1LIBEV(3) libev - high performance full featured event loop LIBEV(3)
2
6 libev - a high performance full-featured event loop written in C
7
9 #include <ev.h>
10 EXAMPLE PROGRAM
12 // a single header file is required
13 #include <ev.h>
14 #include <stdio.h> // for puts
16 // every watcher type has its own typedef'd struct
18 // with the name ev_TYPE
19 ev_io stdin_watcher;
20 ev_timer timeout_watcher;
21 // all watcher callbacks have a similar signature
23 // this callback is called when data is readable on stdin
24 static void
25 stdin_cb (EV_P_ ev_io *w, int revents)
26 {
27 puts ("stdin ready");
28 // for one-shot events, one must manually stop the watcher
29 // with its corresponding stop function.
30 ev_io_stop (EV_A_ w);
31 // this causes all nested ev_run's to stop iterating
33 ev_break (EV_A_ EVBREAK_ALL);
34 }
35 // another callback, this time for a time-out
37 static void
38 timeout_cb (EV_P_ ev_timer *w, int revents)
39 {
40 puts ("timeout");
41 // this causes the innermost ev_run to stop iterating
42 ev_break (EV_A_ EVBREAK_ONE);
43 }
44 int
46 main (void)
47 {
48 // use the default event loop unless you have special needs
49 struct ev_loop *loop = EV_DEFAULT;
50 // initialise an io watcher, then start it
52 // this one will watch for stdin to become readable
53 ev_io_init (&stdin_watcher, stdin_cb, /*STDIN_FILENO*/ 0, EV_READ);
54 ev_io_start (loop, &stdin_watcher);
55 // initialise a timer watcher, then start it
57 // simple non-repeating 5.5 second timeout
58 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
59 ev_timer_start (loop, &timeout_watcher);
60 // now wait for events to arrive
62 ev_run (loop, 0);
63 // break was called, so exit
65 return 0;
66 }
67
69 This document documents the libev software package.
70 The newest version of this document is also available as an html-
72 formatted web page you might find easier to navigate when reading it
73 for the first time:
74 <http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod>.
75 While this document tries to be as complete as possible in documenting
77 libev, its usage and the rationale behind its design, it is not a
78 tutorial on event-based programming, nor will it introduce event-based
79 programming with libev.
80 Familiarity with event based programming techniques in general is
82 assumed throughout this document.
83
85 This manual tries to be very detailed, but unfortunately, this also
86 makes it very long. If you just want to know the basics of libev, I
87 suggest reading "ANATOMY OF A WATCHER", then the "EXAMPLE PROGRAM"
88 above and look up the missing functions in "GLOBAL FUNCTIONS" and the
89 "ev_io" and "ev_timer" sections in "WATCHER TYPES".
90
92 Libev is an event loop: you register interest in certain events (such
93 as a file descriptor being readable or a timeout occurring), and it
94 will manage these event sources and provide your program with events.
95 To do this, it must take more or less complete control over your
97 process (or thread) by executing the event loop handler, and will then
98 communicate events via a callback mechanism.
99 You register interest in certain events by registering so-called event
101 watchers, which are relatively small C structures you initialise with
102 the details of the event, and then hand it over to libev by starting
103 the watcher.
104 FEATURES
106 Libev supports "select", "poll", the Linux-specific aio and "epoll"
107 interfaces, the BSD-specific "kqueue" and the Solaris-specific event
108 port mechanisms for file descriptor events ("ev_io"), the Linux
109 "inotify" interface (for "ev_stat"), Linux eventfd/signalfd (for faster
110 and cleaner inter-thread wakeup ("ev_async")/signal handling
111 ("ev_signal")) relative timers ("ev_timer"), absolute timers with
112 customised rescheduling ("ev_periodic"), synchronous signals
113 ("ev_signal"), process status change events ("ev_child"), and event
114 watchers dealing with the event loop mechanism itself ("ev_idle",
115 "ev_embed", "ev_prepare" and "ev_check" watchers) as well as file
116 watchers ("ev_stat") and even limited support for fork events
117 ("ev_fork").
118 It also is quite fast (see this benchmark
120 <http://libev.schmorp.de/bench.html> comparing it to libevent for
121 example).
122 CONVENTIONS
124 Libev is very configurable. In this manual the default (and most
125 common) configuration will be described, which supports multiple event
126 loops. For more info about various configuration options please have a
127 look at EMBED section in this manual. If libev was configured without
128 support for multiple event loops, then all functions taking an initial
129 argument of name "loop" (which is always of type "struct ev_loop *")
130 will not have this argument.
131 TIME REPRESENTATION
133 Libev represents time as a single floating point number, representing
134 the (fractional) number of seconds since the (POSIX) epoch (in practice
135 somewhere near the beginning of 1970, details are complicated, don't
136 ask). This type is called "ev_tstamp", which is what you should use
137 too. It usually aliases to the "double" type in C. When you need to do
138 any calculations on it, you should treat it as some floating point
139 value.
140 Unlike the name component "stamp" might indicate, it is also used for
142 time differences (e.g. delays) throughout libev.
143
145 Libev knows three classes of errors: operating system errors, usage
146 errors and internal errors (bugs).
147 When libev catches an operating system error it cannot handle (for
149 example a system call indicating a condition libev cannot fix), it
150 calls the callback set via "ev_set_syserr_cb", which is supposed to fix
151 the problem or abort. The default is to print a diagnostic message and
152 to call "abort ()".
153 When libev detects a usage error such as a negative timer interval,
155 then it will print a diagnostic message and abort (via the "assert"
156 mechanism, so "NDEBUG" will disable this checking): these are
157 programming errors in the libev caller and need to be fixed there.
158 Via the "EV_FREQUENT" macro you can compile in and/or enable extensive
160 consistency checking code inside libev that can be used to check for
161 internal inconsistencies, suually caused by application bugs.
162 Libev also has a few internal error-checking "assert"ions. These do not
164 trigger under normal circumstances, as they indicate either a bug in
165 libev or worse.
166
168 These functions can be called anytime, even before initialising the
169 library in any way.
170 ev_tstamp ev_time ()
172 Returns the current time as libev would use it. Please note that
173 the "ev_now" function is usually faster and also often returns the
174 timestamp you actually want to know. Also interesting is the
175 combination of "ev_now_update" and "ev_now".
176 ev_sleep (ev_tstamp interval)
178 Sleep for the given interval: The current thread will be blocked
179 until either it is interrupted or the given time interval has
180 passed (approximately - it might return a bit earlier even if not
181 interrupted). Returns immediately if "interval <= 0".
182 Basically this is a sub-second-resolution "sleep ()".
184 The range of the "interval" is limited - libev only guarantees to
186 work with sleep times of up to one day ("interval <= 86400").
187 int ev_version_major ()
189 int ev_version_minor ()
190 You can find out the major and minor ABI version numbers of the
191 library you linked against by calling the functions
192 "ev_version_major" and "ev_version_minor". If you want, you can
193 compare against the global symbols "EV_VERSION_MAJOR" and
194 "EV_VERSION_MINOR", which specify the version of the library your
195 program was compiled against.
196 These version numbers refer to the ABI version of the library, not
198 the release version.
199 Usually, it's a good idea to terminate if the major versions
201 mismatch, as this indicates an incompatible change. Minor versions
202 are usually compatible to older versions, so a larger minor version
203 alone is usually not a problem.
204 Example: Make sure we haven't accidentally been linked against the
206 wrong version (note, however, that this will not detect other ABI
207 mismatches, such as LFS or reentrancy).
208 assert (("libev version mismatch",
210 ev_version_major () == EV_VERSION_MAJOR
211 && ev_version_minor () >= EV_VERSION_MINOR));
212 unsigned int ev_supported_backends ()
214 Return the set of all backends (i.e. their corresponding
215 "EV_BACKEND_*" value) compiled into this binary of libev
216 (independent of their availability on the system you are running
217 on). See "ev_default_loop" for a description of the set values.
218 Example: make sure we have the epoll method, because yeah this is
220 cool and a must have and can we have a torrent of it please!!!11
221 assert (("sorry, no epoll, no sex",
223 ev_supported_backends () & EVBACKEND_EPOLL));
224 unsigned int ev_recommended_backends ()
226 Return the set of all backends compiled into this binary of libev
227 and also recommended for this platform, meaning it will work for
228 most file descriptor types. This set is often smaller than the one
229 returned by "ev_supported_backends", as for example kqueue is
230 broken on most BSDs and will not be auto-detected unless you
231 explicitly request it (assuming you know what you are doing). This
232 is the set of backends that libev will probe for if you specify no
233 backends explicitly.
234 unsigned int ev_embeddable_backends ()
236 Returns the set of backends that are embeddable in other event
237 loops. This value is platform-specific but can include backends not
238 available on the current system. To find which embeddable backends
239 might be supported on the current system, you would need to look at
240 "ev_embeddable_backends () & ev_supported_backends ()", likewise
241 for recommended ones.
242 See the description of "ev_embed" watchers for more info.
244 ev_set_allocator (void *(*cb)(void *ptr, long size) throw ())
246 Sets the allocation function to use (the prototype is similar - the
247 semantics are identical to the "realloc" C89/SuS/POSIX function).
248 It is used to allocate and free memory (no surprises here). If it
249 returns zero when memory needs to be allocated ("size != 0"), the
250 library might abort or take some potentially destructive action.
251 Since some systems (at least OpenBSD and Darwin) fail to implement
253 correct "realloc" semantics, libev will use a wrapper around the
254 system "realloc" and "free" functions by default.
255 You could override this function in high-availability programs to,
257 say, free some memory if it cannot allocate memory, to use a
258 special allocator, or even to sleep a while and retry until some
259 memory is available.
260 Example: The following is the "realloc" function that libev itself
262 uses which should work with "realloc" and "free" functions of all
263 kinds and is probably a good basis for your own implementation.
264 static void *
266 ev_realloc_emul (void *ptr, long size) EV_NOEXCEPT
267 {
268 if (size)
269 return realloc (ptr, size);
270 free (ptr);
272 return 0;
273 }
274 Example: Replace the libev allocator with one that waits a bit and
276 then retries.
277 static void *
279 persistent_realloc (void *ptr, size_t size)
280 {
281 if (!size)
282 {
283 free (ptr);
284 return 0;
285 }
286 for (;;)
288 {
289 void *newptr = realloc (ptr, size);
290 if (newptr)
292 return newptr;
293 sleep (60);
295 }
296 }
297 ...
299 ev_set_allocator (persistent_realloc);
300 ev_set_syserr_cb (void (*cb)(const char *msg) throw ())
302 Set the callback function to call on a retryable system call error
303 (such as failed select, poll, epoll_wait). The message is a
304 printable string indicating the system call or subsystem causing
305 the problem. If this callback is set, then libev will expect it to
306 remedy the situation, no matter what, when it returns. That is,
307 libev will generally retry the requested operation, or, if the
308 condition doesn't go away, do bad stuff (such as abort).
309 Example: This is basically the same thing that libev does
311 internally, too.
312 static void
314 fatal_error (const char *msg)
315 {
316 perror (msg);
317 abort ();
318 }
319 ...
321 ev_set_syserr_cb (fatal_error);
322 ev_feed_signal (int signum)
324 This function can be used to "simulate" a signal receive. It is
325 completely safe to call this function at any time, from any
326 context, including signal handlers or random threads.
327 Its main use is to customise signal handling in your process,
329 especially in the presence of threads. For example, you could block
330 signals by default in all threads (and specifying
331 "EVFLAG_NOSIGMASK" when creating any loops), and in one thread, use
332 "sigwait" or any other mechanism to wait for signals, then
333 "deliver" them to libev by calling "ev_feed_signal".
334
336 An event loop is described by a "struct ev_loop *" (the "struct" is not
337 optional in this case unless libev 3 compatibility is disabled, as
338 libev 3 had an "ev_loop" function colliding with the struct name).
339 The library knows two types of such loops, the default loop, which
341 supports child process events, and dynamically created event loops
342 which do not.
343 struct ev_loop *ev_default_loop (unsigned int flags)
345 This returns the "default" event loop object, which is what you
346 should normally use when you just need "the event loop". Event loop
347 objects and the "flags" parameter are described in more detail in
348 the entry for "ev_loop_new".
349 If the default loop is already initialised then this function
351 simply returns it (and ignores the flags. If that is troubling you,
352 check "ev_backend ()" afterwards). Otherwise it will create it with
353 the given flags, which should almost always be 0, unless the caller
354 is also the one calling "ev_run" or otherwise qualifies as "the
355 main program".
356 If you don't know what event loop to use, use the one returned from
358 this function (or via the "EV_DEFAULT" macro).
359 Note that this function is not thread-safe, so if you want to use
361 it from multiple threads, you have to employ some kind of mutex
362 (note also that this case is unlikely, as loops cannot be shared
363 easily between threads anyway).
364 The default loop is the only loop that can handle "ev_child"
366 watchers, and to do this, it always registers a handler for
367 "SIGCHLD". If this is a problem for your application you can either
368 create a dynamic loop with "ev_loop_new" which doesn't do that, or
369 you can simply overwrite the "SIGCHLD" signal handler after calling
370 "ev_default_init".
371 Example: This is the most typical usage.
373 if (!ev_default_loop (0))
375 fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
376 Example: Restrict libev to the select and poll backends, and do not
378 allow environment settings to be taken into account:
379 ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
381 struct ev_loop *ev_loop_new (unsigned int flags)
383 This will create and initialise a new event loop object. If the
384 loop could not be initialised, returns false.
385 This function is thread-safe, and one common way to use libev with
387 threads is indeed to create one loop per thread, and using the
388 default loop in the "main" or "initial" thread.
389 The flags argument can be used to specify special behaviour or
391 specific backends to use, and is usually specified as 0 (or
392 "EVFLAG_AUTO").
393 The following flags are supported:
395 "EVFLAG_AUTO"
397 The default flags value. Use this if you have no clue (it's the
398 right thing, believe me).
399 "EVFLAG_NOENV"
401 If this flag bit is or'ed into the flag value (or the program
402 runs setuid or setgid) then libev will not look at the
403 environment variable "LIBEV_FLAGS". Otherwise (the default),
404 this environment variable will override the flags completely if
405 it is found in the environment. This is useful to try out
406 specific backends to test their performance, to work around
407 bugs, or to make libev threadsafe (accessing environment
408 variables cannot be done in a threadsafe way, but usually it
409 works if no other thread modifies them).
410 "EVFLAG_FORKCHECK"
412 Instead of calling "ev_loop_fork" manually after a fork, you
413 can also make libev check for a fork in each iteration by
414 enabling this flag.
415 This works by calling "getpid ()" on every iteration of the
417 loop, and thus this might slow down your event loop if you do a
418 lot of loop iterations and little real work, but is usually not
419 noticeable (on my GNU/Linux system for example, "getpid" is
420 actually a simple 5-insn sequence without a system call and
421 thus very fast, but my GNU/Linux system also has
422 "pthread_atfork" which is even faster). (Update: glibc versions
423 2.25 apparently removed the "getpid" optimisation again).
424 The big advantage of this flag is that you can forget about
426 fork (and forget about forgetting to tell libev about forking,
427 although you still have to ignore "SIGPIPE") when you use this
428 flag.
429 This flag setting cannot be overridden or specified in the
431 "LIBEV_FLAGS" environment variable.
432 "EVFLAG_NOINOTIFY"
434 When this flag is specified, then libev will not attempt to use
435 the inotify API for its "ev_stat" watchers. Apart from
436 debugging and testing, this flag can be useful to conserve
437 inotify file descriptors, as otherwise each loop using
438 "ev_stat" watchers consumes one inotify handle.
439 "EVFLAG_SIGNALFD"
441 When this flag is specified, then libev will attempt to use the
442 signalfd API for its "ev_signal" (and "ev_child") watchers.
443 This API delivers signals synchronously, which makes it both
444 faster and might make it possible to get the queued signal
445 data. It can also simplify signal handling with threads, as
446 long as you properly block signals in your threads that are not
447 interested in handling them.
448 Signalfd will not be used by default as this changes your
450 signal mask, and there are a lot of shoddy libraries and
451 programs (glib's threadpool for example) that can't properly
452 initialise their signal masks.
453 "EVFLAG_NOSIGMASK"
455 When this flag is specified, then libev will avoid to modify
456 the signal mask. Specifically, this means you have to make sure
457 signals are unblocked when you want to receive them.
458 This behaviour is useful when you want to do your own signal
460 handling, or want to handle signals only in specific threads
461 and want to avoid libev unblocking the signals.
462 It's also required by POSIX in a threaded program, as libev
464 calls "sigprocmask", whose behaviour is officially unspecified.
465 This flag's behaviour will become the default in future
467 versions of libev.
468 "EVBACKEND_SELECT" (value 1, portable select backend)
470 This is your standard select(2) backend. Not completely
471 standard, as libev tries to roll its own fd_set with no limits
472 on the number of fds, but if that fails, expect a fairly low
473 limit on the number of fds when using this backend. It doesn't
474 scale too well (O(highest_fd)), but its usually the fastest
475 backend for a low number of (low-numbered :) fds.
476 To get good performance out of this backend you need a high
478 amount of parallelism (most of the file descriptors should be
479 busy). If you are writing a server, you should "accept ()" in a
480 loop to accept as many connections as possible during one
481 iteration. You might also want to have a look at
482 "ev_set_io_collect_interval ()" to increase the amount of
483 readiness notifications you get per iteration.
484 This backend maps "EV_READ" to the "readfds" set and "EV_WRITE"
486 to the "writefds" set (and to work around Microsoft Windows
487 bugs, also onto the "exceptfds" set on that platform).
488 "EVBACKEND_POLL" (value 2, poll backend, available everywhere
490 except on windows)
491 And this is your standard poll(2) backend. It's more
492 complicated than select, but handles sparse fds better and has
493 no artificial limit on the number of fds you can use (except it
494 will slow down considerably with a lot of inactive fds). It
495 scales similarly to select, i.e. O(total_fds). See the entry
496 for "EVBACKEND_SELECT", above, for performance tips.
497 This backend maps "EV_READ" to "POLLIN | POLLERR | POLLHUP",
499 and "EV_WRITE" to "POLLOUT | POLLERR | POLLHUP".
500 "EVBACKEND_EPOLL" (value 4, Linux)
502 Use the Linux-specific epoll(7) interface (for both pre- and
503 post-2.6.9 kernels).
504 For few fds, this backend is a bit little slower than poll and
506 select, but it scales phenomenally better. While poll and
507 select usually scale like O(total_fds) where total_fds is the
508 total number of fds (or the highest fd), epoll scales either
509 O(1) or O(active_fds).
510 The epoll mechanism deserves honorable mention as the most
512 misdesigned of the more advanced event mechanisms: mere
513 annoyances include silently dropping file descriptors,
514 requiring a system call per change per file descriptor (and
515 unnecessary guessing of parameters), problems with dup,
516 returning before the timeout value, resulting in additional
517 iterations (and only giving 5ms accuracy while select on the
518 same platform gives 0.1ms) and so on. The biggest issue is fork
519 races, however - if a program forks then both parent and child
520 process have to recreate the epoll set, which can take
521 considerable time (one syscall per file descriptor) and is of
522 course hard to detect.
523 Epoll is also notoriously buggy - embedding epoll fds should
525 work, but of course doesn't, and epoll just loves to report
526 events for totally different file descriptors (even already
527 closed ones, so one cannot even remove them from the set) than
528 registered in the set (especially on SMP systems). Libev tries
529 to counter these spurious notifications by employing an
530 additional generation counter and comparing that against the
531 events to filter out spurious ones, recreating the set when
532 required. Epoll also erroneously rounds down timeouts, but
533 gives you no way to know when and by how much, so sometimes you
534 have to busy-wait because epoll returns immediately despite a
535 nonzero timeout. And last not least, it also refuses to work
536 with some file descriptors which work perfectly fine with
537 "select" (files, many character devices...).
538 Epoll is truly the train wreck among event poll mechanisms, a
540 frankenpoll, cobbled together in a hurry, no thought to design
541 or interaction with others. Oh, the pain, will it ever stop...
542 While stopping, setting and starting an I/O watcher in the same
544 iteration will result in some caching, there is still a system
545 call per such incident (because the same file descriptor could
546 point to a different file description now), so its best to
547 avoid that. Also, "dup ()"'ed file descriptors might not work
548 very well if you register events for both file descriptors.
549 Best performance from this backend is achieved by not
551 unregistering all watchers for a file descriptor until it has
552 been closed, if possible, i.e. keep at least one watcher active
553 per fd at all times. Stopping and starting a watcher (without
554 re-setting it) also usually doesn't cause extra overhead. A
555 fork can both result in spurious notifications as well as in
556 libev having to destroy and recreate the epoll object, which
557 can take considerable time and thus should be avoided.
558 All this means that, in practice, "EVBACKEND_SELECT" can be as
560 fast or faster than epoll for maybe up to a hundred file
561 descriptors, depending on the usage. So sad.
562 While nominally embeddable in other event loops, this feature
564 is broken in a lot of kernel revisions, but probably(!) works
565 in current versions.
566 This backend maps "EV_READ" and "EV_WRITE" in the same way as
568 "EVBACKEND_POLL".
569 "EVBACKEND_LINUXAIO" (value 64, Linux)
571 Use the Linux-specific Linux AIO (not aio(7) but io_submit(2))
572 event interface available in post-4.18 kernels (but libev only
573 tries to use it in 4.19+).
574 This is another Linux train wreck of an event interface.
576 If this backend works for you (as of this writing, it was very
578 experimental), it is the best event interface available on
579 Linux and might be well worth enabling it - if it isn't
580 available in your kernel this will be detected and this backend
581 will be skipped.
582 This backend can batch oneshot requests and supports a user-
584 space ring buffer to receive events. It also doesn't suffer
585 from most of the design problems of epoll (such as not being
586 able to remove event sources from the epoll set), and generally
587 sounds too good to be true. Because, this being the Linux
588 kernel, of course it suffers from a whole new set of
589 limitations, forcing you to fall back to epoll, inheriting all
590 its design issues.
591 For one, it is not easily embeddable (but probably could be
593 done using an event fd at some extra overhead). It also is
594 subject to a system wide limit that can be configured in
595 /proc/sys/fs/aio-max-nr. If no AIO requests are left, this
596 backend will be skipped during initialisation, and will switch
597 to epoll when the loop is active.
598 Most problematic in practice, however, is that not all file
600 descriptors work with it. For example, in Linux 5.1, TCP
601 sockets, pipes, event fds, files, /dev/null and many others are
602 supported, but ttys do not work properly (a known bug that the
603 kernel developers don't care about, see
604 <https://lore.kernel.org/patchwork/patch/1047453/>), so this is
605 not (yet?) a generic event polling interface.
606 Overall, it seems the Linux developers just don't want it to
608 have a generic event handling mechanism other than "select" or
609 "poll".
610 To work around all these problem, the current version of libev
612 uses its epoll backend as a fallback for file descriptor types
613 that do not work. Or falls back completely to epoll if the
614 kernel acts up.
615 This backend maps "EV_READ" and "EV_WRITE" in the same way as
617 "EVBACKEND_POLL".
618 "EVBACKEND_KQUEUE" (value 8, most BSD clones)
620 Kqueue deserves special mention, as at the time this backend
621 was implemented, it was broken on all BSDs except NetBSD
622 (usually it doesn't work reliably with anything but sockets and
623 pipes, except on Darwin, where of course it's completely
624 useless). Unlike epoll, however, whose brokenness is by design,
625 these kqueue bugs can be (and mostly have been) fixed without
626 API changes to existing programs. For this reason it's not
627 being "auto-detected" on all platforms unless you explicitly
628 specify it in the flags (i.e. using "EVBACKEND_KQUEUE") or
629 libev was compiled on a known-to-be-good (-enough) system like
630 NetBSD.
631 You still can embed kqueue into a normal poll or select backend
633 and use it only for sockets (after having made sure that
634 sockets work with kqueue on the target platform). See
635 "ev_embed" watchers for more info.
636 It scales in the same way as the epoll backend, but the
638 interface to the kernel is more efficient (which says nothing
639 about its actual speed, of course). While stopping, setting and
640 starting an I/O watcher does never cause an extra system call
641 as with "EVBACKEND_EPOLL", it still adds up to two event
642 changes per incident. Support for "fork ()" is very bad (you
643 might have to leak fds on fork, but it's more sane than epoll)
644 and it drops fds silently in similarly hard-to-detect cases.
645 This backend usually performs well under most conditions.
647 While nominally embeddable in other event loops, this doesn't
649 work everywhere, so you might need to test for this. And since
650 it is broken almost everywhere, you should only use it when you
651 have a lot of sockets (for which it usually works), by
652 embedding it into another event loop (e.g. "EVBACKEND_SELECT"
653 or "EVBACKEND_POLL" (but "poll" is of course also broken on OS
654 X)) and, did I mention it, using it only for sockets.
655 This backend maps "EV_READ" into an "EVFILT_READ" kevent with
657 "NOTE_EOF", and "EV_WRITE" into an "EVFILT_WRITE" kevent with
658 "NOTE_EOF".
659 "EVBACKEND_DEVPOLL" (value 16, Solaris 8)
661 This is not implemented yet (and might never be, unless you
662 send me an implementation). According to reports, "/dev/poll"
663 only supports sockets and is not embeddable, which would limit
664 the usefulness of this backend immensely.
665 "EVBACKEND_PORT" (value 32, Solaris 10)
667 This uses the Solaris 10 event port mechanism. As with
668 everything on Solaris, it's really slow, but it still scales
669 very well (O(active_fds)).
670 While this backend scales well, it requires one system call per
672 active file descriptor per loop iteration. For small and medium
673 numbers of file descriptors a "slow" "EVBACKEND_SELECT" or
674 "EVBACKEND_POLL" backend might perform better.
675 On the positive side, this backend actually performed fully to
677 specification in all tests and is fully embeddable, which is a
678 rare feat among the OS-specific backends (I vastly prefer
679 correctness over speed hacks).
680 On the negative side, the interface is bizarre - so bizarre
682 that even sun itself gets it wrong in their code examples: The
683 event polling function sometimes returns events to the caller
684 even though an error occurred, but with no indication whether
685 it has done so or not (yes, it's even documented that way) -
686 deadly for edge-triggered interfaces where you absolutely have
687 to know whether an event occurred or not because you have to
688 re-arm the watcher.
