 =item C<EV_EMBED>
919
920 The embedded event loop specified in the C<ev_embed> watcher needs attention.
921
922 =item C<EV_FORK>
923
924 The event loop has been resumed in the child process after fork (see
925 C<ev_fork>).
926
927 =item C<EV_ASYNC>
928
929 The given async watcher has been asynchronously notified (see C<ev_async>).
930
931 =item C<EV_ERROR>
932
933 An unspecified error has occurred, the watcher has been stopped. This might
934 happen because the watcher could not be properly started because libev
935 ran out of memory, a file descriptor was found to be closed or any other
936 problem. Libev considers these application bugs.
937
938 You best act on it by reporting the problem and somehow coping with the
939 watcher being stopped. Note that well-written programs should not receive
940 an error ever, so when your watcher receives it, this usually indicates a
941 bug in your program.
942
943 Libev will usually signal a few "dummy" events together with an error, for
944 example it might indicate that a fd is readable or writable, and if your
945 callbacks is well-written it can just attempt the operation and cope with
946 the error from read() or write(). This will not work in multi-threaded
947 programs, though, as the fd could already be closed and reused for another
948 thing, so beware.
949
950 =back
951
952 =head2 GENERIC WATCHER FUNCTIONS
953
954 =over 4
955
956 =item C<ev_init> (ev_TYPE *watcher, callback)
957
958 This macro initialises the generic portion of a watcher. The contents
959 of the watcher object can be arbitrary (so C<malloc> will do). Only
960 the generic parts of the watcher are initialised, you I<need> to call
961 the type-specific C<ev_TYPE_set> macro afterwards to initialise the
962 type-specific parts. For each type there is also a C<ev_TYPE_init> macro
963 which rolls both calls into one.
964
965 You can reinitialise a watcher at any time as long as it has been stopped
966 (or never started) and there are no pending events outstanding.
967
968 The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
969 int revents)>.
970
971 Example: Initialise an C<ev_io> watcher in two steps.
972
973 ev_io w;
974 ev_init (&w, my_cb);
975 ev_io_set (&w, STDIN_FILENO, EV_READ);
976
977 =item C<ev_TYPE_set> (ev_TYPE *, [args])
978
979 This macro initialises the type-specific parts of a watcher. You need to
980 call C<ev_init> at least once before you call this macro, but you can
981 call C<ev_TYPE_set> any number of times. You must not, however, call this
982 macro on a watcher that is active (it can be pending, however, which is a
983 difference to the C<ev_init> macro).
984
985 Although some watcher types do not have type-specific arguments
986 (e.g. C<ev_prepare>) you still need to call its C<set> macro.
987
988 See C<ev_init>, above, for an example.
989
990 =item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
991
992 This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
993 calls into a single call. This is the most convenient method to initialise
994 a watcher. The same limitations apply, of course.
995
996 Example: Initialise and set an C<ev_io> watcher in one step.
997
998 ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
999
1000 =item C<ev_TYPE_start> (loop *, ev_TYPE *watcher)
1001
1002 Starts (activates) the given watcher. Only active watchers will receive
1003 events. If the watcher is already active nothing will happen.
1004
1005 Example: Start the C<ev_io> watcher that is being abused as example in this
1006 whole section.
1007
1008 ev_io_start (EV_DEFAULT_UC, &w);
1009
1010 =item C<ev_TYPE_stop> (loop *, ev_TYPE *watcher)
1011
1012 Stops the given watcher if active, and clears the pending status (whether
1013 the watcher was active or not).
1014
1015 It is possible that stopped watchers are pending - for example,
1016 non-repeating timers are being stopped when they become pending - but
1017 calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
1018 pending. If you want to free or reuse the memory used by the watcher it is
1019 therefore a good idea to always call its C<ev_TYPE_stop> function.
1020
1021 =item bool ev_is_active (ev_TYPE *watcher)
1022
1023 Returns a true value iff the watcher is active (i.e. it has been started
1024 and not yet been stopped). As long as a watcher is active you must not modify
1025 it.
1026
1027 =item bool ev_is_pending (ev_TYPE *watcher)
1028
1029 Returns a true value iff the watcher is pending, (i.e. it has outstanding
1030 events but its callback has not yet been invoked). As long as a watcher
1031 is pending (but not active) you must not call an init function on it (but
1032 C<ev_TYPE_set> is safe), you must not change its priority, and you must
1033 make sure the watcher is available to libev (e.g. you cannot C<free ()>
1034 it).
1035
1036 =item callback ev_cb (ev_TYPE *watcher)
1037
1038 Returns the callback currently set on the watcher.
1039
1040 =item ev_cb_set (ev_TYPE *watcher, callback)
1041
1042 Change the callback. You can change the callback at virtually any time
1043 (modulo threads).
1044
1045 =item ev_set_priority (ev_TYPE *watcher, priority)
1046
1047 =item int ev_priority (ev_TYPE *watcher)
1048
1049 Set and query the priority of the watcher. The priority is a small
1050 integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
1051 (default: C<-2>). Pending watchers with higher priority will be invoked
1052 before watchers with lower priority, but priority will not keep watchers
1053 from being executed (except for C<ev_idle> watchers).
1054
1055 This means that priorities are I<only> used for ordering callback
1056 invocation after new events have been received. This is useful, for
1057 example, to reduce latency after idling, or more often, to bind two
1058 watchers on the same event and make sure one is called first.
1059
1060 If you need to suppress invocation when higher priority events are pending
1061 you need to look at C<ev_idle> watchers, which provide this functionality.
1062
1063 You I<must not> change the priority of a watcher as long as it is active or
1064 pending.
1065
1066 The default priority used by watchers when no priority has been set is
1067 always C<0>, which is supposed to not be too high and not be too low :).
1068
1069 Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
1070 fine, as long as you do not mind that the priority value you query might
1071 or might not have been clamped to the valid range.
1072
1073 =item ev_invoke (loop, ev_TYPE *watcher, int revents)
1074
1075 Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
1076 C<loop> nor C<revents> need to be valid as long as the watcher callback
1077 can deal with that fact, as both are simply passed through to the
1078 callback.
1079
1080 =item int ev_clear_pending (loop, ev_TYPE *watcher)
1081
1082 If the watcher is pending, this function clears its pending status and
1083 returns its C<revents> bitset (as if its callback was invoked). If the
1084 watcher isn't pending it does nothing and returns C<0>.
1085
1086 Sometimes it can be useful to "poll" a watcher instead of waiting for its
1087 callback to be invoked, which can be accomplished with this function.
1088
1089 =back
1090
1091
1092 =head2 ASSOCIATING CUSTOM DATA WITH A WATCHER
1093
1094 Each watcher has, by default, a member C<void *data> that you can change
1095 and read at any time: libev will completely ignore it. This can be used
1096 to associate arbitrary data with your watcher. If you need more data and
1097 don't want to allocate memory and store a pointer to it in that data
1098 member, you can also "subclass" the watcher type and provide your own
1099 data:
1100
1101 struct my_io
1102 {
1103 ev_io io;
1104 int otherfd;
1105 void *somedata;
1106 struct whatever *mostinteresting;
1107 };
1108
1109 ...
1110 struct my_io w;
1111 ev_io_init (&w.io, my_cb, fd, EV_READ);
1112
1113 And since your callback will be called with a pointer to the watcher, you
1114 can cast it back to your own type:
1115
1116 static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
1117 {
1118 struct my_io *w = (struct my_io *)w_;
1119 ...
1120 }
1121
1122 More interesting and less C-conformant ways of casting your callback type
1123 instead have been omitted.
1124
1125 Another common scenario is to use some data structure with multiple
1126 embedded watchers:
1127
1128 struct my_biggy
1129 {
1130 int some_data;
1131 ev_timer t1;
1132 ev_timer t2;
1133 }
1134
1135 In this case getting the pointer to C<my_biggy> is a bit more
1136 complicated: Either you store the address of your C<my_biggy> struct
1137 in the C<data> member of the watcher (for woozies), or you need to use
1138 some pointer arithmetic using C<offsetof> inside your watchers (for real
1139 programmers):
1140
1141 #include <stddef.h>
1142
1143 static void
1144 t1_cb (EV_P_ ev_timer *w, int revents)
1145 {
1146 struct my_biggy big = (struct my_biggy *
1147 (((char *)w) - offsetof (struct my_biggy, t1));
1148 }
1149
1150 static void
1151 t2_cb (EV_P_ ev_timer *w, int revents)
1152 {
1153 struct my_biggy big = (struct my_biggy *
1154 (((char *)w) - offsetof (struct my_biggy, t2));
1155 }
1156
1157
1158 =head1 WATCHER TYPES
1159
1160 This section describes each watcher in detail, but will not repeat
1161 information given in the last section. Any initialisation/set macros,
1162 functions and members specific to the watcher type are explained.
1163
1164 Members are additionally marked with either I<[read-only]>, meaning that,
1165 while the watcher is active, you can look at the member and expect some
1166 sensible content, but you must not modify it (you can modify it while the
1167 watcher is stopped to your hearts content), or I<[read-write]>, which
1168 means you can expect it to have some sensible content while the watcher
1169 is active, but you can also modify it. Modifying it may not do something
1170 sensible or take immediate effect (or do anything at all), but libev will
1171 not crash or malfunction in any way.
1172
1173
1174 =head2 C<ev_io> - is this file descriptor readable or writable?
1175
1176 I/O watchers check whether a file descriptor is readable or writable
1177 in each iteration of the event loop, or, more precisely, when reading
1178 would not block the process and writing would at least be able to write
1179 some data. This behaviour is called level-triggering because you keep
1180 receiving events as long as the condition persists. Remember you can stop
1181 the watcher if you don't want to act on the event and neither want to
1182 receive future events.
1183
1184 In general you can register as many read and/or write event watchers per
1185 fd as you want (as long as you don't confuse yourself). Setting all file
1186 descriptors to non-blocking mode is also usually a good idea (but not
1187 required if you know what you are doing).
1188
1189 If you cannot use non-blocking mode, then force the use of a
1190 known-to-be-good backend (at the time of this writing, this includes only
1191 C<EVBACKEND_SELECT> and C<EVBACKEND_POLL>).
1192
1193 Another thing you have to watch out for is that it is quite easy to
1194 receive "spurious" readiness notifications, that is your callback might
1195 be called with C<EV_READ> but a subsequent C<read>(2) will actually block
1196 because there is no data. Not only are some backends known to create a
1197 lot of those (for example Solaris ports), it is very easy to get into
1198 this situation even with a relatively standard program structure. Thus
1199 it is best to always use non-blocking I/O: An extra C<read>(2) returning
1200 C<EAGAIN> is far preferable to a program hanging until some data arrives.
1201
1202 If you cannot run the fd in non-blocking mode (for example you should
1203 not play around with an Xlib connection), then you have to separately
1204 re-test whether a file descriptor is really ready with a known-to-be good
1205 interface such as poll (fortunately in our Xlib example, Xlib already
1206 does this on its own, so its quite safe to use). Some people additionally
1207 use C<SIGALRM> and an interval timer, just to be sure you won't block
1208 indefinitely.
1209
1210 But really, best use non-blocking mode.
1211
1212 =head3 The special problem of disappearing file descriptors
1213
1214 Some backends (e.g. kqueue, epoll) need to be told about closing a file
1215 descriptor (either due to calling C<close> explicitly or any other means,
1216 such as C<dup2>). The reason is that you register interest in some file
1217 descriptor, but when it goes away, the operating system will silently drop
1218 this interest. If another file descriptor with the same number then is
1219 registered with libev, there is no efficient way to see that this is, in
1220 fact, a different file descriptor.
1221
1222 To avoid having to explicitly tell libev about such cases, libev follows
1223 the following policy: Each time C<ev_io_set> is being called, libev
1224 will assume that this is potentially a new file descriptor, otherwise
1225 it is assumed that the file descriptor stays the same. That means that
1226 you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
1227 descriptor even if the file descriptor number itself did not change.
1228
1229 This is how one would do it normally anyway, the important point is that
1230 the libev application should not optimise around libev but should leave
1231 optimisations to libev.
1232
1233 =head3 The special problem of dup'ed file descriptors
1234
1235 Some backends (e.g. epoll), cannot register events for file descriptors,
1236 but only events for the underlying file descriptions. That means when you
1237 have C<dup ()>'ed file descriptors or weirder constellations, and register
1238 events for them, only one file descriptor might actually receive events.
1239
1240 There is no workaround possible except not registering events
1241 for potentially C<dup ()>'ed file descriptors, or to resort to
1242 C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
1243
1244 =head3 The special problem of fork
1245
1246 Some backends (epoll, kqueue) do not support C<fork ()> at all or exhibit
1247 useless behaviour. Libev fully supports fork, but needs to be told about
1248 it in the child.
1249
1250 To support fork in your programs, you either have to call
1251 C<ev_default_fork ()> or C<ev_loop_fork ()> after a fork in the child,
1252 enable C<EVFLAG_FORKCHECK>, or resort to C<EVBACKEND_SELECT> or
1253 C<EVBACKEND_POLL>.
1254
1255 =head3 The special problem of SIGPIPE
1256
1257 While not really specific to libev, it is easy to forget about C<SIGPIPE>:
1258 when writing to a pipe whose other end has been closed, your program gets
1259 sent a SIGPIPE, which, by default, aborts your program. For most programs
1260 this is sensible behaviour, for daemons, this is usually undesirable.
1261
1262 So when you encounter spurious, unexplained daemon exits, make sure you
1263 ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
1264 somewhere, as that would have given you a big clue).
1265
1266
1267 =head3 Watcher-Specific Functions
1268
1269 =over 4
1270
1271 =item ev_io_init (ev_io *, callback, int fd, int events)
1272
1273 =item ev_io_set (ev_io *, int fd, int events)
1274
1275 Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
1276 receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
1277 C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
1278
1279 =item int fd [read-only]
1280
1281 The file descriptor being watched.
1282
1283 =item int events [read-only]
1284
1285 The events being watched.
1286
1287 =back
1288
1289 =head3 Examples
1290
1291 Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
1292 readable, but only once. Since it is likely line-buffered, you could
1293 attempt to read a whole line in the callback.
