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 "epoll", the BSD-
107 specific "kqueue" and the Solaris-specific event port mechanisms for
108 file descriptor events ("ev_io"), the Linux "inotify" interface (for
109 "ev_stat"), Linux eventfd/signalfd (for faster and cleaner inter-thread
110 wakeup ("ev_async")/signal handling ("ev_signal")) relative timers
111 ("ev_timer"), absolute timers with customised rescheduling
112 ("ev_periodic"), synchronous signals ("ev_signal"), process status
113 change events ("ev_child"), and event watchers dealing with the event
114 loop mechanism itself ("ev_idle", "ev_embed", "ev_prepare" and
115 "ev_check" watchers) as well as file watchers ("ev_stat") and even
116 limited support for fork events ("ev_fork").
117 It also is quite fast (see this benchmark
119 <http://libev.schmorp.de/bench.html> comparing it to libevent for
120 example).
121 CONVENTIONS
123 Libev is very configurable. In this manual the default (and most
124 common) configuration will be described, which supports multiple event
125 loops. For more info about various configuration options please have a
126 look at EMBED section in this manual. If libev was configured without
127 support for multiple event loops, then all functions taking an initial
128 argument of name "loop" (which is always of type "struct ev_loop *")
129 will not have this argument.
130 TIME REPRESENTATION
132 Libev represents time as a single floating point number, representing
133 the (fractional) number of seconds since the (POSIX) epoch (in practice
134 somewhere near the beginning of 1970, details are complicated, don't
135 ask). This type is called "ev_tstamp", which is what you should use
136 too. It usually aliases to the "double" type in C. When you need to do
137 any calculations on it, you should treat it as some floating point
138 value.
139 Unlike the name component "stamp" might indicate, it is also used for
141 time differences (e.g. delays) throughout libev.
142
144 Libev knows three classes of errors: operating system errors, usage
145 errors and internal errors (bugs).
146 When libev catches an operating system error it cannot handle (for
148 example a system call indicating a condition libev cannot fix), it
149 calls the callback set via "ev_set_syserr_cb", which is supposed to fix
150 the problem or abort. The default is to print a diagnostic message and
151 to call "abort ()".
152 When libev detects a usage error such as a negative timer interval,
154 then it will print a diagnostic message and abort (via the "assert"
155 mechanism, so "NDEBUG" will disable this checking): these are
156 programming errors in the libev caller and need to be fixed there.
157 Libev also has a few internal error-checking "assert"ions, and also has
159 extensive consistency checking code. These do not trigger under normal
160 circumstances, as they indicate either a bug in libev or worse.
161
163 These functions can be called anytime, even before initialising the
164 library in any way.
165 ev_tstamp ev_time ()
167 Returns the current time as libev would use it. Please note that
168 the "ev_now" function is usually faster and also often returns the
169 timestamp you actually want to know. Also interesting is the
170 combination of "ev_now_update" and "ev_now".
171 ev_sleep (ev_tstamp interval)
173 Sleep for the given interval: The current thread will be blocked
174 until either it is interrupted or the given time interval has
175 passed (approximately - it might return a bit earlier even if not
176 interrupted). Returns immediately if "interval <= 0".
177 Basically this is a sub-second-resolution "sleep ()".
179 The range of the "interval" is limited - libev only guarantees to
181 work with sleep times of up to one day ("interval <= 86400").
182 int ev_version_major ()
184 int ev_version_minor ()
185 You can find out the major and minor ABI version numbers of the
186 library you linked against by calling the functions
187 "ev_version_major" and "ev_version_minor". If you want, you can
188 compare against the global symbols "EV_VERSION_MAJOR" and
189 "EV_VERSION_MINOR", which specify the version of the library your
190 program was compiled against.
191 These version numbers refer to the ABI version of the library, not
193 the release version.
194 Usually, it's a good idea to terminate if the major versions
196 mismatch, as this indicates an incompatible change. Minor versions
197 are usually compatible to older versions, so a larger minor version
198 alone is usually not a problem.
199 Example: Make sure we haven't accidentally been linked against the
201 wrong version (note, however, that this will not detect other ABI
202 mismatches, such as LFS or reentrancy).
203 assert (("libev version mismatch",
205 ev_version_major () == EV_VERSION_MAJOR
206 && ev_version_minor () >= EV_VERSION_MINOR));
207 unsigned int ev_supported_backends ()
209 Return the set of all backends (i.e. their corresponding
210 "EV_BACKEND_*" value) compiled into this binary of libev
211 (independent of their availability on the system you are running
212 on). See "ev_default_loop" for a description of the set values.
213 Example: make sure we have the epoll method, because yeah this is
215 cool and a must have and can we have a torrent of it please!!!11
216 assert (("sorry, no epoll, no sex",
218 ev_supported_backends () & EVBACKEND_EPOLL));
219 unsigned int ev_recommended_backends ()
221 Return the set of all backends compiled into this binary of libev
222 and also recommended for this platform, meaning it will work for
223 most file descriptor types. This set is often smaller than the one
224 returned by "ev_supported_backends", as for example kqueue is
225 broken on most BSDs and will not be auto-detected unless you
226 explicitly request it (assuming you know what you are doing). This
227 is the set of backends that libev will probe for if you specify no
228 backends explicitly.
229 unsigned int ev_embeddable_backends ()
231 Returns the set of backends that are embeddable in other event
232 loops. This value is platform-specific but can include backends not
233 available on the current system. To find which embeddable backends
234 might be supported on the current system, you would need to look at
235 "ev_embeddable_backends () & ev_supported_backends ()", likewise
236 for recommended ones.
237 See the description of "ev_embed" watchers for more info.
239 ev_set_allocator (void *(*cb)(void *ptr, long size) throw ())
241 Sets the allocation function to use (the prototype is similar - the
242 semantics are identical to the "realloc" C89/SuS/POSIX function).
243 It is used to allocate and free memory (no surprises here). If it
244 returns zero when memory needs to be allocated ("size != 0"), the
245 library might abort or take some potentially destructive action.
246 Since some systems (at least OpenBSD and Darwin) fail to implement
248 correct "realloc" semantics, libev will use a wrapper around the
249 system "realloc" and "free" functions by default.
250 You could override this function in high-availability programs to,
252 say, free some memory if it cannot allocate memory, to use a
253 special allocator, or even to sleep a while and retry until some
254 memory is available.
255 Example: Replace the libev allocator with one that waits a bit and
257 then retries (example requires a standards-compliant "realloc").
258 static void *
260 persistent_realloc (void *ptr, size_t size)
261 {
262 for (;;)
263 {
264 void *newptr = realloc (ptr, size);
265 if (newptr)
267 return newptr;
268 sleep (60);
270 }
271 }
272 ...
274 ev_set_allocator (persistent_realloc);
275 ev_set_syserr_cb (void (*cb)(const char *msg) throw ())
277 Set the callback function to call on a retryable system call error
278 (such as failed select, poll, epoll_wait). The message is a
279 printable string indicating the system call or subsystem causing
280 the problem. If this callback is set, then libev will expect it to
281 remedy the situation, no matter what, when it returns. That is,
282 libev will generally retry the requested operation, or, if the
283 condition doesn't go away, do bad stuff (such as abort).
284 Example: This is basically the same thing that libev does
286 internally, too.
287 static void
289 fatal_error (const char *msg)
290 {
291 perror (msg);
292 abort ();
293 }
294 ...
296 ev_set_syserr_cb (fatal_error);
297 ev_feed_signal (int signum)
299 This function can be used to "simulate" a signal receive. It is
300 completely safe to call this function at any time, from any
301 context, including signal handlers or random threads.
302 Its main use is to customise signal handling in your process,
304 especially in the presence of threads. For example, you could block
305 signals by default in all threads (and specifying
306 "EVFLAG_NOSIGMASK" when creating any loops), and in one thread, use
307 "sigwait" or any other mechanism to wait for signals, then
308 "deliver" them to libev by calling "ev_feed_signal".
309
311 An event loop is described by a "struct ev_loop *" (the "struct" is not
312 optional in this case unless libev 3 compatibility is disabled, as
313 libev 3 had an "ev_loop" function colliding with the struct name).
314 The library knows two types of such loops, the default loop, which
316 supports child process events, and dynamically created event loops
317 which do not.
318 struct ev_loop *ev_default_loop (unsigned int flags)
320 This returns the "default" event loop object, which is what you
321 should normally use when you just need "the event loop". Event loop
322 objects and the "flags" parameter are described in more detail in
323 the entry for "ev_loop_new".
324 If the default loop is already initialised then this function
326 simply returns it (and ignores the flags. If that is troubling you,
327 check "ev_backend ()" afterwards). Otherwise it will create it with
328 the given flags, which should almost always be 0, unless the caller
329 is also the one calling "ev_run" or otherwise qualifies as "the
330 main program".
331 If you don't know what event loop to use, use the one returned from
333 this function (or via the "EV_DEFAULT" macro).
334 Note that this function is not thread-safe, so if you want to use
336 it from multiple threads, you have to employ some kind of mutex
337 (note also that this case is unlikely, as loops cannot be shared
338 easily between threads anyway).
339 The default loop is the only loop that can handle "ev_child"
341 watchers, and to do this, it always registers a handler for
342 "SIGCHLD". If this is a problem for your application you can either
343 create a dynamic loop with "ev_loop_new" which doesn't do that, or
344 you can simply overwrite the "SIGCHLD" signal handler after calling
345 "ev_default_init".
346 Example: This is the most typical usage.
348 if (!ev_default_loop (0))
350 fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
351 Example: Restrict libev to the select and poll backends, and do not
353 allow environment settings to be taken into account:
354 ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
356 struct ev_loop *ev_loop_new (unsigned int flags)
358 This will create and initialise a new event loop object. If the
359 loop could not be initialised, returns false.
360 This function is thread-safe, and one common way to use libev with
362 threads is indeed to create one loop per thread, and using the
363 default loop in the "main" or "initial" thread.
364 The flags argument can be used to specify special behaviour or
366 specific backends to use, and is usually specified as 0 (or
367 "EVFLAG_AUTO").
368 The following flags are supported:
370 "EVFLAG_AUTO"
372 The default flags value. Use this if you have no clue (it's the
373 right thing, believe me).
374 "EVFLAG_NOENV"
376 If this flag bit is or'ed into the flag value (or the program
377 runs setuid or setgid) then libev will not look at the
378 environment variable "LIBEV_FLAGS". Otherwise (the default),
379 this environment variable will override the flags completely if
380 it is found in the environment. This is useful to try out
381 specific backends to test their performance, to work around
382 bugs, or to make libev threadsafe (accessing environment
383 variables cannot be done in a threadsafe way, but usually it
384 works if no other thread modifies them).
385 "EVFLAG_FORKCHECK"
387 Instead of calling "ev_loop_fork" manually after a fork, you
388 can also make libev check for a fork in each iteration by
389 enabling this flag.
390 This works by calling "getpid ()" on every iteration of the
392 loop, and thus this might slow down your event loop if you do a
393 lot of loop iterations and little real work, but is usually not
394 noticeable (on my GNU/Linux system for example, "getpid" is
395 actually a simple 5-insn sequence without a system call and
396 thus very fast, but my GNU/Linux system also has
397 "pthread_atfork" which is even faster).
398 The big advantage of this flag is that you can forget about
400 fork (and forget about forgetting to tell libev about forking,
401 although you still have to ignore "SIGPIPE") when you use this
402 flag.
403 This flag setting cannot be overridden or specified in the
405 "LIBEV_FLAGS" environment variable.
406 "EVFLAG_NOINOTIFY"
408 When this flag is specified, then libev will not attempt to use
409 the inotify API for its "ev_stat" watchers. Apart from
410 debugging and testing, this flag can be useful to conserve
411 inotify file descriptors, as otherwise each loop using
412 "ev_stat" watchers consumes one inotify handle.
413 "EVFLAG_SIGNALFD"
415 When this flag is specified, then libev will attempt to use the
416 signalfd API for its "ev_signal" (and "ev_child") watchers.
417 This API delivers signals synchronously, which makes it both
418 faster and might make it possible to get the queued signal
419 data. It can also simplify signal handling with threads, as
420 long as you properly block signals in your threads that are not
421 interested in handling them.
422 Signalfd will not be used by default as this changes your
424 signal mask, and there are a lot of shoddy libraries and
425 programs (glib's threadpool for example) that can't properly
426 initialise their signal masks.
427 "EVFLAG_NOSIGMASK"
429 When this flag is specified, then libev will avoid to modify
430 the signal mask. Specifically, this means you have to make sure
431 signals are unblocked when you want to receive them.
432 This behaviour is useful when you want to do your own signal
434 handling, or want to handle signals only in specific threads
435 and want to avoid libev unblocking the signals.
436 It's also required by POSIX in a threaded program, as libev
438 calls "sigprocmask", whose behaviour is officially unspecified.
439 This flag's behaviour will become the default in future
441 versions of libev.
442 "EVBACKEND_SELECT" (value 1, portable select backend)
444 This is your standard select(2) backend. Not completely
445 standard, as libev tries to roll its own fd_set with no limits
446 on the number of fds, but if that fails, expect a fairly low
447 limit on the number of fds when using this backend. It doesn't
448 scale too well (O(highest_fd)), but its usually the fastest
449 backend for a low number of (low-numbered :) fds.
450 To get good performance out of this backend you need a high
452 amount of parallelism (most of the file descriptors should be
453 busy). If you are writing a server, you should "accept ()" in a
454 loop to accept as many connections as possible during one
455 iteration. You might also want to have a look at
456 "ev_set_io_collect_interval ()" to increase the amount of
457 readiness notifications you get per iteration.
458 This backend maps "EV_READ" to the "readfds" set and "EV_WRITE"
460 to the "writefds" set (and to work around Microsoft Windows
461 bugs, also onto the "exceptfds" set on that platform).
462 "EVBACKEND_POLL" (value 2, poll backend, available everywhere
464 except on windows)
465 And this is your standard poll(2) backend. It's more
466 complicated than select, but handles sparse fds better and has
467 no artificial limit on the number of fds you can use (except it
468 will slow down considerably with a lot of inactive fds). It
469 scales similarly to select, i.e. O(total_fds). See the entry
470 for "EVBACKEND_SELECT", above, for performance tips.
471 This backend maps "EV_READ" to "POLLIN | POLLERR | POLLHUP",
473 and "EV_WRITE" to "POLLOUT | POLLERR | POLLHUP".
474 "EVBACKEND_EPOLL" (value 4, Linux)
476 Use the linux-specific epoll(7) interface (for both pre- and
477 post-2.6.9 kernels).
478 For few fds, this backend is a bit little slower than poll and
480 select, but it scales phenomenally better. While poll and
481 select usually scale like O(total_fds) where total_fds is the
482 total number of fds (or the highest fd), epoll scales either
483 O(1) or O(active_fds).
484 The epoll mechanism deserves honorable mention as the most
486 misdesigned of the more advanced event mechanisms: mere
487 annoyances include silently dropping file descriptors,
488 requiring a system call per change per file descriptor (and
489 unnecessary guessing of parameters), problems with dup,
490 returning before the timeout value, resulting in additional
491 iterations (and only giving 5ms accuracy while select on the
492 same platform gives 0.1ms) and so on. The biggest issue is fork
493 races, however - if a program forks then both parent and child
494 process have to recreate the epoll set, which can take
495 considerable time (one syscall per file descriptor) and is of
496 course hard to detect.
497 Epoll is also notoriously buggy - embedding epoll fds should
499 work, but of course doesn't, and epoll just loves to report
500 events for totally different file descriptors (even already
501 closed ones, so one cannot even remove them from the set) than
502 registered in the set (especially on SMP systems). Libev tries
503 to counter these spurious notifications by employing an
504 additional generation counter and comparing that against the
505 events to filter out spurious ones, recreating the set when
506 required. Epoll also erroneously rounds down timeouts, but
507 gives you no way to know when and by how much, so sometimes you
508 have to busy-wait because epoll returns immediately despite a
509 nonzero timeout. And last not least, it also refuses to work
510 with some file descriptors which work perfectly fine with
511 "select" (files, many character devices...).
512 Epoll is truly the train wreck among event poll mechanisms, a
514 frankenpoll, cobbled together in a hurry, no thought to design
515 or interaction with others. Oh, the pain, will it ever stop...
516 While stopping, setting and starting an I/O watcher in the same
518 iteration will result in some caching, there is still a system
519 call per such incident (because the same file descriptor could
520 point to a different file description now), so its best to
521 avoid that. Also, "dup ()"'ed file descriptors might not work
522 very well if you register events for both file descriptors.
523 Best performance from this backend is achieved by not
525 unregistering all watchers for a file descriptor until it has
526 been closed, if possible, i.e. keep at least one watcher active
527 per fd at all times. Stopping and starting a watcher (without
528 re-setting it) also usually doesn't cause extra overhead. A
529 fork can both result in spurious notifications as well as in
530 libev having to destroy and recreate the epoll object, which
531 can take considerable time and thus should be avoided.
532 All this means that, in practice, "EVBACKEND_SELECT" can be as
534 fast or faster than epoll for maybe up to a hundred file
535 descriptors, depending on the usage. So sad.
536 While nominally embeddable in other event loops, this feature
538 is broken in all kernel versions tested so far.
539 This backend maps "EV_READ" and "EV_WRITE" in the same way as
541 "EVBACKEND_POLL".
542 "EVBACKEND_KQUEUE" (value 8, most BSD clones)
544 Kqueue deserves special mention, as at the time of this
545 writing, it was broken on all BSDs except NetBSD (usually it
546 doesn't work reliably with anything but sockets and pipes,
547 except on Darwin, where of course it's completely useless).
548 Unlike epoll, however, whose brokenness is by design, these
549 kqueue bugs can (and eventually will) be fixed without API
550 changes to existing programs. For this reason it's not being
551 "auto-detected" unless you explicitly specify it in the flags
552 (i.e. using "EVBACKEND_KQUEUE") or libev was compiled on a
553 known-to-be-good (-enough) system like NetBSD.
554 You still can embed kqueue into a normal poll or select backend
556 and use it only for sockets (after having made sure that
557 sockets work with kqueue on the target platform). See
558 "ev_embed" watchers for more info.
559 It scales in the same way as the epoll backend, but the
561 interface to the kernel is more efficient (which says nothing
562 about its actual speed, of course). While stopping, setting and
563 starting an I/O watcher does never cause an extra system call
564 as with "EVBACKEND_EPOLL", it still adds up to two event
565 changes per incident. Support for "fork ()" is very bad (you
566 might have to leak fd's on fork, but it's more sane than epoll)
567 and it drops fds silently in similarly hard-to-detect cases.
568 This backend usually performs well under most conditions.
570 While nominally embeddable in other event loops, this doesn't
572 work everywhere, so you might need to test for this. And since
573 it is broken almost everywhere, you should only use it when you
574 have a lot of sockets (for which it usually works), by
575 embedding it into another event loop (e.g. "EVBACKEND_SELECT"
576 or "EVBACKEND_POLL" (but "poll" is of course also broken on OS
577 X)) and, did I mention it, using it only for sockets.
578 This backend maps "EV_READ" into an "EVFILT_READ" kevent with
580 "NOTE_EOF", and "EV_WRITE" into an "EVFILT_WRITE" kevent with
581 "NOTE_EOF".
582 "EVBACKEND_DEVPOLL" (value 16, Solaris 8)
584 This is not implemented yet (and might never be, unless you
585 send me an implementation). According to reports, "/dev/poll"
586 only supports sockets and is not embeddable, which would limit
587 the usefulness of this backend immensely.
588 "EVBACKEND_PORT" (value 32, Solaris 10)
590 This uses the Solaris 10 event port mechanism. As with
591 everything on Solaris, it's really slow, but it still scales
592 very well (O(active_fds)).
593 While this backend scales well, it requires one system call per
595 active file descriptor per loop iteration. For small and medium
596 numbers of file descriptors a "slow" "EVBACKEND_SELECT" or
597 "EVBACKEND_POLL" backend might perform better.
598 On the positive side, this backend actually performed fully to
600 specification in all tests and is fully embeddable, which is a
601 rare feat among the OS-specific backends (I vastly prefer
602 correctness over speed hacks).
603 On the negative side, the interface is bizarre - so bizarre
605 that even sun itself gets it wrong in their code examples: The
606 event polling function sometimes returns events to the caller
607 even though an error occurred, but with no indication whether
608 it has done so or not (yes, it's even documented that way) -
609 deadly for edge-triggered interfaces where you absolutely have
610 to know whether an event occurred or not because you have to
611 re-arm the watcher.
612 Fortunately libev seems to be able to work around these
614 idiocies.
615 This backend maps "EV_READ" and "EV_WRITE" in the same way as
617 "EVBACKEND_POLL".
618 "EVBACKEND_ALL"
620 Try all backends (even potentially broken ones that wouldn't be
621 tried with "EVFLAG_AUTO"). Since this is a mask, you can do
622 stuff such as "EVBACKEND_ALL & ~EVBACKEND_KQUEUE".
623 It is definitely not recommended to use this flag, use whatever
625 "ev_recommended_backends ()" returns, or simply do not specify
626 a backend at all.
627 "EVBACKEND_MASK"
629 Not a backend at all, but a mask to select all backend bits
630 from a "flags" value, in case you want to mask out any backends
631 from a flags value (e.g. when modifying the "LIBEV_FLAGS"
632 environment variable).
633 If one or more of the backend flags are or'ed into the flags value,
635 then only these backends will be tried (in the reverse order as
636 listed here). If none are specified, all backends in
637 "ev_recommended_backends ()" will be tried.
638 Example: Try to create a event loop that uses epoll and nothing
640 else.
641 struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
643 if (!epoller)
644 fatal ("no epoll found here, maybe it hides under your chair");
645 Example: Use whatever libev has to offer, but make sure that kqueue
647 is used if available.
648 struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_KQUEUE);
650 ev_loop_destroy (loop)
652 Destroys an event loop object (frees all memory and kernel state
653 etc.). None of the active event watchers will be stopped in the
654 normal sense, so e.g. "ev_is_active" might still return true. It is
655 your responsibility to either stop all watchers cleanly yourself
656 before calling this function, or cope with the fact afterwards
657 (which is usually the easiest thing, you can just ignore the
658 watchers and/or "free ()" them for example).
659 Note that certain global state, such as signal state (and installed
661 signal handlers), will not be freed by this function, and related
662 watchers (such as signal and child watchers) would need to be
663 stopped manually.
664 This function is normally used on loop objects allocated by
666 "ev_loop_new", but it can also be used on the default loop returned
667 by "ev_default_loop", in which case it is not thread-safe.
668 Note that it is not advisable to call this function on the default
670 loop except in the rare occasion where you really need to free its
671 resources. If you need dynamically allocated loops it is better to
672 use "ev_loop_new" and "ev_loop_destroy".
673 ev_loop_fork (loop)
675 This function sets a flag that causes subsequent "ev_run"
676 iterations to reinitialise the kernel state for backends that have
677 one. Despite the name, you can call it anytime you are allowed to
678 start or stop watchers (except inside an "ev_prepare" callback),
679 but it makes most sense after forking, in the child process. You
680 must call it (or use "EVFLAG_FORKCHECK") in the child before
681 resuming or calling "ev_run".
682 In addition, if you want to reuse a loop (via this function or
684 "EVFLAG_FORKCHECK"), you also have to ignore "SIGPIPE".
685 Again, you have to call it on any loop that you want to re-use
687 after a fork, even if you do not plan to use the loop in the
688 parent. This is because some kernel interfaces *cough* kqueue
689 *cough* do funny things during fork.
690 On the other hand, you only need to call this function in the child
692 process if and only if you want to use the event loop in the child.
693 If you just fork+exec or create a new loop in the child, you don't
694 have to call it at all (in fact, "epoll" is so badly broken that it
695 makes a difference, but libev will usually detect this case on its
696 own and do a costly reset of the backend).
697 The function itself is quite fast and it's usually not a problem to
699 call it just in case after a fork.
700 Example: Automate calling "ev_loop_fork" on the default loop when
702 using pthreads.
703 static void
705 post_fork_child (void)
706 {
707 ev_loop_fork (EV_DEFAULT);
708 }
709 ...
711 pthread_atfork (0, 0, post_fork_child);
712 int ev_is_default_loop (loop)
714 Returns true when the given loop is, in fact, the default loop, and
715 false otherwise.
716 unsigned int ev_iteration (loop)
718 Returns the current iteration count for the event loop, which is
719 identical to the number of times libev did poll for new events. It
720 starts at 0 and happily wraps around with enough iterations.
721 This value can sometimes be useful as a generation counter of sorts
723 (it "ticks" the number of loop iterations), as it roughly
724 corresponds with "ev_prepare" and "ev_check" calls - and is
725 incremented between the prepare and check phases.
726 unsigned int ev_depth (loop)
728 Returns the number of times "ev_run" was entered minus the number
729 of times "ev_run" was exited normally, in other words, the
730 recursion depth.
731 Outside "ev_run", this number is zero. In a callback, this number
733 is 1, unless "ev_run" was invoked recursively (or from another
734 thread), in which case it is higher.
735 Leaving "ev_run" abnormally (setjmp/longjmp, cancelling the thread,
737 throwing an exception etc.), doesn't count as "exit" - consider
738 this as a hint to avoid such ungentleman-like behaviour unless it's
739 really convenient, in which case it is fully supported.