689 Fortunately libev seems to be able to work around these
691 idiocies.
692 This backend maps "EV_READ" and "EV_WRITE" in the same way as
694 "EVBACKEND_POLL".
695 "EVBACKEND_ALL"
697 Try all backends (even potentially broken ones that wouldn't be
698 tried with "EVFLAG_AUTO"). Since this is a mask, you can do
699 stuff such as "EVBACKEND_ALL & ~EVBACKEND_KQUEUE".
700 It is definitely not recommended to use this flag, use whatever
702 "ev_recommended_backends ()" returns, or simply do not specify
703 a backend at all.
704 "EVBACKEND_MASK"
706 Not a backend at all, but a mask to select all backend bits
707 from a "flags" value, in case you want to mask out any backends
708 from a flags value (e.g. when modifying the "LIBEV_FLAGS"
709 environment variable).
710 If one or more of the backend flags are or'ed into the flags value,
712 then only these backends will be tried (in the reverse order as
713 listed here). If none are specified, all backends in
714 "ev_recommended_backends ()" will be tried.
715 Example: Try to create a event loop that uses epoll and nothing
717 else.
718 struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
720 if (!epoller)
721 fatal ("no epoll found here, maybe it hides under your chair");
722 Example: Use whatever libev has to offer, but make sure that kqueue
724 is used if available.
725 struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_KQUEUE);
727 Example: Similarly, on linux, you mgiht want to take advantage of
729 the linux aio backend if possible, but fall back to something else
730 if that isn't available.
731 struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_LINUXAIO);
733 ev_loop_destroy (loop)
735 Destroys an event loop object (frees all memory and kernel state
736 etc.). None of the active event watchers will be stopped in the
737 normal sense, so e.g. "ev_is_active" might still return true. It is
738 your responsibility to either stop all watchers cleanly yourself
739 before calling this function, or cope with the fact afterwards
740 (which is usually the easiest thing, you can just ignore the
741 watchers and/or "free ()" them for example).
742 Note that certain global state, such as signal state (and installed
744 signal handlers), will not be freed by this function, and related
745 watchers (such as signal and child watchers) would need to be
746 stopped manually.
747 This function is normally used on loop objects allocated by
749 "ev_loop_new", but it can also be used on the default loop returned
750 by "ev_default_loop", in which case it is not thread-safe.
751 Note that it is not advisable to call this function on the default
753 loop except in the rare occasion where you really need to free its
754 resources. If you need dynamically allocated loops it is better to
755 use "ev_loop_new" and "ev_loop_destroy".
756 ev_loop_fork (loop)
758 This function sets a flag that causes subsequent "ev_run"
759 iterations to reinitialise the kernel state for backends that have
760 one. Despite the name, you can call it anytime you are allowed to
761 start or stop watchers (except inside an "ev_prepare" callback),
762 but it makes most sense after forking, in the child process. You
763 must call it (or use "EVFLAG_FORKCHECK") in the child before
764 resuming or calling "ev_run".
765 In addition, if you want to reuse a loop (via this function or
767 "EVFLAG_FORKCHECK"), you also have to ignore "SIGPIPE".
768 Again, you have to call it on any loop that you want to re-use
770 after a fork, even if you do not plan to use the loop in the
771 parent. This is because some kernel interfaces *cough* kqueue
772 *cough* do funny things during fork.
773 On the other hand, you only need to call this function in the child
775 process if and only if you want to use the event loop in the child.
776 If you just fork+exec or create a new loop in the child, you don't
777 have to call it at all (in fact, "epoll" is so badly broken that it
778 makes a difference, but libev will usually detect this case on its
779 own and do a costly reset of the backend).
780 The function itself is quite fast and it's usually not a problem to
782 call it just in case after a fork.
783 Example: Automate calling "ev_loop_fork" on the default loop when
785 using pthreads.
786 static void
788 post_fork_child (void)
789 {
790 ev_loop_fork (EV_DEFAULT);
791 }
792 ...
794 pthread_atfork (0, 0, post_fork_child);
795 int ev_is_default_loop (loop)
797 Returns true when the given loop is, in fact, the default loop, and
798 false otherwise.
799 unsigned int ev_iteration (loop)
801 Returns the current iteration count for the event loop, which is
802 identical to the number of times libev did poll for new events. It
803 starts at 0 and happily wraps around with enough iterations.
804 This value can sometimes be useful as a generation counter of sorts
806 (it "ticks" the number of loop iterations), as it roughly
807 corresponds with "ev_prepare" and "ev_check" calls - and is
808 incremented between the prepare and check phases.
809 unsigned int ev_depth (loop)
811 Returns the number of times "ev_run" was entered minus the number
812 of times "ev_run" was exited normally, in other words, the
813 recursion depth.
814 Outside "ev_run", this number is zero. In a callback, this number
816 is 1, unless "ev_run" was invoked recursively (or from another
817 thread), in which case it is higher.
818 Leaving "ev_run" abnormally (setjmp/longjmp, cancelling the thread,
820 throwing an exception etc.), doesn't count as "exit" - consider
821 this as a hint to avoid such ungentleman-like behaviour unless it's
822 really convenient, in which case it is fully supported.
823 unsigned int ev_backend (loop)
825 Returns one of the "EVBACKEND_*" flags indicating the event backend
826 in use.
827 ev_tstamp ev_now (loop)
829 Returns the current "event loop time", which is the time the event
830 loop received events and started processing them. This timestamp
831 does not change as long as callbacks are being processed, and this
832 is also the base time used for relative timers. You can treat it as
833 the timestamp of the event occurring (or more correctly, libev
834 finding out about it).
835 ev_now_update (loop)
837 Establishes the current time by querying the kernel, updating the
838 time returned by "ev_now ()" in the progress. This is a costly
839 operation and is usually done automatically within "ev_run ()".
840 This function is rarely useful, but when some event callback runs
842 for a very long time without entering the event loop, updating
843 libev's idea of the current time is a good idea.
844 See also "The special problem of time updates" in the "ev_timer"
846 section.
847 ev_suspend (loop)
849 ev_resume (loop)
850 These two functions suspend and resume an event loop, for use when
851 the loop is not used for a while and timeouts should not be
852 processed.
853 A typical use case would be an interactive program such as a game:
855 When the user presses "^Z" to suspend the game and resumes it an
856 hour later it would be best to handle timeouts as if no time had
857 actually passed while the program was suspended. This can be
858 achieved by calling "ev_suspend" in your "SIGTSTP" handler, sending
859 yourself a "SIGSTOP" and calling "ev_resume" directly afterwards to
860 resume timer processing.
861 Effectively, all "ev_timer" watchers will be delayed by the time
863 spend between "ev_suspend" and "ev_resume", and all "ev_periodic"
864 watchers will be rescheduled (that is, they will lose any events
865 that would have occurred while suspended).
866 After calling "ev_suspend" you must not call any function on the
868 given loop other than "ev_resume", and you must not call
869 "ev_resume" without a previous call to "ev_suspend".
870 Calling "ev_suspend"/"ev_resume" has the side effect of updating
872 the event loop time (see "ev_now_update").
873 bool ev_run (loop, int flags)
875 Finally, this is it, the event handler. This function usually is
876 called after you have initialised all your watchers and you want to
877 start handling events. It will ask the operating system for any new
878 events, call the watcher callbacks, and then repeat the whole
879 process indefinitely: This is why event loops are called loops.
880 If the flags argument is specified as 0, it will keep handling
882 events until either no event watchers are active anymore or
883 "ev_break" was called.
884 The return value is false if there are no more active watchers
886 (which usually means "all jobs done" or "deadlock"), and true in
887 all other cases (which usually means " you should call "ev_run"
888 again").
889 Please note that an explicit "ev_break" is usually better than
891 relying on all watchers to be stopped when deciding when a program
892 has finished (especially in interactive programs), but having a
893 program that automatically loops as long as it has to and no longer
894 by virtue of relying on its watchers stopping correctly, that is
895 truly a thing of beauty.
896 This function is mostly exception-safe - you can break out of a
898 "ev_run" call by calling "longjmp" in a callback, throwing a C++
899 exception and so on. This does not decrement the "ev_depth" value,
900 nor will it clear any outstanding "EVBREAK_ONE" breaks.
901 A flags value of "EVRUN_NOWAIT" will look for new events, will
903 handle those events and any already outstanding ones, but will not
904 wait and block your process in case there are no events and will
905 return after one iteration of the loop. This is sometimes useful to
906 poll and handle new events while doing lengthy calculations, to
907 keep the program responsive.
908 A flags value of "EVRUN_ONCE" will look for new events (waiting if
910 necessary) and will handle those and any already outstanding ones.
911 It will block your process until at least one new event arrives
912 (which could be an event internal to libev itself, so there is no
913 guarantee that a user-registered callback will be called), and will
914 return after one iteration of the loop.
915 This is useful if you are waiting for some external event in
917 conjunction with something not expressible using other libev
918 watchers (i.e. "roll your own "ev_run""). However, a pair of
919 "ev_prepare"/"ev_check" watchers is usually a better approach for
920 this kind of thing.
921 Here are the gory details of what "ev_run" does (this is for your
923 understanding, not a guarantee that things will work exactly like
924 this in future versions):
925 - Increment loop depth.
927 - Reset the ev_break status.
928 - Before the first iteration, call any pending watchers.
929 LOOP:
930 - If EVFLAG_FORKCHECK was used, check for a fork.
931 - If a fork was detected (by any means), queue and call all fork watchers.
932 - Queue and call all prepare watchers.
933 - If ev_break was called, goto FINISH.
934 - If we have been forked, detach and recreate the kernel state
935 as to not disturb the other process.
936 - Update the kernel state with all outstanding changes.
937 - Update the "event loop time" (ev_now ()).
938 - Calculate for how long to sleep or block, if at all
939 (active idle watchers, EVRUN_NOWAIT or not having
940 any active watchers at all will result in not sleeping).
941 - Sleep if the I/O and timer collect interval say so.
942 - Increment loop iteration counter.
943 - Block the process, waiting for any events.
944 - Queue all outstanding I/O (fd) events.
945 - Update the "event loop time" (ev_now ()), and do time jump adjustments.
946 - Queue all expired timers.
947 - Queue all expired periodics.
948 - Queue all idle watchers with priority higher than that of pending events.
949 - Queue all check watchers.
950 - Call all queued watchers in reverse order (i.e. check watchers first).
951 Signals and child watchers are implemented as I/O watchers, and will
952 be handled here by queueing them when their watcher gets executed.
953 - If ev_break has been called, or EVRUN_ONCE or EVRUN_NOWAIT
954 were used, or there are no active watchers, goto FINISH, otherwise
955 continue with step LOOP.
956 FINISH:
957 - Reset the ev_break status iff it was EVBREAK_ONE.
958 - Decrement the loop depth.
959 - Return.
960 Example: Queue some jobs and then loop until no events are
962 outstanding anymore.
963 ... queue jobs here, make sure they register event watchers as long
965 ... as they still have work to do (even an idle watcher will do..)
966 ev_run (my_loop, 0);
967 ... jobs done or somebody called break. yeah!
968 ev_break (loop, how)
970 Can be used to make a call to "ev_run" return early (but only after
971 it has processed all outstanding events). The "how" argument must
972 be either "EVBREAK_ONE", which will make the innermost "ev_run"
973 call return, or "EVBREAK_ALL", which will make all nested "ev_run"
974 calls return.
975 This "break state" will be cleared on the next call to "ev_run".
977 It is safe to call "ev_break" from outside any "ev_run" calls, too,
979 in which case it will have no effect.
980 ev_ref (loop)
982 ev_unref (loop)
983 Ref/unref can be used to add or remove a reference count on the
984 event loop: Every watcher keeps one reference, and as long as the
985 reference count is nonzero, "ev_run" will not return on its own.
986 This is useful when you have a watcher that you never intend to
988 unregister, but that nevertheless should not keep "ev_run" from
989 returning. In such a case, call "ev_unref" after starting, and
990 "ev_ref" before stopping it.
991 As an example, libev itself uses this for its internal signal pipe:
993 It is not visible to the libev user and should not keep "ev_run"
994 from exiting if no event watchers registered by it are active. It
995 is also an excellent way to do this for generic recurring timers or
996 from within third-party libraries. Just remember to unref after
997 start and ref before stop (but only if the watcher wasn't active
998 before, or was active before, respectively. Note also that libev
999 might stop watchers itself (e.g. non-repeating timers) in which
1000 case you have to "ev_ref" in the callback).
1001 Example: Create a signal watcher, but keep it from keeping "ev_run"
1003 running when nothing else is active.
1004 ev_signal exitsig;
1006 ev_signal_init (&exitsig, sig_cb, SIGINT);
1007 ev_signal_start (loop, &exitsig);
1008 ev_unref (loop);
1009 Example: For some weird reason, unregister the above signal handler
1011 again.
1012 ev_ref (loop);
1014 ev_signal_stop (loop, &exitsig);
1015 ev_set_io_collect_interval (loop, ev_tstamp interval)
1017 ev_set_timeout_collect_interval (loop, ev_tstamp interval)
1018 These advanced functions influence the time that libev will spend
1019 waiting for events. Both time intervals are by default 0, meaning
1020 that libev will try to invoke timer/periodic callbacks and I/O
1021 callbacks with minimum latency.
1022 Setting these to a higher value (the "interval" must be >= 0)
1024 allows libev to delay invocation of I/O and timer/periodic
1025 callbacks to increase efficiency of loop iterations (or to increase
1026 power-saving opportunities).
1027 The idea is that sometimes your program runs just fast enough to
1029 handle one (or very few) event(s) per loop iteration. While this
1030 makes the program responsive, it also wastes a lot of CPU time to
1031 poll for new events, especially with backends like "select ()"
1032 which have a high overhead for the actual polling but can deliver
1033 many events at once.
1034 By setting a higher io collect interval you allow libev to spend
1036 more time collecting I/O events, so you can handle more events per
1037 iteration, at the cost of increasing latency. Timeouts (both
1038 "ev_periodic" and "ev_timer") will not be affected. Setting this to
1039 a non-null value will introduce an additional "ev_sleep ()" call
1040 into most loop iterations. The sleep time ensures that libev will
1041 not poll for I/O events more often then once per this interval, on
1042 average (as long as the host time resolution is good enough).
1043 Likewise, by setting a higher timeout collect interval you allow
1045 libev to spend more time collecting timeouts, at the expense of
1046 increased latency/jitter/inexactness (the watcher callback will be
1047 called later). "ev_io" watchers will not be affected. Setting this
1048 to a non-null value will not introduce any overhead in libev.
1049 Many (busy) programs can usually benefit by setting the I/O collect
1051 interval to a value near 0.1 or so, which is often enough for
1052 interactive servers (of course not for games), likewise for
1053 timeouts. It usually doesn't make much sense to set it to a lower
1054 value than 0.01, as this approaches the timing granularity of most
1055 systems. Note that if you do transactions with the outside world
1056 and you can't increase the parallelity, then this setting will
1057 limit your transaction rate (if you need to poll once per
1058 transaction and the I/O collect interval is 0.01, then you can't do
1059 more than 100 transactions per second).
1060 Setting the timeout collect interval can improve the opportunity
1062 for saving power, as the program will "bundle" timer callback
1063 invocations that are "near" in time together, by delaying some,
1064 thus reducing the number of times the process sleeps and wakes up
1065 again. Another useful technique to reduce iterations/wake-ups is to
1066 use "ev_periodic" watchers and make sure they fire on, say, one-
1067 second boundaries only.
1068 Example: we only need 0.1s timeout granularity, and we wish not to
1070 poll more often than 100 times per second:
1071 ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
1073 ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
1074 ev_invoke_pending (loop)
1076 This call will simply invoke all pending watchers while resetting
1077 their pending state. Normally, "ev_run" does this automatically
1078 when required, but when overriding the invoke callback this call
1079 comes handy. This function can be invoked from a watcher - this can
1080 be useful for example when you want to do some lengthy calculation
1081 and want to pass further event handling to another thread (you
1082 still have to make sure only one thread executes within
1083 "ev_invoke_pending" or "ev_run" of course).
1084 int ev_pending_count (loop)
1086 Returns the number of pending watchers - zero indicates that no
1087 watchers are pending.
1088 ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
1090 This overrides the invoke pending functionality of the loop:
1091 Instead of invoking all pending watchers when there are any,
1092 "ev_run" will call this callback instead. This is useful, for
1093 example, when you want to invoke the actual watchers inside another
1094 context (another thread etc.).
1095 If you want to reset the callback, use "ev_invoke_pending" as new
1097 callback.
1098 ev_set_loop_release_cb (loop, void (*release)(EV_P) throw (), void
1100 (*acquire)(EV_P) throw ())
1101 Sometimes you want to share the same loop between multiple threads.
1102 This can be done relatively simply by putting mutex_lock/unlock
1103 calls around each call to a libev function.
1104 However, "ev_run" can run an indefinite time, so it is not feasible
1106 to wait for it to return. One way around this is to wake up the
1107 event loop via "ev_break" and "ev_async_send", another way is to
1108 set these release and acquire callbacks on the loop.
1109 When set, then "release" will be called just before the thread is
1111 suspended waiting for new events, and "acquire" is called just
1112 afterwards.
1113 Ideally, "release" will just call your mutex_unlock function, and
1115 "acquire" will just call the mutex_lock function again.
1116 While event loop modifications are allowed between invocations of
1118 "release" and "acquire" (that's their only purpose after all), no
1119 modifications done will affect the event loop, i.e. adding watchers
1120 will have no effect on the set of file descriptors being watched,
1121 or the time waited. Use an "ev_async" watcher to wake up "ev_run"
1122 when you want it to take note of any changes you made.
1123 In theory, threads executing "ev_run" will be async-cancel safe
1125 between invocations of "release" and "acquire".
1126 See also the locking example in the "THREADS" section later in this
1128 document.
1129 ev_set_userdata (loop, void *data)
1131 void *ev_userdata (loop)
1132 Set and retrieve a single "void *" associated with a loop. When
1133 "ev_set_userdata" has never been called, then "ev_userdata" returns
1134 0.
1135 These two functions can be used to associate arbitrary data with a
1137 loop, and are intended solely for the "invoke_pending_cb",
1138 "release" and "acquire" callbacks described above, but of course
1139 can be (ab-)used for any other purpose as well.
1140 ev_verify (loop)
1142 This function only does something when "EV_VERIFY" support has been
1143 compiled in, which is the default for non-minimal builds. It tries
1144 to go through all internal structures and checks them for validity.
1145 If anything is found to be inconsistent, it will print an error
1146 message to standard error and call "abort ()".
1147 This can be used to catch bugs inside libev itself: under normal
1149 circumstances, this function will never abort as of course libev
1150 keeps its data structures consistent.
1151
1153 In the following description, uppercase "TYPE" in names stands for the
1154 watcher type, e.g. "ev_TYPE_start" can mean "ev_timer_start" for timer
1155 watchers and "ev_io_start" for I/O watchers.
1156 A watcher is an opaque structure that you allocate and register to
1158 record your interest in some event. To make a concrete example, imagine
1159 you want to wait for STDIN to become readable, you would create an
1160 "ev_io" watcher for that:
1161 static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
1163 {
1164 ev_io_stop (w);
1165 ev_break (loop, EVBREAK_ALL);
1166 }
1167 struct ev_loop *loop = ev_default_loop (0);
1169 ev_io stdin_watcher;
1171 ev_init (&stdin_watcher, my_cb);
1173 ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
1174 ev_io_start (loop, &stdin_watcher);
1175 ev_run (loop, 0);
1177 As you can see, you are responsible for allocating the memory for your
1179 watcher structures (and it is usually a bad idea to do this on the
1180 stack).
1181 Each watcher has an associated watcher structure (called "struct
1183 ev_TYPE" or simply "ev_TYPE", as typedefs are provided for all watcher
1184 structs).
1185 Each watcher structure must be initialised by a call to "ev_init
1187 (watcher *, callback)", which expects a callback to be provided. This
1188 callback is invoked each time the event occurs (or, in the case of I/O
1189 watchers, each time the event loop detects that the file descriptor
1190 given is readable and/or writable).
1191 Each watcher type further has its own "ev_TYPE_set (watcher *, ...)"
1193 macro to configure it, with arguments specific to the watcher type.
1194 There is also a macro to combine initialisation and setting in one
1195 call: "ev_TYPE_init (watcher *, callback, ...)".
1196 To make the watcher actually watch out for events, you have to start it
1198 with a watcher-specific start function ("ev_TYPE_start (loop, watcher
1199 *)"), and you can stop watching for events at any time by calling the
1200 corresponding stop function ("ev_TYPE_stop (loop, watcher *)".
1201 As long as your watcher is active (has been started but not stopped)
1203 you must not touch the values stored in it. Most specifically you must
1204 never reinitialise it or call its "ev_TYPE_set" macro.
1205 Each and every callback receives the event loop pointer as first, the
1207 registered watcher structure as second, and a bitset of received events
1208 as third argument.
1209 The received events usually include a single bit per event type
1211 received (you can receive multiple events at the same time). The
1212 possible bit masks are:
1213 "EV_READ"
1215 "EV_WRITE"
1216 The file descriptor in the "ev_io" watcher has become readable
1217 and/or writable.
1218 "EV_TIMER"
1220 The "ev_timer" watcher has timed out.
1221 "EV_PERIODIC"
1223 The "ev_periodic" watcher has timed out.
1224 "EV_SIGNAL"
1226 The signal specified in the "ev_signal" watcher has been received
1227 by a thread.
1228 "EV_CHILD"
1230 The pid specified in the "ev_child" watcher has received a status
1231 change.
1232 "EV_STAT"
1234 The path specified in the "ev_stat" watcher changed its attributes
1235 somehow.
1236 "EV_IDLE"
1238 The "ev_idle" watcher has determined that you have nothing better
1239 to do.
1240 "EV_PREPARE"
1242 "EV_CHECK"
1243 All "ev_prepare" watchers are invoked just before "ev_run" starts
1244 to gather new events, and all "ev_check" watchers are queued (not
1245 invoked) just after "ev_run" has gathered them, but before it
1246 queues any callbacks for any received events. That means
1247 "ev_prepare" watchers are the last watchers invoked before the
1248 event loop sleeps or polls for new events, and "ev_check" watchers
1249 will be invoked before any other watchers of the same or lower
1250 priority within an event loop iteration.
1251 Callbacks of both watcher types can start and stop as many watchers
1253 as they want, and all of them will be taken into account (for
1254 example, a "ev_prepare" watcher might start an idle watcher to keep
1255 "ev_run" from blocking).
1256 "EV_EMBED"
1258 The embedded event loop specified in the "ev_embed" watcher needs
1259 attention.
1260 "EV_FORK"
1262 The event loop has been resumed in the child process after fork
1263 (see "ev_fork").
1264 "EV_CLEANUP"
1266 The event loop is about to be destroyed (see "ev_cleanup").
1267 "EV_ASYNC"
1269 The given async watcher has been asynchronously notified (see
1270 "ev_async").
1271 "EV_CUSTOM"
1273 Not ever sent (or otherwise used) by libev itself, but can be
1274 freely used by libev users to signal watchers (e.g. via
1275 "ev_feed_event").
1276 "EV_ERROR"
1278 An unspecified error has occurred, the watcher has been stopped.
1279 This might happen because the watcher could not be properly started
1280 because libev ran out of memory, a file descriptor was found to be
1281 closed or any other problem. Libev considers these application
1282 bugs.
1283 You best act on it by reporting the problem and somehow coping with
1285 the watcher being stopped. Note that well-written programs should
1286 not receive an error ever, so when your watcher receives it, this
1287 usually indicates a bug in your program.
1288 Libev will usually signal a few "dummy" events together with an
1290 error, for example it might indicate that a fd is readable or
1291 writable, and if your callbacks is well-written it can just attempt
1292 the operation and cope with the error from read() or write(). This
1293 will not work in multi-threaded programs, though, as the fd could
1294 already be closed and reused for another thing, so beware.
1295 GENERIC WATCHER FUNCTIONS
1297 "ev_init" (ev_TYPE *watcher, callback)
1298 This macro initialises the generic portion of a watcher. The
1299 contents of the watcher object can be arbitrary (so "malloc" will
1300 do). Only the generic parts of the watcher are initialised, you
1301 need to call the type-specific "ev_TYPE_set" macro afterwards to
1302 initialise the type-specific parts. For each type there is also a
1303 "ev_TYPE_init" macro which rolls both calls into one.
1304 You can reinitialise a watcher at any time as long as it has been
1306 stopped (or never started) and there are no pending events
1307 outstanding.
1308 The callback is always of type "void (*)(struct ev_loop *loop,
1310 ev_TYPE *watcher, int revents)".
1311 Example: Initialise an "ev_io" watcher in two steps.
1313 ev_io w;
1315 ev_init (&w, my_cb);
1316 ev_io_set (&w, STDIN_FILENO, EV_READ);
1317 "ev_TYPE_set" (ev_TYPE *watcher, [args])
1319 This macro initialises the type-specific parts of a watcher. You
1320 need to call "ev_init" at least once before you call this macro,
1321 but you can call "ev_TYPE_set" any number of times. You must not,
1322 however, call this macro on a watcher that is active (it can be
1323 pending, however, which is a difference to the "ev_init" macro).
1324 Although some watcher types do not have type-specific arguments
1326 (e.g. "ev_prepare") you still need to call its "set" macro.
1327 See "ev_init", above, for an example.
1329 "ev_TYPE_init" (ev_TYPE *watcher, callback, [args])
1331 This convenience macro rolls both "ev_init" and "ev_TYPE_set" macro
1332 calls into a single call. This is the most convenient method to
1333 initialise a watcher. The same limitations apply, of course.
1334 Example: Initialise and set an "ev_io" watcher in one step.
1336 ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
1338 "ev_TYPE_start" (loop, ev_TYPE *watcher)
1340 Starts (activates) the given watcher. Only active watchers will
1341 receive events. If the watcher is already active nothing will
1342 happen.
1343 Example: Start the "ev_io" watcher that is being abused as example
1345 in this whole section.