1294
1295 static void
1296 stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
1297 {
1298 ev_io_stop (loop, w);
1299 .. read from stdin here (or from w->fd) and handle any I/O errors
1300 }
1301
1302 ...
1303 struct ev_loop *loop = ev_default_init (0);
1304 ev_io stdin_readable;
1305 ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
1306 ev_io_start (loop, &stdin_readable);
1307 ev_loop (loop, 0);
1308
1309
1310 =head2 C<ev_timer> - relative and optionally repeating timeouts
1311
1312 Timer watchers are simple relative timers that generate an event after a
1313 given time, and optionally repeating in regular intervals after that.
1314
1315 The timers are based on real time, that is, if you register an event that
1316 times out after an hour and you reset your system clock to January last
1317 year, it will still time out after (roughly) one hour. "Roughly" because
1318 detecting time jumps is hard, and some inaccuracies are unavoidable (the
1319 monotonic clock option helps a lot here).
1320
1321 The callback is guaranteed to be invoked only I<after> its timeout has
1322 passed, but if multiple timers become ready during the same loop iteration
1323 then order of execution is undefined.
1324
1325 =head3 Be smart about timeouts
1326
1327 Many real-world problems involve some kind of timeout, usually for error
1328 recovery. A typical example is an HTTP request - if the other side hangs,
1329 you want to raise some error after a while.
1330
1331 What follows are some ways to handle this problem, from obvious and
1332 inefficient to smart and efficient.
1333
1334 In the following, a 60 second activity timeout is assumed - a timeout that
1335 gets reset to 60 seconds each time there is activity (e.g. each time some
1336 data or other life sign was received).
1337
1338 =over 4
1339
1340 =item 1. Use a timer and stop, reinitialise and start it on activity.
1341
1342 This is the most obvious, but not the most simple way: In the beginning,
1343 start the watcher:
1344
1345 ev_timer_init (timer, callback, 60., 0.);
1346 ev_timer_start (loop, timer);
1347
1348 Then, each time there is some activity, C<ev_timer_stop> it, initialise it
1349 and start it again:
1350
1351 ev_timer_stop (loop, timer);
1352 ev_timer_set (timer, 60., 0.);
1353 ev_timer_start (loop, timer);
1354
1355 This is relatively simple to implement, but means that each time there is
1356 some activity, libev will first have to remove the timer from its internal
1357 data structure and then add it again. Libev tries to be fast, but it's
1358 still not a constant-time operation.
1359
1360 =item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
1361
1362 This is the easiest way, and involves using C<ev_timer_again> instead of
1363 C<ev_timer_start>.
1364
1365 To implement this, configure an C<ev_timer> with a C<repeat> value
1366 of C<60> and then call C<ev_timer_again> at start and each time you
1367 successfully read or write some data. If you go into an idle state where
1368 you do not expect data to travel on the socket, you can C<ev_timer_stop>
1369 the timer, and C<ev_timer_again> will automatically restart it if need be.
1370
1371 That means you can ignore both the C<ev_timer_start> function and the
1372 C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
1373 member and C<ev_timer_again>.
1374
1375 At start:
1376
1377 ev_timer_init (timer, callback);
1378 timer->repeat = 60.;
1379 ev_timer_again (loop, timer);
1380
1381 Each time there is some activity:
1382
1383 ev_timer_again (loop, timer);
1384
1385 It is even possible to change the time-out on the fly, regardless of
1386 whether the watcher is active or not:
1387
1388 timer->repeat = 30.;
1389 ev_timer_again (loop, timer);
1390
1391 This is slightly more efficient then stopping/starting the timer each time
1392 you want to modify its timeout value, as libev does not have to completely
1393 remove and re-insert the timer from/into its internal data structure.
1394
1395 It is, however, even simpler than the "obvious" way to do it.
1396
1397 =item 3. Let the timer time out, but then re-arm it as required.
1398
1399 This method is more tricky, but usually most efficient: Most timeouts are
1400 relatively long compared to the intervals between other activity - in
1401 our example, within 60 seconds, there are usually many I/O events with
1402 associated activity resets.
1403
1404 In this case, it would be more efficient to leave the C<ev_timer> alone,
1405 but remember the time of last activity, and check for a real timeout only
1406 within the callback:
1407
1408 ev_tstamp last_activity; // time of last activity
1409
1410 static void
1411 callback (EV_P_ ev_timer *w, int revents)
1412 {
1413 ev_tstamp now = ev_now (EV_A);
1414 ev_tstamp timeout = last_activity + 60.;
1415
1416 // if last_activity + 60. is older than now, we did time out
1417 if (timeout < now)
1418 {
1419 // timeout occured, take action
1420 }
1421 else
1422 {
1423 // callback was invoked, but there was some activity, re-arm
1424 // the watcher to fire in last_activity + 60, which is
1425 // guaranteed to be in the future, so "again" is positive:
1426 w->repeat = timeout - now;
1427 ev_timer_again (EV_A_ w);
1428 }
1429 }
1430
1431 To summarise the callback: first calculate the real timeout (defined
1432 as "60 seconds after the last activity"), then check if that time has
1433 been reached, which means something I<did>, in fact, time out. Otherwise
1434 the callback was invoked too early (C<timeout> is in the future), so
1435 re-schedule the timer to fire at that future time, to see if maybe we have
1436 a timeout then.
1437
1438 Note how C<ev_timer_again> is used, taking advantage of the
1439 C<ev_timer_again> optimisation when the timer is already running.
1440
1441 This scheme causes more callback invocations (about one every 60 seconds
1442 minus half the average time between activity), but virtually no calls to
1443 libev to change the timeout.
1444
1445 To start the timer, simply initialise the watcher and set C<last_activity>
1446 to the current time (meaning we just have some activity :), then call the
1447 callback, which will "do the right thing" and start the timer:
1448
1449 ev_timer_init (timer, callback);
1450 last_activity = ev_now (loop);
1451 callback (loop, timer, EV_TIMEOUT);
1452
1453 And when there is some activity, simply store the current time in
1454 C<last_activity>, no libev calls at all:
1455
1456 last_actiivty = ev_now (loop);
1457
1458 This technique is slightly more complex, but in most cases where the
1459 time-out is unlikely to be triggered, much more efficient.
1460
1461 Changing the timeout is trivial as well (if it isn't hard-coded in the
1462 callback :) - just change the timeout and invoke the callback, which will
1463 fix things for you.
1464
1465 =item 4. Wee, just use a double-linked list for your timeouts.
1466
1467 If there is not one request, but many thousands (millions...), all
1468 employing some kind of timeout with the same timeout value, then one can
1469 do even better:
1470
1471 When starting the timeout, calculate the timeout value and put the timeout
1472 at the I<end> of the list.
1473
1474 Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
1475 the list is expected to fire (for example, using the technique #3).
1476
1477 When there is some activity, remove the timer from the list, recalculate
1478 the timeout, append it to the end of the list again, and make sure to
1479 update the C<ev_timer> if it was taken from the beginning of the list.
1480
1481 This way, one can manage an unlimited number of timeouts in O(1) time for
1482 starting, stopping and updating the timers, at the expense of a major
1483 complication, and having to use a constant timeout. The constant timeout
1484 ensures that the list stays sorted.
1485
1486 =back
1487
1488 So which method the best?
1489
1490 Method #2 is a simple no-brain-required solution that is adequate in most
1491 situations. Method #3 requires a bit more thinking, but handles many cases
1492 better, and isn't very complicated either. In most case, choosing either
1493 one is fine, with #3 being better in typical situations.
1494
1495 Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
1496 rather complicated, but extremely efficient, something that really pays
1497 off after the first million or so of active timers, i.e. it's usually
1498 overkill :)
1499
1500 =head3 The special problem of time updates
1501
1502 Establishing the current time is a costly operation (it usually takes at
1503 least two system calls): EV therefore updates its idea of the current
1504 time only before and after C<ev_loop> collects new events, which causes a
1505 growing difference between C<ev_now ()> and C<ev_time ()> when handling
1506 lots of events in one iteration.
1507
1508 The relative timeouts are calculated relative to the C<ev_now ()>
1509 time. This is usually the right thing as this timestamp refers to the time
1510 of the event triggering whatever timeout you are modifying/starting. If
1511 you suspect event processing to be delayed and you I<need> to base the
1512 timeout on the current time, use something like this to adjust for this:
1513
1514 ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);
1515
1516 If the event loop is suspended for a long time, you can also force an
1517 update of the time returned by C<ev_now ()> by calling C<ev_now_update
1518 ()>.
1519
1520 =head3 Watcher-Specific Functions and Data Members
1521
1522 =over 4
1523
1524 =item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
1525
1526 =item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
1527
1528 Configure the timer to trigger after C<after> seconds. If C<repeat>
1529 is C<0.>, then it will automatically be stopped once the timeout is
1530 reached. If it is positive, then the timer will automatically be
1531 configured to trigger again C<repeat> seconds later, again, and again,
1532 until stopped manually.
1533
1534 The timer itself will do a best-effort at avoiding drift, that is, if
1535 you configure a timer to trigger every 10 seconds, then it will normally
1536 trigger at exactly 10 second intervals. If, however, your program cannot
1537 keep up with the timer (because it takes longer than those 10 seconds to
1538 do stuff) the timer will not fire more than once per event loop iteration.
1539
1540 =item ev_timer_again (loop, ev_timer *)
1541
1542 This will act as if the timer timed out and restart it again if it is
1543 repeating. The exact semantics are:
1544
1545 If the timer is pending, its pending status is cleared.
1546
1547 If the timer is started but non-repeating, stop it (as if it timed out).
1548
1549 If the timer is repeating, either start it if necessary (with the
1550 C<repeat> value), or reset the running timer to the C<repeat> value.
1551
1552 This sounds a bit complicated, see "Be smart about timeouts", above, for a
1553 usage example.
1554
1555 =item ev_tstamp repeat [read-write]
1556
1557 The current C<repeat> value. Will be used each time the watcher times out
1558 or C<ev_timer_again> is called, and determines the next timeout (if any),
1559 which is also when any modifications are taken into account.
1560
1561 =back
1562
1563 =head3 Examples
1564
1565 Example: Create a timer that fires after 60 seconds.
1566
1567 static void
1568 one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
1569 {
1570 .. one minute over, w is actually stopped right here
1571 }
1572
1573 ev_timer mytimer;
1574 ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
1575 ev_timer_start (loop, &mytimer);
1576
1577 Example: Create a timeout timer that times out after 10 seconds of
1578 inactivity.
1579
1580 static void
1581 timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
1582 {
1583 .. ten seconds without any activity
1584 }
1585
1586 ev_timer mytimer;
1587 ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
1588 ev_timer_again (&mytimer); /* start timer */
1589 ev_loop (loop, 0);
1590
1591 // and in some piece of code that gets executed on any "activity":
1592 // reset the timeout to start ticking again at 10 seconds
1593 ev_timer_again (&mytimer);
1594
1595
1596 =head2 C<ev_periodic> - to cron or not to cron?
1597
1598 Periodic watchers are also timers of a kind, but they are very versatile
1599 (and unfortunately a bit complex).
1600
1601 Unlike C<ev_timer>'s, they are not based on real time (or relative time)
1602 but on wall clock time (absolute time). You can tell a periodic watcher
1603 to trigger after some specific point in time. For example, if you tell a
1604 periodic watcher to trigger in 10 seconds (by specifying e.g. C<ev_now ()
1605 + 10.>, that is, an absolute time not a delay) and then reset your system
1606 clock to January of the previous year, then it will take more than year
1607 to trigger the event (unlike an C<ev_timer>, which would still trigger
1608 roughly 10 seconds later as it uses a relative timeout).
1609
1610 C<ev_periodic>s can also be used to implement vastly more complex timers,
1611 such as triggering an event on each "midnight, local time", or other
1612 complicated rules.
1613
1614 As with timers, the callback is guaranteed to be invoked only when the
1615 time (C<at>) has passed, but if multiple periodic timers become ready
1616 during the same loop iteration, then order of execution is undefined.
1617
1618 =head3 Watcher-Specific Functions and Data Members
1619
1620 =over 4
1621
1622 =item ev_periodic_init (ev_periodic *, callback, ev_tstamp at, ev_tstamp interval, reschedule_cb)
1623
1624 =item ev_periodic_set (ev_periodic *, ev_tstamp after, ev_tstamp repeat, reschedule_cb)
1625
1626 Lots of arguments, lets sort it out... There are basically three modes of
1627 operation, and we will explain them from simplest to most complex:
1628
1629 =over 4
1630
1631 =item * absolute timer (at = time, interval = reschedule_cb = 0)
1632
1633 In this configuration the watcher triggers an event after the wall clock
1634 time C<at> has passed. It will not repeat and will not adjust when a time
1635 jump occurs, that is, if it is to be run at January 1st 2011 then it will
1636 only run when the system clock reaches or surpasses this time.
1637
1638 =item * repeating interval timer (at = offset, interval > 0, reschedule_cb = 0)
1639
1640 In this mode the watcher will always be scheduled to time out at the next
1641 C<at + N * interval> time (for some integer N, which can also be negative)
1642 and then repeat, regardless of any time jumps.
1643
1644 This can be used to create timers that do not drift with respect to the
1645 system clock, for example, here is a C<ev_periodic> that triggers each
1646 hour, on the hour:
1647
1648 ev_periodic_set (&periodic, 0., 3600., 0);
1649
1650 This doesn't mean there will always be 3600 seconds in between triggers,
1651 but only that the callback will be called when the system time shows a
1652 full hour (UTC), or more correctly, when the system time is evenly divisible
1653 by 3600.
1654
1655 Another way to think about it (for the mathematically inclined) is that
1656 C<ev_periodic> will try to run the callback in this mode at the next possible
1657 time where C<time = at (mod interval)>, regardless of any time jumps.
1658
1659 For numerical stability it is preferable that the C<at> value is near
1660 C<ev_now ()> (the current time), but there is no range requirement for
1661 this value, and in fact is often specified as zero.
1662
1663 Note also that there is an upper limit to how often a timer can fire (CPU
1664 speed for example), so if C<interval> is very small then timing stability
1665 will of course deteriorate. Libev itself tries to be exact to be about one
1666 millisecond (if the OS supports it and the machine is fast enough).
