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_loop's to stop iterating
33 ev_unloop (EV_A_ EVUNLOOP_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_loop to stop iterating
42 ev_unloop (EV_A_ EVUNLOOP_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_loop (0);
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_loop (loop, 0);
63 // unloop 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 Familarity with event based programming techniques in general is
82 assumed throughout this document.
83
85 Libev is an event loop: you register interest in certain events (such
86 as a file descriptor being readable or a timeout occurring), and it
87 will manage these event sources and provide your program with events.
88 To do this, it must take more or less complete control over your
90 process (or thread) by executing the event loop handler, and will then
91 communicate events via a callback mechanism.
92 You register interest in certain events by registering so-called event
94 watchers, which are relatively small C structures you initialise with
95 the details of the event, and then hand it over to libev by starting
96 the watcher.
97 FEATURES
99 Libev supports "select", "poll", the Linux-specific "epoll", the BSD-
100 specific "kqueue" and the Solaris-specific event port mechanisms for
101 file descriptor events ("ev_io"), the Linux "inotify" interface (for
102 "ev_stat"), Linux eventfd/signalfd (for faster and cleaner inter-thread
103 wakeup ("ev_async")/signal handling ("ev_signal")) relative timers
104 ("ev_timer"), absolute timers with customised rescheduling
105 ("ev_periodic"), synchronous signals ("ev_signal"), process status
106 change events ("ev_child"), and event watchers dealing with the event
107 loop mechanism itself ("ev_idle", "ev_embed", "ev_prepare" and
108 "ev_check" watchers) as well as file watchers ("ev_stat") and even
109 limited support for fork events ("ev_fork").
110 It also is quite fast (see this <benchmark> comparing it to libevent
112 for example).
113 CONVENTIONS
115 Libev is very configurable. In this manual the default (and most
116 common) configuration will be described, which supports multiple event
117 loops. For more info about various configuration options please have a
118 look at EMBED section in this manual. If libev was configured without
119 support for multiple event loops, then all functions taking an initial
120 argument of name "loop" (which is always of type "struct ev_loop *")
121 will not have this argument.
122 TIME REPRESENTATION
124 Libev represents time as a single floating point number, representing
125 the (fractional) number of seconds since the (POSIX) epoch (somewhere
126 near the beginning of 1970, details are complicated, don't ask). This
127 type is called "ev_tstamp", which is what you should use too. It
128 usually aliases to the "double" type in C. When you need to do any
129 calculations on it, you should treat it as some floating point value.
130 Unlike the name component "stamp" might indicate, it is also used for
131 time differences throughout libev.
132
134 Libev knows three classes of errors: operating system errors, usage
135 errors and internal errors (bugs).
136 When libev catches an operating system error it cannot handle (for
138 example a system call indicating a condition libev cannot fix), it
139 calls the callback set via "ev_set_syserr_cb", which is supposed to fix
140 the problem or abort. The default is to print a diagnostic message and
141 to call "abort ()".
142 When libev detects a usage error such as a negative timer interval,
144 then it will print a diagnostic message and abort (via the "assert"
145 mechanism, so "NDEBUG" will disable this checking): these are
146 programming errors in the libev caller and need to be fixed there.
147 Libev also has a few internal error-checking "assert"ions, and also has
149 extensive consistency checking code. These do not trigger under normal
150 circumstances, as they indicate either a bug in libev or worse.
151
153 These functions can be called anytime, even before initialising the
154 library in any way.
155 ev_tstamp ev_time ()
157 Returns the current time as libev would use it. Please note that
158 the "ev_now" function is usually faster and also often returns the
159 timestamp you actually want to know.
160 ev_sleep (ev_tstamp interval)
162 Sleep for the given interval: The current thread will be blocked
163 until either it is interrupted or the given time interval has
164 passed. Basically this is a sub-second-resolution "sleep ()".
165 int ev_version_major ()
167 int ev_version_minor ()
168 You can find out the major and minor ABI version numbers of the
169 library you linked against by calling the functions
170 "ev_version_major" and "ev_version_minor". If you want, you can
171 compare against the global symbols "EV_VERSION_MAJOR" and
172 "EV_VERSION_MINOR", which specify the version of the library your
173 program was compiled against.
174 These version numbers refer to the ABI version of the library, not
176 the release version.
177 Usually, it's a good idea to terminate if the major versions
179 mismatch, as this indicates an incompatible change. Minor versions
180 are usually compatible to older versions, so a larger minor version
181 alone is usually not a problem.
182 Example: Make sure we haven't accidentally been linked against the
184 wrong version.
185 assert (("libev version mismatch",
187 ev_version_major () == EV_VERSION_MAJOR
188 && ev_version_minor () >= EV_VERSION_MINOR));
189 unsigned int ev_supported_backends ()
191 Return the set of all backends (i.e. their corresponding
192 "EV_BACKEND_*" value) compiled into this binary of libev
193 (independent of their availability on the system you are running
194 on). See "ev_default_loop" for a description of the set values.
195 Example: make sure we have the epoll method, because yeah this is
197 cool and a must have and can we have a torrent of it please!!!11
198 assert (("sorry, no epoll, no sex",
200 ev_supported_backends () & EVBACKEND_EPOLL));
201 unsigned int ev_recommended_backends ()
203 Return the set of all backends compiled into this binary of libev
204 and also recommended for this platform. This set is often smaller
205 than the one returned by "ev_supported_backends", as for example
206 kqueue is broken on most BSDs and will not be auto-detected unless
207 you explicitly request it (assuming you know what you are doing).
208 This is the set of backends that libev will probe for if you
209 specify no backends explicitly.
210 unsigned int ev_embeddable_backends ()
212 Returns the set of backends that are embeddable in other event
213 loops. This is the theoretical, all-platform, value. To find which
214 backends might be supported on the current system, you would need
215 to look at "ev_embeddable_backends () & ev_supported_backends ()",
216 likewise for recommended ones.
217 See the description of "ev_embed" watchers for more info.
219 ev_set_allocator (void *(*cb)(void *ptr, long size)) [NOT REENTRANT]
221 Sets the allocation function to use (the prototype is similar - the
222 semantics are identical to the "realloc" C89/SuS/POSIX function).
223 It is used to allocate and free memory (no surprises here). If it
224 returns zero when memory needs to be allocated ("size != 0"), the
225 library might abort or take some potentially destructive action.
226 Since some systems (at least OpenBSD and Darwin) fail to implement
228 correct "realloc" semantics, libev will use a wrapper around the
229 system "realloc" and "free" functions by default.
230 You could override this function in high-availability programs to,
232 say, free some memory if it cannot allocate memory, to use a
233 special allocator, or even to sleep a while and retry until some
234 memory is available.
235 Example: Replace the libev allocator with one that waits a bit and
237 then retries (example requires a standards-compliant "realloc").
238 static void *
240 persistent_realloc (void *ptr, size_t size)
241 {
242 for (;;)
243 {
244 void *newptr = realloc (ptr, size);
245 if (newptr)
247 return newptr;
248 sleep (60);
250 }
251 }
252 ...
254 ev_set_allocator (persistent_realloc);
255 ev_set_syserr_cb (void (*cb)(const char *msg)); [NOT REENTRANT]
257 Set the callback function to call on a retryable system call error
258 (such as failed select, poll, epoll_wait). The message is a
259 printable string indicating the system call or subsystem causing
260 the problem. If this callback is set, then libev will expect it to
261 remedy the situation, no matter what, when it returns. That is,
262 libev will generally retry the requested operation, or, if the
263 condition doesn't go away, do bad stuff (such as abort).
264 Example: This is basically the same thing that libev does
266 internally, too.
267 static void
269 fatal_error (const char *msg)
270 {
271 perror (msg);
272 abort ();
273 }
274 ...
276 ev_set_syserr_cb (fatal_error);
277
279 An event loop is described by a "struct ev_loop *" (the "struct" is not
280 optional in this case, as there is also an "ev_loop" function).
281 The library knows two types of such loops, the default loop, which
283 supports signals and child events, and dynamically created loops which
284 do not.
285 struct ev_loop *ev_default_loop (unsigned int flags)
287 This will initialise the default event loop if it hasn't been
288 initialised yet and return it. If the default loop could not be
289 initialised, returns false. If it already was initialised it simply
290 returns it (and ignores the flags. If that is troubling you, check
291 "ev_backend ()" afterwards).
292 If you don't know what event loop to use, use the one returned from
294 this function.
295 Note that this function is not thread-safe, so if you want to use
297 it from multiple threads, you have to lock (note also that this is
298 unlikely, as loops cannot be shared easily between threads anyway).
299 The default loop is the only loop that can handle "ev_signal" and
301 "ev_child" watchers, and to do this, it always registers a handler
302 for "SIGCHLD". If this is a problem for your application you can
303 either create a dynamic loop with "ev_loop_new" that doesn't do
304 that, or you can simply overwrite the "SIGCHLD" signal handler
305 after calling "ev_default_init".
306 The flags argument can be used to specify special behaviour or
308 specific backends to use, and is usually specified as 0 (or
309 "EVFLAG_AUTO").
310 The following flags are supported:
312 "EVFLAG_AUTO"
314 The default flags value. Use this if you have no clue (it's the
315 right thing, believe me).
316 "EVFLAG_NOENV"
318 If this flag bit is or'ed into the flag value (or the program
319 runs setuid or setgid) then libev will not look at the
320 environment variable "LIBEV_FLAGS". Otherwise (the default),
321 this environment variable will override the flags completely if
322 it is found in the environment. This is useful to try out
323 specific backends to test their performance, or to work around
324 bugs.
325 "EVFLAG_FORKCHECK"
327 Instead of calling "ev_default_fork" or "ev_loop_fork" manually
328 after a fork, you can also make libev check for a fork in each
329 iteration by enabling this flag.
330 This works by calling "getpid ()" on every iteration of the
332 loop, and thus this might slow down your event loop if you do a
333 lot of loop iterations and little real work, but is usually not
334 noticeable (on my GNU/Linux system for example, "getpid" is
335 actually a simple 5-insn sequence without a system call and
336 thus very fast, but my GNU/Linux system also has
337 "pthread_atfork" which is even faster).
338 The big advantage of this flag is that you can forget about
340 fork (and forget about forgetting to tell libev about forking)
341 when you use this flag.
342 This flag setting cannot be overridden or specified in the
344 "LIBEV_FLAGS" environment variable.
345 "EVFLAG_NOINOTIFY"
347 When this flag is specified, then libev will not attempt to use
348 the inotify API for it's "ev_stat" watchers. Apart from
349 debugging and testing, this flag can be useful to conserve
350 inotify file descriptors, as otherwise each loop using
351 "ev_stat" watchers consumes one inotify handle.
352 "EVFLAG_SIGNALFD"
354 When this flag is specified, then libev will attempt to use the
355 signalfd API for it's "ev_signal" (and "ev_child") watchers.
356 This API delivers signals synchronously, which makes it both
357 faster and might make it possible to get the queued signal
358 data. It can also simplify signal handling with threads, as
359 long as you properly block signals in your threads that are not
360 interested in handling them.
361 Signalfd will not be used by default as this changes your
363 signal mask, and there are a lot of shoddy libraries and
364 programs (glib's threadpool for example) that can't properly
365 initialise their signal masks.
366 "EVBACKEND_SELECT" (value 1, portable select backend)
368 This is your standard select(2) backend. Not completely
369 standard, as libev tries to roll its own fd_set with no limits
370 on the number of fds, but if that fails, expect a fairly low
371 limit on the number of fds when using this backend. It doesn't
372 scale too well (O(highest_fd)), but its usually the fastest
373 backend for a low number of (low-numbered :) fds.
374 To get good performance out of this backend you need a high
376 amount of parallelism (most of the file descriptors should be
377 busy). If you are writing a server, you should "accept ()" in a
378 loop to accept as many connections as possible during one
379 iteration. You might also want to have a look at
380 "ev_set_io_collect_interval ()" to increase the amount of
381 readiness notifications you get per iteration.
382 This backend maps "EV_READ" to the "readfds" set and "EV_WRITE"
384 to the "writefds" set (and to work around Microsoft Windows
385 bugs, also onto the "exceptfds" set on that platform).
386 "EVBACKEND_POLL" (value 2, poll backend, available everywhere
388 except on windows)
389 And this is your standard poll(2) backend. It's more
390 complicated than select, but handles sparse fds better and has
391 no artificial limit on the number of fds you can use (except it
392 will slow down considerably with a lot of inactive fds). It
393 scales similarly to select, i.e. O(total_fds). See the entry
394 for "EVBACKEND_SELECT", above, for performance tips.
395 This backend maps "EV_READ" to "POLLIN | POLLERR | POLLHUP",
397 and "EV_WRITE" to "POLLOUT | POLLERR | POLLHUP".
398 "EVBACKEND_EPOLL" (value 4, Linux)
400 Use the linux-specific epoll(7) interface (for both pre- and
401 post-2.6.9 kernels).
402 For few fds, this backend is a bit little slower than poll and
404 select, but it scales phenomenally better. While poll and
405 select usually scale like O(total_fds) where n is the total
406 number of fds (or the highest fd), epoll scales either O(1) or
407 O(active_fds).
408 The epoll mechanism deserves honorable mention as the most
410 misdesigned of the more advanced event mechanisms: mere
411 annoyances include silently dropping file descriptors,
412 requiring a system call per change per file descriptor (and
413 unnecessary guessing of parameters), problems with dup and so
414 on. The biggest issue is fork races, however - if a program
415 forks then both parent and child process have to recreate the
416 epoll set, which can take considerable time (one syscall per
417 file descriptor) and is of course hard to detect.
418 Epoll is also notoriously buggy - embedding epoll fds should
420 work, but of course doesn't, and epoll just loves to report
421 events for totally different file descriptors (even already
422 closed ones, so one cannot even remove them from the set) than
423 registered in the set (especially on SMP systems). Libev tries
424 to counter these spurious notifications by employing an
425 additional generation counter and comparing that against the
426 events to filter out spurious ones, recreating the set when
427 required.
428 While stopping, setting and starting an I/O watcher in the same
430 iteration will result in some caching, there is still a system
431 call per such incident (because the same file descriptor could
432 point to a different file description now), so its best to
433 avoid that. Also, "dup ()"'ed file descriptors might not work
434 very well if you register events for both file descriptors.
435 Best performance from this backend is achieved by not
437 unregistering all watchers for a file descriptor until it has
438 been closed, if possible, i.e. keep at least one watcher active
439 per fd at all times. Stopping and starting a watcher (without
440 re-setting it) also usually doesn't cause extra overhead. A
441 fork can both result in spurious notifications as well as in
442 libev having to destroy and recreate the epoll object, which
443 can take considerable time and thus should be avoided.
444 All this means that, in practice, "EVBACKEND_SELECT" can be as
446 fast or faster than epoll for maybe up to a hundred file
447 descriptors, depending on the usage. So sad.
448 While nominally embeddable in other event loops, this feature
450 is broken in all kernel versions tested so far.
451 This backend maps "EV_READ" and "EV_WRITE" in the same way as
453 "EVBACKEND_POLL".
454 "EVBACKEND_KQUEUE" (value 8, most BSD clones)
456 Kqueue deserves special mention, as at the time of this
457 writing, it was broken on all BSDs except NetBSD (usually it
458 doesn't work reliably with anything but sockets and pipes,
459 except on Darwin, where of course it's completely useless).
460 Unlike epoll, however, whose brokenness is by design, these
461 kqueue bugs can (and eventually will) be fixed without API
462 changes to existing programs. For this reason it's not being
463 "auto-detected" unless you explicitly specify it in the flags
464 (i.e. using "EVBACKEND_KQUEUE") or libev was compiled on a
465 known-to-be-good (-enough) system like NetBSD.
466 You still can embed kqueue into a normal poll or select backend
468 and use it only for sockets (after having made sure that
469 sockets work with kqueue on the target platform). See
470 "ev_embed" watchers for more info.
471 It scales in the same way as the epoll backend, but the
473 interface to the kernel is more efficient (which says nothing
474 about its actual speed, of course). While stopping, setting and
475 starting an I/O watcher does never cause an extra system call
476 as with "EVBACKEND_EPOLL", it still adds up to two event
477 changes per incident. Support for "fork ()" is very bad (but
478 sane, unlike epoll) and it drops fds silently in similarly
479 hard-to-detect cases
480 This backend usually performs well under most conditions.
482 While nominally embeddable in other event loops, this doesn't
484 work everywhere, so you might need to test for this. And since
485 it is broken almost everywhere, you should only use it when you
486 have a lot of sockets (for which it usually works), by
487 embedding it into another event loop (e.g. "EVBACKEND_SELECT"
488 or "EVBACKEND_POLL" (but "poll" is of course also broken on OS
489 X)) and, did I mention it, using it only for sockets.
490 This backend maps "EV_READ" into an "EVFILT_READ" kevent with
492 "NOTE_EOF", and "EV_WRITE" into an "EVFILT_WRITE" kevent with
493 "NOTE_EOF".
494 "EVBACKEND_DEVPOLL" (value 16, Solaris 8)
496 This is not implemented yet (and might never be, unless you
497 send me an implementation). According to reports, "/dev/poll"
498 only supports sockets and is not embeddable, which would limit
499 the usefulness of this backend immensely.
500 "EVBACKEND_PORT" (value 32, Solaris 10)
502 This uses the Solaris 10 event port mechanism. As with
503 everything on Solaris, it's really slow, but it still scales
504 very well (O(active_fds)).
505 Please note that Solaris event ports can deliver a lot of
507 spurious notifications, so you need to use non-blocking I/O or
508 other means to avoid blocking when no data (or space) is
509 available.
510 While this backend scales well, it requires one system call per
512 active file descriptor per loop iteration. For small and medium
513 numbers of file descriptors a "slow" "EVBACKEND_SELECT" or
514 "EVBACKEND_POLL" backend might perform better.
515 On the positive side, with the exception of the spurious
517 readiness notifications, this backend actually performed fully
518 to specification in all tests and is fully embeddable, which is
519 a rare feat among the OS-specific backends (I vastly prefer
520 correctness over speed hacks).
521 This backend maps "EV_READ" and "EV_WRITE" in the same way as
523 "EVBACKEND_POLL".
524 "EVBACKEND_ALL"
526 Try all backends (even potentially broken ones that wouldn't be
527 tried with "EVFLAG_AUTO"). Since this is a mask, you can do
528 stuff such as "EVBACKEND_ALL & ~EVBACKEND_KQUEUE".
529 It is definitely not recommended to use this flag.
531 If one or more of the backend flags are or'ed into the flags value,
533 then only these backends will be tried (in the reverse order as
534 listed here). If none are specified, all backends in
535 "ev_recommended_backends ()" will be tried.
536 Example: This is the most typical usage.
538 if (!ev_default_loop (0))
540 fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
541 Example: Restrict libev to the select and poll backends, and do not
543 allow environment settings to be taken into account:
544 ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
546 Example: Use whatever libev has to offer, but make sure that kqueue
548 is used if available (warning, breaks stuff, best use only with
549 your own private event loop and only if you know the OS supports
550 your types of fds):
551 ev_default_loop (ev_recommended_backends () | EVBACKEND_KQUEUE);
553 struct ev_loop *ev_loop_new (unsigned int flags)
555 Similar to "ev_default_loop", but always creates a new event loop
556 that is always distinct from the default loop. Unlike the default
557 loop, it cannot handle signal and child watchers, and attempts to
558 do so will be greeted by undefined behaviour (or a failed assertion
559 if assertions are enabled).
560 Note that this function is thread-safe, and the recommended way to
562 use libev with threads is indeed to create one loop per thread, and
563 using the default loop in the "main" or "initial" thread.
564 Example: Try to create a event loop that uses epoll and nothing
566 else.
567 struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
569 if (!epoller)
570 fatal ("no epoll found here, maybe it hides under your chair");
571 ev_default_destroy ()
573 Destroys the default loop again (frees all memory and kernel state
574 etc.). None of the active event watchers will be stopped in the
575 normal sense, so e.g. "ev_is_active" might still return true. It is
576 your responsibility to either stop all watchers cleanly yourself
577 before calling this function, or cope with the fact afterwards
578 (which is usually the easiest thing, you can just ignore the
579 watchers and/or "free ()" them for example).
580 Note that certain global state, such as signal state (and installed
582 signal handlers), will not be freed by this function, and related
583 watchers (such as signal and child watchers) would need to be
584 stopped manually.
585 In general it is not advisable to call this function except in the
587 rare occasion where you really need to free e.g. the signal
588 handling pipe fds. If you need dynamically allocated loops it is
589 better to use "ev_loop_new" and "ev_loop_destroy".
590 ev_loop_destroy (loop)
592 Like "ev_default_destroy", but destroys an event loop created by an
593 earlier call to "ev_loop_new".
