ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/ev.pod
(Generate patch)

Comparing libev/ev.pod (file contents):
Revision 1.84 by root, Wed Dec 12 22:26:37 2007 UTC vs.
Revision 1.128 by root, Fri Feb 1 13:53:56 2008 UTC

… …
4 4
5=head1 SYNOPSIS 5=head1 SYNOPSIS
6 6
7 #include <ev.h> 7 #include <ev.h>
8 8
9=head1 EXAMPLE PROGRAM 9=head2 EXAMPLE PROGRAM
10 10
11 #include <ev.h> 11 #include <ev.h>
12 12
13 ev_io stdin_watcher; 13 ev_io stdin_watcher;
14 ev_timer timeout_watcher; 14 ev_timer timeout_watcher;
… …
53The newest version of this document is also available as a html-formatted 53The newest version of this document is also available as a html-formatted
54web page you might find easier to navigate when reading it for the first 54web page you might find easier to navigate when reading it for the first
55time: L<http://cvs.schmorp.de/libev/ev.html>. 55time: L<http://cvs.schmorp.de/libev/ev.html>.
56 56
57Libev is an event loop: you register interest in certain events (such as a 57Libev is an event loop: you register interest in certain events (such as a
58file descriptor being readable or a timeout occuring), and it will manage 58file descriptor being readable or a timeout occurring), and it will manage
59these event sources and provide your program with events. 59these event sources and provide your program with events.
60 60
61To do this, it must take more or less complete control over your process 61To do this, it must take more or less complete control over your process
62(or thread) by executing the I<event loop> handler, and will then 62(or thread) by executing the I<event loop> handler, and will then
63communicate events via a callback mechanism. 63communicate events via a callback mechanism.
… …
65You register interest in certain events by registering so-called I<event 65You register interest in certain events by registering so-called I<event
66watchers>, which are relatively small C structures you initialise with the 66watchers>, which are relatively small C structures you initialise with the
67details of the event, and then hand it over to libev by I<starting> the 67details of the event, and then hand it over to libev by I<starting> the
68watcher. 68watcher.
69 69
70=head1 FEATURES 70=head2 FEATURES
71 71
72Libev supports C<select>, C<poll>, the Linux-specific C<epoll>, the 72Libev supports C<select>, C<poll>, the Linux-specific C<epoll>, the
73BSD-specific C<kqueue> and the Solaris-specific event port mechanisms 73BSD-specific C<kqueue> and the Solaris-specific event port mechanisms
74for file descriptor events (C<ev_io>), the Linux C<inotify> interface 74for file descriptor events (C<ev_io>), the Linux C<inotify> interface
75(for C<ev_stat>), relative timers (C<ev_timer>), absolute timers 75(for C<ev_stat>), relative timers (C<ev_timer>), absolute timers
… …
82 82
83It also is quite fast (see this 83It also is quite fast (see this
84L<benchmark|http://libev.schmorp.de/bench.html> comparing it to libevent 84L<benchmark|http://libev.schmorp.de/bench.html> comparing it to libevent
85for example). 85for example).
86 86
87=head1 CONVENTIONS 87=head2 CONVENTIONS
88 88
89Libev is very configurable. In this manual the default configuration will 89Libev is very configurable. In this manual the default configuration will
90be described, which supports multiple event loops. For more info about 90be described, which supports multiple event loops. For more info about
91various configuration options please have a look at B<EMBED> section in 91various configuration options please have a look at B<EMBED> section in
92this manual. If libev was configured without support for multiple event 92this manual. If libev was configured without support for multiple event
93loops, then all functions taking an initial argument of name C<loop> 93loops, then all functions taking an initial argument of name C<loop>
94(which is always of type C<struct ev_loop *>) will not have this argument. 94(which is always of type C<struct ev_loop *>) will not have this argument.
95 95
96=head1 TIME REPRESENTATION 96=head2 TIME REPRESENTATION
97 97
98Libev represents time as a single floating point number, representing the 98Libev represents time as a single floating point number, representing the
99(fractional) number of seconds since the (POSIX) epoch (somewhere near 99(fractional) number of seconds since the (POSIX) epoch (somewhere near
100the beginning of 1970, details are complicated, don't ask). This type is 100the beginning of 1970, details are complicated, don't ask). This type is
101called C<ev_tstamp>, which is what you should use too. It usually aliases 101called C<ev_tstamp>, which is what you should use too. It usually aliases
102to the C<double> type in C, and when you need to do any calculations on 102to the C<double> type in C, and when you need to do any calculations on
103it, you should treat it as such. 103it, you should treat it as some floatingpoint value. Unlike the name
104component C<stamp> might indicate, it is also used for time differences
105throughout libev.