1667
1668 =item * manual reschedule mode (at and interval ignored, reschedule_cb = callback)
1669
1670 In this mode the values for C<interval> and C<at> are both being
1671 ignored. Instead, each time the periodic watcher gets scheduled, the
1672 reschedule callback will be called with the watcher as first, and the
1673 current time as second argument.
1674
1675 NOTE: I<This callback MUST NOT stop or destroy any periodic watcher,
1676 ever, or make ANY event loop modifications whatsoever>.
1677
1678 If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
1679 it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
1680 only event loop modification you are allowed to do).
1681
1682 The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
1683 *w, ev_tstamp now)>, e.g.:
1684
1685 static ev_tstamp
1686 my_rescheduler (ev_periodic *w, ev_tstamp now)
1687 {
1688 return now + 60.;
1689 }
1690
1691 It must return the next time to trigger, based on the passed time value
1692 (that is, the lowest time value larger than to the second argument). It
1693 will usually be called just before the callback will be triggered, but
1694 might be called at other times, too.
1695
1696 NOTE: I<< This callback must always return a time that is higher than or
1697 equal to the passed C<now> value >>.
1698
1699 This can be used to create very complex timers, such as a timer that
1700 triggers on "next midnight, local time". To do this, you would calculate the
1701 next midnight after C<now> and return the timestamp value for this. How
1702 you do this is, again, up to you (but it is not trivial, which is the main
1703 reason I omitted it as an example).
1704
1705 =back
1706
1707 =item ev_periodic_again (loop, ev_periodic *)
1708
1709 Simply stops and restarts the periodic watcher again. This is only useful
1710 when you changed some parameters or the reschedule callback would return
1711 a different time than the last time it was called (e.g. in a crond like
1712 program when the crontabs have changed).
1713
1714 =item ev_tstamp ev_periodic_at (ev_periodic *)
1715
1716 When active, returns the absolute time that the watcher is supposed to
1717 trigger next.
1718
1719 =item ev_tstamp offset [read-write]
1720
1721 When repeating, this contains the offset value, otherwise this is the
1722 absolute point in time (the C<at> value passed to C<ev_periodic_set>).
1723
1724 Can be modified any time, but changes only take effect when the periodic
1725 timer fires or C<ev_periodic_again> is being called.
1726
1727 =item ev_tstamp interval [read-write]
1728
1729 The current interval value. Can be modified any time, but changes only
1730 take effect when the periodic timer fires or C<ev_periodic_again> is being
1731 called.
1732
1733 =item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
1734
1735 The current reschedule callback, or C<0>, if this functionality is
1736 switched off. Can be changed any time, but changes only take effect when
1737 the periodic timer fires or C<ev_periodic_again> is being called.
1738
1739 =back
1740
1741 =head3 Examples
1742
1743 Example: Call a callback every hour, or, more precisely, whenever the
1744 system time is divisible by 3600. The callback invocation times have
1745 potentially a lot of jitter, but good long-term stability.
1746
1747 static void
1748 clock_cb (struct ev_loop *loop, ev_io *w, int revents)
1749 {
1750 ... its now a full hour (UTC, or TAI or whatever your clock follows)
1751 }
1752
1753 ev_periodic hourly_tick;
1754 ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
1755 ev_periodic_start (loop, &hourly_tick);
1756
1757 Example: The same as above, but use a reschedule callback to do it:
1758
1759 #include <math.h>
1760
1761 static ev_tstamp
1762 my_scheduler_cb (ev_periodic *w, ev_tstamp now)
1763 {
1764 return now + (3600. - fmod (now, 3600.));
1765 }
1766
1767 ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
1768
1769 Example: Call a callback every hour, starting now:
1770
1771 ev_periodic hourly_tick;
1772 ev_periodic_init (&hourly_tick, clock_cb,
1773 fmod (ev_now (loop), 3600.), 3600., 0);
1774 ev_periodic_start (loop, &hourly_tick);
1775
1776
1777 =head2 C<ev_signal> - signal me when a signal gets signalled!
1778
1779 Signal watchers will trigger an event when the process receives a specific
1780 signal one or more times. Even though signals are very asynchronous, libev
1781 will try it's best to deliver signals synchronously, i.e. as part of the
1782 normal event processing, like any other event.
1783
1784 If you want signals asynchronously, just use C<sigaction> as you would
1785 do without libev and forget about sharing the signal. You can even use
1786 C<ev_async> from a signal handler to synchronously wake up an event loop.
1787
1788 You can configure as many watchers as you like per signal. Only when the
1789 first watcher gets started will libev actually register a signal handler
1790 with the kernel (thus it coexists with your own signal handlers as long as
1791 you don't register any with libev for the same signal). Similarly, when
1792 the last signal watcher for a signal is stopped, libev will reset the
1793 signal handler to SIG_DFL (regardless of what it was set to before).
1794
1795 If possible and supported, libev will install its handlers with
1796 C<SA_RESTART> behaviour enabled, so system calls should not be unduly
1797 interrupted. If you have a problem with system calls getting interrupted by
1798 signals you can block all signals in an C<ev_check> watcher and unblock
1799 them in an C<ev_prepare> watcher.
1800
1801 =head3 Watcher-Specific Functions and Data Members
1802
1803 =over 4
1804
1805 =item ev_signal_init (ev_signal *, callback, int signum)
1806
1807 =item ev_signal_set (ev_signal *, int signum)
1808
1809 Configures the watcher to trigger on the given signal number (usually one
1810 of the C<SIGxxx> constants).
1811
1812 =item int signum [read-only]
1813
1814 The signal the watcher watches out for.
1815
1816 =back
1817
1818 =head3 Examples
1819
1820 Example: Try to exit cleanly on SIGINT.
1821
1822 static void
1823 sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
1824 {
1825 ev_unloop (loop, EVUNLOOP_ALL);
1826 }
1827
1828 ev_signal signal_watcher;
1829 ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
1830 ev_signal_start (loop, &signal_watcher);
1831
1832
1833 =head2 C<ev_child> - watch out for process status changes
1834
1835 Child watchers trigger when your process receives a SIGCHLD in response to
1836 some child status changes (most typically when a child of yours dies or
1837 exits). It is permissible to install a child watcher I<after> the child
1838 has been forked (which implies it might have already exited), as long
1839 as the event loop isn't entered (or is continued from a watcher), i.e.,
1840 forking and then immediately registering a watcher for the child is fine,
1841 but forking and registering a watcher a few event loop iterations later is
1842 not.
1843
1844 Only the default event loop is capable of handling signals, and therefore
1845 you can only register child watchers in the default event loop.
1846
1847 =head3 Process Interaction
1848
1849 Libev grabs C<SIGCHLD> as soon as the default event loop is
1850 initialised. This is necessary to guarantee proper behaviour even if
1851 the first child watcher is started after the child exits. The occurrence
1852 of C<SIGCHLD> is recorded asynchronously, but child reaping is done
1853 synchronously as part of the event loop processing. Libev always reaps all
1854 children, even ones not watched.
1855
1856 =head3 Overriding the Built-In Processing
1857
1858 Libev offers no special support for overriding the built-in child
1859 processing, but if your application collides with libev's default child
1860 handler, you can override it easily by installing your own handler for
1861 C<SIGCHLD> after initialising the default loop, and making sure the
1862 default loop never gets destroyed. You are encouraged, however, to use an
1863 event-based approach to child reaping and thus use libev's support for
1864 that, so other libev users can use C<ev_child> watchers freely.
1865
1866 =head3 Stopping the Child Watcher
1867
1868 Currently, the child watcher never gets stopped, even when the
1869 child terminates, so normally one needs to stop the watcher in the
1870 callback. Future versions of libev might stop the watcher automatically
1871 when a child exit is detected.
1872
1873 =head3 Watcher-Specific Functions and Data Members
1874
1875 =over 4
1876
1877 =item ev_child_init (ev_child *, callback, int pid, int trace)
1878
1879 =item ev_child_set (ev_child *, int pid, int trace)
1880
1881 Configures the watcher to wait for status changes of process C<pid> (or
1882 I<any> process if C<pid> is specified as C<0>). The callback can look
1883 at the C<rstatus> member of the C<ev_child> watcher structure to see
1884 the status word (use the macros from C<sys/wait.h> and see your systems
1885 C<waitpid> documentation). The C<rpid> member contains the pid of the
1886 process causing the status change. C<trace> must be either C<0> (only
1887 activate the watcher when the process terminates) or C<1> (additionally
1888 activate the watcher when the process is stopped or continued).
1889
1890 =item int pid [read-only]
1891
1892 The process id this watcher watches out for, or C<0>, meaning any process id.
1893
1894 =item int rpid [read-write]
1895
1896 The process id that detected a status change.
1897
1898 =item int rstatus [read-write]
1899
1900 The process exit/trace status caused by C<rpid> (see your systems
1901 C<waitpid> and C<sys/wait.h> documentation for details).
1902
1903 =back
1904
1905 =head3 Examples
1906
1907 Example: C<fork()> a new process and install a child handler to wait for
1908 its completion.
1909
1910 ev_child cw;
1911
1912 static void
1913 child_cb (EV_P_ ev_child *w, int revents)
1914 {
1915 ev_child_stop (EV_A_ w);
1916 printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
1917 }
1918
1919 pid_t pid = fork ();
1920
1921 if (pid < 0)
1922 // error
1923 else if (pid == 0)
1924 {
1925 // the forked child executes here
1926 exit (1);
1927 }
1928 else
1929 {
1930 ev_child_init (&cw, child_cb, pid, 0);
1931 ev_child_start (EV_DEFAULT_ &cw);
1932 }
1933
1934
1935 =head2 C<ev_stat> - did the file attributes just change?
1936
1937 This watches a file system path for attribute changes. That is, it calls
1938 C<stat> on that path in regular intervals (or when the OS says it changed)
1939 and sees if it changed compared to the last time, invoking the callback if
1940 it did.
1941
1942 The path does not need to exist: changing from "path exists" to "path does
1943 not exist" is a status change like any other. The condition "path does not
1944 exist" (or more correctly "path cannot be stat'ed") is signified by the
1945 C<st_nlink> field being zero (which is otherwise always forced to be at
1946 least one) and all the other fields of the stat buffer having unspecified
1947 contents.
1948
1949 The path I<must not> end in a slash or contain special components such as
1950 C<.> or C<..>. The path I<should> be absolute: If it is relative and
1951 your working directory changes, then the behaviour is undefined.
1952
1953 Since there is no portable change notification interface available, the
1954 portable implementation simply calls C<stat(2)> regularly on the path
1955 to see if it changed somehow. You can specify a recommended polling
1956 interval for this case. If you specify a polling interval of C<0> (highly
1957 recommended!) then a I<suitable, unspecified default> value will be used
1958 (which you can expect to be around five seconds, although this might
1959 change dynamically). Libev will also impose a minimum interval which is
1960 currently around C<0.1>, but that's usually overkill.
1961
1962 This watcher type is not meant for massive numbers of stat watchers,
1963 as even with OS-supported change notifications, this can be
1964 resource-intensive.
1965
1966 At the time of this writing, the only OS-specific interface implemented
1967 is the Linux inotify interface (implementing kqueue support is left as an
1968 exercise for the reader. Note, however, that the author sees no way of
1969 implementing C<ev_stat> semantics with kqueue, except as a hint).
1970
1971 =head3 ABI Issues (Largefile Support)
1972
1973 Libev by default (unless the user overrides this) uses the default
1974 compilation environment, which means that on systems with large file
1975 support disabled by default, you get the 32 bit version of the stat
1976 structure. When using the library from programs that change the ABI to
1977 use 64 bit file offsets the programs will fail. In that case you have to
1978 compile libev with the same flags to get binary compatibility. This is
1979 obviously the case with any flags that change the ABI, but the problem is
1980 most noticeably displayed with ev_stat and large file support.
1981
1982 The solution for this is to lobby your distribution maker to make large
1983 file interfaces available by default (as e.g. FreeBSD does) and not
1984 optional. Libev cannot simply switch on large file support because it has
1985 to exchange stat structures with application programs compiled using the
1986 default compilation environment.
1987
1988 =head3 Inotify and Kqueue
1989
1990 When C<inotify (7)> support has been compiled into libev and present at
1991 runtime, it will be used to speed up change detection where possible. The
1992 inotify descriptor will be created lazily when the first C<ev_stat>
1993 watcher is being started.
1994
1995 Inotify presence does not change the semantics of C<ev_stat> watchers
1996 except that changes might be detected earlier, and in some cases, to avoid
1997 making regular C<stat> calls. Even in the presence of inotify support
1998 there are many cases where libev has to resort to regular C<stat> polling,
1999 but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
2000 many bugs), the path exists (i.e. stat succeeds), and the path resides on
2001 a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
2002 xfs are fully working) libev usually gets away without polling.
2003
2004 There is no support for kqueue, as apparently it cannot be used to
2005 implement this functionality, due to the requirement of having a file
2006 descriptor open on the object at all times, and detecting renames, unlinks
2007 etc. is difficult.
2008
2009 =head3 C<stat ()> is a synchronous operation
2010
2011 Libev doesn't normally do any kind of I/O itself, and so is not blocking
2012 the process. The exception are C<ev_stat> watchers - those call C<stat
2013 ()>, which is a synchronous operation.
2014
2015 For local paths, this usually doesn't matter: unless the system is very
2016 busy or the intervals between stat's are large, a stat call will be fast,
2017 as the path data is usually in memory already (except when starting the
2018 watcher).
2019
2020 For networked file systems, calling C<stat ()> can block an indefinite
2021 time due to network issues, and even under good conditions, a stat call
2022 often takes multiple milliseconds.
2023
2024 Therefore, it is best to avoid using C<ev_stat> watchers on networked
2025 paths, although this is fully supported by libev.
2026
2027 =head3 The special problem of stat time resolution
2028
2029 The C<stat ()> system call only supports full-second resolution portably,
2030 and even on systems where the resolution is higher, most file systems
2031 still only support whole seconds.