1346 ev_io_start (EV_DEFAULT_UC, &w);
1348 "ev_TYPE_stop" (loop, ev_TYPE *watcher)
1350 Stops the given watcher if active, and clears the pending status
1351 (whether the watcher was active or not).
1352 It is possible that stopped watchers are pending - for example,
1354 non-repeating timers are being stopped when they become pending -
1355 but calling "ev_TYPE_stop" ensures that the watcher is neither
1356 active nor pending. If you want to free or reuse the memory used by
1357 the watcher it is therefore a good idea to always call its
1358 "ev_TYPE_stop" function.
1359 bool ev_is_active (ev_TYPE *watcher)
1361 Returns a true value iff the watcher is active (i.e. it has been
1362 started and not yet been stopped). As long as a watcher is active
1363 you must not modify it.
1364 bool ev_is_pending (ev_TYPE *watcher)
1366 Returns a true value iff the watcher is pending, (i.e. it has
1367 outstanding events but its callback has not yet been invoked). As
1368 long as a watcher is pending (but not active) you must not call an
1369 init function on it (but "ev_TYPE_set" is safe), you must not
1370 change its priority, and you must make sure the watcher is
1371 available to libev (e.g. you cannot "free ()" it).
1372 callback ev_cb (ev_TYPE *watcher)
1374 Returns the callback currently set on the watcher.
1375 ev_set_cb (ev_TYPE *watcher, callback)
1377 Change the callback. You can change the callback at virtually any
1378 time (modulo threads).
1379 ev_set_priority (ev_TYPE *watcher, int priority)
1381 int ev_priority (ev_TYPE *watcher)
1382 Set and query the priority of the watcher. The priority is a small
1383 integer between "EV_MAXPRI" (default: 2) and "EV_MINPRI" (default:
1384 "-2"). Pending watchers with higher priority will be invoked before
1385 watchers with lower priority, but priority will not keep watchers
1386 from being executed (except for "ev_idle" watchers).
1387 If you need to suppress invocation when higher priority events are
1389 pending you need to look at "ev_idle" watchers, which provide this
1390 functionality.
1391 You must not change the priority of a watcher as long as it is
1393 active or pending.
1394 Setting a priority outside the range of "EV_MINPRI" to "EV_MAXPRI"
1396 is fine, as long as you do not mind that the priority value you
1397 query might or might not have been clamped to the valid range.
1398 The default priority used by watchers when no priority has been set
1400 is always 0, which is supposed to not be too high and not be too
1401 low :).
1402 See "WATCHER PRIORITY MODELS", below, for a more thorough treatment
1404 of priorities.
1405 ev_invoke (loop, ev_TYPE *watcher, int revents)
1407 Invoke the "watcher" with the given "loop" and "revents". Neither
1408 "loop" nor "revents" need to be valid as long as the watcher
1409 callback can deal with that fact, as both are simply passed through
1410 to the callback.
1411 int ev_clear_pending (loop, ev_TYPE *watcher)
1413 If the watcher is pending, this function clears its pending status
1414 and returns its "revents" bitset (as if its callback was invoked).
1415 If the watcher isn't pending it does nothing and returns 0.
1416 Sometimes it can be useful to "poll" a watcher instead of waiting
1418 for its callback to be invoked, which can be accomplished with this
1419 function.
1420 ev_feed_event (loop, ev_TYPE *watcher, int revents)
1422 Feeds the given event set into the event loop, as if the specified
1423 event had happened for the specified watcher (which must be a
1424 pointer to an initialised but not necessarily started event
1425 watcher). Obviously you must not free the watcher as long as it has
1426 pending events.
1427 Stopping the watcher, letting libev invoke it, or calling
1429 "ev_clear_pending" will clear the pending event, even if the
1430 watcher was not started in the first place.
1431 See also "ev_feed_fd_event" and "ev_feed_signal_event" for related
1433 functions that do not need a watcher.
1434 See also the "ASSOCIATING CUSTOM DATA WITH A WATCHER" and "BUILDING
1436 YOUR OWN COMPOSITE WATCHERS" idioms.
1437 WATCHER STATES
1439 There are various watcher states mentioned throughout this manual -
1440 active, pending and so on. In this section these states and the rules
1441 to transition between them will be described in more detail - and while
1442 these rules might look complicated, they usually do "the right thing".
1443 initialised
1445 Before a watcher can be registered with the event loop it has to be
1446 initialised. This can be done with a call to "ev_TYPE_init", or
1447 calls to "ev_init" followed by the watcher-specific "ev_TYPE_set"
1448 function.
1449 In this state it is simply some block of memory that is suitable
1451 for use in an event loop. It can be moved around, freed, reused
1452 etc. at will - as long as you either keep the memory contents
1453 intact, or call "ev_TYPE_init" again.
1454 started/running/active
1456 Once a watcher has been started with a call to "ev_TYPE_start" it
1457 becomes property of the event loop, and is actively waiting for
1458 events. While in this state it cannot be accessed (except in a few
1459 documented ways), moved, freed or anything else - the only legal
1460 thing is to keep a pointer to it, and call libev functions on it
1461 that are documented to work on active watchers.
1462 pending
1464 If a watcher is active and libev determines that an event it is
1465 interested in has occurred (such as a timer expiring), it will
1466 become pending. It will stay in this pending state until either it
1467 is stopped or its callback is about to be invoked, so it is not
1468 normally pending inside the watcher callback.
1469 The watcher might or might not be active while it is pending (for
1471 example, an expired non-repeating timer can be pending but no
1472 longer active). If it is stopped, it can be freely accessed (e.g.
1473 by calling "ev_TYPE_set"), but it is still property of the event
1474 loop at this time, so cannot be moved, freed or reused. And if it
1475 is active the rules described in the previous item still apply.
1476 It is also possible to feed an event on a watcher that is not
1478 active (e.g. via "ev_feed_event"), in which case it becomes
1479 pending without being active.
1480 stopped
1482 A watcher can be stopped implicitly by libev (in which case it
1483 might still be pending), or explicitly by calling its
1484 "ev_TYPE_stop" function. The latter will clear any pending state
1485 the watcher might be in, regardless of whether it was active or
1486 not, so stopping a watcher explicitly before freeing it is often a
1487 good idea.
1488 While stopped (and not pending) the watcher is essentially in the
1490 initialised state, that is, it can be reused, moved, modified in
1491 any way you wish (but when you trash the memory block, you need to
1492 "ev_TYPE_init" it again).
1493 WATCHER PRIORITY MODELS
1495 Many event loops support watcher priorities, which are usually small
1496 integers that influence the ordering of event callback invocation
1497 between watchers in some way, all else being equal.
1498 In libev, Watcher priorities can be set using "ev_set_priority". See
1500 its description for the more technical details such as the actual
1501 priority range.
1502 There are two common ways how these these priorities are being
1504 interpreted by event loops:
1505 In the more common lock-out model, higher priorities "lock out"
1507 invocation of lower priority watchers, which means as long as higher
1508 priority watchers receive events, lower priority watchers are not being
1509 invoked.
1510 The less common only-for-ordering model uses priorities solely to order
1512 callback invocation within a single event loop iteration: Higher
1513 priority watchers are invoked before lower priority ones, but they all
1514 get invoked before polling for new events.
1515 Libev uses the second (only-for-ordering) model for all its watchers
1517 except for idle watchers (which use the lock-out model).
1518 The rationale behind this is that implementing the lock-out model for
1520 watchers is not well supported by most kernel interfaces, and most
1521 event libraries will just poll for the same events again and again as
1522 long as their callbacks have not been executed, which is very
1523 inefficient in the common case of one high-priority watcher locking out
1524 a mass of lower priority ones.
1525 Static (ordering) priorities are most useful when you have two or more
1527 watchers handling the same resource: a typical usage example is having
1528 an "ev_io" watcher to receive data, and an associated "ev_timer" to
1529 handle timeouts. Under load, data might be received while the program
1530 handles other jobs, but since timers normally get invoked first, the
1531 timeout handler will be executed before checking for data. In that
1532 case, giving the timer a lower priority than the I/O watcher ensures
1533 that I/O will be handled first even under adverse conditions (which is
1534 usually, but not always, what you want).
1535 Since idle watchers use the "lock-out" model, meaning that idle
1537 watchers will only be executed when no same or higher priority watchers
1538 have received events, they can be used to implement the "lock-out"
1539 model when required.
1540 For example, to emulate how many other event libraries handle
1542 priorities, you can associate an "ev_idle" watcher to each such
1543 watcher, and in the normal watcher callback, you just start the idle
1544 watcher. The real processing is done in the idle watcher callback. This
1545 causes libev to continuously poll and process kernel event data for the
1546 watcher, but when the lock-out case is known to be rare (which in turn
1547 is rare :), this is workable.
1548 Usually, however, the lock-out model implemented that way will perform
1550 miserably under the type of load it was designed to handle. In that
1551 case, it might be preferable to stop the real watcher before starting
1552 the idle watcher, so the kernel will not have to process the event in
1553 case the actual processing will be delayed for considerable time.
1554 Here is an example of an I/O watcher that should run at a strictly
1556 lower priority than the default, and which should only process data
1557 when no other events are pending:
1558 ev_idle idle; // actual processing watcher
1560 ev_io io; // actual event watcher
1561 static void
1563 io_cb (EV_P_ ev_io *w, int revents)
1564 {
1565 // stop the I/O watcher, we received the event, but
1566 // are not yet ready to handle it.
1567 ev_io_stop (EV_A_ w);
1568 // start the idle watcher to handle the actual event.
1570 // it will not be executed as long as other watchers
1571 // with the default priority are receiving events.
1572 ev_idle_start (EV_A_ &idle);
1573 }
1574 static void
1576 idle_cb (EV_P_ ev_idle *w, int revents)
1577 {
1578 // actual processing
1579 read (STDIN_FILENO, ...);
1580 // have to start the I/O watcher again, as
1582 // we have handled the event
1583 ev_io_start (EV_P_ &io);
1584 }
1585 // initialisation
1587 ev_idle_init (&idle, idle_cb);
1588 ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
1589 ev_io_start (EV_DEFAULT_ &io);
1590 In the "real" world, it might also be beneficial to start a timer, so
1592 that low-priority connections can not be locked out forever under load.
1593 This enables your program to keep a lower latency for important
1594 connections during short periods of high load, while not completely
1595 locking out less important ones.
1596
1598 This section describes each watcher in detail, but will not repeat
1599 information given in the last section. Any initialisation/set macros,
1600 functions and members specific to the watcher type are explained.
1601 Members are additionally marked with either [read-only], meaning that,
1603 while the watcher is active, you can look at the member and expect some
1604 sensible content, but you must not modify it (you can modify it while
1605 the watcher is stopped to your hearts content), or [read-write], which
1606 means you can expect it to have some sensible content while the watcher
1607 is active, but you can also modify it. Modifying it may not do
1608 something sensible or take immediate effect (or do anything at all),
1609 but libev will not crash or malfunction in any way.
1610 "ev_io" - is this file descriptor readable or writable?
1612 I/O watchers check whether a file descriptor is readable or writable in
1613 each iteration of the event loop, or, more precisely, when reading
1614 would not block the process and writing would at least be able to write
1615 some data. This behaviour is called level-triggering because you keep
1616 receiving events as long as the condition persists. Remember you can
1617 stop the watcher if you don't want to act on the event and neither want
1618 to receive future events.
1619 In general you can register as many read and/or write event watchers
1621 per fd as you want (as long as you don't confuse yourself). Setting all
1622 file descriptors to non-blocking mode is also usually a good idea (but
1623 not required if you know what you are doing).
1624 Another thing you have to watch out for is that it is quite easy to
1626 receive "spurious" readiness notifications, that is, your callback
1627 might be called with "EV_READ" but a subsequent "read"(2) will actually
1628 block because there is no data. It is very easy to get into this
1629 situation even with a relatively standard program structure. Thus it is
1630 best to always use non-blocking I/O: An extra "read"(2) returning
1631 "EAGAIN" is far preferable to a program hanging until some data
1632 arrives.
1633 If you cannot run the fd in non-blocking mode (for example you should
1635 not play around with an Xlib connection), then you have to separately
1636 re-test whether a file descriptor is really ready with a known-to-be
1637 good interface such as poll (fortunately in the case of Xlib, it
1638 already does this on its own, so its quite safe to use). Some people
1639 additionally use "SIGALRM" and an interval timer, just to be sure you
1640 won't block indefinitely.
1641 But really, best use non-blocking mode.
1643 The special problem of disappearing file descriptors
1645 Some backends (e.g. kqueue, epoll, linuxaio) need to be told about
1647 closing a file descriptor (either due to calling "close" explicitly or
1648 any other means, such as "dup2"). The reason is that you register
1649 interest in some file descriptor, but when it goes away, the operating
1650 system will silently drop this interest. If another file descriptor
1651 with the same number then is registered with libev, there is no
1652 efficient way to see that this is, in fact, a different file
1653 descriptor.
1654 To avoid having to explicitly tell libev about such cases, libev
1656 follows the following policy: Each time "ev_io_set" is being called,
1657 libev will assume that this is potentially a new file descriptor,
1658 otherwise it is assumed that the file descriptor stays the same. That
1659 means that you have to call "ev_io_set" (or "ev_io_init") when you
1660 change the descriptor even if the file descriptor number itself did not
1661 change.
1662 This is how one would do it normally anyway, the important point is
1664 that the libev application should not optimise around libev but should
1665 leave optimisations to libev.
1666 The special problem of dup'ed file descriptors
1668 Some backends (e.g. epoll), cannot register events for file
1670 descriptors, but only events for the underlying file descriptions. That
1671 means when you have "dup ()"'ed file descriptors or weirder
1672 constellations, and register events for them, only one file descriptor
1673 might actually receive events.
1674 There is no workaround possible except not registering events for
1676 potentially "dup ()"'ed file descriptors, or to resort to
1677 "EVBACKEND_SELECT" or "EVBACKEND_POLL".
1678 The special problem of files
1680 Many people try to use "select" (or libev) on file descriptors
1682 representing files, and expect it to become ready when their program
1683 doesn't block on disk accesses (which can take a long time on their
1684 own).
1685 However, this cannot ever work in the "expected" way - you get a
1687 readiness notification as soon as the kernel knows whether and how much
1688 data is there, and in the case of open files, that's always the case,
1689 so you always get a readiness notification instantly, and your read (or
1690 possibly write) will still block on the disk I/O.
1691 Another way to view it is that in the case of sockets, pipes, character
1693 devices and so on, there is another party (the sender) that delivers
1694 data on its own, but in the case of files, there is no such thing: the
1695 disk will not send data on its own, simply because it doesn't know what
1696 you wish to read - you would first have to request some data.
1697 Since files are typically not-so-well supported by advanced
1699 notification mechanism, libev tries hard to emulate POSIX behaviour
1700 with respect to files, even though you should not use it. The reason
1701 for this is convenience: sometimes you want to watch STDIN or STDOUT,
1702 which is usually a tty, often a pipe, but also sometimes files or
1703 special devices (for example, "epoll" on Linux works with /dev/random
1704 but not with /dev/urandom), and even though the file might better be
1705 served with asynchronous I/O instead of with non-blocking I/O, it is
1706 still useful when it "just works" instead of freezing.
1707 So avoid file descriptors pointing to files when you know it (e.g. use
1709 libeio), but use them when it is convenient, e.g. for STDIN/STDOUT, or
1710 when you rarely read from a file instead of from a socket, and want to
1711 reuse the same code path.
1712 The special problem of fork
1714 Some backends (epoll, kqueue, probably linuxaio) do not support "fork
1716 ()" at all or exhibit useless behaviour. Libev fully supports fork, but
1717 needs to be told about it in the child if you want to continue to use
1718 it in the child.
1719 To support fork in your child processes, you have to call "ev_loop_fork
1721 ()" after a fork in the child, enable "EVFLAG_FORKCHECK", or resort to
1722 "EVBACKEND_SELECT" or "EVBACKEND_POLL".
1723 The special problem of SIGPIPE
1725 While not really specific to libev, it is easy to forget about
1727 "SIGPIPE": when writing to a pipe whose other end has been closed, your
1728 program gets sent a SIGPIPE, which, by default, aborts your program.
1729 For most programs this is sensible behaviour, for daemons, this is
1730 usually undesirable.
1731 So when you encounter spurious, unexplained daemon exits, make sure you
1733 ignore SIGPIPE (and maybe make sure you log the exit status of your
1734 daemon somewhere, as that would have given you a big clue).
1735 The special problem of accept()ing when you can't
1737 Many implementations of the POSIX "accept" function (for example, found
1739 in post-2004 Linux) have the peculiar behaviour of not removing a
1740 connection from the pending queue in all error cases.
1741 For example, larger servers often run out of file descriptors (because
1743 of resource limits), causing "accept" to fail with "ENFILE" but not
1744 rejecting the connection, leading to libev signalling readiness on the
1745 next iteration again (the connection still exists after all), and
1746 typically causing the program to loop at 100% CPU usage.
1747 Unfortunately, the set of errors that cause this issue differs between
1749 operating systems, there is usually little the app can do to remedy the
1750 situation, and no known thread-safe method of removing the connection
1751 to cope with overload is known (to me).
1752 One of the easiest ways to handle this situation is to just ignore it -
1754 when the program encounters an overload, it will just loop until the
1755 situation is over. While this is a form of busy waiting, no OS offers
1756 an event-based way to handle this situation, so it's the best one can
1757 do.
1758 A better way to handle the situation is to log any errors other than
1760 "EAGAIN" and "EWOULDBLOCK", making sure not to flood the log with such
1761 messages, and continue as usual, which at least gives the user an idea
1762 of what could be wrong ("raise the ulimit!"). For extra points one
1763 could stop the "ev_io" watcher on the listening fd "for a while", which
1764 reduces CPU usage.
1765 If your program is single-threaded, then you could also keep a dummy
1767 file descriptor for overload situations (e.g. by opening /dev/null),
1768 and when you run into "ENFILE" or "EMFILE", close it, run "accept",
1769 close that fd, and create a new dummy fd. This will gracefully refuse
1770 clients under typical overload conditions.
1771 The last way to handle it is to simply log the error and "exit", as is
1773 often done with "malloc" failures, but this results in an easy
1774 opportunity for a DoS attack.
1775 Watcher-Specific Functions
1777 ev_io_init (ev_io *, callback, int fd, int events)
1779 ev_io_set (ev_io *, int fd, int events)
1780 Configures an "ev_io" watcher. The "fd" is the file descriptor to
1781 receive events for and "events" is either "EV_READ", "EV_WRITE" or
1782 "EV_READ | EV_WRITE", to express the desire to receive the given
1783 events.
1784 int fd [read-only]
1786 The file descriptor being watched.
1787 int events [read-only]
1789 The events being watched.
1790 Examples
1792 Example: Call "stdin_readable_cb" when STDIN_FILENO has become, well
1794 readable, but only once. Since it is likely line-buffered, you could
1795 attempt to read a whole line in the callback.
1796 static void
1798 stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
1799 {
1800 ev_io_stop (loop, w);
1801 .. read from stdin here (or from w->fd) and handle any I/O errors
1802 }
1803 ...
1805 struct ev_loop *loop = ev_default_init (0);
1806 ev_io stdin_readable;
1807 ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
1808 ev_io_start (loop, &stdin_readable);
1809 ev_run (loop, 0);
1810 "ev_timer" - relative and optionally repeating timeouts
1812 Timer watchers are simple relative timers that generate an event after
1813 a given time, and optionally repeating in regular intervals after that.
1814 The timers are based on real time, that is, if you register an event
1816 that times out after an hour and you reset your system clock to January
1817 last year, it will still time out after (roughly) one hour. "Roughly"
1818 because detecting time jumps is hard, and some inaccuracies are
1819 unavoidable (the monotonic clock option helps a lot here).
1820 The callback is guaranteed to be invoked only after its timeout has
1822 passed (not at, so on systems with very low-resolution clocks this
1823 might introduce a small delay, see "the special problem of being too
1824 early", below). If multiple timers become ready during the same loop
1825 iteration then the ones with earlier time-out values are invoked before
1826 ones of the same priority with later time-out values (but this is no
1827 longer true when a callback calls "ev_run" recursively).
1828 Be smart about timeouts
1830 Many real-world problems involve some kind of timeout, usually for
1832 error recovery. A typical example is an HTTP request - if the other
1833 side hangs, you want to raise some error after a while.
1834 What follows are some ways to handle this problem, from obvious and
1836 inefficient to smart and efficient.
1837 In the following, a 60 second activity timeout is assumed - a timeout
1839 that gets reset to 60 seconds each time there is activity (e.g. each
1840 time some data or other life sign was received).
1841 1. Use a timer and stop, reinitialise and start it on activity.
1843 This is the most obvious, but not the most simple way: In the
1844 beginning, start the watcher:
1845 ev_timer_init (timer, callback, 60., 0.);
1847 ev_timer_start (loop, timer);
1848 Then, each time there is some activity, "ev_timer_stop" it,
1850 initialise it and start it again:
1851 ev_timer_stop (loop, timer);
1853 ev_timer_set (timer, 60., 0.);
1854 ev_timer_start (loop, timer);
1855 This is relatively simple to implement, but means that each time
1857 there is some activity, libev will first have to remove the timer
1858 from its internal data structure and then add it again. Libev tries
1859 to be fast, but it's still not a constant-time operation.
1860 2. Use a timer and re-start it with "ev_timer_again" inactivity.
1862 This is the easiest way, and involves using "ev_timer_again"
1863 instead of "ev_timer_start".
1864 To implement this, configure an "ev_timer" with a "repeat" value of
1866 60 and then call "ev_timer_again" at start and each time you
1867 successfully read or write some data. If you go into an idle state
1868 where you do not expect data to travel on the socket, you can
1869 "ev_timer_stop" the timer, and "ev_timer_again" will automatically
1870 restart it if need be.
1871 That means you can ignore both the "ev_timer_start" function and
1873 the "after" argument to "ev_timer_set", and only ever use the
1874 "repeat" member and "ev_timer_again".
1875 At start:
1877 ev_init (timer, callback);
1879 timer->repeat = 60.;
1880 ev_timer_again (loop, timer);
1881 Each time there is some activity:
1883 ev_timer_again (loop, timer);
1885 It is even possible to change the time-out on the fly, regardless
1887 of whether the watcher is active or not:
1888 timer->repeat = 30.;
1890 ev_timer_again (loop, timer);
1891 This is slightly more efficient then stopping/starting the timer
1893 each time you want to modify its timeout value, as libev does not
1894 have to completely remove and re-insert the timer from/into its
1895 internal data structure.
1896 It is, however, even simpler than the "obvious" way to do it.
1898 3. Let the timer time out, but then re-arm it as required.
1900 This method is more tricky, but usually most efficient: Most
1901 timeouts are relatively long compared to the intervals between
1902 other activity - in our example, within 60 seconds, there are
1903 usually many I/O events with associated activity resets.
1904 In this case, it would be more efficient to leave the "ev_timer"
1906 alone, but remember the time of last activity, and check for a real
1907 timeout only within the callback:
1908 ev_tstamp timeout = 60.;
1910 ev_tstamp last_activity; // time of last activity
1911 ev_timer timer;
1912 static void
1914 callback (EV_P_ ev_timer *w, int revents)
1915 {
1916 // calculate when the timeout would happen
1917 ev_tstamp after = last_activity - ev_now (EV_A) + timeout;
1918 // if negative, it means we the timeout already occurred
1920 if (after < 0.)
1921 {
1922 // timeout occurred, take action
1923 }
1924 else
1925 {
1926 // callback was invoked, but there was some recent
1927 // activity. simply restart the timer to time out
1928 // after "after" seconds, which is the earliest time
1929 // the timeout can occur.
1930 ev_timer_set (w, after, 0.);
1931 ev_timer_start (EV_A_ w);
1932 }
1933 }
1934 To summarise the callback: first calculate in how many seconds the
1936 timeout will occur (by calculating the absolute time when it would
1937 occur, "last_activity + timeout", and subtracting the current time,
1938 "ev_now (EV_A)" from that).
1939 If this value is negative, then we are already past the timeout,
1941 i.e. we timed out, and need to do whatever is needed in this case.
1942 Otherwise, we now the earliest time at which the timeout would
1944 trigger, and simply start the timer with this timeout value.
1945 In other words, each time the callback is invoked it will check
1947 whether the timeout occurred. If not, it will simply reschedule
1948 itself to check again at the earliest time it could time out.
1949 Rinse. Repeat.
1950 This scheme causes more callback invocations (about one every 60
1952 seconds minus half the average time between activity), but
1953 virtually no calls to libev to change the timeout.
1954 To start the machinery, simply initialise the watcher and set
1956 "last_activity" to the current time (meaning there was some
1957 activity just now), then call the callback, which will "do the
1958 right thing" and start the timer:
1959 last_activity = ev_now (EV_A);
1961 ev_init (&timer, callback);
1962 callback (EV_A_ &timer, 0);
1963 When there is some activity, simply store the current time in
1965 "last_activity", no libev calls at all:
1966 if (activity detected)
1968 last_activity = ev_now (EV_A);
1969 When your timeout value changes, then the timeout can be changed by
1971 simply providing a new value, stopping the timer and calling the
1972 callback, which will again do the right thing (for example, time
1973 out immediately :).
1974 timeout = new_value;
1976 ev_timer_stop (EV_A_ &timer);
1977 callback (EV_A_ &timer, 0);
1978 This technique is slightly more complex, but in most cases where
1980 the time-out is unlikely to be triggered, much more efficient.
1981 4. Wee, just use a double-linked list for your timeouts.
1983 If there is not one request, but many thousands (millions...), all
1984 employing some kind of timeout with the same timeout value, then
1985 one can do even better:
1986 When starting the timeout, calculate the timeout value and put the
1988 timeout at the end of the list.
1989 Then use an "ev_timer" to fire when the timeout at the beginning of
1991 the list is expected to fire (for example, using the technique #3).
1992 When there is some activity, remove the timer from the list,
1994 recalculate the timeout, append it to the end of the list again,
1995 and make sure to update the "ev_timer" if it was taken from the
1996 beginning of the list.