104 106
105=head1 GLOBAL FUNCTIONS 107=head1 GLOBAL FUNCTIONS
106 108
107These functions can be called anytime, even before initialising the 109These functions can be called anytime, even before initialising the
108library in any way. 110library in any way.
… …
112=item ev_tstamp ev_time () 114=item ev_tstamp ev_time ()
113 115
114Returns the current time as libev would use it. Please note that the 116Returns the current time as libev would use it. Please note that the
115C<ev_now> function is usually faster and also often returns the timestamp 117C<ev_now> function is usually faster and also often returns the timestamp
116you actually want to know. 118you actually want to know.
119
120=item ev_sleep (ev_tstamp interval)
121
122Sleep for the given interval: The current thread will be blocked until
123either it is interrupted or the given time interval has passed. Basically
124this is a subsecond-resolution C<sleep ()>.
117 125
118=item int ev_version_major () 126=item int ev_version_major ()
119 127
120=item int ev_version_minor () 128=item int ev_version_minor ()
121 129
… …
252flags. If that is troubling you, check C<ev_backend ()> afterwards). 260flags. If that is troubling you, check C<ev_backend ()> afterwards).
253 261
254If you don't know what event loop to use, use the one returned from this 262If you don't know what event loop to use, use the one returned from this
255function. 263function.
256 264
265The default loop is the only loop that can handle C<ev_signal> and
266C<ev_child> watchers, and to do this, it always registers a handler
267for C<SIGCHLD>. If this is a problem for your app you can either
268create a dynamic loop with C<ev_loop_new> that doesn't do that, or you
269can simply overwrite the C<SIGCHLD> signal handler I<after> calling
270C<ev_default_init>.
271
257The flags argument can be used to specify special behaviour or specific 272The flags argument can be used to specify special behaviour or specific
258backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>). 273backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
259 274
260The following flags are supported: 275The following flags are supported:
261 276
… …
298=item C<EVBACKEND_SELECT> (value 1, portable select backend) 313=item C<EVBACKEND_SELECT> (value 1, portable select backend)
299 314
300This is your standard select(2) backend. Not I<completely> standard, as 315This is your standard select(2) backend. Not I<completely> standard, as
301libev tries to roll its own fd_set with no limits on the number of fds, 316libev tries to roll its own fd_set with no limits on the number of fds,
302but if that fails, expect a fairly low limit on the number of fds when 317but if that fails, expect a fairly low limit on the number of fds when
303using this backend. It doesn't scale too well (O(highest_fd)), but its usually 318using this backend. It doesn't scale too well (O(highest_fd)), but its
304the fastest backend for a low number of fds. 319usually the fastest backend for a low number of (low-numbered :) fds.
320
321To get good performance out of this backend you need a high amount of
322parallelity (most of the file descriptors should be busy). If you are
323writing a server, you should C<accept ()> in a loop to accept as many
324connections as possible during one iteration. You might also want to have
325a look at C<ev_set_io_collect_interval ()> to increase the amount of
326readyness notifications you get per iteration.
305 327
306=item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows) 328=item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
307 329
308And this is your standard poll(2) backend. It's more complicated than 330And this is your standard poll(2) backend. It's more complicated
309select, but handles sparse fds better and has no artificial limit on the 331than select, but handles sparse fds better and has no artificial
310number of fds you can use (except it will slow down considerably with a 332limit on the number of fds you can use (except it will slow down
311lot of inactive fds). It scales similarly to select, i.e. O(total_fds). 333considerably with a lot of inactive fds). It scales similarly to select,
334i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
335performance tips.
312 336
313=item C<EVBACKEND_EPOLL> (value 4, Linux) 337=item C<EVBACKEND_EPOLL> (value 4, Linux)
314 338
315For few fds, this backend is a bit little slower than poll and select, 339For few fds, this backend is a bit little slower than poll and select,
316but it scales phenomenally better. While poll and select usually scale like 340but it scales phenomenally better. While poll and select usually scale
317O(total_fds) where n is the total number of fds (or the highest fd), epoll scales 341like O(total_fds) where n is the total number of fds (or the highest fd),
318either O(1) or O(active_fds). 342epoll scales either O(1) or O(active_fds). The epoll design has a number
343of shortcomings, such as silently dropping events in some hard-to-detect
344cases and rewiring a syscall per fd change, no fork support and bad
345support for dup.