2032
2033 That means that, if the time is the only thing that changes, you can
2034 easily miss updates: on the first update, C<ev_stat> detects a change and
2035 calls your callback, which does something. When there is another update
2036 within the same second, C<ev_stat> will be unable to detect unless the
2037 stat data does change in other ways (e.g. file size).
2038
2039 The solution to this is to delay acting on a change for slightly more
2040 than a second (or till slightly after the next full second boundary), using
2041 a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
2042 ev_timer_again (loop, w)>).
2043
2044 The C<.02> offset is added to work around small timing inconsistencies
2045 of some operating systems (where the second counter of the current time
2046 might be be delayed. One such system is the Linux kernel, where a call to
2047 C<gettimeofday> might return a timestamp with a full second later than
2048 a subsequent C<time> call - if the equivalent of C<time ()> is used to
2049 update file times then there will be a small window where the kernel uses
2050 the previous second to update file times but libev might already execute
2051 the timer callback).
2052
2053 =head3 Watcher-Specific Functions and Data Members
2054
2055 =over 4
2056
2057 =item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
2058
2059 =item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
2060
2061 Configures the watcher to wait for status changes of the given
2062 C<path>. The C<interval> is a hint on how quickly a change is expected to
2063 be detected and should normally be specified as C<0> to let libev choose
2064 a suitable value. The memory pointed to by C<path> must point to the same
2065 path for as long as the watcher is active.
2066
2067 The callback will receive an C<EV_STAT> event when a change was detected,
2068 relative to the attributes at the time the watcher was started (or the
2069 last change was detected).
2070
2071 =item ev_stat_stat (loop, ev_stat *)
2072
2073 Updates the stat buffer immediately with new values. If you change the
2074 watched path in your callback, you could call this function to avoid
2075 detecting this change (while introducing a race condition if you are not
2076 the only one changing the path). Can also be useful simply to find out the
2077 new values.
2078
2079 =item ev_statdata attr [read-only]
2080
2081 The most-recently detected attributes of the file. Although the type is
2082 C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
2083 suitable for your system, but you can only rely on the POSIX-standardised
2084 members to be present. If the C<st_nlink> member is C<0>, then there was
2085 some error while C<stat>ing the file.
2086
2087 =item ev_statdata prev [read-only]
2088
2089 The previous attributes of the file. The callback gets invoked whenever
2090 C<prev> != C<attr>, or, more precisely, one or more of these members
2091 differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
2092 C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
2093
2094 =item ev_tstamp interval [read-only]
2095
2096 The specified interval.
2097
2098 =item const char *path [read-only]
2099
2100 The file system path that is being watched.
2101
2102 =back
2103
2104 =head3 Examples
2105
2106 Example: Watch C</etc/passwd> for attribute changes.
2107
2108 static void
2109 passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
2110 {
2111 /* /etc/passwd changed in some way */
2112 if (w->attr.st_nlink)
2113 {
2114 printf ("passwd current size %ld\n", (long)w->attr.st_size);
2115 printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
2116 printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
2117 }
2118 else
2119 /* you shalt not abuse printf for puts */
2120 puts ("wow, /etc/passwd is not there, expect problems. "
2121 "if this is windows, they already arrived\n");
2122 }
2123
2124 ...
2125 ev_stat passwd;
2126
2127 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
2128 ev_stat_start (loop, &passwd);
2129
2130 Example: Like above, but additionally use a one-second delay so we do not
2131 miss updates (however, frequent updates will delay processing, too, so
2132 one might do the work both on C<ev_stat> callback invocation I<and> on
2133 C<ev_timer> callback invocation).
2134
2135 static ev_stat passwd;
2136 static ev_timer timer;
2137
2138 static void
2139 timer_cb (EV_P_ ev_timer *w, int revents)
2140 {
2141 ev_timer_stop (EV_A_ w);
2142
2143 /* now it's one second after the most recent passwd change */
2144 }
2145
2146 static void
2147 stat_cb (EV_P_ ev_stat *w, int revents)
2148 {
2149 /* reset the one-second timer */
2150 ev_timer_again (EV_A_ &timer);
2151 }
2152
2153 ...
2154 ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
2155 ev_stat_start (loop, &passwd);
2156 ev_timer_init (&timer, timer_cb, 0., 1.02);
2157
2158
2159 =head2 C<ev_idle> - when you've got nothing better to do...
2160
2161 Idle watchers trigger events when no other events of the same or higher
2162 priority are pending (prepare, check and other idle watchers do not count
2163 as receiving "events").
2164
2165 That is, as long as your process is busy handling sockets or timeouts
2166 (or even signals, imagine) of the same or higher priority it will not be
2167 triggered. But when your process is idle (or only lower-priority watchers
2168 are pending), the idle watchers are being called once per event loop
2169 iteration - until stopped, that is, or your process receives more events
2170 and becomes busy again with higher priority stuff.
2171
2172 The most noteworthy effect is that as long as any idle watchers are
2173 active, the process will not block when waiting for new events.
2174
2175 Apart from keeping your process non-blocking (which is a useful
2176 effect on its own sometimes), idle watchers are a good place to do
2177 "pseudo-background processing", or delay processing stuff to after the
2178 event loop has handled all outstanding events.
2179
2180 =head3 Watcher-Specific Functions and Data Members
2181
2182 =over 4
2183
2184 =item ev_idle_init (ev_signal *, callback)
2185
2186 Initialises and configures the idle watcher - it has no parameters of any
2187 kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
2188 believe me.
2189
2190 =back
2191
2192 =head3 Examples
2193
2194 Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
2195 callback, free it. Also, use no error checking, as usual.
2196
2197 static void
2198 idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
2199 {
2200 free (w);
2201 // now do something you wanted to do when the program has
2202 // no longer anything immediate to do.
2203 }
2204
2205 ev_idle *idle_watcher = malloc (sizeof (ev_idle));
2206 ev_idle_init (idle_watcher, idle_cb);
2207 ev_idle_start (loop, idle_cb);
2208
2209
2210 =head2 C<ev_prepare> and C<ev_check> - customise your event loop!
2211
2212 Prepare and check watchers are usually (but not always) used in pairs:
2213 prepare watchers get invoked before the process blocks and check watchers
2214 afterwards.
2215
2216 You I<must not> call C<ev_loop> or similar functions that enter
2217 the current event loop from either C<ev_prepare> or C<ev_check>
2218 watchers. Other loops than the current one are fine, however. The
2219 rationale behind this is that you do not need to check for recursion in
2220 those watchers, i.e. the sequence will always be C<ev_prepare>, blocking,
2221 C<ev_check> so if you have one watcher of each kind they will always be
2222 called in pairs bracketing the blocking call.
2223
2224 Their main purpose is to integrate other event mechanisms into libev and
2225 their use is somewhat advanced. They could be used, for example, to track
2226 variable changes, implement your own watchers, integrate net-snmp or a
2227 coroutine library and lots more. They are also occasionally useful if
2228 you cache some data and want to flush it before blocking (for example,
2229 in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
2230 watcher).
2231
2232 This is done by examining in each prepare call which file descriptors
2233 need to be watched by the other library, registering C<ev_io> watchers
2234 for them and starting an C<ev_timer> watcher for any timeouts (many
2235 libraries provide exactly this functionality). Then, in the check watcher,
2236 you check for any events that occurred (by checking the pending status
2237 of all watchers and stopping them) and call back into the library. The
2238 I/O and timer callbacks will never actually be called (but must be valid
2239 nevertheless, because you never know, you know?).
2240
2241 As another example, the Perl Coro module uses these hooks to integrate
2242 coroutines into libev programs, by yielding to other active coroutines
2243 during each prepare and only letting the process block if no coroutines
2244 are ready to run (it's actually more complicated: it only runs coroutines
2245 with priority higher than or equal to the event loop and one coroutine
2246 of lower priority, but only once, using idle watchers to keep the event
2247 loop from blocking if lower-priority coroutines are active, thus mapping
2248 low-priority coroutines to idle/background tasks).
2249
2250 It is recommended to give C<ev_check> watchers highest (C<EV_MAXPRI>)
2251 priority, to ensure that they are being run before any other watchers
2252 after the poll (this doesn't matter for C<ev_prepare> watchers).
2253
2254 Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
2255 activate ("feed") events into libev. While libev fully supports this, they
2256 might get executed before other C<ev_check> watchers did their job. As
2257 C<ev_check> watchers are often used to embed other (non-libev) event
2258 loops those other event loops might be in an unusable state until their
2259 C<ev_check> watcher ran (always remind yourself to coexist peacefully with
2260 others).
2261
2262 =head3 Watcher-Specific Functions and Data Members
2263
2264 =over 4
2265
2266 =item ev_prepare_init (ev_prepare *, callback)
2267
2268 =item ev_check_init (ev_check *, callback)
2269
2270 Initialises and configures the prepare or check watcher - they have no
2271 parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
2272 macros, but using them is utterly, utterly, utterly and completely
2273 pointless.
2274
2275 =back
2276
2277 =head3 Examples
2278
2279 There are a number of principal ways to embed other event loops or modules
2280 into libev. Here are some ideas on how to include libadns into libev
2281 (there is a Perl module named C<EV::ADNS> that does this, which you could
2282 use as a working example. Another Perl module named C<EV::Glib> embeds a
2283 Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
2284 Glib event loop).
2285
2286 Method 1: Add IO watchers and a timeout watcher in a prepare handler,
2287 and in a check watcher, destroy them and call into libadns. What follows
2288 is pseudo-code only of course. This requires you to either use a low
2289 priority for the check watcher or use C<ev_clear_pending> explicitly, as
2290 the callbacks for the IO/timeout watchers might not have been called yet.
2291
2292 static ev_io iow [nfd];
2293 static ev_timer tw;
2294
2295 static void
2296 io_cb (struct ev_loop *loop, ev_io *w, int revents)
2297 {
2298 }
2299
2300 // create io watchers for each fd and a timer before blocking
2301 static void
2302 adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
2303 {
2304 int timeout = 3600000;
2305 struct pollfd fds [nfd];
2306 // actual code will need to loop here and realloc etc.
2307 adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
2308
2309 /* the callback is illegal, but won't be called as we stop during check */
2310 ev_timer_init (&tw, 0, timeout * 1e-3);
2311 ev_timer_start (loop, &tw);
2312
2313 // create one ev_io per pollfd
2314 for (int i = 0; i < nfd; ++i)
2315 {
2316 ev_io_init (iow + i, io_cb, fds [i].fd,
2317 ((fds [i].events & POLLIN ? EV_READ : 0)
2318 | (fds [i].events & POLLOUT ? EV_WRITE : 0)));
2319
2320 fds [i].revents = 0;
2321 ev_io_start (loop, iow + i);
2322 }
2323 }
2324
2325 // stop all watchers after blocking
2326 static void
2327 adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
2328 {
2329 ev_timer_stop (loop, &tw);
2330
2331 for (int i = 0; i < nfd; ++i)
2332 {
2333 // set the relevant poll flags
2334 // could also call adns_processreadable etc. here
2335 struct pollfd *fd = fds + i;
2336 int revents = ev_clear_pending (iow + i);
2337 if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
2338 if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
2339
2340 // now stop the watcher
2341 ev_io_stop (loop, iow + i);
2342 }
2343
2344 adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
2345 }
2346
2347 Method 2: This would be just like method 1, but you run C<adns_afterpoll>
2348 in the prepare watcher and would dispose of the check watcher.
2349
2350 Method 3: If the module to be embedded supports explicit event
2351 notification (libadns does), you can also make use of the actual watcher
2352 callbacks, and only destroy/create the watchers in the prepare watcher.
2353
2354 static void
2355 timer_cb (EV_P_ ev_timer *w, int revents)
2356 {
2357 adns_state ads = (adns_state)w->data;
2358 update_now (EV_A);
2359
2360 adns_processtimeouts (ads, &tv_now);
2361 }
2362
2363 static void
2364 io_cb (EV_P_ ev_io *w, int revents)
2365 {
2366 adns_state ads = (adns_state)w->data;
2367 update_now (EV_A);
2368
2369 if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
2370 if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
2371 }
2372
2373 // do not ever call adns_afterpoll
2374
2375 Method 4: Do not use a prepare or check watcher because the module you
2376 want to embed is not flexible enough to support it. Instead, you can
2377 override their poll function. The drawback with this solution is that the
2378 main loop is now no longer controllable by EV. The C<Glib::EV> module uses
2379 this approach, effectively embedding EV as a client into the horrible
2380 libglib event loop.
2381
2382 static gint
2383 event_poll_func (GPollFD *fds, guint nfds, gint timeout)
2384 {
2385 int got_events = 0;
2386
2387 for (n = 0; n < nfds; ++n)
2388 // create/start io watcher that sets the relevant bits in fds[n] and increment got_events
2389
2390 if (timeout >= 0)
2391 // create/start timer
2392
2393 // poll
2394 ev_loop (EV_A_ 0);
2395
2396 // stop timer again
2397 if (timeout >= 0)
2398 ev_timer_stop (EV_A_ &to);
2399
2400 // stop io watchers again - their callbacks should have set
2401 for (n = 0; n < nfds; ++n)
2402 ev_io_stop (EV_A_ iow [n]);
2403
2404 return got_events;
2405 }
2406
2407
2408 =head2 C<ev_embed> - when one backend isn't enough...
2409
2410 This is a rather advanced watcher type that lets you embed one event loop
2411 into another (currently only C<ev_io> events are supported in the embedded
2412 loop, other types of watchers might be handled in a delayed or incorrect
2413 fashion and must not be used).
2414
2415 There are primarily two reasons you would want that: work around bugs and
2416 prioritise I/O.
2417
2418 As an example for a bug workaround, the kqueue backend might only support
2419 sockets on some platform, so it is unusable as generic backend, but you
2420 still want to make use of it because you have many sockets and it scales
2421 so nicely. In this case, you would create a kqueue-based loop and embed
2422 it into your default loop (which might use e.g. poll). Overall operation
2423 will be a bit slower because first libev has to call C<poll> and then
2424 C<kevent>, but at least you can use both mechanisms for what they are
2425 best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
2426
2427 As for prioritising I/O: under rare circumstances you have the case where
2428 some fds have to be watched and handled very quickly (with low latency),
2429 and even priorities and idle watchers might have too much overhead. In
2430 this case you would put all the high priority stuff in one loop and all
2431 the rest in a second one, and embed the second one in the first.