1997 This way, one can manage an unlimited number of timeouts in O(1)
1999 time for starting, stopping and updating the timers, at the expense
2000 of a major complication, and having to use a constant timeout. The
2001 constant timeout ensures that the list stays sorted.
2002 So which method the best?
2004 Method #2 is a simple no-brain-required solution that is adequate in
2006 most situations. Method #3 requires a bit more thinking, but handles
2007 many cases better, and isn't very complicated either. In most case,
2008 choosing either one is fine, with #3 being better in typical
2009 situations.
2010 Method #1 is almost always a bad idea, and buys you nothing. Method #4
2012 is rather complicated, but extremely efficient, something that really
2013 pays off after the first million or so of active timers, i.e. it's
2014 usually overkill :)
2015 The special problem of being too early
2017 If you ask a timer to call your callback after three seconds, then you
2019 expect it to be invoked after three seconds - but of course, this
2020 cannot be guaranteed to infinite precision. Less obviously, it cannot
2021 be guaranteed to any precision by libev - imagine somebody suspending
2022 the process with a STOP signal for a few hours for example.
2023 So, libev tries to invoke your callback as soon as possible after the
2025 delay has occurred, but cannot guarantee this.
2026 A less obvious failure mode is calling your callback too early: many
2028 event loops compare timestamps with a "elapsed delay >= requested
2029 delay", but this can cause your callback to be invoked much earlier
2030 than you would expect.
2031 To see why, imagine a system with a clock that only offers full second
2033 resolution (think windows if you can't come up with a broken enough OS
2034 yourself). If you schedule a one-second timer at the time 500.9, then
2035 the event loop will schedule your timeout to elapse at a system time of
2036 500 (500.9 truncated to the resolution) + 1, or 501.
2037 If an event library looks at the timeout 0.1s later, it will see "501
2039 >= 501" and invoke the callback 0.1s after it was started, even though
2040 a one-second delay was requested - this is being "too early", despite
2041 best intentions.
2042 This is the reason why libev will never invoke the callback if the
2044 elapsed delay equals the requested delay, but only when the elapsed
2045 delay is larger than the requested delay. In the example above, libev
2046 would only invoke the callback at system time 502, or 1.1s after the
2047 timer was started.
2048 So, while libev cannot guarantee that your callback will be invoked
2050 exactly when requested, it can and does guarantee that the requested
2051 delay has actually elapsed, or in other words, it always errs on the
2052 "too late" side of things.
2053 The special problem of time updates
2055 Establishing the current time is a costly operation (it usually takes
2057 at least one system call): EV therefore updates its idea of the current
2058 time only before and after "ev_run" collects new events, which causes a
2059 growing difference between "ev_now ()" and "ev_time ()" when handling
2060 lots of events in one iteration.
2061 The relative timeouts are calculated relative to the "ev_now ()" time.
2063 This is usually the right thing as this timestamp refers to the time of
2064 the event triggering whatever timeout you are modifying/starting. If
2065 you suspect event processing to be delayed and you need to base the
2066 timeout on the current time, use something like the following to adjust
2067 for it:
2068 ev_timer_set (&timer, after + (ev_time () - ev_now ()), 0.);
2070 If the event loop is suspended for a long time, you can also force an
2072 update of the time returned by "ev_now ()" by calling "ev_now_update
2073 ()", although that will push the event time of all outstanding events
2074 further into the future.
2075 The special problem of unsynchronised clocks
2077 Modern systems have a variety of clocks - libev itself uses the normal
2079 "wall clock" clock and, if available, the monotonic clock (to avoid
2080 time jumps).
2081 Neither of these clocks is synchronised with each other or any other
2083 clock on the system, so "ev_time ()" might return a considerably
2084 different time than "gettimeofday ()" or "time ()". On a GNU/Linux
2085 system, for example, a call to "gettimeofday" might return a second
2086 count that is one higher than a directly following call to "time".
2087 The moral of this is to only compare libev-related timestamps with
2089 "ev_time ()" and "ev_now ()", at least if you want better precision
2090 than a second or so.
2091 One more problem arises due to this lack of synchronisation: if libev
2093 uses the system monotonic clock and you compare timestamps from
2094 "ev_time" or "ev_now" from when you started your timer and when your
2095 callback is invoked, you will find that sometimes the callback is a bit
2096 "early".
2097 This is because "ev_timer"s work in real time, not wall clock time, so
2099 libev makes sure your callback is not invoked before the delay
2100 happened, measured according to the real time, not the system clock.
2101 If your timeouts are based on a physical timescale (e.g. "time out this
2103 connection after 100 seconds") then this shouldn't bother you as it is
2104 exactly the right behaviour.
2105 If you want to compare wall clock/system timestamps to your timers,
2107 then you need to use "ev_periodic"s, as these are based on the wall
2108 clock time, where your comparisons will always generate correct
2109 results.
2110 The special problems of suspended animation
2112 When you leave the server world it is quite customary to hit machines
2114 that can suspend/hibernate - what happens to the clocks during such a
2115 suspend?
2116 Some quick tests made with a Linux 2.6.28 indicate that a suspend
2118 freezes all processes, while the clocks ("times", "CLOCK_MONOTONIC")
2119 continue to run until the system is suspended, but they will not
2120 advance while the system is suspended. That means, on resume, it will
2121 be as if the program was frozen for a few seconds, but the suspend time
2122 will not be counted towards "ev_timer" when a monotonic clock source is
2123 used. The real time clock advanced as expected, but if it is used as
2124 sole clocksource, then a long suspend would be detected as a time jump
2125 by libev, and timers would be adjusted accordingly.
2126 I would not be surprised to see different behaviour in different
2128 between operating systems, OS versions or even different hardware.
2129 The other form of suspend (job control, or sending a SIGSTOP) will see
2131 a time jump in the monotonic clocks and the realtime clock. If the
2132 program is suspended for a very long time, and monotonic clock sources
2133 are in use, then you can expect "ev_timer"s to expire as the full
2134 suspension time will be counted towards the timers. When no monotonic
2135 clock source is in use, then libev will again assume a timejump and
2136 adjust accordingly.
2137 It might be beneficial for this latter case to call "ev_suspend" and
2139 "ev_resume" in code that handles "SIGTSTP", to at least get
2140 deterministic behaviour in this case (you can do nothing against
2141 "SIGSTOP").
2142 Watcher-Specific Functions and Data Members
2144 ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
2146 ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
2147 Configure the timer to trigger after "after" seconds (fractional
2148 and negative values are supported). If "repeat" is 0., then it will
2149 automatically be stopped once the timeout is reached. If it is
2150 positive, then the timer will automatically be configured to
2151 trigger again "repeat" seconds later, again, and again, until
2152 stopped manually.
2153 The timer itself will do a best-effort at avoiding drift, that is,
2155 if you configure a timer to trigger every 10 seconds, then it will
2156 normally trigger at exactly 10 second intervals. If, however, your
2157 program cannot keep up with the timer (because it takes longer than
2158 those 10 seconds to do stuff) the timer will not fire more than
2159 once per event loop iteration.
2160 ev_timer_again (loop, ev_timer *)
2162 This will act as if the timer timed out, and restarts it again if
2163 it is repeating. It basically works like calling "ev_timer_stop",
2164 updating the timeout to the "repeat" value and calling
2165 "ev_timer_start".
2166 The exact semantics are as in the following rules, all of which
2168 will be applied to the watcher:
2169 If the timer is pending, the pending status is always cleared.
2171 If the timer is started but non-repeating, stop it (as if it timed
2172 out, without invoking it).
2173 If the timer is repeating, make the "repeat" value the new timeout
2174 and start the timer, if necessary.
2175 This sounds a bit complicated, see "Be smart about timeouts",
2177 above, for a usage example.
2178 ev_tstamp ev_timer_remaining (loop, ev_timer *)
2180 Returns the remaining time until a timer fires. If the timer is
2181 active, then this time is relative to the current event loop time,
2182 otherwise it's the timeout value currently configured.
2183 That is, after an "ev_timer_set (w, 5, 7)", "ev_timer_remaining"
2185 returns 5. When the timer is started and one second passes,
2186 "ev_timer_remaining" will return 4. When the timer expires and is
2187 restarted, it will return roughly 7 (likely slightly less as
2188 callback invocation takes some time, too), and so on.
2189 ev_tstamp repeat [read-write]
2191 The current "repeat" value. Will be used each time the watcher
2192 times out or "ev_timer_again" is called, and determines the next
2193 timeout (if any), which is also when any modifications are taken
2194 into account.
2195 Examples
2197 Example: Create a timer that fires after 60 seconds.
2199 static void
2201 one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
2202 {
2203 .. one minute over, w is actually stopped right here
2204 }
2205 ev_timer mytimer;
2207 ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
2208 ev_timer_start (loop, &mytimer);
2209 Example: Create a timeout timer that times out after 10 seconds of
2211 inactivity.
2212 static void
2214 timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
2215 {
2216 .. ten seconds without any activity
2217 }
2218 ev_timer mytimer;
2220 ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
2221 ev_timer_again (&mytimer); /* start timer */
2222 ev_run (loop, 0);
2223 // and in some piece of code that gets executed on any "activity":
2225 // reset the timeout to start ticking again at 10 seconds
2226 ev_timer_again (&mytimer);
2227 "ev_periodic" - to cron or not to cron?
2229 Periodic watchers are also timers of a kind, but they are very
2230 versatile (and unfortunately a bit complex).
2231 Unlike "ev_timer", periodic watchers are not based on real time (or
2233 relative time, the physical time that passes) but on wall clock time
2234 (absolute time, the thing you can read on your calendar or clock). The
2235 difference is that wall clock time can run faster or slower than real
2236 time, and time jumps are not uncommon (e.g. when you adjust your wrist-
2237 watch).
2238 You can tell a periodic watcher to trigger after some specific point in
2240 time: for example, if you tell a periodic watcher to trigger "in 10
2241 seconds" (by specifying e.g. "ev_now () + 10.", that is, an absolute
2242 time not a delay) and then reset your system clock to January of the
2243 previous year, then it will take a year or more to trigger the event
2244 (unlike an "ev_timer", which would still trigger roughly 10 seconds
2245 after starting it, as it uses a relative timeout).
2246 "ev_periodic" watchers can also be used to implement vastly more
2248 complex timers, such as triggering an event on each "midnight, local
2249 time", or other complicated rules. This cannot easily be done with
2250 "ev_timer" watchers, as those cannot react to time jumps.
2251 As with timers, the callback is guaranteed to be invoked only when the
2253 point in time where it is supposed to trigger has passed. If multiple
2254 timers become ready during the same loop iteration then the ones with
2255 earlier time-out values are invoked before ones with later time-out
2256 values (but this is no longer true when a callback calls "ev_run"
2257 recursively).
2258 Watcher-Specific Functions and Data Members
2260 ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp
2262 interval, reschedule_cb)
2263 ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval,
2264 reschedule_cb)
2265 Lots of arguments, let's sort it out... There are basically three
2266 modes of operation, and we will explain them from simplest to most
2267 complex:
2268 · absolute timer (offset = absolute time, interval = 0,
2270 reschedule_cb = 0)
2271 In this configuration the watcher triggers an event after the
2273 wall clock time "offset" has passed. It will not repeat and
2274 will not adjust when a time jump occurs, that is, if it is to
2275 be run at January 1st 2011 then it will be stopped and invoked
2276 when the system clock reaches or surpasses this point in time.
2277 · repeating interval timer (offset = offset within interval,
2279 interval > 0, reschedule_cb = 0)
2280 In this mode the watcher will always be scheduled to time out
2282 at the next "offset + N * interval" time (for some integer N,
2283 which can also be negative) and then repeat, regardless of any
2284 time jumps. The "offset" argument is merely an offset into the
2285 "interval" periods.
2286 This can be used to create timers that do not drift with
2288 respect to the system clock, for example, here is an
2289 "ev_periodic" that triggers each hour, on the hour (with
2290 respect to UTC):
2291 ev_periodic_set (&periodic, 0., 3600., 0);
2293 This doesn't mean there will always be 3600 seconds in between
2295 triggers, but only that the callback will be called when the
2296 system time shows a full hour (UTC), or more correctly, when
2297 the system time is evenly divisible by 3600.
2298 Another way to think about it (for the mathematically inclined)
2300 is that "ev_periodic" will try to run the callback in this mode
2301 at the next possible time where "time = offset (mod interval)",
2302 regardless of any time jumps.
2303 The "interval" MUST be positive, and for numerical stability,
2305 the interval value should be higher than "1/8192" (which is
2306 around 100 microseconds) and "offset" should be higher than 0
2307 and should have at most a similar magnitude as the current time
2308 (say, within a factor of ten). Typical values for offset are,
2309 in fact, 0 or something between 0 and "interval", which is also
2310 the recommended range.
2311 Note also that there is an upper limit to how often a timer can
2313 fire (CPU speed for example), so if "interval" is very small
2314 then timing stability will of course deteriorate. Libev itself
2315 tries to be exact to be about one millisecond (if the OS
2316 supports it and the machine is fast enough).
2317 · manual reschedule mode (offset ignored, interval ignored,
2319 reschedule_cb = callback)
2320 In this mode the values for "interval" and "offset" are both
2322 being ignored. Instead, each time the periodic watcher gets
2323 scheduled, the reschedule callback will be called with the
2324 watcher as first, and the current time as second argument.
2325 NOTE: This callback MUST NOT stop or destroy any periodic
2327 watcher, ever, or make ANY other event loop modifications
2328 whatsoever, unless explicitly allowed by documentation here.
2329 If you need to stop it, return "now + 1e30" (or so, fudge
2331 fudge) and stop it afterwards (e.g. by starting an "ev_prepare"
2332 watcher, which is the only event loop modification you are
2333 allowed to do).
2334 The callback prototype is "ev_tstamp
2336 (*reschedule_cb)(ev_periodic *w, ev_tstamp now)", e.g.:
2337 static ev_tstamp
2339 my_rescheduler (ev_periodic *w, ev_tstamp now)
2340 {
2341 return now + 60.;
2342 }
2343 It must return the next time to trigger, based on the passed
2345 time value (that is, the lowest time value larger than to the
2346 second argument). It will usually be called just before the
2347 callback will be triggered, but might be called at other times,
2348 too.
2349 NOTE: This callback must always return a time that is higher
2351 than or equal to the passed "now" value.
2352 This can be used to create very complex timers, such as a timer
2354 that triggers on "next midnight, local time". To do this, you
2355 would calculate the next midnight after "now" and return the
2356 timestamp value for this. Here is a (completely untested, no
2357 error checking) example on how to do this:
2358 #include <time.h>
2360 static ev_tstamp
2362 my_rescheduler (ev_periodic *w, ev_tstamp now)
2363 {
2364 time_t tnow = (time_t)now;
2365 struct tm tm;
2366 localtime_r (&tnow, &tm);
2367 tm.tm_sec = tm.tm_min = tm.tm_hour = 0; // midnight current day
2369 ++tm.tm_mday; // midnight next day
2370 return mktime (&tm);
2372 }
2373 Note: this code might run into trouble on days that have more
2375 then two midnights (beginning and end).
2376 ev_periodic_again (loop, ev_periodic *)
2378 Simply stops and restarts the periodic watcher again. This is only
2379 useful when you changed some parameters or the reschedule callback
2380 would return a different time than the last time it was called
2381 (e.g. in a crond like program when the crontabs have changed).
2382 ev_tstamp ev_periodic_at (ev_periodic *)
2384 When active, returns the absolute time that the watcher is supposed
2385 to trigger next. This is not the same as the "offset" argument to
2386 "ev_periodic_set", but indeed works even in interval and manual
2387 rescheduling modes.
2388 ev_tstamp offset [read-write]
2390 When repeating, this contains the offset value, otherwise this is
2391 the absolute point in time (the "offset" value passed to
2392 "ev_periodic_set", although libev might modify this value for
2393 better numerical stability).
2394 Can be modified any time, but changes only take effect when the
2396 periodic timer fires or "ev_periodic_again" is being called.
2397 ev_tstamp interval [read-write]
2399 The current interval value. Can be modified any time, but changes
2400 only take effect when the periodic timer fires or
2401 "ev_periodic_again" is being called.
2402 ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
2404 The current reschedule callback, or 0, if this functionality is
2405 switched off. Can be changed any time, but changes only take effect
2406 when the periodic timer fires or "ev_periodic_again" is being
2407 called.
2408 Examples
2410 Example: Call a callback every hour, or, more precisely, whenever the
2412 system time is divisible by 3600. The callback invocation times have
2413 potentially a lot of jitter, but good long-term stability.
2414 static void
2416 clock_cb (struct ev_loop *loop, ev_periodic *w, int revents)
2417 {
2418 ... its now a full hour (UTC, or TAI or whatever your clock follows)
2419 }
2420 ev_periodic hourly_tick;
2422 ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
2423 ev_periodic_start (loop, &hourly_tick);
2424 Example: The same as above, but use a reschedule callback to do it:
2426 #include <math.h>
2428 static ev_tstamp
2430 my_scheduler_cb (ev_periodic *w, ev_tstamp now)
2431 {
2432 return now + (3600. - fmod (now, 3600.));
2433 }
2434 ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
2436 Example: Call a callback every hour, starting now:
2438 ev_periodic hourly_tick;
2440 ev_periodic_init (&hourly_tick, clock_cb,
2441 fmod (ev_now (loop), 3600.), 3600., 0);
2442 ev_periodic_start (loop, &hourly_tick);
2443 "ev_signal" - signal me when a signal gets signalled!
2445 Signal watchers will trigger an event when the process receives a
2446 specific signal one or more times. Even though signals are very
2447 asynchronous, libev will try its best to deliver signals synchronously,
2448 i.e. as part of the normal event processing, like any other event.
2449 If you want signals to be delivered truly asynchronously, just use
2451 "sigaction" as you would do without libev and forget about sharing the
2452 signal. You can even use "ev_async" from a signal handler to
2453 synchronously wake up an event loop.
2454 You can configure as many watchers as you like for the same signal, but
2456 only within the same loop, i.e. you can watch for "SIGINT" in your
2457 default loop and for "SIGIO" in another loop, but you cannot watch for
2458 "SIGINT" in both the default loop and another loop at the same time. At
2459 the moment, "SIGCHLD" is permanently tied to the default loop.
2460 Only after the first watcher for a signal is started will libev
2462 actually register something with the kernel. It thus coexists with your
2463 own signal handlers as long as you don't register any with libev for
2464 the same signal.
2465 If possible and supported, libev will install its handlers with
2467 "SA_RESTART" (or equivalent) behaviour enabled, so system calls should
2468 not be unduly interrupted. If you have a problem with system calls
2469 getting interrupted by signals you can block all signals in an
2470 "ev_check" watcher and unblock them in an "ev_prepare" watcher.
2471 The special problem of inheritance over fork/execve/pthread_create
2473 Both the signal mask ("sigprocmask") and the signal disposition
2475 ("sigaction") are unspecified after starting a signal watcher (and
2476 after stopping it again), that is, libev might or might not block the
2477 signal, and might or might not set or restore the installed signal
2478 handler (but see "EVFLAG_NOSIGMASK").
2479 While this does not matter for the signal disposition (libev never sets
2481 signals to "SIG_IGN", so handlers will be reset to "SIG_DFL" on
2482 "execve"), this matters for the signal mask: many programs do not
2483 expect certain signals to be blocked.
2484 This means that before calling "exec" (from the child) you should reset
2486 the signal mask to whatever "default" you expect (all clear is a good
2487 choice usually).
2488 The simplest way to ensure that the signal mask is reset in the child
2490 is to install a fork handler with "pthread_atfork" that resets it. That
2491 will catch fork calls done by libraries (such as the libc) as well.
2492 In current versions of libev, the signal will not be blocked
2494 indefinitely unless you use the "signalfd" API ("EV_SIGNALFD"). While
2495 this reduces the window of opportunity for problems, it will not go
2496 away, as libev has to modify the signal mask, at least temporarily.
2497 So I can't stress this enough: If you do not reset your signal mask
2499 when you expect it to be empty, you have a race condition in your code.
2500 This is not a libev-specific thing, this is true for most event
2501 libraries.
2502 The special problem of threads signal handling
2504 POSIX threads has problematic signal handling semantics, specifically,
2506 a lot of functionality (sigfd, sigwait etc.) only really works if all
2507 threads in a process block signals, which is hard to achieve.
2508 When you want to use sigwait (or mix libev signal handling with your
2510 own for the same signals), you can tackle this problem by globally
2511 blocking all signals before creating any threads (or creating them with
2512 a fully set sigprocmask) and also specifying the "EVFLAG_NOSIGMASK"
2513 when creating loops. Then designate one thread as "signal receiver
2514 thread" which handles these signals. You can pass on any signals that
2515 libev might be interested in by calling "ev_feed_signal".
2516 Watcher-Specific Functions and Data Members
2518 ev_signal_init (ev_signal *, callback, int signum)
2520 ev_signal_set (ev_signal *, int signum)
2521 Configures the watcher to trigger on the given signal number
2522 (usually one of the "SIGxxx" constants).
2523 int signum [read-only]
2525 The signal the watcher watches out for.
2526 Examples
2528 Example: Try to exit cleanly on SIGINT.
2530 static void
2532 sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
2533 {
2534 ev_break (loop, EVBREAK_ALL);
2535 }
2536 ev_signal signal_watcher;
2538 ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
2539 ev_signal_start (loop, &signal_watcher);
2540 "ev_child" - watch out for process status changes
2542 Child watchers trigger when your process receives a SIGCHLD in response
2543 to some child status changes (most typically when a child of yours dies
2544 or exits). It is permissible to install a child watcher after the child
2545 has been forked (which implies it might have already exited), as long
2546 as the event loop isn't entered (or is continued from a watcher), i.e.,
2547 forking and then immediately registering a watcher for the child is
2548 fine, but forking and registering a watcher a few event loop iterations
2549 later or in the next callback invocation is not.
2550 Only the default event loop is capable of handling signals, and
2552 therefore you can only register child watchers in the default event
2553 loop.
2554 Due to some design glitches inside libev, child watchers will always be
2556 handled at maximum priority (their priority is set to "EV_MAXPRI" by
2557 libev)
2558 Process Interaction
2560 Libev grabs "SIGCHLD" as soon as the default event loop is initialised.
2562 This is necessary to guarantee proper behaviour even if the first child
2563 watcher is started after the child exits. The occurrence of "SIGCHLD"
2564 is recorded asynchronously, but child reaping is done synchronously as
2565 part of the event loop processing. Libev always reaps all children,
2566 even ones not watched.
2567 Overriding the Built-In Processing
2569 Libev offers no special support for overriding the built-in child
2571 processing, but if your application collides with libev's default child
2572 handler, you can override it easily by installing your own handler for
2573 "SIGCHLD" after initialising the default loop, and making sure the
2574 default loop never gets destroyed. You are encouraged, however, to use
2575 an event-based approach to child reaping and thus use libev's support
2576 for that, so other libev users can use "ev_child" watchers freely.
2577 Stopping the Child Watcher
2579 Currently, the child watcher never gets stopped, even when the child
2581 terminates, so normally one needs to stop the watcher in the callback.
2582 Future versions of libev might stop the watcher automatically when a
2583 child exit is detected (calling "ev_child_stop" twice is not a
2584 problem).
2585 Watcher-Specific Functions and Data Members
2587 ev_child_init (ev_child *, callback, int pid, int trace)
2589 ev_child_set (ev_child *, int pid, int trace)
2590 Configures the watcher to wait for status changes of process "pid"
2591 (or any process if "pid" is specified as 0). The callback can look
2592 at the "rstatus" member of the "ev_child" watcher structure to see
2593 the status word (use the macros from "sys/wait.h" and see your
2594 systems "waitpid" documentation). The "rpid" member contains the
2595 pid of the process causing the status change. "trace" must be
2596 either 0 (only activate the watcher when the process terminates) or
2597 1 (additionally activate the watcher when the process is stopped or
2598 continued).
2599 int pid [read-only]
2601 The process id this watcher watches out for, or 0, meaning any
2602 process id.
2603 int rpid [read-write]
2605 The process id that detected a status change.
2606 int rstatus [read-write]
2608 The process exit/trace status caused by "rpid" (see your systems
2609 "waitpid" and "sys/wait.h" documentation for details).
2610 Examples
2612 Example: "fork()" a new process and install a child handler to wait for
2614 its completion.
2615 ev_child cw;
2617 static void
2619 child_cb (EV_P_ ev_child *w, int revents)
2620 {
2621 ev_child_stop (EV_A_ w);
2622 printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
2623 }
2624 pid_t pid = fork ();
2626 if (pid < 0)
2628 // error
2629 else if (pid == 0)
2630 {
2631 // the forked child executes here
2632 exit (1);
2633 }
2634 else
2635 {
2636 ev_child_init (&cw, child_cb, pid, 0);
2637 ev_child_start (EV_DEFAULT_ &cw);
2638 }
2639 "ev_stat" - did the file attributes just change?
2641 This watches a file system path for attribute changes. That is, it
2642 calls "stat" on that path in regular intervals (or when the OS says it
2643 changed) and sees if it changed compared to the last time, invoking the
2644 callback if it did. Starting the watcher "stat"'s the file, so only
2645 changes that happen after the watcher has been started will be
2646 reported.
2647 The path does not need to exist: changing from "path exists" to "path
2649 does not exist" is a status change like any other. The condition "path
2650 does not exist" (or more correctly "path cannot be stat'ed") is
2651 signified by the "st_nlink" field being zero (which is otherwise always
2652 forced to be at least one) and all the other fields of the stat buffer
2653 having unspecified contents.
2654 The path must not end in a slash or contain special components such as
2656 "." or "..". The path should be absolute: If it is relative and your
2657 working directory changes, then the behaviour is undefined.
2658 Since there is no portable change notification interface available, the
2660 portable implementation simply calls stat(2) regularly on the path to
2661 see if it changed somehow. You can specify a recommended polling
2662 interval for this case. If you specify a polling interval of 0 (highly
2663 recommended!) then a suitable, unspecified default value will be used
2664 (which you can expect to be around five seconds, although this might
2665 change dynamically). Libev will also impose a minimum interval which is
2666 currently around 0.1, but that's usually overkill.
2667 This watcher type is not meant for massive numbers of stat watchers, as
2669 even with OS-supported change notifications, this can be resource-
2670 intensive.