740 unsigned int ev_backend (loop)
742 Returns one of the "EVBACKEND_*" flags indicating the event backend
743 in use.
744 ev_tstamp ev_now (loop)
746 Returns the current "event loop time", which is the time the event
747 loop received events and started processing them. This timestamp
748 does not change as long as callbacks are being processed, and this
749 is also the base time used for relative timers. You can treat it as
750 the timestamp of the event occurring (or more correctly, libev
751 finding out about it).
752 ev_now_update (loop)
754 Establishes the current time by querying the kernel, updating the
755 time returned by "ev_now ()" in the progress. This is a costly
756 operation and is usually done automatically within "ev_run ()".
757 This function is rarely useful, but when some event callback runs
759 for a very long time without entering the event loop, updating
760 libev's idea of the current time is a good idea.
761 See also "The special problem of time updates" in the "ev_timer"
763 section.
764 ev_suspend (loop)
766 ev_resume (loop)
767 These two functions suspend and resume an event loop, for use when
768 the loop is not used for a while and timeouts should not be
769 processed.
770 A typical use case would be an interactive program such as a game:
772 When the user presses "^Z" to suspend the game and resumes it an
773 hour later it would be best to handle timeouts as if no time had
774 actually passed while the program was suspended. This can be
775 achieved by calling "ev_suspend" in your "SIGTSTP" handler, sending
776 yourself a "SIGSTOP" and calling "ev_resume" directly afterwards to
777 resume timer processing.
778 Effectively, all "ev_timer" watchers will be delayed by the time
780 spend between "ev_suspend" and "ev_resume", and all "ev_periodic"
781 watchers will be rescheduled (that is, they will lose any events
782 that would have occurred while suspended).
783 After calling "ev_suspend" you must not call any function on the
785 given loop other than "ev_resume", and you must not call
786 "ev_resume" without a previous call to "ev_suspend".
787 Calling "ev_suspend"/"ev_resume" has the side effect of updating
789 the event loop time (see "ev_now_update").
790 bool ev_run (loop, int flags)
792 Finally, this is it, the event handler. This function usually is
793 called after you have initialised all your watchers and you want to
794 start handling events. It will ask the operating system for any new
795 events, call the watcher callbacks, and then repeat the whole
796 process indefinitely: This is why event loops are called loops.
797 If the flags argument is specified as 0, it will keep handling
799 events until either no event watchers are active anymore or
800 "ev_break" was called.
801 The return value is false if there are no more active watchers
803 (which usually means "all jobs done" or "deadlock"), and true in
804 all other cases (which usually means " you should call "ev_run"
805 again").
806 Please note that an explicit "ev_break" is usually better than
808 relying on all watchers to be stopped when deciding when a program
809 has finished (especially in interactive programs), but having a
810 program that automatically loops as long as it has to and no longer
811 by virtue of relying on its watchers stopping correctly, that is
812 truly a thing of beauty.
813 This function is mostly exception-safe - you can break out of a
815 "ev_run" call by calling "longjmp" in a callback, throwing a C++
816 exception and so on. This does not decrement the "ev_depth" value,
817 nor will it clear any outstanding "EVBREAK_ONE" breaks.
818 A flags value of "EVRUN_NOWAIT" will look for new events, will
820 handle those events and any already outstanding ones, but will not
821 wait and block your process in case there are no events and will
822 return after one iteration of the loop. This is sometimes useful to
823 poll and handle new events while doing lengthy calculations, to
824 keep the program responsive.
825 A flags value of "EVRUN_ONCE" will look for new events (waiting if
827 necessary) and will handle those and any already outstanding ones.
828 It will block your process until at least one new event arrives
829 (which could be an event internal to libev itself, so there is no
830 guarantee that a user-registered callback will be called), and will
831 return after one iteration of the loop.
832 This is useful if you are waiting for some external event in
834 conjunction with something not expressible using other libev
835 watchers (i.e. "roll your own "ev_run""). However, a pair of
836 "ev_prepare"/"ev_check" watchers is usually a better approach for
837 this kind of thing.
838 Here are the gory details of what "ev_run" does (this is for your
840 understanding, not a guarantee that things will work exactly like
841 this in future versions):
842 - Increment loop depth.
844 - Reset the ev_break status.
845 - Before the first iteration, call any pending watchers.
846 LOOP:
847 - If EVFLAG_FORKCHECK was used, check for a fork.
848 - If a fork was detected (by any means), queue and call all fork watchers.
849 - Queue and call all prepare watchers.
850 - If ev_break was called, goto FINISH.
851 - If we have been forked, detach and recreate the kernel state
852 as to not disturb the other process.
853 - Update the kernel state with all outstanding changes.
854 - Update the "event loop time" (ev_now ()).
855 - Calculate for how long to sleep or block, if at all
856 (active idle watchers, EVRUN_NOWAIT or not having
857 any active watchers at all will result in not sleeping).
858 - Sleep if the I/O and timer collect interval say so.
859 - Increment loop iteration counter.
860 - Block the process, waiting for any events.
861 - Queue all outstanding I/O (fd) events.
862 - Update the "event loop time" (ev_now ()), and do time jump adjustments.
863 - Queue all expired timers.
864 - Queue all expired periodics.
865 - Queue all idle watchers with priority higher than that of pending events.
866 - Queue all check watchers.
867 - Call all queued watchers in reverse order (i.e. check watchers first).
868 Signals and child watchers are implemented as I/O watchers, and will
869 be handled here by queueing them when their watcher gets executed.
870 - If ev_break has been called, or EVRUN_ONCE or EVRUN_NOWAIT
871 were used, or there are no active watchers, goto FINISH, otherwise
872 continue with step LOOP.
873 FINISH:
874 - Reset the ev_break status iff it was EVBREAK_ONE.
875 - Decrement the loop depth.
876 - Return.
877 Example: Queue some jobs and then loop until no events are
879 outstanding anymore.
880 ... queue jobs here, make sure they register event watchers as long
882 ... as they still have work to do (even an idle watcher will do..)
883 ev_run (my_loop, 0);
884 ... jobs done or somebody called break. yeah!
885 ev_break (loop, how)
887 Can be used to make a call to "ev_run" return early (but only after
888 it has processed all outstanding events). The "how" argument must
889 be either "EVBREAK_ONE", which will make the innermost "ev_run"
890 call return, or "EVBREAK_ALL", which will make all nested "ev_run"
891 calls return.
892 This "break state" will be cleared on the next call to "ev_run".
894 It is safe to call "ev_break" from outside any "ev_run" calls, too,
896 in which case it will have no effect.
897 ev_ref (loop)
899 ev_unref (loop)
900 Ref/unref can be used to add or remove a reference count on the
901 event loop: Every watcher keeps one reference, and as long as the
902 reference count is nonzero, "ev_run" will not return on its own.
903 This is useful when you have a watcher that you never intend to
905 unregister, but that nevertheless should not keep "ev_run" from
906 returning. In such a case, call "ev_unref" after starting, and
907 "ev_ref" before stopping it.
908 As an example, libev itself uses this for its internal signal pipe:
910 It is not visible to the libev user and should not keep "ev_run"
911 from exiting if no event watchers registered by it are active. It
912 is also an excellent way to do this for generic recurring timers or
913 from within third-party libraries. Just remember to unref after
914 start and ref before stop (but only if the watcher wasn't active
915 before, or was active before, respectively. Note also that libev
916 might stop watchers itself (e.g. non-repeating timers) in which
917 case you have to "ev_ref" in the callback).
918 Example: Create a signal watcher, but keep it from keeping "ev_run"
920 running when nothing else is active.
921 ev_signal exitsig;
923 ev_signal_init (&exitsig, sig_cb, SIGINT);
924 ev_signal_start (loop, &exitsig);
925 ev_unref (loop);
926 Example: For some weird reason, unregister the above signal handler
928 again.
929 ev_ref (loop);
931 ev_signal_stop (loop, &exitsig);
932 ev_set_io_collect_interval (loop, ev_tstamp interval)
934 ev_set_timeout_collect_interval (loop, ev_tstamp interval)
935 These advanced functions influence the time that libev will spend
936 waiting for events. Both time intervals are by default 0, meaning
937 that libev will try to invoke timer/periodic callbacks and I/O
938 callbacks with minimum latency.
939 Setting these to a higher value (the "interval" must be >= 0)
941 allows libev to delay invocation of I/O and timer/periodic
942 callbacks to increase efficiency of loop iterations (or to increase
943 power-saving opportunities).
944 The idea is that sometimes your program runs just fast enough to
946 handle one (or very few) event(s) per loop iteration. While this
947 makes the program responsive, it also wastes a lot of CPU time to
948 poll for new events, especially with backends like "select ()"
949 which have a high overhead for the actual polling but can deliver
950 many events at once.
951 By setting a higher io collect interval you allow libev to spend
953 more time collecting I/O events, so you can handle more events per
954 iteration, at the cost of increasing latency. Timeouts (both
955 "ev_periodic" and "ev_timer") will not be affected. Setting this to
956 a non-null value will introduce an additional "ev_sleep ()" call
957 into most loop iterations. The sleep time ensures that libev will
958 not poll for I/O events more often then once per this interval, on
959 average (as long as the host time resolution is good enough).
960 Likewise, by setting a higher timeout collect interval you allow
962 libev to spend more time collecting timeouts, at the expense of
963 increased latency/jitter/inexactness (the watcher callback will be
964 called later). "ev_io" watchers will not be affected. Setting this
965 to a non-null value will not introduce any overhead in libev.
966 Many (busy) programs can usually benefit by setting the I/O collect
968 interval to a value near 0.1 or so, which is often enough for
969 interactive servers (of course not for games), likewise for
970 timeouts. It usually doesn't make much sense to set it to a lower
971 value than 0.01, as this approaches the timing granularity of most
972 systems. Note that if you do transactions with the outside world
973 and you can't increase the parallelity, then this setting will
974 limit your transaction rate (if you need to poll once per
975 transaction and the I/O collect interval is 0.01, then you can't do
976 more than 100 transactions per second).
977 Setting the timeout collect interval can improve the opportunity
979 for saving power, as the program will "bundle" timer callback
980 invocations that are "near" in time together, by delaying some,
981 thus reducing the number of times the process sleeps and wakes up
982 again. Another useful technique to reduce iterations/wake-ups is to
983 use "ev_periodic" watchers and make sure they fire on, say, one-
984 second boundaries only.
985 Example: we only need 0.1s timeout granularity, and we wish not to
987 poll more often than 100 times per second:
988 ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
990 ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
991 ev_invoke_pending (loop)
993 This call will simply invoke all pending watchers while resetting
994 their pending state. Normally, "ev_run" does this automatically
995 when required, but when overriding the invoke callback this call
996 comes handy. This function can be invoked from a watcher - this can
997 be useful for example when you want to do some lengthy calculation
998 and want to pass further event handling to another thread (you
999 still have to make sure only one thread executes within
1000 "ev_invoke_pending" or "ev_run" of course).
1001 int ev_pending_count (loop)
1003 Returns the number of pending watchers - zero indicates that no
1004 watchers are pending.
1005 ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
1007 This overrides the invoke pending functionality of the loop:
1008 Instead of invoking all pending watchers when there are any,
1009 "ev_run" will call this callback instead. This is useful, for
1010 example, when you want to invoke the actual watchers inside another
1011 context (another thread etc.).
1012 If you want to reset the callback, use "ev_invoke_pending" as new
1014 callback.
1015 ev_set_loop_release_cb (loop, void (*release)(EV_P) throw (), void
1017 (*acquire)(EV_P) throw ())
1018 Sometimes you want to share the same loop between multiple threads.
1019 This can be done relatively simply by putting mutex_lock/unlock
1020 calls around each call to a libev function.
1021 However, "ev_run" can run an indefinite time, so it is not feasible
1023 to wait for it to return. One way around this is to wake up the
1024 event loop via "ev_break" and "ev_async_send", another way is to
1025 set these release and acquire callbacks on the loop.
1026 When set, then "release" will be called just before the thread is
1028 suspended waiting for new events, and "acquire" is called just
1029 afterwards.
1030 Ideally, "release" will just call your mutex_unlock function, and
1032 "acquire" will just call the mutex_lock function again.
1033 While event loop modifications are allowed between invocations of
1035 "release" and "acquire" (that's their only purpose after all), no
1036 modifications done will affect the event loop, i.e. adding watchers
1037 will have no effect on the set of file descriptors being watched,
1038 or the time waited. Use an "ev_async" watcher to wake up "ev_run"
1039 when you want it to take note of any changes you made.
1040 In theory, threads executing "ev_run" will be async-cancel safe
1042 between invocations of "release" and "acquire".
1043 See also the locking example in the "THREADS" section later in this
1045 document.
1046 ev_set_userdata (loop, void *data)
1048 void *ev_userdata (loop)
1049 Set and retrieve a single "void *" associated with a loop. When
1050 "ev_set_userdata" has never been called, then "ev_userdata" returns
1051 0.
1052 These two functions can be used to associate arbitrary data with a
1054 loop, and are intended solely for the "invoke_pending_cb",
1055 "release" and "acquire" callbacks described above, but of course
1056 can be (ab-)used for any other purpose as well.
1057 ev_verify (loop)
1059 This function only does something when "EV_VERIFY" support has been
1060 compiled in, which is the default for non-minimal builds. It tries
1061 to go through all internal structures and checks them for validity.
1062 If anything is found to be inconsistent, it will print an error
1063 message to standard error and call "abort ()".
1064 This can be used to catch bugs inside libev itself: under normal
1066 circumstances, this function will never abort as of course libev
1067 keeps its data structures consistent.
1068
1070 In the following description, uppercase "TYPE" in names stands for the
1071 watcher type, e.g. "ev_TYPE_start" can mean "ev_timer_start" for timer
1072 watchers and "ev_io_start" for I/O watchers.
1073 A watcher is an opaque structure that you allocate and register to
1075 record your interest in some event. To make a concrete example, imagine
1076 you want to wait for STDIN to become readable, you would create an
1077 "ev_io" watcher for that:
1078 static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
1080 {
1081 ev_io_stop (w);
1082 ev_break (loop, EVBREAK_ALL);
1083 }
1084 struct ev_loop *loop = ev_default_loop (0);
1086 ev_io stdin_watcher;
1088 ev_init (&stdin_watcher, my_cb);
1090 ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
1091 ev_io_start (loop, &stdin_watcher);
1092 ev_run (loop, 0);
1094 As you can see, you are responsible for allocating the memory for your
1096 watcher structures (and it is usually a bad idea to do this on the
1097 stack).
1098 Each watcher has an associated watcher structure (called "struct
1100 ev_TYPE" or simply "ev_TYPE", as typedefs are provided for all watcher
1101 structs).
1102 Each watcher structure must be initialised by a call to "ev_init
1104 (watcher *, callback)", which expects a callback to be provided. This
1105 callback is invoked each time the event occurs (or, in the case of I/O
1106 watchers, each time the event loop detects that the file descriptor
1107 given is readable and/or writable).
1108 Each watcher type further has its own "ev_TYPE_set (watcher *, ...)"
1110 macro to configure it, with arguments specific to the watcher type.
1111 There is also a macro to combine initialisation and setting in one
1112 call: "ev_TYPE_init (watcher *, callback, ...)".
1113 To make the watcher actually watch out for events, you have to start it
1115 with a watcher-specific start function ("ev_TYPE_start (loop, watcher
1116 *)"), and you can stop watching for events at any time by calling the
1117 corresponding stop function ("ev_TYPE_stop (loop, watcher *)".
1118 As long as your watcher is active (has been started but not stopped)
1120 you must not touch the values stored in it. Most specifically you must
1121 never reinitialise it or call its "ev_TYPE_set" macro.
1122 Each and every callback receives the event loop pointer as first, the
1124 registered watcher structure as second, and a bitset of received events
1125 as third argument.
1126 The received events usually include a single bit per event type
1128 received (you can receive multiple events at the same time). The
1129 possible bit masks are:
1130 "EV_READ"
1132 "EV_WRITE"
1133 The file descriptor in the "ev_io" watcher has become readable
1134 and/or writable.
1135 "EV_TIMER"
1137 The "ev_timer" watcher has timed out.
1138 "EV_PERIODIC"
1140 The "ev_periodic" watcher has timed out.
1141 "EV_SIGNAL"
1143 The signal specified in the "ev_signal" watcher has been received
1144 by a thread.
1145 "EV_CHILD"
1147 The pid specified in the "ev_child" watcher has received a status
1148 change.
1149 "EV_STAT"
1151 The path specified in the "ev_stat" watcher changed its attributes
1152 somehow.
1153 "EV_IDLE"
1155 The "ev_idle" watcher has determined that you have nothing better
1156 to do.
1157 "EV_PREPARE"
1159 "EV_CHECK"
1160 All "ev_prepare" watchers are invoked just before "ev_run" starts
1161 to gather new events, and all "ev_check" watchers are queued (not
1162 invoked) just after "ev_run" has gathered them, but before it
1163 queues any callbacks for any received events. That means
1164 "ev_prepare" watchers are the last watchers invoked before the
1165 event loop sleeps or polls for new events, and "ev_check" watchers
1166 will be invoked before any other watchers of the same or lower
1167 priority within an event loop iteration.
1168 Callbacks of both watcher types can start and stop as many watchers
1170 as they want, and all of them will be taken into account (for
1171 example, a "ev_prepare" watcher might start an idle watcher to keep
1172 "ev_run" from blocking).
1173 "EV_EMBED"
1175 The embedded event loop specified in the "ev_embed" watcher needs
1176 attention.
1177 "EV_FORK"
1179 The event loop has been resumed in the child process after fork
1180 (see "ev_fork").
1181 "EV_CLEANUP"
1183 The event loop is about to be destroyed (see "ev_cleanup").
1184 "EV_ASYNC"
1186 The given async watcher has been asynchronously notified (see
1187 "ev_async").
1188 "EV_CUSTOM"
1190 Not ever sent (or otherwise used) by libev itself, but can be
1191 freely used by libev users to signal watchers (e.g. via
1192 "ev_feed_event").
1193 "EV_ERROR"
1195 An unspecified error has occurred, the watcher has been stopped.
1196 This might happen because the watcher could not be properly started
1197 because libev ran out of memory, a file descriptor was found to be
1198 closed or any other problem. Libev considers these application
1199 bugs.
1200 You best act on it by reporting the problem and somehow coping with
1202 the watcher being stopped. Note that well-written programs should
1203 not receive an error ever, so when your watcher receives it, this
1204 usually indicates a bug in your program.
1205 Libev will usually signal a few "dummy" events together with an
1207 error, for example it might indicate that a fd is readable or
1208 writable, and if your callbacks is well-written it can just attempt
1209 the operation and cope with the error from read() or write(). This
1210 will not work in multi-threaded programs, though, as the fd could
1211 already be closed and reused for another thing, so beware.
1212 GENERIC WATCHER FUNCTIONS
1214 "ev_init" (ev_TYPE *watcher, callback)
1215 This macro initialises the generic portion of a watcher. The
1216 contents of the watcher object can be arbitrary (so "malloc" will
1217 do). Only the generic parts of the watcher are initialised, you
1218 need to call the type-specific "ev_TYPE_set" macro afterwards to
1219 initialise the type-specific parts. For each type there is also a
1220 "ev_TYPE_init" macro which rolls both calls into one.
1221 You can reinitialise a watcher at any time as long as it has been
1223 stopped (or never started) and there are no pending events
1224 outstanding.
1225 The callback is always of type "void (*)(struct ev_loop *loop,
1227 ev_TYPE *watcher, int revents)".
1228 Example: Initialise an "ev_io" watcher in two steps.
1230 ev_io w;
1232 ev_init (&w, my_cb);
1233 ev_io_set (&w, STDIN_FILENO, EV_READ);
1234 "ev_TYPE_set" (ev_TYPE *watcher, [args])
1236 This macro initialises the type-specific parts of a watcher. You
1237 need to call "ev_init" at least once before you call this macro,
1238 but you can call "ev_TYPE_set" any number of times. You must not,
1239 however, call this macro on a watcher that is active (it can be
1240 pending, however, which is a difference to the "ev_init" macro).
1241 Although some watcher types do not have type-specific arguments
1243 (e.g. "ev_prepare") you still need to call its "set" macro.
1244 See "ev_init", above, for an example.
1246 "ev_TYPE_init" (ev_TYPE *watcher, callback, [args])
1248 This convenience macro rolls both "ev_init" and "ev_TYPE_set" macro
1249 calls into a single call. This is the most convenient method to
1250 initialise a watcher. The same limitations apply, of course.
1251 Example: Initialise and set an "ev_io" watcher in one step.
1253 ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
1255 "ev_TYPE_start" (loop, ev_TYPE *watcher)
1257 Starts (activates) the given watcher. Only active watchers will
1258 receive events. If the watcher is already active nothing will
1259 happen.
1260 Example: Start the "ev_io" watcher that is being abused as example
1262 in this whole section.
1263 ev_io_start (EV_DEFAULT_UC, &w);
1265 "ev_TYPE_stop" (loop, ev_TYPE *watcher)
1267 Stops the given watcher if active, and clears the pending status
1268 (whether the watcher was active or not).
1269 It is possible that stopped watchers are pending - for example,
1271 non-repeating timers are being stopped when they become pending -
1272 but calling "ev_TYPE_stop" ensures that the watcher is neither
1273 active nor pending. If you want to free or reuse the memory used by
1274 the watcher it is therefore a good idea to always call its
1275 "ev_TYPE_stop" function.
1276 bool ev_is_active (ev_TYPE *watcher)
1278 Returns a true value iff the watcher is active (i.e. it has been
1279 started and not yet been stopped). As long as a watcher is active
1280 you must not modify it.
1281 bool ev_is_pending (ev_TYPE *watcher)
1283 Returns a true value iff the watcher is pending, (i.e. it has
1284 outstanding events but its callback has not yet been invoked). As
1285 long as a watcher is pending (but not active) you must not call an
1286 init function on it (but "ev_TYPE_set" is safe), you must not
1287 change its priority, and you must make sure the watcher is
1288 available to libev (e.g. you cannot "free ()" it).
1289 callback ev_cb (ev_TYPE *watcher)
1291 Returns the callback currently set on the watcher.
1292 ev_set_cb (ev_TYPE *watcher, callback)
1294 Change the callback. You can change the callback at virtually any
1295 time (modulo threads).
1296 ev_set_priority (ev_TYPE *watcher, int priority)
1298 int ev_priority (ev_TYPE *watcher)
1299 Set and query the priority of the watcher. The priority is a small
1300 integer between "EV_MAXPRI" (default: 2) and "EV_MINPRI" (default:
1301 "-2"). Pending watchers with higher priority will be invoked before
1302 watchers with lower priority, but priority will not keep watchers
1303 from being executed (except for "ev_idle" watchers).
1304 If you need to suppress invocation when higher priority events are
1306 pending you need to look at "ev_idle" watchers, which provide this
1307 functionality.
1308 You must not change the priority of a watcher as long as it is
1310 active or pending.
1311 Setting a priority outside the range of "EV_MINPRI" to "EV_MAXPRI"
1313 is fine, as long as you do not mind that the priority value you
1314 query might or might not have been clamped to the valid range.
1315 The default priority used by watchers when no priority has been set
1317 is always 0, which is supposed to not be too high and not be too
1318 low :).
1319 See "WATCHER PRIORITY MODELS", below, for a more thorough treatment
1321 of priorities.
1322 ev_invoke (loop, ev_TYPE *watcher, int revents)
1324 Invoke the "watcher" with the given "loop" and "revents". Neither
1325 "loop" nor "revents" need to be valid as long as the watcher
1326 callback can deal with that fact, as both are simply passed through
1327 to the callback.
1328 int ev_clear_pending (loop, ev_TYPE *watcher)
1330 If the watcher is pending, this function clears its pending status
1331 and returns its "revents" bitset (as if its callback was invoked).
1332 If the watcher isn't pending it does nothing and returns 0.
1333 Sometimes it can be useful to "poll" a watcher instead of waiting
1335 for its callback to be invoked, which can be accomplished with this
1336 function.
1337 ev_feed_event (loop, ev_TYPE *watcher, int revents)
1339 Feeds the given event set into the event loop, as if the specified
1340 event had happened for the specified watcher (which must be a
1341 pointer to an initialised but not necessarily started event
1342 watcher). Obviously you must not free the watcher as long as it has
1343 pending events.
1344 Stopping the watcher, letting libev invoke it, or calling
1346 "ev_clear_pending" will clear the pending event, even if the
1347 watcher was not started in the first place.
1348 See also "ev_feed_fd_event" and "ev_feed_signal_event" for related
1350 functions that do not need a watcher.
1351 See also the "ASSOCIATING CUSTOM DATA WITH A WATCHER" and "BUILDING
1353 YOUR OWN COMPOSITE WATCHERS" idioms.
1354 WATCHER STATES
1356 There are various watcher states mentioned throughout this manual -
1357 active, pending and so on. In this section these states and the rules
1358 to transition between them will be described in more detail - and while
1359 these rules might look complicated, they usually do "the right thing".
1360 initialised
1362 Before a watcher can be registered with the event loop it has to be
1363 initialised. This can be done with a call to "ev_TYPE_init", or
1364 calls to "ev_init" followed by the watcher-specific "ev_TYPE_set"
1365 function.