594 ev_default_fork ()
596 This function sets a flag that causes subsequent "ev_loop"
597 iterations to reinitialise the kernel state for backends that have
598 one. Despite the name, you can call it anytime, but it makes most
599 sense after forking, in the child process (or both child and
600 parent, but that again makes little sense). You must call it in the
601 child before using any of the libev functions, and it will only
602 take effect at the next "ev_loop" iteration.
603 On the other hand, you only need to call this function in the child
605 process if and only if you want to use the event library in the
606 child. If you just fork+exec, you don't have to call it at all.
607 The function itself is quite fast and it's usually not a problem to
609 call it just in case after a fork. To make this easy, the function
610 will fit in quite nicely into a call to "pthread_atfork":
611 pthread_atfork (0, 0, ev_default_fork);
613 ev_loop_fork (loop)
615 Like "ev_default_fork", but acts on an event loop created by
616 "ev_loop_new". Yes, you have to call this on every allocated event
617 loop after fork that you want to re-use in the child, and how you
618 do this is entirely your own problem.
619 int ev_is_default_loop (loop)
621 Returns true when the given loop is, in fact, the default loop, and
622 false otherwise.
623 unsigned int ev_loop_count (loop)
625 Returns the count of loop iterations for the loop, which is
626 identical to the number of times libev did poll for new events. It
627 starts at 0 and happily wraps around with enough iterations.
628 This value can sometimes be useful as a generation counter of sorts
630 (it "ticks" the number of loop iterations), as it roughly
631 corresponds with "ev_prepare" and "ev_check" calls.
632 unsigned int ev_loop_depth (loop)
634 Returns the number of times "ev_loop" was entered minus the number
635 of times "ev_loop" was exited, in other words, the recursion depth.
636 Outside "ev_loop", this number is zero. In a callback, this number
638 is 1, unless "ev_loop" was invoked recursively (or from another
639 thread), in which case it is higher.
640 Leaving "ev_loop" abnormally (setjmp/longjmp, cancelling the thread
642 etc.), doesn't count as exit.
643 unsigned int ev_backend (loop)
645 Returns one of the "EVBACKEND_*" flags indicating the event backend
646 in use.
647 ev_tstamp ev_now (loop)
649 Returns the current "event loop time", which is the time the event
650 loop received events and started processing them. This timestamp
651 does not change as long as callbacks are being processed, and this
652 is also the base time used for relative timers. You can treat it as
653 the timestamp of the event occurring (or more correctly, libev
654 finding out about it).
655 ev_now_update (loop)
657 Establishes the current time by querying the kernel, updating the
658 time returned by "ev_now ()" in the progress. This is a costly
659 operation and is usually done automatically within "ev_loop ()".
660 This function is rarely useful, but when some event callback runs
662 for a very long time without entering the event loop, updating
663 libev's idea of the current time is a good idea.
664 See also "The special problem of time updates" in the "ev_timer"
666 section.
667 ev_suspend (loop)
669 ev_resume (loop)
670 These two functions suspend and resume a loop, for use when the
671 loop is not used for a while and timeouts should not be processed.
672 A typical use case would be an interactive program such as a game:
674 When the user presses "^Z" to suspend the game and resumes it an
675 hour later it would be best to handle timeouts as if no time had
676 actually passed while the program was suspended. This can be
677 achieved by calling "ev_suspend" in your "SIGTSTP" handler, sending
678 yourself a "SIGSTOP" and calling "ev_resume" directly afterwards to
679 resume timer processing.
680 Effectively, all "ev_timer" watchers will be delayed by the time
682 spend between "ev_suspend" and "ev_resume", and all "ev_periodic"
683 watchers will be rescheduled (that is, they will lose any events
684 that would have occured while suspended).
685 After calling "ev_suspend" you must not call any function on the
687 given loop other than "ev_resume", and you must not call
688 "ev_resume" without a previous call to "ev_suspend".
689 Calling "ev_suspend"/"ev_resume" has the side effect of updating
691 the event loop time (see "ev_now_update").
692 ev_loop (loop, int flags)
694 Finally, this is it, the event handler. This function usually is
695 called after you have initialised all your watchers and you want to
696 start handling events.
697 If the flags argument is specified as 0, it will not return until
699 either no event watchers are active anymore or "ev_unloop" was
700 called.
701 Please note that an explicit "ev_unloop" is usually better than
703 relying on all watchers to be stopped when deciding when a program
704 has finished (especially in interactive programs), but having a
705 program that automatically loops as long as it has to and no longer
706 by virtue of relying on its watchers stopping correctly, that is
707 truly a thing of beauty.
708 A flags value of "EVLOOP_NONBLOCK" will look for new events, will
710 handle those events and any already outstanding ones, but will not
711 block your process in case there are no events and will return
712 after one iteration of the loop.
713 A flags value of "EVLOOP_ONESHOT" will look for new events (waiting
715 if necessary) and will handle those and any already outstanding
716 ones. It will block your process until at least one new event
717 arrives (which could be an event internal to libev itself, so there
718 is no guarantee that a user-registered callback will be called),
719 and will return after one iteration of the loop.
720 This is useful if you are waiting for some external event in
722 conjunction with something not expressible using other libev
723 watchers (i.e. "roll your own "ev_loop""). However, a pair of
724 "ev_prepare"/"ev_check" watchers is usually a better approach for
725 this kind of thing.
726 Here are the gory details of what "ev_loop" does:
728 - Before the first iteration, call any pending watchers.
730 * If EVFLAG_FORKCHECK was used, check for a fork.
731 - If a fork was detected (by any means), queue and call all fork watchers.
732 - Queue and call all prepare watchers.
733 - If we have been forked, detach and recreate the kernel state
734 as to not disturb the other process.
735 - Update the kernel state with all outstanding changes.
736 - Update the "event loop time" (ev_now ()).
737 - Calculate for how long to sleep or block, if at all
738 (active idle watchers, EVLOOP_NONBLOCK or not having
739 any active watchers at all will result in not sleeping).
740 - Sleep if the I/O and timer collect interval say so.
741 - Block the process, waiting for any events.
742 - Queue all outstanding I/O (fd) events.
743 - Update the "event loop time" (ev_now ()), and do time jump adjustments.
744 - Queue all expired timers.
745 - Queue all expired periodics.
746 - Unless any events are pending now, queue all idle watchers.
747 - Queue all check watchers.
748 - Call all queued watchers in reverse order (i.e. check watchers first).
749 Signals and child watchers are implemented as I/O watchers, and will
750 be handled here by queueing them when their watcher gets executed.
751 - If ev_unloop has been called, or EVLOOP_ONESHOT or EVLOOP_NONBLOCK
752 were used, or there are no active watchers, return, otherwise
753 continue with step *.
754 Example: Queue some jobs and then loop until no events are
756 outstanding anymore.
757 ... queue jobs here, make sure they register event watchers as long
759 ... as they still have work to do (even an idle watcher will do..)
760 ev_loop (my_loop, 0);
761 ... jobs done or somebody called unloop. yeah!
762 ev_unloop (loop, how)
764 Can be used to make a call to "ev_loop" return early (but only
765 after it has processed all outstanding events). The "how" argument
766 must be either "EVUNLOOP_ONE", which will make the innermost
767 "ev_loop" call return, or "EVUNLOOP_ALL", which will make all
768 nested "ev_loop" calls return.
769 This "unloop state" will be cleared when entering "ev_loop" again.
771 It is safe to call "ev_unloop" from otuside any "ev_loop" calls.
773 ev_ref (loop)
775 ev_unref (loop)
776 Ref/unref can be used to add or remove a reference count on the
777 event loop: Every watcher keeps one reference, and as long as the
778 reference count is nonzero, "ev_loop" will not return on its own.
779 This is useful when you have a watcher that you never intend to
781 unregister, but that nevertheless should not keep "ev_loop" from
782 returning. In such a case, call "ev_unref" after starting, and
783 "ev_ref" before stopping it.
784 As an example, libev itself uses this for its internal signal pipe:
786 It is not visible to the libev user and should not keep "ev_loop"
787 from exiting if no event watchers registered by it are active. It
788 is also an excellent way to do this for generic recurring timers or
789 from within third-party libraries. Just remember to unref after
790 start and ref before stop (but only if the watcher wasn't active
791 before, or was active before, respectively. Note also that libev
792 might stop watchers itself (e.g. non-repeating timers) in which
793 case you have to "ev_ref" in the callback).
794 Example: Create a signal watcher, but keep it from keeping
796 "ev_loop" running when nothing else is active.
797 ev_signal exitsig;
799 ev_signal_init (&exitsig, sig_cb, SIGINT);
800 ev_signal_start (loop, &exitsig);
801 evf_unref (loop);
802 Example: For some weird reason, unregister the above signal handler
804 again.
805 ev_ref (loop);
807 ev_signal_stop (loop, &exitsig);
808 ev_set_io_collect_interval (loop, ev_tstamp interval)
810 ev_set_timeout_collect_interval (loop, ev_tstamp interval)
811 These advanced functions influence the time that libev will spend
812 waiting for events. Both time intervals are by default 0, meaning
813 that libev will try to invoke timer/periodic callbacks and I/O
814 callbacks with minimum latency.
815 Setting these to a higher value (the "interval" must be >= 0)
817 allows libev to delay invocation of I/O and timer/periodic
818 callbacks to increase efficiency of loop iterations (or to increase
819 power-saving opportunities).
820 The idea is that sometimes your program runs just fast enough to
822 handle one (or very few) event(s) per loop iteration. While this
823 makes the program responsive, it also wastes a lot of CPU time to
824 poll for new events, especially with backends like "select ()"
825 which have a high overhead for the actual polling but can deliver
826 many events at once.
827 By setting a higher io collect interval you allow libev to spend
829 more time collecting I/O events, so you can handle more events per
830 iteration, at the cost of increasing latency. Timeouts (both
831 "ev_periodic" and "ev_timer") will be not affected. Setting this to
832 a non-null value will introduce an additional "ev_sleep ()" call
833 into most loop iterations. The sleep time ensures that libev will
834 not poll for I/O events more often then once per this interval, on
835 average.
836 Likewise, by setting a higher timeout collect interval you allow
838 libev to spend more time collecting timeouts, at the expense of
839 increased latency/jitter/inexactness (the watcher callback will be
840 called later). "ev_io" watchers will not be affected. Setting this
841 to a non-null value will not introduce any overhead in libev.
842 Many (busy) programs can usually benefit by setting the I/O collect
844 interval to a value near 0.1 or so, which is often enough for
845 interactive servers (of course not for games), likewise for
846 timeouts. It usually doesn't make much sense to set it to a lower
847 value than 0.01, as this approaches the timing granularity of most
848 systems. Note that if you do transactions with the outside world
849 and you can't increase the parallelity, then this setting will
850 limit your transaction rate (if you need to poll once per
851 transaction and the I/O collect interval is 0.01, then you can't do
852 more than 100 transations per second).
853 Setting the timeout collect interval can improve the opportunity
855 for saving power, as the program will "bundle" timer callback
856 invocations that are "near" in time together, by delaying some,
857 thus reducing the number of times the process sleeps and wakes up
858 again. Another useful technique to reduce iterations/wake-ups is to
859 use "ev_periodic" watchers and make sure they fire on, say, one-
860 second boundaries only.
861 Example: we only need 0.1s timeout granularity, and we wish not to
863 poll more often than 100 times per second:
864 ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
866 ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
867 ev_invoke_pending (loop)
869 This call will simply invoke all pending watchers while resetting
870 their pending state. Normally, "ev_loop" does this automatically
871 when required, but when overriding the invoke callback this call
872 comes handy.
873 int ev_pending_count (loop)
875 Returns the number of pending watchers - zero indicates that no
876 watchers are pending.
877 ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
879 This overrides the invoke pending functionality of the loop:
880 Instead of invoking all pending watchers when there are any,
881 "ev_loop" will call this callback instead. This is useful, for
882 example, when you want to invoke the actual watchers inside another
883 context (another thread etc.).
884 If you want to reset the callback, use "ev_invoke_pending" as new
886 callback.
887 ev_set_loop_release_cb (loop, void (*release)(EV_P), void
889 (*acquire)(EV_P))
890 Sometimes you want to share the same loop between multiple threads.
891 This can be done relatively simply by putting mutex_lock/unlock
892 calls around each call to a libev function.
893 However, "ev_loop" can run an indefinite time, so it is not
895 feasible to wait for it to return. One way around this is to wake
896 up the loop via "ev_unloop" and "av_async_send", another way is to
897 set these release and acquire callbacks on the loop.
898 When set, then "release" will be called just before the thread is
900 suspended waiting for new events, and "acquire" is called just
901 afterwards.
902 Ideally, "release" will just call your mutex_unlock function, and
904 "acquire" will just call the mutex_lock function again.
905 While event loop modifications are allowed between invocations of
907 "release" and "acquire" (that's their only purpose after all), no
908 modifications done will affect the event loop, i.e. adding watchers
909 will have no effect on the set of file descriptors being watched,
910 or the time waited. Use an "ev_async" watcher to wake up "ev_loop"
911 when you want it to take note of any changes you made.
912 In theory, threads executing "ev_loop" will be async-cancel safe
914 between invocations of "release" and "acquire".
915 See also the locking example in the "THREADS" section later in this
917 document.
918 ev_set_userdata (loop, void *data)
920 ev_userdata (loop)
921 Set and retrieve a single "void *" associated with a loop. When
922 "ev_set_userdata" has never been called, then "ev_userdata" returns
923 0.
924 These two functions can be used to associate arbitrary data with a
926 loop, and are intended solely for the "invoke_pending_cb",
927 "release" and "acquire" callbacks described above, but of course
928 can be (ab-)used for any other purpose as well.
929 ev_loop_verify (loop)
931 This function only does something when "EV_VERIFY" support has been
932 compiled in, which is the default for non-minimal builds. It tries
933 to go through all internal structures and checks them for validity.
934 If anything is found to be inconsistent, it will print an error
935 message to standard error and call "abort ()".
936 This can be used to catch bugs inside libev itself: under normal
938 circumstances, this function will never abort as of course libev
939 keeps its data structures consistent.
940
942 In the following description, uppercase "TYPE" in names stands for the
943 watcher type, e.g. "ev_TYPE_start" can mean "ev_timer_start" for timer
944 watchers and "ev_io_start" for I/O watchers.
945 A watcher is a structure that you create and register to record your
947 interest in some event. For instance, if you want to wait for STDIN to
948 become readable, you would create an "ev_io" watcher for that:
949 static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
951 {
952 ev_io_stop (w);
953 ev_unloop (loop, EVUNLOOP_ALL);
954 }
955 struct ev_loop *loop = ev_default_loop (0);
957 ev_io stdin_watcher;
959 ev_init (&stdin_watcher, my_cb);
961 ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
962 ev_io_start (loop, &stdin_watcher);
963 ev_loop (loop, 0);
965 As you can see, you are responsible for allocating the memory for your
967 watcher structures (and it is usually a bad idea to do this on the
968 stack).
969 Each watcher has an associated watcher structure (called "struct
971 ev_TYPE" or simply "ev_TYPE", as typedefs are provided for all watcher
972 structs).
973 Each watcher structure must be initialised by a call to "ev_init
975 (watcher *, callback)", which expects a callback to be provided. This
976 callback gets invoked each time the event occurs (or, in the case of
977 I/O watchers, each time the event loop detects that the file descriptor
978 given is readable and/or writable).
979 Each watcher type further has its own "ev_TYPE_set (watcher *, ...)"
981 macro to configure it, with arguments specific to the watcher type.
982 There is also a macro to combine initialisation and setting in one
983 call: "ev_TYPE_init (watcher *, callback, ...)".
984 To make the watcher actually watch out for events, you have to start it
986 with a watcher-specific start function ("ev_TYPE_start (loop, watcher
987 *)"), and you can stop watching for events at any time by calling the
988 corresponding stop function ("ev_TYPE_stop (loop, watcher *)".
989 As long as your watcher is active (has been started but not stopped)
991 you must not touch the values stored in it. Most specifically you must
992 never reinitialise it or call its "ev_TYPE_set" macro.
993 Each and every callback receives the event loop pointer as first, the
995 registered watcher structure as second, and a bitset of received events
996 as third argument.
997 The received events usually include a single bit per event type
999 received (you can receive multiple events at the same time). The
1000 possible bit masks are:
1001 "EV_READ"
1003 "EV_WRITE"
1004 The file descriptor in the "ev_io" watcher has become readable
1005 and/or writable.
1006 "EV_TIMEOUT"
1008 The "ev_timer" watcher has timed out.
1009 "EV_PERIODIC"
1011 The "ev_periodic" watcher has timed out.
1012 "EV_SIGNAL"
1014 The signal specified in the "ev_signal" watcher has been received
1015 by a thread.
1016 "EV_CHILD"
1018 The pid specified in the "ev_child" watcher has received a status
1019 change.
1020 "EV_STAT"
1022 The path specified in the "ev_stat" watcher changed its attributes
1023 somehow.
1024 "EV_IDLE"
1026 The "ev_idle" watcher has determined that you have nothing better
1027 to do.
1028 "EV_PREPARE"
1030 "EV_CHECK"
1031 All "ev_prepare" watchers are invoked just before "ev_loop" starts
1032 to gather new events, and all "ev_check" watchers are invoked just
1033 after "ev_loop" has gathered them, but before it invokes any
1034 callbacks for any received events. Callbacks of both watcher types
1035 can start and stop as many watchers as they want, and all of them
1036 will be taken into account (for example, a "ev_prepare" watcher
1037 might start an idle watcher to keep "ev_loop" from blocking).
1038 "EV_EMBED"
1040 The embedded event loop specified in the "ev_embed" watcher needs
1041 attention.
1042 "EV_FORK"
1044 The event loop has been resumed in the child process after fork
1045 (see "ev_fork").
1046 "EV_ASYNC"
1048 The given async watcher has been asynchronously notified (see
1049 "ev_async").
1050 "EV_CUSTOM"
1052 Not ever sent (or otherwise used) by libev itself, but can be
1053 freely used by libev users to signal watchers (e.g. via
1054 "ev_feed_event").
1055 "EV_ERROR"
1057 An unspecified error has occurred, the watcher has been stopped.
1058 This might happen because the watcher could not be properly started
1059 because libev ran out of memory, a file descriptor was found to be
1060 closed or any other problem. Libev considers these application
1061 bugs.
1062 You best act on it by reporting the problem and somehow coping with
1064 the watcher being stopped. Note that well-written programs should
1065 not receive an error ever, so when your watcher receives it, this
1066 usually indicates a bug in your program.
1067 Libev will usually signal a few "dummy" events together with an
1069 error, for example it might indicate that a fd is readable or
1070 writable, and if your callbacks is well-written it can just attempt
1071 the operation and cope with the error from read() or write(). This
1072 will not work in multi-threaded programs, though, as the fd could
1073 already be closed and reused for another thing, so beware.
1074 GENERIC WATCHER FUNCTIONS
1076 "ev_init" (ev_TYPE *watcher, callback)
1077 This macro initialises the generic portion of a watcher. The
1078 contents of the watcher object can be arbitrary (so "malloc" will
1079 do). Only the generic parts of the watcher are initialised, you
1080 need to call the type-specific "ev_TYPE_set" macro afterwards to
1081 initialise the type-specific parts. For each type there is also a
1082 "ev_TYPE_init" macro which rolls both calls into one.
1083 You can reinitialise a watcher at any time as long as it has been
1085 stopped (or never started) and there are no pending events
1086 outstanding.
1087 The callback is always of type "void (*)(struct ev_loop *loop,
1089 ev_TYPE *watcher, int revents)".
1090 Example: Initialise an "ev_io" watcher in two steps.
1092 ev_io w;
1094 ev_init (&w, my_cb);
1095 ev_io_set (&w, STDIN_FILENO, EV_READ);
1096 "ev_TYPE_set" (ev_TYPE *watcher, [args])
1098 This macro initialises the type-specific parts of a watcher. You
1099 need to call "ev_init" at least once before you call this macro,
1100 but you can call "ev_TYPE_set" any number of times. You must not,
1101 however, call this macro on a watcher that is active (it can be
1102 pending, however, which is a difference to the "ev_init" macro).
1103 Although some watcher types do not have type-specific arguments
1105 (e.g. "ev_prepare") you still need to call its "set" macro.
1106 See "ev_init", above, for an example.
1108 "ev_TYPE_init" (ev_TYPE *watcher, callback, [args])
1110 This convenience macro rolls both "ev_init" and "ev_TYPE_set" macro
1111 calls into a single call. This is the most convenient method to
1112 initialise a watcher. The same limitations apply, of course.