2671 At the time of this writing, the only OS-specific interface implemented
2673 is the Linux inotify interface (implementing kqueue support is left as
2674 an exercise for the reader. Note, however, that the author sees no way
2675 of implementing "ev_stat" semantics with kqueue, except as a hint).
2676 ABI Issues (Largefile Support)
2678 Libev by default (unless the user overrides this) uses the default
2680 compilation environment, which means that on systems with large file
2681 support disabled by default, you get the 32 bit version of the stat
2682 structure. When using the library from programs that change the ABI to
2683 use 64 bit file offsets the programs will fail. In that case you have
2684 to compile libev with the same flags to get binary compatibility. This
2685 is obviously the case with any flags that change the ABI, but the
2686 problem is most noticeably displayed with ev_stat and large file
2687 support.
2688 The solution for this is to lobby your distribution maker to make large
2690 file interfaces available by default (as e.g. FreeBSD does) and not
2691 optional. Libev cannot simply switch on large file support because it
2692 has to exchange stat structures with application programs compiled
2693 using the default compilation environment.
2694 Inotify and Kqueue
2696 When "inotify (7)" support has been compiled into libev and present at
2698 runtime, it will be used to speed up change detection where possible.
2699 The inotify descriptor will be created lazily when the first "ev_stat"
2700 watcher is being started.
2701 Inotify presence does not change the semantics of "ev_stat" watchers
2703 except that changes might be detected earlier, and in some cases, to
2704 avoid making regular "stat" calls. Even in the presence of inotify
2705 support there are many cases where libev has to resort to regular
2706 "stat" polling, but as long as kernel 2.6.25 or newer is used (2.6.24
2707 and older have too many bugs), the path exists (i.e. stat succeeds),
2708 and the path resides on a local filesystem (libev currently assumes
2709 only ext2/3, jfs, reiserfs and xfs are fully working) libev usually
2710 gets away without polling.
2711 There is no support for kqueue, as apparently it cannot be used to
2713 implement this functionality, due to the requirement of having a file
2714 descriptor open on the object at all times, and detecting renames,
2715 unlinks etc. is difficult.
2716 "stat ()" is a synchronous operation
2718 Libev doesn't normally do any kind of I/O itself, and so is not
2720 blocking the process. The exception are "ev_stat" watchers - those call
2721 "stat ()", which is a synchronous operation.
2722 For local paths, this usually doesn't matter: unless the system is very
2724 busy or the intervals between stat's are large, a stat call will be
2725 fast, as the path data is usually in memory already (except when
2726 starting the watcher).
2727 For networked file systems, calling "stat ()" can block an indefinite
2729 time due to network issues, and even under good conditions, a stat call
2730 often takes multiple milliseconds.
2731 Therefore, it is best to avoid using "ev_stat" watchers on networked
2733 paths, although this is fully supported by libev.
2734 The special problem of stat time resolution
2736 The "stat ()" system call only supports full-second resolution
2738 portably, and even on systems where the resolution is higher, most file
2739 systems still only support whole seconds.
2740 That means that, if the time is the only thing that changes, you can
2742 easily miss updates: on the first update, "ev_stat" detects a change
2743 and calls your callback, which does something. When there is another
2744 update within the same second, "ev_stat" will be unable to detect
2745 unless the stat data does change in other ways (e.g. file size).
2746 The solution to this is to delay acting on a change for slightly more
2748 than a second (or till slightly after the next full second boundary),
2749 using a roughly one-second-delay "ev_timer" (e.g. "ev_timer_set (w, 0.,
2750 1.02); ev_timer_again (loop, w)").
2751 The .02 offset is added to work around small timing inconsistencies of
2753 some operating systems (where the second counter of the current time
2754 might be be delayed. One such system is the Linux kernel, where a call
2755 to "gettimeofday" might return a timestamp with a full second later
2756 than a subsequent "time" call - if the equivalent of "time ()" is used
2757 to update file times then there will be a small window where the kernel
2758 uses the previous second to update file times but libev might already
2759 execute the timer callback).
2760 Watcher-Specific Functions and Data Members
2762 ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp
2764 interval)
2765 ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
2766 Configures the watcher to wait for status changes of the given
2767 "path". The "interval" is a hint on how quickly a change is
2768 expected to be detected and should normally be specified as 0 to
2769 let libev choose a suitable value. The memory pointed to by "path"
2770 must point to the same path for as long as the watcher is active.
2771 The callback will receive an "EV_STAT" event when a change was
2773 detected, relative to the attributes at the time the watcher was
2774 started (or the last change was detected).
2775 ev_stat_stat (loop, ev_stat *)
2777 Updates the stat buffer immediately with new values. If you change
2778 the watched path in your callback, you could call this function to
2779 avoid detecting this change (while introducing a race condition if
2780 you are not the only one changing the path). Can also be useful
2781 simply to find out the new values.
2782 ev_statdata attr [read-only]
2784 The most-recently detected attributes of the file. Although the
2785 type is "ev_statdata", this is usually the (or one of the) "struct
2786 stat" types suitable for your system, but you can only rely on the
2787 POSIX-standardised members to be present. If the "st_nlink" member
2788 is 0, then there was some error while "stat"ing the file.
2789 ev_statdata prev [read-only]
2791 The previous attributes of the file. The callback gets invoked
2792 whenever "prev" != "attr", or, more precisely, one or more of these
2793 members differ: "st_dev", "st_ino", "st_mode", "st_nlink",
2794 "st_uid", "st_gid", "st_rdev", "st_size", "st_atime", "st_mtime",
2795 "st_ctime".
2796 ev_tstamp interval [read-only]
2798 The specified interval.
2799 const char *path [read-only]
2801 The file system path that is being watched.
2802 Examples
2804 Example: Watch "/etc/passwd" for attribute changes.
2806 static void
2808 passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
2809 {
2810 /* /etc/passwd changed in some way */
2811 if (w->attr.st_nlink)
2812 {
2813 printf ("passwd current size %ld\n", (long)w->attr.st_size);
2814 printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
2815 printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
2816 }
2817 else
2818 /* you shalt not abuse printf for puts */
2819 puts ("wow, /etc/passwd is not there, expect problems. "
2820 "if this is windows, they already arrived\n");
2821 }
2822 ...
2824 ev_stat passwd;
2825 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
2827 ev_stat_start (loop, &passwd);
2828 Example: Like above, but additionally use a one-second delay so we do
2830 not miss updates (however, frequent updates will delay processing, too,
2831 so one might do the work both on "ev_stat" callback invocation and on
2832 "ev_timer" callback invocation).
2833 static ev_stat passwd;
2835 static ev_timer timer;
2836 static void
2838 timer_cb (EV_P_ ev_timer *w, int revents)
2839 {
2840 ev_timer_stop (EV_A_ w);
2841 /* now it's one second after the most recent passwd change */
2843 }
2844 static void
2846 stat_cb (EV_P_ ev_stat *w, int revents)
2847 {
2848 /* reset the one-second timer */
2849 ev_timer_again (EV_A_ &timer);
2850 }
2851 ...
2853 ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
2854 ev_stat_start (loop, &passwd);
2855 ev_timer_init (&timer, timer_cb, 0., 1.02);
2856 "ev_idle" - when you've got nothing better to do...
2858 Idle watchers trigger events when no other events of the same or higher
2859 priority are pending (prepare, check and other idle watchers do not
2860 count as receiving "events").
2861 That is, as long as your process is busy handling sockets or timeouts
2863 (or even signals, imagine) of the same or higher priority it will not
2864 be triggered. But when your process is idle (or only lower-priority
2865 watchers are pending), the idle watchers are being called once per
2866 event loop iteration - until stopped, that is, or your process receives
2867 more events and becomes busy again with higher priority stuff.
2868 The most noteworthy effect is that as long as any idle watchers are
2870 active, the process will not block when waiting for new events.
2871 Apart from keeping your process non-blocking (which is a useful effect
2873 on its own sometimes), idle watchers are a good place to do "pseudo-
2874 background processing", or delay processing stuff to after the event
2875 loop has handled all outstanding events.
2876 Abusing an "ev_idle" watcher for its side-effect
2878 As long as there is at least one active idle watcher, libev will never
2880 sleep unnecessarily. Or in other words, it will loop as fast as
2881 possible. For this to work, the idle watcher doesn't need to be
2882 invoked at all - the lowest priority will do.
2883 This mode of operation can be useful together with an "ev_check"
2885 watcher, to do something on each event loop iteration - for example to
2886 balance load between different connections.
2887 See "Abusing an ev_check watcher for its side-effect" for a longer
2889 example.
2890 Watcher-Specific Functions and Data Members
2892 ev_idle_init (ev_idle *, callback)
2894 Initialises and configures the idle watcher - it has no parameters
2895 of any kind. There is a "ev_idle_set" macro, but using it is
2896 utterly pointless, believe me.
2897 Examples
2899 Example: Dynamically allocate an "ev_idle" watcher, start it, and in
2901 the callback, free it. Also, use no error checking, as usual.
2902 static void
2904 idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
2905 {
2906 // stop the watcher
2907 ev_idle_stop (loop, w);
2908 // now we can free it
2910 free (w);
2911 // now do something you wanted to do when the program has
2913 // no longer anything immediate to do.
2914 }
2915 ev_idle *idle_watcher = malloc (sizeof (ev_idle));
2917 ev_idle_init (idle_watcher, idle_cb);
2918 ev_idle_start (loop, idle_watcher);
2919 "ev_prepare" and "ev_check" - customise your event loop!
2921 Prepare and check watchers are often (but not always) used in pairs:
2922 prepare watchers get invoked before the process blocks and check
2923 watchers afterwards.
2924 You must not call "ev_run" (or similar functions that enter the current
2926 event loop) or "ev_loop_fork" from either "ev_prepare" or "ev_check"
2927 watchers. Other loops than the current one are fine, however. The
2928 rationale behind this is that you do not need to check for recursion in
2929 those watchers, i.e. the sequence will always be "ev_prepare",
2930 blocking, "ev_check" so if you have one watcher of each kind they will
2931 always be called in pairs bracketing the blocking call.
2932 Their main purpose is to integrate other event mechanisms into libev
2934 and their use is somewhat advanced. They could be used, for example, to
2935 track variable changes, implement your own watchers, integrate net-snmp
2936 or a coroutine library and lots more. They are also occasionally useful
2937 if you cache some data and want to flush it before blocking (for
2938 example, in X programs you might want to do an "XFlush ()" in an
2939 "ev_prepare" watcher).
2940 This is done by examining in each prepare call which file descriptors
2942 need to be watched by the other library, registering "ev_io" watchers
2943 for them and starting an "ev_timer" watcher for any timeouts (many
2944 libraries provide exactly this functionality). Then, in the check
2945 watcher, you check for any events that occurred (by checking the
2946 pending status of all watchers and stopping them) and call back into
2947 the library. The I/O and timer callbacks will never actually be called
2948 (but must be valid nevertheless, because you never know, you know?).
2949 As another example, the Perl Coro module uses these hooks to integrate
2951 coroutines into libev programs, by yielding to other active coroutines
2952 during each prepare and only letting the process block if no coroutines
2953 are ready to run (it's actually more complicated: it only runs
2954 coroutines with priority higher than or equal to the event loop and one
2955 coroutine of lower priority, but only once, using idle watchers to keep
2956 the event loop from blocking if lower-priority coroutines are active,
2957 thus mapping low-priority coroutines to idle/background tasks).
2958 When used for this purpose, it is recommended to give "ev_check"
2960 watchers highest ("EV_MAXPRI") priority, to ensure that they are being
2961 run before any other watchers after the poll (this doesn't matter for
2962 "ev_prepare" watchers).
2963 Also, "ev_check" watchers (and "ev_prepare" watchers, too) should not
2965 activate ("feed") events into libev. While libev fully supports this,
2966 they might get executed before other "ev_check" watchers did their job.
2967 As "ev_check" watchers are often used to embed other (non-libev) event
2968 loops those other event loops might be in an unusable state until their
2969 "ev_check" watcher ran (always remind yourself to coexist peacefully
2970 with others).
2971 Abusing an "ev_check" watcher for its side-effect
2973 "ev_check" (and less often also "ev_prepare") watchers can also be
2975 useful because they are called once per event loop iteration. For
2976 example, if you want to handle a large number of connections fairly,
2977 you normally only do a bit of work for each active connection, and if
2978 there is more work to do, you wait for the next event loop iteration,
2979 so other connections have a chance of making progress.
2980 Using an "ev_check" watcher is almost enough: it will be called on the
2982 next event loop iteration. However, that isn't as soon as possible -
2983 without external events, your "ev_check" watcher will not be invoked.
2984 This is where "ev_idle" watchers come in handy - all you need is a
2986 single global idle watcher that is active as long as you have one
2987 active "ev_check" watcher. The "ev_idle" watcher makes sure the event
2988 loop will not sleep, and the "ev_check" watcher makes sure a callback
2989 gets invoked. Neither watcher alone can do that.
2990 Watcher-Specific Functions and Data Members
2992 ev_prepare_init (ev_prepare *, callback)
2994 ev_check_init (ev_check *, callback)
2995 Initialises and configures the prepare or check watcher - they have
2996 no parameters of any kind. There are "ev_prepare_set" and
2997 "ev_check_set" macros, but using them is utterly, utterly, utterly
2998 and completely pointless.
2999 Examples
3001 There are a number of principal ways to embed other event loops or
3003 modules into libev. Here are some ideas on how to include libadns into
3004 libev (there is a Perl module named "EV::ADNS" that does this, which
3005 you could use as a working example. Another Perl module named
3006 "EV::Glib" embeds a Glib main context into libev, and finally,
3007 "Glib::EV" embeds EV into the Glib event loop).
3008 Method 1: Add IO watchers and a timeout watcher in a prepare handler,
3010 and in a check watcher, destroy them and call into libadns. What
3011 follows is pseudo-code only of course. This requires you to either use
3012 a low priority for the check watcher or use "ev_clear_pending"
3013 explicitly, as the callbacks for the IO/timeout watchers might not have
3014 been called yet.
3015 static ev_io iow [nfd];
3017 static ev_timer tw;
3018 static void
3020 io_cb (struct ev_loop *loop, ev_io *w, int revents)
3021 {
3022 }
3023 // create io watchers for each fd and a timer before blocking
3025 static void
3026 adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
3027 {
3028 int timeout = 3600000;
3029 struct pollfd fds [nfd];
3030 // actual code will need to loop here and realloc etc.
3031 adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
3032 /* the callback is illegal, but won't be called as we stop during check */
3034 ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
3035 ev_timer_start (loop, &tw);
3036 // create one ev_io per pollfd
3038 for (int i = 0; i < nfd; ++i)
3039 {
3040 ev_io_init (iow + i, io_cb, fds [i].fd,
3041 ((fds [i].events & POLLIN ? EV_READ : 0)
3042 | (fds [i].events & POLLOUT ? EV_WRITE : 0)));
3043 fds [i].revents = 0;
3045 ev_io_start (loop, iow + i);
3046 }
3047 }
3048 // stop all watchers after blocking
3050 static void
3051 adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
3052 {
3053 ev_timer_stop (loop, &tw);
3054 for (int i = 0; i < nfd; ++i)
3056 {
3057 // set the relevant poll flags
3058 // could also call adns_processreadable etc. here
3059 struct pollfd *fd = fds + i;
3060 int revents = ev_clear_pending (iow + i);
3061 if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
3062 if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
3063 // now stop the watcher
3065 ev_io_stop (loop, iow + i);
3066 }
3067 adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
3069 }
3070 Method 2: This would be just like method 1, but you run
3072 "adns_afterpoll" in the prepare watcher and would dispose of the check
3073 watcher.
3074 Method 3: If the module to be embedded supports explicit event
3076 notification (libadns does), you can also make use of the actual
3077 watcher callbacks, and only destroy/create the watchers in the prepare
3078 watcher.
3079 static void
3081 timer_cb (EV_P_ ev_timer *w, int revents)
3082 {
3083 adns_state ads = (adns_state)w->data;
3084 update_now (EV_A);
3085 adns_processtimeouts (ads, &tv_now);
3087 }
3088 static void
3090 io_cb (EV_P_ ev_io *w, int revents)
3091 {
3092 adns_state ads = (adns_state)w->data;
3093 update_now (EV_A);
3094 if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
3096 if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
3097 }
3098 // do not ever call adns_afterpoll
3100 Method 4: Do not use a prepare or check watcher because the module you
3102 want to embed is not flexible enough to support it. Instead, you can
3103 override their poll function. The drawback with this solution is that
3104 the main loop is now no longer controllable by EV. The "Glib::EV"
3105 module uses this approach, effectively embedding EV as a client into
3106 the horrible libglib event loop.
3107 static gint
3109 event_poll_func (GPollFD *fds, guint nfds, gint timeout)
3110 {
3111 int got_events = 0;
3112 for (n = 0; n < nfds; ++n)
3114 // create/start io watcher that sets the relevant bits in fds[n] and increment got_events
3115 if (timeout >= 0)
3117 // create/start timer
3118 // poll
3120 ev_run (EV_A_ 0);
3121 // stop timer again
3123 if (timeout >= 0)
3124 ev_timer_stop (EV_A_ &to);
3125 // stop io watchers again - their callbacks should have set
3127 for (n = 0; n < nfds; ++n)
3128 ev_io_stop (EV_A_ iow [n]);
3129 return got_events;
3131 }
3132 "ev_embed" - when one backend isn't enough...
3134 This is a rather advanced watcher type that lets you embed one event
3135 loop into another (currently only "ev_io" events are supported in the
3136 embedded loop, other types of watchers might be handled in a delayed or
3137 incorrect fashion and must not be used).
3138 There are primarily two reasons you would want that: work around bugs
3140 and prioritise I/O.
3141 As an example for a bug workaround, the kqueue backend might only
3143 support sockets on some platform, so it is unusable as generic backend,
3144 but you still want to make use of it because you have many sockets and
3145 it scales so nicely. In this case, you would create a kqueue-based loop
3146 and embed it into your default loop (which might use e.g. poll).
3147 Overall operation will be a bit slower because first libev has to call
3148 "poll" and then "kevent", but at least you can use both mechanisms for
3149 what they are best: "kqueue" for scalable sockets and "poll" if you
3150 want it to work :)
3151 As for prioritising I/O: under rare circumstances you have the case
3153 where some fds have to be watched and handled very quickly (with low
3154 latency), and even priorities and idle watchers might have too much
3155 overhead. In this case you would put all the high priority stuff in one
3156 loop and all the rest in a second one, and embed the second one in the
3157 first.
3158 As long as the watcher is active, the callback will be invoked every
3160 time there might be events pending in the embedded loop. The callback
3161 must then call "ev_embed_sweep (mainloop, watcher)" to make a single
3162 sweep and invoke their callbacks (the callback doesn't need to invoke
3163 the "ev_embed_sweep" function directly, it could also start an idle
3164 watcher to give the embedded loop strictly lower priority for example).
3165 You can also set the callback to 0, in which case the embed watcher
3167 will automatically execute the embedded loop sweep whenever necessary.
3168 Fork detection will be handled transparently while the "ev_embed"
3170 watcher is active, i.e., the embedded loop will automatically be forked
3171 when the embedding loop forks. In other cases, the user is responsible
3172 for calling "ev_loop_fork" on the embedded loop.
3173 Unfortunately, not all backends are embeddable: only the ones returned
3175 by "ev_embeddable_backends" are, which, unfortunately, does not include
3176 any portable one.
3177 So when you want to use this feature you will always have to be
3179 prepared that you cannot get an embeddable loop. The recommended way to
3180 get around this is to have a separate variables for your embeddable
3181 loop, try to create it, and if that fails, use the normal loop for
3182 everything.
3183 "ev_embed" and fork
3185 While the "ev_embed" watcher is running, forks in the embedding loop
3187 will automatically be applied to the embedded loop as well, so no
3188 special fork handling is required in that case. When the watcher is not
3189 running, however, it is still the task of the libev user to call
3190 "ev_loop_fork ()" as applicable.
3191 Watcher-Specific Functions and Data Members
3193 ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
3195 ev_embed_set (ev_embed *, struct ev_loop *embedded_loop)
3196 Configures the watcher to embed the given loop, which must be
3197 embeddable. If the callback is 0, then "ev_embed_sweep" will be
3198 invoked automatically, otherwise it is the responsibility of the
3199 callback to invoke it (it will continue to be called until the
3200 sweep has been done, if you do not want that, you need to
3201 temporarily stop the embed watcher).
3202 ev_embed_sweep (loop, ev_embed *)
3204 Make a single, non-blocking sweep over the embedded loop. This
3205 works similarly to "ev_run (embedded_loop, EVRUN_NOWAIT)", but in
3206 the most appropriate way for embedded loops.
3207 struct ev_loop *other [read-only]
3209 The embedded event loop.
3210 Examples
3212 Example: Try to get an embeddable event loop and embed it into the
3214 default event loop. If that is not possible, use the default loop. The
3215 default loop is stored in "loop_hi", while the embeddable loop is
3216 stored in "loop_lo" (which is "loop_hi" in the case no embeddable loop
3217 can be used).
3218 struct ev_loop *loop_hi = ev_default_init (0);
3220 struct ev_loop *loop_lo = 0;
3221 ev_embed embed;
3222 // see if there is a chance of getting one that works
3224 // (remember that a flags value of 0 means autodetection)
3225 loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
3226 ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
3227 : 0;
3228 // if we got one, then embed it, otherwise default to loop_hi
3230 if (loop_lo)
3231 {
3232 ev_embed_init (&embed, 0, loop_lo);
3233 ev_embed_start (loop_hi, &embed);
3234 }
3235 else
3236 loop_lo = loop_hi;
3237 Example: Check if kqueue is available but not recommended and create a
3239 kqueue backend for use with sockets (which usually work with any kqueue
3240 implementation). Store the kqueue/socket-only event loop in
3241 "loop_socket". (One might optionally use "EVFLAG_NOENV", too).
3242 struct ev_loop *loop = ev_default_init (0);
3244 struct ev_loop *loop_socket = 0;
3245 ev_embed embed;
3246 if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
3248 if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
3249 {
3250 ev_embed_init (&embed, 0, loop_socket);
3251 ev_embed_start (loop, &embed);
3252 }
3253 if (!loop_socket)
3255 loop_socket = loop;
3256 // now use loop_socket for all sockets, and loop for everything else
3258 "ev_fork" - the audacity to resume the event loop after a fork
3260 Fork watchers are called when a "fork ()" was detected (usually because
3261 whoever is a good citizen cared to tell libev about it by calling
3262 "ev_loop_fork"). The invocation is done before the event loop blocks
3263 next and before "ev_check" watchers are being called, and only in the
3264 child after the fork. If whoever good citizen calling "ev_default_fork"
3265 cheats and calls it in the wrong process, the fork handlers will be
3266 invoked, too, of course.
3267 The special problem of life after fork - how is it possible?
3269 Most uses of "fork ()" consist of forking, then some simple calls to
3271 set up/change the process environment, followed by a call to "exec()".
3272 This sequence should be handled by libev without any problems.
3273 This changes when the application actually wants to do event handling
3275 in the child, or both parent in child, in effect "continuing" after the
3276 fork.
3277 The default mode of operation (for libev, with application help to
3279 detect forks) is to duplicate all the state in the child, as would be
3280 expected when either the parent or the child process continues.
3281 When both processes want to continue using libev, then this is usually
3283 the wrong result. In that case, usually one process (typically the
3284 parent) is supposed to continue with all watchers in place as before,
3285 while the other process typically wants to start fresh, i.e. without
3286 any active watchers.
3287 The cleanest and most efficient way to achieve that with libev is to
3289 simply create a new event loop, which of course will be "empty", and
3290 use that for new watchers. This has the advantage of not touching more
3291 memory than necessary, and thus avoiding the copy-on-write, and the
3292 disadvantage of having to use multiple event loops (which do not
3293 support signal watchers).
3294 When this is not possible, or you want to use the default loop for
3296 other reasons, then in the process that wants to start "fresh", call
3297 "ev_loop_destroy (EV_DEFAULT)" followed by "ev_default_loop (...)".
3298 Destroying the default loop will "orphan" (not stop) all registered
3299 watchers, so you have to be careful not to execute code that modifies
3300 those watchers. Note also that in that case, you have to re-register
3301 any signal watchers.
3302 Watcher-Specific Functions and Data Members
3304 ev_fork_init (ev_fork *, callback)
3306 Initialises and configures the fork watcher - it has no parameters
3307 of any kind. There is a "ev_fork_set" macro, but using it is
3308 utterly pointless, really.
3309 "ev_cleanup" - even the best things end
3311 Cleanup watchers are called just before the event loop is being
3312 destroyed by a call to "ev_loop_destroy".
3313 While there is no guarantee that the event loop gets destroyed, cleanup
3315 watchers provide a convenient method to install cleanup hooks for your
3316 program, worker threads and so on - you just to make sure to destroy
3317 the loop when you want them to be invoked.
3318 Cleanup watchers are invoked in the same way as any other watcher.
3320 Unlike all other watchers, they do not keep a reference to the event
3321 loop (which makes a lot of sense if you think about it). Like all other
3322 watchers, you can call libev functions in the callback, except
3323 "ev_cleanup_start".
3324 Watcher-Specific Functions and Data Members
3326 ev_cleanup_init (ev_cleanup *, callback)
3328 Initialises and configures the cleanup watcher - it has no
3329 parameters of any kind. There is a "ev_cleanup_set" macro, but
3330 using it is utterly pointless, I assure you.
3331 Example: Register an atexit handler to destroy the default loop, so any
3333 cleanup functions are called.
3334 static void
3336 program_exits (void)
3337 {
3338 ev_loop_destroy (EV_DEFAULT_UC);
3339 }
3340 ...
3342 atexit (program_exits);
3343 "ev_async" - how to wake up an event loop
3345 In general, you cannot use an "ev_loop" from multiple threads or other
3346 asynchronous sources such as signal handlers (as opposed to multiple
3347 event loops - those are of course safe to use in different threads).
3348 Sometimes, however, you need to wake up an event loop you do not
3350 control, for example because it belongs to another thread. This is what
3351 "ev_async" watchers do: as long as the "ev_async" watcher is active,
3352 you can signal it by calling "ev_async_send", which is thread- and
3353 signal safe.