1366 In this state it is simply some block of memory that is suitable
1368 for use in an event loop. It can be moved around, freed, reused
1369 etc. at will - as long as you either keep the memory contents
1370 intact, or call "ev_TYPE_init" again.
1371 started/running/active
1373 Once a watcher has been started with a call to "ev_TYPE_start" it
1374 becomes property of the event loop, and is actively waiting for
1375 events. While in this state it cannot be accessed (except in a few
1376 documented ways), moved, freed or anything else - the only legal
1377 thing is to keep a pointer to it, and call libev functions on it
1378 that are documented to work on active watchers.
1379 pending
1381 If a watcher is active and libev determines that an event it is
1382 interested in has occurred (such as a timer expiring), it will
1383 become pending. It will stay in this pending state until either it
1384 is stopped or its callback is about to be invoked, so it is not
1385 normally pending inside the watcher callback.
1386 The watcher might or might not be active while it is pending (for
1388 example, an expired non-repeating timer can be pending but no
1389 longer active). If it is stopped, it can be freely accessed (e.g.
1390 by calling "ev_TYPE_set"), but it is still property of the event
1391 loop at this time, so cannot be moved, freed or reused. And if it
1392 is active the rules described in the previous item still apply.
1393 It is also possible to feed an event on a watcher that is not
1395 active (e.g. via "ev_feed_event"), in which case it becomes
1396 pending without being active.
1397 stopped
1399 A watcher can be stopped implicitly by libev (in which case it
1400 might still be pending), or explicitly by calling its
1401 "ev_TYPE_stop" function. The latter will clear any pending state
1402 the watcher might be in, regardless of whether it was active or
1403 not, so stopping a watcher explicitly before freeing it is often a
1404 good idea.
1405 While stopped (and not pending) the watcher is essentially in the
1407 initialised state, that is, it can be reused, moved, modified in
1408 any way you wish (but when you trash the memory block, you need to
1409 "ev_TYPE_init" it again).
1410 WATCHER PRIORITY MODELS
1412 Many event loops support watcher priorities, which are usually small
1413 integers that influence the ordering of event callback invocation
1414 between watchers in some way, all else being equal.
1415 In libev, Watcher priorities can be set using "ev_set_priority". See
1417 its description for the more technical details such as the actual
1418 priority range.
1419 There are two common ways how these these priorities are being
1421 interpreted by event loops:
1422 In the more common lock-out model, higher priorities "lock out"
1424 invocation of lower priority watchers, which means as long as higher
1425 priority watchers receive events, lower priority watchers are not being
1426 invoked.
1427 The less common only-for-ordering model uses priorities solely to order
1429 callback invocation within a single event loop iteration: Higher
1430 priority watchers are invoked before lower priority ones, but they all
1431 get invoked before polling for new events.
1432 Libev uses the second (only-for-ordering) model for all its watchers
1434 except for idle watchers (which use the lock-out model).
1435 The rationale behind this is that implementing the lock-out model for
1437 watchers is not well supported by most kernel interfaces, and most
1438 event libraries will just poll for the same events again and again as
1439 long as their callbacks have not been executed, which is very
1440 inefficient in the common case of one high-priority watcher locking out
1441 a mass of lower priority ones.
1442 Static (ordering) priorities are most useful when you have two or more
1444 watchers handling the same resource: a typical usage example is having
1445 an "ev_io" watcher to receive data, and an associated "ev_timer" to
1446 handle timeouts. Under load, data might be received while the program
1447 handles other jobs, but since timers normally get invoked first, the
1448 timeout handler will be executed before checking for data. In that
1449 case, giving the timer a lower priority than the I/O watcher ensures
1450 that I/O will be handled first even under adverse conditions (which is
1451 usually, but not always, what you want).
1452 Since idle watchers use the "lock-out" model, meaning that idle
1454 watchers will only be executed when no same or higher priority watchers
1455 have received events, they can be used to implement the "lock-out"
1456 model when required.
1457 For example, to emulate how many other event libraries handle
1459 priorities, you can associate an "ev_idle" watcher to each such
1460 watcher, and in the normal watcher callback, you just start the idle
1461 watcher. The real processing is done in the idle watcher callback. This
1462 causes libev to continuously poll and process kernel event data for the
1463 watcher, but when the lock-out case is known to be rare (which in turn
1464 is rare :), this is workable.
1465 Usually, however, the lock-out model implemented that way will perform
1467 miserably under the type of load it was designed to handle. In that
1468 case, it might be preferable to stop the real watcher before starting
1469 the idle watcher, so the kernel will not have to process the event in
1470 case the actual processing will be delayed for considerable time.
1471 Here is an example of an I/O watcher that should run at a strictly
1473 lower priority than the default, and which should only process data
1474 when no other events are pending:
1475 ev_idle idle; // actual processing watcher
1477 ev_io io; // actual event watcher
1478 static void
1480 io_cb (EV_P_ ev_io *w, int revents)
1481 {
1482 // stop the I/O watcher, we received the event, but
1483 // are not yet ready to handle it.
1484 ev_io_stop (EV_A_ w);
1485 // start the idle watcher to handle the actual event.
1487 // it will not be executed as long as other watchers
1488 // with the default priority are receiving events.
1489 ev_idle_start (EV_A_ &idle);
1490 }
1491 static void
1493 idle_cb (EV_P_ ev_idle *w, int revents)
1494 {
1495 // actual processing
1496 read (STDIN_FILENO, ...);
1497 // have to start the I/O watcher again, as
1499 // we have handled the event
1500 ev_io_start (EV_P_ &io);
1501 }
1502 // initialisation
1504 ev_idle_init (&idle, idle_cb);
1505 ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
1506 ev_io_start (EV_DEFAULT_ &io);
1507 In the "real" world, it might also be beneficial to start a timer, so
1509 that low-priority connections can not be locked out forever under load.
1510 This enables your program to keep a lower latency for important
1511 connections during short periods of high load, while not completely
1512 locking out less important ones.
1513
1515 This section describes each watcher in detail, but will not repeat
1516 information given in the last section. Any initialisation/set macros,
1517 functions and members specific to the watcher type are explained.
1518 Members are additionally marked with either [read-only], meaning that,
1520 while the watcher is active, you can look at the member and expect some
1521 sensible content, but you must not modify it (you can modify it while
1522 the watcher is stopped to your hearts content), or [read-write], which
1523 means you can expect it to have some sensible content while the watcher
1524 is active, but you can also modify it. Modifying it may not do
1525 something sensible or take immediate effect (or do anything at all),
1526 but libev will not crash or malfunction in any way.
1527 "ev_io" - is this file descriptor readable or writable?
1529 I/O watchers check whether a file descriptor is readable or writable in
1530 each iteration of the event loop, or, more precisely, when reading
1531 would not block the process and writing would at least be able to write
1532 some data. This behaviour is called level-triggering because you keep
1533 receiving events as long as the condition persists. Remember you can
1534 stop the watcher if you don't want to act on the event and neither want
1535 to receive future events.
1536 In general you can register as many read and/or write event watchers
1538 per fd as you want (as long as you don't confuse yourself). Setting all
1539 file descriptors to non-blocking mode is also usually a good idea (but
1540 not required if you know what you are doing).
1541 Another thing you have to watch out for is that it is quite easy to
1543 receive "spurious" readiness notifications, that is, your callback
1544 might be called with "EV_READ" but a subsequent "read"(2) will actually
1545 block because there is no data. It is very easy to get into this
1546 situation even with a relatively standard program structure. Thus it is
1547 best to always use non-blocking I/O: An extra "read"(2) returning
1548 "EAGAIN" is far preferable to a program hanging until some data
1549 arrives.
1550 If you cannot run the fd in non-blocking mode (for example you should
1552 not play around with an Xlib connection), then you have to separately
1553 re-test whether a file descriptor is really ready with a known-to-be
1554 good interface such as poll (fortunately in the case of Xlib, it
1555 already does this on its own, so its quite safe to use). Some people
1556 additionally use "SIGALRM" and an interval timer, just to be sure you
1557 won't block indefinitely.
1558 But really, best use non-blocking mode.
1560 The special problem of disappearing file descriptors
1562 Some backends (e.g. kqueue, epoll) need to be told about closing a file
1564 descriptor (either due to calling "close" explicitly or any other
1565 means, such as "dup2"). The reason is that you register interest in
1566 some file descriptor, but when it goes away, the operating system will
1567 silently drop this interest. If another file descriptor with the same
1568 number then is registered with libev, there is no efficient way to see
1569 that this is, in fact, a different file descriptor.
1570 To avoid having to explicitly tell libev about such cases, libev
1572 follows the following policy: Each time "ev_io_set" is being called,
1573 libev will assume that this is potentially a new file descriptor,
1574 otherwise it is assumed that the file descriptor stays the same. That
1575 means that you have to call "ev_io_set" (or "ev_io_init") when you
1576 change the descriptor even if the file descriptor number itself did not
1577 change.
1578 This is how one would do it normally anyway, the important point is
1580 that the libev application should not optimise around libev but should
1581 leave optimisations to libev.
1582 The special problem of dup'ed file descriptors
1584 Some backends (e.g. epoll), cannot register events for file
1586 descriptors, but only events for the underlying file descriptions. That
1587 means when you have "dup ()"'ed file descriptors or weirder
1588 constellations, and register events for them, only one file descriptor
1589 might actually receive events.
1590 There is no workaround possible except not registering events for
1592 potentially "dup ()"'ed file descriptors, or to resort to
1593 "EVBACKEND_SELECT" or "EVBACKEND_POLL".
1594 The special problem of files
1596 Many people try to use "select" (or libev) on file descriptors
1598 representing files, and expect it to become ready when their program
1599 doesn't block on disk accesses (which can take a long time on their
1600 own).
1601 However, this cannot ever work in the "expected" way - you get a
1603 readiness notification as soon as the kernel knows whether and how much
1604 data is there, and in the case of open files, that's always the case,
1605 so you always get a readiness notification instantly, and your read (or
1606 possibly write) will still block on the disk I/O.
1607 Another way to view it is that in the case of sockets, pipes, character
1609 devices and so on, there is another party (the sender) that delivers
1610 data on its own, but in the case of files, there is no such thing: the
1611 disk will not send data on its own, simply because it doesn't know what
1612 you wish to read - you would first have to request some data.
1613 Since files are typically not-so-well supported by advanced
1615 notification mechanism, libev tries hard to emulate POSIX behaviour
1616 with respect to files, even though you should not use it. The reason
1617 for this is convenience: sometimes you want to watch STDIN or STDOUT,
1618 which is usually a tty, often a pipe, but also sometimes files or
1619 special devices (for example, "epoll" on Linux works with /dev/random
1620 but not with /dev/urandom), and even though the file might better be
1621 served with asynchronous I/O instead of with non-blocking I/O, it is
1622 still useful when it "just works" instead of freezing.
1623 So avoid file descriptors pointing to files when you know it (e.g. use
1625 libeio), but use them when it is convenient, e.g. for STDIN/STDOUT, or
1626 when you rarely read from a file instead of from a socket, and want to
1627 reuse the same code path.
1628 The special problem of fork
1630 Some backends (epoll, kqueue) do not support "fork ()" at all or
1632 exhibit useless behaviour. Libev fully supports fork, but needs to be
1633 told about it in the child if you want to continue to use it in the
1634 child.
1635 To support fork in your child processes, you have to call "ev_loop_fork
1637 ()" after a fork in the child, enable "EVFLAG_FORKCHECK", or resort to
1638 "EVBACKEND_SELECT" or "EVBACKEND_POLL".
1639 The special problem of SIGPIPE
1641 While not really specific to libev, it is easy to forget about
1643 "SIGPIPE": when writing to a pipe whose other end has been closed, your
1644 program gets sent a SIGPIPE, which, by default, aborts your program.
1645 For most programs this is sensible behaviour, for daemons, this is
1646 usually undesirable.
1647 So when you encounter spurious, unexplained daemon exits, make sure you
1649 ignore SIGPIPE (and maybe make sure you log the exit status of your
1650 daemon somewhere, as that would have given you a big clue).
1651 The special problem of accept()ing when you can't
1653 Many implementations of the POSIX "accept" function (for example, found
1655 in post-2004 Linux) have the peculiar behaviour of not removing a
1656 connection from the pending queue in all error cases.
1657 For example, larger servers often run out of file descriptors (because
1659 of resource limits), causing "accept" to fail with "ENFILE" but not
1660 rejecting the connection, leading to libev signalling readiness on the
1661 next iteration again (the connection still exists after all), and
1662 typically causing the program to loop at 100% CPU usage.
1663 Unfortunately, the set of errors that cause this issue differs between
1665 operating systems, there is usually little the app can do to remedy the
1666 situation, and no known thread-safe method of removing the connection
1667 to cope with overload is known (to me).
1668 One of the easiest ways to handle this situation is to just ignore it -
1670 when the program encounters an overload, it will just loop until the
1671 situation is over. While this is a form of busy waiting, no OS offers
1672 an event-based way to handle this situation, so it's the best one can
1673 do.
1674 A better way to handle the situation is to log any errors other than
1676 "EAGAIN" and "EWOULDBLOCK", making sure not to flood the log with such
1677 messages, and continue as usual, which at least gives the user an idea
1678 of what could be wrong ("raise the ulimit!"). For extra points one
1679 could stop the "ev_io" watcher on the listening fd "for a while", which
1680 reduces CPU usage.
1681 If your program is single-threaded, then you could also keep a dummy
1683 file descriptor for overload situations (e.g. by opening /dev/null),
1684 and when you run into "ENFILE" or "EMFILE", close it, run "accept",
1685 close that fd, and create a new dummy fd. This will gracefully refuse
1686 clients under typical overload conditions.
1687 The last way to handle it is to simply log the error and "exit", as is
1689 often done with "malloc" failures, but this results in an easy
1690 opportunity for a DoS attack.
1691 Watcher-Specific Functions
1693 ev_io_init (ev_io *, callback, int fd, int events)
1695 ev_io_set (ev_io *, int fd, int events)
1696 Configures an "ev_io" watcher. The "fd" is the file descriptor to
1697 receive events for and "events" is either "EV_READ", "EV_WRITE" or
1698 "EV_READ | EV_WRITE", to express the desire to receive the given
1699 events.
1700 int fd [read-only]
1702 The file descriptor being watched.
1703 int events [read-only]
1705 The events being watched.
1706 Examples
1708 Example: Call "stdin_readable_cb" when STDIN_FILENO has become, well
1710 readable, but only once. Since it is likely line-buffered, you could
1711 attempt to read a whole line in the callback.
1712 static void
1714 stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
1715 {
1716 ev_io_stop (loop, w);
1717 .. read from stdin here (or from w->fd) and handle any I/O errors
1718 }
1719 ...
1721 struct ev_loop *loop = ev_default_init (0);
1722 ev_io stdin_readable;
1723 ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
1724 ev_io_start (loop, &stdin_readable);
1725 ev_run (loop, 0);
1726 "ev_timer" - relative and optionally repeating timeouts
1728 Timer watchers are simple relative timers that generate an event after
1729 a given time, and optionally repeating in regular intervals after that.
1730 The timers are based on real time, that is, if you register an event
1732 that times out after an hour and you reset your system clock to January
1733 last year, it will still time out after (roughly) one hour. "Roughly"
1734 because detecting time jumps is hard, and some inaccuracies are
1735 unavoidable (the monotonic clock option helps a lot here).
1736 The callback is guaranteed to be invoked only after its timeout has
1738 passed (not at, so on systems with very low-resolution clocks this
1739 might introduce a small delay, see "the special problem of being too
1740 early", below). If multiple timers become ready during the same loop
1741 iteration then the ones with earlier time-out values are invoked before
1742 ones of the same priority with later time-out values (but this is no
1743 longer true when a callback calls "ev_run" recursively).
1744 Be smart about timeouts
1746 Many real-world problems involve some kind of timeout, usually for
1748 error recovery. A typical example is an HTTP request - if the other
1749 side hangs, you want to raise some error after a while.
1750 What follows are some ways to handle this problem, from obvious and
1752 inefficient to smart and efficient.
1753 In the following, a 60 second activity timeout is assumed - a timeout
1755 that gets reset to 60 seconds each time there is activity (e.g. each
1756 time some data or other life sign was received).
1757 1. Use a timer and stop, reinitialise and start it on activity.
1759 This is the most obvious, but not the most simple way: In the
1760 beginning, start the watcher:
1761 ev_timer_init (timer, callback, 60., 0.);
1763 ev_timer_start (loop, timer);
1764 Then, each time there is some activity, "ev_timer_stop" it,
1766 initialise it and start it again:
1767 ev_timer_stop (loop, timer);
1769 ev_timer_set (timer, 60., 0.);
1770 ev_timer_start (loop, timer);
1771 This is relatively simple to implement, but means that each time
1773 there is some activity, libev will first have to remove the timer
1774 from its internal data structure and then add it again. Libev tries
1775 to be fast, but it's still not a constant-time operation.
1776 2. Use a timer and re-start it with "ev_timer_again" inactivity.
1778 This is the easiest way, and involves using "ev_timer_again"
1779 instead of "ev_timer_start".
1780 To implement this, configure an "ev_timer" with a "repeat" value of
1782 60 and then call "ev_timer_again" at start and each time you
1783 successfully read or write some data. If you go into an idle state
1784 where you do not expect data to travel on the socket, you can
1785 "ev_timer_stop" the timer, and "ev_timer_again" will automatically
1786 restart it if need be.
1787 That means you can ignore both the "ev_timer_start" function and
1789 the "after" argument to "ev_timer_set", and only ever use the
1790 "repeat" member and "ev_timer_again".
1791 At start:
1793 ev_init (timer, callback);
1795 timer->repeat = 60.;
1796 ev_timer_again (loop, timer);
1797 Each time there is some activity:
1799 ev_timer_again (loop, timer);
1801 It is even possible to change the time-out on the fly, regardless
1803 of whether the watcher is active or not:
1804 timer->repeat = 30.;
1806 ev_timer_again (loop, timer);
1807 This is slightly more efficient then stopping/starting the timer
1809 each time you want to modify its timeout value, as libev does not
1810 have to completely remove and re-insert the timer from/into its
1811 internal data structure.
1812 It is, however, even simpler than the "obvious" way to do it.
1814 3. Let the timer time out, but then re-arm it as required.
1816 This method is more tricky, but usually most efficient: Most
1817 timeouts are relatively long compared to the intervals between
1818 other activity - in our example, within 60 seconds, there are
1819 usually many I/O events with associated activity resets.
1820 In this case, it would be more efficient to leave the "ev_timer"
1822 alone, but remember the time of last activity, and check for a real
1823 timeout only within the callback:
1824 ev_tstamp timeout = 60.;
1826 ev_tstamp last_activity; // time of last activity
1827 ev_timer timer;
1828 static void
1830 callback (EV_P_ ev_timer *w, int revents)
1831 {
1832 // calculate when the timeout would happen
1833 ev_tstamp after = last_activity - ev_now (EV_A) + timeout;
1834 // if negative, it means we the timeout already occurred
1836 if (after < 0.)
1837 {
1838 // timeout occurred, take action
1839 }
1840 else
1841 {
1842 // callback was invoked, but there was some recent
1843 // activity. simply restart the timer to time out
1844 // after "after" seconds, which is the earliest time
1845 // the timeout can occur.
1846 ev_timer_set (w, after, 0.);
1847 ev_timer_start (EV_A_ w);
1848 }
1849 }
1850 To summarise the callback: first calculate in how many seconds the
1852 timeout will occur (by calculating the absolute time when it would
1853 occur, "last_activity + timeout", and subtracting the current time,
1854 "ev_now (EV_A)" from that).
1855 If this value is negative, then we are already past the timeout,
1857 i.e. we timed out, and need to do whatever is needed in this case.
1858 Otherwise, we now the earliest time at which the timeout would
1860 trigger, and simply start the timer with this timeout value.
1861 In other words, each time the callback is invoked it will check
1863 whether the timeout occurred. If not, it will simply reschedule
1864 itself to check again at the earliest time it could time out.
1865 Rinse. Repeat.
1866 This scheme causes more callback invocations (about one every 60
1868 seconds minus half the average time between activity), but
1869 virtually no calls to libev to change the timeout.
1870 To start the machinery, simply initialise the watcher and set
1872 "last_activity" to the current time (meaning there was some
1873 activity just now), then call the callback, which will "do the
1874 right thing" and start the timer:
1875 last_activity = ev_now (EV_A);
1877 ev_init (&timer, callback);
1878 callback (EV_A_ &timer, 0);
1879 When there is some activity, simply store the current time in
1881 "last_activity", no libev calls at all:
1882 if (activity detected)
1884 last_activity = ev_now (EV_A);
1885 When your timeout value changes, then the timeout can be changed by
1887 simply providing a new value, stopping the timer and calling the
1888 callback, which will again do the right thing (for example, time
1889 out immediately :).
1890 timeout = new_value;
1892 ev_timer_stop (EV_A_ &timer);
1893 callback (EV_A_ &timer, 0);
1894 This technique is slightly more complex, but in most cases where
1896 the time-out is unlikely to be triggered, much more efficient.
1897 4. Wee, just use a double-linked list for your timeouts.
1899 If there is not one request, but many thousands (millions...), all
1900 employing some kind of timeout with the same timeout value, then
1901 one can do even better:
1902 When starting the timeout, calculate the timeout value and put the
1904 timeout at the end of the list.
1905 Then use an "ev_timer" to fire when the timeout at the beginning of
1907 the list is expected to fire (for example, using the technique #3).
1908 When there is some activity, remove the timer from the list,
1910 recalculate the timeout, append it to the end of the list again,
1911 and make sure to update the "ev_timer" if it was taken from the
1912 beginning of the list.
1913 This way, one can manage an unlimited number of timeouts in O(1)
1915 time for starting, stopping and updating the timers, at the expense
1916 of a major complication, and having to use a constant timeout. The
1917 constant timeout ensures that the list stays sorted.
1918 So which method the best?
1920 Method #2 is a simple no-brain-required solution that is adequate in
1922 most situations. Method #3 requires a bit more thinking, but handles
1923 many cases better, and isn't very complicated either. In most case,
1924 choosing either one is fine, with #3 being better in typical
1925 situations.
1926 Method #1 is almost always a bad idea, and buys you nothing. Method #4
1928 is rather complicated, but extremely efficient, something that really
1929 pays off after the first million or so of active timers, i.e. it's
1930 usually overkill :)
1931 The special problem of being too early
1933 If you ask a timer to call your callback after three seconds, then you
1935 expect it to be invoked after three seconds - but of course, this
1936 cannot be guaranteed to infinite precision. Less obviously, it cannot
1937 be guaranteed to any precision by libev - imagine somebody suspending
1938 the process with a STOP signal for a few hours for example.
1939 So, libev tries to invoke your callback as soon as possible after the
1941 delay has occurred, but cannot guarantee this.
1942 A less obvious failure mode is calling your callback too early: many
1944 event loops compare timestamps with a "elapsed delay >= requested
1945 delay", but this can cause your callback to be invoked much earlier
1946 than you would expect.
1947 To see why, imagine a system with a clock that only offers full second
1949 resolution (think windows if you can't come up with a broken enough OS
1950 yourself). If you schedule a one-second timer at the time 500.9, then
1951 the event loop will schedule your timeout to elapse at a system time of
1952 500 (500.9 truncated to the resolution) + 1, or 501.
1953 If an event library looks at the timeout 0.1s later, it will see "501
1955 >= 501" and invoke the callback 0.1s after it was started, even though
1956 a one-second delay was requested - this is being "too early", despite
1957 best intentions.
1958 This is the reason why libev will never invoke the callback if the
1960 elapsed delay equals the requested delay, but only when the elapsed
1961 delay is larger than the requested delay. In the example above, libev
1962 would only invoke the callback at system time 502, or 1.1s after the
1963 timer was started.
1964 So, while libev cannot guarantee that your callback will be invoked
1966 exactly when requested, it can and does guarantee that the requested
1967 delay has actually elapsed, or in other words, it always errs on the
1968 "too late" side of things.
1969 The special problem of time updates
1971 Establishing the current time is a costly operation (it usually takes
1973 at least one system call): EV therefore updates its idea of the current
1974 time only before and after "ev_run" collects new events, which causes a
1975 growing difference between "ev_now ()" and "ev_time ()" when handling
1976 lots of events in one iteration.
1977 The relative timeouts are calculated relative to the "ev_now ()" time.
1979 This is usually the right thing as this timestamp refers to the time of
1980 the event triggering whatever timeout you are modifying/starting. If
1981 you suspect event processing to be delayed and you need to base the
1982 timeout on the current time, use something like the following to adjust
1983 for it:
1984 ev_timer_set (&timer, after + (ev_time () - ev_now ()), 0.);
1986 If the event loop is suspended for a long time, you can also force an
1988 update of the time returned by "ev_now ()" by calling "ev_now_update
1989 ()", although that will push the event time of all outstanding events
1990 further into the future.
1991 The special problem of unsynchronised clocks
1993 Modern systems have a variety of clocks - libev itself uses the normal
1995 "wall clock" clock and, if available, the monotonic clock (to avoid
1996 time jumps).
1997 Neither of these clocks is synchronised with each other or any other
1999 clock on the system, so "ev_time ()" might return a considerably
2000 different time than "gettimeofday ()" or "time ()". On a GNU/Linux
2001 system, for example, a call to "gettimeofday" might return a second
2002 count that is one higher than a directly following call to "time".