1113 Example: Initialise and set an "ev_io" watcher in one step.
1115 ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
1117 "ev_TYPE_start" (loop, ev_TYPE *watcher)
1119 Starts (activates) the given watcher. Only active watchers will
1120 receive events. If the watcher is already active nothing will
1121 happen.
1122 Example: Start the "ev_io" watcher that is being abused as example
1124 in this whole section.
1125 ev_io_start (EV_DEFAULT_UC, &w);
1127 "ev_TYPE_stop" (loop, ev_TYPE *watcher)
1129 Stops the given watcher if active, and clears the pending status
1130 (whether the watcher was active or not).
1131 It is possible that stopped watchers are pending - for example,
1133 non-repeating timers are being stopped when they become pending -
1134 but calling "ev_TYPE_stop" ensures that the watcher is neither
1135 active nor pending. If you want to free or reuse the memory used by
1136 the watcher it is therefore a good idea to always call its
1137 "ev_TYPE_stop" function.
1138 bool ev_is_active (ev_TYPE *watcher)
1140 Returns a true value iff the watcher is active (i.e. it has been
1141 started and not yet been stopped). As long as a watcher is active
1142 you must not modify it.
1143 bool ev_is_pending (ev_TYPE *watcher)
1145 Returns a true value iff the watcher is pending, (i.e. it has
1146 outstanding events but its callback has not yet been invoked). As
1147 long as a watcher is pending (but not active) you must not call an
1148 init function on it (but "ev_TYPE_set" is safe), you must not
1149 change its priority, and you must make sure the watcher is
1150 available to libev (e.g. you cannot "free ()" it).
1151 callback ev_cb (ev_TYPE *watcher)
1153 Returns the callback currently set on the watcher.
1154 ev_cb_set (ev_TYPE *watcher, callback)
1156 Change the callback. You can change the callback at virtually any
1157 time (modulo threads).
1158 ev_set_priority (ev_TYPE *watcher, int priority)
1160 int ev_priority (ev_TYPE *watcher)
1161 Set and query the priority of the watcher. The priority is a small
1162 integer between "EV_MAXPRI" (default: 2) and "EV_MINPRI" (default:
1163 "-2"). Pending watchers with higher priority will be invoked before
1164 watchers with lower priority, but priority will not keep watchers
1165 from being executed (except for "ev_idle" watchers).
1166 If you need to suppress invocation when higher priority events are
1168 pending you need to look at "ev_idle" watchers, which provide this
1169 functionality.
1170 You must not change the priority of a watcher as long as it is
1172 active or pending.
1173 Setting a priority outside the range of "EV_MINPRI" to "EV_MAXPRI"
1175 is fine, as long as you do not mind that the priority value you
1176 query might or might not have been clamped to the valid range.
1177 The default priority used by watchers when no priority has been set
1179 is always 0, which is supposed to not be too high and not be too
1180 low :).
1181 See "WATCHER PRIORITY MODELS", below, for a more thorough treatment
1183 of priorities.
1184 ev_invoke (loop, ev_TYPE *watcher, int revents)
1186 Invoke the "watcher" with the given "loop" and "revents". Neither
1187 "loop" nor "revents" need to be valid as long as the watcher
1188 callback can deal with that fact, as both are simply passed through
1189 to the callback.
1190 int ev_clear_pending (loop, ev_TYPE *watcher)
1192 If the watcher is pending, this function clears its pending status
1193 and returns its "revents" bitset (as if its callback was invoked).
1194 If the watcher isn't pending it does nothing and returns 0.
1195 Sometimes it can be useful to "poll" a watcher instead of waiting
1197 for its callback to be invoked, which can be accomplished with this
1198 function.
1199 ev_feed_event (loop, ev_TYPE *watcher, int revents)
1201 Feeds the given event set into the event loop, as if the specified
1202 event had happened for the specified watcher (which must be a
1203 pointer to an initialised but not necessarily started event
1204 watcher). Obviously you must not free the watcher as long as it has
1205 pending events.
1206 Stopping the watcher, letting libev invoke it, or calling
1208 "ev_clear_pending" will clear the pending event, even if the
1209 watcher was not started in the first place.
1210 See also "ev_feed_fd_event" and "ev_feed_signal_event" for related
1212 functions that do not need a watcher.
1213 ASSOCIATING CUSTOM DATA WITH A WATCHER
1215 Each watcher has, by default, a member "void *data" that you can change
1216 and read at any time: libev will completely ignore it. This can be used
1217 to associate arbitrary data with your watcher. If you need more data
1218 and don't want to allocate memory and store a pointer to it in that
1219 data member, you can also "subclass" the watcher type and provide your
1220 own data:
1221 struct my_io
1223 {
1224 ev_io io;
1225 int otherfd;
1226 void *somedata;
1227 struct whatever *mostinteresting;
1228 };
1229 ...
1231 struct my_io w;
1232 ev_io_init (&w.io, my_cb, fd, EV_READ);
1233 And since your callback will be called with a pointer to the watcher,
1235 you can cast it back to your own type:
1236 static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
1238 {
1239 struct my_io *w = (struct my_io *)w_;
1240 ...
1241 }
1242 More interesting and less C-conformant ways of casting your callback
1244 type instead have been omitted.
1245 Another common scenario is to use some data structure with multiple
1247 embedded watchers:
1248 struct my_biggy
1250 {
1251 int some_data;
1252 ev_timer t1;
1253 ev_timer t2;
1254 }
1255 In this case getting the pointer to "my_biggy" is a bit more
1257 complicated: Either you store the address of your "my_biggy" struct in
1258 the "data" member of the watcher (for woozies), or you need to use some
1259 pointer arithmetic using "offsetof" inside your watchers (for real
1260 programmers):
1261 #include <stddef.h>
1263 static void
1265 t1_cb (EV_P_ ev_timer *w, int revents)
1266 {
1267 struct my_biggy big = (struct my_biggy *)
1268 (((char *)w) - offsetof (struct my_biggy, t1));
1269 }
1270 static void
1272 t2_cb (EV_P_ ev_timer *w, int revents)
1273 {
1274 struct my_biggy big = (struct my_biggy *)
1275 (((char *)w) - offsetof (struct my_biggy, t2));
1276 }
1277 WATCHER PRIORITY MODELS
1279 Many event loops support watcher priorities, which are usually small
1280 integers that influence the ordering of event callback invocation
1281 between watchers in some way, all else being equal.
1282 In libev, Watcher priorities can be set using "ev_set_priority". See
1284 its description for the more technical details such as the actual
1285 priority range.
1286 There are two common ways how these these priorities are being
1288 interpreted by event loops:
1289 In the more common lock-out model, higher priorities "lock out"
1291 invocation of lower priority watchers, which means as long as higher
1292 priority watchers receive events, lower priority watchers are not being
1293 invoked.
1294 The less common only-for-ordering model uses priorities solely to order
1296 callback invocation within a single event loop iteration: Higher
1297 priority watchers are invoked before lower priority ones, but they all
1298 get invoked before polling for new events.
1299 Libev uses the second (only-for-ordering) model for all its watchers
1301 except for idle watchers (which use the lock-out model).
1302 The rationale behind this is that implementing the lock-out model for
1304 watchers is not well supported by most kernel interfaces, and most
1305 event libraries will just poll for the same events again and again as
1306 long as their callbacks have not been executed, which is very
1307 inefficient in the common case of one high-priority watcher locking out
1308 a mass of lower priority ones.
1309 Static (ordering) priorities are most useful when you have two or more
1311 watchers handling the same resource: a typical usage example is having
1312 an "ev_io" watcher to receive data, and an associated "ev_timer" to
1313 handle timeouts. Under load, data might be received while the program
1314 handles other jobs, but since timers normally get invoked first, the
1315 timeout handler will be executed before checking for data. In that
1316 case, giving the timer a lower priority than the I/O watcher ensures
1317 that I/O will be handled first even under adverse conditions (which is
1318 usually, but not always, what you want).
1319 Since idle watchers use the "lock-out" model, meaning that idle
1321 watchers will only be executed when no same or higher priority watchers
1322 have received events, they can be used to implement the "lock-out"
1323 model when required.
1324 For example, to emulate how many other event libraries handle
1326 priorities, you can associate an "ev_idle" watcher to each such
1327 watcher, and in the normal watcher callback, you just start the idle
1328 watcher. The real processing is done in the idle watcher callback. This
1329 causes libev to continously poll and process kernel event data for the
1330 watcher, but when the lock-out case is known to be rare (which in turn
1331 is rare :), this is workable.
1332 Usually, however, the lock-out model implemented that way will perform
1334 miserably under the type of load it was designed to handle. In that
1335 case, it might be preferable to stop the real watcher before starting
1336 the idle watcher, so the kernel will not have to process the event in
1337 case the actual processing will be delayed for considerable time.
1338 Here is an example of an I/O watcher that should run at a strictly
1340 lower priority than the default, and which should only process data
1341 when no other events are pending:
1342 ev_idle idle; // actual processing watcher
1344 ev_io io; // actual event watcher
1345 static void
1347 io_cb (EV_P_ ev_io *w, int revents)
1348 {
1349 // stop the I/O watcher, we received the event, but
1350 // are not yet ready to handle it.
1351 ev_io_stop (EV_A_ w);
1352 // start the idle watcher to ahndle the actual event.
1354 // it will not be executed as long as other watchers
1355 // with the default priority are receiving events.
1356 ev_idle_start (EV_A_ &idle);
1357 }
1358 static void
1360 idle_cb (EV_P_ ev_idle *w, int revents)
1361 {
1362 // actual processing
1363 read (STDIN_FILENO, ...);
1364 // have to start the I/O watcher again, as
1366 // we have handled the event
1367 ev_io_start (EV_P_ &io);
1368 }
1369 // initialisation
1371 ev_idle_init (&idle, idle_cb);
1372 ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
1373 ev_io_start (EV_DEFAULT_ &io);
1374 In the "real" world, it might also be beneficial to start a timer, so
1376 that low-priority connections can not be locked out forever under load.
1377 This enables your program to keep a lower latency for important
1378 connections during short periods of high load, while not completely
1379 locking out less important ones.
1380
1382 This section describes each watcher in detail, but will not repeat
1383 information given in the last section. Any initialisation/set macros,
1384 functions and members specific to the watcher type are explained.
1385 Members are additionally marked with either [read-only], meaning that,
1387 while the watcher is active, you can look at the member and expect some
1388 sensible content, but you must not modify it (you can modify it while
1389 the watcher is stopped to your hearts content), or [read-write], which
1390 means you can expect it to have some sensible content while the watcher
1391 is active, but you can also modify it. Modifying it may not do
1392 something sensible or take immediate effect (or do anything at all),
1393 but libev will not crash or malfunction in any way.
1394 "ev_io" - is this file descriptor readable or writable?
1396 I/O watchers check whether a file descriptor is readable or writable in
1397 each iteration of the event loop, or, more precisely, when reading
1398 would not block the process and writing would at least be able to write
1399 some data. This behaviour is called level-triggering because you keep
1400 receiving events as long as the condition persists. Remember you can
1401 stop the watcher if you don't want to act on the event and neither want
1402 to receive future events.
1403 In general you can register as many read and/or write event watchers
1405 per fd as you want (as long as you don't confuse yourself). Setting all
1406 file descriptors to non-blocking mode is also usually a good idea (but
1407 not required if you know what you are doing).
1408 If you cannot use non-blocking mode, then force the use of a known-to-
1410 be-good backend (at the time of this writing, this includes only
1411 "EVBACKEND_SELECT" and "EVBACKEND_POLL"). The same applies to file
1412 descriptors for which non-blocking operation makes no sense (such as
1413 files) - libev doesn't guarentee any specific behaviour in that case.
1414 Another thing you have to watch out for is that it is quite easy to
1416 receive "spurious" readiness notifications, that is your callback might
1417 be called with "EV_READ" but a subsequent "read"(2) will actually block
1418 because there is no data. Not only are some backends known to create a
1419 lot of those (for example Solaris ports), it is very easy to get into
1420 this situation even with a relatively standard program structure. Thus
1421 it is best to always use non-blocking I/O: An extra "read"(2) returning
1422 "EAGAIN" is far preferable to a program hanging until some data
1423 arrives.
1424 If you cannot run the fd in non-blocking mode (for example you should
1426 not play around with an Xlib connection), then you have to separately
1427 re-test whether a file descriptor is really ready with a known-to-be
1428 good interface such as poll (fortunately in our Xlib example, Xlib
1429 already does this on its own, so its quite safe to use). Some people
1430 additionally use "SIGALRM" and an interval timer, just to be sure you
1431 won't block indefinitely.
1432 But really, best use non-blocking mode.
1434 The special problem of disappearing file descriptors
1436 Some backends (e.g. kqueue, epoll) need to be told about closing a file
1438 descriptor (either due to calling "close" explicitly or any other
1439 means, such as "dup2"). The reason is that you register interest in
1440 some file descriptor, but when it goes away, the operating system will
1441 silently drop this interest. If another file descriptor with the same
1442 number then is registered with libev, there is no efficient way to see
1443 that this is, in fact, a different file descriptor.
1444 To avoid having to explicitly tell libev about such cases, libev
1446 follows the following policy: Each time "ev_io_set" is being called,
1447 libev will assume that this is potentially a new file descriptor,
1448 otherwise it is assumed that the file descriptor stays the same. That
1449 means that you have to call "ev_io_set" (or "ev_io_init") when you
1450 change the descriptor even if the file descriptor number itself did not
1451 change.
1452 This is how one would do it normally anyway, the important point is
1454 that the libev application should not optimise around libev but should
1455 leave optimisations to libev.
1456 The special problem of dup'ed file descriptors
1458 Some backends (e.g. epoll), cannot register events for file
1460 descriptors, but only events for the underlying file descriptions. That
1461 means when you have "dup ()"'ed file descriptors or weirder
1462 constellations, and register events for them, only one file descriptor
1463 might actually receive events.
1464 There is no workaround possible except not registering events for
1466 potentially "dup ()"'ed file descriptors, or to resort to
1467 "EVBACKEND_SELECT" or "EVBACKEND_POLL".
1468 The special problem of fork
1470 Some backends (epoll, kqueue) do not support "fork ()" at all or
1472 exhibit useless behaviour. Libev fully supports fork, but needs to be
1473 told about it in the child.
1474 To support fork in your programs, you either have to call
1476 "ev_default_fork ()" or "ev_loop_fork ()" after a fork in the child,
1477 enable "EVFLAG_FORKCHECK", or resort to "EVBACKEND_SELECT" or
1478 "EVBACKEND_POLL".
1479 The special problem of SIGPIPE
1481 While not really specific to libev, it is easy to forget about
1483 "SIGPIPE": when writing to a pipe whose other end has been closed, your
1484 program gets sent a SIGPIPE, which, by default, aborts your program.
1485 For most programs this is sensible behaviour, for daemons, this is
1486 usually undesirable.
1487 So when you encounter spurious, unexplained daemon exits, make sure you
1489 ignore SIGPIPE (and maybe make sure you log the exit status of your
1490 daemon somewhere, as that would have given you a big clue).
1491 Watcher-Specific Functions
1493 ev_io_init (ev_io *, callback, int fd, int events)
1495 ev_io_set (ev_io *, int fd, int events)
1496 Configures an "ev_io" watcher. The "fd" is the file descriptor to
1497 receive events for and "events" is either "EV_READ", "EV_WRITE" or
1498 "EV_READ | EV_WRITE", to express the desire to receive the given
1499 events.
1500 int fd [read-only]
1502 The file descriptor being watched.
1503 int events [read-only]
1505 The events being watched.
1506 Examples
1508 Example: Call "stdin_readable_cb" when STDIN_FILENO has become, well
1510 readable, but only once. Since it is likely line-buffered, you could
1511 attempt to read a whole line in the callback.
1512 static void
1514 stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
1515 {
1516 ev_io_stop (loop, w);
1517 .. read from stdin here (or from w->fd) and handle any I/O errors
1518 }
1519 ...
1521 struct ev_loop *loop = ev_default_init (0);
1522 ev_io stdin_readable;
1523 ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
1524 ev_io_start (loop, &stdin_readable);
1525 ev_loop (loop, 0);
1526 "ev_timer" - relative and optionally repeating timeouts
1528 Timer watchers are simple relative timers that generate an event after
1529 a given time, and optionally repeating in regular intervals after that.
1530 The timers are based on real time, that is, if you register an event
1532 that times out after an hour and you reset your system clock to January
1533 last year, it will still time out after (roughly) one hour. "Roughly"
1534 because detecting time jumps is hard, and some inaccuracies are
1535 unavoidable (the monotonic clock option helps a lot here).
1536 The callback is guaranteed to be invoked only after its timeout has
1538 passed (not at, so on systems with very low-resolution clocks this
1539 might introduce a small delay). If multiple timers become ready during
1540 the same loop iteration then the ones with earlier time-out values are
1541 invoked before ones of the same priority with later time-out values
1542 (but this is no longer true when a callback calls "ev_loop"
1543 recursively).
1544 Be smart about timeouts
1546 Many real-world problems involve some kind of timeout, usually for
1548 error recovery. A typical example is an HTTP request - if the other
1549 side hangs, you want to raise some error after a while.
1550 What follows are some ways to handle this problem, from obvious and
1552 inefficient to smart and efficient.
1553 In the following, a 60 second activity timeout is assumed - a timeout
1555 that gets reset to 60 seconds each time there is activity (e.g. each
1556 time some data or other life sign was received).
1557 1. Use a timer and stop, reinitialise and start it on activity.
1559 This is the most obvious, but not the most simple way: In the
1560 beginning, start the watcher:
1561 ev_timer_init (timer, callback, 60., 0.);
1563 ev_timer_start (loop, timer);
1564 Then, each time there is some activity, "ev_timer_stop" it,
1566 initialise it and start it again:
1567 ev_timer_stop (loop, timer);
1569 ev_timer_set (timer, 60., 0.);
1570 ev_timer_start (loop, timer);
1571 This is relatively simple to implement, but means that each time
1573 there is some activity, libev will first have to remove the timer
1574 from its internal data structure and then add it again. Libev tries
1575 to be fast, but it's still not a constant-time operation.
1576 2. Use a timer and re-start it with "ev_timer_again" inactivity.
1578 This is the easiest way, and involves using "ev_timer_again"
1579 instead of "ev_timer_start".
1580 To implement this, configure an "ev_timer" with a "repeat" value of
1582 60 and then call "ev_timer_again" at start and each time you
1583 successfully read or write some data. If you go into an idle state
1584 where you do not expect data to travel on the socket, you can
1585 "ev_timer_stop" the timer, and "ev_timer_again" will automatically
1586 restart it if need be.
1587 That means you can ignore both the "ev_timer_start" function and
1589 the "after" argument to "ev_timer_set", and only ever use the
1590 "repeat" member and "ev_timer_again".
1591 At start:
1593 ev_init (timer, callback);
1595 timer->repeat = 60.;
1596 ev_timer_again (loop, timer);
1597 Each time there is some activity:
1599 ev_timer_again (loop, timer);
1601 It is even possible to change the time-out on the fly, regardless
1603 of whether the watcher is active or not:
1604 timer->repeat = 30.;
1606 ev_timer_again (loop, timer);
1607 This is slightly more efficient then stopping/starting the timer
1609 each time you want to modify its timeout value, as libev does not
1610 have to completely remove and re-insert the timer from/into its
1611 internal data structure.
1612 It is, however, even simpler than the "obvious" way to do it.
1614 3. Let the timer time out, but then re-arm it as required.
1616 This method is more tricky, but usually most efficient: Most
1617 timeouts are relatively long compared to the intervals between
1618 other activity - in our example, within 60 seconds, there are
1619 usually many I/O events with associated activity resets.
1620 In this case, it would be more efficient to leave the "ev_timer"
1622 alone, but remember the time of last activity, and check for a real
1623 timeout only within the callback:
1624 ev_tstamp last_activity; // time of last activity
1626 static void
1628 callback (EV_P_ ev_timer *w, int revents)
1629 {
1630 ev_tstamp now = ev_now (EV_A);
1631 ev_tstamp timeout = last_activity + 60.;
1632 // if last_activity + 60. is older than now, we did time out
1634 if (timeout < now)
1635 {
1636 // timeout occured, take action
1637 }
1638 else
1639 {
1640 // callback was invoked, but there was some activity, re-arm
1641 // the watcher to fire in last_activity + 60, which is
1642 // guaranteed to be in the future, so "again" is positive:
1643 w->repeat = timeout - now;
1644 ev_timer_again (EV_A_ w);
1645 }
1646 }
1647 To summarise the callback: first calculate the real timeout
1649 (defined as "60 seconds after the last activity"), then check if
1650 that time has been reached, which means something did, in fact,
1651 time out. Otherwise the callback was invoked too early ("timeout"
1652 is in the future), so re-schedule the timer to fire at that future
1653 time, to see if maybe we have a timeout then.