319 346
320While stopping and starting an I/O watcher in the same iteration will 347While stopping, setting and starting an I/O watcher in the same iteration
321result in some caching, there is still a syscall per such incident 348will result in some caching, there is still a syscall per such incident
322(because the fd could point to a different file description now), so its 349(because the fd could point to a different file description now), so its
323best to avoid that. Also, dup()ed file descriptors might not work very 350best to avoid that. Also, C<dup ()>'ed file descriptors might not work
324well if you register events for both fds. 351very well if you register events for both fds.
325 352
326Please note that epoll sometimes generates spurious notifications, so you 353Please note that epoll sometimes generates spurious notifications, so you
327need to use non-blocking I/O or other means to avoid blocking when no data 354need to use non-blocking I/O or other means to avoid blocking when no data
328(or space) is available. 355(or space) is available.
329 356
357Best performance from this backend is achieved by not unregistering all
358watchers for a file descriptor until it has been closed, if possible, i.e.
359keep at least one watcher active per fd at all times.
360
361While nominally embeddeble in other event loops, this feature is broken in
362all kernel versions tested so far.
363
330=item C<EVBACKEND_KQUEUE> (value 8, most BSD clones) 364=item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
331 365
332Kqueue deserves special mention, as at the time of this writing, it 366Kqueue deserves special mention, as at the time of this writing, it
333was broken on all BSDs except NetBSD (usually it doesn't work with 367was broken on all BSDs except NetBSD (usually it doesn't work reliably
334anything but sockets and pipes, except on Darwin, where of course its 368with anything but sockets and pipes, except on Darwin, where of course
335completely useless). For this reason its not being "autodetected" 369it's completely useless). For this reason it's not being "autodetected"
336unless you explicitly specify it explicitly in the flags (i.e. using 370unless you explicitly specify it explicitly in the flags (i.e. using
337C<EVBACKEND_KQUEUE>). 371C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
372system like NetBSD.
373
374You still can embed kqueue into a normal poll or select backend and use it
375only for sockets (after having made sure that sockets work with kqueue on
376the target platform). See C<ev_embed> watchers for more info.
338 377
339It scales in the same way as the epoll backend, but the interface to the 378It scales in the same way as the epoll backend, but the interface to the
340kernel is more efficient (which says nothing about its actual speed, of 379kernel is more efficient (which says nothing about its actual speed, of
341course). While starting and stopping an I/O watcher does not cause an 380course). While stopping, setting and starting an I/O watcher does never
342extra syscall as with epoll, it still adds up to four event changes per 381cause an extra syscall as with C<EVBACKEND_EPOLL>, it still adds up to
343incident, so its best to avoid that. 382two event changes per incident, support for C<fork ()> is very bad and it
383drops fds silently in similarly hard-to-detect cases.
384
385This backend usually performs well under most conditions.
386
387While nominally embeddable in other event loops, this doesn't work
388everywhere, so you might need to test for this. And since it is broken
389almost everywhere, you should only use it when you have a lot of sockets
390(for which it usually works), by embedding it into another event loop
391(e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>) and using it only for
392sockets.
344 393
345=item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8) 394=item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
346 395
347This is not implemented yet (and might never be). 396This is not implemented yet (and might never be, unless you send me an
397implementation). According to reports, C</dev/poll> only supports sockets
398and is not embeddable, which would limit the usefulness of this backend
399immensely.
348 400
349=item C<EVBACKEND_PORT> (value 32, Solaris 10) 401=item C<EVBACKEND_PORT> (value 32, Solaris 10)
350 402
351This uses the Solaris 10 port mechanism. As with everything on Solaris, 403This uses the Solaris 10 event port mechanism. As with everything on Solaris,
352it's really slow, but it still scales very well (O(active_fds)). 404it's really slow, but it still scales very well (O(active_fds)).
353 405
354Please note that solaris ports can result in a lot of spurious 406Please note that solaris event ports can deliver a lot of spurious
355notifications, so you need to use non-blocking I/O or other means to avoid 407notifications, so you need to use non-blocking I/O or other means to avoid
356blocking when no data (or space) is available. 408blocking when no data (or space) is available.
409
410While this backend scales well, it requires one system call per active
411file descriptor per loop iteration. For small and medium numbers of file
412descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
413might perform better.
414
415On the positive side, ignoring the spurious readyness notifications, this
416backend actually performed to specification in all tests and is fully
417embeddable, which is a rare feat among the OS-specific backends.