2003 The moral of this is to only compare libev-related timestamps with
2005 "ev_time ()" and "ev_now ()", at least if you want better precision
2006 than a second or so.
2007 One more problem arises due to this lack of synchronisation: if libev
2009 uses the system monotonic clock and you compare timestamps from
2010 "ev_time" or "ev_now" from when you started your timer and when your
2011 callback is invoked, you will find that sometimes the callback is a bit
2012 "early".
2013 This is because "ev_timer"s work in real time, not wall clock time, so
2015 libev makes sure your callback is not invoked before the delay
2016 happened, measured according to the real time, not the system clock.
2017 If your timeouts are based on a physical timescale (e.g. "time out this
2019 connection after 100 seconds") then this shouldn't bother you as it is
2020 exactly the right behaviour.
2021 If you want to compare wall clock/system timestamps to your timers,
2023 then you need to use "ev_periodic"s, as these are based on the wall
2024 clock time, where your comparisons will always generate correct
2025 results.
2026 The special problems of suspended animation
2028 When you leave the server world it is quite customary to hit machines
2030 that can suspend/hibernate - what happens to the clocks during such a
2031 suspend?
2032 Some quick tests made with a Linux 2.6.28 indicate that a suspend
2034 freezes all processes, while the clocks ("times", "CLOCK_MONOTONIC")
2035 continue to run until the system is suspended, but they will not
2036 advance while the system is suspended. That means, on resume, it will
2037 be as if the program was frozen for a few seconds, but the suspend time
2038 will not be counted towards "ev_timer" when a monotonic clock source is
2039 used. The real time clock advanced as expected, but if it is used as
2040 sole clocksource, then a long suspend would be detected as a time jump
2041 by libev, and timers would be adjusted accordingly.
2042 I would not be surprised to see different behaviour in different
2044 between operating systems, OS versions or even different hardware.
2045 The other form of suspend (job control, or sending a SIGSTOP) will see
2047 a time jump in the monotonic clocks and the realtime clock. If the
2048 program is suspended for a very long time, and monotonic clock sources
2049 are in use, then you can expect "ev_timer"s to expire as the full
2050 suspension time will be counted towards the timers. When no monotonic
2051 clock source is in use, then libev will again assume a timejump and
2052 adjust accordingly.
2053 It might be beneficial for this latter case to call "ev_suspend" and
2055 "ev_resume" in code that handles "SIGTSTP", to at least get
2056 deterministic behaviour in this case (you can do nothing against
2057 "SIGSTOP").
2058 Watcher-Specific Functions and Data Members
2060 ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
2062 ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
2063 Configure the timer to trigger after "after" seconds. If "repeat"
2064 is 0., then it will automatically be stopped once the timeout is
2065 reached. If it is positive, then the timer will automatically be
2066 configured to trigger again "repeat" seconds later, again, and
2067 again, until stopped manually.
2068 The timer itself will do a best-effort at avoiding drift, that is,
2070 if you configure a timer to trigger every 10 seconds, then it will
2071 normally trigger at exactly 10 second intervals. If, however, your
2072 program cannot keep up with the timer (because it takes longer than
2073 those 10 seconds to do stuff) the timer will not fire more than
2074 once per event loop iteration.
2075 ev_timer_again (loop, ev_timer *)
2077 This will act as if the timer timed out, and restarts it again if
2078 it is repeating. It basically works like calling "ev_timer_stop",
2079 updating the timeout to the "repeat" value and calling
2080 "ev_timer_start".
2081 The exact semantics are as in the following rules, all of which
2083 will be applied to the watcher:
2084 If the timer is pending, the pending status is always cleared.
2086 If the timer is started but non-repeating, stop it (as if it timed
2087 out, without invoking it).
2088 If the timer is repeating, make the "repeat" value the new timeout
2089 and start the timer, if necessary.
2090 This sounds a bit complicated, see "Be smart about timeouts",
2092 above, for a usage example.
2093 ev_tstamp ev_timer_remaining (loop, ev_timer *)
2095 Returns the remaining time until a timer fires. If the timer is
2096 active, then this time is relative to the current event loop time,
2097 otherwise it's the timeout value currently configured.
2098 That is, after an "ev_timer_set (w, 5, 7)", "ev_timer_remaining"
2100 returns 5. When the timer is started and one second passes,
2101 "ev_timer_remaining" will return 4. When the timer expires and is
2102 restarted, it will return roughly 7 (likely slightly less as
2103 callback invocation takes some time, too), and so on.
2104 ev_tstamp repeat [read-write]
2106 The current "repeat" value. Will be used each time the watcher
2107 times out or "ev_timer_again" is called, and determines the next
2108 timeout (if any), which is also when any modifications are taken
2109 into account.
2110 Examples
2112 Example: Create a timer that fires after 60 seconds.
2114 static void
2116 one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
2117 {
2118 .. one minute over, w is actually stopped right here
2119 }
2120 ev_timer mytimer;
2122 ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
2123 ev_timer_start (loop, &mytimer);
2124 Example: Create a timeout timer that times out after 10 seconds of
2126 inactivity.
2127 static void
2129 timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
2130 {
2131 .. ten seconds without any activity
2132 }
2133 ev_timer mytimer;
2135 ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
2136 ev_timer_again (&mytimer); /* start timer */
2137 ev_run (loop, 0);
2138 // and in some piece of code that gets executed on any "activity":
2140 // reset the timeout to start ticking again at 10 seconds
2141 ev_timer_again (&mytimer);
2142 "ev_periodic" - to cron or not to cron?
2144 Periodic watchers are also timers of a kind, but they are very
2145 versatile (and unfortunately a bit complex).
2146 Unlike "ev_timer", periodic watchers are not based on real time (or
2148 relative time, the physical time that passes) but on wall clock time
2149 (absolute time, the thing you can read on your calendar or clock). The
2150 difference is that wall clock time can run faster or slower than real
2151 time, and time jumps are not uncommon (e.g. when you adjust your wrist-
2152 watch).
2153 You can tell a periodic watcher to trigger after some specific point in
2155 time: for example, if you tell a periodic watcher to trigger "in 10
2156 seconds" (by specifying e.g. "ev_now () + 10.", that is, an absolute
2157 time not a delay) and then reset your system clock to January of the
2158 previous year, then it will take a year or more to trigger the event
2159 (unlike an "ev_timer", which would still trigger roughly 10 seconds
2160 after starting it, as it uses a relative timeout).
2161 "ev_periodic" watchers can also be used to implement vastly more
2163 complex timers, such as triggering an event on each "midnight, local
2164 time", or other complicated rules. This cannot be done with "ev_timer"
2165 watchers, as those cannot react to time jumps.
2166 As with timers, the callback is guaranteed to be invoked only when the
2168 point in time where it is supposed to trigger has passed. If multiple
2169 timers become ready during the same loop iteration then the ones with
2170 earlier time-out values are invoked before ones with later time-out
2171 values (but this is no longer true when a callback calls "ev_run"
2172 recursively).
2173 Watcher-Specific Functions and Data Members
2175 ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp
2177 interval, reschedule_cb)
2178 ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval,
2179 reschedule_cb)
2180 Lots of arguments, let's sort it out... There are basically three
2181 modes of operation, and we will explain them from simplest to most
2182 complex:
2183 · absolute timer (offset = absolute time, interval = 0,
2185 reschedule_cb = 0)
2186 In this configuration the watcher triggers an event after the
2188 wall clock time "offset" has passed. It will not repeat and
2189 will not adjust when a time jump occurs, that is, if it is to
2190 be run at January 1st 2011 then it will be stopped and invoked
2191 when the system clock reaches or surpasses this point in time.
2192 · repeating interval timer (offset = offset within interval,
2194 interval > 0, reschedule_cb = 0)
2195 In this mode the watcher will always be scheduled to time out
2197 at the next "offset + N * interval" time (for some integer N,
2198 which can also be negative) and then repeat, regardless of any
2199 time jumps. The "offset" argument is merely an offset into the
2200 "interval" periods.
2201 This can be used to create timers that do not drift with
2203 respect to the system clock, for example, here is an
2204 "ev_periodic" that triggers each hour, on the hour (with
2205 respect to UTC):
2206 ev_periodic_set (&periodic, 0., 3600., 0);
2208 This doesn't mean there will always be 3600 seconds in between
2210 triggers, but only that the callback will be called when the
2211 system time shows a full hour (UTC), or more correctly, when
2212 the system time is evenly divisible by 3600.
2213 Another way to think about it (for the mathematically inclined)
2215 is that "ev_periodic" will try to run the callback in this mode
2216 at the next possible time where "time = offset (mod interval)",
2217 regardless of any time jumps.
2218 The "interval" MUST be positive, and for numerical stability,
2220 the interval value should be higher than "1/8192" (which is
2221 around 100 microseconds) and "offset" should be higher than 0
2222 and should have at most a similar magnitude as the current time
2223 (say, within a factor of ten). Typical values for offset are,
2224 in fact, 0 or something between 0 and "interval", which is also
2225 the recommended range.
2226 Note also that there is an upper limit to how often a timer can
2228 fire (CPU speed for example), so if "interval" is very small
2229 then timing stability will of course deteriorate. Libev itself
2230 tries to be exact to be about one millisecond (if the OS
2231 supports it and the machine is fast enough).
2232 · manual reschedule mode (offset ignored, interval ignored,
2234 reschedule_cb = callback)
2235 In this mode the values for "interval" and "offset" are both
2237 being ignored. Instead, each time the periodic watcher gets
2238 scheduled, the reschedule callback will be called with the
2239 watcher as first, and the current time as second argument.
2240 NOTE: This callback MUST NOT stop or destroy any periodic
2242 watcher, ever, or make ANY other event loop modifications
2243 whatsoever, unless explicitly allowed by documentation here.
2244 If you need to stop it, return "now + 1e30" (or so, fudge
2246 fudge) and stop it afterwards (e.g. by starting an "ev_prepare"
2247 watcher, which is the only event loop modification you are
2248 allowed to do).
2249 The callback prototype is "ev_tstamp
2251 (*reschedule_cb)(ev_periodic *w, ev_tstamp now)", e.g.:
2252 static ev_tstamp
2254 my_rescheduler (ev_periodic *w, ev_tstamp now)
2255 {
2256 return now + 60.;
2257 }
2258 It must return the next time to trigger, based on the passed
2260 time value (that is, the lowest time value larger than to the
2261 second argument). It will usually be called just before the
2262 callback will be triggered, but might be called at other times,
2263 too.
2264 NOTE: This callback must always return a time that is higher
2266 than or equal to the passed "now" value.
2267 This can be used to create very complex timers, such as a timer
2269 that triggers on "next midnight, local time". To do this, you
2270 would calculate the next midnight after "now" and return the
2271 timestamp value for this. How you do this is, again, up to you
2272 (but it is not trivial, which is the main reason I omitted it
2273 as an example).
2274 ev_periodic_again (loop, ev_periodic *)
2276 Simply stops and restarts the periodic watcher again. This is only
2277 useful when you changed some parameters or the reschedule callback
2278 would return a different time than the last time it was called
2279 (e.g. in a crond like program when the crontabs have changed).
2280 ev_tstamp ev_periodic_at (ev_periodic *)
2282 When active, returns the absolute time that the watcher is supposed
2283 to trigger next. This is not the same as the "offset" argument to
2284 "ev_periodic_set", but indeed works even in interval and manual
2285 rescheduling modes.
2286 ev_tstamp offset [read-write]
2288 When repeating, this contains the offset value, otherwise this is
2289 the absolute point in time (the "offset" value passed to
2290 "ev_periodic_set", although libev might modify this value for
2291 better numerical stability).
2292 Can be modified any time, but changes only take effect when the
2294 periodic timer fires or "ev_periodic_again" is being called.
2295 ev_tstamp interval [read-write]
2297 The current interval value. Can be modified any time, but changes
2298 only take effect when the periodic timer fires or
2299 "ev_periodic_again" is being called.
2300 ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
2302 The current reschedule callback, or 0, if this functionality is
2303 switched off. Can be changed any time, but changes only take effect
2304 when the periodic timer fires or "ev_periodic_again" is being
2305 called.
2306 Examples
2308 Example: Call a callback every hour, or, more precisely, whenever the
2310 system time is divisible by 3600. The callback invocation times have
2311 potentially a lot of jitter, but good long-term stability.
2312 static void
2314 clock_cb (struct ev_loop *loop, ev_periodic *w, int revents)
2315 {
2316 ... its now a full hour (UTC, or TAI or whatever your clock follows)
2317 }
2318 ev_periodic hourly_tick;
2320 ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
2321 ev_periodic_start (loop, &hourly_tick);
2322 Example: The same as above, but use a reschedule callback to do it:
2324 #include <math.h>
2326 static ev_tstamp
2328 my_scheduler_cb (ev_periodic *w, ev_tstamp now)
2329 {
2330 return now + (3600. - fmod (now, 3600.));
2331 }
2332 ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
2334 Example: Call a callback every hour, starting now:
2336 ev_periodic hourly_tick;
2338 ev_periodic_init (&hourly_tick, clock_cb,
2339 fmod (ev_now (loop), 3600.), 3600., 0);
2340 ev_periodic_start (loop, &hourly_tick);
2341 "ev_signal" - signal me when a signal gets signalled!
2343 Signal watchers will trigger an event when the process receives a
2344 specific signal one or more times. Even though signals are very
2345 asynchronous, libev will try its best to deliver signals synchronously,
2346 i.e. as part of the normal event processing, like any other event.
2347 If you want signals to be delivered truly asynchronously, just use
2349 "sigaction" as you would do without libev and forget about sharing the
2350 signal. You can even use "ev_async" from a signal handler to
2351 synchronously wake up an event loop.
2352 You can configure as many watchers as you like for the same signal, but
2354 only within the same loop, i.e. you can watch for "SIGINT" in your
2355 default loop and for "SIGIO" in another loop, but you cannot watch for
2356 "SIGINT" in both the default loop and another loop at the same time. At
2357 the moment, "SIGCHLD" is permanently tied to the default loop.
2358 Only after the first watcher for a signal is started will libev
2360 actually register something with the kernel. It thus coexists with your
2361 own signal handlers as long as you don't register any with libev for
2362 the same signal.
2363 If possible and supported, libev will install its handlers with
2365 "SA_RESTART" (or equivalent) behaviour enabled, so system calls should
2366 not be unduly interrupted. If you have a problem with system calls
2367 getting interrupted by signals you can block all signals in an
2368 "ev_check" watcher and unblock them in an "ev_prepare" watcher.
2369 The special problem of inheritance over fork/execve/pthread_create
2371 Both the signal mask ("sigprocmask") and the signal disposition
2373 ("sigaction") are unspecified after starting a signal watcher (and
2374 after stopping it again), that is, libev might or might not block the
2375 signal, and might or might not set or restore the installed signal
2376 handler (but see "EVFLAG_NOSIGMASK").
2377 While this does not matter for the signal disposition (libev never sets
2379 signals to "SIG_IGN", so handlers will be reset to "SIG_DFL" on
2380 "execve"), this matters for the signal mask: many programs do not
2381 expect certain signals to be blocked.
2382 This means that before calling "exec" (from the child) you should reset
2384 the signal mask to whatever "default" you expect (all clear is a good
2385 choice usually).
2386 The simplest way to ensure that the signal mask is reset in the child
2388 is to install a fork handler with "pthread_atfork" that resets it. That
2389 will catch fork calls done by libraries (such as the libc) as well.
2390 In current versions of libev, the signal will not be blocked
2392 indefinitely unless you use the "signalfd" API ("EV_SIGNALFD"). While
2393 this reduces the window of opportunity for problems, it will not go
2394 away, as libev has to modify the signal mask, at least temporarily.
2395 So I can't stress this enough: If you do not reset your signal mask
2397 when you expect it to be empty, you have a race condition in your code.
2398 This is not a libev-specific thing, this is true for most event
2399 libraries.
2400 The special problem of threads signal handling
2402 POSIX threads has problematic signal handling semantics, specifically,
2404 a lot of functionality (sigfd, sigwait etc.) only really works if all
2405 threads in a process block signals, which is hard to achieve.
2406 When you want to use sigwait (or mix libev signal handling with your
2408 own for the same signals), you can tackle this problem by globally
2409 blocking all signals before creating any threads (or creating them with
2410 a fully set sigprocmask) and also specifying the "EVFLAG_NOSIGMASK"
2411 when creating loops. Then designate one thread as "signal receiver
2412 thread" which handles these signals. You can pass on any signals that
2413 libev might be interested in by calling "ev_feed_signal".
2414 Watcher-Specific Functions and Data Members
2416 ev_signal_init (ev_signal *, callback, int signum)
2418 ev_signal_set (ev_signal *, int signum)
2419 Configures the watcher to trigger on the given signal number
2420 (usually one of the "SIGxxx" constants).
2421 int signum [read-only]
2423 The signal the watcher watches out for.
2424 Examples
2426 Example: Try to exit cleanly on SIGINT.
2428 static void
2430 sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
2431 {
2432 ev_break (loop, EVBREAK_ALL);
2433 }
2434 ev_signal signal_watcher;
2436 ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
2437 ev_signal_start (loop, &signal_watcher);
2438 "ev_child" - watch out for process status changes
2440 Child watchers trigger when your process receives a SIGCHLD in response
2441 to some child status changes (most typically when a child of yours dies
2442 or exits). It is permissible to install a child watcher after the child
2443 has been forked (which implies it might have already exited), as long
2444 as the event loop isn't entered (or is continued from a watcher), i.e.,
2445 forking and then immediately registering a watcher for the child is
2446 fine, but forking and registering a watcher a few event loop iterations
2447 later or in the next callback invocation is not.
2448 Only the default event loop is capable of handling signals, and
2450 therefore you can only register child watchers in the default event
2451 loop.
2452 Due to some design glitches inside libev, child watchers will always be
2454 handled at maximum priority (their priority is set to "EV_MAXPRI" by
2455 libev)
2456 Process Interaction
2458 Libev grabs "SIGCHLD" as soon as the default event loop is initialised.
2460 This is necessary to guarantee proper behaviour even if the first child
2461 watcher is started after the child exits. The occurrence of "SIGCHLD"
2462 is recorded asynchronously, but child reaping is done synchronously as
2463 part of the event loop processing. Libev always reaps all children,
2464 even ones not watched.
2465 Overriding the Built-In Processing
2467 Libev offers no special support for overriding the built-in child
2469 processing, but if your application collides with libev's default child
2470 handler, you can override it easily by installing your own handler for
2471 "SIGCHLD" after initialising the default loop, and making sure the
2472 default loop never gets destroyed. You are encouraged, however, to use
2473 an event-based approach to child reaping and thus use libev's support
2474 for that, so other libev users can use "ev_child" watchers freely.
2475 Stopping the Child Watcher
2477 Currently, the child watcher never gets stopped, even when the child
2479 terminates, so normally one needs to stop the watcher in the callback.
2480 Future versions of libev might stop the watcher automatically when a
2481 child exit is detected (calling "ev_child_stop" twice is not a
2482 problem).
2483 Watcher-Specific Functions and Data Members
2485 ev_child_init (ev_child *, callback, int pid, int trace)
2487 ev_child_set (ev_child *, int pid, int trace)
2488 Configures the watcher to wait for status changes of process "pid"
2489 (or any process if "pid" is specified as 0). The callback can look
2490 at the "rstatus" member of the "ev_child" watcher structure to see
2491 the status word (use the macros from "sys/wait.h" and see your
2492 systems "waitpid" documentation). The "rpid" member contains the
2493 pid of the process causing the status change. "trace" must be
2494 either 0 (only activate the watcher when the process terminates) or
2495 1 (additionally activate the watcher when the process is stopped or
2496 continued).
2497 int pid [read-only]
2499 The process id this watcher watches out for, or 0, meaning any
2500 process id.
2501 int rpid [read-write]
2503 The process id that detected a status change.
2504 int rstatus [read-write]
2506 The process exit/trace status caused by "rpid" (see your systems
2507 "waitpid" and "sys/wait.h" documentation for details).
2508 Examples
2510 Example: "fork()" a new process and install a child handler to wait for
2512 its completion.
2513 ev_child cw;
2515 static void
2517 child_cb (EV_P_ ev_child *w, int revents)
2518 {
2519 ev_child_stop (EV_A_ w);
2520 printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
2521 }
2522 pid_t pid = fork ();
2524 if (pid < 0)
2526 // error
2527 else if (pid == 0)
2528 {
2529 // the forked child executes here
2530 exit (1);
2531 }
2532 else
2533 {
2534 ev_child_init (&cw, child_cb, pid, 0);
2535 ev_child_start (EV_DEFAULT_ &cw);
2536 }
2537 "ev_stat" - did the file attributes just change?
2539 This watches a file system path for attribute changes. That is, it
2540 calls "stat" on that path in regular intervals (or when the OS says it
2541 changed) and sees if it changed compared to the last time, invoking the
2542 callback if it did. Starting the watcher "stat"'s the file, so only
2543 changes that happen after the watcher has been started will be
2544 reported.
2545 The path does not need to exist: changing from "path exists" to "path
2547 does not exist" is a status change like any other. The condition "path
2548 does not exist" (or more correctly "path cannot be stat'ed") is
2549 signified by the "st_nlink" field being zero (which is otherwise always
2550 forced to be at least one) and all the other fields of the stat buffer
2551 having unspecified contents.
2552 The path must not end in a slash or contain special components such as
2554 "." or "..". The path should be absolute: If it is relative and your
2555 working directory changes, then the behaviour is undefined.
2556 Since there is no portable change notification interface available, the
2558 portable implementation simply calls stat(2) regularly on the path to
2559 see if it changed somehow. You can specify a recommended polling
2560 interval for this case. If you specify a polling interval of 0 (highly
2561 recommended!) then a suitable, unspecified default value will be used
2562 (which you can expect to be around five seconds, although this might
2563 change dynamically). Libev will also impose a minimum interval which is
2564 currently around 0.1, but that's usually overkill.
2565 This watcher type is not meant for massive numbers of stat watchers, as
2567 even with OS-supported change notifications, this can be resource-
2568 intensive.
2569 At the time of this writing, the only OS-specific interface implemented
2571 is the Linux inotify interface (implementing kqueue support is left as
2572 an exercise for the reader. Note, however, that the author sees no way
2573 of implementing "ev_stat" semantics with kqueue, except as a hint).
2574 ABI Issues (Largefile Support)
2576 Libev by default (unless the user overrides this) uses the default
2578 compilation environment, which means that on systems with large file
2579 support disabled by default, you get the 32 bit version of the stat
2580 structure. When using the library from programs that change the ABI to
2581 use 64 bit file offsets the programs will fail. In that case you have
2582 to compile libev with the same flags to get binary compatibility. This
2583 is obviously the case with any flags that change the ABI, but the
2584 problem is most noticeably displayed with ev_stat and large file
2585 support.
2586 The solution for this is to lobby your distribution maker to make large
2588 file interfaces available by default (as e.g. FreeBSD does) and not
2589 optional. Libev cannot simply switch on large file support because it
2590 has to exchange stat structures with application programs compiled
2591 using the default compilation environment.
2592 Inotify and Kqueue
2594 When "inotify (7)" support has been compiled into libev and present at
2596 runtime, it will be used to speed up change detection where possible.
2597 The inotify descriptor will be created lazily when the first "ev_stat"
2598 watcher is being started.
2599 Inotify presence does not change the semantics of "ev_stat" watchers
2601 except that changes might be detected earlier, and in some cases, to
2602 avoid making regular "stat" calls. Even in the presence of inotify
2603 support there are many cases where libev has to resort to regular
2604 "stat" polling, but as long as kernel 2.6.25 or newer is used (2.6.24
2605 and older have too many bugs), the path exists (i.e. stat succeeds),
2606 and the path resides on a local filesystem (libev currently assumes
2607 only ext2/3, jfs, reiserfs and xfs are fully working) libev usually
2608 gets away without polling.
2609 There is no support for kqueue, as apparently it cannot be used to
2611 implement this functionality, due to the requirement of having a file
2612 descriptor open on the object at all times, and detecting renames,
2613 unlinks etc. is difficult.
2614 "stat ()" is a synchronous operation
2616 Libev doesn't normally do any kind of I/O itself, and so is not
2618 blocking the process. The exception are "ev_stat" watchers - those call
2619 "stat ()", which is a synchronous operation.
2620 For local paths, this usually doesn't matter: unless the system is very
2622 busy or the intervals between stat's are large, a stat call will be
2623 fast, as the path data is usually in memory already (except when
2624 starting the watcher).
2625 For networked file systems, calling "stat ()" can block an indefinite
2627 time due to network issues, and even under good conditions, a stat call
2628 often takes multiple milliseconds.
2629 Therefore, it is best to avoid using "ev_stat" watchers on networked
2631 paths, although this is fully supported by libev.
2632 The special problem of stat time resolution
2634 The "stat ()" system call only supports full-second resolution
2636 portably, and even on systems where the resolution is higher, most file
2637 systems still only support whole seconds.
2638 That means that, if the time is the only thing that changes, you can
2640 easily miss updates: on the first update, "ev_stat" detects a change
2641 and calls your callback, which does something. When there is another
2642 update within the same second, "ev_stat" will be unable to detect
2643 unless the stat data does change in other ways (e.g. file size).