2432
2433 As long as the watcher is active, the callback will be invoked every
2434 time there might be events pending in the embedded loop. The callback
2435 must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
2436 sweep and invoke their callbacks (the callback doesn't need to invoke the
2437 C<ev_embed_sweep> function directly, it could also start an idle watcher
2438 to give the embedded loop strictly lower priority for example).
2439
2440 You can also set the callback to C<0>, in which case the embed watcher
2441 will automatically execute the embedded loop sweep whenever necessary.
2442
2443 Fork detection will be handled transparently while the C<ev_embed> watcher
2444 is active, i.e., the embedded loop will automatically be forked when the
2445 embedding loop forks. In other cases, the user is responsible for calling
2446 C<ev_loop_fork> on the embedded loop.
2447
2448 Unfortunately, not all backends are embeddable: only the ones returned by
2449 C<ev_embeddable_backends> are, which, unfortunately, does not include any
2450 portable one.
2451
2452 So when you want to use this feature you will always have to be prepared
2453 that you cannot get an embeddable loop. The recommended way to get around
2454 this is to have a separate variables for your embeddable loop, try to
2455 create it, and if that fails, use the normal loop for everything.
2456
2457 =head3 C<ev_embed> and fork
2458
2459 While the C<ev_embed> watcher is running, forks in the embedding loop will
2460 automatically be applied to the embedded loop as well, so no special
2461 fork handling is required in that case. When the watcher is not running,
2462 however, it is still the task of the libev user to call C<ev_loop_fork ()>
2463 as applicable.
2464
2465 =head3 Watcher-Specific Functions and Data Members
2466
2467 =over 4
2468
2469 =item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
2470
2471 =item ev_embed_set (ev_embed *, callback, struct ev_loop *embedded_loop)
2472
2473 Configures the watcher to embed the given loop, which must be
2474 embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
2475 invoked automatically, otherwise it is the responsibility of the callback
2476 to invoke it (it will continue to be called until the sweep has been done,
2477 if you do not want that, you need to temporarily stop the embed watcher).
2478
2479 =item ev_embed_sweep (loop, ev_embed *)
2480
2481 Make a single, non-blocking sweep over the embedded loop. This works
2482 similarly to C<ev_loop (embedded_loop, EVLOOP_NONBLOCK)>, but in the most
2483 appropriate way for embedded loops.
2484
2485 =item struct ev_loop *other [read-only]
2486
2487 The embedded event loop.
2488
2489 =back
2490
2491 =head3 Examples
2492
2493 Example: Try to get an embeddable event loop and embed it into the default
2494 event loop. If that is not possible, use the default loop. The default
2495 loop is stored in C<loop_hi>, while the embeddable loop is stored in
2496 C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
2497 used).
2498
2499 struct ev_loop *loop_hi = ev_default_init (0);
2500 struct ev_loop *loop_lo = 0;
2501 ev_embed embed;
2502
2503 // see if there is a chance of getting one that works
2504 // (remember that a flags value of 0 means autodetection)
2505 loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
2506 ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
2507 : 0;
2508
2509 // if we got one, then embed it, otherwise default to loop_hi
2510 if (loop_lo)
2511 {
2512 ev_embed_init (&embed, 0, loop_lo);
2513 ev_embed_start (loop_hi, &embed);
2514 }
2515 else
2516 loop_lo = loop_hi;
2517
2518 Example: Check if kqueue is available but not recommended and create
2519 a kqueue backend for use with sockets (which usually work with any
2520 kqueue implementation). Store the kqueue/socket-only event loop in
2521 C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
2522
2523 struct ev_loop *loop = ev_default_init (0);
2524 struct ev_loop *loop_socket = 0;
2525 ev_embed embed;
2526
2527 if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
2528 if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
2529 {
2530 ev_embed_init (&embed, 0, loop_socket);
2531 ev_embed_start (loop, &embed);
2532 }
2533
2534 if (!loop_socket)
2535 loop_socket = loop;
2536
2537 // now use loop_socket for all sockets, and loop for everything else
2538
2539
2540 =head2 C<ev_fork> - the audacity to resume the event loop after a fork
2541
2542 Fork watchers are called when a C<fork ()> was detected (usually because
2543 whoever is a good citizen cared to tell libev about it by calling
2544 C<ev_default_fork> or C<ev_loop_fork>). The invocation is done before the
2545 event loop blocks next and before C<ev_check> watchers are being called,
2546 and only in the child after the fork. If whoever good citizen calling
2547 C<ev_default_fork> cheats and calls it in the wrong process, the fork
2548 handlers will be invoked, too, of course.
2549
2550 =head3 Watcher-Specific Functions and Data Members
2551
2552 =over 4
2553
2554 =item ev_fork_init (ev_signal *, callback)
2555
2556 Initialises and configures the fork watcher - it has no parameters of any
2557 kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
2558 believe me.
2559
2560 =back
2561
2562
2563 =head2 C<ev_async> - how to wake up another event loop
2564
2565 In general, you cannot use an C<ev_loop> from multiple threads or other
2566 asynchronous sources such as signal handlers (as opposed to multiple event
2567 loops - those are of course safe to use in different threads).
2568
2569 Sometimes, however, you need to wake up another event loop you do not
2570 control, for example because it belongs to another thread. This is what
2571 C<ev_async> watchers do: as long as the C<ev_async> watcher is active, you
2572 can signal it by calling C<ev_async_send>, which is thread- and signal
2573 safe.
2574
2575 This functionality is very similar to C<ev_signal> watchers, as signals,
2576 too, are asynchronous in nature, and signals, too, will be compressed
2577 (i.e. the number of callback invocations may be less than the number of
2578 C<ev_async_sent> calls).
2579
2580 Unlike C<ev_signal> watchers, C<ev_async> works with any event loop, not
2581 just the default loop.
2582
2583 =head3 Queueing
2584
2585 C<ev_async> does not support queueing of data in any way. The reason
2586 is that the author does not know of a simple (or any) algorithm for a
2587 multiple-writer-single-reader queue that works in all cases and doesn't
2588 need elaborate support such as pthreads.
2589
2590 That means that if you want to queue data, you have to provide your own
2591 queue. But at least I can tell you how to implement locking around your
2592 queue:
2593
2594 =over 4
2595
2596 =item queueing from a signal handler context
2597
2598 To implement race-free queueing, you simply add to the queue in the signal
2599 handler but you block the signal handler in the watcher callback. Here is
2600 an example that does that for some fictitious SIGUSR1 handler:
2601
2602 static ev_async mysig;
2603
2604 static void
2605 sigusr1_handler (void)
2606 {
2607 sometype data;
2608
2609 // no locking etc.
2610 queue_put (data);
2611 ev_async_send (EV_DEFAULT_ &mysig);
2612 }
2613
2614 static void
2615 mysig_cb (EV_P_ ev_async *w, int revents)
2616 {
2617 sometype data;
2618 sigset_t block, prev;
2619
2620 sigemptyset (&block);
2621 sigaddset (&block, SIGUSR1);
2622 sigprocmask (SIG_BLOCK, &block, &prev);
2623
2624 while (queue_get (&data))
2625 process (data);
2626
2627 if (sigismember (&prev, SIGUSR1)
2628 sigprocmask (SIG_UNBLOCK, &block, 0);
2629 }
2630
2631 (Note: pthreads in theory requires you to use C<pthread_setmask>
2632 instead of C<sigprocmask> when you use threads, but libev doesn't do it
2633 either...).
2634
2635 =item queueing from a thread context
2636
2637 The strategy for threads is different, as you cannot (easily) block
2638 threads but you can easily preempt them, so to queue safely you need to
2639 employ a traditional mutex lock, such as in this pthread example:
2640
2641 static ev_async mysig;
2642 static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
2643
2644 static void
2645 otherthread (void)
2646 {
2647 // only need to lock the actual queueing operation
2648 pthread_mutex_lock (&mymutex);
2649 queue_put (data);
2650 pthread_mutex_unlock (&mymutex);
2651
2652 ev_async_send (EV_DEFAULT_ &mysig);
2653 }
2654
2655 static void
2656 mysig_cb (EV_P_ ev_async *w, int revents)
2657 {
2658 pthread_mutex_lock (&mymutex);
2659
2660 while (queue_get (&data))
2661 process (data);
2662
2663 pthread_mutex_unlock (&mymutex);
2664 }
2665
2666 =back
2667
2668
2669 =head3 Watcher-Specific Functions and Data Members
2670
2671 =over 4
2672
2673 =item ev_async_init (ev_async *, callback)
2674
2675 Initialises and configures the async watcher - it has no parameters of any
2676 kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
2677 trust me.
2678
2679 =item ev_async_send (loop, ev_async *)
2680
2681 Sends/signals/activates the given C<ev_async> watcher, that is, feeds
2682 an C<EV_ASYNC> event on the watcher into the event loop. Unlike
2683 C<ev_feed_event>, this call is safe to do from other threads, signal or
2684 similar contexts (see the discussion of C<EV_ATOMIC_T> in the embedding
2685 section below on what exactly this means).
2686
2687 This call incurs the overhead of a system call only once per loop iteration,
2688 so while the overhead might be noticeable, it doesn't apply to repeated
2689 calls to C<ev_async_send>.
2690
2691 =item bool = ev_async_pending (ev_async *)
2692
2693 Returns a non-zero value when C<ev_async_send> has been called on the
2694 watcher but the event has not yet been processed (or even noted) by the
2695 event loop.
2696
2697 C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
2698 the loop iterates next and checks for the watcher to have become active,
2699 it will reset the flag again. C<ev_async_pending> can be used to very
2700 quickly check whether invoking the loop might be a good idea.
2701
2702 Not that this does I<not> check whether the watcher itself is pending, only
2703 whether it has been requested to make this watcher pending.
2704
2705 =back
2706
2707
2708 =head1 OTHER FUNCTIONS
2709
2710 There are some other functions of possible interest. Described. Here. Now.
2711
2712 =over 4
2713
2714 =item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)
2715
2716 This function combines a simple timer and an I/O watcher, calls your
2717 callback on whichever event happens first and automatically stops both
2718 watchers. This is useful if you want to wait for a single event on an fd
2719 or timeout without having to allocate/configure/start/stop/free one or
2720 more watchers yourself.
2721
2722 If C<fd> is less than 0, then no I/O watcher will be started and the
2723 C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
2724 the given C<fd> and C<events> set will be created and started.
2725
2726 If C<timeout> is less than 0, then no timeout watcher will be
2727 started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
2728 repeat = 0) will be started. C<0> is a valid timeout.
2729
2730 The callback has the type C<void (*cb)(int revents, void *arg)> and gets
2731 passed an C<revents> set like normal event callbacks (a combination of
2732 C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMEOUT>) and the C<arg>
2733 value passed to C<ev_once>. Note that it is possible to receive I<both>
2734 a timeout and an io event at the same time - you probably should give io
2735 events precedence.
2736
2737 Example: wait up to ten seconds for data to appear on STDIN_FILENO.
2738
2739 static void stdin_ready (int revents, void *arg)
2740 {
2741 if (revents & EV_READ)
2742 /* stdin might have data for us, joy! */;
2743 else if (revents & EV_TIMEOUT)
2744 /* doh, nothing entered */;
2745 }
2746
2747 ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
2748
2749 =item ev_feed_event (struct ev_loop *, watcher *, int revents)
2750
2751 Feeds the given event set into the event loop, as if the specified event
2752 had happened for the specified watcher (which must be a pointer to an
2753 initialised but not necessarily started event watcher).
2754
2755 =item ev_feed_fd_event (struct ev_loop *, int fd, int revents)
2756
2757 Feed an event on the given fd, as if a file descriptor backend detected
2758 the given events it.
2759
2760 =item ev_feed_signal_event (struct ev_loop *loop, int signum)
2761
2762 Feed an event as if the given signal occurred (C<loop> must be the default
2763 loop!).
2764
2765 =back
2766
2767
2768 =head1 LIBEVENT EMULATION
2769
2770 Libev offers a compatibility emulation layer for libevent. It cannot
2771 emulate the internals of libevent, so here are some usage hints:
2772
2773 =over 4
2774
2775 =item * Use it by including <event.h>, as usual.
2776
2777 =item * The following members are fully supported: ev_base, ev_callback,
2778 ev_arg, ev_fd, ev_res, ev_events.
2779
2780 =item * Avoid using ev_flags and the EVLIST_*-macros, while it is
2781 maintained by libev, it does not work exactly the same way as in libevent (consider
2782 it a private API).
2783
2784 =item * Priorities are not currently supported. Initialising priorities
2785 will fail and all watchers will have the same priority, even though there
2786 is an ev_pri field.
2787
2788 =item * In libevent, the last base created gets the signals, in libev, the
2789 first base created (== the default loop) gets the signals.
2790
2791 =item * Other members are not supported.
2792
2793 =item * The libev emulation is I<not> ABI compatible to libevent, you need
2794 to use the libev header file and library.
2795
2796 =back
2797
2798 =head1 C++ SUPPORT
2799
2800 Libev comes with some simplistic wrapper classes for C++ that mainly allow
2801 you to use some convenience methods to start/stop watchers and also change
2802 the callback model to a model using method callbacks on objects.
2803
2804 To use it,
2805
2806 #include <ev++.h>
2807
2808 This automatically includes F<ev.h> and puts all of its definitions (many
2809 of them macros) into the global namespace. All C++ specific things are
2810 put into the C<ev> namespace. It should support all the same embedding
2811 options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
2812
2813 Care has been taken to keep the overhead low. The only data member the C++
2814 classes add (compared to plain C-style watchers) is the event loop pointer
2815 that the watcher is associated with (or no additional members at all if
2816 you disable C<EV_MULTIPLICITY> when embedding libev).
2817
2818 Currently, functions, and static and non-static member functions can be
2819 used as callbacks. Other types should be easy to add as long as they only
2820 need one additional pointer for context. If you need support for other
2821 types of functors please contact the author (preferably after implementing
2822 it).