3354 This functionality is very similar to "ev_signal" watchers, as signals,
3356 too, are asynchronous in nature, and signals, too, will be compressed
3357 (i.e. the number of callback invocations may be less than the number of
3358 "ev_async_send" calls). In fact, you could use signal watchers as a
3359 kind of "global async watchers" by using a watcher on an otherwise
3360 unused signal, and "ev_feed_signal" to signal this watcher from another
3361 thread, even without knowing which loop owns the signal.
3362 Queueing
3364 "ev_async" does not support queueing of data in any way. The reason is
3366 that the author does not know of a simple (or any) algorithm for a
3367 multiple-writer-single-reader queue that works in all cases and doesn't
3368 need elaborate support such as pthreads or unportable memory access
3369 semantics.
3370 That means that if you want to queue data, you have to provide your own
3372 queue. But at least I can tell you how to implement locking around your
3373 queue:
3374 queueing from a signal handler context
3376 To implement race-free queueing, you simply add to the queue in the
3377 signal handler but you block the signal handler in the watcher
3378 callback. Here is an example that does that for some fictitious
3379 SIGUSR1 handler:
3380 static ev_async mysig;
3382 static void
3384 sigusr1_handler (void)
3385 {
3386 sometype data;
3387 // no locking etc.
3389 queue_put (data);
3390 ev_async_send (EV_DEFAULT_ &mysig);
3391 }
3392 static void
3394 mysig_cb (EV_P_ ev_async *w, int revents)
3395 {
3396 sometype data;
3397 sigset_t block, prev;
3398 sigemptyset (&block);
3400 sigaddset (&block, SIGUSR1);
3401 sigprocmask (SIG_BLOCK, &block, &prev);
3402 while (queue_get (&data))
3404 process (data);
3405 if (sigismember (&prev, SIGUSR1)
3407 sigprocmask (SIG_UNBLOCK, &block, 0);
3408 }
3409 (Note: pthreads in theory requires you to use "pthread_setmask"
3411 instead of "sigprocmask" when you use threads, but libev doesn't do
3412 it either...).
3413 queueing from a thread context
3415 The strategy for threads is different, as you cannot (easily) block
3416 threads but you can easily preempt them, so to queue safely you
3417 need to employ a traditional mutex lock, such as in this pthread
3418 example:
3419 static ev_async mysig;
3421 static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
3422 static void
3424 otherthread (void)
3425 {
3426 // only need to lock the actual queueing operation
3427 pthread_mutex_lock (&mymutex);
3428 queue_put (data);
3429 pthread_mutex_unlock (&mymutex);
3430 ev_async_send (EV_DEFAULT_ &mysig);
3432 }
3433 static void
3435 mysig_cb (EV_P_ ev_async *w, int revents)
3436 {
3437 pthread_mutex_lock (&mymutex);
3438 while (queue_get (&data))
3440 process (data);
3441 pthread_mutex_unlock (&mymutex);
3443 }
3444 Watcher-Specific Functions and Data Members
3446 ev_async_init (ev_async *, callback)
3448 Initialises and configures the async watcher - it has no parameters
3449 of any kind. There is a "ev_async_set" macro, but using it is
3450 utterly pointless, trust me.
3451 ev_async_send (loop, ev_async *)
3453 Sends/signals/activates the given "ev_async" watcher, that is,
3454 feeds an "EV_ASYNC" event on the watcher into the event loop, and
3455 instantly returns.
3456 Unlike "ev_feed_event", this call is safe to do from other threads,
3458 signal or similar contexts (see the discussion of "EV_ATOMIC_T" in
3459 the embedding section below on what exactly this means).
3460 Note that, as with other watchers in libev, multiple events might
3462 get compressed into a single callback invocation (another way to
3463 look at this is that "ev_async" watchers are level-triggered: they
3464 are set on "ev_async_send", reset when the event loop detects
3465 that).
3466 This call incurs the overhead of at most one extra system call per
3468 event loop iteration, if the event loop is blocked, and no syscall
3469 at all if the event loop (or your program) is processing events.
3470 That means that repeated calls are basically free (there is no need
3471 to avoid calls for performance reasons) and that the overhead
3472 becomes smaller (typically zero) under load.
3473 bool = ev_async_pending (ev_async *)
3475 Returns a non-zero value when "ev_async_send" has been called on
3476 the watcher but the event has not yet been processed (or even
3477 noted) by the event loop.
3478 "ev_async_send" sets a flag in the watcher and wakes up the loop.
3480 When the loop iterates next and checks for the watcher to have
3481 become active, it will reset the flag again. "ev_async_pending" can
3482 be used to very quickly check whether invoking the loop might be a
3483 good idea.
3484 Not that this does not check whether the watcher itself is pending,
3486 only whether it has been requested to make this watcher pending:
3487 there is a time window between the event loop checking and
3488 resetting the async notification, and the callback being invoked.
3489
3491 There are some other functions of possible interest. Described. Here.
3492 Now.
3493 ev_once (loop, int fd, int events, ev_tstamp timeout, callback, arg)
3495 This function combines a simple timer and an I/O watcher, calls
3496 your callback on whichever event happens first and automatically
3497 stops both watchers. This is useful if you want to wait for a
3498 single event on an fd or timeout without having to
3499 allocate/configure/start/stop/free one or more watchers yourself.
3500 If "fd" is less than 0, then no I/O watcher will be started and the
3502 "events" argument is being ignored. Otherwise, an "ev_io" watcher
3503 for the given "fd" and "events" set will be created and started.
3504 If "timeout" is less than 0, then no timeout watcher will be
3506 started. Otherwise an "ev_timer" watcher with after = "timeout"
3507 (and repeat = 0) will be started. 0 is a valid timeout.
3508 The callback has the type "void (*cb)(int revents, void *arg)" and
3510 is passed an "revents" set like normal event callbacks (a
3511 combination of "EV_ERROR", "EV_READ", "EV_WRITE" or "EV_TIMER") and
3512 the "arg" value passed to "ev_once". Note that it is possible to
3513 receive both a timeout and an io event at the same time - you
3514 probably should give io events precedence.
3515 Example: wait up to ten seconds for data to appear on STDIN_FILENO.
3517 static void stdin_ready (int revents, void *arg)
3519 {
3520 if (revents & EV_READ)
3521 /* stdin might have data for us, joy! */;
3522 else if (revents & EV_TIMER)
3523 /* doh, nothing entered */;
3524 }
3525 ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
3527 ev_feed_fd_event (loop, int fd, int revents)
3529 Feed an event on the given fd, as if a file descriptor backend
3530 detected the given events.
3531 ev_feed_signal_event (loop, int signum)
3533 Feed an event as if the given signal occurred. See also
3534 "ev_feed_signal", which is async-safe.
3535
3537 This section explains some common idioms that are not immediately
3538 obvious. Note that examples are sprinkled over the whole manual, and
3539 this section only contains stuff that wouldn't fit anywhere else.
3540 ASSOCIATING CUSTOM DATA WITH A WATCHER
3542 Each watcher has, by default, a "void *data" member that you can read
3543 or modify at any time: libev will completely ignore it. This can be
3544 used to associate arbitrary data with your watcher. If you need more
3545 data and don't want to allocate memory separately and store a pointer
3546 to it in that data member, you can also "subclass" the watcher type and
3547 provide your own data:
3548 struct my_io
3550 {
3551 ev_io io;
3552 int otherfd;
3553 void *somedata;
3554 struct whatever *mostinteresting;
3555 };
3556 ...
3558 struct my_io w;
3559 ev_io_init (&w.io, my_cb, fd, EV_READ);
3560 And since your callback will be called with a pointer to the watcher,
3562 you can cast it back to your own type:
3563 static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
3565 {
3566 struct my_io *w = (struct my_io *)w_;
3567 ...
3568 }
3569 More interesting and less C-conformant ways of casting your callback
3571 function type instead have been omitted.
3572 BUILDING YOUR OWN COMPOSITE WATCHERS
3574 Another common scenario is to use some data structure with multiple
3575 embedded watchers, in effect creating your own watcher that combines
3576 multiple libev event sources into one "super-watcher":
3577 struct my_biggy
3579 {
3580 int some_data;
3581 ev_timer t1;
3582 ev_timer t2;
3583 }
3584 In this case getting the pointer to "my_biggy" is a bit more
3586 complicated: Either you store the address of your "my_biggy" struct in
3587 the "data" member of the watcher (for woozies or C++ coders), or you
3588 need to use some pointer arithmetic using "offsetof" inside your
3589 watchers (for real programmers):
3590 #include <stddef.h>
3592 static void
3594 t1_cb (EV_P_ ev_timer *w, int revents)
3595 {
3596 struct my_biggy big = (struct my_biggy *)
3597 (((char *)w) - offsetof (struct my_biggy, t1));
3598 }
3599 static void
3601 t2_cb (EV_P_ ev_timer *w, int revents)
3602 {
3603 struct my_biggy big = (struct my_biggy *)
3604 (((char *)w) - offsetof (struct my_biggy, t2));
3605 }
3606 AVOIDING FINISHING BEFORE RETURNING
3608 Often you have structures like this in event-based programs:
3609 callback ()
3611 {
3612 free (request);
3613 }
3614 request = start_new_request (..., callback);
3616 The intent is to start some "lengthy" operation. The "request" could be
3618 used to cancel the operation, or do other things with it.
3619 It's not uncommon to have code paths in "start_new_request" that
3621 immediately invoke the callback, for example, to report errors. Or you
3622 add some caching layer that finds that it can skip the lengthy aspects
3623 of the operation and simply invoke the callback with the result.
3624 The problem here is that this will happen before "start_new_request"
3626 has returned, so "request" is not set.
3627 Even if you pass the request by some safer means to the callback, you
3629 might want to do something to the request after starting it, such as
3630 canceling it, which probably isn't working so well when the callback
3631 has already been invoked.
3632 A common way around all these issues is to make sure that
3634 "start_new_request" always returns before the callback is invoked. If
3635 "start_new_request" immediately knows the result, it can artificially
3636 delay invoking the callback by using a "prepare" or "idle" watcher for
3637 example, or more sneakily, by reusing an existing (stopped) watcher and
3638 pushing it into the pending queue:
3639 ev_set_cb (watcher, callback);
3641 ev_feed_event (EV_A_ watcher, 0);
3642 This way, "start_new_request" can safely return before the callback is
3644 invoked, while not delaying callback invocation too much.
3645 MODEL/NESTED EVENT LOOP INVOCATIONS AND EXIT CONDITIONS
3647 Often (especially in GUI toolkits) there are places where you have
3648 modal interaction, which is most easily implemented by recursively
3649 invoking "ev_run".
3650 This brings the problem of exiting - a callback might want to finish
3652 the main "ev_run" call, but not the nested one (e.g. user clicked
3653 "Quit", but a modal "Are you sure?" dialog is still waiting), or just
3654 the nested one and not the main one (e.g. user clocked "Ok" in a modal
3655 dialog), or some other combination: In these cases, a simple "ev_break"
3656 will not work.
3657 The solution is to maintain "break this loop" variable for each
3659 "ev_run" invocation, and use a loop around "ev_run" until the condition
3660 is triggered, using "EVRUN_ONCE":
3661 // main loop
3663 int exit_main_loop = 0;
3664 while (!exit_main_loop)
3666 ev_run (EV_DEFAULT_ EVRUN_ONCE);
3667 // in a modal watcher
3669 int exit_nested_loop = 0;
3670 while (!exit_nested_loop)
3672 ev_run (EV_A_ EVRUN_ONCE);
3673 To exit from any of these loops, just set the corresponding exit
3675 variable:
3676 // exit modal loop
3678 exit_nested_loop = 1;
3679 // exit main program, after modal loop is finished
3681 exit_main_loop = 1;
3682 // exit both
3684 exit_main_loop = exit_nested_loop = 1;
3685 THREAD LOCKING EXAMPLE
3687 Here is a fictitious example of how to run an event loop in a different
3688 thread from where callbacks are being invoked and watchers are
3689 created/added/removed.
3690 For a real-world example, see the "EV::Loop::Async" perl module, which
3692 uses exactly this technique (which is suited for many high-level
3693 languages).
3694 The example uses a pthread mutex to protect the loop data, a condition
3696 variable to wait for callback invocations, an async watcher to notify
3697 the event loop thread and an unspecified mechanism to wake up the main
3698 thread.
3699 First, you need to associate some data with the event loop:
3701 typedef struct {
3703 mutex_t lock; /* global loop lock */
3704 ev_async async_w;
3705 thread_t tid;
3706 cond_t invoke_cv;
3707 } userdata;
3708 void prepare_loop (EV_P)
3710 {
3711 // for simplicity, we use a static userdata struct.
3712 static userdata u;
3713 ev_async_init (&u->async_w, async_cb);
3715 ev_async_start (EV_A_ &u->async_w);
3716 pthread_mutex_init (&u->lock, 0);
3718 pthread_cond_init (&u->invoke_cv, 0);
3719 // now associate this with the loop
3721 ev_set_userdata (EV_A_ u);
3722 ev_set_invoke_pending_cb (EV_A_ l_invoke);
3723 ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
3724 // then create the thread running ev_run
3726 pthread_create (&u->tid, 0, l_run, EV_A);
3727 }
3728 The callback for the "ev_async" watcher does nothing: the watcher is
3730 used solely to wake up the event loop so it takes notice of any new
3731 watchers that might have been added:
3732 static void
3734 async_cb (EV_P_ ev_async *w, int revents)
3735 {
3736 // just used for the side effects
3737 }
3738 The "l_release" and "l_acquire" callbacks simply unlock/lock the mutex
3740 protecting the loop data, respectively.
3741 static void
3743 l_release (EV_P)
3744 {
3745 userdata *u = ev_userdata (EV_A);
3746 pthread_mutex_unlock (&u->lock);
3747 }
3748 static void
3750 l_acquire (EV_P)
3751 {
3752 userdata *u = ev_userdata (EV_A);
3753 pthread_mutex_lock (&u->lock);
3754 }
3755 The event loop thread first acquires the mutex, and then jumps straight
3757 into "ev_run":
3758 void *
3760 l_run (void *thr_arg)
3761 {
3762 struct ev_loop *loop = (struct ev_loop *)thr_arg;
3763 l_acquire (EV_A);
3765 pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
3766 ev_run (EV_A_ 0);
3767 l_release (EV_A);
3768 return 0;
3770 }
3771 Instead of invoking all pending watchers, the "l_invoke" callback will
3773 signal the main thread via some unspecified mechanism (signals? pipe
3774 writes? "Async::Interrupt"?) and then waits until all pending watchers
3775 have been called (in a while loop because a) spurious wakeups are
3776 possible and b) skipping inter-thread-communication when there are no
3777 pending watchers is very beneficial):
3778 static void
3780 l_invoke (EV_P)
3781 {
3782 userdata *u = ev_userdata (EV_A);
3783 while (ev_pending_count (EV_A))
3785 {
3786 wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
3787 pthread_cond_wait (&u->invoke_cv, &u->lock);
3788 }
3789 }
3790 Now, whenever the main thread gets told to invoke pending watchers, it
3792 will grab the lock, call "ev_invoke_pending" and then signal the loop
3793 thread to continue:
3794 static void
3796 real_invoke_pending (EV_P)
3797 {
3798 userdata *u = ev_userdata (EV_A);
3799 pthread_mutex_lock (&u->lock);
3801 ev_invoke_pending (EV_A);
3802 pthread_cond_signal (&u->invoke_cv);
3803 pthread_mutex_unlock (&u->lock);
3804 }
3805 Whenever you want to start/stop a watcher or do other modifications to
3807 an event loop, you will now have to lock:
3808 ev_timer timeout_watcher;
3810 userdata *u = ev_userdata (EV_A);
3811 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
3813 pthread_mutex_lock (&u->lock);
3815 ev_timer_start (EV_A_ &timeout_watcher);
3816 ev_async_send (EV_A_ &u->async_w);
3817 pthread_mutex_unlock (&u->lock);
3818 Note that sending the "ev_async" watcher is required because otherwise
3820 an event loop currently blocking in the kernel will have no knowledge
3821 about the newly added timer. By waking up the loop it will pick up any
3822 new watchers in the next event loop iteration.
3823 THREADS, COROUTINES, CONTINUATIONS, QUEUES... INSTEAD OF CALLBACKS
3825 While the overhead of a callback that e.g. schedules a thread is small,
3826 it is still an overhead. If you embed libev, and your main usage is
3827 with some kind of threads or coroutines, you might want to customise
3828 libev so that doesn't need callbacks anymore.
3829 Imagine you have coroutines that you can switch to using a function
3831 "switch_to (coro)", that libev runs in a coroutine called "libev_coro"
3832 and that due to some magic, the currently active coroutine is stored in
3833 a global called "current_coro". Then you can build your own "wait for
3834 libev event" primitive by changing "EV_CB_DECLARE" and "EV_CB_INVOKE"
3835 (note the differing ";" conventions):
3836 #define EV_CB_DECLARE(type) struct my_coro *cb;
3838 #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
3839 That means instead of having a C callback function, you store the
3841 coroutine to switch to in each watcher, and instead of having libev
3842 call your callback, you instead have it switch to that coroutine.
3843 A coroutine might now wait for an event with a function called
3845 "wait_for_event". (the watcher needs to be started, as always, but it
3846 doesn't matter when, or whether the watcher is active or not when this
3847 function is called):
3848 void
3850 wait_for_event (ev_watcher *w)
3851 {
3852 ev_set_cb (w, current_coro);
3853 switch_to (libev_coro);
3854 }
3855 That basically suspends the coroutine inside "wait_for_event" and
3857 continues the libev coroutine, which, when appropriate, switches back
3858 to this or any other coroutine.
3859 You can do similar tricks if you have, say, threads with an event queue
3861 - instead of storing a coroutine, you store the queue object and
3862 instead of switching to a coroutine, you push the watcher onto the
3863 queue and notify any waiters.
3864 To embed libev, see "EMBEDDING", but in short, it's easiest to create
3866 two files, my_ev.h and my_ev.c that include the respective libev files:
3867 // my_ev.h
3869 #define EV_CB_DECLARE(type) struct my_coro *cb;
3870 #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
3871 #include "../libev/ev.h"
3872 // my_ev.c
3874 #define EV_H "my_ev.h"
3875 #include "../libev/ev.c"
3876 And then use my_ev.h when you would normally use ev.h, and compile
3878 my_ev.c into your project. When properly specifying include paths, you
3879 can even use ev.h as header file name directly.
3880
3882 Libev offers a compatibility emulation layer for libevent. It cannot
3883 emulate the internals of libevent, so here are some usage hints:
3884 · Only the libevent-1.4.1-beta API is being emulated.
3886 This was the newest libevent version available when libev was
3888 implemented, and is still mostly unchanged in 2010.
3889 · Use it by including <event.h>, as usual.
3891 · The following members are fully supported: ev_base, ev_callback,
3893 ev_arg, ev_fd, ev_res, ev_events.
3894 · Avoid using ev_flags and the EVLIST_*-macros, while it is
3896 maintained by libev, it does not work exactly the same way as in
3897 libevent (consider it a private API).
3898 · Priorities are not currently supported. Initialising priorities
3900 will fail and all watchers will have the same priority, even though
3901 there is an ev_pri field.
3902 · In libevent, the last base created gets the signals, in libev, the
3904 base that registered the signal gets the signals.
3905 · Other members are not supported.
3907 · The libev emulation is not ABI compatible to libevent, you need to
3909 use the libev header file and library.
3910
3912 C API
3913 The normal C API should work fine when used from C++: both ev.h and the
3914 libev sources can be compiled as C++. Therefore, code that uses the C
3915 API will work fine.
3916 Proper exception specifications might have to be added to callbacks
3918 passed to libev: exceptions may be thrown only from watcher callbacks,
3919 all other callbacks (allocator, syserr, loop acquire/release and
3920 periodic reschedule callbacks) must not throw exceptions, and might
3921 need a "noexcept" specification. If you have code that needs to be
3922 compiled as both C and C++ you can use the "EV_NOEXCEPT" macro for
3923 this:
3924 static void
3926 fatal_error (const char *msg) EV_NOEXCEPT
3927 {
3928 perror (msg);
3929 abort ();
3930 }
3931 ...
3933 ev_set_syserr_cb (fatal_error);
3934 The only API functions that can currently throw exceptions are
3936 "ev_run", "ev_invoke", "ev_invoke_pending" and "ev_loop_destroy" (the
3937 latter because it runs cleanup watchers).
3938 Throwing exceptions in watcher callbacks is only supported if libev
3940 itself is compiled with a C++ compiler or your C and C++ environments
3941 allow throwing exceptions through C libraries (most do).
3942 C++ API
3944 Libev comes with some simplistic wrapper classes for C++ that mainly
3945 allow you to use some convenience methods to start/stop watchers and
3946 also change the callback model to a model using method callbacks on
3947 objects.
3948 To use it,
3950 #include <ev++.h>
3952 This automatically includes ev.h and puts all of its definitions (many
3954 of them macros) into the global namespace. All C++ specific things are
3955 put into the "ev" namespace. It should support all the same embedding
3956 options as ev.h, most notably "EV_MULTIPLICITY".
3957 Care has been taken to keep the overhead low. The only data member the
3959 C++ classes add (compared to plain C-style watchers) is the event loop
3960 pointer that the watcher is associated with (or no additional members
3961 at all if you disable "EV_MULTIPLICITY" when embedding libev).
3962 Currently, functions, static and non-static member functions and
3964 classes with "operator ()" can be used as callbacks. Other types should
3965 be easy to add as long as they only need one additional pointer for
3966 context. If you need support for other types of functors please contact
3967 the author (preferably after implementing it).
3968 For all this to work, your C++ compiler either has to use the same
3970 calling conventions as your C compiler (for static member functions),
3971 or you have to embed libev and compile libev itself as C++.
3972 Here is a list of things available in the "ev" namespace:
3974 "ev::READ", "ev::WRITE" etc.
3976 These are just enum values with the same values as the "EV_READ"
3977 etc. macros from ev.h.
3978 "ev::tstamp", "ev::now"
3980 Aliases to the same types/functions as with the "ev_" prefix.
3981 "ev::io", "ev::timer", "ev::periodic", "ev::idle", "ev::sig" etc.
3983 For each "ev_TYPE" watcher in ev.h there is a corresponding class
3984 of the same name in the "ev" namespace, with the exception of
3985 "ev_signal" which is called "ev::sig" to avoid clashes with the
3986 "signal" macro defined by many implementations.
3987 All of those classes have these methods:
3989 ev::TYPE::TYPE ()
3991 ev::TYPE::TYPE (loop)
3992 ev::TYPE::~TYPE
3993 The constructor (optionally) takes an event loop to associate
3994 the watcher with. If it is omitted, it will use "EV_DEFAULT".
3995 The constructor calls "ev_init" for you, which means you have
3997 to call the "set" method before starting it.
3998 It will not set a callback, however: You have to call the
4000 templated "set" method to set a callback before you can start
4001 the watcher.
4002 (The reason why you have to use a method is a limitation in C++
4004 which does not allow explicit template arguments for
4005 constructors).
4006 The destructor automatically stops the watcher if it is active.
4008 w->set<class, &class::method> (object *)
4010 This method sets the callback method to call. The method has to
4011 have a signature of "void (*)(ev_TYPE &, int)", it receives the
4012 watcher as first argument and the "revents" as second. The
4013 object must be given as parameter and is stored in the "data"
4014 member of the watcher.
4015 This method synthesizes efficient thunking code to call your
4017 method from the C callback that libev requires. If your
4018 compiler can inline your callback (i.e. it is visible to it at
4019 the place of the "set" call and your compiler is good :), then
4020 the method will be fully inlined into the thunking function,
4021 making it as fast as a direct C callback.
4022 Example: simple class declaration and watcher initialisation
4024 struct myclass
4026 {
4027 void io_cb (ev::io &w, int revents) { }
4028 }
4029 myclass obj;
4031 ev::io iow;
4032 iow.set <myclass, &myclass::io_cb> (&obj);
4033 w->set (object *)
4035 This is a variation of a method callback - leaving out the
4036 method to call will default the method to "operator ()", which
4037 makes it possible to use functor objects without having to
4038 manually specify the "operator ()" all the time. Incidentally,
4039 you can then also leave out the template argument list.
4040 The "operator ()" method prototype must be "void operator
4042 ()(watcher &w, int revents)".
4043 See the method-"set" above for more details.
4045 Example: use a functor object as callback.
4047 struct myfunctor
4049 {
4050 void operator() (ev::io &w, int revents)
4051 {
4052 ...
4053 }
4054 }
4055 myfunctor f;
4057 ev::io w;
4059 w.set (&f);
4060 w->set<function> (void *data = 0)
4062 Also sets a callback, but uses a static method or plain
4063 function as callback. The optional "data" argument will be
4064 stored in the watcher's "data" member and is free for you to
4065 use.
4066 The prototype of the "function" must be "void (*)(ev::TYPE &w,
4068 int)".
4069 See the method-"set" above for more details.
4071 Example: Use a plain function as callback.
4073 static void io_cb (ev::io &w, int revents) { }
4075 iow.set <io_cb> ();
4076 w->set (loop)
4078 Associates a different "struct ev_loop" with this watcher. You
4079 can only do this when the watcher is inactive (and not pending
4080 either).
4081 w->set ([arguments])
4083 Basically the same as "ev_TYPE_set" (except for "ev::embed"
4084 watchers>), with the same arguments. Either this method or a
4085 suitable start method must be called at least once. Unlike the
4086 C counterpart, an active watcher gets automatically stopped and
4087 restarted when reconfiguring it with this method.
4088 For "ev::embed" watchers this method is called "set_embed", to
4090 avoid clashing with the "set (loop)" method.
4091 w->start ()
4093 Starts the watcher. Note that there is no "loop" argument, as
4094 the constructor already stores the event loop.
4095 w->start ([arguments])
4097 Instead of calling "set" and "start" methods separately, it is
4098 often convenient to wrap them in one call. Uses the same type
4099 of arguments as the configure "set" method of the watcher.