2644 The solution to this is to delay acting on a change for slightly more
2646 than a second (or till slightly after the next full second boundary),
2647 using a roughly one-second-delay "ev_timer" (e.g. "ev_timer_set (w, 0.,
2648 1.02); ev_timer_again (loop, w)").
2649 The .02 offset is added to work around small timing inconsistencies of
2651 some operating systems (where the second counter of the current time
2652 might be be delayed. One such system is the Linux kernel, where a call
2653 to "gettimeofday" might return a timestamp with a full second later
2654 than a subsequent "time" call - if the equivalent of "time ()" is used
2655 to update file times then there will be a small window where the kernel
2656 uses the previous second to update file times but libev might already
2657 execute the timer callback).
2658 Watcher-Specific Functions and Data Members
2660 ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp
2662 interval)
2663 ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
2664 Configures the watcher to wait for status changes of the given
2665 "path". The "interval" is a hint on how quickly a change is
2666 expected to be detected and should normally be specified as 0 to
2667 let libev choose a suitable value. The memory pointed to by "path"
2668 must point to the same path for as long as the watcher is active.
2669 The callback will receive an "EV_STAT" event when a change was
2671 detected, relative to the attributes at the time the watcher was
2672 started (or the last change was detected).
2673 ev_stat_stat (loop, ev_stat *)
2675 Updates the stat buffer immediately with new values. If you change
2676 the watched path in your callback, you could call this function to
2677 avoid detecting this change (while introducing a race condition if
2678 you are not the only one changing the path). Can also be useful
2679 simply to find out the new values.
2680 ev_statdata attr [read-only]
2682 The most-recently detected attributes of the file. Although the
2683 type is "ev_statdata", this is usually the (or one of the) "struct
2684 stat" types suitable for your system, but you can only rely on the
2685 POSIX-standardised members to be present. If the "st_nlink" member
2686 is 0, then there was some error while "stat"ing the file.
2687 ev_statdata prev [read-only]
2689 The previous attributes of the file. The callback gets invoked
2690 whenever "prev" != "attr", or, more precisely, one or more of these
2691 members differ: "st_dev", "st_ino", "st_mode", "st_nlink",
2692 "st_uid", "st_gid", "st_rdev", "st_size", "st_atime", "st_mtime",
2693 "st_ctime".
2694 ev_tstamp interval [read-only]
2696 The specified interval.
2697 const char *path [read-only]
2699 The file system path that is being watched.
2700 Examples
2702 Example: Watch "/etc/passwd" for attribute changes.
2704 static void
2706 passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
2707 {
2708 /* /etc/passwd changed in some way */
2709 if (w->attr.st_nlink)
2710 {
2711 printf ("passwd current size %ld\n", (long)w->attr.st_size);
2712 printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
2713 printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
2714 }
2715 else
2716 /* you shalt not abuse printf for puts */
2717 puts ("wow, /etc/passwd is not there, expect problems. "
2718 "if this is windows, they already arrived\n");
2719 }
2720 ...
2722 ev_stat passwd;
2723 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
2725 ev_stat_start (loop, &passwd);
2726 Example: Like above, but additionally use a one-second delay so we do
2728 not miss updates (however, frequent updates will delay processing, too,
2729 so one might do the work both on "ev_stat" callback invocation and on
2730 "ev_timer" callback invocation).
2731 static ev_stat passwd;
2733 static ev_timer timer;
2734 static void
2736 timer_cb (EV_P_ ev_timer *w, int revents)
2737 {
2738 ev_timer_stop (EV_A_ w);
2739 /* now it's one second after the most recent passwd change */
2741 }
2742 static void
2744 stat_cb (EV_P_ ev_stat *w, int revents)
2745 {
2746 /* reset the one-second timer */
2747 ev_timer_again (EV_A_ &timer);
2748 }
2749 ...
2751 ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
2752 ev_stat_start (loop, &passwd);
2753 ev_timer_init (&timer, timer_cb, 0., 1.02);
2754 "ev_idle" - when you've got nothing better to do...
2756 Idle watchers trigger events when no other events of the same or higher
2757 priority are pending (prepare, check and other idle watchers do not
2758 count as receiving "events").
2759 That is, as long as your process is busy handling sockets or timeouts
2761 (or even signals, imagine) of the same or higher priority it will not
2762 be triggered. But when your process is idle (or only lower-priority
2763 watchers are pending), the idle watchers are being called once per
2764 event loop iteration - until stopped, that is, or your process receives
2765 more events and becomes busy again with higher priority stuff.
2766 The most noteworthy effect is that as long as any idle watchers are
2768 active, the process will not block when waiting for new events.
2769 Apart from keeping your process non-blocking (which is a useful effect
2771 on its own sometimes), idle watchers are a good place to do "pseudo-
2772 background processing", or delay processing stuff to after the event
2773 loop has handled all outstanding events.
2774 Abusing an "ev_idle" watcher for its side-effect
2776 As long as there is at least one active idle watcher, libev will never
2778 sleep unnecessarily. Or in other words, it will loop as fast as
2779 possible. For this to work, the idle watcher doesn't need to be
2780 invoked at all - the lowest priority will do.
2781 This mode of operation can be useful together with an "ev_check"
2783 watcher, to do something on each event loop iteration - for example to
2784 balance load between different connections.
2785 See "Abusing an ev_check watcher for its side-effect" for a longer
2787 example.
2788 Watcher-Specific Functions and Data Members
2790 ev_idle_init (ev_idle *, callback)
2792 Initialises and configures the idle watcher - it has no parameters
2793 of any kind. There is a "ev_idle_set" macro, but using it is
2794 utterly pointless, believe me.
2795 Examples
2797 Example: Dynamically allocate an "ev_idle" watcher, start it, and in
2799 the callback, free it. Also, use no error checking, as usual.
2800 static void
2802 idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
2803 {
2804 // stop the watcher
2805 ev_idle_stop (loop, w);
2806 // now we can free it
2808 free (w);
2809 // now do something you wanted to do when the program has
2811 // no longer anything immediate to do.
2812 }
2813 ev_idle *idle_watcher = malloc (sizeof (ev_idle));
2815 ev_idle_init (idle_watcher, idle_cb);
2816 ev_idle_start (loop, idle_watcher);
2817 "ev_prepare" and "ev_check" - customise your event loop!
2819 Prepare and check watchers are often (but not always) used in pairs:
2820 prepare watchers get invoked before the process blocks and check
2821 watchers afterwards.
2822 You must not call "ev_run" (or similar functions that enter the current
2824 event loop) or "ev_loop_fork" from either "ev_prepare" or "ev_check"
2825 watchers. Other loops than the current one are fine, however. The
2826 rationale behind this is that you do not need to check for recursion in
2827 those watchers, i.e. the sequence will always be "ev_prepare",
2828 blocking, "ev_check" so if you have one watcher of each kind they will
2829 always be called in pairs bracketing the blocking call.
2830 Their main purpose is to integrate other event mechanisms into libev
2832 and their use is somewhat advanced. They could be used, for example, to
2833 track variable changes, implement your own watchers, integrate net-snmp
2834 or a coroutine library and lots more. They are also occasionally useful
2835 if you cache some data and want to flush it before blocking (for
2836 example, in X programs you might want to do an "XFlush ()" in an
2837 "ev_prepare" watcher).
2838 This is done by examining in each prepare call which file descriptors
2840 need to be watched by the other library, registering "ev_io" watchers
2841 for them and starting an "ev_timer" watcher for any timeouts (many
2842 libraries provide exactly this functionality). Then, in the check
2843 watcher, you check for any events that occurred (by checking the
2844 pending status of all watchers and stopping them) and call back into
2845 the library. The I/O and timer callbacks will never actually be called
2846 (but must be valid nevertheless, because you never know, you know?).
2847 As another example, the Perl Coro module uses these hooks to integrate
2849 coroutines into libev programs, by yielding to other active coroutines
2850 during each prepare and only letting the process block if no coroutines
2851 are ready to run (it's actually more complicated: it only runs
2852 coroutines with priority higher than or equal to the event loop and one
2853 coroutine of lower priority, but only once, using idle watchers to keep
2854 the event loop from blocking if lower-priority coroutines are active,
2855 thus mapping low-priority coroutines to idle/background tasks).
2856 When used for this purpose, it is recommended to give "ev_check"
2858 watchers highest ("EV_MAXPRI") priority, to ensure that they are being
2859 run before any other watchers after the poll (this doesn't matter for
2860 "ev_prepare" watchers).
2861 Also, "ev_check" watchers (and "ev_prepare" watchers, too) should not
2863 activate ("feed") events into libev. While libev fully supports this,
2864 they might get executed before other "ev_check" watchers did their job.
2865 As "ev_check" watchers are often used to embed other (non-libev) event
2866 loops those other event loops might be in an unusable state until their
2867 "ev_check" watcher ran (always remind yourself to coexist peacefully
2868 with others).
2869 Abusing an "ev_check" watcher for its side-effect
2871 "ev_check" (and less often also "ev_prepare") watchers can also be
2873 useful because they are called once per event loop iteration. For
2874 example, if you want to handle a large number of connections fairly,
2875 you normally only do a bit of work for each active connection, and if
2876 there is more work to do, you wait for the next event loop iteration,
2877 so other connections have a chance of making progress.
2878 Using an "ev_check" watcher is almost enough: it will be called on the
2880 next event loop iteration. However, that isn't as soon as possible -
2881 without external events, your "ev_check" watcher will not be invoked.
2882 This is where "ev_idle" watchers come in handy - all you need is a
2884 single global idle watcher that is active as long as you have one
2885 active "ev_check" watcher. The "ev_idle" watcher makes sure the event
2886 loop will not sleep, and the "ev_check" watcher makes sure a callback
2887 gets invoked. Neither watcher alone can do that.
2888 Watcher-Specific Functions and Data Members
2890 ev_prepare_init (ev_prepare *, callback)
2892 ev_check_init (ev_check *, callback)
2893 Initialises and configures the prepare or check watcher - they have
2894 no parameters of any kind. There are "ev_prepare_set" and
2895 "ev_check_set" macros, but using them is utterly, utterly, utterly
2896 and completely pointless.
2897 Examples
2899 There are a number of principal ways to embed other event loops or
2901 modules into libev. Here are some ideas on how to include libadns into
2902 libev (there is a Perl module named "EV::ADNS" that does this, which
2903 you could use as a working example. Another Perl module named
2904 "EV::Glib" embeds a Glib main context into libev, and finally,
2905 "Glib::EV" embeds EV into the Glib event loop).
2906 Method 1: Add IO watchers and a timeout watcher in a prepare handler,
2908 and in a check watcher, destroy them and call into libadns. What
2909 follows is pseudo-code only of course. This requires you to either use
2910 a low priority for the check watcher or use "ev_clear_pending"
2911 explicitly, as the callbacks for the IO/timeout watchers might not have
2912 been called yet.
2913 static ev_io iow [nfd];
2915 static ev_timer tw;
2916 static void
2918 io_cb (struct ev_loop *loop, ev_io *w, int revents)
2919 {
2920 }
2921 // create io watchers for each fd and a timer before blocking
2923 static void
2924 adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
2925 {
2926 int timeout = 3600000;
2927 struct pollfd fds [nfd];
2928 // actual code will need to loop here and realloc etc.
2929 adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
2930 /* the callback is illegal, but won't be called as we stop during check */
2932 ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
2933 ev_timer_start (loop, &tw);
2934 // create one ev_io per pollfd
2936 for (int i = 0; i < nfd; ++i)
2937 {
2938 ev_io_init (iow + i, io_cb, fds [i].fd,
2939 ((fds [i].events & POLLIN ? EV_READ : 0)
2940 | (fds [i].events & POLLOUT ? EV_WRITE : 0)));
2941 fds [i].revents = 0;
2943 ev_io_start (loop, iow + i);
2944 }
2945 }
2946 // stop all watchers after blocking
2948 static void
2949 adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
2950 {
2951 ev_timer_stop (loop, &tw);
2952 for (int i = 0; i < nfd; ++i)
2954 {
2955 // set the relevant poll flags
2956 // could also call adns_processreadable etc. here
2957 struct pollfd *fd = fds + i;
2958 int revents = ev_clear_pending (iow + i);
2959 if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
2960 if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
2961 // now stop the watcher
2963 ev_io_stop (loop, iow + i);
2964 }
2965 adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
2967 }
2968 Method 2: This would be just like method 1, but you run
2970 "adns_afterpoll" in the prepare watcher and would dispose of the check
2971 watcher.
2972 Method 3: If the module to be embedded supports explicit event
2974 notification (libadns does), you can also make use of the actual
2975 watcher callbacks, and only destroy/create the watchers in the prepare
2976 watcher.
2977 static void
2979 timer_cb (EV_P_ ev_timer *w, int revents)
2980 {
2981 adns_state ads = (adns_state)w->data;
2982 update_now (EV_A);
2983 adns_processtimeouts (ads, &tv_now);
2985 }
2986 static void
2988 io_cb (EV_P_ ev_io *w, int revents)
2989 {
2990 adns_state ads = (adns_state)w->data;
2991 update_now (EV_A);
2992 if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
2994 if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
2995 }
2996 // do not ever call adns_afterpoll
2998 Method 4: Do not use a prepare or check watcher because the module you
3000 want to embed is not flexible enough to support it. Instead, you can
3001 override their poll function. The drawback with this solution is that
3002 the main loop is now no longer controllable by EV. The "Glib::EV"
3003 module uses this approach, effectively embedding EV as a client into
3004 the horrible libglib event loop.
3005 static gint
3007 event_poll_func (GPollFD *fds, guint nfds, gint timeout)
3008 {
3009 int got_events = 0;
3010 for (n = 0; n < nfds; ++n)
3012 // create/start io watcher that sets the relevant bits in fds[n] and increment got_events
3013 if (timeout >= 0)
3015 // create/start timer
3016 // poll
3018 ev_run (EV_A_ 0);
3019 // stop timer again
3021 if (timeout >= 0)
3022 ev_timer_stop (EV_A_ &to);
3023 // stop io watchers again - their callbacks should have set
3025 for (n = 0; n < nfds; ++n)
3026 ev_io_stop (EV_A_ iow [n]);
3027 return got_events;
3029 }
3030 "ev_embed" - when one backend isn't enough...
3032 This is a rather advanced watcher type that lets you embed one event
3033 loop into another (currently only "ev_io" events are supported in the
3034 embedded loop, other types of watchers might be handled in a delayed or
3035 incorrect fashion and must not be used).
3036 There are primarily two reasons you would want that: work around bugs
3038 and prioritise I/O.
3039 As an example for a bug workaround, the kqueue backend might only
3041 support sockets on some platform, so it is unusable as generic backend,
3042 but you still want to make use of it because you have many sockets and
3043 it scales so nicely. In this case, you would create a kqueue-based loop
3044 and embed it into your default loop (which might use e.g. poll).
3045 Overall operation will be a bit slower because first libev has to call
3046 "poll" and then "kevent", but at least you can use both mechanisms for
3047 what they are best: "kqueue" for scalable sockets and "poll" if you
3048 want it to work :)
3049 As for prioritising I/O: under rare circumstances you have the case
3051 where some fds have to be watched and handled very quickly (with low
3052 latency), and even priorities and idle watchers might have too much
3053 overhead. In this case you would put all the high priority stuff in one
3054 loop and all the rest in a second one, and embed the second one in the
3055 first.
3056 As long as the watcher is active, the callback will be invoked every
3058 time there might be events pending in the embedded loop. The callback
3059 must then call "ev_embed_sweep (mainloop, watcher)" to make a single
3060 sweep and invoke their callbacks (the callback doesn't need to invoke
3061 the "ev_embed_sweep" function directly, it could also start an idle
3062 watcher to give the embedded loop strictly lower priority for example).
3063 You can also set the callback to 0, in which case the embed watcher
3065 will automatically execute the embedded loop sweep whenever necessary.
3066 Fork detection will be handled transparently while the "ev_embed"
3068 watcher is active, i.e., the embedded loop will automatically be forked
3069 when the embedding loop forks. In other cases, the user is responsible
3070 for calling "ev_loop_fork" on the embedded loop.
3071 Unfortunately, not all backends are embeddable: only the ones returned
3073 by "ev_embeddable_backends" are, which, unfortunately, does not include
3074 any portable one.
3075 So when you want to use this feature you will always have to be
3077 prepared that you cannot get an embeddable loop. The recommended way to
3078 get around this is to have a separate variables for your embeddable
3079 loop, try to create it, and if that fails, use the normal loop for
3080 everything.
3081 "ev_embed" and fork
3083 While the "ev_embed" watcher is running, forks in the embedding loop
3085 will automatically be applied to the embedded loop as well, so no
3086 special fork handling is required in that case. When the watcher is not
3087 running, however, it is still the task of the libev user to call
3088 "ev_loop_fork ()" as applicable.
3089 Watcher-Specific Functions and Data Members
3091 ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
3093 ev_embed_set (ev_embed *, struct ev_loop *embedded_loop)
3094 Configures the watcher to embed the given loop, which must be
3095 embeddable. If the callback is 0, then "ev_embed_sweep" will be
3096 invoked automatically, otherwise it is the responsibility of the
3097 callback to invoke it (it will continue to be called until the
3098 sweep has been done, if you do not want that, you need to
3099 temporarily stop the embed watcher).
3100 ev_embed_sweep (loop, ev_embed *)
3102 Make a single, non-blocking sweep over the embedded loop. This
3103 works similarly to "ev_run (embedded_loop, EVRUN_NOWAIT)", but in
3104 the most appropriate way for embedded loops.
3105 struct ev_loop *other [read-only]
3107 The embedded event loop.
3108 Examples
3110 Example: Try to get an embeddable event loop and embed it into the
3112 default event loop. If that is not possible, use the default loop. The
3113 default loop is stored in "loop_hi", while the embeddable loop is
3114 stored in "loop_lo" (which is "loop_hi" in the case no embeddable loop
3115 can be used).
3116 struct ev_loop *loop_hi = ev_default_init (0);
3118 struct ev_loop *loop_lo = 0;
3119 ev_embed embed;
3120 // see if there is a chance of getting one that works
3122 // (remember that a flags value of 0 means autodetection)
3123 loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
3124 ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
3125 : 0;
3126 // if we got one, then embed it, otherwise default to loop_hi
3128 if (loop_lo)
3129 {
3130 ev_embed_init (&embed, 0, loop_lo);
3131 ev_embed_start (loop_hi, &embed);
3132 }
3133 else
3134 loop_lo = loop_hi;
3135 Example: Check if kqueue is available but not recommended and create a
3137 kqueue backend for use with sockets (which usually work with any kqueue
3138 implementation). Store the kqueue/socket-only event loop in
3139 "loop_socket". (One might optionally use "EVFLAG_NOENV", too).
3140 struct ev_loop *loop = ev_default_init (0);
3142 struct ev_loop *loop_socket = 0;
3143 ev_embed embed;
3144 if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
3146 if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
3147 {
3148 ev_embed_init (&embed, 0, loop_socket);
3149 ev_embed_start (loop, &embed);
3150 }
3151 if (!loop_socket)
3153 loop_socket = loop;
3154 // now use loop_socket for all sockets, and loop for everything else
3156 "ev_fork" - the audacity to resume the event loop after a fork
3158 Fork watchers are called when a "fork ()" was detected (usually because
3159 whoever is a good citizen cared to tell libev about it by calling
3160 "ev_loop_fork"). The invocation is done before the event loop blocks
3161 next and before "ev_check" watchers are being called, and only in the
3162 child after the fork. If whoever good citizen calling "ev_default_fork"
3163 cheats and calls it in the wrong process, the fork handlers will be
3164 invoked, too, of course.
3165 The special problem of life after fork - how is it possible?
3167 Most uses of "fork ()" consist of forking, then some simple calls to
3169 set up/change the process environment, followed by a call to "exec()".
3170 This sequence should be handled by libev without any problems.
3171 This changes when the application actually wants to do event handling
3173 in the child, or both parent in child, in effect "continuing" after the
3174 fork.
3175 The default mode of operation (for libev, with application help to
3177 detect forks) is to duplicate all the state in the child, as would be
3178 expected when either the parent or the child process continues.
3179 When both processes want to continue using libev, then this is usually
3181 the wrong result. In that case, usually one process (typically the
3182 parent) is supposed to continue with all watchers in place as before,
3183 while the other process typically wants to start fresh, i.e. without
3184 any active watchers.
3185 The cleanest and most efficient way to achieve that with libev is to
3187 simply create a new event loop, which of course will be "empty", and
3188 use that for new watchers. This has the advantage of not touching more
3189 memory than necessary, and thus avoiding the copy-on-write, and the
3190 disadvantage of having to use multiple event loops (which do not
3191 support signal watchers).
3192 When this is not possible, or you want to use the default loop for
3194 other reasons, then in the process that wants to start "fresh", call
3195 "ev_loop_destroy (EV_DEFAULT)" followed by "ev_default_loop (...)".
3196 Destroying the default loop will "orphan" (not stop) all registered
3197 watchers, so you have to be careful not to execute code that modifies
3198 those watchers. Note also that in that case, you have to re-register
3199 any signal watchers.
3200 Watcher-Specific Functions and Data Members
3202 ev_fork_init (ev_fork *, callback)
3204 Initialises and configures the fork watcher - it has no parameters
3205 of any kind. There is a "ev_fork_set" macro, but using it is
3206 utterly pointless, really.
3207 "ev_cleanup" - even the best things end
3209 Cleanup watchers are called just before the event loop is being
3210 destroyed by a call to "ev_loop_destroy".
3211 While there is no guarantee that the event loop gets destroyed, cleanup
3213 watchers provide a convenient method to install cleanup hooks for your
3214 program, worker threads and so on - you just to make sure to destroy
3215 the loop when you want them to be invoked.
3216 Cleanup watchers are invoked in the same way as any other watcher.
3218 Unlike all other watchers, they do not keep a reference to the event
3219 loop (which makes a lot of sense if you think about it). Like all other
3220 watchers, you can call libev functions in the callback, except
3221 "ev_cleanup_start".
3222 Watcher-Specific Functions and Data Members
3224 ev_cleanup_init (ev_cleanup *, callback)
3226 Initialises and configures the cleanup watcher - it has no
3227 parameters of any kind. There is a "ev_cleanup_set" macro, but
3228 using it is utterly pointless, I assure you.
3229 Example: Register an atexit handler to destroy the default loop, so any
3231 cleanup functions are called.
3232 static void
3234 program_exits (void)
3235 {
3236 ev_loop_destroy (EV_DEFAULT_UC);
3237 }
3238 ...
3240 atexit (program_exits);
3241 "ev_async" - how to wake up an event loop
3243 In general, you cannot use an "ev_loop" from multiple threads or other
3244 asynchronous sources such as signal handlers (as opposed to multiple
3245 event loops - those are of course safe to use in different threads).
3246 Sometimes, however, you need to wake up an event loop you do not
3248 control, for example because it belongs to another thread. This is what
3249 "ev_async" watchers do: as long as the "ev_async" watcher is active,
3250 you can signal it by calling "ev_async_send", which is thread- and
3251 signal safe.
3252 This functionality is very similar to "ev_signal" watchers, as signals,
3254 too, are asynchronous in nature, and signals, too, will be compressed
3255 (i.e. the number of callback invocations may be less than the number of
3256 "ev_async_send" calls). In fact, you could use signal watchers as a
3257 kind of "global async watchers" by using a watcher on an otherwise
3258 unused signal, and "ev_feed_signal" to signal this watcher from another
3259 thread, even without knowing which loop owns the signal.
3260 Queueing
3262 "ev_async" does not support queueing of data in any way. The reason is
3264 that the author does not know of a simple (or any) algorithm for a
3265 multiple-writer-single-reader queue that works in all cases and doesn't
3266 need elaborate support such as pthreads or unportable memory access
3267 semantics.
3268 That means that if you want to queue data, you have to provide your own
3270 queue. But at least I can tell you how to implement locking around your
3271 queue:
3272 queueing from a signal handler context
3274 To implement race-free queueing, you simply add to the queue in the
3275 signal handler but you block the signal handler in the watcher
3276 callback. Here is an example that does that for some fictitious
3277 SIGUSR1 handler:
3278 static ev_async mysig;
3280 static void
3282 sigusr1_handler (void)
3283 {
3284 sometype data;
3285 // no locking etc.
3287 queue_put (data);
3288 ev_async_send (EV_DEFAULT_ &mysig);
3289 }
3290 static void
3292 mysig_cb (EV_P_ ev_async *w, int revents)
3293 {
3294 sometype data;
3295 sigset_t block, prev;
3296 sigemptyset (&block);
3298 sigaddset (&block, SIGUSR1);
3299 sigprocmask (SIG_BLOCK, &block, &prev);
3300 while (queue_get (&data))
3302 process (data);
3303 if (sigismember (&prev, SIGUSR1)
3305 sigprocmask (SIG_UNBLOCK, &block, 0);
3306 }
3307 (Note: pthreads in theory requires you to use "pthread_setmask"
3309 instead of "sigprocmask" when you use threads, but libev doesn't do
3310 it either...).