357 418
358=item C<EVBACKEND_ALL> 419=item C<EVBACKEND_ALL>
359 420
360Try all backends (even potentially broken ones that wouldn't be tried 421Try all backends (even potentially broken ones that wouldn't be tried
361with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as 422with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
362C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>. 423C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.
363 424
425It is definitely not recommended to use this flag.
426
364=back 427=back
365 428
366If one or more of these are ored into the flags value, then only these 429If one or more of these are ored into the flags value, then only these
367backends will be tried (in the reverse order as given here). If none are 430backends will be tried (in the reverse order as listed here). If none are
368specified, most compiled-in backend will be tried, usually in reverse 431specified, all backends in C<ev_recommended_backends ()> will be tried.
369order of their flag values :)
370 432
371The most typical usage is like this: 433The most typical usage is like this:
372 434
373 if (!ev_default_loop (0)) 435 if (!ev_default_loop (0))
374 fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?"); 436 fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
… …
402Destroys the default loop again (frees all memory and kernel state 464Destroys the default loop again (frees all memory and kernel state
403etc.). None of the active event watchers will be stopped in the normal 465etc.). None of the active event watchers will be stopped in the normal
404sense, so e.g. C<ev_is_active> might still return true. It is your 466sense, so e.g. C<ev_is_active> might still return true. It is your
405responsibility to either stop all watchers cleanly yoursef I<before> 467responsibility to either stop all watchers cleanly yoursef I<before>
406calling this function, or cope with the fact afterwards (which is usually 468calling this function, or cope with the fact afterwards (which is usually
407the easiest thing, youc na just ignore the watchers and/or C<free ()> them 469the easiest thing, you can just ignore the watchers and/or C<free ()> them
408for example). 470for example).
471
472Note that certain global state, such as signal state, will not be freed by
473this function, and related watchers (such as signal and child watchers)
474would need to be stopped manually.
475
476In general it is not advisable to call this function except in the
477rare occasion where you really need to free e.g. the signal handling
478pipe fds. If you need dynamically allocated loops it is better to use
479C<ev_loop_new> and C<ev_loop_destroy>).
409 480
410=item ev_loop_destroy (loop) 481=item ev_loop_destroy (loop)
411 482
412Like C<ev_default_destroy>, but destroys an event loop created by an 483Like C<ev_default_destroy>, but destroys an event loop created by an
413earlier call to C<ev_loop_new>. 484earlier call to C<ev_loop_new>.
414 485
415=item ev_default_fork () 486=item ev_default_fork ()
416 487
488This function sets a flag that causes subsequent C<ev_loop> iterations
417This function reinitialises the kernel state for backends that have 489to reinitialise the kernel state for backends that have one. Despite the
418one. Despite the name, you can call it anytime, but it makes most sense 490name, you can call it anytime, but it makes most sense after forking, in
419after forking, in either the parent or child process (or both, but that 491the child process (or both child and parent, but that again makes little
420again makes little sense). 492sense). You I<must> call it in the child before using any of the libev
493functions, and it will only take effect at the next C<ev_loop> iteration.
421 494
422You I<must> call this function in the child process after forking if and 495On the other hand, you only need to call this function in the child
423only if you want to use the event library in both processes. If you just 496process if and only if you want to use the event library in the child. If
424fork+exec, you don't have to call it. 497you just fork+exec, you don't have to call it at all.
425 498
426The function itself is quite fast and it's usually not a problem to call 499The function itself is quite fast and it's usually not a problem to call
427it just in case after a fork. To make this easy, the function will fit in 500it just in case after a fork. To make this easy, the function will fit in
428quite nicely into a call to C<pthread_atfork>: 501quite nicely into a call to C<pthread_atfork>:
429 502
430 pthread_atfork (0, 0, ev_default_fork); 503 pthread_atfork (0, 0, ev_default_fork);
431
432At the moment, C<EVBACKEND_SELECT> and C<EVBACKEND_POLL> are safe to use
433without calling this function, so if you force one of those backends you
434do not need to care.
435 504
436=item ev_loop_fork (loop) 505=item ev_loop_fork (loop)
437 506
438Like C<ev_default_fork>, but acts on an event loop created by 507Like C<ev_default_fork>, but acts on an event loop created by
439C<ev_loop_new>. Yes, you have to call this on every allocated event loop 508C<ev_loop_new>. Yes, you have to call this on every allocated event loop
… …
458 527
459Returns the current "event loop time", which is the time the event loop 528Returns the current "event loop time", which is the time the event loop
460received events and started processing them. This timestamp does not 529received events and started processing them. This timestamp does not
461change as long as callbacks are being processed, and this is also the base 530change as long as callbacks are being processed, and this is also the base
462time used for relative timers. You can treat it as the timestamp of the 531time used for relative timers. You can treat it as the timestamp of the
463event occuring (or more correctly, libev finding out about it). 532event occurring (or more correctly, libev finding out about it).