2823
2824 Here is a list of things available in the C<ev> namespace:
2825
2826 =over 4
2827
2828 =item C<ev::READ>, C<ev::WRITE> etc.
2829
2830 These are just enum values with the same values as the C<EV_READ> etc.
2831 macros from F<ev.h>.
2832
2833 =item C<ev::tstamp>, C<ev::now>
2834
2835 Aliases to the same types/functions as with the C<ev_> prefix.
2836
2837 =item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
2838
2839 For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
2840 the same name in the C<ev> namespace, with the exception of C<ev_signal>
2841 which is called C<ev::sig> to avoid clashes with the C<signal> macro
2842 defines by many implementations.
2843
2844 All of those classes have these methods:
2845
2846 =over 4
2847
2848 =item ev::TYPE::TYPE ()
2849
2850 =item ev::TYPE::TYPE (struct ev_loop *)
2851
2852 =item ev::TYPE::~TYPE
2853
2854 The constructor (optionally) takes an event loop to associate the watcher
2855 with. If it is omitted, it will use C<EV_DEFAULT>.
2856
2857 The constructor calls C<ev_init> for you, which means you have to call the
2858 C<set> method before starting it.
2859
2860 It will not set a callback, however: You have to call the templated C<set>
2861 method to set a callback before you can start the watcher.
2862
2863 (The reason why you have to use a method is a limitation in C++ which does
2864 not allow explicit template arguments for constructors).
2865
2866 The destructor automatically stops the watcher if it is active.
2867
2868 =item w->set<class, &class::method> (object *)
2869
2870 This method sets the callback method to call. The method has to have a
2871 signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
2872 first argument and the C<revents> as second. The object must be given as
2873 parameter and is stored in the C<data> member of the watcher.
2874
2875 This method synthesizes efficient thunking code to call your method from
2876 the C callback that libev requires. If your compiler can inline your
2877 callback (i.e. it is visible to it at the place of the C<set> call and
2878 your compiler is good :), then the method will be fully inlined into the
2879 thunking function, making it as fast as a direct C callback.
2880
2881 Example: simple class declaration and watcher initialisation
2882
2883 struct myclass
2884 {
2885 void io_cb (ev::io &w, int revents) { }
2886 }
2887
2888 myclass obj;
2889 ev::io iow;
2890 iow.set <myclass, &myclass::io_cb> (&obj);
2891
2892 =item w->set (object *)
2893
2894 This is an B<experimental> feature that might go away in a future version.
2895
2896 This is a variation of a method callback - leaving out the method to call
2897 will default the method to C<operator ()>, which makes it possible to use
2898 functor objects without having to manually specify the C<operator ()> all
2899 the time. Incidentally, you can then also leave out the template argument
2900 list.
2901
2902 The C<operator ()> method prototype must be C<void operator ()(watcher &w,
2903 int revents)>.
2904
2905 See the method-C<set> above for more details.
2906
2907 Example: use a functor object as callback.
2908
2909 struct myfunctor
2910 {
2911 void operator() (ev::io &w, int revents)
2912 {
2913 ...
2914 }
2915 }
2916
2917 myfunctor f;
2918
2919 ev::io w;
2920 w.set (&f);
2921
2922 =item w->set<function> (void *data = 0)
2923
2924 Also sets a callback, but uses a static method or plain function as
2925 callback. The optional C<data> argument will be stored in the watcher's
2926 C<data> member and is free for you to use.
2927
2928 The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
2929
2930 See the method-C<set> above for more details.
2931
2932 Example: Use a plain function as callback.
2933
2934 static void io_cb (ev::io &w, int revents) { }
2935 iow.set <io_cb> ();
2936
2937 =item w->set (struct ev_loop *)
2938
2939 Associates a different C<struct ev_loop> with this watcher. You can only
2940 do this when the watcher is inactive (and not pending either).
2941
2942 =item w->set ([arguments])
2943
2944 Basically the same as C<ev_TYPE_set>, with the same arguments. Must be
2945 called at least once. Unlike the C counterpart, an active watcher gets
2946 automatically stopped and restarted when reconfiguring it with this
2947 method.
2948
2949 =item w->start ()
2950
2951 Starts the watcher. Note that there is no C<loop> argument, as the
2952 constructor already stores the event loop.
2953
2954 =item w->stop ()
2955
2956 Stops the watcher if it is active. Again, no C<loop> argument.
2957
2958 =item w->again () (C<ev::timer>, C<ev::periodic> only)
2959
2960 For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
2961 C<ev_TYPE_again> function.
2962
2963 =item w->sweep () (C<ev::embed> only)
2964
2965 Invokes C<ev_embed_sweep>.
2966
2967 =item w->update () (C<ev::stat> only)
2968
2969 Invokes C<ev_stat_stat>.
2970
2971 =back
2972
2973 =back
2974
2975 Example: Define a class with an IO and idle watcher, start one of them in
2976 the constructor.
2977
2978 class myclass
2979 {
2980 ev::io io ; void io_cb (ev::io &w, int revents);
2981 ev::idle idle; void idle_cb (ev::idle &w, int revents);
2982
2983 myclass (int fd)
2984 {
2985 io .set <myclass, &myclass::io_cb > (this);
2986 idle.set <myclass, &myclass::idle_cb> (this);
2987
2988 io.start (fd, ev::READ);
2989 }
2990 };
2991
2992
2993 =head1 OTHER LANGUAGE BINDINGS
2994
2995 Libev does not offer other language bindings itself, but bindings for a
2996 number of languages exist in the form of third-party packages. If you know
2997 any interesting language binding in addition to the ones listed here, drop
2998 me a note.
2999
3000 =over 4
3001
3002 =item Perl
3003
3004 The EV module implements the full libev API and is actually used to test
3005 libev. EV is developed together with libev. Apart from the EV core module,
3006 there are additional modules that implement libev-compatible interfaces
3007 to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
3008 C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
3009 and C<EV::Glib>).
3010
3011 It can be found and installed via CPAN, its homepage is at
3012 L<http://software.schmorp.de/pkg/EV>.
3013
3014 =item Python
3015
3016 Python bindings can be found at L<http://code.google.com/p/pyev/>. It
3017 seems to be quite complete and well-documented. Note, however, that the
3018 patch they require for libev is outright dangerous as it breaks the ABI
3019 for everybody else, and therefore, should never be applied in an installed
3020 libev (if python requires an incompatible ABI then it needs to embed
3021 libev).
3022
3023 =item Ruby
3024
3025 Tony Arcieri has written a ruby extension that offers access to a subset
3026 of the libev API and adds file handle abstractions, asynchronous DNS and
3027 more on top of it. It can be found via gem servers. Its homepage is at
3028 L<http://rev.rubyforge.org/>.
3029
3030 Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
3031 makes rev work even on mingw.
3032
3033 =item D
3034
3035 Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
3036 be found at L<http://proj.llucax.com.ar/wiki/evd>.
3037
3038 =item Ocaml
3039
3040 Erkki Seppala has written Ocaml bindings for libev, to be found at
3041 L<http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
3042
3043 =back
3044
3045
3046 =head1 MACRO MAGIC
3047
3048 Libev can be compiled with a variety of options, the most fundamental
3049 of which is C<EV_MULTIPLICITY>. This option determines whether (most)
3050 functions and callbacks have an initial C<struct ev_loop *> argument.
3051
3052 To make it easier to write programs that cope with either variant, the
3053 following macros are defined:
3054
3055 =over 4
3056
3057 =item C<EV_A>, C<EV_A_>
3058
3059 This provides the loop I<argument> for functions, if one is required ("ev
3060 loop argument"). The C<EV_A> form is used when this is the sole argument,
3061 C<EV_A_> is used when other arguments are following. Example:
3062
3063 ev_unref (EV_A);
3064 ev_timer_add (EV_A_ watcher);
3065 ev_loop (EV_A_ 0);
3066
3067 It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
3068 which is often provided by the following macro.
3069
3070 =item C<EV_P>, C<EV_P_>
3071
3072 This provides the loop I<parameter> for functions, if one is required ("ev
3073 loop parameter"). The C<EV_P> form is used when this is the sole parameter,
3074 C<EV_P_> is used when other parameters are following. Example:
3075
3076 // this is how ev_unref is being declared
3077 static void ev_unref (EV_P);
3078
3079 // this is how you can declare your typical callback
3080 static void cb (EV_P_ ev_timer *w, int revents)
3081
3082 It declares a parameter C<loop> of type C<struct ev_loop *>, quite
3083 suitable for use with C<EV_A>.
3084
3085 =item C<EV_DEFAULT>, C<EV_DEFAULT_>
3086
3087 Similar to the other two macros, this gives you the value of the default
3088 loop, if multiple loops are supported ("ev loop default").
3089
3090 =item C<EV_DEFAULT_UC>, C<EV_DEFAULT_UC_>
3091
3092 Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
3093 default loop has been initialised (C<UC> == unchecked). Their behaviour
3094 is undefined when the default loop has not been initialised by a previous
3095 execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
3096
3097 It is often prudent to use C<EV_DEFAULT> when initialising the first
3098 watcher in a function but use C<EV_DEFAULT_UC> afterwards.
3099
3100 =back
3101
3102 Example: Declare and initialise a check watcher, utilising the above
3103 macros so it will work regardless of whether multiple loops are supported
3104 or not.
3105
3106 static void
3107 check_cb (EV_P_ ev_timer *w, int revents)
3108 {
3109 ev_check_stop (EV_A_ w);
3110 }
3111
3112 ev_check check;
3113 ev_check_init (&check, check_cb);
3114 ev_check_start (EV_DEFAULT_ &check);
3115 ev_loop (EV_DEFAULT_ 0);
3116
3117 =head1 EMBEDDING
3118
3119 Libev can (and often is) directly embedded into host
3120 applications. Examples of applications that embed it include the Deliantra
3121 Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
3122 and rxvt-unicode.
3123
3124 The goal is to enable you to just copy the necessary files into your
3125 source directory without having to change even a single line in them, so
3126 you can easily upgrade by simply copying (or having a checked-out copy of
3127 libev somewhere in your source tree).
3128
3129 =head2 FILESETS
3130
3131 Depending on what features you need you need to include one or more sets of files
3132 in your application.
3133
3134 =head3 CORE EVENT LOOP
3135
3136 To include only the libev core (all the C<ev_*> functions), with manual
3137 configuration (no autoconf):
3138
3139 #define EV_STANDALONE 1
3140 #include "ev.c"
3141
3142 This will automatically include F<ev.h>, too, and should be done in a
3143 single C source file only to provide the function implementations. To use
3144 it, do the same for F<ev.h> in all files wishing to use this API (best
3145 done by writing a wrapper around F<ev.h> that you can include instead and
3146 where you can put other configuration options):
3147
3148 #define EV_STANDALONE 1
3149 #include "ev.h"
3150
3151 Both header files and implementation files can be compiled with a C++
3152 compiler (at least, that's a stated goal, and breakage will be treated
3153 as a bug).
3154
3155 You need the following files in your source tree, or in a directory
3156 in your include path (e.g. in libev/ when using -Ilibev):
3157
3158 ev.h
3159 ev.c
3160 ev_vars.h
3161 ev_wrap.h
3162
3163 ev_win32.c required on win32 platforms only
3164
3165 ev_select.c only when select backend is enabled (which is enabled by default)
3166 ev_poll.c only when poll backend is enabled (disabled by default)
3167 ev_epoll.c only when the epoll backend is enabled (disabled by default)
3168 ev_kqueue.c only when the kqueue backend is enabled (disabled by default)
3169 ev_port.c only when the solaris port backend is enabled (disabled by default)
3170
3171 F<ev.c> includes the backend files directly when enabled, so you only need
3172 to compile this single file.
3173
3174 =head3 LIBEVENT COMPATIBILITY API
3175
3176 To include the libevent compatibility API, also include:
3177
3178 #include "event.c"
3179
3180 in the file including F<ev.c>, and:
3181
3182 #include "event.h"
3183
3184 in the files that want to use the libevent API. This also includes F<ev.h>.
3185
3186 You need the following additional files for this:
3187
3188 event.h
3189 event.c
3190
3191 =head3 AUTOCONF SUPPORT
3192
3193 Instead of using C<EV_STANDALONE=1> and providing your configuration in
3194 whatever way you want, you can also C<m4_include([libev.m4])> in your
3195 F<configure.ac> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
3196 include F<config.h> and configure itself accordingly.
3197
3198 For this of course you need the m4 file:
3199
3200 libev.m4
3201
3202 =head2 PREPROCESSOR SYMBOLS/MACROS
3203
3204 Libev can be configured via a variety of preprocessor symbols you have to
3205 define before including any of its files. The default in the absence of
3206 autoconf is documented for every option.
3207
3208 =over 4
3209
3210 =item EV_STANDALONE
3211
3212 Must always be C<1> if you do not use autoconf configuration, which
3213 keeps libev from including F<config.h>, and it also defines dummy
3214 implementations for some libevent functions (such as logging, which is not
3215 supported). It will also not define any of the structs usually found in
3216 F<event.h> that are not directly supported by the libev core alone.
3217
3218 In stanbdalone mode, libev will still try to automatically deduce the
3219 configuration, but has to be more conservative.
3220
3221 =item EV_USE_MONOTONIC
3222
3223 If defined to be C<1>, libev will try to detect the availability of the
3224 monotonic clock option at both compile time and runtime. Otherwise no
3225 use of the monotonic clock option will be attempted. If you enable this,
3226 you usually have to link against librt or something similar. Enabling it
3227 when the functionality isn't available is safe, though, although you have
3228 to make sure you link against any libraries where the C<clock_gettime>
3229 function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
3230
3231 =item EV_USE_REALTIME
3232
3233 If defined to be C<1>, libev will try to detect the availability of the
3234 real-time clock option at compile time (and assume its availability at
3235 runtime if successful). Otherwise no use of the real-time clock option will
3236 be attempted. This effectively replaces C<gettimeofday> by C<clock_get
3237 (CLOCK_REALTIME, ...)> and will not normally affect correctness. See the
3238 note about libraries in the description of C<EV_USE_MONOTONIC>, though.