3311 queueing from a thread context
3313 The strategy for threads is different, as you cannot (easily) block
3314 threads but you can easily preempt them, so to queue safely you
3315 need to employ a traditional mutex lock, such as in this pthread
3316 example:
3317 static ev_async mysig;
3319 static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
3320 static void
3322 otherthread (void)
3323 {
3324 // only need to lock the actual queueing operation
3325 pthread_mutex_lock (&mymutex);
3326 queue_put (data);
3327 pthread_mutex_unlock (&mymutex);
3328 ev_async_send (EV_DEFAULT_ &mysig);
3330 }
3331 static void
3333 mysig_cb (EV_P_ ev_async *w, int revents)
3334 {
3335 pthread_mutex_lock (&mymutex);
3336 while (queue_get (&data))
3338 process (data);
3339 pthread_mutex_unlock (&mymutex);
3341 }
3342 Watcher-Specific Functions and Data Members
3344 ev_async_init (ev_async *, callback)
3346 Initialises and configures the async watcher - it has no parameters
3347 of any kind. There is a "ev_async_set" macro, but using it is
3348 utterly pointless, trust me.
3349 ev_async_send (loop, ev_async *)
3351 Sends/signals/activates the given "ev_async" watcher, that is,
3352 feeds an "EV_ASYNC" event on the watcher into the event loop, and
3353 instantly returns.
3354 Unlike "ev_feed_event", this call is safe to do from other threads,
3356 signal or similar contexts (see the discussion of "EV_ATOMIC_T" in
3357 the embedding section below on what exactly this means).
3358 Note that, as with other watchers in libev, multiple events might
3360 get compressed into a single callback invocation (another way to
3361 look at this is that "ev_async" watchers are level-triggered: they
3362 are set on "ev_async_send", reset when the event loop detects
3363 that).
3364 This call incurs the overhead of at most one extra system call per
3366 event loop iteration, if the event loop is blocked, and no syscall
3367 at all if the event loop (or your program) is processing events.
3368 That means that repeated calls are basically free (there is no need
3369 to avoid calls for performance reasons) and that the overhead
3370 becomes smaller (typically zero) under load.
3371 bool = ev_async_pending (ev_async *)
3373 Returns a non-zero value when "ev_async_send" has been called on
3374 the watcher but the event has not yet been processed (or even
3375 noted) by the event loop.
3376 "ev_async_send" sets a flag in the watcher and wakes up the loop.
3378 When the loop iterates next and checks for the watcher to have
3379 become active, it will reset the flag again. "ev_async_pending" can
3380 be used to very quickly check whether invoking the loop might be a
3381 good idea.
3382 Not that this does not check whether the watcher itself is pending,
3384 only whether it has been requested to make this watcher pending:
3385 there is a time window between the event loop checking and
3386 resetting the async notification, and the callback being invoked.
3387
3389 There are some other functions of possible interest. Described. Here.
3390 Now.
3391 ev_once (loop, int fd, int events, ev_tstamp timeout, callback)
3393 This function combines a simple timer and an I/O watcher, calls
3394 your callback on whichever event happens first and automatically
3395 stops both watchers. This is useful if you want to wait for a
3396 single event on an fd or timeout without having to
3397 allocate/configure/start/stop/free one or more watchers yourself.
3398 If "fd" is less than 0, then no I/O watcher will be started and the
3400 "events" argument is being ignored. Otherwise, an "ev_io" watcher
3401 for the given "fd" and "events" set will be created and started.
3402 If "timeout" is less than 0, then no timeout watcher will be
3404 started. Otherwise an "ev_timer" watcher with after = "timeout"
3405 (and repeat = 0) will be started. 0 is a valid timeout.
3406 The callback has the type "void (*cb)(int revents, void *arg)" and
3408 is passed an "revents" set like normal event callbacks (a
3409 combination of "EV_ERROR", "EV_READ", "EV_WRITE" or "EV_TIMER") and
3410 the "arg" value passed to "ev_once". Note that it is possible to
3411 receive both a timeout and an io event at the same time - you
3412 probably should give io events precedence.
3413 Example: wait up to ten seconds for data to appear on STDIN_FILENO.
3415 static void stdin_ready (int revents, void *arg)
3417 {
3418 if (revents & EV_READ)
3419 /* stdin might have data for us, joy! */;
3420 else if (revents & EV_TIMER)
3421 /* doh, nothing entered */;
3422 }
3423 ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
3425 ev_feed_fd_event (loop, int fd, int revents)
3427 Feed an event on the given fd, as if a file descriptor backend
3428 detected the given events.
3429 ev_feed_signal_event (loop, int signum)
3431 Feed an event as if the given signal occurred. See also
3432 "ev_feed_signal", which is async-safe.
3433
3435 This section explains some common idioms that are not immediately
3436 obvious. Note that examples are sprinkled over the whole manual, and
3437 this section only contains stuff that wouldn't fit anywhere else.
3438 ASSOCIATING CUSTOM DATA WITH A WATCHER
3440 Each watcher has, by default, a "void *data" member that you can read
3441 or modify at any time: libev will completely ignore it. This can be
3442 used to associate arbitrary data with your watcher. If you need more
3443 data and don't want to allocate memory separately and store a pointer
3444 to it in that data member, you can also "subclass" the watcher type and
3445 provide your own data:
3446 struct my_io
3448 {
3449 ev_io io;
3450 int otherfd;
3451 void *somedata;
3452 struct whatever *mostinteresting;
3453 };
3454 ...
3456 struct my_io w;
3457 ev_io_init (&w.io, my_cb, fd, EV_READ);
3458 And since your callback will be called with a pointer to the watcher,
3460 you can cast it back to your own type:
3461 static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
3463 {
3464 struct my_io *w = (struct my_io *)w_;
3465 ...
3466 }
3467 More interesting and less C-conformant ways of casting your callback
3469 function type instead have been omitted.
3470 BUILDING YOUR OWN COMPOSITE WATCHERS
3472 Another common scenario is to use some data structure with multiple
3473 embedded watchers, in effect creating your own watcher that combines
3474 multiple libev event sources into one "super-watcher":
3475 struct my_biggy
3477 {
3478 int some_data;
3479 ev_timer t1;
3480 ev_timer t2;
3481 }
3482 In this case getting the pointer to "my_biggy" is a bit more
3484 complicated: Either you store the address of your "my_biggy" struct in
3485 the "data" member of the watcher (for woozies or C++ coders), or you
3486 need to use some pointer arithmetic using "offsetof" inside your
3487 watchers (for real programmers):
3488 #include <stddef.h>
3490 static void
3492 t1_cb (EV_P_ ev_timer *w, int revents)
3493 {
3494 struct my_biggy big = (struct my_biggy *)
3495 (((char *)w) - offsetof (struct my_biggy, t1));
3496 }
3497 static void
3499 t2_cb (EV_P_ ev_timer *w, int revents)
3500 {
3501 struct my_biggy big = (struct my_biggy *)
3502 (((char *)w) - offsetof (struct my_biggy, t2));
3503 }
3504 AVOIDING FINISHING BEFORE RETURNING
3506 Often you have structures like this in event-based programs:
3507 callback ()
3509 {
3510 free (request);
3511 }
3512 request = start_new_request (..., callback);
3514 The intent is to start some "lengthy" operation. The "request" could be
3516 used to cancel the operation, or do other things with it.
3517 It's not uncommon to have code paths in "start_new_request" that
3519 immediately invoke the callback, for example, to report errors. Or you
3520 add some caching layer that finds that it can skip the lengthy aspects
3521 of the operation and simply invoke the callback with the result.
3522 The problem here is that this will happen before "start_new_request"
3524 has returned, so "request" is not set.
3525 Even if you pass the request by some safer means to the callback, you
3527 might want to do something to the request after starting it, such as
3528 canceling it, which probably isn't working so well when the callback
3529 has already been invoked.
3530 A common way around all these issues is to make sure that
3532 "start_new_request" always returns before the callback is invoked. If
3533 "start_new_request" immediately knows the result, it can artificially
3534 delay invoking the callback by using a "prepare" or "idle" watcher for
3535 example, or more sneakily, by reusing an existing (stopped) watcher and
3536 pushing it into the pending queue:
3537 ev_set_cb (watcher, callback);
3539 ev_feed_event (EV_A_ watcher, 0);
3540 This way, "start_new_request" can safely return before the callback is
3542 invoked, while not delaying callback invocation too much.
3543 MODEL/NESTED EVENT LOOP INVOCATIONS AND EXIT CONDITIONS
3545 Often (especially in GUI toolkits) there are places where you have
3546 modal interaction, which is most easily implemented by recursively
3547 invoking "ev_run".
3548 This brings the problem of exiting - a callback might want to finish
3550 the main "ev_run" call, but not the nested one (e.g. user clicked
3551 "Quit", but a modal "Are you sure?" dialog is still waiting), or just
3552 the nested one and not the main one (e.g. user clocked "Ok" in a modal
3553 dialog), or some other combination: In these cases, a simple "ev_break"
3554 will not work.
3555 The solution is to maintain "break this loop" variable for each
3557 "ev_run" invocation, and use a loop around "ev_run" until the condition
3558 is triggered, using "EVRUN_ONCE":
3559 // main loop
3561 int exit_main_loop = 0;
3562 while (!exit_main_loop)
3564 ev_run (EV_DEFAULT_ EVRUN_ONCE);
3565 // in a modal watcher
3567 int exit_nested_loop = 0;
3568 while (!exit_nested_loop)
3570 ev_run (EV_A_ EVRUN_ONCE);
3571 To exit from any of these loops, just set the corresponding exit
3573 variable:
3574 // exit modal loop
3576 exit_nested_loop = 1;
3577 // exit main program, after modal loop is finished
3579 exit_main_loop = 1;
3580 // exit both
3582 exit_main_loop = exit_nested_loop = 1;
3583 THREAD LOCKING EXAMPLE
3585 Here is a fictitious example of how to run an event loop in a different
3586 thread from where callbacks are being invoked and watchers are
3587 created/added/removed.
3588 For a real-world example, see the "EV::Loop::Async" perl module, which
3590 uses exactly this technique (which is suited for many high-level
3591 languages).
3592 The example uses a pthread mutex to protect the loop data, a condition
3594 variable to wait for callback invocations, an async watcher to notify
3595 the event loop thread and an unspecified mechanism to wake up the main
3596 thread.
3597 First, you need to associate some data with the event loop:
3599 typedef struct {
3601 mutex_t lock; /* global loop lock */
3602 ev_async async_w;
3603 thread_t tid;
3604 cond_t invoke_cv;
3605 } userdata;
3606 void prepare_loop (EV_P)
3608 {
3609 // for simplicity, we use a static userdata struct.
3610 static userdata u;
3611 ev_async_init (&u->async_w, async_cb);
3613 ev_async_start (EV_A_ &u->async_w);
3614 pthread_mutex_init (&u->lock, 0);
3616 pthread_cond_init (&u->invoke_cv, 0);
3617 // now associate this with the loop
3619 ev_set_userdata (EV_A_ u);
3620 ev_set_invoke_pending_cb (EV_A_ l_invoke);
3621 ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
3622 // then create the thread running ev_run
3624 pthread_create (&u->tid, 0, l_run, EV_A);
3625 }
3626 The callback for the "ev_async" watcher does nothing: the watcher is
3628 used solely to wake up the event loop so it takes notice of any new
3629 watchers that might have been added:
3630 static void
3632 async_cb (EV_P_ ev_async *w, int revents)
3633 {
3634 // just used for the side effects
3635 }
3636 The "l_release" and "l_acquire" callbacks simply unlock/lock the mutex
3638 protecting the loop data, respectively.
3639 static void
3641 l_release (EV_P)
3642 {
3643 userdata *u = ev_userdata (EV_A);
3644 pthread_mutex_unlock (&u->lock);
3645 }
3646 static void
3648 l_acquire (EV_P)
3649 {
3650 userdata *u = ev_userdata (EV_A);
3651 pthread_mutex_lock (&u->lock);
3652 }
3653 The event loop thread first acquires the mutex, and then jumps straight
3655 into "ev_run":
3656 void *
3658 l_run (void *thr_arg)
3659 {
3660 struct ev_loop *loop = (struct ev_loop *)thr_arg;
3661 l_acquire (EV_A);
3663 pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
3664 ev_run (EV_A_ 0);
3665 l_release (EV_A);
3666 return 0;
3668 }
3669 Instead of invoking all pending watchers, the "l_invoke" callback will
3671 signal the main thread via some unspecified mechanism (signals? pipe
3672 writes? "Async::Interrupt"?) and then waits until all pending watchers
3673 have been called (in a while loop because a) spurious wakeups are
3674 possible and b) skipping inter-thread-communication when there are no
3675 pending watchers is very beneficial):
3676 static void
3678 l_invoke (EV_P)
3679 {
3680 userdata *u = ev_userdata (EV_A);
3681 while (ev_pending_count (EV_A))
3683 {
3684 wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
3685 pthread_cond_wait (&u->invoke_cv, &u->lock);
3686 }
3687 }
3688 Now, whenever the main thread gets told to invoke pending watchers, it
3690 will grab the lock, call "ev_invoke_pending" and then signal the loop
3691 thread to continue:
3692 static void
3694 real_invoke_pending (EV_P)
3695 {
3696 userdata *u = ev_userdata (EV_A);
3697 pthread_mutex_lock (&u->lock);
3699 ev_invoke_pending (EV_A);
3700 pthread_cond_signal (&u->invoke_cv);
3701 pthread_mutex_unlock (&u->lock);
3702 }
3703 Whenever you want to start/stop a watcher or do other modifications to
3705 an event loop, you will now have to lock:
3706 ev_timer timeout_watcher;
3708 userdata *u = ev_userdata (EV_A);
3709 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
3711 pthread_mutex_lock (&u->lock);
3713 ev_timer_start (EV_A_ &timeout_watcher);
3714 ev_async_send (EV_A_ &u->async_w);
3715 pthread_mutex_unlock (&u->lock);
3716 Note that sending the "ev_async" watcher is required because otherwise
3718 an event loop currently blocking in the kernel will have no knowledge
3719 about the newly added timer. By waking up the loop it will pick up any
3720 new watchers in the next event loop iteration.
3721 THREADS, COROUTINES, CONTINUATIONS, QUEUES... INSTEAD OF CALLBACKS
3723 While the overhead of a callback that e.g. schedules a thread is small,
3724 it is still an overhead. If you embed libev, and your main usage is
3725 with some kind of threads or coroutines, you might want to customise
3726 libev so that doesn't need callbacks anymore.
3727 Imagine you have coroutines that you can switch to using a function
3729 "switch_to (coro)", that libev runs in a coroutine called "libev_coro"
3730 and that due to some magic, the currently active coroutine is stored in
3731 a global called "current_coro". Then you can build your own "wait for
3732 libev event" primitive by changing "EV_CB_DECLARE" and "EV_CB_INVOKE"
3733 (note the differing ";" conventions):
3734 #define EV_CB_DECLARE(type) struct my_coro *cb;
3736 #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
3737 That means instead of having a C callback function, you store the
3739 coroutine to switch to in each watcher, and instead of having libev
3740 call your callback, you instead have it switch to that coroutine.
3741 A coroutine might now wait for an event with a function called
3743 "wait_for_event". (the watcher needs to be started, as always, but it
3744 doesn't matter when, or whether the watcher is active or not when this
3745 function is called):
3746 void
3748 wait_for_event (ev_watcher *w)
3749 {
3750 ev_set_cb (w, current_coro);
3751 switch_to (libev_coro);
3752 }
3753 That basically suspends the coroutine inside "wait_for_event" and
3755 continues the libev coroutine, which, when appropriate, switches back
3756 to this or any other coroutine.
3757 You can do similar tricks if you have, say, threads with an event queue
3759 - instead of storing a coroutine, you store the queue object and
3760 instead of switching to a coroutine, you push the watcher onto the
3761 queue and notify any waiters.
3762 To embed libev, see "EMBEDDING", but in short, it's easiest to create
3764 two files, my_ev.h and my_ev.c that include the respective libev files:
3765 // my_ev.h
3767 #define EV_CB_DECLARE(type) struct my_coro *cb;
3768 #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
3769 #include "../libev/ev.h"
3770 // my_ev.c
3772 #define EV_H "my_ev.h"
3773 #include "../libev/ev.c"
3774 And then use my_ev.h when you would normally use ev.h, and compile
3776 my_ev.c into your project. When properly specifying include paths, you
3777 can even use ev.h as header file name directly.
3778
3780 Libev offers a compatibility emulation layer for libevent. It cannot
3781 emulate the internals of libevent, so here are some usage hints:
3782 · Only the libevent-1.4.1-beta API is being emulated.
3784 This was the newest libevent version available when libev was
3786 implemented, and is still mostly unchanged in 2010.
3787 · Use it by including <event.h>, as usual.
3789 · The following members are fully supported: ev_base, ev_callback,
3791 ev_arg, ev_fd, ev_res, ev_events.
3792 · Avoid using ev_flags and the EVLIST_*-macros, while it is
3794 maintained by libev, it does not work exactly the same way as in
3795 libevent (consider it a private API).
3796 · Priorities are not currently supported. Initialising priorities
3798 will fail and all watchers will have the same priority, even though
3799 there is an ev_pri field.
3800 · In libevent, the last base created gets the signals, in libev, the
3802 base that registered the signal gets the signals.
3803 · Other members are not supported.
3805 · The libev emulation is not ABI compatible to libevent, you need to
3807 use the libev header file and library.
3808
3810 C API
3811 The normal C API should work fine when used from C++: both ev.h and the
3812 libev sources can be compiled as C++. Therefore, code that uses the C
3813 API will work fine.
3814 Proper exception specifications might have to be added to callbacks
3816 passed to libev: exceptions may be thrown only from watcher callbacks,
3817 all other callbacks (allocator, syserr, loop acquire/release and
3818 periodic reschedule callbacks) must not throw exceptions, and might
3819 need a "throw ()" specification. If you have code that needs to be
3820 compiled as both C and C++ you can use the "EV_THROW" macro for this:
3821 static void
3823 fatal_error (const char *msg) EV_THROW
3824 {
3825 perror (msg);
3826 abort ();
3827 }
3828 ...
3830 ev_set_syserr_cb (fatal_error);
3831 The only API functions that can currently throw exceptions are
3833 "ev_run", "ev_invoke", "ev_invoke_pending" and "ev_loop_destroy" (the
3834 latter because it runs cleanup watchers).
3835 Throwing exceptions in watcher callbacks is only supported if libev
3837 itself is compiled with a C++ compiler or your C and C++ environments
3838 allow throwing exceptions through C libraries (most do).
3839 C++ API
3841 Libev comes with some simplistic wrapper classes for C++ that mainly
3842 allow you to use some convenience methods to start/stop watchers and
3843 also change the callback model to a model using method callbacks on
3844 objects.
3845 To use it,
3847 #include <ev++.h>
3849 This automatically includes ev.h and puts all of its definitions (many
3851 of them macros) into the global namespace. All C++ specific things are
3852 put into the "ev" namespace. It should support all the same embedding
3853 options as ev.h, most notably "EV_MULTIPLICITY".
3854 Care has been taken to keep the overhead low. The only data member the
3856 C++ classes add (compared to plain C-style watchers) is the event loop
3857 pointer that the watcher is associated with (or no additional members
3858 at all if you disable "EV_MULTIPLICITY" when embedding libev).
3859 Currently, functions, static and non-static member functions and
3861 classes with "operator ()" can be used as callbacks. Other types should
3862 be easy to add as long as they only need one additional pointer for
3863 context. If you need support for other types of functors please contact
3864 the author (preferably after implementing it).
3865 For all this to work, your C++ compiler either has to use the same
3867 calling conventions as your C compiler (for static member functions),
3868 or you have to embed libev and compile libev itself as C++.
3869 Here is a list of things available in the "ev" namespace:
3871 "ev::READ", "ev::WRITE" etc.
3873 These are just enum values with the same values as the "EV_READ"
3874 etc. macros from ev.h.
3875 "ev::tstamp", "ev::now"
3877 Aliases to the same types/functions as with the "ev_" prefix.
3878 "ev::io", "ev::timer", "ev::periodic", "ev::idle", "ev::sig" etc.
3880 For each "ev_TYPE" watcher in ev.h there is a corresponding class
3881 of the same name in the "ev" namespace, with the exception of
3882 "ev_signal" which is called "ev::sig" to avoid clashes with the
3883 "signal" macro defined by many implementations.
3884 All of those classes have these methods:
3886 ev::TYPE::TYPE ()
3888 ev::TYPE::TYPE (loop)
3889 ev::TYPE::~TYPE
3890 The constructor (optionally) takes an event loop to associate
3891 the watcher with. If it is omitted, it will use "EV_DEFAULT".
3892 The constructor calls "ev_init" for you, which means you have
3894 to call the "set" method before starting it.
3895 It will not set a callback, however: You have to call the
3897 templated "set" method to set a callback before you can start
3898 the watcher.
3899 (The reason why you have to use a method is a limitation in C++
3901 which does not allow explicit template arguments for
3902 constructors).
3903 The destructor automatically stops the watcher if it is active.
3905 w->set<class, &class::method> (object *)
3907 This method sets the callback method to call. The method has to
3908 have a signature of "void (*)(ev_TYPE &, int)", it receives the
3909 watcher as first argument and the "revents" as second. The
3910 object must be given as parameter and is stored in the "data"
3911 member of the watcher.
3912 This method synthesizes efficient thunking code to call your
3914 method from the C callback that libev requires. If your
3915 compiler can inline your callback (i.e. it is visible to it at
3916 the place of the "set" call and your compiler is good :), then
3917 the method will be fully inlined into the thunking function,
3918 making it as fast as a direct C callback.
3919 Example: simple class declaration and watcher initialisation
3921 struct myclass
3923 {
3924 void io_cb (ev::io &w, int revents) { }
3925 }
3926 myclass obj;
3928 ev::io iow;
3929 iow.set <myclass, &myclass::io_cb> (&obj);
3930 w->set (object *)
3932 This is a variation of a method callback - leaving out the
3933 method to call will default the method to "operator ()", which
3934 makes it possible to use functor objects without having to
3935 manually specify the "operator ()" all the time. Incidentally,
3936 you can then also leave out the template argument list.
3937 The "operator ()" method prototype must be "void operator
3939 ()(watcher &w, int revents)".
3940 See the method-"set" above for more details.
3942 Example: use a functor object as callback.
3944 struct myfunctor
3946 {
3947 void operator() (ev::io &w, int revents)
3948 {
3949 ...
3950 }
3951 }
3952 myfunctor f;
3954 ev::io w;
3956 w.set (&f);
3957 w->set<function> (void *data = 0)
3959 Also sets a callback, but uses a static method or plain
3960 function as callback. The optional "data" argument will be
3961 stored in the watcher's "data" member and is free for you to
3962 use.
3963 The prototype of the "function" must be "void (*)(ev::TYPE &w,
3965 int)".
3966 See the method-"set" above for more details.
3968 Example: Use a plain function as callback.
3970 static void io_cb (ev::io &w, int revents) { }
3972 iow.set <io_cb> ();
3973 w->set (loop)
3975 Associates a different "struct ev_loop" with this watcher. You
3976 can only do this when the watcher is inactive (and not pending
3977 either).
3978 w->set ([arguments])
3980 Basically the same as "ev_TYPE_set" (except for "ev::embed"
3981 watchers>), with the same arguments. Either this method or a
3982 suitable start method must be called at least once. Unlike the
3983 C counterpart, an active watcher gets automatically stopped and
3984 restarted when reconfiguring it with this method.
3985 For "ev::embed" watchers this method is called "set_embed", to
3987 avoid clashing with the "set (loop)" method.
3988 w->start ()
3990 Starts the watcher. Note that there is no "loop" argument, as
3991 the constructor already stores the event loop.
3992 w->start ([arguments])
3994 Instead of calling "set" and "start" methods separately, it is
3995 often convenient to wrap them in one call. Uses the same type
3996 of arguments as the configure "set" method of the watcher.
3997 w->stop ()
3999 Stops the watcher if it is active. Again, no "loop" argument.
4000 w->again () ("ev::timer", "ev::periodic" only)
4002 For "ev::timer" and "ev::periodic", this invokes the
4003 corresponding "ev_TYPE_again" function.
4004 w->sweep () ("ev::embed" only)
4006 Invokes "ev_embed_sweep".
4007 w->update () ("ev::stat" only)
4009 Invokes "ev_stat_stat".
4010 Example: Define a class with two I/O and idle watchers, start the I/O
4012 watchers in the constructor.
4013 class myclass
4015 {
4016 ev::io io ; void io_cb (ev::io &w, int revents);
4017 ev::io io2 ; void io2_cb (ev::io &w, int revents);
4018 ev::idle idle; void idle_cb (ev::idle &w, int revents);
4019 myclass (int fd)
4021 {
4022 io .set <myclass, &myclass::io_cb > (this);
4023 io2 .set <myclass, &myclass::io2_cb > (this);
4024 idle.set <myclass, &myclass::idle_cb> (this);
4025 io.set (fd, ev::WRITE); // configure the watcher
4027 io.start (); // start it whenever convenient
4028 io2.start (fd, ev::READ); // set + start in one call
4030 }
4031 };
4032
4034 Libev does not offer other language bindings itself, but bindings for a
4035 number of languages exist in the form of third-party packages. If you
4036 know any interesting language binding in addition to the ones listed
4037 here, drop me a note.