464 533
465=item ev_loop (loop, int flags) 534=item ev_loop (loop, int flags)
466 535
467Finally, this is it, the event handler. This function usually is called 536Finally, this is it, the event handler. This function usually is called
468after you initialised all your watchers and you want to start handling 537after you initialised all your watchers and you want to start handling
… …
490usually a better approach for this kind of thing. 559usually a better approach for this kind of thing.
491 560
492Here are the gory details of what C<ev_loop> does: 561Here are the gory details of what C<ev_loop> does:
493 562
494 - Before the first iteration, call any pending watchers. 563 - Before the first iteration, call any pending watchers.
495 * If there are no active watchers (reference count is zero), return. 564 * If EVFLAG_FORKCHECK was used, check for a fork.
496 - Queue all prepare watchers and then call all outstanding watchers. 565 - If a fork was detected, queue and call all fork watchers.
566 - Queue and call all prepare watchers.
497 - If we have been forked, recreate the kernel state. 567 - If we have been forked, recreate the kernel state.
498 - Update the kernel state with all outstanding changes. 568 - Update the kernel state with all outstanding changes.
499 - Update the "event loop time". 569 - Update the "event loop time".
500 - Calculate for how long to block. 570 - Calculate for how long to sleep or block, if at all
571 (active idle watchers, EVLOOP_NONBLOCK or not having
572 any active watchers at all will result in not sleeping).
573 - Sleep if the I/O and timer collect interval say so.
501 - Block the process, waiting for any events. 574 - Block the process, waiting for any events.
502 - Queue all outstanding I/O (fd) events. 575 - Queue all outstanding I/O (fd) events.
503 - Update the "event loop time" and do time jump handling. 576 - Update the "event loop time" and do time jump handling.
504 - Queue all outstanding timers. 577 - Queue all outstanding timers.
505 - Queue all outstanding periodics. 578 - Queue all outstanding periodics.
506 - If no events are pending now, queue all idle watchers. 579 - If no events are pending now, queue all idle watchers.
507 - Queue all check watchers. 580 - Queue all check watchers.
508 - Call all queued watchers in reverse order (i.e. check watchers first). 581 - Call all queued watchers in reverse order (i.e. check watchers first).
509 Signals and child watchers are implemented as I/O watchers, and will 582 Signals and child watchers are implemented as I/O watchers, and will
510 be handled here by queueing them when their watcher gets executed. 583 be handled here by queueing them when their watcher gets executed.
511 - If ev_unloop has been called or EVLOOP_ONESHOT or EVLOOP_NONBLOCK 584 - If ev_unloop has been called, or EVLOOP_ONESHOT or EVLOOP_NONBLOCK
512 were used, return, otherwise continue with step *. 585 were used, or there are no active watchers, return, otherwise
586 continue with step *.
513 587
514Example: Queue some jobs and then loop until no events are outsanding 588Example: Queue some jobs and then loop until no events are outstanding
515anymore. 589anymore.
516 590
517 ... queue jobs here, make sure they register event watchers as long 591 ... queue jobs here, make sure they register event watchers as long
518 ... as they still have work to do (even an idle watcher will do..) 592 ... as they still have work to do (even an idle watcher will do..)
519 ev_loop (my_loop, 0); 593 ev_loop (my_loop, 0);
… …
523 597
524Can be used to make a call to C<ev_loop> return early (but only after it 598Can be used to make a call to C<ev_loop> return early (but only after it
525has processed all outstanding events). The C<how> argument must be either 599has processed all outstanding events). The C<how> argument must be either
526C<EVUNLOOP_ONE>, which will make the innermost C<ev_loop> call return, or 600C<EVUNLOOP_ONE>, which will make the innermost C<ev_loop> call return, or
527C<EVUNLOOP_ALL>, which will make all nested C<ev_loop> calls return. 601C<EVUNLOOP_ALL>, which will make all nested C<ev_loop> calls return.
602
603This "unloop state" will be cleared when entering C<ev_loop> again.