1654 Note how "ev_timer_again" is used, taking advantage of the
1656 "ev_timer_again" optimisation when the timer is already running.
1657 This scheme causes more callback invocations (about one every 60
1659 seconds minus half the average time between activity), but
1660 virtually no calls to libev to change the timeout.
1661 To start the timer, simply initialise the watcher and set
1663 "last_activity" to the current time (meaning we just have some
1664 activity :), then call the callback, which will "do the right
1665 thing" and start the timer:
1666 ev_init (timer, callback);
1668 last_activity = ev_now (loop);
1669 callback (loop, timer, EV_TIMEOUT);
1670 And when there is some activity, simply store the current time in
1672 "last_activity", no libev calls at all:
1673 last_actiivty = ev_now (loop);
1675 This technique is slightly more complex, but in most cases where
1677 the time-out is unlikely to be triggered, much more efficient.
1678 Changing the timeout is trivial as well (if it isn't hard-coded in
1680 the callback :) - just change the timeout and invoke the callback,
1681 which will fix things for you.
1682 4. Wee, just use a double-linked list for your timeouts.
1684 If there is not one request, but many thousands (millions...), all
1685 employing some kind of timeout with the same timeout value, then
1686 one can do even better:
1687 When starting the timeout, calculate the timeout value and put the
1689 timeout at the end of the list.
1690 Then use an "ev_timer" to fire when the timeout at the beginning of
1692 the list is expected to fire (for example, using the technique #3).
1693 When there is some activity, remove the timer from the list,
1695 recalculate the timeout, append it to the end of the list again,
1696 and make sure to update the "ev_timer" if it was taken from the
1697 beginning of the list.
1698 This way, one can manage an unlimited number of timeouts in O(1)
1700 time for starting, stopping and updating the timers, at the expense
1701 of a major complication, and having to use a constant timeout. The
1702 constant timeout ensures that the list stays sorted.
1703 So which method the best?
1705 Method #2 is a simple no-brain-required solution that is adequate in
1707 most situations. Method #3 requires a bit more thinking, but handles
1708 many cases better, and isn't very complicated either. In most case,
1709 choosing either one is fine, with #3 being better in typical
1710 situations.
1711 Method #1 is almost always a bad idea, and buys you nothing. Method #4
1713 is rather complicated, but extremely efficient, something that really
1714 pays off after the first million or so of active timers, i.e. it's
1715 usually overkill :)
1716 The special problem of time updates
1718 Establishing the current time is a costly operation (it usually takes
1720 at least two system calls): EV therefore updates its idea of the
1721 current time only before and after "ev_loop" collects new events, which
1722 causes a growing difference between "ev_now ()" and "ev_time ()" when
1723 handling lots of events in one iteration.
1724 The relative timeouts are calculated relative to the "ev_now ()" time.
1726 This is usually the right thing as this timestamp refers to the time of
1727 the event triggering whatever timeout you are modifying/starting. If
1728 you suspect event processing to be delayed and you need to base the
1729 timeout on the current time, use something like this to adjust for
1730 this:
1731 ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);
1733 If the event loop is suspended for a long time, you can also force an
1735 update of the time returned by "ev_now ()" by calling "ev_now_update
1736 ()".
1737 The special problems of suspended animation
1739 When you leave the server world it is quite customary to hit machines
1741 that can suspend/hibernate - what happens to the clocks during such a
1742 suspend?
1743 Some quick tests made with a Linux 2.6.28 indicate that a suspend
1745 freezes all processes, while the clocks ("times", "CLOCK_MONOTONIC")
1746 continue to run until the system is suspended, but they will not
1747 advance while the system is suspended. That means, on resume, it will
1748 be as if the program was frozen for a few seconds, but the suspend time
1749 will not be counted towards "ev_timer" when a monotonic clock source is
1750 used. The real time clock advanced as expected, but if it is used as
1751 sole clocksource, then a long suspend would be detected as a time jump
1752 by libev, and timers would be adjusted accordingly.
1753 I would not be surprised to see different behaviour in different
1755 between operating systems, OS versions or even different hardware.
1756 The other form of suspend (job control, or sending a SIGSTOP) will see
1758 a time jump in the monotonic clocks and the realtime clock. If the
1759 program is suspended for a very long time, and monotonic clock sources
1760 are in use, then you can expect "ev_timer"s to expire as the full
1761 suspension time will be counted towards the timers. When no monotonic
1762 clock source is in use, then libev will again assume a timejump and
1763 adjust accordingly.
1764 It might be beneficial for this latter case to call "ev_suspend" and
1766 "ev_resume" in code that handles "SIGTSTP", to at least get
1767 deterministic behaviour in this case (you can do nothing against
1768 "SIGSTOP").
1769 Watcher-Specific Functions and Data Members
1771 ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
1773 ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
1774 Configure the timer to trigger after "after" seconds. If "repeat"
1775 is 0., then it will automatically be stopped once the timeout is
1776 reached. If it is positive, then the timer will automatically be
1777 configured to trigger again "repeat" seconds later, again, and
1778 again, until stopped manually.
1779 The timer itself will do a best-effort at avoiding drift, that is,
1781 if you configure a timer to trigger every 10 seconds, then it will
1782 normally trigger at exactly 10 second intervals. If, however, your
1783 program cannot keep up with the timer (because it takes longer than
1784 those 10 seconds to do stuff) the timer will not fire more than
1785 once per event loop iteration.
1786 ev_timer_again (loop, ev_timer *)
1788 This will act as if the timer timed out and restart it again if it
1789 is repeating. The exact semantics are:
1790 If the timer is pending, its pending status is cleared.
1792 If the timer is started but non-repeating, stop it (as if it timed
1794 out).
1795 If the timer is repeating, either start it if necessary (with the
1797 "repeat" value), or reset the running timer to the "repeat" value.
1798 This sounds a bit complicated, see "Be smart about timeouts",
1800 above, for a usage example.
1801 ev_tstamp ev_timer_remaining (loop, ev_timer *)
1803 Returns the remaining time until a timer fires. If the timer is
1804 active, then this time is relative to the current event loop time,
1805 otherwise it's the timeout value currently configured.
1806 That is, after an "ev_timer_set (w, 5, 7)", "ev_timer_remaining"
1808 returns 5. When the timer is started and one second passes,
1809 "ev_timer_remain" will return 4. When the timer expires and is
1810 restarted, it will return roughly 7 (likely slightly less as
1811 callback invocation takes some time, too), and so on.
1812 ev_tstamp repeat [read-write]
1814 The current "repeat" value. Will be used each time the watcher
1815 times out or "ev_timer_again" is called, and determines the next
1816 timeout (if any), which is also when any modifications are taken
1817 into account.
1818 Examples
1820 Example: Create a timer that fires after 60 seconds.
1822 static void
1824 one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
1825 {
1826 .. one minute over, w is actually stopped right here
1827 }
1828 ev_timer mytimer;
1830 ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
1831 ev_timer_start (loop, &mytimer);
1832 Example: Create a timeout timer that times out after 10 seconds of
1834 inactivity.
1835 static void
1837 timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
1838 {
1839 .. ten seconds without any activity
1840 }
1841 ev_timer mytimer;
1843 ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
1844 ev_timer_again (&mytimer); /* start timer */
1845 ev_loop (loop, 0);
1846 // and in some piece of code that gets executed on any "activity":
1848 // reset the timeout to start ticking again at 10 seconds
1849 ev_timer_again (&mytimer);
1850 "ev_periodic" - to cron or not to cron?
1852 Periodic watchers are also timers of a kind, but they are very
1853 versatile (and unfortunately a bit complex).
1854 Unlike "ev_timer", periodic watchers are not based on real time (or
1856 relative time, the physical time that passes) but on wall clock time
1857 (absolute time, the thing you can read on your calender or clock). The
1858 difference is that wall clock time can run faster or slower than real
1859 time, and time jumps are not uncommon (e.g. when you adjust your wrist-
1860 watch).
1861 You can tell a periodic watcher to trigger after some specific point in
1863 time: for example, if you tell a periodic watcher to trigger "in 10
1864 seconds" (by specifying e.g. "ev_now () + 10.", that is, an absolute
1865 time not a delay) and then reset your system clock to January of the
1866 previous year, then it will take a year or more to trigger the event
1867 (unlike an "ev_timer", which would still trigger roughly 10 seconds
1868 after starting it, as it uses a relative timeout).
1869 "ev_periodic" watchers can also be used to implement vastly more
1871 complex timers, such as triggering an event on each "midnight, local
1872 time", or other complicated rules. This cannot be done with "ev_timer"
1873 watchers, as those cannot react to time jumps.
1874 As with timers, the callback is guaranteed to be invoked only when the
1876 point in time where it is supposed to trigger has passed. If multiple
1877 timers become ready during the same loop iteration then the ones with
1878 earlier time-out values are invoked before ones with later time-out
1879 values (but this is no longer true when a callback calls "ev_loop"
1880 recursively).
1881 Watcher-Specific Functions and Data Members
1883 ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp
1885 interval, reschedule_cb)
1886 ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval,
1887 reschedule_cb)
1888 Lots of arguments, let's sort it out... There are basically three
1889 modes of operation, and we will explain them from simplest to most
1890 complex:
1891 · absolute timer (offset = absolute time, interval = 0,
1893 reschedule_cb = 0)
1894 In this configuration the watcher triggers an event after the
1896 wall clock time "offset" has passed. It will not repeat and
1897 will not adjust when a time jump occurs, that is, if it is to
1898 be run at January 1st 2011 then it will be stopped and invoked
1899 when the system clock reaches or surpasses this point in time.
1900 · repeating interval timer (offset = offset within interval,
1902 interval > 0, reschedule_cb = 0)
1903 In this mode the watcher will always be scheduled to time out
1905 at the next "offset + N * interval" time (for some integer N,
1906 which can also be negative) and then repeat, regardless of any
1907 time jumps. The "offset" argument is merely an offset into the
1908 "interval" periods.
1909 This can be used to create timers that do not drift with
1911 respect to the system clock, for example, here is an
1912 "ev_periodic" that triggers each hour, on the hour (with
1913 respect to UTC):
1914 ev_periodic_set (&periodic, 0., 3600., 0);
1916 This doesn't mean there will always be 3600 seconds in between
1918 triggers, but only that the callback will be called when the
1919 system time shows a full hour (UTC), or more correctly, when
1920 the system time is evenly divisible by 3600.
1921 Another way to think about it (for the mathematically inclined)
1923 is that "ev_periodic" will try to run the callback in this mode
1924 at the next possible time where "time = offset (mod interval)",
1925 regardless of any time jumps.
1926 For numerical stability it is preferable that the "offset"
1928 value is near "ev_now ()" (the current time), but there is no
1929 range requirement for this value, and in fact is often
1930 specified as zero.
1931 Note also that there is an upper limit to how often a timer can
1933 fire (CPU speed for example), so if "interval" is very small
1934 then timing stability will of course deteriorate. Libev itself
1935 tries to be exact to be about one millisecond (if the OS
1936 supports it and the machine is fast enough).
1937 · manual reschedule mode (offset ignored, interval ignored,
1939 reschedule_cb = callback)
1940 In this mode the values for "interval" and "offset" are both
1942 being ignored. Instead, each time the periodic watcher gets
1943 scheduled, the reschedule callback will be called with the
1944 watcher as first, and the current time as second argument.
1945 NOTE: This callback MUST NOT stop or destroy any periodic
1947 watcher, ever, or make ANY other event loop modifications
1948 whatsoever, unless explicitly allowed by documentation here.
1949 If you need to stop it, return "now + 1e30" (or so, fudge
1951 fudge) and stop it afterwards (e.g. by starting an "ev_prepare"
1952 watcher, which is the only event loop modification you are
1953 allowed to do).
1954 The callback prototype is "ev_tstamp
1956 (*reschedule_cb)(ev_periodic *w, ev_tstamp now)", e.g.:
1957 static ev_tstamp
1959 my_rescheduler (ev_periodic *w, ev_tstamp now)
1960 {
1961 return now + 60.;
1962 }
1963 It must return the next time to trigger, based on the passed
1965 time value (that is, the lowest time value larger than to the
1966 second argument). It will usually be called just before the
1967 callback will be triggered, but might be called at other times,
1968 too.
1969 NOTE: This callback must always return a time that is higher
1971 than or equal to the passed "now" value.
1972 This can be used to create very complex timers, such as a timer
1974 that triggers on "next midnight, local time". To do this, you
1975 would calculate the next midnight after "now" and return the
1976 timestamp value for this. How you do this is, again, up to you
1977 (but it is not trivial, which is the main reason I omitted it
1978 as an example).
1979 ev_periodic_again (loop, ev_periodic *)
1981 Simply stops and restarts the periodic watcher again. This is only
1982 useful when you changed some parameters or the reschedule callback
1983 would return a different time than the last time it was called
1984 (e.g. in a crond like program when the crontabs have changed).
1985 ev_tstamp ev_periodic_at (ev_periodic *)
1987 When active, returns the absolute time that the watcher is supposed
1988 to trigger next. This is not the same as the "offset" argument to
1989 "ev_periodic_set", but indeed works even in interval and manual
1990 rescheduling modes.
1991 ev_tstamp offset [read-write]
1993 When repeating, this contains the offset value, otherwise this is
1994 the absolute point in time (the "offset" value passed to
1995 "ev_periodic_set", although libev might modify this value for
1996 better numerical stability).
1997 Can be modified any time, but changes only take effect when the
1999 periodic timer fires or "ev_periodic_again" is being called.
2000 ev_tstamp interval [read-write]
2002 The current interval value. Can be modified any time, but changes
2003 only take effect when the periodic timer fires or
2004 "ev_periodic_again" is being called.
2005 ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
2007 The current reschedule callback, or 0, if this functionality is
2008 switched off. Can be changed any time, but changes only take effect
2009 when the periodic timer fires or "ev_periodic_again" is being
2010 called.
2011 Examples
2013 Example: Call a callback every hour, or, more precisely, whenever the
2015 system time is divisible by 3600. The callback invocation times have
2016 potentially a lot of jitter, but good long-term stability.
2017 static void
2019 clock_cb (struct ev_loop *loop, ev_io *w, int revents)
2020 {
2021 ... its now a full hour (UTC, or TAI or whatever your clock follows)
2022 }
2023 ev_periodic hourly_tick;
2025 ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
2026 ev_periodic_start (loop, &hourly_tick);
2027 Example: The same as above, but use a reschedule callback to do it:
2029 #include <math.h>
2031 static ev_tstamp
2033 my_scheduler_cb (ev_periodic *w, ev_tstamp now)
2034 {
2035 return now + (3600. - fmod (now, 3600.));
2036 }
2037 ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
2039 Example: Call a callback every hour, starting now:
2041 ev_periodic hourly_tick;
2043 ev_periodic_init (&hourly_tick, clock_cb,
2044 fmod (ev_now (loop), 3600.), 3600., 0);
2045 ev_periodic_start (loop, &hourly_tick);
2046 "ev_signal" - signal me when a signal gets signalled!
2048 Signal watchers will trigger an event when the process receives a
2049 specific signal one or more times. Even though signals are very
2050 asynchronous, libev will try it's best to deliver signals
2051 synchronously, i.e. as part of the normal event processing, like any
2052 other event.
2053 If you want signals to be delivered truly asynchronously, just use
2055 "sigaction" as you would do without libev and forget about sharing the
2056 signal. You can even use "ev_async" from a signal handler to
2057 synchronously wake up an event loop.
2058 You can configure as many watchers as you like for the same signal, but
2060 only within the same loop, i.e. you can watch for "SIGINT" in your
2061 default loop and for "SIGIO" in another loop, but you cannot watch for
2062 "SIGINT" in both the default loop and another loop at the same time. At
2063 the moment, "SIGCHLD" is permanently tied to the default loop.
2064 When the first watcher gets started will libev actually register
2066 something with the kernel (thus it coexists with your own signal
2067 handlers as long as you don't register any with libev for the same
2068 signal).
2069 If possible and supported, libev will install its handlers with
2071 "SA_RESTART" (or equivalent) behaviour enabled, so system calls should
2072 not be unduly interrupted. If you have a problem with system calls
2073 getting interrupted by signals you can block all signals in an
2074 "ev_check" watcher and unblock them in an "ev_prepare" watcher.
2075 The special problem of inheritance over fork/execve/pthread_create
2077 Both the signal mask ("sigprocmask") and the signal disposition
2079 ("sigaction") are unspecified after starting a signal watcher (and
2080 after stopping it again), that is, libev might or might not block the
2081 signal, and might or might not set or restore the installed signal
2082 handler.
2083 While this does not matter for the signal disposition (libev never sets
2085 signals to "SIG_IGN", so handlers will be reset to "SIG_DFL" on
2086 "execve"), this matters for the signal mask: many programs do not
2087 expect certain signals to be blocked.
2088 This means that before calling "exec" (from the child) you should reset
2090 the signal mask to whatever "default" you expect (all clear is a good
2091 choice usually).
2092 The simplest way to ensure that the signal mask is reset in the child
2094 is to install a fork handler with "pthread_atfork" that resets it. That
2095 will catch fork calls done by libraries (such as the libc) as well.
2096 In current versions of libev, the signal will not be blocked
2098 indefinitely unless you use the "signalfd" API ("EV_SIGNALFD"). While
2099 this reduces the window of opportunity for problems, it will not go
2100 away, as libev has to modify the signal mask, at least temporarily.
2101 So I can't stress this enough: If you do not reset your signal mask
2103 when you expect it to be empty, you have a race condition in your code.
2104 This is not a libev-specific thing, this is true for most event
2105 libraries.
2106 Watcher-Specific Functions and Data Members
2108 ev_signal_init (ev_signal *, callback, int signum)
2110 ev_signal_set (ev_signal *, int signum)
2111 Configures the watcher to trigger on the given signal number
2112 (usually one of the "SIGxxx" constants).
2113 int signum [read-only]
2115 The signal the watcher watches out for.
2116 Examples
2118 Example: Try to exit cleanly on SIGINT.
2120 static void
2122 sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
2123 {
2124 ev_unloop (loop, EVUNLOOP_ALL);
2125 }
2126 ev_signal signal_watcher;
2128 ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
2129 ev_signal_start (loop, &signal_watcher);
2130 "ev_child" - watch out for process status changes
2132 Child watchers trigger when your process receives a SIGCHLD in response
2133 to some child status changes (most typically when a child of yours dies
2134 or exits). It is permissible to install a child watcher after the child
2135 has been forked (which implies it might have already exited), as long
2136 as the event loop isn't entered (or is continued from a watcher), i.e.,
2137 forking and then immediately registering a watcher for the child is
2138 fine, but forking and registering a watcher a few event loop iterations
2139 later or in the next callback invocation is not.
2140 Only the default event loop is capable of handling signals, and
2142 therefore you can only register child watchers in the default event
2143 loop.
2144 Due to some design glitches inside libev, child watchers will always be
2146 handled at maximum priority (their priority is set to "EV_MAXPRI" by
2147 libev)
2148 Process Interaction
2150 Libev grabs "SIGCHLD" as soon as the default event loop is initialised.
2152 This is necessary to guarantee proper behaviour even if the first child
2153 watcher is started after the child exits. The occurrence of "SIGCHLD"
2154 is recorded asynchronously, but child reaping is done synchronously as
2155 part of the event loop processing. Libev always reaps all children,
2156 even ones not watched.
2157 Overriding the Built-In Processing
2159 Libev offers no special support for overriding the built-in child
2161 processing, but if your application collides with libev's default child
2162 handler, you can override it easily by installing your own handler for
2163 "SIGCHLD" after initialising the default loop, and making sure the
2164 default loop never gets destroyed. You are encouraged, however, to use
2165 an event-based approach to child reaping and thus use libev's support
2166 for that, so other libev users can use "ev_child" watchers freely.
2167 Stopping the Child Watcher
2169 Currently, the child watcher never gets stopped, even when the child
2171 terminates, so normally one needs to stop the watcher in the callback.
2172 Future versions of libev might stop the watcher automatically when a
2173 child exit is detected (calling "ev_child_stop" twice is not a
2174 problem).
2175 Watcher-Specific Functions and Data Members
2177 ev_child_init (ev_child *, callback, int pid, int trace)
2179 ev_child_set (ev_child *, int pid, int trace)
2180 Configures the watcher to wait for status changes of process "pid"
2181 (or any process if "pid" is specified as 0). The callback can look
2182 at the "rstatus" member of the "ev_child" watcher structure to see
2183 the status word (use the macros from "sys/wait.h" and see your
2184 systems "waitpid" documentation). The "rpid" member contains the
2185 pid of the process causing the status change. "trace" must be
2186 either 0 (only activate the watcher when the process terminates) or
2187 1 (additionally activate the watcher when the process is stopped or
2188 continued).