4038 Perl
4040 The EV module implements the full libev API and is actually used to
4041 test libev. EV is developed together with libev. Apart from the EV
4042 core module, there are additional modules that implement libev-
4043 compatible interfaces to "libadns" ("EV::ADNS", but "AnyEvent::DNS"
4044 is preferred nowadays), "Net::SNMP" ("Net::SNMP::EV") and the
4045 "libglib" event core ("Glib::EV" and "EV::Glib").
4046 It can be found and installed via CPAN, its homepage is at
4048 <http://software.schmorp.de/pkg/EV>.
4049 Python
4051 Python bindings can be found at <http://code.google.com/p/pyev/>.
4052 It seems to be quite complete and well-documented.
4053 Ruby
4055 Tony Arcieri has written a ruby extension that offers access to a
4056 subset of the libev API and adds file handle abstractions,
4057 asynchronous DNS and more on top of it. It can be found via gem
4058 servers. Its homepage is at <http://rev.rubyforge.org/>.
4059 Roger Pack reports that using the link order "-lws2_32
4061 -lmsvcrt-ruby-190" makes rev work even on mingw.
4062 Haskell
4064 A haskell binding to libev is available at
4065 <http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
4066 D Leandro Lucarella has written a D language binding (ev.d) for
4068 libev, to be found at
4069 <http://www.llucax.com.ar/proj/ev.d/index.html>.
4070 Ocaml
4072 Erkki Seppala has written Ocaml bindings for libev, to be found at
4073 <http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
4074 Lua Brian Maher has written a partial interface to libev for lua (at
4076 the time of this writing, only "ev_io" and "ev_timer"), to be found
4077 at <http://github.com/brimworks/lua-ev>.
4078 Javascript
4080 Node.js (<http://nodejs.org>) uses libev as the underlying event
4081 library.
4082 Others
4084 There are others, and I stopped counting.
4085
4087 Libev can be compiled with a variety of options, the most fundamental
4088 of which is "EV_MULTIPLICITY". This option determines whether (most)
4089 functions and callbacks have an initial "struct ev_loop *" argument.
4090 To make it easier to write programs that cope with either variant, the
4092 following macros are defined:
4093 "EV_A", "EV_A_"
4095 This provides the loop argument for functions, if one is required
4096 ("ev loop argument"). The "EV_A" form is used when this is the sole
4097 argument, "EV_A_" is used when other arguments are following.
4098 Example:
4099 ev_unref (EV_A);
4101 ev_timer_add (EV_A_ watcher);
4102 ev_run (EV_A_ 0);
4103 It assumes the variable "loop" of type "struct ev_loop *" is in
4105 scope, which is often provided by the following macro.
4106 "EV_P", "EV_P_"
4108 This provides the loop parameter for functions, if one is required
4109 ("ev loop parameter"). The "EV_P" form is used when this is the
4110 sole parameter, "EV_P_" is used when other parameters are
4111 following. Example:
4112 // this is how ev_unref is being declared
4114 static void ev_unref (EV_P);
4115 // this is how you can declare your typical callback
4117 static void cb (EV_P_ ev_timer *w, int revents)
4118 It declares a parameter "loop" of type "struct ev_loop *", quite
4120 suitable for use with "EV_A".
4121 "EV_DEFAULT", "EV_DEFAULT_"
4123 Similar to the other two macros, this gives you the value of the
4124 default loop, if multiple loops are supported ("ev loop default").
4125 The default loop will be initialised if it isn't already
4126 initialised.
4127 For non-multiplicity builds, these macros do nothing, so you always
4129 have to initialise the loop somewhere.
4130 "EV_DEFAULT_UC", "EV_DEFAULT_UC_"
4132 Usage identical to "EV_DEFAULT" and "EV_DEFAULT_", but requires
4133 that the default loop has been initialised ("UC" == unchecked).
4134 Their behaviour is undefined when the default loop has not been
4135 initialised by a previous execution of "EV_DEFAULT", "EV_DEFAULT_"
4136 or "ev_default_init (...)".
4137 It is often prudent to use "EV_DEFAULT" when initialising the first
4139 watcher in a function but use "EV_DEFAULT_UC" afterwards.
4140 Example: Declare and initialise a check watcher, utilising the above
4142 macros so it will work regardless of whether multiple loops are
4143 supported or not.
4144 static void
4146 check_cb (EV_P_ ev_timer *w, int revents)
4147 {
4148 ev_check_stop (EV_A_ w);
4149 }
4150 ev_check check;
4152 ev_check_init (&check, check_cb);
4153 ev_check_start (EV_DEFAULT_ &check);
4154 ev_run (EV_DEFAULT_ 0);
4155
4157 Libev can (and often is) directly embedded into host applications.
4158 Examples of applications that embed it include the Deliantra Game
4159 Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe) and
4160 rxvt-unicode.
4161 The goal is to enable you to just copy the necessary files into your
4163 source directory without having to change even a single line in them,
4164 so you can easily upgrade by simply copying (or having a checked-out
4165 copy of libev somewhere in your source tree).
4166 FILESETS
4168 Depending on what features you need you need to include one or more
4169 sets of files in your application.
4170 CORE EVENT LOOP
4172 To include only the libev core (all the "ev_*" functions), with manual
4174 configuration (no autoconf):
4175 #define EV_STANDALONE 1
4177 #include "ev.c"
4178 This will automatically include ev.h, too, and should be done in a
4180 single C source file only to provide the function implementations. To
4181 use it, do the same for ev.h in all files wishing to use this API (best
4182 done by writing a wrapper around ev.h that you can include instead and
4183 where you can put other configuration options):
4184 #define EV_STANDALONE 1
4186 #include "ev.h"
4187 Both header files and implementation files can be compiled with a C++
4189 compiler (at least, that's a stated goal, and breakage will be treated
4190 as a bug).
4191 You need the following files in your source tree, or in a directory in
4193 your include path (e.g. in libev/ when using -Ilibev):
4194 ev.h
4196 ev.c
4197 ev_vars.h
4198 ev_wrap.h
4199 ev_win32.c required on win32 platforms only
4201 ev_select.c only when select backend is enabled (which is enabled by default)
4203 ev_poll.c only when poll backend is enabled (disabled by default)
4204 ev_epoll.c only when the epoll backend is enabled (disabled by default)
4205 ev_kqueue.c only when the kqueue backend is enabled (disabled by default)
4206 ev_port.c only when the solaris port backend is enabled (disabled by default)
4207 ev.c includes the backend files directly when enabled, so you only need
4209 to compile this single file.
4210 LIBEVENT COMPATIBILITY API
4212 To include the libevent compatibility API, also include:
4214 #include "event.c"
4216 in the file including ev.c, and:
4218 #include "event.h"
4220 in the files that want to use the libevent API. This also includes
4222 ev.h.
4223 You need the following additional files for this:
4225 event.h
4227 event.c
4228 AUTOCONF SUPPORT
4230 Instead of using "EV_STANDALONE=1" and providing your configuration in
4232 whatever way you want, you can also "m4_include([libev.m4])" in your
4233 configure.ac and leave "EV_STANDALONE" undefined. ev.c will then
4234 include config.h and configure itself accordingly.
4235 For this of course you need the m4 file:
4237 libev.m4
4239 PREPROCESSOR SYMBOLS/MACROS
4241 Libev can be configured via a variety of preprocessor symbols you have
4242 to define before including (or compiling) any of its files. The default
4243 in the absence of autoconf is documented for every option.
4244 Symbols marked with "(h)" do not change the ABI, and can have different
4246 values when compiling libev vs. including ev.h, so it is permissible to
4247 redefine them before including ev.h without breaking compatibility to a
4248 compiled library. All other symbols change the ABI, which means all
4249 users of libev and the libev code itself must be compiled with
4250 compatible settings.
4251 EV_COMPAT3 (h)
4253 Backwards compatibility is a major concern for libev. This is why
4254 this release of libev comes with wrappers for the functions and
4255 symbols that have been renamed between libev version 3 and 4.
4256 You can disable these wrappers (to test compatibility with future
4258 versions) by defining "EV_COMPAT3" to 0 when compiling your
4259 sources. This has the additional advantage that you can drop the
4260 "struct" from "struct ev_loop" declarations, as libev will provide
4261 an "ev_loop" typedef in that case.
4262 In some future version, the default for "EV_COMPAT3" will become 0,
4264 and in some even more future version the compatibility code will be
4265 removed completely.
4266 EV_STANDALONE (h)
4268 Must always be 1 if you do not use autoconf configuration, which
4269 keeps libev from including config.h, and it also defines dummy
4270 implementations for some libevent functions (such as logging, which
4271 is not supported). It will also not define any of the structs
4272 usually found in event.h that are not directly supported by the
4273 libev core alone.
4274 In standalone mode, libev will still try to automatically deduce
4276 the configuration, but has to be more conservative.
4277 EV_USE_FLOOR
4279 If defined to be 1, libev will use the "floor ()" function for its
4280 periodic reschedule calculations, otherwise libev will fall back on
4281 a portable (slower) implementation. If you enable this, you usually
4282 have to link against libm or something equivalent. Enabling this
4283 when the "floor" function is not available will fail, so the safe
4284 default is to not enable this.
4285 EV_USE_MONOTONIC
4287 If defined to be 1, libev will try to detect the availability of
4288 the monotonic clock option at both compile time and runtime.
4289 Otherwise no use of the monotonic clock option will be attempted.
4290 If you enable this, you usually have to link against librt or
4291 something similar. Enabling it when the functionality isn't
4292 available is safe, though, although you have to make sure you link
4293 against any libraries where the "clock_gettime" function is hiding
4294 in (often -lrt). See also "EV_USE_CLOCK_SYSCALL".
4295 EV_USE_REALTIME
4297 If defined to be 1, libev will try to detect the availability of
4298 the real-time clock option at compile time (and assume its
4299 availability at runtime if successful). Otherwise no use of the
4300 real-time clock option will be attempted. This effectively replaces
4301 "gettimeofday" by "clock_get (CLOCK_REALTIME, ...)" and will not
4302 normally affect correctness. See the note about libraries in the
4303 description of "EV_USE_MONOTONIC", though. Defaults to the opposite
4304 value of "EV_USE_CLOCK_SYSCALL".
4305 EV_USE_CLOCK_SYSCALL
4307 If defined to be 1, libev will try to use a direct syscall instead
4308 of calling the system-provided "clock_gettime" function. This
4309 option exists because on GNU/Linux, "clock_gettime" is in "librt",
4310 but "librt" unconditionally pulls in "libpthread", slowing down
4311 single-threaded programs needlessly. Using a direct syscall is
4312 slightly slower (in theory), because no optimised vdso
4313 implementation can be used, but avoids the pthread dependency.
4314 Defaults to 1 on GNU/Linux with glibc 2.x or higher, as it
4315 simplifies linking (no need for "-lrt").
4316 EV_USE_NANOSLEEP
4318 If defined to be 1, libev will assume that "nanosleep ()" is
4319 available and will use it for delays. Otherwise it will use "select
4320 ()".
4321 EV_USE_EVENTFD
4323 If defined to be 1, then libev will assume that "eventfd ()" is
4324 available and will probe for kernel support at runtime. This will
4325 improve "ev_signal" and "ev_async" performance and reduce resource
4326 consumption. If undefined, it will be enabled if the headers
4327 indicate GNU/Linux + Glibc 2.7 or newer, otherwise disabled.
4328 EV_USE_SELECT
4330 If undefined or defined to be 1, libev will compile in support for
4331 the "select"(2) backend. No attempt at auto-detection will be done:
4332 if no other method takes over, select will be it. Otherwise the
4333 select backend will not be compiled in.
4334 EV_SELECT_USE_FD_SET
4336 If defined to 1, then the select backend will use the system
4337 "fd_set" structure. This is useful if libev doesn't compile due to
4338 a missing "NFDBITS" or "fd_mask" definition or it mis-guesses the
4339 bitset layout on exotic systems. This usually limits the range of
4340 file descriptors to some low limit such as 1024 or might have other
4341 limitations (winsocket only allows 64 sockets). The "FD_SETSIZE"
4342 macro, set before compilation, configures the maximum size of the
4343 "fd_set".
4344 EV_SELECT_IS_WINSOCKET
4346 When defined to 1, the select backend will assume that
4347 select/socket/connect etc. don't understand file descriptors but
4348 wants osf handles on win32 (this is the case when the select to be
4349 used is the winsock select). This means that it will call
4350 "_get_osfhandle" on the fd to convert it to an OS handle.
4351 Otherwise, it is assumed that all these functions actually work on
4352 fds, even on win32. Should not be defined on non-win32 platforms.
4353 EV_FD_TO_WIN32_HANDLE(fd)
4355 If "EV_SELECT_IS_WINSOCKET" is enabled, then libev needs a way to
4356 map file descriptors to socket handles. When not defining this
4357 symbol (the default), then libev will call "_get_osfhandle", which
4358 is usually correct. In some cases, programs use their own file
4359 descriptor management, in which case they can provide this function
4360 to map fds to socket handles.
4361 EV_WIN32_HANDLE_TO_FD(handle)
4363 If "EV_SELECT_IS_WINSOCKET" then libev maps handles to file
4364 descriptors using the standard "_open_osfhandle" function. For
4365 programs implementing their own fd to handle mapping, overwriting
4366 this function makes it easier to do so. This can be done by
4367 defining this macro to an appropriate value.
4368 EV_WIN32_CLOSE_FD(fd)
4370 If programs implement their own fd to handle mapping on win32, then
4371 this macro can be used to override the "close" function, useful to
4372 unregister file descriptors again. Note that the replacement
4373 function has to close the underlying OS handle.
4374 EV_USE_WSASOCKET
4376 If defined to be 1, libev will use "WSASocket" to create its
4377 internal communication socket, which works better in some
4378 environments. Otherwise, the normal "socket" function will be used,
4379 which works better in other environments.
4380 EV_USE_POLL
4382 If defined to be 1, libev will compile in support for the "poll"(2)
4383 backend. Otherwise it will be enabled on non-win32 platforms. It
4384 takes precedence over select.
4385 EV_USE_EPOLL
4387 If defined to be 1, libev will compile in support for the Linux
4388 "epoll"(7) backend. Its availability will be detected at runtime,
4389 otherwise another method will be used as fallback. This is the
4390 preferred backend for GNU/Linux systems. If undefined, it will be
4391 enabled if the headers indicate GNU/Linux + Glibc 2.4 or newer,
4392 otherwise disabled.
4393 EV_USE_KQUEUE
4395 If defined to be 1, libev will compile in support for the BSD style
4396 "kqueue"(2) backend. Its actual availability will be detected at
4397 runtime, otherwise another method will be used as fallback. This is
4398 the preferred backend for BSD and BSD-like systems, although on
4399 most BSDs kqueue only supports some types of fds correctly (the
4400 only platform we found that supports ptys for example was NetBSD),
4401 so kqueue might be compiled in, but not be used unless explicitly
4402 requested. The best way to use it is to find out whether kqueue
4403 supports your type of fd properly and use an embedded kqueue loop.
4404 EV_USE_PORT
4406 If defined to be 1, libev will compile in support for the Solaris
4407 10 port style backend. Its availability will be detected at
4408 runtime, otherwise another method will be used as fallback. This is
4409 the preferred backend for Solaris 10 systems.
4410 EV_USE_DEVPOLL
4412 Reserved for future expansion, works like the USE symbols above.
4413 EV_USE_INOTIFY
4415 If defined to be 1, libev will compile in support for the Linux
4416 inotify interface to speed up "ev_stat" watchers. Its actual
4417 availability will be detected at runtime. If undefined, it will be
4418 enabled if the headers indicate GNU/Linux + Glibc 2.4 or newer,
4419 otherwise disabled.
4420 EV_NO_SMP
4422 If defined to be 1, libev will assume that memory is always
4423 coherent between threads, that is, threads can be used, but threads
4424 never run on different cpus (or different cpu cores). This reduces
4425 dependencies and makes libev faster.
4426 EV_NO_THREADS
4428 If defined to be 1, libev will assume that it will never be called
4429 from different threads (that includes signal handlers), which is a
4430 stronger assumption than "EV_NO_SMP", above. This reduces
4431 dependencies and makes libev faster.
4432 EV_ATOMIC_T
4434 Libev requires an integer type (suitable for storing 0 or 1) whose
4435 access is atomic with respect to other threads or signal contexts.
4436 No such type is easily found in the C language, so you can provide
4437 your own type that you know is safe for your purposes. It is used
4438 both for signal handler "locking" as well as for signal and thread
4439 safety in "ev_async" watchers.
4440 In the absence of this define, libev will use "sig_atomic_t
4442 volatile" (from signal.h), which is usually good enough on most
4443 platforms.
4444 EV_H (h)
4446 The name of the ev.h header file used to include it. The default if
4447 undefined is "ev.h" in event.h, ev.c and ev++.h. This can be used
4448 to virtually rename the ev.h header file in case of conflicts.
4449 EV_CONFIG_H (h)
4451 If "EV_STANDALONE" isn't 1, this variable can be used to override
4452 ev.c's idea of where to find the config.h file, similarly to
4453 "EV_H", above.
4454 EV_EVENT_H (h)
4456 Similarly to "EV_H", this macro can be used to override event.c's
4457 idea of how the event.h header can be found, the default is
4458 "event.h".
4459 EV_PROTOTYPES (h)
4461 If defined to be 0, then ev.h will not define any function
4462 prototypes, but still define all the structs and other symbols.
4463 This is occasionally useful if you want to provide your own wrapper
4464 functions around libev functions.
4465 EV_MULTIPLICITY
4467 If undefined or defined to 1, then all event-loop-specific
4468 functions will have the "struct ev_loop *" as first argument, and
4469 you can create additional independent event loops. Otherwise there
4470 will be no support for multiple event loops and there is no first
4471 event loop pointer argument. Instead, all functions act on the
4472 single default loop.
4473 Note that "EV_DEFAULT" and "EV_DEFAULT_" will no longer provide a
4475 default loop when multiplicity is switched off - you always have to
4476 initialise the loop manually in this case.
4477 EV_MINPRI
4479 EV_MAXPRI
4480 The range of allowed priorities. "EV_MINPRI" must be smaller or
4481 equal to "EV_MAXPRI", but otherwise there are no non-obvious
4482 limitations. You can provide for more priorities by overriding
4483 those symbols (usually defined to be "-2" and 2, respectively).
4484 When doing priority-based operations, libev usually has to linearly
4486 search all the priorities, so having many of them (hundreds) uses a
4487 lot of space and time, so using the defaults of five priorities (-2
4488 .. +2) is usually fine.
4489 If your embedding application does not need any priorities,
4491 defining these both to 0 will save some memory and CPU.
4492 EV_PERIODIC_ENABLE, EV_IDLE_ENABLE, EV_EMBED_ENABLE, EV_STAT_ENABLE,
4494 EV_PREPARE_ENABLE, EV_CHECK_ENABLE, EV_FORK_ENABLE, EV_SIGNAL_ENABLE,
4495 EV_ASYNC_ENABLE, EV_CHILD_ENABLE.
4496 If undefined or defined to be 1 (and the platform supports it),
4497 then the respective watcher type is supported. If defined to be 0,
4498 then it is not. Disabling watcher types mainly saves code size.
4499 EV_FEATURES
4501 If you need to shave off some kilobytes of code at the expense of
4502 some speed (but with the full API), you can define this symbol to
4503 request certain subsets of functionality. The default is to enable
4504 all features that can be enabled on the platform.
4505 A typical way to use this symbol is to define it to 0 (or to a
4507 bitset with some broad features you want) and then selectively re-
4508 enable additional parts you want, for example if you want
4509 everything minimal, but multiple event loop support, async and
4510 child watchers and the poll backend, use this:
4511 #define EV_FEATURES 0
4513 #define EV_MULTIPLICITY 1
4514 #define EV_USE_POLL 1
4515 #define EV_CHILD_ENABLE 1
4516 #define EV_ASYNC_ENABLE 1
4517 The actual value is a bitset, it can be a combination of the
4519 following values (by default, all of these are enabled):
4520 1 - faster/larger code
4522 Use larger code to speed up some operations.
4523 Currently this is used to override some inlining decisions
4525 (enlarging the code size by roughly 30% on amd64).
4526 When optimising for size, use of compiler flags such as "-Os"
4528 with gcc is recommended, as well as "-DNDEBUG", as libev
4529 contains a number of assertions.
4530 The default is off when "__OPTIMIZE_SIZE__" is defined by your
4532 compiler (e.g. gcc with "-Os").
4533 2 - faster/larger data structures
4535 Replaces the small 2-heap for timer management by a faster
4536 4-heap, larger hash table sizes and so on. This will usually
4537 further increase code size and can additionally have an effect
4538 on the size of data structures at runtime.
4539 The default is off when "__OPTIMIZE_SIZE__" is defined by your
4541 compiler (e.g. gcc with "-Os").
4542 4 - full API configuration
4544 This enables priorities (sets "EV_MAXPRI"=2 and
4545 "EV_MINPRI"=-2), and enables multiplicity
4546 ("EV_MULTIPLICITY"=1).
4547 8 - full API
4549 This enables a lot of the "lesser used" API functions. See
4550 "ev.h" for details on which parts of the API are still
4551 available without this feature, and do not complain if this
4552 subset changes over time.
4553 16 - enable all optional watcher types
4555 Enables all optional watcher types. If you want to selectively
4556 enable only some watcher types other than I/O and timers (e.g.
4557 prepare, embed, async, child...) you can enable them manually
4558 by defining "EV_watchertype_ENABLE" to 1 instead.
4559 32 - enable all backends
4561 This enables all backends - without this feature, you need to
4562 enable at least one backend manually ("EV_USE_SELECT" is a good
4563 choice).
4564 64 - enable OS-specific "helper" APIs
4566 Enable inotify, eventfd, signalfd and similar OS-specific
4567 helper APIs by default.
4568 Compiling with "gcc -Os -DEV_STANDALONE -DEV_USE_EPOLL=1
4570 -DEV_FEATURES=0" reduces the compiled size of libev from 24.7Kb
4571 code/2.8Kb data to 6.5Kb code/0.3Kb data on my GNU/Linux amd64
4572 system, while still giving you I/O watchers, timers and monotonic
4573 clock support.
4574 With an intelligent-enough linker (gcc+binutils are intelligent
4576 enough when you use "-Wl,--gc-sections -ffunction-sections")
4577 functions unused by your program might be left out as well - a
4578 binary starting a timer and an I/O watcher then might come out at
4579 only 5Kb.
4580 EV_API_STATIC
4582 If this symbol is defined (by default it is not), then all
4583 identifiers will have static linkage. This means that libev will
4584 not export any identifiers, and you cannot link against libev
4585 anymore. This can be useful when you embed libev, only want to use
4586 libev functions in a single file, and do not want its identifiers
4587 to be visible.
4588 To use this, define "EV_API_STATIC" and include ev.c in the file
4590 that wants to use libev.
4591 This option only works when libev is compiled with a C compiler, as
4593 C++ doesn't support the required declaration syntax.
4594 EV_AVOID_STDIO
4596 If this is set to 1 at compiletime, then libev will avoid using
4597 stdio functions (printf, scanf, perror etc.). This will increase
4598 the code size somewhat, but if your program doesn't otherwise
4599 depend on stdio and your libc allows it, this avoids linking in the
4600 stdio library which is quite big.
4601 Note that error messages might become less precise when this option
4603 is enabled.
4604 EV_NSIG
4606 The highest supported signal number, +1 (or, the number of
4607 signals): Normally, libev tries to deduce the maximum number of
4608 signals automatically, but sometimes this fails, in which case it
4609 can be specified. Also, using a lower number than detected (32
4610 should be good for about any system in existence) can save some
4611 memory, as libev statically allocates some 12-24 bytes per signal
4612 number.
4613 EV_PID_HASHSIZE
4615 "ev_child" watchers use a small hash table to distribute workload
4616 by pid. The default size is 16 (or 1 with "EV_FEATURES" disabled),
4617 usually more than enough. If you need to manage thousands of
4618 children you might want to increase this value (must be a power of
4619 two).
4620 EV_INOTIFY_HASHSIZE
4622 "ev_stat" watchers use a small hash table to distribute workload by
4623 inotify watch id. The default size is 16 (or 1 with "EV_FEATURES"
4624 disabled), usually more than enough. If you need to manage
4625 thousands of "ev_stat" watchers you might want to increase this
4626 value (must be a power of two).
4627 EV_USE_4HEAP
4629 Heaps are not very cache-efficient. To improve the cache-efficiency
4630 of the timer and periodics heaps, libev uses a 4-heap when this
4631 symbol is defined to 1. The 4-heap uses more complicated (longer)
4632 code but has noticeably faster performance with many (thousands) of
4633 watchers.
4634 The default is 1, unless "EV_FEATURES" overrides it, in which case
4636 it will be 0.
4637 EV_HEAP_CACHE_AT
4639 Heaps are not very cache-efficient. To improve the cache-efficiency
4640 of the timer and periodics heaps, libev can cache the timestamp
4641 (at) within the heap structure (selected by defining
4642 "EV_HEAP_CACHE_AT" to 1), which uses 8-12 bytes more per watcher
4643 and a few hundred bytes more code, but avoids random read accesses
4644 on heap changes. This improves performance noticeably with many
4645 (hundreds) of watchers.
4646 The default is 1, unless "EV_FEATURES" overrides it, in which case
4648 it will be 0.