528 604
529=item ev_ref (loop) 605=item ev_ref (loop)
530 606
531=item ev_unref (loop) 607=item ev_unref (loop)
532 608
… …
537returning, ev_unref() after starting, and ev_ref() before stopping it. For 613returning, ev_unref() after starting, and ev_ref() before stopping it. For
538example, libev itself uses this for its internal signal pipe: It is not 614example, libev itself uses this for its internal signal pipe: It is not
539visible to the libev user and should not keep C<ev_loop> from exiting if 615visible to the libev user and should not keep C<ev_loop> from exiting if
540no event watchers registered by it are active. It is also an excellent 616no event watchers registered by it are active. It is also an excellent
541way to do this for generic recurring timers or from within third-party 617way to do this for generic recurring timers or from within third-party
542libraries. Just remember to I<unref after start> and I<ref before stop>. 618libraries. Just remember to I<unref after start> and I<ref before stop>
619(but only if the watcher wasn't active before, or was active before,
620respectively).
543 621
544Example: Create a signal watcher, but keep it from keeping C<ev_loop> 622Example: Create a signal watcher, but keep it from keeping C<ev_loop>
545running when nothing else is active. 623running when nothing else is active.
546 624
547 struct ev_signal exitsig; 625 struct ev_signal exitsig;
… …
551 629
552Example: For some weird reason, unregister the above signal handler again. 630Example: For some weird reason, unregister the above signal handler again.
553 631
554 ev_ref (loop); 632 ev_ref (loop);
555 ev_signal_stop (loop, &exitsig); 633 ev_signal_stop (loop, &exitsig);
634
635=item ev_set_io_collect_interval (loop, ev_tstamp interval)
636
637=item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
638
639These advanced functions influence the time that libev will spend waiting
640for events. Both are by default C<0>, meaning that libev will try to
641invoke timer/periodic callbacks and I/O callbacks with minimum latency.
642
643Setting these to a higher value (the C<interval> I<must> be >= C<0>)
644allows libev to delay invocation of I/O and timer/periodic callbacks to
645increase efficiency of loop iterations.
646
647The background is that sometimes your program runs just fast enough to
648handle one (or very few) event(s) per loop iteration. While this makes
649the program responsive, it also wastes a lot of CPU time to poll for new
650events, especially with backends like C<select ()> which have a high
651overhead for the actual polling but can deliver many events at once.
652
653By setting a higher I<io collect interval> you allow libev to spend more
654time collecting I/O events, so you can handle more events per iteration,
655at the cost of increasing latency. Timeouts (both C<ev_periodic> and
656C<ev_timer>) will be not affected. Setting this to a non-null value will
657introduce an additional C<ev_sleep ()> call into most loop iterations.
658
659Likewise, by setting a higher I<timeout collect interval> you allow libev
660to spend more time collecting timeouts, at the expense of increased
661latency (the watcher callback will be called later). C<ev_io> watchers
662will not be affected. Setting this to a non-null value will not introduce
663any overhead in libev.
664
665Many (busy) programs can usually benefit by setting the io collect
666interval to a value near C<0.1> or so, which is often enough for
667interactive servers (of course not for games), likewise for timeouts. It
668usually doesn't make much sense to set it to a lower value than C<0.01>,
669as this approsaches the timing granularity of most systems.
556 670
557=back 671=back
558 672
559 673
560=head1 ANATOMY OF A WATCHER 674=head1 ANATOMY OF A WATCHER
… …
659 773
660=item C<EV_FORK> 774=item C<EV_FORK>
661 775
662The event loop has been resumed in the child process after fork (see 776The event loop has been resumed in the child process after fork (see
663C<ev_fork>). 777C<ev_fork>).
778
779=item C<EV_ASYNC>
780
781The given async watcher has been asynchronously notified (see C<ev_async>).
664 782
665=item C<EV_ERROR> 783=item C<EV_ERROR>
666 784
667An unspecified error has occured, the watcher has been stopped. This might 785An unspecified error has occured, the watcher has been stopped. This might
668happen because the watcher could not be properly started because libev 786happen because the watcher could not be properly started because libev
… …
886In general you can register as many read and/or write event watchers per 1004In general you can register as many read and/or write event watchers per
887fd as you want (as long as you don't confuse yourself). Setting all file 1005fd as you want (as long as you don't confuse yourself). Setting all file
888descriptors to non-blocking mode is also usually a good idea (but not 1006descriptors to non-blocking mode is also usually a good idea (but not
889required if you know what you are doing). 1007required if you know what you are doing).