2189 int pid [read-only]
2191 The process id this watcher watches out for, or 0, meaning any
2192 process id.
2193 int rpid [read-write]
2195 The process id that detected a status change.
2196 int rstatus [read-write]
2198 The process exit/trace status caused by "rpid" (see your systems
2199 "waitpid" and "sys/wait.h" documentation for details).
2200 Examples
2202 Example: "fork()" a new process and install a child handler to wait for
2204 its completion.
2205 ev_child cw;
2207 static void
2209 child_cb (EV_P_ ev_child *w, int revents)
2210 {
2211 ev_child_stop (EV_A_ w);
2212 printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
2213 }
2214 pid_t pid = fork ();
2216 if (pid < 0)
2218 // error
2219 else if (pid == 0)
2220 {
2221 // the forked child executes here
2222 exit (1);
2223 }
2224 else
2225 {
2226 ev_child_init (&cw, child_cb, pid, 0);
2227 ev_child_start (EV_DEFAULT_ &cw);
2228 }
2229 "ev_stat" - did the file attributes just change?
2231 This watches a file system path for attribute changes. That is, it
2232 calls "stat" on that path in regular intervals (or when the OS says it
2233 changed) and sees if it changed compared to the last time, invoking the
2234 callback if it did.
2235 The path does not need to exist: changing from "path exists" to "path
2237 does not exist" is a status change like any other. The condition "path
2238 does not exist" (or more correctly "path cannot be stat'ed") is
2239 signified by the "st_nlink" field being zero (which is otherwise always
2240 forced to be at least one) and all the other fields of the stat buffer
2241 having unspecified contents.
2242 The path must not end in a slash or contain special components such as
2244 "." or "..". The path should be absolute: If it is relative and your
2245 working directory changes, then the behaviour is undefined.
2246 Since there is no portable change notification interface available, the
2248 portable implementation simply calls stat(2) regularly on the path to
2249 see if it changed somehow. You can specify a recommended polling
2250 interval for this case. If you specify a polling interval of 0 (highly
2251 recommended!) then a suitable, unspecified default value will be used
2252 (which you can expect to be around five seconds, although this might
2253 change dynamically). Libev will also impose a minimum interval which is
2254 currently around 0.1, but that's usually overkill.
2255 This watcher type is not meant for massive numbers of stat watchers, as
2257 even with OS-supported change notifications, this can be resource-
2258 intensive.
2259 At the time of this writing, the only OS-specific interface implemented
2261 is the Linux inotify interface (implementing kqueue support is left as
2262 an exercise for the reader. Note, however, that the author sees no way
2263 of implementing "ev_stat" semantics with kqueue, except as a hint).
2264 ABI Issues (Largefile Support)
2266 Libev by default (unless the user overrides this) uses the default
2268 compilation environment, which means that on systems with large file
2269 support disabled by default, you get the 32 bit version of the stat
2270 structure. When using the library from programs that change the ABI to
2271 use 64 bit file offsets the programs will fail. In that case you have
2272 to compile libev with the same flags to get binary compatibility. This
2273 is obviously the case with any flags that change the ABI, but the
2274 problem is most noticeably displayed with ev_stat and large file
2275 support.
2276 The solution for this is to lobby your distribution maker to make large
2278 file interfaces available by default (as e.g. FreeBSD does) and not
2279 optional. Libev cannot simply switch on large file support because it
2280 has to exchange stat structures with application programs compiled
2281 using the default compilation environment.
2282 Inotify and Kqueue
2284 When "inotify (7)" support has been compiled into libev and present at
2286 runtime, it will be used to speed up change detection where possible.
2287 The inotify descriptor will be created lazily when the first "ev_stat"
2288 watcher is being started.
2289 Inotify presence does not change the semantics of "ev_stat" watchers
2291 except that changes might be detected earlier, and in some cases, to
2292 avoid making regular "stat" calls. Even in the presence of inotify
2293 support there are many cases where libev has to resort to regular
2294 "stat" polling, but as long as kernel 2.6.25 or newer is used (2.6.24
2295 and older have too many bugs), the path exists (i.e. stat succeeds),
2296 and the path resides on a local filesystem (libev currently assumes
2297 only ext2/3, jfs, reiserfs and xfs are fully working) libev usually
2298 gets away without polling.
2299 There is no support for kqueue, as apparently it cannot be used to
2301 implement this functionality, due to the requirement of having a file
2302 descriptor open on the object at all times, and detecting renames,
2303 unlinks etc. is difficult.
2304 "stat ()" is a synchronous operation
2306 Libev doesn't normally do any kind of I/O itself, and so is not
2308 blocking the process. The exception are "ev_stat" watchers - those call
2309 "stat ()", which is a synchronous operation.
2310 For local paths, this usually doesn't matter: unless the system is very
2312 busy or the intervals between stat's are large, a stat call will be
2313 fast, as the path data is usually in memory already (except when
2314 starting the watcher).
2315 For networked file systems, calling "stat ()" can block an indefinite
2317 time due to network issues, and even under good conditions, a stat call
2318 often takes multiple milliseconds.
2319 Therefore, it is best to avoid using "ev_stat" watchers on networked
2321 paths, although this is fully supported by libev.
2322 The special problem of stat time resolution
2324 The "stat ()" system call only supports full-second resolution
2326 portably, and even on systems where the resolution is higher, most file
2327 systems still only support whole seconds.
2328 That means that, if the time is the only thing that changes, you can
2330 easily miss updates: on the first update, "ev_stat" detects a change
2331 and calls your callback, which does something. When there is another
2332 update within the same second, "ev_stat" will be unable to detect
2333 unless the stat data does change in other ways (e.g. file size).
2334 The solution to this is to delay acting on a change for slightly more
2336 than a second (or till slightly after the next full second boundary),
2337 using a roughly one-second-delay "ev_timer" (e.g. "ev_timer_set (w, 0.,
2338 1.02); ev_timer_again (loop, w)").
2339 The .02 offset is added to work around small timing inconsistencies of
2341 some operating systems (where the second counter of the current time
2342 might be be delayed. One such system is the Linux kernel, where a call
2343 to "gettimeofday" might return a timestamp with a full second later
2344 than a subsequent "time" call - if the equivalent of "time ()" is used
2345 to update file times then there will be a small window where the kernel
2346 uses the previous second to update file times but libev might already
2347 execute the timer callback).
2348 Watcher-Specific Functions and Data Members
2350 ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp
2352 interval)
2353 ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
2354 Configures the watcher to wait for status changes of the given
2355 "path". The "interval" is a hint on how quickly a change is
2356 expected to be detected and should normally be specified as 0 to
2357 let libev choose a suitable value. The memory pointed to by "path"
2358 must point to the same path for as long as the watcher is active.
2359 The callback will receive an "EV_STAT" event when a change was
2361 detected, relative to the attributes at the time the watcher was
2362 started (or the last change was detected).
2363 ev_stat_stat (loop, ev_stat *)
2365 Updates the stat buffer immediately with new values. If you change
2366 the watched path in your callback, you could call this function to
2367 avoid detecting this change (while introducing a race condition if
2368 you are not the only one changing the path). Can also be useful
2369 simply to find out the new values.
2370 ev_statdata attr [read-only]
2372 The most-recently detected attributes of the file. Although the
2373 type is "ev_statdata", this is usually the (or one of the) "struct
2374 stat" types suitable for your system, but you can only rely on the
2375 POSIX-standardised members to be present. If the "st_nlink" member
2376 is 0, then there was some error while "stat"ing the file.
2377 ev_statdata prev [read-only]
2379 The previous attributes of the file. The callback gets invoked
2380 whenever "prev" != "attr", or, more precisely, one or more of these
2381 members differ: "st_dev", "st_ino", "st_mode", "st_nlink",
2382 "st_uid", "st_gid", "st_rdev", "st_size", "st_atime", "st_mtime",
2383 "st_ctime".
2384 ev_tstamp interval [read-only]
2386 The specified interval.
2387 const char *path [read-only]
2389 The file system path that is being watched.
2390 Examples
2392 Example: Watch "/etc/passwd" for attribute changes.
2394 static void
2396 passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
2397 {
2398 /* /etc/passwd changed in some way */
2399 if (w->attr.st_nlink)
2400 {
2401 printf ("passwd current size %ld\n", (long)w->attr.st_size);
2402 printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
2403 printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
2404 }
2405 else
2406 /* you shalt not abuse printf for puts */
2407 puts ("wow, /etc/passwd is not there, expect problems. "
2408 "if this is windows, they already arrived\n");
2409 }
2410 ...
2412 ev_stat passwd;
2413 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
2415 ev_stat_start (loop, &passwd);
2416 Example: Like above, but additionally use a one-second delay so we do
2418 not miss updates (however, frequent updates will delay processing, too,
2419 so one might do the work both on "ev_stat" callback invocation and on
2420 "ev_timer" callback invocation).
2421 static ev_stat passwd;
2423 static ev_timer timer;
2424 static void
2426 timer_cb (EV_P_ ev_timer *w, int revents)
2427 {
2428 ev_timer_stop (EV_A_ w);
2429 /* now it's one second after the most recent passwd change */
2431 }
2432 static void
2434 stat_cb (EV_P_ ev_stat *w, int revents)
2435 {
2436 /* reset the one-second timer */
2437 ev_timer_again (EV_A_ &timer);
2438 }
2439 ...
2441 ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
2442 ev_stat_start (loop, &passwd);
2443 ev_timer_init (&timer, timer_cb, 0., 1.02);
2444 "ev_idle" - when you've got nothing better to do...
2446 Idle watchers trigger events when no other events of the same or higher
2447 priority are pending (prepare, check and other idle watchers do not
2448 count as receiving "events").
2449 That is, as long as your process is busy handling sockets or timeouts
2451 (or even signals, imagine) of the same or higher priority it will not
2452 be triggered. But when your process is idle (or only lower-priority
2453 watchers are pending), the idle watchers are being called once per
2454 event loop iteration - until stopped, that is, or your process receives
2455 more events and becomes busy again with higher priority stuff.
2456 The most noteworthy effect is that as long as any idle watchers are
2458 active, the process will not block when waiting for new events.
2459 Apart from keeping your process non-blocking (which is a useful effect
2461 on its own sometimes), idle watchers are a good place to do "pseudo-
2462 background processing", or delay processing stuff to after the event
2463 loop has handled all outstanding events.
2464 Watcher-Specific Functions and Data Members
2466 ev_idle_init (ev_idle *, callback)
2468 Initialises and configures the idle watcher - it has no parameters
2469 of any kind. There is a "ev_idle_set" macro, but using it is
2470 utterly pointless, believe me.
2471 Examples
2473 Example: Dynamically allocate an "ev_idle" watcher, start it, and in
2475 the callback, free it. Also, use no error checking, as usual.
2476 static void
2478 idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
2479 {
2480 free (w);
2481 // now do something you wanted to do when the program has
2482 // no longer anything immediate to do.
2483 }
2484 ev_idle *idle_watcher = malloc (sizeof (ev_idle));
2486 ev_idle_init (idle_watcher, idle_cb);
2487 ev_idle_start (loop, idle_watcher);
2488 "ev_prepare" and "ev_check" - customise your event loop!
2490 Prepare and check watchers are usually (but not always) used in pairs:
2491 prepare watchers get invoked before the process blocks and check
2492 watchers afterwards.
2493 You must not call "ev_loop" or similar functions that enter the current
2495 event loop from either "ev_prepare" or "ev_check" watchers. Other loops
2496 than the current one are fine, however. The rationale behind this is
2497 that you do not need to check for recursion in those watchers, i.e. the
2498 sequence will always be "ev_prepare", blocking, "ev_check" so if you
2499 have one watcher of each kind they will always be called in pairs
2500 bracketing the blocking call.
2501 Their main purpose is to integrate other event mechanisms into libev
2503 and their use is somewhat advanced. They could be used, for example, to
2504 track variable changes, implement your own watchers, integrate net-snmp
2505 or a coroutine library and lots more. They are also occasionally useful
2506 if you cache some data and want to flush it before blocking (for
2507 example, in X programs you might want to do an "XFlush ()" in an
2508 "ev_prepare" watcher).
2509 This is done by examining in each prepare call which file descriptors
2511 need to be watched by the other library, registering "ev_io" watchers
2512 for them and starting an "ev_timer" watcher for any timeouts (many
2513 libraries provide exactly this functionality). Then, in the check
2514 watcher, you check for any events that occurred (by checking the
2515 pending status of all watchers and stopping them) and call back into
2516 the library. The I/O and timer callbacks will never actually be called
2517 (but must be valid nevertheless, because you never know, you know?).
2518 As another example, the Perl Coro module uses these hooks to integrate
2520 coroutines into libev programs, by yielding to other active coroutines
2521 during each prepare and only letting the process block if no coroutines
2522 are ready to run (it's actually more complicated: it only runs
2523 coroutines with priority higher than or equal to the event loop and one
2524 coroutine of lower priority, but only once, using idle watchers to keep
2525 the event loop from blocking if lower-priority coroutines are active,
2526 thus mapping low-priority coroutines to idle/background tasks).
2527 It is recommended to give "ev_check" watchers highest ("EV_MAXPRI")
2529 priority, to ensure that they are being run before any other watchers
2530 after the poll (this doesn't matter for "ev_prepare" watchers).
2531 Also, "ev_check" watchers (and "ev_prepare" watchers, too) should not
2533 activate ("feed") events into libev. While libev fully supports this,
2534 they might get executed before other "ev_check" watchers did their job.
2535 As "ev_check" watchers are often used to embed other (non-libev) event
2536 loops those other event loops might be in an unusable state until their
2537 "ev_check" watcher ran (always remind yourself to coexist peacefully
2538 with others).
2539 Watcher-Specific Functions and Data Members
2541 ev_prepare_init (ev_prepare *, callback)
2543 ev_check_init (ev_check *, callback)
2544 Initialises and configures the prepare or check watcher - they have
2545 no parameters of any kind. There are "ev_prepare_set" and
2546 "ev_check_set" macros, but using them is utterly, utterly, utterly
2547 and completely pointless.
2548 Examples
2550 There are a number of principal ways to embed other event loops or
2552 modules into libev. Here are some ideas on how to include libadns into
2553 libev (there is a Perl module named "EV::ADNS" that does this, which
2554 you could use as a working example. Another Perl module named
2555 "EV::Glib" embeds a Glib main context into libev, and finally,
2556 "Glib::EV" embeds EV into the Glib event loop).
2557 Method 1: Add IO watchers and a timeout watcher in a prepare handler,
2559 and in a check watcher, destroy them and call into libadns. What
2560 follows is pseudo-code only of course. This requires you to either use
2561 a low priority for the check watcher or use "ev_clear_pending"
2562 explicitly, as the callbacks for the IO/timeout watchers might not have
2563 been called yet.
2564 static ev_io iow [nfd];
2566 static ev_timer tw;
2567 static void
2569 io_cb (struct ev_loop *loop, ev_io *w, int revents)
2570 {
2571 }
2572 // create io watchers for each fd and a timer before blocking
2574 static void
2575 adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
2576 {
2577 int timeout = 3600000;
2578 struct pollfd fds [nfd];
2579 // actual code will need to loop here and realloc etc.
2580 adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
2581 /* the callback is illegal, but won't be called as we stop during check */
2583 ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
2584 ev_timer_start (loop, &tw);
2585 // create one ev_io per pollfd
2587 for (int i = 0; i < nfd; ++i)
2588 {
2589 ev_io_init (iow + i, io_cb, fds [i].fd,
2590 ((fds [i].events & POLLIN ? EV_READ : 0)
2591 | (fds [i].events & POLLOUT ? EV_WRITE : 0)));
2592 fds [i].revents = 0;
2594 ev_io_start (loop, iow + i);
2595 }
2596 }
2597 // stop all watchers after blocking
2599 static void
2600 adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
2601 {
2602 ev_timer_stop (loop, &tw);
2603 for (int i = 0; i < nfd; ++i)
2605 {
2606 // set the relevant poll flags
2607 // could also call adns_processreadable etc. here
2608 struct pollfd *fd = fds + i;
2609 int revents = ev_clear_pending (iow + i);
2610 if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
2611 if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
2612 // now stop the watcher
2614 ev_io_stop (loop, iow + i);
2615 }
2616 adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
2618 }
2619 Method 2: This would be just like method 1, but you run
2621 "adns_afterpoll" in the prepare watcher and would dispose of the check
2622 watcher.
2623 Method 3: If the module to be embedded supports explicit event
2625 notification (libadns does), you can also make use of the actual
2626 watcher callbacks, and only destroy/create the watchers in the prepare
2627 watcher.
2628 static void
2630 timer_cb (EV_P_ ev_timer *w, int revents)
2631 {
2632 adns_state ads = (adns_state)w->data;
2633 update_now (EV_A);
2634 adns_processtimeouts (ads, &tv_now);
2636 }
2637 static void
2639 io_cb (EV_P_ ev_io *w, int revents)
2640 {
2641 adns_state ads = (adns_state)w->data;
2642 update_now (EV_A);
2643 if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
2645 if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
2646 }
2647 // do not ever call adns_afterpoll
2649 Method 4: Do not use a prepare or check watcher because the module you
2651 want to embed is not flexible enough to support it. Instead, you can
2652 override their poll function. The drawback with this solution is that
2653 the main loop is now no longer controllable by EV. The "Glib::EV"
2654 module uses this approach, effectively embedding EV as a client into
2655 the horrible libglib event loop.
2656 static gint
2658 event_poll_func (GPollFD *fds, guint nfds, gint timeout)
2659 {
2660 int got_events = 0;
2661 for (n = 0; n < nfds; ++n)
2663 // create/start io watcher that sets the relevant bits in fds[n] and increment got_events
2664 if (timeout >= 0)
2666 // create/start timer
2667 // poll
2669 ev_loop (EV_A_ 0);
2670 // stop timer again
2672 if (timeout >= 0)
2673 ev_timer_stop (EV_A_ &to);
2674 // stop io watchers again - their callbacks should have set
2676 for (n = 0; n < nfds; ++n)
2677 ev_io_stop (EV_A_ iow [n]);
2678 return got_events;
2680 }
2681 "ev_embed" - when one backend isn't enough...
2683 This is a rather advanced watcher type that lets you embed one event
2684 loop into another (currently only "ev_io" events are supported in the
2685 embedded loop, other types of watchers might be handled in a delayed or
2686 incorrect fashion and must not be used).
2687 There are primarily two reasons you would want that: work around bugs
2689 and prioritise I/O.
2690 As an example for a bug workaround, the kqueue backend might only
2692 support sockets on some platform, so it is unusable as generic backend,
2693 but you still want to make use of it because you have many sockets and
2694 it scales so nicely. In this case, you would create a kqueue-based loop
2695 and embed it into your default loop (which might use e.g. poll).
2696 Overall operation will be a bit slower because first libev has to call
2697 "poll" and then "kevent", but at least you can use both mechanisms for
2698 what they are best: "kqueue" for scalable sockets and "poll" if you
2699 want it to work :)
2700 As for prioritising I/O: under rare circumstances you have the case
2702 where some fds have to be watched and handled very quickly (with low
2703 latency), and even priorities and idle watchers might have too much
2704 overhead. In this case you would put all the high priority stuff in one
2705 loop and all the rest in a second one, and embed the second one in the
2706 first.
2707 As long as the watcher is active, the callback will be invoked every
2709 time there might be events pending in the embedded loop. The callback
2710 must then call "ev_embed_sweep (mainloop, watcher)" to make a single
2711 sweep and invoke their callbacks (the callback doesn't need to invoke
2712 the "ev_embed_sweep" function directly, it could also start an idle
2713 watcher to give the embedded loop strictly lower priority for example).
2714 You can also set the callback to 0, in which case the embed watcher
2716 will automatically execute the embedded loop sweep whenever necessary.
2717 Fork detection will be handled transparently while the "ev_embed"
2719 watcher is active, i.e., the embedded loop will automatically be forked
2720 when the embedding loop forks. In other cases, the user is responsible
2721 for calling "ev_loop_fork" on the embedded loop.
2722 Unfortunately, not all backends are embeddable: only the ones returned
2724 by "ev_embeddable_backends" are, which, unfortunately, does not include
2725 any portable one.
2726 So when you want to use this feature you will always have to be
2728 prepared that you cannot get an embeddable loop. The recommended way to
2729 get around this is to have a separate variables for your embeddable
2730 loop, try to create it, and if that fails, use the normal loop for
2731 everything.