4649 EV_VERIFY
4651 Controls how much internal verification (see "ev_verify ()") will
4652 be done: If set to 0, no internal verification code will be
4653 compiled in. If set to 1, then verification code will be compiled
4654 in, but not called. If set to 2, then the internal verification
4655 code will be called once per loop, which can slow down libev. If
4656 set to 3, then the verification code will be called very
4657 frequently, which will slow down libev considerably.
4658 The default is 1, unless "EV_FEATURES" overrides it, in which case
4660 it will be 0.
4661 EV_COMMON
4663 By default, all watchers have a "void *data" member. By redefining
4664 this macro to something else you can include more and other types
4665 of members. You have to define it each time you include one of the
4666 files, though, and it must be identical each time.
4667 For example, the perl EV module uses something like this:
4669 #define EV_COMMON \
4671 SV *self; /* contains this struct */ \
4672 SV *cb_sv, *fh /* note no trailing ";" */
4673 EV_CB_DECLARE (type)
4675 EV_CB_INVOKE (watcher, revents)
4676 ev_set_cb (ev, cb)
4677 Can be used to change the callback member declaration in each
4678 watcher, and the way callbacks are invoked and set. Must expand to
4679 a struct member definition and a statement, respectively. See the
4680 ev.h header file for their default definitions. One possible use
4681 for overriding these is to avoid the "struct ev_loop *" as first
4682 argument in all cases, or to use method calls instead of plain
4683 function calls in C++.
4684 EXPORTED API SYMBOLS
4686 If you need to re-export the API (e.g. via a DLL) and you need a list
4687 of exported symbols, you can use the provided Symbol.* files which list
4688 all public symbols, one per line:
4689 Symbols.ev for libev proper
4691 Symbols.event for the libevent emulation
4692 This can also be used to rename all public symbols to avoid clashes
4694 with multiple versions of libev linked together (which is obviously bad
4695 in itself, but sometimes it is inconvenient to avoid this).
4696 A sed command like this will create wrapper "#define"'s that you need
4698 to include before including ev.h:
4699 <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
4701 This would create a file wrap.h which essentially looks like this:
4703 #define ev_backend myprefix_ev_backend
4705 #define ev_check_start myprefix_ev_check_start
4706 #define ev_check_stop myprefix_ev_check_stop
4707 ...
4708 EXAMPLES
4710 For a real-world example of a program the includes libev verbatim, you
4711 can have a look at the EV perl module
4712 (<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
4713 the libev/ subdirectory and includes them in the EV/EVAPI.h (public
4714 interface) and EV.xs (implementation) files. Only the EV.xs file will
4715 be compiled. It is pretty complex because it provides its own header
4716 file.
4717 The usage in rxvt-unicode is simpler. It has a ev_cpp.h header file
4719 that everybody includes and which overrides some configure choices:
4720 #define EV_FEATURES 8
4722 #define EV_USE_SELECT 1
4723 #define EV_PREPARE_ENABLE 1
4724 #define EV_IDLE_ENABLE 1
4725 #define EV_SIGNAL_ENABLE 1
4726 #define EV_CHILD_ENABLE 1
4727 #define EV_USE_STDEXCEPT 0
4728 #define EV_CONFIG_H <config.h>
4729 #include "ev++.h"
4731 And a ev_cpp.C implementation file that contains libev proper and is
4733 compiled:
4734 #include "ev_cpp.h"
4736 #include "ev.c"
4737
4739 THREADS AND COROUTINES
4740 THREADS
4741 All libev functions are reentrant and thread-safe unless explicitly
4743 documented otherwise, but libev implements no locking itself. This
4744 means that you can use as many loops as you want in parallel, as long
4745 as there are no concurrent calls into any libev function with the same
4746 loop parameter ("ev_default_*" calls have an implicit default loop
4747 parameter, of course): libev guarantees that different event loops
4748 share no data structures that need any locking.
4749 Or to put it differently: calls with different loop parameters can be
4751 done concurrently from multiple threads, calls with the same loop
4752 parameter must be done serially (but can be done from different
4753 threads, as long as only one thread ever is inside a call at any point
4754 in time, e.g. by using a mutex per loop).
4755 Specifically to support threads (and signal handlers), libev implements
4757 so-called "ev_async" watchers, which allow some limited form of
4758 concurrency on the same event loop, namely waking it up "from the
4759 outside".
4760 If you want to know which design (one loop, locking, or multiple loops
4762 without or something else still) is best for your problem, then I
4763 cannot help you, but here is some generic advice:
4764 · most applications have a main thread: use the default libev loop in
4766 that thread, or create a separate thread running only the default
4767 loop.
4768 This helps integrating other libraries or software modules that use
4770 libev themselves and don't care/know about threading.
4771 · one loop per thread is usually a good model.
4773 Doing this is almost never wrong, sometimes a better-performance
4775 model exists, but it is always a good start.
4776 · other models exist, such as the leader/follower pattern, where one
4778 loop is handed through multiple threads in a kind of round-robin
4779 fashion.
4780 Choosing a model is hard - look around, learn, know that usually
4782 you can do better than you currently do :-)
4783 · often you need to talk to some other thread which blocks in the
4785 event loop.
4786 "ev_async" watchers can be used to wake them up from other threads
4788 safely (or from signal contexts...).
4789 An example use would be to communicate signals or other events that
4791 only work in the default loop by registering the signal watcher
4792 with the default loop and triggering an "ev_async" watcher from the
4793 default loop watcher callback into the event loop interested in the
4794 signal.
4795 See also "THREAD LOCKING EXAMPLE".
4797 COROUTINES
4799 Libev is very accommodating to coroutines ("cooperative threads"):
4801 libev fully supports nesting calls to its functions from different
4802 coroutines (e.g. you can call "ev_run" on the same loop from two
4803 different coroutines, and switch freely between both coroutines running
4804 the loop, as long as you don't confuse yourself). The only exception is
4805 that you must not do this from "ev_periodic" reschedule callbacks.
4806 Care has been taken to ensure that libev does not keep local state
4808 inside "ev_run", and other calls do not usually allow for coroutine
4809 switches as they do not call any callbacks.
4810 COMPILER WARNINGS
4812 Depending on your compiler and compiler settings, you might get no or a
4813 lot of warnings when compiling libev code. Some people are apparently
4814 scared by this.
4815 However, these are unavoidable for many reasons. For one, each compiler
4817 has different warnings, and each user has different tastes regarding
4818 warning options. "Warn-free" code therefore cannot be a goal except
4819 when targeting a specific compiler and compiler-version.
4820 Another reason is that some compiler warnings require elaborate
4822 workarounds, or other changes to the code that make it less clear and
4823 less maintainable.
4824 And of course, some compiler warnings are just plain stupid, or simply
4826 wrong (because they don't actually warn about the condition their
4827 message seems to warn about). For example, certain older gcc versions
4828 had some warnings that resulted in an extreme number of false
4829 positives. These have been fixed, but some people still insist on
4830 making code warn-free with such buggy versions.
4831 While libev is written to generate as few warnings as possible, "warn-
4833 free" code is not a goal, and it is recommended not to build libev with
4834 any compiler warnings enabled unless you are prepared to cope with them
4835 (e.g. by ignoring them). Remember that warnings are just that:
4836 warnings, not errors, or proof of bugs.
4837 VALGRIND
4839 Valgrind has a special section here because it is a popular tool that
4840 is highly useful. Unfortunately, valgrind reports are very hard to
4841 interpret.
4842 If you think you found a bug (memory leak, uninitialised data access
4844 etc.) in libev, then check twice: If valgrind reports something like:
4845 ==2274== definitely lost: 0 bytes in 0 blocks.
4847 ==2274== possibly lost: 0 bytes in 0 blocks.
4848 ==2274== still reachable: 256 bytes in 1 blocks.
4849 Then there is no memory leak, just as memory accounted to global
4851 variables is not a memleak - the memory is still being referenced, and
4852 didn't leak.
4853 Similarly, under some circumstances, valgrind might report kernel bugs
4855 as if it were a bug in libev (e.g. in realloc or in the poll backend,
4856 although an acceptable workaround has been found here), or it might be
4857 confused.
4858 Keep in mind that valgrind is a very good tool, but only a tool. Don't
4860 make it into some kind of religion.
4861 If you are unsure about something, feel free to contact the mailing
4863 list with the full valgrind report and an explanation on why you think
4864 this is a bug in libev (best check the archives, too :). However, don't
4865 be annoyed when you get a brisk "this is no bug" answer and take the
4866 chance of learning how to interpret valgrind properly.
4867 If you need, for some reason, empty reports from valgrind for your
4869 project I suggest using suppression lists.
4870
4872 GNU/LINUX 32 BIT LIMITATIONS
4873 GNU/Linux is the only common platform that supports 64 bit file/large
4874 file interfaces but disables them by default.
4875 That means that libev compiled in the default environment doesn't
4877 support files larger than 2GiB or so, which mainly affects "ev_stat"
4878 watchers.
4879 Unfortunately, many programs try to work around this GNU/Linux issue by
4881 enabling the large file API, which makes them incompatible with the
4882 standard libev compiled for their system.
4883 Likewise, libev cannot enable the large file API itself as this would
4885 suddenly make it incompatible to the default compile time environment,
4886 i.e. all programs not using special compile switches.
4887 OS/X AND DARWIN BUGS
4889 The whole thing is a bug if you ask me - basically any system interface
4890 you touch is broken, whether it is locales, poll, kqueue or even the
4891 OpenGL drivers.
4892 "kqueue" is buggy
4894 The kqueue syscall is broken in all known versions - most versions
4896 support only sockets, many support pipes.
4897 Libev tries to work around this by not using "kqueue" by default on
4899 this rotten platform, but of course you can still ask for it when
4900 creating a loop - embedding a socket-only kqueue loop into a select-
4901 based one is probably going to work well.
4902 "poll" is buggy
4904 Instead of fixing "kqueue", Apple replaced their (working) "poll"
4906 implementation by something calling "kqueue" internally around the
4907 10.5.6 release, so now "kqueue" and "poll" are broken.
4908 Libev tries to work around this by not using "poll" by default on this
4910 rotten platform, but of course you can still ask for it when creating a
4911 loop.
4912 "select" is buggy
4914 All that's left is "select", and of course Apple found a way to fuck
4916 this one up as well: On OS/X, "select" actively limits the number of
4917 file descriptors you can pass in to 1024 - your program suddenly
4918 crashes when you use more.
4919 There is an undocumented "workaround" for this - defining
4921 "_DARWIN_UNLIMITED_SELECT", which libev tries to use, so select should
4922 work on OS/X.
4923 SOLARIS PROBLEMS AND WORKAROUNDS
4925 "errno" reentrancy
4926 The default compile environment on Solaris is unfortunately so thread-
4928 unsafe that you can't even use components/libraries compiled without
4929 "-D_REENTRANT" in a threaded program, which, of course, isn't defined
4930 by default. A valid, if stupid, implementation choice.
4931 If you want to use libev in threaded environments you have to make sure
4933 it's compiled with "_REENTRANT" defined.
4934 Event port backend
4936 The scalable event interface for Solaris is called "event ports".
4938 Unfortunately, this mechanism is very buggy in all major releases. If
4939 you run into high CPU usage, your program freezes or you get a large
4940 number of spurious wakeups, make sure you have all the relevant and
4941 latest kernel patches applied. No, I don't know which ones, but there
4942 are multiple ones to apply, and afterwards, event ports actually work
4943 great.
4944 If you can't get it to work, you can try running the program by setting
4946 the environment variable "LIBEV_FLAGS=3" to only allow "poll" and
4947 "select" backends.
4948 AIX POLL BUG
4950 AIX unfortunately has a broken "poll.h" header. Libev works around this
4951 by trying to avoid the poll backend altogether (i.e. it's not even
4952 compiled in), which normally isn't a big problem as "select" works fine
4953 with large bitsets on AIX, and AIX is dead anyway.
4954 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
4956 General issues
4957 Win32 doesn't support any of the standards (e.g. POSIX) that libev
4959 requires, and its I/O model is fundamentally incompatible with the
4960 POSIX model. Libev still offers limited functionality on this platform
4961 in the form of the "EVBACKEND_SELECT" backend, and only supports socket
4962 descriptors. This only applies when using Win32 natively, not when
4963 using e.g. cygwin. Actually, it only applies to the microsofts own
4964 compilers, as every compiler comes with a slightly differently
4965 broken/incompatible environment.
4966 Lifting these limitations would basically require the full re-
4968 implementation of the I/O system. If you are into this kind of thing,
4969 then note that glib does exactly that for you in a very portable way
4970 (note also that glib is the slowest event library known to man).
4971 There is no supported compilation method available on windows except
4973 embedding it into other applications.
4974 Sensible signal handling is officially unsupported by Microsoft - libev
4976 tries its best, but under most conditions, signals will simply not
4977 work.
4978 Not a libev limitation but worth mentioning: windows apparently doesn't
4980 accept large writes: instead of resulting in a partial write, windows
4981 will either accept everything or return "ENOBUFS" if the buffer is too
4982 large, so make sure you only write small amounts into your sockets
4983 (less than a megabyte seems safe, but this apparently depends on the
4984 amount of memory available).
4985 Due to the many, low, and arbitrary limits on the win32 platform and
4987 the abysmal performance of winsockets, using a large number of sockets
4988 is not recommended (and not reasonable). If your program needs to use
4989 more than a hundred or so sockets, then likely it needs to use a
4990 totally different implementation for windows, as libev offers the POSIX
4991 readiness notification model, which cannot be implemented efficiently
4992 on windows (due to Microsoft monopoly games).
4993 A typical way to use libev under windows is to embed it (see the
4995 embedding section for details) and use the following evwrap.h header
4996 file instead of ev.h:
4997 #define EV_STANDALONE /* keeps ev from requiring config.h */
4999 #define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
5000 #include "ev.h"
5002 And compile the following evwrap.c file into your project (make sure
5004 you do not compile the ev.c or any other embedded source files!):
5005 #include "evwrap.h"
5007 #include "ev.c"
5008 The winsocket "select" function
5010 The winsocket "select" function doesn't follow POSIX in that it
5012 requires socket handles and not socket file descriptors (it is also
5013 extremely buggy). This makes select very inefficient, and also requires
5014 a mapping from file descriptors to socket handles (the Microsoft C
5015 runtime provides the function "_open_osfhandle" for this). See the
5016 discussion of the "EV_SELECT_USE_FD_SET", "EV_SELECT_IS_WINSOCKET" and
5017 "EV_FD_TO_WIN32_HANDLE" preprocessor symbols for more info.
5018 The configuration for a "naked" win32 using the Microsoft runtime
5020 libraries and raw winsocket select is:
5021 #define EV_USE_SELECT 1
5023 #define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
5024 Note that winsockets handling of fd sets is O(n), so you can easily get
5026 a complexity in the O(nX) range when using win32.
5027 Limited number of file descriptors
5029 Windows has numerous arbitrary (and low) limits on things.
5031 Early versions of winsocket's select only supported waiting for a
5033 maximum of 64 handles (probably owning to the fact that all windows
5034 kernels can only wait for 64 things at the same time internally;
5035 Microsoft recommends spawning a chain of threads and wait for 63
5036 handles and the previous thread in each. Sounds great!).
5037 Newer versions support more handles, but you need to define
5039 "FD_SETSIZE" to some high number (e.g. 2048) before compiling the
5040 winsocket select call (which might be in libev or elsewhere, for
5041 example, perl and many other interpreters do their own select emulation
5042 on windows).
5043 Another limit is the number of file descriptors in the Microsoft
5045 runtime libraries, which by default is 64 (there must be a hidden 64
5046 fetish or something like this inside Microsoft). You can increase this
5047 by calling "_setmaxstdio", which can increase this limit to 2048
5048 (another arbitrary limit), but is broken in many versions of the
5049 Microsoft runtime libraries. This might get you to about 512 or 2048
5050 sockets (depending on windows version and/or the phase of the moon). To
5051 get more, you need to wrap all I/O functions and provide your own fd
5052 management, but the cost of calling select (O(nX)) will likely make
5053 this unworkable.
5054 PORTABILITY REQUIREMENTS
5056 In addition to a working ISO-C implementation and of course the
5057 backend-specific APIs, libev relies on a few additional extensions:
5058 "void (*)(ev_watcher_type *, int revents)" must have compatible calling
5060 conventions regardless of "ev_watcher_type *".
5061 Libev assumes not only that all watcher pointers have the same
5062 internal structure (guaranteed by POSIX but not by ISO C for
5063 example), but it also assumes that the same (machine) code can be
5064 used to call any watcher callback: The watcher callbacks have
5065 different type signatures, but libev calls them using an
5066 "ev_watcher *" internally.
5067 null pointers and integer zero are represented by 0 bytes
5069 Libev uses "memset" to initialise structs and arrays to 0 bytes,
5070 and relies on this setting pointers and integers to null.
5071 pointer accesses must be thread-atomic
5073 Accessing a pointer value must be atomic, it must both be readable
5074 and writable in one piece - this is the case on all current
5075 architectures.
5076 "sig_atomic_t volatile" must be thread-atomic as well
5078 The type "sig_atomic_t volatile" (or whatever is defined as
5079 "EV_ATOMIC_T") must be atomic with respect to accesses from
5080 different threads. This is not part of the specification for
5081 "sig_atomic_t", but is believed to be sufficiently portable.
5082 "sigprocmask" must work in a threaded environment
5084 Libev uses "sigprocmask" to temporarily block signals. This is not
5085 allowed in a threaded program ("pthread_sigmask" has to be used).
5086 Typical pthread implementations will either allow "sigprocmask" in
5087 the "main thread" or will block signals process-wide, both
5088 behaviours would be compatible with libev. Interaction between
5089 "sigprocmask" and "pthread_sigmask" could complicate things,
5090 however.
5091 The most portable way to handle signals is to block signals in all
5093 threads except the initial one, and run the signal handling loop in
5094 the initial thread as well.
5095 "long" must be large enough for common memory allocation sizes
5097 To improve portability and simplify its API, libev uses "long"
5098 internally instead of "size_t" when allocating its data structures.
5099 On non-POSIX systems (Microsoft...) this might be unexpectedly low,
5100 but is still at least 31 bits everywhere, which is enough for
5101 hundreds of millions of watchers.
5102 "double" must hold a time value in seconds with enough accuracy
5104 The type "double" is used to represent timestamps. It is required
5105 to have at least 51 bits of mantissa (and 9 bits of exponent),
5106 which is good enough for at least into the year 4000 with
5107 millisecond accuracy (the design goal for libev). This requirement
5108 is overfulfilled by implementations using IEEE 754, which is
5109 basically all existing ones.
5110 With IEEE 754 doubles, you get microsecond accuracy until at least
5112 the year 2255 (and millisecond accuracy till the year 287396 - by
5113 then, libev is either obsolete or somebody patched it to use "long
5114 double" or something like that, just kidding).
5115 If you know of other additional requirements drop me a note.
5117
5119 In this section the complexities of (many of) the algorithms used
5120 inside libev will be documented. For complexity discussions about
5121 backends see the documentation for "ev_default_init".
5122 All of the following are about amortised time: If an array needs to be
5124 extended, libev needs to realloc and move the whole array, but this
5125 happens asymptotically rarer with higher number of elements, so O(1)
5126 might mean that libev does a lengthy realloc operation in rare cases,
5127 but on average it is much faster and asymptotically approaches constant
5128 time.
5129 Starting and stopping timer/periodic watchers: O(log
5131 skipped_other_timers)
5132 This means that, when you have a watcher that triggers in one hour
5133 and there are 100 watchers that would trigger before that, then
5134 inserting will have to skip roughly seven ("ld 100") of these
5135 watchers.
5136 Changing timer/periodic watchers (by autorepeat or calling again):
5138 O(log skipped_other_timers)
5139 That means that changing a timer costs less than removing/adding
5140 them, as only the relative motion in the event queue has to be paid
5141 for.
5142 Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
5144 These just add the watcher into an array or at the head of a list.
5145 Stopping check/prepare/idle/fork/async watchers: O(1)
5147 Stopping an io/signal/child watcher:
5148 O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
5149 These watchers are stored in lists, so they need to be walked to
5150 find the correct watcher to remove. The lists are usually short
5151 (you don't usually have many watchers waiting for the same fd or
5152 signal: one is typical, two is rare).
5153 Finding the next timer in each loop iteration: O(1)
5155 By virtue of using a binary or 4-heap, the next timer is always
5156 found at a fixed position in the storage array.
5157 Each change on a file descriptor per loop iteration:
5159 O(number_of_watchers_for_this_fd)
5160 A change means an I/O watcher gets started or stopped, which
5161 requires libev to recalculate its status (and possibly tell the
5162 kernel, depending on backend and whether "ev_io_set" was used).
5163 Activating one watcher (putting it into the pending state): O(1)
5165 Priority handling: O(number_of_priorities)
5166 Priorities are implemented by allocating some space for each
5167 priority. When doing priority-based operations, libev usually has
5168 to linearly search all the priorities, but starting/stopping and
5169 activating watchers becomes O(1) with respect to priority handling.
5170 Sending an ev_async: O(1)
5172 Processing ev_async_send: O(number_of_async_watchers)
5173 Processing signals: O(max_signal_number)
5174 Sending involves a system call iff there were no other
5175 "ev_async_send" calls in the current loop iteration and the loop is
5176 currently blocked. Checking for async and signal events involves
5177 iterating over all running async watchers or all signal numbers.
5178
5180 The major version 4 introduced some incompatible changes to the API.
5181 At the moment, the "ev.h" header file provides compatibility
5183 definitions for all changes, so most programs should still compile. The
5184 compatibility layer might be removed in later versions of libev, so
5185 better update to the new API early than late.
5186 "EV_COMPAT3" backwards compatibility mechanism
5188 The backward compatibility mechanism can be controlled by
5189 "EV_COMPAT3". See "PREPROCESSOR SYMBOLS/MACROS" in the "EMBEDDING"
5190 section.
5191 "ev_default_destroy" and "ev_default_fork" have been removed
5193 These calls can be replaced easily by their "ev_loop_xxx"
5194 counterparts:
5195 ev_loop_destroy (EV_DEFAULT_UC);
5197 ev_loop_fork (EV_DEFAULT);
5198 function/symbol renames
5200 A number of functions and symbols have been renamed:
5201 ev_loop => ev_run
5203 EVLOOP_NONBLOCK => EVRUN_NOWAIT
5204 EVLOOP_ONESHOT => EVRUN_ONCE
5205 ev_unloop => ev_break
5207 EVUNLOOP_CANCEL => EVBREAK_CANCEL
5208 EVUNLOOP_ONE => EVBREAK_ONE
5209 EVUNLOOP_ALL => EVBREAK_ALL
5210 EV_TIMEOUT => EV_TIMER
5212 ev_loop_count => ev_iteration
5214 ev_loop_depth => ev_depth
5215 ev_loop_verify => ev_verify
5216 Most functions working on "struct ev_loop" objects don't have an
5218 "ev_loop_" prefix, so it was removed; "ev_loop", "ev_unloop" and
5219 associated constants have been renamed to not collide with the
5220 "struct ev_loop" anymore and "EV_TIMER" now follows the same naming
5221 scheme as all other watcher types. Note that "ev_loop_fork" is
5222 still called "ev_loop_fork" because it would otherwise clash with
5223 the "ev_fork" typedef.
5224 "EV_MINIMAL" mechanism replaced by "EV_FEATURES"
5226 The preprocessor symbol "EV_MINIMAL" has been replaced by a
5227 different mechanism, "EV_FEATURES". Programs using "EV_MINIMAL"
5228 usually compile and work, but the library code will of course be
5229 larger.
5230
5232 active
5233 A watcher is active as long as it has been started and not yet
5234 stopped. See "WATCHER STATES" for details.
5235 application
5237 In this document, an application is whatever is using libev.
5238 backend
5240 The part of the code dealing with the operating system interfaces.
5241 callback
5243 The address of a function that is called when some event has been
5244 detected. Callbacks are being passed the event loop, the watcher
5245 that received the event, and the actual event bitset.
5246 callback/watcher invocation
5248 The act of calling the callback associated with a watcher.
5249 event
5251 A change of state of some external event, such as data now being
5252 available for reading on a file descriptor, time having passed or
5253 simply not having any other events happening anymore.
5254 In libev, events are represented as single bits (such as "EV_READ"
5256 or "EV_TIMER").
5257 event library
5259 A software package implementing an event model and loop.
5260 event loop
5262 An entity that handles and processes external events and converts
5263 them into callback invocations.
5264 event model
5266 The model used to describe how an event loop handles and processes
5267 watchers and events.
5268 pending
5270 A watcher is pending as soon as the corresponding event has been
5271 detected. See "WATCHER STATES" for details.
5272 real time
5274 The physical time that is observed. It is apparently strictly
5275 monotonic :)
5276 wall-clock time
5278 The time and date as shown on clocks. Unlike real time, it can
5279 actually be wrong and jump forwards and backwards, e.g. when you
5280 adjust your clock.
5281 watcher
5283 A data structure that describes interest in certain events.
5284 Watchers need to be started (attached to an event loop) before they
5285 can receive events.
5286
5288 Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael
5289 Magnusson and Emanuele Giaquinta, and minor corrections by many others.
5290libev-4.23 2016-11-16 LIBEV(3)