890 1008
891You have to be careful with dup'ed file descriptors, though. Some backends
892(the linux epoll backend is a notable example) cannot handle dup'ed file
893descriptors correctly if you register interest in two or more fds pointing
894to the same underlying file/socket/etc. description (that is, they share
895the same underlying "file open").
896
897If you must do this, then force the use of a known-to-be-good backend 1009If you must do this, then force the use of a known-to-be-good backend
898(at the time of this writing, this includes only C<EVBACKEND_SELECT> and 1010(at the time of this writing, this includes only C<EVBACKEND_SELECT> and
899C<EVBACKEND_POLL>). 1011C<EVBACKEND_POLL>).
900 1012
901Another thing you have to watch out for is that it is quite easy to 1013Another thing you have to watch out for is that it is quite easy to
… …
913such as poll (fortunately in our Xlib example, Xlib already does this on 1025such as poll (fortunately in our Xlib example, Xlib already does this on
914its own, so its quite safe to use). 1026its own, so its quite safe to use).
915 1027
916=head3 The special problem of disappearing file descriptors 1028=head3 The special problem of disappearing file descriptors
917 1029
918Some backends (e.g kqueue, epoll) need to be told about closing a file 1030Some backends (e.g. kqueue, epoll) need to be told about closing a file
919descriptor (either by calling C<close> explicitly or by any other means, 1031descriptor (either by calling C<close> explicitly or by any other means,
920such as C<dup>). The reason is that you register interest in some file 1032such as C<dup>). The reason is that you register interest in some file
921descriptor, but when it goes away, the operating system will silently drop 1033descriptor, but when it goes away, the operating system will silently drop
922this interest. If another file descriptor with the same number then is 1034this interest. If another file descriptor with the same number then is
923registered with libev, there is no efficient way to see that this is, in 1035registered with libev, there is no efficient way to see that this is, in
… …
932 1044
933This is how one would do it normally anyway, the important point is that 1045This is how one would do it normally anyway, the important point is that
934the libev application should not optimise around libev but should leave 1046the libev application should not optimise around libev but should leave
935optimisations to libev. 1047optimisations to libev.
936 1048
1049=head3 The special problem of dup'ed file descriptors
1050
1051Some backends (e.g. epoll), cannot register events for file descriptors,
1052but only events for the underlying file descriptions. That means when you
1053have C<dup ()>'ed file descriptors or weirder constellations, and register
1054events for them, only one file descriptor might actually receive events.
1055
1056There is no workaround possible except not registering events
1057for potentially C<dup ()>'ed file descriptors, or to resort to
1058C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
1059
1060=head3 The special problem of fork
1061
1062Some backends (epoll, kqueue) do not support C<fork ()> at all or exhibit
1063useless behaviour. Libev fully supports fork, but needs to be told about
1064it in the child.
1065
1066To support fork in your programs, you either have to call
1067C<ev_default_fork ()> or C<ev_loop_fork ()> after a fork in the child,
1068enable C<EVFLAG_FORKCHECK>, or resort to C<EVBACKEND_SELECT> or
1069C<EVBACKEND_POLL>.
1070
937 1071
938=head3 Watcher-Specific Functions 1072=head3 Watcher-Specific Functions
939 1073
940=over 4 1074=over 4
941 1075
… …
954=item int events [read-only] 1088=item int events [read-only]
955 1089
956The events being watched. 1090The events being watched.
957 1091
958=back 1092=back
1093
1094=head3 Examples
959 1095
960Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well 1096Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
961readable, but only once. Since it is likely line-buffered, you could 1097readable, but only once. Since it is likely line-buffered, you could
962attempt to read a whole line in the callback. 1098attempt to read a whole line in the callback.
963 1099
… …
1061or C<ev_timer_again> is called and determines the next timeout (if any), 1197or C<ev_timer_again> is called and determines the next timeout (if any),
1062which is also when any modifications are taken into account. 1198which is also when any modifications are taken into account.
1063 1199
1064=back 1200=back
1065 1201
1202=head3 Examples
1203
1066Example: Create a timer that fires after 60 seconds. 1204Example: Create a timer that fires after 60 seconds.
1067 1205
1068 static void 1206 static void
1069 one_minute_cb (struct ev_loop *loop, struct ev_timer *w, int revents) 1207 one_minute_cb (struct ev_loop *loop, struct ev_timer *w, int revents)
1070 { 1208 {
… …
1220 1358
1221The current reschedule callback, or C<0>, if this functionality is 1359The current reschedule callback, or C<0>, if this functionality is
1222switched off. Can be changed any time, but changes only take effect when 1360switched off. Can be changed any time, but changes only take effect when
1223the periodic timer fires or C<ev_periodic_again> is being called. 1361the periodic timer fires or C<ev_periodic_again> is being called.