2732 "ev_embed" and fork
2734 While the "ev_embed" watcher is running, forks in the embedding loop
2736 will automatically be applied to the embedded loop as well, so no
2737 special fork handling is required in that case. When the watcher is not
2738 running, however, it is still the task of the libev user to call
2739 "ev_loop_fork ()" as applicable.
2740 Watcher-Specific Functions and Data Members
2742 ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
2744 ev_embed_set (ev_embed *, callback, struct ev_loop *embedded_loop)
2745 Configures the watcher to embed the given loop, which must be
2746 embeddable. If the callback is 0, then "ev_embed_sweep" will be
2747 invoked automatically, otherwise it is the responsibility of the
2748 callback to invoke it (it will continue to be called until the
2749 sweep has been done, if you do not want that, you need to
2750 temporarily stop the embed watcher).
2751 ev_embed_sweep (loop, ev_embed *)
2753 Make a single, non-blocking sweep over the embedded loop. This
2754 works similarly to "ev_loop (embedded_loop, EVLOOP_NONBLOCK)", but
2755 in the most appropriate way for embedded loops.
2756 struct ev_loop *other [read-only]
2758 The embedded event loop.
2759 Examples
2761 Example: Try to get an embeddable event loop and embed it into the
2763 default event loop. If that is not possible, use the default loop. The
2764 default loop is stored in "loop_hi", while the embeddable loop is
2765 stored in "loop_lo" (which is "loop_hi" in the case no embeddable loop
2766 can be used).
2767 struct ev_loop *loop_hi = ev_default_init (0);
2769 struct ev_loop *loop_lo = 0;
2770 ev_embed embed;
2771 // see if there is a chance of getting one that works
2773 // (remember that a flags value of 0 means autodetection)
2774 loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
2775 ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
2776 : 0;
2777 // if we got one, then embed it, otherwise default to loop_hi
2779 if (loop_lo)
2780 {
2781 ev_embed_init (&embed, 0, loop_lo);
2782 ev_embed_start (loop_hi, &embed);
2783 }
2784 else
2785 loop_lo = loop_hi;
2786 Example: Check if kqueue is available but not recommended and create a
2788 kqueue backend for use with sockets (which usually work with any kqueue
2789 implementation). Store the kqueue/socket-only event loop in
2790 "loop_socket". (One might optionally use "EVFLAG_NOENV", too).
2791 struct ev_loop *loop = ev_default_init (0);
2793 struct ev_loop *loop_socket = 0;
2794 ev_embed embed;
2795 if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
2797 if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
2798 {
2799 ev_embed_init (&embed, 0, loop_socket);
2800 ev_embed_start (loop, &embed);
2801 }
2802 if (!loop_socket)
2804 loop_socket = loop;
2805 // now use loop_socket for all sockets, and loop for everything else
2807 "ev_fork" - the audacity to resume the event loop after a fork
2809 Fork watchers are called when a "fork ()" was detected (usually because
2810 whoever is a good citizen cared to tell libev about it by calling
2811 "ev_default_fork" or "ev_loop_fork"). The invocation is done before the
2812 event loop blocks next and before "ev_check" watchers are being called,
2813 and only in the child after the fork. If whoever good citizen calling
2814 "ev_default_fork" cheats and calls it in the wrong process, the fork
2815 handlers will be invoked, too, of course.
2816 The special problem of life after fork - how is it possible?
2818 Most uses of "fork()" consist of forking, then some simple calls to ste
2820 up/change the process environment, followed by a call to "exec()". This
2821 sequence should be handled by libev without any problems.
2822 This changes when the application actually wants to do event handling
2824 in the child, or both parent in child, in effect "continuing" after the
2825 fork.
2826 The default mode of operation (for libev, with application help to
2828 detect forks) is to duplicate all the state in the child, as would be
2829 expected when either the parent or the child process continues.
2830 When both processes want to continue using libev, then this is usually
2832 the wrong result. In that case, usually one process (typically the
2833 parent) is supposed to continue with all watchers in place as before,
2834 while the other process typically wants to start fresh, i.e. without
2835 any active watchers.
2836 The cleanest and most efficient way to achieve that with libev is to
2838 simply create a new event loop, which of course will be "empty", and
2839 use that for new watchers. This has the advantage of not touching more
2840 memory than necessary, and thus avoiding the copy-on-write, and the
2841 disadvantage of having to use multiple event loops (which do not
2842 support signal watchers).
2843 When this is not possible, or you want to use the default loop for
2845 other reasons, then in the process that wants to start "fresh", call
2846 "ev_default_destroy ()" followed by "ev_default_loop (...)". Destroying
2847 the default loop will "orphan" (not stop) all registered watchers, so
2848 you have to be careful not to execute code that modifies those
2849 watchers. Note also that in that case, you have to re-register any
2850 signal watchers.
2851 Watcher-Specific Functions and Data Members
2853 ev_fork_init (ev_signal *, callback)
2855 Initialises and configures the fork watcher - it has no parameters
2856 of any kind. There is a "ev_fork_set" macro, but using it is
2857 utterly pointless, believe me.
2858 "ev_async" - how to wake up another event loop
2860 In general, you cannot use an "ev_loop" from multiple threads or other
2861 asynchronous sources such as signal handlers (as opposed to multiple
2862 event loops - those are of course safe to use in different threads).
2863 Sometimes, however, you need to wake up another event loop you do not
2865 control, for example because it belongs to another thread. This is what
2866 "ev_async" watchers do: as long as the "ev_async" watcher is active,
2867 you can signal it by calling "ev_async_send", which is thread- and
2868 signal safe.
2869 This functionality is very similar to "ev_signal" watchers, as signals,
2871 too, are asynchronous in nature, and signals, too, will be compressed
2872 (i.e. the number of callback invocations may be less than the number of
2873 "ev_async_sent" calls).
2874 Unlike "ev_signal" watchers, "ev_async" works with any event loop, not
2876 just the default loop.
2877 Queueing
2879 "ev_async" does not support queueing of data in any way. The reason is
2881 that the author does not know of a simple (or any) algorithm for a
2882 multiple-writer-single-reader queue that works in all cases and doesn't
2883 need elaborate support such as pthreads or unportable memory access
2884 semantics.
2885 That means that if you want to queue data, you have to provide your own
2887 queue. But at least I can tell you how to implement locking around your
2888 queue:
2889 queueing from a signal handler context
2891 To implement race-free queueing, you simply add to the queue in the
2892 signal handler but you block the signal handler in the watcher
2893 callback. Here is an example that does that for some fictitious
2894 SIGUSR1 handler:
2895 static ev_async mysig;
2897 static void
2899 sigusr1_handler (void)
2900 {
2901 sometype data;
2902 // no locking etc.
2904 queue_put (data);
2905 ev_async_send (EV_DEFAULT_ &mysig);
2906 }
2907 static void
2909 mysig_cb (EV_P_ ev_async *w, int revents)
2910 {
2911 sometype data;
2912 sigset_t block, prev;
2913 sigemptyset (&block);
2915 sigaddset (&block, SIGUSR1);
2916 sigprocmask (SIG_BLOCK, &block, &prev);
2917 while (queue_get (&data))
2919 process (data);
2920 if (sigismember (&prev, SIGUSR1)
2922 sigprocmask (SIG_UNBLOCK, &block, 0);
2923 }
2924 (Note: pthreads in theory requires you to use "pthread_setmask"
2926 instead of "sigprocmask" when you use threads, but libev doesn't do
2927 it either...).
2928 queueing from a thread context
2930 The strategy for threads is different, as you cannot (easily) block
2931 threads but you can easily preempt them, so to queue safely you
2932 need to employ a traditional mutex lock, such as in this pthread
2933 example:
2934 static ev_async mysig;
2936 static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
2937 static void
2939 otherthread (void)
2940 {
2941 // only need to lock the actual queueing operation
2942 pthread_mutex_lock (&mymutex);
2943 queue_put (data);
2944 pthread_mutex_unlock (&mymutex);
2945 ev_async_send (EV_DEFAULT_ &mysig);
2947 }
2948 static void
2950 mysig_cb (EV_P_ ev_async *w, int revents)
2951 {
2952 pthread_mutex_lock (&mymutex);
2953 while (queue_get (&data))
2955 process (data);
2956 pthread_mutex_unlock (&mymutex);
2958 }
2959 Watcher-Specific Functions and Data Members
2961 ev_async_init (ev_async *, callback)
2963 Initialises and configures the async watcher - it has no parameters
2964 of any kind. There is a "ev_async_set" macro, but using it is
2965 utterly pointless, trust me.
2966 ev_async_send (loop, ev_async *)
2968 Sends/signals/activates the given "ev_async" watcher, that is,
2969 feeds an "EV_ASYNC" event on the watcher into the event loop.
2970 Unlike "ev_feed_event", this call is safe to do from other threads,
2971 signal or similar contexts (see the discussion of "EV_ATOMIC_T" in
2972 the embedding section below on what exactly this means).
2973 Note that, as with other watchers in libev, multiple events might
2975 get compressed into a single callback invocation (another way to
2976 look at this is that "ev_async" watchers are level-triggered, set
2977 on "ev_async_send", reset when the event loop detects that).
2978 This call incurs the overhead of a system call only once per event
2980 loop iteration, so while the overhead might be noticeable, it
2981 doesn't apply to repeated calls to "ev_async_send" for the same
2982 event loop.
2983 bool = ev_async_pending (ev_async *)
2985 Returns a non-zero value when "ev_async_send" has been called on
2986 the watcher but the event has not yet been processed (or even
2987 noted) by the event loop.
2988 "ev_async_send" sets a flag in the watcher and wakes up the loop.
2990 When the loop iterates next and checks for the watcher to have
2991 become active, it will reset the flag again. "ev_async_pending" can
2992 be used to very quickly check whether invoking the loop might be a
2993 good idea.
2994 Not that this does not check whether the watcher itself is pending,
2996 only whether it has been requested to make this watcher pending:
2997 there is a time window between the event loop checking and
2998 resetting the async notification, and the callback being invoked.
2999
3001 There are some other functions of possible interest. Described. Here.
3002 Now.
3003 ev_once (loop, int fd, int events, ev_tstamp timeout, callback)
3005 This function combines a simple timer and an I/O watcher, calls
3006 your callback on whichever event happens first and automatically
3007 stops both watchers. This is useful if you want to wait for a
3008 single event on an fd or timeout without having to
3009 allocate/configure/start/stop/free one or more watchers yourself.
3010 If "fd" is less than 0, then no I/O watcher will be started and the
3012 "events" argument is being ignored. Otherwise, an "ev_io" watcher
3013 for the given "fd" and "events" set will be created and started.
3014 If "timeout" is less than 0, then no timeout watcher will be
3016 started. Otherwise an "ev_timer" watcher with after = "timeout"
3017 (and repeat = 0) will be started. 0 is a valid timeout.
3018 The callback has the type "void (*cb)(int revents, void *arg)" and
3020 gets passed an "revents" set like normal event callbacks (a
3021 combination of "EV_ERROR", "EV_READ", "EV_WRITE" or "EV_TIMEOUT")
3022 and the "arg" value passed to "ev_once". Note that it is possible
3023 to receive both a timeout and an io event at the same time - you
3024 probably should give io events precedence.
3025 Example: wait up to ten seconds for data to appear on STDIN_FILENO.
3027 static void stdin_ready (int revents, void *arg)
3029 {
3030 if (revents & EV_READ)
3031 /* stdin might have data for us, joy! */;
3032 else if (revents & EV_TIMEOUT)
3033 /* doh, nothing entered */;
3034 }
3035 ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
3037 ev_feed_fd_event (loop, int fd, int revents)
3039 Feed an event on the given fd, as if a file descriptor backend
3040 detected the given events it.
3041 ev_feed_signal_event (loop, int signum)
3043 Feed an event as if the given signal occurred ("loop" must be the
3044 default loop!).
3045
3047 Libev offers a compatibility emulation layer for libevent. It cannot
3048 emulate the internals of libevent, so here are some usage hints:
3049 · Use it by including <event.h>, as usual.
3051 · The following members are fully supported: ev_base, ev_callback,
3053 ev_arg, ev_fd, ev_res, ev_events.
3054 · Avoid using ev_flags and the EVLIST_*-macros, while it is
3056 maintained by libev, it does not work exactly the same way as in
3057 libevent (consider it a private API).
3058 · Priorities are not currently supported. Initialising priorities
3060 will fail and all watchers will have the same priority, even though
3061 there is an ev_pri field.
3062 · In libevent, the last base created gets the signals, in libev, the
3064 first base created (== the default loop) gets the signals.
3065 · Other members are not supported.
3067 · The libev emulation is not ABI compatible to libevent, you need to
3069 use the libev header file and library.
3070
3072 Libev comes with some simplistic wrapper classes for C++ that mainly
3073 allow you to use some convenience methods to start/stop watchers and
3074 also change the callback model to a model using method callbacks on
3075 objects.
3076 To use it,
3078 #include <ev++.h>
3080 This automatically includes ev.h and puts all of its definitions (many
3082 of them macros) into the global namespace. All C++ specific things are
3083 put into the "ev" namespace. It should support all the same embedding
3084 options as ev.h, most notably "EV_MULTIPLICITY".
3085 Care has been taken to keep the overhead low. The only data member the
3087 C++ classes add (compared to plain C-style watchers) is the event loop
3088 pointer that the watcher is associated with (or no additional members
3089 at all if you disable "EV_MULTIPLICITY" when embedding libev).
3090 Currently, functions, and static and non-static member functions can be
3092 used as callbacks. Other types should be easy to add as long as they
3093 only need one additional pointer for context. If you need support for
3094 other types of functors please contact the author (preferably after
3095 implementing it).
3096 Here is a list of things available in the "ev" namespace:
3098 "ev::READ", "ev::WRITE" etc.
3100 These are just enum values with the same values as the "EV_READ"
3101 etc. macros from ev.h.
3102 "ev::tstamp", "ev::now"
3104 Aliases to the same types/functions as with the "ev_" prefix.
3105 "ev::io", "ev::timer", "ev::periodic", "ev::idle", "ev::sig" etc.
3107 For each "ev_TYPE" watcher in ev.h there is a corresponding class
3108 of the same name in the "ev" namespace, with the exception of
3109 "ev_signal" which is called "ev::sig" to avoid clashes with the
3110 "signal" macro defines by many implementations.
3111 All of those classes have these methods:
3113 ev::TYPE::TYPE ()
3115 ev::TYPE::TYPE (loop)
3116 ev::TYPE::~TYPE
3117 The constructor (optionally) takes an event loop to associate
3118 the watcher with. If it is omitted, it will use "EV_DEFAULT".
3119 The constructor calls "ev_init" for you, which means you have
3121 to call the "set" method before starting it.
3122 It will not set a callback, however: You have to call the
3124 templated "set" method to set a callback before you can start
3125 the watcher.
3126 (The reason why you have to use a method is a limitation in C++
3128 which does not allow explicit template arguments for
3129 constructors).
3130 The destructor automatically stops the watcher if it is active.
3132 w->set<class, &class::method> (object *)
3134 This method sets the callback method to call. The method has to
3135 have a signature of "void (*)(ev_TYPE &, int)", it receives the
3136 watcher as first argument and the "revents" as second. The
3137 object must be given as parameter and is stored in the "data"
3138 member of the watcher.
3139 This method synthesizes efficient thunking code to call your
3141 method from the C callback that libev requires. If your
3142 compiler can inline your callback (i.e. it is visible to it at
3143 the place of the "set" call and your compiler is good :), then
3144 the method will be fully inlined into the thunking function,
3145 making it as fast as a direct C callback.
3146 Example: simple class declaration and watcher initialisation
3148 struct myclass
3150 {
3151 void io_cb (ev::io &w, int revents) { }
3152 }
3153 myclass obj;
3155 ev::io iow;
3156 iow.set <myclass, &myclass::io_cb> (&obj);
3157 w->set (object *)
3159 This is an experimental feature that might go away in a future
3160 version.
3161 This is a variation of a method callback - leaving out the
3163 method to call will default the method to "operator ()", which
3164 makes it possible to use functor objects without having to
3165 manually specify the "operator ()" all the time. Incidentally,
3166 you can then also leave out the template argument list.
3167 The "operator ()" method prototype must be "void operator
3169 ()(watcher &w, int revents)".
3170 See the method-"set" above for more details.
3172 Example: use a functor object as callback.
3174 struct myfunctor
3176 {
3177 void operator() (ev::io &w, int revents)
3178 {
3179 ...
3180 }
3181 }
3182 myfunctor f;
3184 ev::io w;
3186 w.set (&f);
3187 w->set<function> (void *data = 0)
3189 Also sets a callback, but uses a static method or plain
3190 function as callback. The optional "data" argument will be
3191 stored in the watcher's "data" member and is free for you to
3192 use.
3193 The prototype of the "function" must be "void (*)(ev::TYPE &w,
3195 int)".
3196 See the method-"set" above for more details.
3198 Example: Use a plain function as callback.
3200 static void io_cb (ev::io &w, int revents) { }
3202 iow.set <io_cb> ();
3203 w->set (loop)
3205 Associates a different "struct ev_loop" with this watcher. You
3206 can only do this when the watcher is inactive (and not pending
3207 either).
3208 w->set ([arguments])
3210 Basically the same as "ev_TYPE_set", with the same arguments.
3211 Must be called at least once. Unlike the C counterpart, an
3212 active watcher gets automatically stopped and restarted when
3213 reconfiguring it with this method.
3214 w->start ()
3216 Starts the watcher. Note that there is no "loop" argument, as
3217 the constructor already stores the event loop.
3218 w->stop ()
3220 Stops the watcher if it is active. Again, no "loop" argument.
3221 w->again () ("ev::timer", "ev::periodic" only)
3223 For "ev::timer" and "ev::periodic", this invokes the
3224 corresponding "ev_TYPE_again" function.
3225 w->sweep () ("ev::embed" only)
3227 Invokes "ev_embed_sweep".
3228 w->update () ("ev::stat" only)
3230 Invokes "ev_stat_stat".
3231 Example: Define a class with an IO and idle watcher, start one of them
3233 in the constructor.
3234 class myclass
3236 {
3237 ev::io io ; void io_cb (ev::io &w, int revents);
3238 ev::idle idle; void idle_cb (ev::idle &w, int revents);
3239 myclass (int fd)
3241 {
3242 io .set <myclass, &myclass::io_cb > (this);
3243 idle.set <myclass, &myclass::idle_cb> (this);
3244 io.start (fd, ev::READ);
3246 }
3247 };
3248
3250 Libev does not offer other language bindings itself, but bindings for a
3251 number of languages exist in the form of third-party packages. If you
3252 know any interesting language binding in addition to the ones listed
3253 here, drop me a note.
3254 Perl
3256 The EV module implements the full libev API and is actually used to
3257 test libev. EV is developed together with libev. Apart from the EV
3258 core module, there are additional modules that implement libev-
3259 compatible interfaces to "libadns" ("EV::ADNS", but "AnyEvent::DNS"
3260 is preferred nowadays), "Net::SNMP" ("Net::SNMP::EV") and the
3261 "libglib" event core ("Glib::EV" and "EV::Glib").
3262 It can be found and installed via CPAN, its homepage is at
3264 <http://software.schmorp.de/pkg/EV>.
3265 Python
3267 Python bindings can be found at <http://code.google.com/p/pyev/>.
3268 It seems to be quite complete and well-documented.
3269 Ruby
3271 Tony Arcieri has written a ruby extension that offers access to a
3272 subset of the libev API and adds file handle abstractions,
3273 asynchronous DNS and more on top of it. It can be found via gem
3274 servers. Its homepage is at <http://rev.rubyforge.org/>.
3275 Roger Pack reports that using the link order "-lws2_32
3277 -lmsvcrt-ruby-190" makes rev work even on mingw.
3278 Haskell
3280 A haskell binding to libev is available at
3281 <http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
3282 D Leandro Lucarella has written a D language binding (ev.d) for
3284 libev, to be found at <http://proj.llucax.com.ar/wiki/evd>.
3285 Ocaml
3287 Erkki Seppala has written Ocaml bindings for libev, to be found at
3288 <http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
3289 Lua Brian Maher has written a partial interface to libev for lua (only
3291 "ev_io" and "ev_timer"), to be found at
3292 <http://github.com/brimworks/lua-ev>.
3293
3295 Libev can be compiled with a variety of options, the most fundamental
3296 of which is "EV_MULTIPLICITY". This option determines whether (most)
3297 functions and callbacks have an initial "struct ev_loop *" argument.
3298 To make it easier to write programs that cope with either variant, the
3300 following macros are defined:
3301 "EV_A", "EV_A_"
3303 This provides the loop argument for functions, if one is required
3304 ("ev loop argument"). The "EV_A" form is used when this is the sole
3305 argument, "EV_A_" is used when other arguments are following.