3239
3240 =item EV_USE_CLOCK_SYSCALL
3241
3242 If defined to be C<1>, libev will try to use a direct syscall instead
3243 of calling the system-provided C<clock_gettime> function. This option
3244 exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
3245 unconditionally pulls in C<libpthread>, slowing down single-threaded
3246 programs needlessly. Using a direct syscall is slightly slower (in
3247 theory), because no optimised vdso implementation can be used, but avoids
3248 the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
3249 higher, as it simplifies linking (no need for C<-lrt>).
3250
3251 =item EV_USE_NANOSLEEP
3252
3253 If defined to be C<1>, libev will assume that C<nanosleep ()> is available
3254 and will use it for delays. Otherwise it will use C<select ()>.
3255
3256 =item EV_USE_EVENTFD
3257
3258 If defined to be C<1>, then libev will assume that C<eventfd ()> is
3259 available and will probe for kernel support at runtime. This will improve
3260 C<ev_signal> and C<ev_async> performance and reduce resource consumption.
3261 If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
3262 2.7 or newer, otherwise disabled.
3263
3264 =item EV_USE_SELECT
3265
3266 If undefined or defined to be C<1>, libev will compile in support for the
3267 C<select>(2) backend. No attempt at auto-detection will be done: if no
3268 other method takes over, select will be it. Otherwise the select backend
3269 will not be compiled in.
3270
3271 =item EV_SELECT_USE_FD_SET
3272
3273 If defined to C<1>, then the select backend will use the system C<fd_set>
3274 structure. This is useful if libev doesn't compile due to a missing
3275 C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
3276 on exotic systems. This usually limits the range of file descriptors to
3277 some low limit such as 1024 or might have other limitations (winsocket
3278 only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
3279 configures the maximum size of the C<fd_set>.
3280
3281 =item EV_SELECT_IS_WINSOCKET
3282
3283 When defined to C<1>, the select backend will assume that
3284 select/socket/connect etc. don't understand file descriptors but
3285 wants osf handles on win32 (this is the case when the select to
3286 be used is the winsock select). This means that it will call
3287 C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
3288 it is assumed that all these functions actually work on fds, even
3289 on win32. Should not be defined on non-win32 platforms.
3290
3291 =item EV_FD_TO_WIN32_HANDLE
3292
3293 If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
3294 file descriptors to socket handles. When not defining this symbol (the
3295 default), then libev will call C<_get_osfhandle>, which is usually
3296 correct. In some cases, programs use their own file descriptor management,
3297 in which case they can provide this function to map fds to socket handles.
3298
3299 =item EV_USE_POLL
3300
3301 If defined to be C<1>, libev will compile in support for the C<poll>(2)
3302 backend. Otherwise it will be enabled on non-win32 platforms. It
3303 takes precedence over select.
3304
3305 =item EV_USE_EPOLL
3306
3307 If defined to be C<1>, libev will compile in support for the Linux
3308 C<epoll>(7) backend. Its availability will be detected at runtime,
3309 otherwise another method will be used as fallback. This is the preferred
3310 backend for GNU/Linux systems. If undefined, it will be enabled if the
3311 headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
3312
3313 =item EV_USE_KQUEUE
3314
3315 If defined to be C<1>, libev will compile in support for the BSD style
3316 C<kqueue>(2) backend. Its actual availability will be detected at runtime,
3317 otherwise another method will be used as fallback. This is the preferred
3318 backend for BSD and BSD-like systems, although on most BSDs kqueue only
3319 supports some types of fds correctly (the only platform we found that
3320 supports ptys for example was NetBSD), so kqueue might be compiled in, but
3321 not be used unless explicitly requested. The best way to use it is to find
3322 out whether kqueue supports your type of fd properly and use an embedded
3323 kqueue loop.
3324
3325 =item EV_USE_PORT
3326
3327 If defined to be C<1>, libev will compile in support for the Solaris
3328 10 port style backend. Its availability will be detected at runtime,
3329 otherwise another method will be used as fallback. This is the preferred
3330 backend for Solaris 10 systems.
3331
3332 =item EV_USE_DEVPOLL
3333
3334 Reserved for future expansion, works like the USE symbols above.
3335
3336 =item EV_USE_INOTIFY
3337
3338 If defined to be C<1>, libev will compile in support for the Linux inotify
3339 interface to speed up C<ev_stat> watchers. Its actual availability will
3340 be detected at runtime. If undefined, it will be enabled if the headers
3341 indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
3342
3343 =item EV_ATOMIC_T
3344
3345 Libev requires an integer type (suitable for storing C<0> or C<1>) whose
3346 access is atomic with respect to other threads or signal contexts. No such
3347 type is easily found in the C language, so you can provide your own type
3348 that you know is safe for your purposes. It is used both for signal handler "locking"
3349 as well as for signal and thread safety in C<ev_async> watchers.
3350
3351 In the absence of this define, libev will use C<sig_atomic_t volatile>
3352 (from F<signal.h>), which is usually good enough on most platforms.
3353
3354 =item EV_H
3355
3356 The name of the F<ev.h> header file used to include it. The default if
3357 undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
3358 used to virtually rename the F<ev.h> header file in case of conflicts.
3359
3360 =item EV_CONFIG_H
3361
3362 If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
3363 F<ev.c>'s idea of where to find the F<config.h> file, similarly to
3364 C<EV_H>, above.
3365
3366 =item EV_EVENT_H
3367
3368 Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
3369 of how the F<event.h> header can be found, the default is C<"event.h">.
3370
3371 =item EV_PROTOTYPES
3372
3373 If defined to be C<0>, then F<ev.h> will not define any function
3374 prototypes, but still define all the structs and other symbols. This is
3375 occasionally useful if you want to provide your own wrapper functions
3376 around libev functions.
3377
3378 =item EV_MULTIPLICITY
3379
3380 If undefined or defined to C<1>, then all event-loop-specific functions
3381 will have the C<struct ev_loop *> as first argument, and you can create
3382 additional independent event loops. Otherwise there will be no support
3383 for multiple event loops and there is no first event loop pointer
3384 argument. Instead, all functions act on the single default loop.
3385
3386 =item EV_MINPRI
3387
3388 =item EV_MAXPRI
3389
3390 The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
3391 C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
3392 provide for more priorities by overriding those symbols (usually defined
3393 to be C<-2> and C<2>, respectively).
3394
3395 When doing priority-based operations, libev usually has to linearly search
3396 all the priorities, so having many of them (hundreds) uses a lot of space
3397 and time, so using the defaults of five priorities (-2 .. +2) is usually
3398 fine.
3399
3400 If your embedding application does not need any priorities, defining these
3401 both to C<0> will save some memory and CPU.
3402
3403 =item EV_PERIODIC_ENABLE
3404
3405 If undefined or defined to be C<1>, then periodic timers are supported. If
3406 defined to be C<0>, then they are not. Disabling them saves a few kB of
3407 code.
3408
3409 =item EV_IDLE_ENABLE
3410
3411 If undefined or defined to be C<1>, then idle watchers are supported. If
3412 defined to be C<0>, then they are not. Disabling them saves a few kB of
3413 code.
3414
3415 =item EV_EMBED_ENABLE
3416
3417 If undefined or defined to be C<1>, then embed watchers are supported. If
3418 defined to be C<0>, then they are not. Embed watchers rely on most other
3419 watcher types, which therefore must not be disabled.
3420
3421 =item EV_STAT_ENABLE
3422
3423 If undefined or defined to be C<1>, then stat watchers are supported. If
3424 defined to be C<0>, then they are not.
3425
3426 =item EV_FORK_ENABLE
3427
3428 If undefined or defined to be C<1>, then fork watchers are supported. If
3429 defined to be C<0>, then they are not.
3430
3431 =item EV_ASYNC_ENABLE
3432
3433 If undefined or defined to be C<1>, then async watchers are supported. If
3434 defined to be C<0>, then they are not.
3435
3436 =item EV_MINIMAL
3437
3438 If you need to shave off some kilobytes of code at the expense of some
3439 speed, define this symbol to C<1>. Currently this is used to override some
3440 inlining decisions, saves roughly 30% code size on amd64. It also selects a
3441 much smaller 2-heap for timer management over the default 4-heap.
3442
3443 =item EV_PID_HASHSIZE
3444
3445 C<ev_child> watchers use a small hash table to distribute workload by
3446 pid. The default size is C<16> (or C<1> with C<EV_MINIMAL>), usually more
3447 than enough. If you need to manage thousands of children you might want to
3448 increase this value (I<must> be a power of two).
3449
3450 =item EV_INOTIFY_HASHSIZE
3451
3452 C<ev_stat> watchers use a small hash table to distribute workload by
3453 inotify watch id. The default size is C<16> (or C<1> with C<EV_MINIMAL>),
3454 usually more than enough. If you need to manage thousands of C<ev_stat>
3455 watchers you might want to increase this value (I<must> be a power of
3456 two).
3457
3458 =item EV_USE_4HEAP
3459
3460 Heaps are not very cache-efficient. To improve the cache-efficiency of the
3461 timer and periodics heaps, libev uses a 4-heap when this symbol is defined
3462 to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
3463 faster performance with many (thousands) of watchers.
3464
3465 The default is C<1> unless C<EV_MINIMAL> is set in which case it is C<0>
3466 (disabled).
3467
3468 =item EV_HEAP_CACHE_AT
3469
3470 Heaps are not very cache-efficient. To improve the cache-efficiency of the
3471 timer and periodics heaps, libev can cache the timestamp (I<at>) within
3472 the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
3473 which uses 8-12 bytes more per watcher and a few hundred bytes more code,
3474 but avoids random read accesses on heap changes. This improves performance
3475 noticeably with many (hundreds) of watchers.
3476
3477 The default is C<1> unless C<EV_MINIMAL> is set in which case it is C<0>
3478 (disabled).
3479
3480 =item EV_VERIFY
3481
3482 Controls how much internal verification (see C<ev_loop_verify ()>) will
3483 be done: If set to C<0>, no internal verification code will be compiled
3484 in. If set to C<1>, then verification code will be compiled in, but not
3485 called. If set to C<2>, then the internal verification code will be
3486 called once per loop, which can slow down libev. If set to C<3>, then the
3487 verification code will be called very frequently, which will slow down
3488 libev considerably.
3489
3490 The default is C<1>, unless C<EV_MINIMAL> is set, in which case it will be
3491 C<0>.
3492
3493 =item EV_COMMON
3494
3495 By default, all watchers have a C<void *data> member. By redefining
3496 this macro to a something else you can include more and other types of
3497 members. You have to define it each time you include one of the files,
3498 though, and it must be identical each time.
3499
3500 For example, the perl EV module uses something like this:
3501
3502 #define EV_COMMON \
3503 SV *self; /* contains this struct */ \
3504 SV *cb_sv, *fh /* note no trailing ";" */
3505
3506 =item EV_CB_DECLARE (type)
3507
3508 =item EV_CB_INVOKE (watcher, revents)
3509
3510 =item ev_set_cb (ev, cb)
3511
3512 Can be used to change the callback member declaration in each watcher,
3513 and the way callbacks are invoked and set. Must expand to a struct member
3514 definition and a statement, respectively. See the F<ev.h> header file for
3515 their default definitions. One possible use for overriding these is to
3516 avoid the C<struct ev_loop *> as first argument in all cases, or to use
3517 method calls instead of plain function calls in C++.
3518
3519 =back
3520
3521 =head2 EXPORTED API SYMBOLS
3522
3523 If you need to re-export the API (e.g. via a DLL) and you need a list of
3524 exported symbols, you can use the provided F<Symbol.*> files which list
3525 all public symbols, one per line:
3526
3527 Symbols.ev for libev proper
3528 Symbols.event for the libevent emulation
3529
3530 This can also be used to rename all public symbols to avoid clashes with
3531 multiple versions of libev linked together (which is obviously bad in
3532 itself, but sometimes it is inconvenient to avoid this).
3533
3534 A sed command like this will create wrapper C<#define>'s that you need to
3535 include before including F<ev.h>:
3536
3537 <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
3538
3539 This would create a file F<wrap.h> which essentially looks like this:
3540
3541 #define ev_backend myprefix_ev_backend
3542 #define ev_check_start myprefix_ev_check_start
3543 #define ev_check_stop myprefix_ev_check_stop
3544 ...
3545
3546 =head2 EXAMPLES
3547
3548 For a real-world example of a program the includes libev
3549 verbatim, you can have a look at the EV perl module
3550 (L<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
3551 the F<libev/> subdirectory and includes them in the F<EV/EVAPI.h> (public
3552 interface) and F<EV.xs> (implementation) files. Only the F<EV.xs> file
3553 will be compiled. It is pretty complex because it provides its own header
3554 file.
3555
3556 The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
3557 that everybody includes and which overrides some configure choices:
3558
3559 #define EV_MINIMAL 1
3560 #define EV_USE_POLL 0
3561 #define EV_MULTIPLICITY 0
3562 #define EV_PERIODIC_ENABLE 0
3563 #define EV_STAT_ENABLE 0
3564 #define EV_FORK_ENABLE 0
3565 #define EV_CONFIG_H <config.h>
3566 #define EV_MINPRI 0
3567 #define EV_MAXPRI 0
3568
3569 #include "ev++.h"
3570
3571 And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
3572
3573 #include "ev_cpp.h"
3574 #include "ev.c"
3575
3576 =head1 INTERACTION WITH OTHER PROGRAMS OR LIBRARIES
3577
3578 =head2 THREADS AND COROUTINES
3579
3580 =head3 THREADS
3581
3582 All libev functions are reentrant and thread-safe unless explicitly
3583 documented otherwise, but libev implements no locking itself. This means
3584 that you can use as many loops as you want in parallel, as long as there
3585 are no concurrent calls into any libev function with the same loop
3586 parameter (C<ev_default_*> calls have an implicit default loop parameter,
3587 of course): libev guarantees that different event loops share no data
3588 structures that need any locking.
3589
3590 Or to put it differently: calls with different loop parameters can be done
3591 concurrently from multiple threads, calls with the same loop parameter
3592 must be done serially (but can be done from different threads, as long as
3593 only one thread ever is inside a call at any point in time, e.g. by using
3594 a mutex per loop).