4100 w->stop ()
4102 Stops the watcher if it is active. Again, no "loop" argument.
4103 w->again () ("ev::timer", "ev::periodic" only)
4105 For "ev::timer" and "ev::periodic", this invokes the
4106 corresponding "ev_TYPE_again" function.
4107 w->sweep () ("ev::embed" only)
4109 Invokes "ev_embed_sweep".
4110 w->update () ("ev::stat" only)
4112 Invokes "ev_stat_stat".
4113 Example: Define a class with two I/O and idle watchers, start the I/O
4115 watchers in the constructor.
4116 class myclass
4118 {
4119 ev::io io ; void io_cb (ev::io &w, int revents);
4120 ev::io io2 ; void io2_cb (ev::io &w, int revents);
4121 ev::idle idle; void idle_cb (ev::idle &w, int revents);
4122 myclass (int fd)
4124 {
4125 io .set <myclass, &myclass::io_cb > (this);
4126 io2 .set <myclass, &myclass::io2_cb > (this);
4127 idle.set <myclass, &myclass::idle_cb> (this);
4128 io.set (fd, ev::WRITE); // configure the watcher
4130 io.start (); // start it whenever convenient
4131 io2.start (fd, ev::READ); // set + start in one call
4133 }
4134 };
4135
4137 Libev does not offer other language bindings itself, but bindings for a
4138 number of languages exist in the form of third-party packages. If you
4139 know any interesting language binding in addition to the ones listed
4140 here, drop me a note.
4141 Perl
4143 The EV module implements the full libev API and is actually used to
4144 test libev. EV is developed together with libev. Apart from the EV
4145 core module, there are additional modules that implement libev-
4146 compatible interfaces to "libadns" ("EV::ADNS", but "AnyEvent::DNS"
4147 is preferred nowadays), "Net::SNMP" ("Net::SNMP::EV") and the
4148 "libglib" event core ("Glib::EV" and "EV::Glib").
4149 It can be found and installed via CPAN, its homepage is at
4151 <http://software.schmorp.de/pkg/EV>.
4152 Python
4154 Python bindings can be found at <http://code.google.com/p/pyev/>.
4155 It seems to be quite complete and well-documented.
4156 Ruby
4158 Tony Arcieri has written a ruby extension that offers access to a
4159 subset of the libev API and adds file handle abstractions,
4160 asynchronous DNS and more on top of it. It can be found via gem
4161 servers. Its homepage is at <http://rev.rubyforge.org/>.
4162 Roger Pack reports that using the link order "-lws2_32
4164 -lmsvcrt-ruby-190" makes rev work even on mingw.
4165 Haskell
4167 A haskell binding to libev is available at
4168 <http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
4169 D Leandro Lucarella has written a D language binding (ev.d) for
4171 libev, to be found at
4172 <http://www.llucax.com.ar/proj/ev.d/index.html>.
4173 Ocaml
4175 Erkki Seppala has written Ocaml bindings for libev, to be found at
4176 <http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
4177 Lua Brian Maher has written a partial interface to libev for lua (at
4179 the time of this writing, only "ev_io" and "ev_timer"), to be found
4180 at <http://github.com/brimworks/lua-ev>.
4181 Javascript
4183 Node.js (<http://nodejs.org>) uses libev as the underlying event
4184 library.
4185 Others
4187 There are others, and I stopped counting.
4188
4190 Libev can be compiled with a variety of options, the most fundamental
4191 of which is "EV_MULTIPLICITY". This option determines whether (most)
4192 functions and callbacks have an initial "struct ev_loop *" argument.
4193 To make it easier to write programs that cope with either variant, the
4195 following macros are defined:
4196 "EV_A", "EV_A_"
4198 This provides the loop argument for functions, if one is required
4199 ("ev loop argument"). The "EV_A" form is used when this is the sole
4200 argument, "EV_A_" is used when other arguments are following.
4201 Example:
4202 ev_unref (EV_A);
4204 ev_timer_add (EV_A_ watcher);
4205 ev_run (EV_A_ 0);
4206 It assumes the variable "loop" of type "struct ev_loop *" is in
4208 scope, which is often provided by the following macro.
4209 "EV_P", "EV_P_"
4211 This provides the loop parameter for functions, if one is required
4212 ("ev loop parameter"). The "EV_P" form is used when this is the
4213 sole parameter, "EV_P_" is used when other parameters are
4214 following. Example:
4215 // this is how ev_unref is being declared
4217 static void ev_unref (EV_P);
4218 // this is how you can declare your typical callback
4220 static void cb (EV_P_ ev_timer *w, int revents)
4221 It declares a parameter "loop" of type "struct ev_loop *", quite
4223 suitable for use with "EV_A".
4224 "EV_DEFAULT", "EV_DEFAULT_"
4226 Similar to the other two macros, this gives you the value of the
4227 default loop, if multiple loops are supported ("ev loop default").
4228 The default loop will be initialised if it isn't already
4229 initialised.
4230 For non-multiplicity builds, these macros do nothing, so you always
4232 have to initialise the loop somewhere.
4233 "EV_DEFAULT_UC", "EV_DEFAULT_UC_"
4235 Usage identical to "EV_DEFAULT" and "EV_DEFAULT_", but requires
4236 that the default loop has been initialised ("UC" == unchecked).
4237 Their behaviour is undefined when the default loop has not been
4238 initialised by a previous execution of "EV_DEFAULT", "EV_DEFAULT_"
4239 or "ev_default_init (...)".
4240 It is often prudent to use "EV_DEFAULT" when initialising the first
4242 watcher in a function but use "EV_DEFAULT_UC" afterwards.
4243 Example: Declare and initialise a check watcher, utilising the above
4245 macros so it will work regardless of whether multiple loops are
4246 supported or not.
4247 static void
4249 check_cb (EV_P_ ev_timer *w, int revents)
4250 {
4251 ev_check_stop (EV_A_ w);
4252 }
4253 ev_check check;
4255 ev_check_init (&check, check_cb);
4256 ev_check_start (EV_DEFAULT_ &check);
4257 ev_run (EV_DEFAULT_ 0);
4258
4260 Libev can (and often is) directly embedded into host applications.
4261 Examples of applications that embed it include the Deliantra Game
4262 Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe) and
4263 rxvt-unicode.
4264 The goal is to enable you to just copy the necessary files into your
4266 source directory without having to change even a single line in them,
4267 so you can easily upgrade by simply copying (or having a checked-out
4268 copy of libev somewhere in your source tree).
4269 FILESETS
4271 Depending on what features you need you need to include one or more
4272 sets of files in your application.
4273 CORE EVENT LOOP
4275 To include only the libev core (all the "ev_*" functions), with manual
4277 configuration (no autoconf):
4278 #define EV_STANDALONE 1
4280 #include "ev.c"
4281 This will automatically include ev.h, too, and should be done in a
4283 single C source file only to provide the function implementations. To
4284 use it, do the same for ev.h in all files wishing to use this API (best
4285 done by writing a wrapper around ev.h that you can include instead and
4286 where you can put other configuration options):
4287 #define EV_STANDALONE 1
4289 #include "ev.h"
4290 Both header files and implementation files can be compiled with a C++
4292 compiler (at least, that's a stated goal, and breakage will be treated
4293 as a bug).
4294 You need the following files in your source tree, or in a directory in
4296 your include path (e.g. in libev/ when using -Ilibev):
4297 ev.h
4299 ev.c
4300 ev_vars.h
4301 ev_wrap.h
4302 ev_win32.c required on win32 platforms only
4304 ev_select.c only when select backend is enabled
4306 ev_poll.c only when poll backend is enabled
4307 ev_epoll.c only when the epoll backend is enabled
4308 ev_linuxaio.c only when the linux aio backend is enabled
4309 ev_kqueue.c only when the kqueue backend is enabled
4310 ev_port.c only when the solaris port backend is enabled
4311 ev.c includes the backend files directly when enabled, so you only need
4313 to compile this single file.
4314 LIBEVENT COMPATIBILITY API
4316 To include the libevent compatibility API, also include:
4318 #include "event.c"
4320 in the file including ev.c, and:
4322 #include "event.h"
4324 in the files that want to use the libevent API. This also includes
4326 ev.h.
4327 You need the following additional files for this:
4329 event.h
4331 event.c
4332 AUTOCONF SUPPORT
4334 Instead of using "EV_STANDALONE=1" and providing your configuration in
4336 whatever way you want, you can also "m4_include([libev.m4])" in your
4337 configure.ac and leave "EV_STANDALONE" undefined. ev.c will then
4338 include config.h and configure itself accordingly.
4339 For this of course you need the m4 file:
4341 libev.m4
4343 PREPROCESSOR SYMBOLS/MACROS
4345 Libev can be configured via a variety of preprocessor symbols you have
4346 to define before including (or compiling) any of its files. The default
4347 in the absence of autoconf is documented for every option.
4348 Symbols marked with "(h)" do not change the ABI, and can have different
4350 values when compiling libev vs. including ev.h, so it is permissible to
4351 redefine them before including ev.h without breaking compatibility to a
4352 compiled library. All other symbols change the ABI, which means all
4353 users of libev and the libev code itself must be compiled with
4354 compatible settings.
4355 EV_COMPAT3 (h)
4357 Backwards compatibility is a major concern for libev. This is why
4358 this release of libev comes with wrappers for the functions and
4359 symbols that have been renamed between libev version 3 and 4.
4360 You can disable these wrappers (to test compatibility with future
4362 versions) by defining "EV_COMPAT3" to 0 when compiling your
4363 sources. This has the additional advantage that you can drop the
4364 "struct" from "struct ev_loop" declarations, as libev will provide
4365 an "ev_loop" typedef in that case.
4366 In some future version, the default for "EV_COMPAT3" will become 0,
4368 and in some even more future version the compatibility code will be
4369 removed completely.
4370 EV_STANDALONE (h)
4372 Must always be 1 if you do not use autoconf configuration, which
4373 keeps libev from including config.h, and it also defines dummy
4374 implementations for some libevent functions (such as logging, which
4375 is not supported). It will also not define any of the structs
4376 usually found in event.h that are not directly supported by the
4377 libev core alone.
4378 In standalone mode, libev will still try to automatically deduce
4380 the configuration, but has to be more conservative.
4381 EV_USE_FLOOR
4383 If defined to be 1, libev will use the "floor ()" function for its
4384 periodic reschedule calculations, otherwise libev will fall back on
4385 a portable (slower) implementation. If you enable this, you usually
4386 have to link against libm or something equivalent. Enabling this
4387 when the "floor" function is not available will fail, so the safe
4388 default is to not enable this.
4389 EV_USE_MONOTONIC
4391 If defined to be 1, libev will try to detect the availability of
4392 the monotonic clock option at both compile time and runtime.
4393 Otherwise no use of the monotonic clock option will be attempted.
4394 If you enable this, you usually have to link against librt or
4395 something similar. Enabling it when the functionality isn't
4396 available is safe, though, although you have to make sure you link
4397 against any libraries where the "clock_gettime" function is hiding
4398 in (often -lrt). See also "EV_USE_CLOCK_SYSCALL".
4399 EV_USE_REALTIME
4401 If defined to be 1, libev will try to detect the availability of
4402 the real-time clock option at compile time (and assume its
4403 availability at runtime if successful). Otherwise no use of the
4404 real-time clock option will be attempted. This effectively replaces
4405 "gettimeofday" by "clock_get (CLOCK_REALTIME, ...)" and will not
4406 normally affect correctness. See the note about libraries in the
4407 description of "EV_USE_MONOTONIC", though. Defaults to the opposite
4408 value of "EV_USE_CLOCK_SYSCALL".
4409 EV_USE_CLOCK_SYSCALL
4411 If defined to be 1, libev will try to use a direct syscall instead
4412 of calling the system-provided "clock_gettime" function. This
4413 option exists because on GNU/Linux, "clock_gettime" is in "librt",
4414 but "librt" unconditionally pulls in "libpthread", slowing down
4415 single-threaded programs needlessly. Using a direct syscall is
4416 slightly slower (in theory), because no optimised vdso
4417 implementation can be used, but avoids the pthread dependency.
4418 Defaults to 1 on GNU/Linux with glibc 2.x or higher, as it
4419 simplifies linking (no need for "-lrt").
4420 EV_USE_NANOSLEEP
4422 If defined to be 1, libev will assume that "nanosleep ()" is
4423 available and will use it for delays. Otherwise it will use "select
4424 ()".
4425 EV_USE_EVENTFD
4427 If defined to be 1, then libev will assume that "eventfd ()" is
4428 available and will probe for kernel support at runtime. This will
4429 improve "ev_signal" and "ev_async" performance and reduce resource
4430 consumption. If undefined, it will be enabled if the headers
4431 indicate GNU/Linux + Glibc 2.7 or newer, otherwise disabled.
4432 EV_USE_SELECT
4434 If undefined or defined to be 1, libev will compile in support for
4435 the "select"(2) backend. No attempt at auto-detection will be done:
4436 if no other method takes over, select will be it. Otherwise the
4437 select backend will not be compiled in.
4438 EV_SELECT_USE_FD_SET
4440 If defined to 1, then the select backend will use the system
4441 "fd_set" structure. This is useful if libev doesn't compile due to
4442 a missing "NFDBITS" or "fd_mask" definition or it mis-guesses the
4443 bitset layout on exotic systems. This usually limits the range of
4444 file descriptors to some low limit such as 1024 or might have other
4445 limitations (winsocket only allows 64 sockets). The "FD_SETSIZE"
4446 macro, set before compilation, configures the maximum size of the
4447 "fd_set".
4448 EV_SELECT_IS_WINSOCKET
4450 When defined to 1, the select backend will assume that
4451 select/socket/connect etc. don't understand file descriptors but
4452 wants osf handles on win32 (this is the case when the select to be
4453 used is the winsock select). This means that it will call
4454 "_get_osfhandle" on the fd to convert it to an OS handle.
4455 Otherwise, it is assumed that all these functions actually work on
4456 fds, even on win32. Should not be defined on non-win32 platforms.
4457 EV_FD_TO_WIN32_HANDLE(fd)
4459 If "EV_SELECT_IS_WINSOCKET" is enabled, then libev needs a way to
4460 map file descriptors to socket handles. When not defining this
4461 symbol (the default), then libev will call "_get_osfhandle", which
4462 is usually correct. In some cases, programs use their own file
4463 descriptor management, in which case they can provide this function
4464 to map fds to socket handles.
4465 EV_WIN32_HANDLE_TO_FD(handle)
4467 If "EV_SELECT_IS_WINSOCKET" then libev maps handles to file
4468 descriptors using the standard "_open_osfhandle" function. For
4469 programs implementing their own fd to handle mapping, overwriting
4470 this function makes it easier to do so. This can be done by
4471 defining this macro to an appropriate value.
4472 EV_WIN32_CLOSE_FD(fd)
4474 If programs implement their own fd to handle mapping on win32, then
4475 this macro can be used to override the "close" function, useful to
4476 unregister file descriptors again. Note that the replacement
4477 function has to close the underlying OS handle.
4478 EV_USE_WSASOCKET
4480 If defined to be 1, libev will use "WSASocket" to create its
4481 internal communication socket, which works better in some
4482 environments. Otherwise, the normal "socket" function will be used,
4483 which works better in other environments.
4484 EV_USE_POLL
4486 If defined to be 1, libev will compile in support for the "poll"(2)
4487 backend. Otherwise it will be enabled on non-win32 platforms. It
4488 takes precedence over select.
4489 EV_USE_EPOLL
4491 If defined to be 1, libev will compile in support for the Linux
4492 "epoll"(7) backend. Its availability will be detected at runtime,
4493 otherwise another method will be used as fallback. This is the
4494 preferred backend for GNU/Linux systems. If undefined, it will be
4495 enabled if the headers indicate GNU/Linux + Glibc 2.4 or newer,
4496 otherwise disabled.
4497 EV_USE_LINUXAIO
4499 If defined to be 1, libev will compile in support for the Linux aio
4500 backend. Due to it's currenbt limitations it has to be requested
4501 explicitly. If undefined, it will be enabled on linux, otherwise
4502 disabled.
4503 EV_USE_KQUEUE
4505 If defined to be 1, libev will compile in support for the BSD style
4506 "kqueue"(2) backend. Its actual availability will be detected at
4507 runtime, otherwise another method will be used as fallback. This is
4508 the preferred backend for BSD and BSD-like systems, although on
4509 most BSDs kqueue only supports some types of fds correctly (the
4510 only platform we found that supports ptys for example was NetBSD),
4511 so kqueue might be compiled in, but not be used unless explicitly
4512 requested. The best way to use it is to find out whether kqueue
4513 supports your type of fd properly and use an embedded kqueue loop.
4514 EV_USE_PORT
4516 If defined to be 1, libev will compile in support for the Solaris
4517 10 port style backend. Its availability will be detected at
4518 runtime, otherwise another method will be used as fallback. This is
4519 the preferred backend for Solaris 10 systems.
4520 EV_USE_DEVPOLL
4522 Reserved for future expansion, works like the USE symbols above.
4523 EV_USE_INOTIFY
4525 If defined to be 1, libev will compile in support for the Linux
4526 inotify interface to speed up "ev_stat" watchers. Its actual
4527 availability will be detected at runtime. If undefined, it will be
4528 enabled if the headers indicate GNU/Linux + Glibc 2.4 or newer,
4529 otherwise disabled.
4530 EV_NO_SMP
4532 If defined to be 1, libev will assume that memory is always
4533 coherent between threads, that is, threads can be used, but threads
4534 never run on different cpus (or different cpu cores). This reduces
4535 dependencies and makes libev faster.
4536 EV_NO_THREADS
4538 If defined to be 1, libev will assume that it will never be called
4539 from different threads (that includes signal handlers), which is a
4540 stronger assumption than "EV_NO_SMP", above. This reduces
4541 dependencies and makes libev faster.
4542 EV_ATOMIC_T
4544 Libev requires an integer type (suitable for storing 0 or 1) whose
4545 access is atomic with respect to other threads or signal contexts.
4546 No such type is easily found in the C language, so you can provide
4547 your own type that you know is safe for your purposes. It is used
4548 both for signal handler "locking" as well as for signal and thread
4549 safety in "ev_async" watchers.
4550 In the absence of this define, libev will use "sig_atomic_t
4552 volatile" (from signal.h), which is usually good enough on most
4553 platforms.
4554 EV_H (h)
4556 The name of the ev.h header file used to include it. The default if
4557 undefined is "ev.h" in event.h, ev.c and ev++.h. This can be used
4558 to virtually rename the ev.h header file in case of conflicts.
4559 EV_CONFIG_H (h)
4561 If "EV_STANDALONE" isn't 1, this variable can be used to override
4562 ev.c's idea of where to find the config.h file, similarly to
4563 "EV_H", above.
4564 EV_EVENT_H (h)
4566 Similarly to "EV_H", this macro can be used to override event.c's
4567 idea of how the event.h header can be found, the default is
4568 "event.h".
4569 EV_PROTOTYPES (h)
4571 If defined to be 0, then ev.h will not define any function
4572 prototypes, but still define all the structs and other symbols.
4573 This is occasionally useful if you want to provide your own wrapper
4574 functions around libev functions.
4575 EV_MULTIPLICITY
4577 If undefined or defined to 1, then all event-loop-specific
4578 functions will have the "struct ev_loop *" as first argument, and
4579 you can create additional independent event loops. Otherwise there
4580 will be no support for multiple event loops and there is no first
4581 event loop pointer argument. Instead, all functions act on the
4582 single default loop.
4583 Note that "EV_DEFAULT" and "EV_DEFAULT_" will no longer provide a
4585 default loop when multiplicity is switched off - you always have to
4586 initialise the loop manually in this case.
4587 EV_MINPRI
4589 EV_MAXPRI
4590 The range of allowed priorities. "EV_MINPRI" must be smaller or
4591 equal to "EV_MAXPRI", but otherwise there are no non-obvious
4592 limitations. You can provide for more priorities by overriding
4593 those symbols (usually defined to be "-2" and 2, respectively).
4594 When doing priority-based operations, libev usually has to linearly
4596 search all the priorities, so having many of them (hundreds) uses a
4597 lot of space and time, so using the defaults of five priorities (-2
4598 .. +2) is usually fine.
4599 If your embedding application does not need any priorities,
4601 defining these both to 0 will save some memory and CPU.
4602 EV_PERIODIC_ENABLE, EV_IDLE_ENABLE, EV_EMBED_ENABLE, EV_STAT_ENABLE,
4604 EV_PREPARE_ENABLE, EV_CHECK_ENABLE, EV_FORK_ENABLE, EV_SIGNAL_ENABLE,
4605 EV_ASYNC_ENABLE, EV_CHILD_ENABLE.
4606 If undefined or defined to be 1 (and the platform supports it),
4607 then the respective watcher type is supported. If defined to be 0,
4608 then it is not. Disabling watcher types mainly saves code size.
4609 EV_FEATURES
4611 If you need to shave off some kilobytes of code at the expense of
4612 some speed (but with the full API), you can define this symbol to
4613 request certain subsets of functionality. The default is to enable
4614 all features that can be enabled on the platform.
4615 A typical way to use this symbol is to define it to 0 (or to a
4617 bitset with some broad features you want) and then selectively re-
4618 enable additional parts you want, for example if you want
4619 everything minimal, but multiple event loop support, async and
4620 child watchers and the poll backend, use this:
4621 #define EV_FEATURES 0
4623 #define EV_MULTIPLICITY 1
4624 #define EV_USE_POLL 1
4625 #define EV_CHILD_ENABLE 1
4626 #define EV_ASYNC_ENABLE 1
4627 The actual value is a bitset, it can be a combination of the
4629 following values (by default, all of these are enabled):
4630 1 - faster/larger code
4632 Use larger code to speed up some operations.
4633 Currently this is used to override some inlining decisions
4635 (enlarging the code size by roughly 30% on amd64).
4636 When optimising for size, use of compiler flags such as "-Os"
4638 with gcc is recommended, as well as "-DNDEBUG", as libev
4639 contains a number of assertions.
4640 The default is off when "__OPTIMIZE_SIZE__" is defined by your
4642 compiler (e.g. gcc with "-Os").
4643 2 - faster/larger data structures
4645 Replaces the small 2-heap for timer management by a faster
4646 4-heap, larger hash table sizes and so on. This will usually
4647 further increase code size and can additionally have an effect
4648 on the size of data structures at runtime.
4649 The default is off when "__OPTIMIZE_SIZE__" is defined by your
4651 compiler (e.g. gcc with "-Os").
4652 4 - full API configuration
4654 This enables priorities (sets "EV_MAXPRI"=2 and
4655 "EV_MINPRI"=-2), and enables multiplicity
4656 ("EV_MULTIPLICITY"=1).
4657 8 - full API
4659 This enables a lot of the "lesser used" API functions. See
4660 "ev.h" for details on which parts of the API are still
4661 available without this feature, and do not complain if this
4662 subset changes over time.
4663 16 - enable all optional watcher types
4665 Enables all optional watcher types. If you want to selectively
4666 enable only some watcher types other than I/O and timers (e.g.
4667 prepare, embed, async, child...) you can enable them manually
4668 by defining "EV_watchertype_ENABLE" to 1 instead.
4669 32 - enable all backends
4671 This enables all backends - without this feature, you need to
4672 enable at least one backend manually ("EV_USE_SELECT" is a good
4673 choice).
4674 64 - enable OS-specific "helper" APIs
4676 Enable inotify, eventfd, signalfd and similar OS-specific
4677 helper APIs by default.
4678 Compiling with "gcc -Os -DEV_STANDALONE -DEV_USE_EPOLL=1
4680 -DEV_FEATURES=0" reduces the compiled size of libev from 24.7Kb
4681 code/2.8Kb data to 6.5Kb code/0.3Kb data on my GNU/Linux amd64
4682 system, while still giving you I/O watchers, timers and monotonic
4683 clock support.
4684 With an intelligent-enough linker (gcc+binutils are intelligent
4686 enough when you use "-Wl,--gc-sections -ffunction-sections")
4687 functions unused by your program might be left out as well - a
4688 binary starting a timer and an I/O watcher then might come out at
4689 only 5Kb.
4690 EV_API_STATIC
4692 If this symbol is defined (by default it is not), then all
4693 identifiers will have static linkage. This means that libev will
4694 not export any identifiers, and you cannot link against libev
4695 anymore. This can be useful when you embed libev, only want to use
4696 libev functions in a single file, and do not want its identifiers
4697 to be visible.
4698 To use this, define "EV_API_STATIC" and include ev.c in the file
4700 that wants to use libev.
4701 This option only works when libev is compiled with a C compiler, as
4703 C++ doesn't support the required declaration syntax.
4704 EV_AVOID_STDIO
4706 If this is set to 1 at compiletime, then libev will avoid using
4707 stdio functions (printf, scanf, perror etc.). This will increase
4708 the code size somewhat, but if your program doesn't otherwise
4709 depend on stdio and your libc allows it, this avoids linking in the
4710 stdio library which is quite big.
4711 Note that error messages might become less precise when this option
4713 is enabled.
4714 EV_NSIG
4716 The highest supported signal number, +1 (or, the number of
4717 signals): Normally, libev tries to deduce the maximum number of
4718 signals automatically, but sometimes this fails, in which case it
4719 can be specified. Also, using a lower number than detected (32
4720 should be good for about any system in existence) can save some
4721 memory, as libev statically allocates some 12-24 bytes per signal
4722 number.
4723 EV_PID_HASHSIZE
4725 "ev_child" watchers use a small hash table to distribute workload
4726 by pid. The default size is 16 (or 1 with "EV_FEATURES" disabled),
4727 usually more than enough. If you need to manage thousands of
4728 children you might want to increase this value (must be a power of
4729 two).
4730 EV_INOTIFY_HASHSIZE
4732 "ev_stat" watchers use a small hash table to distribute workload by
4733 inotify watch id. The default size is 16 (or 1 with "EV_FEATURES"
4734 disabled), usually more than enough. If you need to manage
4735 thousands of "ev_stat" watchers you might want to increase this
4736 value (must be a power of two).
4737 EV_USE_4HEAP
4739 Heaps are not very cache-efficient. To improve the cache-efficiency
4740 of the timer and periodics heaps, libev uses a 4-heap when this
4741 symbol is defined to 1. The 4-heap uses more complicated (longer)
4742 code but has noticeably faster performance with many (thousands) of
4743 watchers.