3306 Example:
3307 ev_unref (EV_A);
3309 ev_timer_add (EV_A_ watcher);
3310 ev_loop (EV_A_ 0);
3311 It assumes the variable "loop" of type "struct ev_loop *" is in
3313 scope, which is often provided by the following macro.
3314 "EV_P", "EV_P_"
3316 This provides the loop parameter for functions, if one is required
3317 ("ev loop parameter"). The "EV_P" form is used when this is the
3318 sole parameter, "EV_P_" is used when other parameters are
3319 following. Example:
3320 // this is how ev_unref is being declared
3322 static void ev_unref (EV_P);
3323 // this is how you can declare your typical callback
3325 static void cb (EV_P_ ev_timer *w, int revents)
3326 It declares a parameter "loop" of type "struct ev_loop *", quite
3328 suitable for use with "EV_A".
3329 "EV_DEFAULT", "EV_DEFAULT_"
3331 Similar to the other two macros, this gives you the value of the
3332 default loop, if multiple loops are supported ("ev loop default").
3333 "EV_DEFAULT_UC", "EV_DEFAULT_UC_"
3335 Usage identical to "EV_DEFAULT" and "EV_DEFAULT_", but requires
3336 that the default loop has been initialised ("UC" == unchecked).
3337 Their behaviour is undefined when the default loop has not been
3338 initialised by a previous execution of "EV_DEFAULT", "EV_DEFAULT_"
3339 or "ev_default_init (...)".
3340 It is often prudent to use "EV_DEFAULT" when initialising the first
3342 watcher in a function but use "EV_DEFAULT_UC" afterwards.
3343 Example: Declare and initialise a check watcher, utilising the above
3345 macros so it will work regardless of whether multiple loops are
3346 supported or not.
3347 static void
3349 check_cb (EV_P_ ev_timer *w, int revents)
3350 {
3351 ev_check_stop (EV_A_ w);
3352 }
3353 ev_check check;
3355 ev_check_init (&check, check_cb);
3356 ev_check_start (EV_DEFAULT_ &check);
3357 ev_loop (EV_DEFAULT_ 0);
3358
3360 Libev can (and often is) directly embedded into host applications.
3361 Examples of applications that embed it include the Deliantra Game
3362 Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe) and
3363 rxvt-unicode.
3364 The goal is to enable you to just copy the necessary files into your
3366 source directory without having to change even a single line in them,
3367 so you can easily upgrade by simply copying (or having a checked-out
3368 copy of libev somewhere in your source tree).
3369 FILESETS
3371 Depending on what features you need you need to include one or more
3372 sets of files in your application.
3373 CORE EVENT LOOP
3375 To include only the libev core (all the "ev_*" functions), with manual
3377 configuration (no autoconf):
3378 #define EV_STANDALONE 1
3380 #include "ev.c"
3381 This will automatically include ev.h, too, and should be done in a
3383 single C source file only to provide the function implementations. To
3384 use it, do the same for ev.h in all files wishing to use this API (best
3385 done by writing a wrapper around ev.h that you can include instead and
3386 where you can put other configuration options):
3387 #define EV_STANDALONE 1
3389 #include "ev.h"
3390 Both header files and implementation files can be compiled with a C++
3392 compiler (at least, that's a stated goal, and breakage will be treated
3393 as a bug).
3394 You need the following files in your source tree, or in a directory in
3396 your include path (e.g. in libev/ when using -Ilibev):
3397 ev.h
3399 ev.c
3400 ev_vars.h
3401 ev_wrap.h
3402 ev_win32.c required on win32 platforms only
3404 ev_select.c only when select backend is enabled (which is enabled by default)
3406 ev_poll.c only when poll backend is enabled (disabled by default)
3407 ev_epoll.c only when the epoll backend is enabled (disabled by default)
3408 ev_kqueue.c only when the kqueue backend is enabled (disabled by default)
3409 ev_port.c only when the solaris port backend is enabled (disabled by default)
3410 ev.c includes the backend files directly when enabled, so you only need
3412 to compile this single file.
3413 LIBEVENT COMPATIBILITY API
3415 To include the libevent compatibility API, also include:
3417 #include "event.c"
3419 in the file including ev.c, and:
3421 #include "event.h"
3423 in the files that want to use the libevent API. This also includes
3425 ev.h.
3426 You need the following additional files for this:
3428 event.h
3430 event.c
3431 AUTOCONF SUPPORT
3433 Instead of using "EV_STANDALONE=1" and providing your configuration in
3435 whatever way you want, you can also "m4_include([libev.m4])" in your
3436 configure.ac and leave "EV_STANDALONE" undefined. ev.c will then
3437 include config.h and configure itself accordingly.
3438 For this of course you need the m4 file:
3440 libev.m4
3442 PREPROCESSOR SYMBOLS/MACROS
3444 Libev can be configured via a variety of preprocessor symbols you have
3445 to define before including any of its files. The default in the absence
3446 of autoconf is documented for every option.
3447 EV_STANDALONE
3449 Must always be 1 if you do not use autoconf configuration, which
3450 keeps libev from including config.h, and it also defines dummy
3451 implementations for some libevent functions (such as logging, which
3452 is not supported). It will also not define any of the structs
3453 usually found in event.h that are not directly supported by the
3454 libev core alone.
3455 In standalone mode, libev will still try to automatically deduce
3457 the configuration, but has to be more conservative.
3458 EV_USE_MONOTONIC
3460 If defined to be 1, libev will try to detect the availability of
3461 the monotonic clock option at both compile time and runtime.
3462 Otherwise no use of the monotonic clock option will be attempted.
3463 If you enable this, you usually have to link against librt or
3464 something similar. Enabling it when the functionality isn't
3465 available is safe, though, although you have to make sure you link
3466 against any libraries where the "clock_gettime" function is hiding
3467 in (often -lrt). See also "EV_USE_CLOCK_SYSCALL".
3468 EV_USE_REALTIME
3470 If defined to be 1, libev will try to detect the availability of
3471 the real-time clock option at compile time (and assume its
3472 availability at runtime if successful). Otherwise no use of the
3473 real-time clock option will be attempted. This effectively replaces
3474 "gettimeofday" by "clock_get (CLOCK_REALTIME, ...)" and will not
3475 normally affect correctness. See the note about libraries in the
3476 description of "EV_USE_MONOTONIC", though. Defaults to the opposite
3477 value of "EV_USE_CLOCK_SYSCALL".
3478 EV_USE_CLOCK_SYSCALL
3480 If defined to be 1, libev will try to use a direct syscall instead
3481 of calling the system-provided "clock_gettime" function. This
3482 option exists because on GNU/Linux, "clock_gettime" is in "librt",
3483 but "librt" unconditionally pulls in "libpthread", slowing down
3484 single-threaded programs needlessly. Using a direct syscall is
3485 slightly slower (in theory), because no optimised vdso
3486 implementation can be used, but avoids the pthread dependency.
3487 Defaults to 1 on GNU/Linux with glibc 2.x or higher, as it
3488 simplifies linking (no need for "-lrt").
3489 EV_USE_NANOSLEEP
3491 If defined to be 1, libev will assume that "nanosleep ()" is
3492 available and will use it for delays. Otherwise it will use "select
3493 ()".
3494 EV_USE_EVENTFD
3496 If defined to be 1, then libev will assume that "eventfd ()" is
3497 available and will probe for kernel support at runtime. This will
3498 improve "ev_signal" and "ev_async" performance and reduce resource
3499 consumption. If undefined, it will be enabled if the headers
3500 indicate GNU/Linux + Glibc 2.7 or newer, otherwise disabled.
3501 EV_USE_SELECT
3503 If undefined or defined to be 1, libev will compile in support for
3504 the "select"(2) backend. No attempt at auto-detection will be done:
3505 if no other method takes over, select will be it. Otherwise the
3506 select backend will not be compiled in.
3507 EV_SELECT_USE_FD_SET
3509 If defined to 1, then the select backend will use the system
3510 "fd_set" structure. This is useful if libev doesn't compile due to
3511 a missing "NFDBITS" or "fd_mask" definition or it mis-guesses the
3512 bitset layout on exotic systems. This usually limits the range of
3513 file descriptors to some low limit such as 1024 or might have other
3514 limitations (winsocket only allows 64 sockets). The "FD_SETSIZE"
3515 macro, set before compilation, configures the maximum size of the
3516 "fd_set".
3517 EV_SELECT_IS_WINSOCKET
3519 When defined to 1, the select backend will assume that
3520 select/socket/connect etc. don't understand file descriptors but
3521 wants osf handles on win32 (this is the case when the select to be
3522 used is the winsock select). This means that it will call
3523 "_get_osfhandle" on the fd to convert it to an OS handle.
3524 Otherwise, it is assumed that all these functions actually work on
3525 fds, even on win32. Should not be defined on non-win32 platforms.
3526 EV_FD_TO_WIN32_HANDLE(fd)
3528 If "EV_SELECT_IS_WINSOCKET" is enabled, then libev needs a way to
3529 map file descriptors to socket handles. When not defining this
3530 symbol (the default), then libev will call "_get_osfhandle", which
3531 is usually correct. In some cases, programs use their own file
3532 descriptor management, in which case they can provide this function
3533 to map fds to socket handles.
3534 EV_WIN32_HANDLE_TO_FD(handle)
3536 If "EV_SELECT_IS_WINSOCKET" then libev maps handles to file
3537 descriptors using the standard "_open_osfhandle" function. For
3538 programs implementing their own fd to handle mapping, overwriting
3539 this function makes it easier to do so. This can be done by
3540 defining this macro to an appropriate value.
3541 EV_WIN32_CLOSE_FD(fd)
3543 If programs implement their own fd to handle mapping on win32, then
3544 this macro can be used to override the "close" function, useful to
3545 unregister file descriptors again. Note that the replacement
3546 function has to close the underlying OS handle.
3547 EV_USE_POLL
3549 If defined to be 1, libev will compile in support for the "poll"(2)
3550 backend. Otherwise it will be enabled on non-win32 platforms. It
3551 takes precedence over select.
3552 EV_USE_EPOLL
3554 If defined to be 1, libev will compile in support for the Linux
3555 "epoll"(7) backend. Its availability will be detected at runtime,
3556 otherwise another method will be used as fallback. This is the
3557 preferred backend for GNU/Linux systems. If undefined, it will be
3558 enabled if the headers indicate GNU/Linux + Glibc 2.4 or newer,
3559 otherwise disabled.
3560 EV_USE_KQUEUE
3562 If defined to be 1, libev will compile in support for the BSD style
3563 "kqueue"(2) backend. Its actual availability will be detected at
3564 runtime, otherwise another method will be used as fallback. This is
3565 the preferred backend for BSD and BSD-like systems, although on
3566 most BSDs kqueue only supports some types of fds correctly (the
3567 only platform we found that supports ptys for example was NetBSD),
3568 so kqueue might be compiled in, but not be used unless explicitly
3569 requested. The best way to use it is to find out whether kqueue
3570 supports your type of fd properly and use an embedded kqueue loop.
3571 EV_USE_PORT
3573 If defined to be 1, libev will compile in support for the Solaris
3574 10 port style backend. Its availability will be detected at
3575 runtime, otherwise another method will be used as fallback. This is
3576 the preferred backend for Solaris 10 systems.
3577 EV_USE_DEVPOLL
3579 Reserved for future expansion, works like the USE symbols above.
3580 EV_USE_INOTIFY
3582 If defined to be 1, libev will compile in support for the Linux
3583 inotify interface to speed up "ev_stat" watchers. Its actual
3584 availability will be detected at runtime. If undefined, it will be
3585 enabled if the headers indicate GNU/Linux + Glibc 2.4 or newer,
3586 otherwise disabled.
3587 EV_ATOMIC_T
3589 Libev requires an integer type (suitable for storing 0 or 1) whose
3590 access is atomic with respect to other threads or signal contexts.
3591 No such type is easily found in the C language, so you can provide
3592 your own type that you know is safe for your purposes. It is used
3593 both for signal handler "locking" as well as for signal and thread
3594 safety in "ev_async" watchers.
3595 In the absence of this define, libev will use "sig_atomic_t
3597 volatile" (from signal.h), which is usually good enough on most
3598 platforms.
3599 EV_H
3601 The name of the ev.h header file used to include it. The default if
3602 undefined is "ev.h" in event.h, ev.c and ev++.h. This can be used
3603 to virtually rename the ev.h header file in case of conflicts.
3604 EV_CONFIG_H
3606 If "EV_STANDALONE" isn't 1, this variable can be used to override
3607 ev.c's idea of where to find the config.h file, similarly to
3608 "EV_H", above.
3609 EV_EVENT_H
3611 Similarly to "EV_H", this macro can be used to override event.c's
3612 idea of how the event.h header can be found, the default is
3613 "event.h".
3614 EV_PROTOTYPES
3616 If defined to be 0, then ev.h will not define any function
3617 prototypes, but still define all the structs and other symbols.
3618 This is occasionally useful if you want to provide your own wrapper
3619 functions around libev functions.
3620 EV_MULTIPLICITY
3622 If undefined or defined to 1, then all event-loop-specific
3623 functions will have the "struct ev_loop *" as first argument, and
3624 you can create additional independent event loops. Otherwise there
3625 will be no support for multiple event loops and there is no first
3626 event loop pointer argument. Instead, all functions act on the
3627 single default loop.
3628 EV_MINPRI
3630 EV_MAXPRI
3631 The range of allowed priorities. "EV_MINPRI" must be smaller or
3632 equal to "EV_MAXPRI", but otherwise there are no non-obvious
3633 limitations. You can provide for more priorities by overriding
3634 those symbols (usually defined to be "-2" and 2, respectively).
3635 When doing priority-based operations, libev usually has to linearly
3637 search all the priorities, so having many of them (hundreds) uses a
3638 lot of space and time, so using the defaults of five priorities (-2
3639 .. +2) is usually fine.
3640 If your embedding application does not need any priorities,
3642 defining these both to 0 will save some memory and CPU.
3643 EV_PERIODIC_ENABLE
3645 If undefined or defined to be 1, then periodic timers are
3646 supported. If defined to be 0, then they are not. Disabling them
3647 saves a few kB of code.
3648 EV_IDLE_ENABLE
3650 If undefined or defined to be 1, then idle watchers are supported.
3651 If defined to be 0, then they are not. Disabling them saves a few
3652 kB of code.
3653 EV_EMBED_ENABLE
3655 If undefined or defined to be 1, then embed watchers are supported.
3656 If defined to be 0, then they are not. Embed watchers rely on most
3657 other watcher types, which therefore must not be disabled.
3658 EV_STAT_ENABLE
3660 If undefined or defined to be 1, then stat watchers are supported.
3661 If defined to be 0, then they are not.
3662 EV_FORK_ENABLE
3664 If undefined or defined to be 1, then fork watchers are supported.
3665 If defined to be 0, then they are not.
3666 EV_ASYNC_ENABLE
3668 If undefined or defined to be 1, then async watchers are supported.
3669 If defined to be 0, then they are not.
3670 EV_MINIMAL
3672 If you need to shave off some kilobytes of code at the expense of
3673 some speed (but with the full API), define this symbol to 1.
3674 Currently this is used to override some inlining decisions, saves
3675 roughly 30% code size on amd64. It also selects a much smaller
3676 2-heap for timer management over the default 4-heap.
3677 You can save even more by disabling watcher types you do not need
3679 and setting "EV_MAXPRI" == "EV_MINPRI". Also, disabling "assert"
3680 ("-DNDEBUG") will usually reduce code size a lot.
3681 Defining "EV_MINIMAL" to 2 will additionally reduce the core API to
3683 provide a bare-bones event library. See "ev.h" for details on what
3684 parts of the API are still available, and do not complain if this
3685 subset changes over time.
3686 EV_NSIG
3688 The highest supported signal number, +1 (or, the number of
3689 signals): Normally, libev tries to deduce the maximum number of
3690 signals automatically, but sometimes this fails, in which case it
3691 can be specified. Also, using a lower number than detected (32
3692 should be good for about any system in existance) can save some
3693 memory, as libev statically allocates some 12-24 bytes per signal
3694 number.
3695 EV_PID_HASHSIZE
3697 "ev_child" watchers use a small hash table to distribute workload
3698 by pid. The default size is 16 (or 1 with "EV_MINIMAL"), usually
3699 more than enough. If you need to manage thousands of children you
3700 might want to increase this value (must be a power of two).
3701 EV_INOTIFY_HASHSIZE
3703 "ev_stat" watchers use a small hash table to distribute workload by
3704 inotify watch id. The default size is 16 (or 1 with "EV_MINIMAL"),
3705 usually more than enough. If you need to manage thousands of
3706 "ev_stat" watchers you might want to increase this value (must be a
3707 power of two).
3708 EV_USE_4HEAP
3710 Heaps are not very cache-efficient. To improve the cache-efficiency
3711 of the timer and periodics heaps, libev uses a 4-heap when this
3712 symbol is defined to 1. The 4-heap uses more complicated (longer)
3713 code but has noticeably faster performance with many (thousands) of
3714 watchers.
3715 The default is 1 unless "EV_MINIMAL" is set in which case it is 0
3717 (disabled).
3718 EV_HEAP_CACHE_AT
3720 Heaps are not very cache-efficient. To improve the cache-efficiency
3721 of the timer and periodics heaps, libev can cache the timestamp
3722 (at) within the heap structure (selected by defining
3723 "EV_HEAP_CACHE_AT" to 1), which uses 8-12 bytes more per watcher
3724 and a few hundred bytes more code, but avoids random read accesses
3725 on heap changes. This improves performance noticeably with many
3726 (hundreds) of watchers.
3727 The default is 1 unless "EV_MINIMAL" is set in which case it is 0
3729 (disabled).
3730 EV_VERIFY
3732 Controls how much internal verification (see "ev_loop_verify ()")
3733 will be done: If set to 0, no internal verification code will be
3734 compiled in. If set to 1, then verification code will be compiled
3735 in, but not called. If set to 2, then the internal verification
3736 code will be called once per loop, which can slow down libev. If
3737 set to 3, then the verification code will be called very
3738 frequently, which will slow down libev considerably.
3739 The default is 1, unless "EV_MINIMAL" is set, in which case it will
3741 be 0.
3742 EV_COMMON
3744 By default, all watchers have a "void *data" member. By redefining
3745 this macro to a something else you can include more and other types
3746 of members. You have to define it each time you include one of the
3747 files, though, and it must be identical each time.
3748 For example, the perl EV module uses something like this:
3750 #define EV_COMMON \
3752 SV *self; /* contains this struct */ \
3753 SV *cb_sv, *fh /* note no trailing ";" */
3754 EV_CB_DECLARE (type)
3756 EV_CB_INVOKE (watcher, revents)
3757 ev_set_cb (ev, cb)
3758 Can be used to change the callback member declaration in each
3759 watcher, and the way callbacks are invoked and set. Must expand to
3760 a struct member definition and a statement, respectively. See the
3761 ev.h header file for their default definitions. One possible use
3762 for overriding these is to avoid the "struct ev_loop *" as first
3763 argument in all cases, or to use method calls instead of plain
3764 function calls in C++.
3765 EXPORTED API SYMBOLS
3767 If you need to re-export the API (e.g. via a DLL) and you need a list
3768 of exported symbols, you can use the provided Symbol.* files which list
3769 all public symbols, one per line:
3770 Symbols.ev for libev proper
3772 Symbols.event for the libevent emulation
3773 This can also be used to rename all public symbols to avoid clashes
3775 with multiple versions of libev linked together (which is obviously bad
3776 in itself, but sometimes it is inconvenient to avoid this).
3777 A sed command like this will create wrapper "#define"'s that you need
3779 to include before including ev.h:
3780 <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
3782 This would create a file wrap.h which essentially looks like this:
3784 #define ev_backend myprefix_ev_backend
3786 #define ev_check_start myprefix_ev_check_start
3787 #define ev_check_stop myprefix_ev_check_stop
3788 ...
3789 EXAMPLES
3791 For a real-world example of a program the includes libev verbatim, you
3792 can have a look at the EV perl module
3793 (<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
3794 the libev/ subdirectory and includes them in the EV/EVAPI.h (public
3795 interface) and EV.xs (implementation) files. Only the EV.xs file will
3796 be compiled. It is pretty complex because it provides its own header
3797 file.