3595
3596 Specifically to support threads (and signal handlers), libev implements
3597 so-called C<ev_async> watchers, which allow some limited form of
3598 concurrency on the same event loop, namely waking it up "from the
3599 outside".
3600
3601 If you want to know which design (one loop, locking, or multiple loops
3602 without or something else still) is best for your problem, then I cannot
3603 help you, but here is some generic advice:
3604
3605 =over 4
3606
3607 =item * most applications have a main thread: use the default libev loop
3608 in that thread, or create a separate thread running only the default loop.
3609
3610 This helps integrating other libraries or software modules that use libev
3611 themselves and don't care/know about threading.
3612
3613 =item * one loop per thread is usually a good model.
3614
3615 Doing this is almost never wrong, sometimes a better-performance model
3616 exists, but it is always a good start.
3617
3618 =item * other models exist, such as the leader/follower pattern, where one
3619 loop is handed through multiple threads in a kind of round-robin fashion.
3620
3621 Choosing a model is hard - look around, learn, know that usually you can do
3622 better than you currently do :-)
3623
3624 =item * often you need to talk to some other thread which blocks in the
3625 event loop.
3626
3627 C<ev_async> watchers can be used to wake them up from other threads safely
3628 (or from signal contexts...).
3629
3630 An example use would be to communicate signals or other events that only
3631 work in the default loop by registering the signal watcher with the
3632 default loop and triggering an C<ev_async> watcher from the default loop
3633 watcher callback into the event loop interested in the signal.
3634
3635 =back
3636
3637 =head3 COROUTINES
3638
3639 Libev is very accommodating to coroutines ("cooperative threads"):
3640 libev fully supports nesting calls to its functions from different
3641 coroutines (e.g. you can call C<ev_loop> on the same loop from two
3642 different coroutines, and switch freely between both coroutines running the
3643 loop, as long as you don't confuse yourself). The only exception is that
3644 you must not do this from C<ev_periodic> reschedule callbacks.
3645
3646 Care has been taken to ensure that libev does not keep local state inside
3647 C<ev_loop>, and other calls do not usually allow for coroutine switches as
3648 they do not call any callbacks.
3649
3650 =head2 COMPILER WARNINGS
3651
3652 Depending on your compiler and compiler settings, you might get no or a
3653 lot of warnings when compiling libev code. Some people are apparently
3654 scared by this.
3655
3656 However, these are unavoidable for many reasons. For one, each compiler
3657 has different warnings, and each user has different tastes regarding
3658 warning options. "Warn-free" code therefore cannot be a goal except when
3659 targeting a specific compiler and compiler-version.
3660
3661 Another reason is that some compiler warnings require elaborate
3662 workarounds, or other changes to the code that make it less clear and less
3663 maintainable.
3664
3665 And of course, some compiler warnings are just plain stupid, or simply
3666 wrong (because they don't actually warn about the condition their message
3667 seems to warn about). For example, certain older gcc versions had some
3668 warnings that resulted an extreme number of false positives. These have
3669 been fixed, but some people still insist on making code warn-free with
3670 such buggy versions.
3671
3672 While libev is written to generate as few warnings as possible,
3673 "warn-free" code is not a goal, and it is recommended not to build libev
3674 with any compiler warnings enabled unless you are prepared to cope with
3675 them (e.g. by ignoring them). Remember that warnings are just that:
3676 warnings, not errors, or proof of bugs.
3677
3678
3679 =head2 VALGRIND
3680
3681 Valgrind has a special section here because it is a popular tool that is
3682 highly useful. Unfortunately, valgrind reports are very hard to interpret.
3683
3684 If you think you found a bug (memory leak, uninitialised data access etc.)
3685 in libev, then check twice: If valgrind reports something like:
3686
3687 ==2274== definitely lost: 0 bytes in 0 blocks.
3688 ==2274== possibly lost: 0 bytes in 0 blocks.
3689 ==2274== still reachable: 256 bytes in 1 blocks.
3690
3691 Then there is no memory leak, just as memory accounted to global variables
3692 is not a memleak - the memory is still being referenced, and didn't leak.
3693
3694 Similarly, under some circumstances, valgrind might report kernel bugs
3695 as if it were a bug in libev (e.g. in realloc or in the poll backend,
3696 although an acceptable workaround has been found here), or it might be
3697 confused.
3698
3699 Keep in mind that valgrind is a very good tool, but only a tool. Don't
3700 make it into some kind of religion.
3701
3702 If you are unsure about something, feel free to contact the mailing list
3703 with the full valgrind report and an explanation on why you think this
3704 is a bug in libev (best check the archives, too :). However, don't be
3705 annoyed when you get a brisk "this is no bug" answer and take the chance
3706 of learning how to interpret valgrind properly.
3707
3708 If you need, for some reason, empty reports from valgrind for your project
3709 I suggest using suppression lists.
3710
3711
3712 =head1 PORTABILITY NOTES
3713
3714 =head2 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
3715
3716 Win32 doesn't support any of the standards (e.g. POSIX) that libev
3717 requires, and its I/O model is fundamentally incompatible with the POSIX
3718 model. Libev still offers limited functionality on this platform in
3719 the form of the C<EVBACKEND_SELECT> backend, and only supports socket
3720 descriptors. This only applies when using Win32 natively, not when using
3721 e.g. cygwin.
3722
3723 Lifting these limitations would basically require the full
3724 re-implementation of the I/O system. If you are into these kinds of
3725 things, then note that glib does exactly that for you in a very portable
3726 way (note also that glib is the slowest event library known to man).
3727
3728 There is no supported compilation method available on windows except
3729 embedding it into other applications.
3730
3731 Not a libev limitation but worth mentioning: windows apparently doesn't
3732 accept large writes: instead of resulting in a partial write, windows will
3733 either accept everything or return C<ENOBUFS> if the buffer is too large,
3734 so make sure you only write small amounts into your sockets (less than a
3735 megabyte seems safe, but this apparently depends on the amount of memory
3736 available).
3737
3738 Due to the many, low, and arbitrary limits on the win32 platform and
3739 the abysmal performance of winsockets, using a large number of sockets
3740 is not recommended (and not reasonable). If your program needs to use
3741 more than a hundred or so sockets, then likely it needs to use a totally
3742 different implementation for windows, as libev offers the POSIX readiness
3743 notification model, which cannot be implemented efficiently on windows
3744 (Microsoft monopoly games).
3745
3746 A typical way to use libev under windows is to embed it (see the embedding
3747 section for details) and use the following F<evwrap.h> header file instead
3748 of F<ev.h>:
3749
3750 #define EV_STANDALONE /* keeps ev from requiring config.h */
3751 #define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
3752
3753 #include "ev.h"
3754
3755 And compile the following F<evwrap.c> file into your project (make sure
3756 you do I<not> compile the F<ev.c> or any other embedded source files!):
3757
3758 #include "evwrap.h"
3759 #include "ev.c"
3760
3761 =over 4
3762
3763 =item The winsocket select function
3764
3765 The winsocket C<select> function doesn't follow POSIX in that it
3766 requires socket I<handles> and not socket I<file descriptors> (it is
3767 also extremely buggy). This makes select very inefficient, and also
3768 requires a mapping from file descriptors to socket handles (the Microsoft
3769 C runtime provides the function C<_open_osfhandle> for this). See the
3770 discussion of the C<EV_SELECT_USE_FD_SET>, C<EV_SELECT_IS_WINSOCKET> and
3771 C<EV_FD_TO_WIN32_HANDLE> preprocessor symbols for more info.
3772
3773 The configuration for a "naked" win32 using the Microsoft runtime
3774 libraries and raw winsocket select is:
3775
3776 #define EV_USE_SELECT 1
3777 #define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
3778
3779 Note that winsockets handling of fd sets is O(n), so you can easily get a
3780 complexity in the O(n²) range when using win32.
3781
3782 =item Limited number of file descriptors
3783
3784 Windows has numerous arbitrary (and low) limits on things.
3785
3786 Early versions of winsocket's select only supported waiting for a maximum
3787 of C<64> handles (probably owning to the fact that all windows kernels
3788 can only wait for C<64> things at the same time internally; Microsoft
3789 recommends spawning a chain of threads and wait for 63 handles and the
3790 previous thread in each. Great).
3791
3792 Newer versions support more handles, but you need to define C<FD_SETSIZE>
3793 to some high number (e.g. C<2048>) before compiling the winsocket select
3794 call (which might be in libev or elsewhere, for example, perl does its own
3795 select emulation on windows).
3796
3797 Another limit is the number of file descriptors in the Microsoft runtime
3798 libraries, which by default is C<64> (there must be a hidden I<64> fetish
3799 or something like this inside Microsoft). You can increase this by calling
3800 C<_setmaxstdio>, which can increase this limit to C<2048> (another
3801 arbitrary limit), but is broken in many versions of the Microsoft runtime
3802 libraries.
3803
3804 This might get you to about C<512> or C<2048> sockets (depending on
3805 windows version and/or the phase of the moon). To get more, you need to
3806 wrap all I/O functions and provide your own fd management, but the cost of
3807 calling select (O(n²)) will likely make this unworkable.
3808
3809 =back
3810
3811 =head2 PORTABILITY REQUIREMENTS
3812
3813 In addition to a working ISO-C implementation and of course the
3814 backend-specific APIs, libev relies on a few additional extensions:
3815
3816 =over 4
3817
3818 =item C<void (*)(ev_watcher_type *, int revents)> must have compatible
3819 calling conventions regardless of C<ev_watcher_type *>.
3820
3821 Libev assumes not only that all watcher pointers have the same internal
3822 structure (guaranteed by POSIX but not by ISO C for example), but it also
3823 assumes that the same (machine) code can be used to call any watcher
3824 callback: The watcher callbacks have different type signatures, but libev
3825 calls them using an C<ev_watcher *> internally.
3826
3827 =item C<sig_atomic_t volatile> must be thread-atomic as well
3828
3829 The type C<sig_atomic_t volatile> (or whatever is defined as
3830 C<EV_ATOMIC_T>) must be atomic with respect to accesses from different
3831 threads. This is not part of the specification for C<sig_atomic_t>, but is
3832 believed to be sufficiently portable.
3833
3834 =item C<sigprocmask> must work in a threaded environment
3835
3836 Libev uses C<sigprocmask> to temporarily block signals. This is not
3837 allowed in a threaded program (C<pthread_sigmask> has to be used). Typical
3838 pthread implementations will either allow C<sigprocmask> in the "main
3839 thread" or will block signals process-wide, both behaviours would
3840 be compatible with libev. Interaction between C<sigprocmask> and
3841 C<pthread_sigmask> could complicate things, however.
3842
3843 The most portable way to handle signals is to block signals in all threads
3844 except the initial one, and run the default loop in the initial thread as
3845 well.
3846
3847 =item C<long> must be large enough for common memory allocation sizes
3848
3849 To improve portability and simplify its API, libev uses C<long> internally
3850 instead of C<size_t> when allocating its data structures. On non-POSIX
3851 systems (Microsoft...) this might be unexpectedly low, but is still at
3852 least 31 bits everywhere, which is enough for hundreds of millions of
3853 watchers.
3854
3855 =item C<double> must hold a time value in seconds with enough accuracy
3856
3857 The type C<double> is used to represent timestamps. It is required to
3858 have at least 51 bits of mantissa (and 9 bits of exponent), which is good
3859 enough for at least into the year 4000. This requirement is fulfilled by
3860 implementations implementing IEEE 754 (basically all existing ones).
3861
3862 =back
3863
3864 If you know of other additional requirements drop me a note.
3865
3866
3867 =head1 ALGORITHMIC COMPLEXITIES
3868
3869 In this section the complexities of (many of) the algorithms used inside
3870 libev will be documented. For complexity discussions about backends see
3871 the documentation for C<ev_default_init>.
3872
3873 All of the following are about amortised time: If an array needs to be
3874 extended, libev needs to realloc and move the whole array, but this
3875 happens asymptotically rarer with higher number of elements, so O(1) might
3876 mean that libev does a lengthy realloc operation in rare cases, but on
3877 average it is much faster and asymptotically approaches constant time.
3878
3879 =over 4
3880
3881 =item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
3882
3883 This means that, when you have a watcher that triggers in one hour and
3884 there are 100 watchers that would trigger before that, then inserting will
3885 have to skip roughly seven (C<ld 100>) of these watchers.
3886
3887 =item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
3888
3889 That means that changing a timer costs less than removing/adding them,
3890 as only the relative motion in the event queue has to be paid for.
3891
3892 =item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
3893
3894 These just add the watcher into an array or at the head of a list.
3895
3896 =item Stopping check/prepare/idle/fork/async watchers: O(1)
3897
3898 =item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
3899
3900 These watchers are stored in lists, so they need to be walked to find the
3901 correct watcher to remove. The lists are usually short (you don't usually
3902 have many watchers waiting for the same fd or signal: one is typical, two
3903 is rare).
3904
3905 =item Finding the next timer in each loop iteration: O(1)
3906
3907 By virtue of using a binary or 4-heap, the next timer is always found at a
3908 fixed position in the storage array.
3909
3910 =item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
3911
3912 A change means an I/O watcher gets started or stopped, which requires
3913 libev to recalculate its status (and possibly tell the kernel, depending
3914 on backend and whether C<ev_io_set> was used).
3915
3916 =item Activating one watcher (putting it into the pending state): O(1)
3917
3918 =item Priority handling: O(number_of_priorities)
3919
3920 Priorities are implemented by allocating some space for each
3921 priority. When doing priority-based operations, libev usually has to
3922 linearly search all the priorities, but starting/stopping and activating
3923 watchers becomes O(1) with respect to priority handling.
3924
3925 =item Sending an ev_async: O(1)
3926
3927 =item Processing ev_async_send: O(number_of_async_watchers)
3928
3929 =item Processing signals: O(max_signal_number)
3930
3931 Sending involves a system call I<iff> there were no other C<ev_async_send>
3932 calls in the current loop iteration. Checking for async and signal events
3933 involves iterating over all running async watchers or all signal numbers.
3934
3935 =back
3936
3937
3938 =head1 AUTHOR
3939
3940 Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael Magnusson.
3941