4744 The default is 1, unless "EV_FEATURES" overrides it, in which case
4746 it will be 0.
4747 EV_HEAP_CACHE_AT
4749 Heaps are not very cache-efficient. To improve the cache-efficiency
4750 of the timer and periodics heaps, libev can cache the timestamp
4751 (at) within the heap structure (selected by defining
4752 "EV_HEAP_CACHE_AT" to 1), which uses 8-12 bytes more per watcher
4753 and a few hundred bytes more code, but avoids random read accesses
4754 on heap changes. This improves performance noticeably with many
4755 (hundreds) of watchers.
4756 The default is 1, unless "EV_FEATURES" overrides it, in which case
4758 it will be 0.
4759 EV_VERIFY
4761 Controls how much internal verification (see "ev_verify ()") will
4762 be done: If set to 0, no internal verification code will be
4763 compiled in. If set to 1, then verification code will be compiled
4764 in, but not called. If set to 2, then the internal verification
4765 code will be called once per loop, which can slow down libev. If
4766 set to 3, then the verification code will be called very
4767 frequently, which will slow down libev considerably.
4768 Verification errors are reported via C's "assert" mechanism, so if
4770 you disable that (e.g. by defining "NDEBUG") then no errors will be
4771 reported.
4772 The default is 1, unless "EV_FEATURES" overrides it, in which case
4774 it will be 0.
4775 EV_COMMON
4777 By default, all watchers have a "void *data" member. By redefining
4778 this macro to something else you can include more and other types
4779 of members. You have to define it each time you include one of the
4780 files, though, and it must be identical each time.
4781 For example, the perl EV module uses something like this:
4783 #define EV_COMMON \
4785 SV *self; /* contains this struct */ \
4786 SV *cb_sv, *fh /* note no trailing ";" */
4787 EV_CB_DECLARE (type)
4789 EV_CB_INVOKE (watcher, revents)
4790 ev_set_cb (ev, cb)
4791 Can be used to change the callback member declaration in each
4792 watcher, and the way callbacks are invoked and set. Must expand to
4793 a struct member definition and a statement, respectively. See the
4794 ev.h header file for their default definitions. One possible use
4795 for overriding these is to avoid the "struct ev_loop *" as first
4796 argument in all cases, or to use method calls instead of plain
4797 function calls in C++.
4798 EXPORTED API SYMBOLS
4800 If you need to re-export the API (e.g. via a DLL) and you need a list
4801 of exported symbols, you can use the provided Symbol.* files which list
4802 all public symbols, one per line:
4803 Symbols.ev for libev proper
4805 Symbols.event for the libevent emulation
4806 This can also be used to rename all public symbols to avoid clashes
4808 with multiple versions of libev linked together (which is obviously bad
4809 in itself, but sometimes it is inconvenient to avoid this).
4810 A sed command like this will create wrapper "#define"'s that you need
4812 to include before including ev.h:
4813 <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
4815 This would create a file wrap.h which essentially looks like this:
4817 #define ev_backend myprefix_ev_backend
4819 #define ev_check_start myprefix_ev_check_start
4820 #define ev_check_stop myprefix_ev_check_stop
4821 ...
4822 EXAMPLES
4824 For a real-world example of a program the includes libev verbatim, you
4825 can have a look at the EV perl module
4826 (<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
4827 the libev/ subdirectory and includes them in the EV/EVAPI.h (public
4828 interface) and EV.xs (implementation) files. Only the EV.xs file will
4829 be compiled. It is pretty complex because it provides its own header
4830 file.
4831 The usage in rxvt-unicode is simpler. It has a ev_cpp.h header file
4833 that everybody includes and which overrides some configure choices:
4834 #define EV_FEATURES 8
4836 #define EV_USE_SELECT 1
4837 #define EV_PREPARE_ENABLE 1
4838 #define EV_IDLE_ENABLE 1
4839 #define EV_SIGNAL_ENABLE 1
4840 #define EV_CHILD_ENABLE 1
4841 #define EV_USE_STDEXCEPT 0
4842 #define EV_CONFIG_H <config.h>
4843 #include "ev++.h"
4845 And a ev_cpp.C implementation file that contains libev proper and is
4847 compiled:
4848 #include "ev_cpp.h"
4850 #include "ev.c"
4851
4853 THREADS AND COROUTINES
4854 THREADS
4855 All libev functions are reentrant and thread-safe unless explicitly
4857 documented otherwise, but libev implements no locking itself. This
4858 means that you can use as many loops as you want in parallel, as long
4859 as there are no concurrent calls into any libev function with the same
4860 loop parameter ("ev_default_*" calls have an implicit default loop
4861 parameter, of course): libev guarantees that different event loops
4862 share no data structures that need any locking.
4863 Or to put it differently: calls with different loop parameters can be
4865 done concurrently from multiple threads, calls with the same loop
4866 parameter must be done serially (but can be done from different
4867 threads, as long as only one thread ever is inside a call at any point
4868 in time, e.g. by using a mutex per loop).
4869 Specifically to support threads (and signal handlers), libev implements
4871 so-called "ev_async" watchers, which allow some limited form of
4872 concurrency on the same event loop, namely waking it up "from the
4873 outside".
4874 If you want to know which design (one loop, locking, or multiple loops
4876 without or something else still) is best for your problem, then I
4877 cannot help you, but here is some generic advice:
4878 · most applications have a main thread: use the default libev loop in
4880 that thread, or create a separate thread running only the default
4881 loop.
4882 This helps integrating other libraries or software modules that use
4884 libev themselves and don't care/know about threading.
4885 · one loop per thread is usually a good model.
4887 Doing this is almost never wrong, sometimes a better-performance
4889 model exists, but it is always a good start.
4890 · other models exist, such as the leader/follower pattern, where one
4892 loop is handed through multiple threads in a kind of round-robin
4893 fashion.
4894 Choosing a model is hard - look around, learn, know that usually
4896 you can do better than you currently do :-)
4897 · often you need to talk to some other thread which blocks in the
4899 event loop.
4900 "ev_async" watchers can be used to wake them up from other threads
4902 safely (or from signal contexts...).
4903 An example use would be to communicate signals or other events that
4905 only work in the default loop by registering the signal watcher
4906 with the default loop and triggering an "ev_async" watcher from the
4907 default loop watcher callback into the event loop interested in the
4908 signal.
4909 See also "THREAD LOCKING EXAMPLE".
4911 COROUTINES
4913 Libev is very accommodating to coroutines ("cooperative threads"):
4915 libev fully supports nesting calls to its functions from different
4916 coroutines (e.g. you can call "ev_run" on the same loop from two
4917 different coroutines, and switch freely between both coroutines running
4918 the loop, as long as you don't confuse yourself). The only exception is
4919 that you must not do this from "ev_periodic" reschedule callbacks.
4920 Care has been taken to ensure that libev does not keep local state
4922 inside "ev_run", and other calls do not usually allow for coroutine
4923 switches as they do not call any callbacks.
4924 COMPILER WARNINGS
4926 Depending on your compiler and compiler settings, you might get no or a
4927 lot of warnings when compiling libev code. Some people are apparently
4928 scared by this.
4929 However, these are unavoidable for many reasons. For one, each compiler
4931 has different warnings, and each user has different tastes regarding
4932 warning options. "Warn-free" code therefore cannot be a goal except
4933 when targeting a specific compiler and compiler-version.
4934 Another reason is that some compiler warnings require elaborate
4936 workarounds, or other changes to the code that make it less clear and
4937 less maintainable.
4938 And of course, some compiler warnings are just plain stupid, or simply
4940 wrong (because they don't actually warn about the condition their
4941 message seems to warn about). For example, certain older gcc versions
4942 had some warnings that resulted in an extreme number of false
4943 positives. These have been fixed, but some people still insist on
4944 making code warn-free with such buggy versions.
4945 While libev is written to generate as few warnings as possible, "warn-
4947 free" code is not a goal, and it is recommended not to build libev with
4948 any compiler warnings enabled unless you are prepared to cope with them
4949 (e.g. by ignoring them). Remember that warnings are just that:
4950 warnings, not errors, or proof of bugs.
4951 VALGRIND
4953 Valgrind has a special section here because it is a popular tool that
4954 is highly useful. Unfortunately, valgrind reports are very hard to
4955 interpret.
4956 If you think you found a bug (memory leak, uninitialised data access
4958 etc.) in libev, then check twice: If valgrind reports something like:
4959 ==2274== definitely lost: 0 bytes in 0 blocks.
4961 ==2274== possibly lost: 0 bytes in 0 blocks.
4962 ==2274== still reachable: 256 bytes in 1 blocks.
4963 Then there is no memory leak, just as memory accounted to global
4965 variables is not a memleak - the memory is still being referenced, and
4966 didn't leak.
4967 Similarly, under some circumstances, valgrind might report kernel bugs
4969 as if it were a bug in libev (e.g. in realloc or in the poll backend,
4970 although an acceptable workaround has been found here), or it might be
4971 confused.
4972 Keep in mind that valgrind is a very good tool, but only a tool. Don't
4974 make it into some kind of religion.
4975 If you are unsure about something, feel free to contact the mailing
4977 list with the full valgrind report and an explanation on why you think
4978 this is a bug in libev (best check the archives, too :). However, don't
4979 be annoyed when you get a brisk "this is no bug" answer and take the
4980 chance of learning how to interpret valgrind properly.
4981 If you need, for some reason, empty reports from valgrind for your
4983 project I suggest using suppression lists.
4984
4986 GNU/LINUX 32 BIT LIMITATIONS
4987 GNU/Linux is the only common platform that supports 64 bit file/large
4988 file interfaces but disables them by default.
4989 That means that libev compiled in the default environment doesn't
4991 support files larger than 2GiB or so, which mainly affects "ev_stat"
4992 watchers.
4993 Unfortunately, many programs try to work around this GNU/Linux issue by
4995 enabling the large file API, which makes them incompatible with the
4996 standard libev compiled for their system.
4997 Likewise, libev cannot enable the large file API itself as this would
4999 suddenly make it incompatible to the default compile time environment,
5000 i.e. all programs not using special compile switches.
5001 OS/X AND DARWIN BUGS
5003 The whole thing is a bug if you ask me - basically any system interface
5004 you touch is broken, whether it is locales, poll, kqueue or even the
5005 OpenGL drivers.
5006 "kqueue" is buggy
5008 The kqueue syscall is broken in all known versions - most versions
5010 support only sockets, many support pipes.
5011 Libev tries to work around this by not using "kqueue" by default on
5013 this rotten platform, but of course you can still ask for it when
5014 creating a loop - embedding a socket-only kqueue loop into a select-
5015 based one is probably going to work well.
5016 "poll" is buggy
5018 Instead of fixing "kqueue", Apple replaced their (working) "poll"
5020 implementation by something calling "kqueue" internally around the
5021 10.5.6 release, so now "kqueue" and "poll" are broken.
5022 Libev tries to work around this by not using "poll" by default on this
5024 rotten platform, but of course you can still ask for it when creating a
5025 loop.
5026 "select" is buggy
5028 All that's left is "select", and of course Apple found a way to fuck
5030 this one up as well: On OS/X, "select" actively limits the number of
5031 file descriptors you can pass in to 1024 - your program suddenly
5032 crashes when you use more.
5033 There is an undocumented "workaround" for this - defining
5035 "_DARWIN_UNLIMITED_SELECT", which libev tries to use, so select should
5036 work on OS/X.
5037 SOLARIS PROBLEMS AND WORKAROUNDS
5039 "errno" reentrancy
5040 The default compile environment on Solaris is unfortunately so thread-
5042 unsafe that you can't even use components/libraries compiled without
5043 "-D_REENTRANT" in a threaded program, which, of course, isn't defined
5044 by default. A valid, if stupid, implementation choice.
5045 If you want to use libev in threaded environments you have to make sure
5047 it's compiled with "_REENTRANT" defined.
5048 Event port backend
5050 The scalable event interface for Solaris is called "event ports".
5052 Unfortunately, this mechanism is very buggy in all major releases. If
5053 you run into high CPU usage, your program freezes or you get a large
5054 number of spurious wakeups, make sure you have all the relevant and
5055 latest kernel patches applied. No, I don't know which ones, but there
5056 are multiple ones to apply, and afterwards, event ports actually work
5057 great.
5058 If you can't get it to work, you can try running the program by setting
5060 the environment variable "LIBEV_FLAGS=3" to only allow "poll" and
5061 "select" backends.
5062 AIX POLL BUG
5064 AIX unfortunately has a broken "poll.h" header. Libev works around this
5065 by trying to avoid the poll backend altogether (i.e. it's not even
5066 compiled in), which normally isn't a big problem as "select" works fine
5067 with large bitsets on AIX, and AIX is dead anyway.
5068 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
5070 General issues
5071 Win32 doesn't support any of the standards (e.g. POSIX) that libev
5073 requires, and its I/O model is fundamentally incompatible with the
5074 POSIX model. Libev still offers limited functionality on this platform
5075 in the form of the "EVBACKEND_SELECT" backend, and only supports socket
5076 descriptors. This only applies when using Win32 natively, not when
5077 using e.g. cygwin. Actually, it only applies to the microsofts own
5078 compilers, as every compiler comes with a slightly differently
5079 broken/incompatible environment.
5080 Lifting these limitations would basically require the full re-
5082 implementation of the I/O system. If you are into this kind of thing,
5083 then note that glib does exactly that for you in a very portable way
5084 (note also that glib is the slowest event library known to man).
5085 There is no supported compilation method available on windows except
5087 embedding it into other applications.
5088 Sensible signal handling is officially unsupported by Microsoft - libev
5090 tries its best, but under most conditions, signals will simply not
5091 work.
5092 Not a libev limitation but worth mentioning: windows apparently doesn't
5094 accept large writes: instead of resulting in a partial write, windows
5095 will either accept everything or return "ENOBUFS" if the buffer is too
5096 large, so make sure you only write small amounts into your sockets
5097 (less than a megabyte seems safe, but this apparently depends on the
5098 amount of memory available).
5099 Due to the many, low, and arbitrary limits on the win32 platform and
5101 the abysmal performance of winsockets, using a large number of sockets
5102 is not recommended (and not reasonable). If your program needs to use
5103 more than a hundred or so sockets, then likely it needs to use a
5104 totally different implementation for windows, as libev offers the POSIX
5105 readiness notification model, which cannot be implemented efficiently
5106 on windows (due to Microsoft monopoly games).
5107 A typical way to use libev under windows is to embed it (see the
5109 embedding section for details) and use the following evwrap.h header
5110 file instead of ev.h:
5111 #define EV_STANDALONE /* keeps ev from requiring config.h */
5113 #define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
5114 #include "ev.h"
5116 And compile the following evwrap.c file into your project (make sure
5118 you do not compile the ev.c or any other embedded source files!):
5119 #include "evwrap.h"
5121 #include "ev.c"
5122 The winsocket "select" function
5124 The winsocket "select" function doesn't follow POSIX in that it
5126 requires socket handles and not socket file descriptors (it is also
5127 extremely buggy). This makes select very inefficient, and also requires
5128 a mapping from file descriptors to socket handles (the Microsoft C
5129 runtime provides the function "_open_osfhandle" for this). See the
5130 discussion of the "EV_SELECT_USE_FD_SET", "EV_SELECT_IS_WINSOCKET" and
5131 "EV_FD_TO_WIN32_HANDLE" preprocessor symbols for more info.
5132 The configuration for a "naked" win32 using the Microsoft runtime
5134 libraries and raw winsocket select is:
5135 #define EV_USE_SELECT 1
5137 #define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
5138 Note that winsockets handling of fd sets is O(n), so you can easily get
5140 a complexity in the O(nX) range when using win32.
5141 Limited number of file descriptors
5143 Windows has numerous arbitrary (and low) limits on things.
5145 Early versions of winsocket's select only supported waiting for a
5147 maximum of 64 handles (probably owning to the fact that all windows
5148 kernels can only wait for 64 things at the same time internally;
5149 Microsoft recommends spawning a chain of threads and wait for 63
5150 handles and the previous thread in each. Sounds great!).
5151 Newer versions support more handles, but you need to define
5153 "FD_SETSIZE" to some high number (e.g. 2048) before compiling the
5154 winsocket select call (which might be in libev or elsewhere, for
5155 example, perl and many other interpreters do their own select emulation
5156 on windows).
5157 Another limit is the number of file descriptors in the Microsoft
5159 runtime libraries, which by default is 64 (there must be a hidden 64
5160 fetish or something like this inside Microsoft). You can increase this
5161 by calling "_setmaxstdio", which can increase this limit to 2048
5162 (another arbitrary limit), but is broken in many versions of the
5163 Microsoft runtime libraries. This might get you to about 512 or 2048
5164 sockets (depending on windows version and/or the phase of the moon). To
5165 get more, you need to wrap all I/O functions and provide your own fd
5166 management, but the cost of calling select (O(nX)) will likely make
5167 this unworkable.
5168 PORTABILITY REQUIREMENTS
5170 In addition to a working ISO-C implementation and of course the
5171 backend-specific APIs, libev relies on a few additional extensions:
5172 "void (*)(ev_watcher_type *, int revents)" must have compatible calling
5174 conventions regardless of "ev_watcher_type *".
5175 Libev assumes not only that all watcher pointers have the same
5176 internal structure (guaranteed by POSIX but not by ISO C for
5177 example), but it also assumes that the same (machine) code can be
5178 used to call any watcher callback: The watcher callbacks have
5179 different type signatures, but libev calls them using an
5180 "ev_watcher *" internally.
5181 null pointers and integer zero are represented by 0 bytes
5183 Libev uses "memset" to initialise structs and arrays to 0 bytes,
5184 and relies on this setting pointers and integers to null.
5185 pointer accesses must be thread-atomic
5187 Accessing a pointer value must be atomic, it must both be readable
5188 and writable in one piece - this is the case on all current
5189 architectures.
5190 "sig_atomic_t volatile" must be thread-atomic as well
5192 The type "sig_atomic_t volatile" (or whatever is defined as
5193 "EV_ATOMIC_T") must be atomic with respect to accesses from
5194 different threads. This is not part of the specification for
5195 "sig_atomic_t", but is believed to be sufficiently portable.
5196 "sigprocmask" must work in a threaded environment
5198 Libev uses "sigprocmask" to temporarily block signals. This is not
5199 allowed in a threaded program ("pthread_sigmask" has to be used).
5200 Typical pthread implementations will either allow "sigprocmask" in
5201 the "main thread" or will block signals process-wide, both
5202 behaviours would be compatible with libev. Interaction between
5203 "sigprocmask" and "pthread_sigmask" could complicate things,
5204 however.
5205 The most portable way to handle signals is to block signals in all
5207 threads except the initial one, and run the signal handling loop in
5208 the initial thread as well.
5209 "long" must be large enough for common memory allocation sizes
5211 To improve portability and simplify its API, libev uses "long"
5212 internally instead of "size_t" when allocating its data structures.
5213 On non-POSIX systems (Microsoft...) this might be unexpectedly low,
5214 but is still at least 31 bits everywhere, which is enough for
5215 hundreds of millions of watchers.
5216 "double" must hold a time value in seconds with enough accuracy
5218 The type "double" is used to represent timestamps. It is required
5219 to have at least 51 bits of mantissa (and 9 bits of exponent),
5220 which is good enough for at least into the year 4000 with
5221 millisecond accuracy (the design goal for libev). This requirement
5222 is overfulfilled by implementations using IEEE 754, which is
5223 basically all existing ones.
5224 With IEEE 754 doubles, you get microsecond accuracy until at least
5226 the year 2255 (and millisecond accuracy till the year 287396 - by
5227 then, libev is either obsolete or somebody patched it to use "long
5228 double" or something like that, just kidding).
5229 If you know of other additional requirements drop me a note.
5231
5233 In this section the complexities of (many of) the algorithms used
5234 inside libev will be documented. For complexity discussions about
5235 backends see the documentation for "ev_default_init".
5236 All of the following are about amortised time: If an array needs to be
5238 extended, libev needs to realloc and move the whole array, but this
5239 happens asymptotically rarer with higher number of elements, so O(1)
5240 might mean that libev does a lengthy realloc operation in rare cases,
5241 but on average it is much faster and asymptotically approaches constant
5242 time.
5243 Starting and stopping timer/periodic watchers: O(log
5245 skipped_other_timers)
5246 This means that, when you have a watcher that triggers in one hour
5247 and there are 100 watchers that would trigger before that, then
5248 inserting will have to skip roughly seven ("ld 100") of these
5249 watchers.
5250 Changing timer/periodic watchers (by autorepeat or calling again):
5252 O(log skipped_other_timers)
5253 That means that changing a timer costs less than removing/adding
5254 them, as only the relative motion in the event queue has to be paid
5255 for.
5256 Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
5258 These just add the watcher into an array or at the head of a list.
5259 Stopping check/prepare/idle/fork/async watchers: O(1)
5261 Stopping an io/signal/child watcher:
5262 O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
5263 These watchers are stored in lists, so they need to be walked to
5264 find the correct watcher to remove. The lists are usually short
5265 (you don't usually have many watchers waiting for the same fd or
5266 signal: one is typical, two is rare).
5267 Finding the next timer in each loop iteration: O(1)
5269 By virtue of using a binary or 4-heap, the next timer is always
5270 found at a fixed position in the storage array.
5271 Each change on a file descriptor per loop iteration:
5273 O(number_of_watchers_for_this_fd)
5274 A change means an I/O watcher gets started or stopped, which
5275 requires libev to recalculate its status (and possibly tell the
5276 kernel, depending on backend and whether "ev_io_set" was used).
5277 Activating one watcher (putting it into the pending state): O(1)
5279 Priority handling: O(number_of_priorities)
5280 Priorities are implemented by allocating some space for each
5281 priority. When doing priority-based operations, libev usually has
5282 to linearly search all the priorities, but starting/stopping and
5283 activating watchers becomes O(1) with respect to priority handling.
5284 Sending an ev_async: O(1)
5286 Processing ev_async_send: O(number_of_async_watchers)
5287 Processing signals: O(max_signal_number)
5288 Sending involves a system call iff there were no other
5289 "ev_async_send" calls in the current loop iteration and the loop is
5290 currently blocked. Checking for async and signal events involves
5291 iterating over all running async watchers or all signal numbers.
5292
5294 The major version 4 introduced some incompatible changes to the API.
5295 At the moment, the "ev.h" header file provides compatibility
5297 definitions for all changes, so most programs should still compile. The
5298 compatibility layer might be removed in later versions of libev, so
5299 better update to the new API early than late.
5300 "EV_COMPAT3" backwards compatibility mechanism
5302 The backward compatibility mechanism can be controlled by
5303 "EV_COMPAT3". See "PREPROCESSOR SYMBOLS/MACROS" in the "EMBEDDING"
5304 section.
5305 "ev_default_destroy" and "ev_default_fork" have been removed
5307 These calls can be replaced easily by their "ev_loop_xxx"
5308 counterparts:
5309 ev_loop_destroy (EV_DEFAULT_UC);
5311 ev_loop_fork (EV_DEFAULT);
5312 function/symbol renames
5314 A number of functions and symbols have been renamed:
5315 ev_loop => ev_run
5317 EVLOOP_NONBLOCK => EVRUN_NOWAIT
5318 EVLOOP_ONESHOT => EVRUN_ONCE
5319 ev_unloop => ev_break
5321 EVUNLOOP_CANCEL => EVBREAK_CANCEL
5322 EVUNLOOP_ONE => EVBREAK_ONE
5323 EVUNLOOP_ALL => EVBREAK_ALL
5324 EV_TIMEOUT => EV_TIMER
5326 ev_loop_count => ev_iteration
5328 ev_loop_depth => ev_depth
5329 ev_loop_verify => ev_verify
5330 Most functions working on "struct ev_loop" objects don't have an
5332 "ev_loop_" prefix, so it was removed; "ev_loop", "ev_unloop" and
5333 associated constants have been renamed to not collide with the
5334 "struct ev_loop" anymore and "EV_TIMER" now follows the same naming
5335 scheme as all other watcher types. Note that "ev_loop_fork" is
5336 still called "ev_loop_fork" because it would otherwise clash with
5337 the "ev_fork" typedef.
5338 "EV_MINIMAL" mechanism replaced by "EV_FEATURES"
5340 The preprocessor symbol "EV_MINIMAL" has been replaced by a
5341 different mechanism, "EV_FEATURES". Programs using "EV_MINIMAL"
5342 usually compile and work, but the library code will of course be
5343 larger.
5344
5346 active
5347 A watcher is active as long as it has been started and not yet
5348 stopped. See "WATCHER STATES" for details.
5349 application
5351 In this document, an application is whatever is using libev.
5352 backend
5354 The part of the code dealing with the operating system interfaces.
5355 callback
5357 The address of a function that is called when some event has been
5358 detected. Callbacks are being passed the event loop, the watcher
5359 that received the event, and the actual event bitset.
5360 callback/watcher invocation
5362 The act of calling the callback associated with a watcher.
5363 event
5365 A change of state of some external event, such as data now being
5366 available for reading on a file descriptor, time having passed or
5367 simply not having any other events happening anymore.
5368 In libev, events are represented as single bits (such as "EV_READ"
5370 or "EV_TIMER").
5371 event library
5373 A software package implementing an event model and loop.
5374 event loop
5376 An entity that handles and processes external events and converts
5377 them into callback invocations.
5378 event model
5380 The model used to describe how an event loop handles and processes
5381 watchers and events.
5382 pending
5384 A watcher is pending as soon as the corresponding event has been
5385 detected. See "WATCHER STATES" for details.
5386 real time
5388 The physical time that is observed. It is apparently strictly
5389 monotonic :)
5390 wall-clock time
5392 The time and date as shown on clocks. Unlike real time, it can
5393 actually be wrong and jump forwards and backwards, e.g. when you
5394 adjust your clock.
5395 watcher
5397 A data structure that describes interest in certain events.
5398 Watchers need to be started (attached to an event loop) before they
5399 can receive events.
5400
5402 Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael
5403 Magnusson and Emanuele Giaquinta, and minor corrections by many others.
5404libev-4.25 2019-06-25 LIBEV(3)