3798 The usage in rxvt-unicode is simpler. It has a ev_cpp.h header file
3800 that everybody includes and which overrides some configure choices:
3801 #define EV_MINIMAL 1
3803 #define EV_USE_POLL 0
3804 #define EV_MULTIPLICITY 0
3805 #define EV_PERIODIC_ENABLE 0
3806 #define EV_STAT_ENABLE 0
3807 #define EV_FORK_ENABLE 0
3808 #define EV_CONFIG_H <config.h>
3809 #define EV_MINPRI 0
3810 #define EV_MAXPRI 0
3811 #include "ev++.h"
3813 And a ev_cpp.C implementation file that contains libev proper and is
3815 compiled:
3816 #include "ev_cpp.h"
3818 #include "ev.c"
3819
3821 THREADS AND COROUTINES
3822 THREADS
3823 All libev functions are reentrant and thread-safe unless explicitly
3825 documented otherwise, but libev implements no locking itself. This
3826 means that you can use as many loops as you want in parallel, as long
3827 as there are no concurrent calls into any libev function with the same
3828 loop parameter ("ev_default_*" calls have an implicit default loop
3829 parameter, of course): libev guarantees that different event loops
3830 share no data structures that need any locking.
3831 Or to put it differently: calls with different loop parameters can be
3833 done concurrently from multiple threads, calls with the same loop
3834 parameter must be done serially (but can be done from different
3835 threads, as long as only one thread ever is inside a call at any point
3836 in time, e.g. by using a mutex per loop).
3837 Specifically to support threads (and signal handlers), libev implements
3839 so-called "ev_async" watchers, which allow some limited form of
3840 concurrency on the same event loop, namely waking it up "from the
3841 outside".
3842 If you want to know which design (one loop, locking, or multiple loops
3844 without or something else still) is best for your problem, then I
3845 cannot help you, but here is some generic advice:
3846 · most applications have a main thread: use the default libev loop in
3848 that thread, or create a separate thread running only the default
3849 loop.
3850 This helps integrating other libraries or software modules that use
3852 libev themselves and don't care/know about threading.
3853 · one loop per thread is usually a good model.
3855 Doing this is almost never wrong, sometimes a better-performance
3857 model exists, but it is always a good start.
3858 · other models exist, such as the leader/follower pattern, where one
3860 loop is handed through multiple threads in a kind of round-robin
3861 fashion.
3862 Choosing a model is hard - look around, learn, know that usually
3864 you can do better than you currently do :-)
3865 · often you need to talk to some other thread which blocks in the
3867 event loop.
3868 "ev_async" watchers can be used to wake them up from other threads
3870 safely (or from signal contexts...).
3871 An example use would be to communicate signals or other events that
3873 only work in the default loop by registering the signal watcher
3874 with the default loop and triggering an "ev_async" watcher from the
3875 default loop watcher callback into the event loop interested in the
3876 signal.
3877 THREAD LOCKING EXAMPLE
3879 Here is a fictitious example of how to run an event loop in a different
3881 thread than where callbacks are being invoked and watchers are
3882 created/added/removed.
3883 For a real-world example, see the "EV::Loop::Async" perl module, which
3885 uses exactly this technique (which is suited for many high-level
3886 languages).
3887 The example uses a pthread mutex to protect the loop data, a condition
3889 variable to wait for callback invocations, an async watcher to notify
3890 the event loop thread and an unspecified mechanism to wake up the main
3891 thread.
3892 First, you need to associate some data with the event loop:
3894 typedef struct {
3896 mutex_t lock; /* global loop lock */
3897 ev_async async_w;
3898 thread_t tid;
3899 cond_t invoke_cv;
3900 } userdata;
3901 void prepare_loop (EV_P)
3903 {
3904 // for simplicity, we use a static userdata struct.
3905 static userdata u;
3906 ev_async_init (&u->async_w, async_cb);
3908 ev_async_start (EV_A_ &u->async_w);
3909 pthread_mutex_init (&u->lock, 0);
3911 pthread_cond_init (&u->invoke_cv, 0);
3912 // now associate this with the loop
3914 ev_set_userdata (EV_A_ u);
3915 ev_set_invoke_pending_cb (EV_A_ l_invoke);
3916 ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
3917 // then create the thread running ev_loop
3919 pthread_create (&u->tid, 0, l_run, EV_A);
3920 }
3921 The callback for the "ev_async" watcher does nothing: the watcher is
3923 used solely to wake up the event loop so it takes notice of any new
3924 watchers that might have been added:
3925 static void
3927 async_cb (EV_P_ ev_async *w, int revents)
3928 {
3929 // just used for the side effects
3930 }
3931 The "l_release" and "l_acquire" callbacks simply unlock/lock the mutex
3933 protecting the loop data, respectively.
3934 static void
3936 l_release (EV_P)
3937 {
3938 userdata *u = ev_userdata (EV_A);
3939 pthread_mutex_unlock (&u->lock);
3940 }
3941 static void
3943 l_acquire (EV_P)
3944 {
3945 userdata *u = ev_userdata (EV_A);
3946 pthread_mutex_lock (&u->lock);
3947 }
3948 The event loop thread first acquires the mutex, and then jumps straight
3950 into "ev_loop":
3951 void *
3953 l_run (void *thr_arg)
3954 {
3955 struct ev_loop *loop = (struct ev_loop *)thr_arg;
3956 l_acquire (EV_A);
3958 pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
3959 ev_loop (EV_A_ 0);
3960 l_release (EV_A);
3961 return 0;
3963 }
3964 Instead of invoking all pending watchers, the "l_invoke" callback will
3966 signal the main thread via some unspecified mechanism (signals? pipe
3967 writes? "Async::Interrupt"?) and then waits until all pending watchers
3968 have been called (in a while loop because a) spurious wakeups are
3969 possible and b) skipping inter-thread-communication when there are no
3970 pending watchers is very beneficial):
3971 static void
3973 l_invoke (EV_P)
3974 {
3975 userdata *u = ev_userdata (EV_A);
3976 while (ev_pending_count (EV_A))
3978 {
3979 wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
3980 pthread_cond_wait (&u->invoke_cv, &u->lock);
3981 }
3982 }
3983 Now, whenever the main thread gets told to invoke pending watchers, it
3985 will grab the lock, call "ev_invoke_pending" and then signal the loop
3986 thread to continue:
3987 static void
3989 real_invoke_pending (EV_P)
3990 {
3991 userdata *u = ev_userdata (EV_A);
3992 pthread_mutex_lock (&u->lock);
3994 ev_invoke_pending (EV_A);
3995 pthread_cond_signal (&u->invoke_cv);
3996 pthread_mutex_unlock (&u->lock);
3997 }
3998 Whenever you want to start/stop a watcher or do other modifications to
4000 an event loop, you will now have to lock:
4001 ev_timer timeout_watcher;
4003 userdata *u = ev_userdata (EV_A);
4004 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
4006 pthread_mutex_lock (&u->lock);
4008 ev_timer_start (EV_A_ &timeout_watcher);
4009 ev_async_send (EV_A_ &u->async_w);
4010 pthread_mutex_unlock (&u->lock);
4011 Note that sending the "ev_async" watcher is required because otherwise
4013 an event loop currently blocking in the kernel will have no knowledge
4014 about the newly added timer. By waking up the loop it will pick up any
4015 new watchers in the next event loop iteration.
4016 COROUTINES
4018 Libev is very accommodating to coroutines ("cooperative threads"):
4020 libev fully supports nesting calls to its functions from different
4021 coroutines (e.g. you can call "ev_loop" on the same loop from two
4022 different coroutines, and switch freely between both coroutines running
4023 the loop, as long as you don't confuse yourself). The only exception is
4024 that you must not do this from "ev_periodic" reschedule callbacks.
4025 Care has been taken to ensure that libev does not keep local state
4027 inside "ev_loop", and other calls do not usually allow for coroutine
4028 switches as they do not call any callbacks.
4029 COMPILER WARNINGS
4031 Depending on your compiler and compiler settings, you might get no or a
4032 lot of warnings when compiling libev code. Some people are apparently
4033 scared by this.
4034 However, these are unavoidable for many reasons. For one, each compiler
4036 has different warnings, and each user has different tastes regarding
4037 warning options. "Warn-free" code therefore cannot be a goal except
4038 when targeting a specific compiler and compiler-version.
4039 Another reason is that some compiler warnings require elaborate
4041 workarounds, or other changes to the code that make it less clear and
4042 less maintainable.
4043 And of course, some compiler warnings are just plain stupid, or simply
4045 wrong (because they don't actually warn about the condition their
4046 message seems to warn about). For example, certain older gcc versions
4047 had some warnings that resulted an extreme number of false positives.
4048 These have been fixed, but some people still insist on making code
4049 warn-free with such buggy versions.
4050 While libev is written to generate as few warnings as possible, "warn-
4052 free" code is not a goal, and it is recommended not to build libev with
4053 any compiler warnings enabled unless you are prepared to cope with them
4054 (e.g. by ignoring them). Remember that warnings are just that:
4055 warnings, not errors, or proof of bugs.
4056 VALGRIND
4058 Valgrind has a special section here because it is a popular tool that
4059 is highly useful. Unfortunately, valgrind reports are very hard to
4060 interpret.
4061 If you think you found a bug (memory leak, uninitialised data access
4063 etc.) in libev, then check twice: If valgrind reports something like:
4064 ==2274== definitely lost: 0 bytes in 0 blocks.
4066 ==2274== possibly lost: 0 bytes in 0 blocks.
4067 ==2274== still reachable: 256 bytes in 1 blocks.
4068 Then there is no memory leak, just as memory accounted to global
4070 variables is not a memleak - the memory is still being referenced, and
4071 didn't leak.
4072 Similarly, under some circumstances, valgrind might report kernel bugs
4074 as if it were a bug in libev (e.g. in realloc or in the poll backend,
4075 although an acceptable workaround has been found here), or it might be
4076 confused.
4077 Keep in mind that valgrind is a very good tool, but only a tool. Don't
4079 make it into some kind of religion.
4080 If you are unsure about something, feel free to contact the mailing
4082 list with the full valgrind report and an explanation on why you think
4083 this is a bug in libev (best check the archives, too :). However, don't
4084 be annoyed when you get a brisk "this is no bug" answer and take the
4085 chance of learning how to interpret valgrind properly.
4086 If you need, for some reason, empty reports from valgrind for your
4088 project I suggest using suppression lists.
4089
4091 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
4092 Win32 doesn't support any of the standards (e.g. POSIX) that libev
4093 requires, and its I/O model is fundamentally incompatible with the
4094 POSIX model. Libev still offers limited functionality on this platform
4095 in the form of the "EVBACKEND_SELECT" backend, and only supports socket
4096 descriptors. This only applies when using Win32 natively, not when
4097 using e.g. cygwin.
4098 Lifting these limitations would basically require the full re-
4100 implementation of the I/O system. If you are into these kinds of
4101 things, then note that glib does exactly that for you in a very
4102 portable way (note also that glib is the slowest event library known to
4103 man).
4104 There is no supported compilation method available on windows except
4106 embedding it into other applications.
4107 Sensible signal handling is officially unsupported by Microsoft - libev
4109 tries its best, but under most conditions, signals will simply not
4110 work.
4111 Not a libev limitation but worth mentioning: windows apparently doesn't
4113 accept large writes: instead of resulting in a partial write, windows
4114 will either accept everything or return "ENOBUFS" if the buffer is too
4115 large, so make sure you only write small amounts into your sockets
4116 (less than a megabyte seems safe, but this apparently depends on the
4117 amount of memory available).
4118 Due to the many, low, and arbitrary limits on the win32 platform and
4120 the abysmal performance of winsockets, using a large number of sockets
4121 is not recommended (and not reasonable). If your program needs to use
4122 more than a hundred or so sockets, then likely it needs to use a
4123 totally different implementation for windows, as libev offers the POSIX
4124 readiness notification model, which cannot be implemented efficiently
4125 on windows (due to Microsoft monopoly games).
4126 A typical way to use libev under windows is to embed it (see the
4128 embedding section for details) and use the following evwrap.h header
4129 file instead of ev.h:
4130 #define EV_STANDALONE /* keeps ev from requiring config.h */
4132 #define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
4133 #include "ev.h"
4135 And compile the following evwrap.c file into your project (make sure
4137 you do not compile the ev.c or any other embedded source files!):
4138 #include "evwrap.h"
4140 #include "ev.c"
4141 The winsocket select function
4143 The winsocket "select" function doesn't follow POSIX in that it
4144 requires socket handles and not socket file descriptors (it is also
4145 extremely buggy). This makes select very inefficient, and also
4146 requires a mapping from file descriptors to socket handles (the
4147 Microsoft C runtime provides the function "_open_osfhandle" for
4148 this). See the discussion of the "EV_SELECT_USE_FD_SET",
4149 "EV_SELECT_IS_WINSOCKET" and "EV_FD_TO_WIN32_HANDLE" preprocessor
4150 symbols for more info.
4151 The configuration for a "naked" win32 using the Microsoft runtime
4153 libraries and raw winsocket select is:
4154 #define EV_USE_SELECT 1
4156 #define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
4157 Note that winsockets handling of fd sets is O(n), so you can easily
4159 get a complexity in the O(nAX) range when using win32.
4160 Limited number of file descriptors
4162 Windows has numerous arbitrary (and low) limits on things.
4163 Early versions of winsocket's select only supported waiting for a
4165 maximum of 64 handles (probably owning to the fact that all windows
4166 kernels can only wait for 64 things at the same time internally;
4167 Microsoft recommends spawning a chain of threads and wait for 63
4168 handles and the previous thread in each. Sounds great!).
4169 Newer versions support more handles, but you need to define
4171 "FD_SETSIZE" to some high number (e.g. 2048) before compiling the
4172 winsocket select call (which might be in libev or elsewhere, for
4173 example, perl and many other interpreters do their own select
4174 emulation on windows).
4175 Another limit is the number of file descriptors in the Microsoft
4177 runtime libraries, which by default is 64 (there must be a hidden
4178 64 fetish or something like this inside Microsoft). You can
4179 increase this by calling "_setmaxstdio", which can increase this
4180 limit to 2048 (another arbitrary limit), but is broken in many
4181 versions of the Microsoft runtime libraries. This might get you to
4182 about 512 or 2048 sockets (depending on windows version and/or the
4183 phase of the moon). To get more, you need to wrap all I/O functions
4184 and provide your own fd management, but the cost of calling select
4185 (O(nAX)) will likely make this unworkable.
4186 PORTABILITY REQUIREMENTS
4188 In addition to a working ISO-C implementation and of course the
4189 backend-specific APIs, libev relies on a few additional extensions:
4190 "void (*)(ev_watcher_type *, int revents)" must have compatible calling
4192 conventions regardless of "ev_watcher_type *".
4193 Libev assumes not only that all watcher pointers have the same
4194 internal structure (guaranteed by POSIX but not by ISO C for
4195 example), but it also assumes that the same (machine) code can be
4196 used to call any watcher callback: The watcher callbacks have
4197 different type signatures, but libev calls them using an
4198 "ev_watcher *" internally.
4199 "sig_atomic_t volatile" must be thread-atomic as well
4201 The type "sig_atomic_t volatile" (or whatever is defined as
4202 "EV_ATOMIC_T") must be atomic with respect to accesses from
4203 different threads. This is not part of the specification for
4204 "sig_atomic_t", but is believed to be sufficiently portable.
4205 "sigprocmask" must work in a threaded environment
4207 Libev uses "sigprocmask" to temporarily block signals. This is not
4208 allowed in a threaded program ("pthread_sigmask" has to be used).
4209 Typical pthread implementations will either allow "sigprocmask" in
4210 the "main thread" or will block signals process-wide, both
4211 behaviours would be compatible with libev. Interaction between
4212 "sigprocmask" and "pthread_sigmask" could complicate things,
4213 however.
4214 The most portable way to handle signals is to block signals in all
4216 threads except the initial one, and run the default loop in the
4217 initial thread as well.
4218 "long" must be large enough for common memory allocation sizes
4220 To improve portability and simplify its API, libev uses "long"
4221 internally instead of "size_t" when allocating its data structures.
4222 On non-POSIX systems (Microsoft...) this might be unexpectedly low,
4223 but is still at least 31 bits everywhere, which is enough for
4224 hundreds of millions of watchers.
4225 "double" must hold a time value in seconds with enough accuracy
4227 The type "double" is used to represent timestamps. It is required
4228 to have at least 51 bits of mantissa (and 9 bits of exponent),
4229 which is good enough for at least into the year 4000. This
4230 requirement is fulfilled by implementations implementing IEEE 754,
4231 which is basically all existing ones. With IEEE 754 doubles, you
4232 get microsecond accuracy until at least 2200.
4233 If you know of other additional requirements drop me a note.
4235
4237 In this section the complexities of (many of) the algorithms used
4238 inside libev will be documented. For complexity discussions about
4239 backends see the documentation for "ev_default_init".
4240 All of the following are about amortised time: If an array needs to be
4242 extended, libev needs to realloc and move the whole array, but this
4243 happens asymptotically rarer with higher number of elements, so O(1)
4244 might mean that libev does a lengthy realloc operation in rare cases,
4245 but on average it is much faster and asymptotically approaches constant
4246 time.
4247 Starting and stopping timer/periodic watchers: O(log
4249 skipped_other_timers)
4250 This means that, when you have a watcher that triggers in one hour
4251 and there are 100 watchers that would trigger before that, then
4252 inserting will have to skip roughly seven ("ld 100") of these
4253 watchers.
4254 Changing timer/periodic watchers (by autorepeat or calling again):
4256 O(log skipped_other_timers)
4257 That means that changing a timer costs less than removing/adding
4258 them, as only the relative motion in the event queue has to be paid
4259 for.
4260 Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
4262 These just add the watcher into an array or at the head of a list.
4263 Stopping check/prepare/idle/fork/async watchers: O(1)
4265 Stopping an io/signal/child watcher:
4266 O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
4267 These watchers are stored in lists, so they need to be walked to
4268 find the correct watcher to remove. The lists are usually short
4269 (you don't usually have many watchers waiting for the same fd or
4270 signal: one is typical, two is rare).
4271 Finding the next timer in each loop iteration: O(1)
4273 By virtue of using a binary or 4-heap, the next timer is always
4274 found at a fixed position in the storage array.
4275 Each change on a file descriptor per loop iteration:
4277 O(number_of_watchers_for_this_fd)
4278 A change means an I/O watcher gets started or stopped, which
4279 requires libev to recalculate its status (and possibly tell the
4280 kernel, depending on backend and whether "ev_io_set" was used).
4281 Activating one watcher (putting it into the pending state): O(1)
4283 Priority handling: O(number_of_priorities)
4284 Priorities are implemented by allocating some space for each
4285 priority. When doing priority-based operations, libev usually has
4286 to linearly search all the priorities, but starting/stopping and
4287 activating watchers becomes O(1) with respect to priority handling.
4288 Sending an ev_async: O(1)
4290 Processing ev_async_send: O(number_of_async_watchers)
4291 Processing signals: O(max_signal_number)
4292 Sending involves a system call iff there were no other
4293 "ev_async_send" calls in the current loop iteration. Checking for
4294 async and signal events involves iterating over all running async
4295 watchers or all signal numbers.
4296
4298 active
4299 A watcher is active as long as it has been started (has been
4300 attached to an event loop) but not yet stopped (disassociated from
4301 the event loop).
4302 application
4304 In this document, an application is whatever is using libev.
4305 callback
4307 The address of a function that is called when some event has been
4308 detected. Callbacks are being passed the event loop, the watcher
4309 that received the event, and the actual event bitset.
4310 callback invocation
4312 The act of calling the callback associated with a watcher.
4313 event
4315 A change of state of some external event, such as data now being
4316 available for reading on a file descriptor, time having passed or
4317 simply not having any other events happening anymore.
4318 In libev, events are represented as single bits (such as "EV_READ"
4320 or "EV_TIMEOUT").
4321 event library
4323 A software package implementing an event model and loop.
4324 event loop
4326 An entity that handles and processes external events and converts
4327 them into callback invocations.
4328 event model
4330 The model used to describe how an event loop handles and processes
4331 watchers and events.
4332 pending
4334 A watcher is pending as soon as the corresponding event has been
4335 detected, and stops being pending as soon as the watcher will be
4336 invoked or its pending status is explicitly cleared by the
4337 application.
4338 A watcher can be pending, but not active. Stopping a watcher also
4340 clears its pending status.
4341 real time
4343 The physical time that is observed. It is apparently strictly
4344 monotonic :)
4345 wall-clock time
4347 The time and date as shown on clocks. Unlike real time, it can
4348 actually be wrong and jump forwards and backwards, e.g. when the
4349 you adjust your clock.
4350 watcher
4352 A data structure that describes interest in certain events.
4353 Watchers need to be started (attached to an event loop) before they
4354 can receive events.
4355 watcher invocation
4357 The act of calling the callback associated with a watcher.
4358
4360 Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael
4361 Magnusson.
4362libev-3.9 2009-12-31 LIBEV(3)