1224 1362
1363=item ev_tstamp at [read-only]
1364
1365When active, contains the absolute time that the watcher is supposed to
1366trigger next.
1367
1225=back 1368=back
1369
1370=head3 Examples
1226 1371
1227Example: Call a callback every hour, or, more precisely, whenever the 1372Example: Call a callback every hour, or, more precisely, whenever the
1228system clock is divisible by 3600. The callback invocation times have 1373system clock is divisible by 3600. The callback invocation times have
1229potentially a lot of jittering, but good long-term stability. 1374potentially a lot of jittering, but good long-term stability.
1230 1375
… …
1297 1442
1298=head3 Watcher-Specific Functions and Data Members 1443=head3 Watcher-Specific Functions and Data Members
1299 1444
1300=over 4 1445=over 4
1301 1446
1302=item ev_child_init (ev_child *, callback, int pid) 1447=item ev_child_init (ev_child *, callback, int pid, int trace)
1303 1448
1304=item ev_child_set (ev_child *, int pid) 1449=item ev_child_set (ev_child *, int pid, int trace)
1305 1450
1306Configures the watcher to wait for status changes of process C<pid> (or 1451Configures the watcher to wait for status changes of process C<pid> (or
1307I<any> process if C<pid> is specified as C<0>). The callback can look 1452I<any> process if C<pid> is specified as C<0>). The callback can look
1308at the C<rstatus> member of the C<ev_child> watcher structure to see 1453at the C<rstatus> member of the C<ev_child> watcher structure to see
1309the status word (use the macros from C<sys/wait.h> and see your systems 1454the status word (use the macros from C<sys/wait.h> and see your systems
1310C<waitpid> documentation). The C<rpid> member contains the pid of the 1455C<waitpid> documentation). The C<rpid> member contains the pid of the
1311process causing the status change. 1456process causing the status change. C<trace> must be either C<0> (only
1457activate the watcher when the process terminates) or C<1> (additionally
1458activate the watcher when the process is stopped or continued).
1312 1459
1313=item int pid [read-only] 1460=item int pid [read-only]
1314 1461
1315The process id this watcher watches out for, or C<0>, meaning any process id. 1462The process id this watcher watches out for, or C<0>, meaning any process id.
1316 1463
… …
1322 1469
1323The process exit/trace status caused by C<rpid> (see your systems 1470The process exit/trace status caused by C<rpid> (see your systems
1324C<waitpid> and C<sys/wait.h> documentation for details). 1471C<waitpid> and C<sys/wait.h> documentation for details).
1325 1472
1326=back 1473=back
1474
1475=head3 Examples
1327 1476
1328Example: Try to exit cleanly on SIGINT and SIGTERM. 1477Example: Try to exit cleanly on SIGINT and SIGTERM.
1329 1478
1330 static void 1479 static void
1331 sigint_cb (struct ev_loop *loop, struct ev_signal *w, int revents) 1480 sigint_cb (struct ev_loop *loop, struct ev_signal *w, int revents)
… …
1372semantics of C<ev_stat> watchers, which means that libev sometimes needs 1521semantics of C<ev_stat> watchers, which means that libev sometimes needs
1373to fall back to regular polling again even with inotify, but changes are 1522to fall back to regular polling again even with inotify, but changes are
1374usually detected immediately, and if the file exists there will be no 1523usually detected immediately, and if the file exists there will be no
1375polling. 1524polling.
1376 1525
1526=head3 Inotify
1527
1528When C<inotify (7)> support has been compiled into libev (generally only
1529available on Linux) and present at runtime, it will be used to speed up
1530change detection where possible. The inotify descriptor will be created lazily
1531when the first C<ev_stat> watcher is being started.
1532
1533Inotify presense does not change the semantics of C<ev_stat> watchers
1534except that changes might be detected earlier, and in some cases, to avoid
1535making regular C<stat> calls. Even in the presense of inotify support
1536there are many cases where libev has to resort to regular C<stat> polling.
1537
1538(There is no support for kqueue, as apparently it cannot be used to
1539implement this functionality, due to the requirement of having a file
1540descriptor open on the object at all times).
1541
1542=head3 The special problem of stat time resolution
1543
1544The C<stat ()> syscall only supports full-second resolution portably, and
1545even on systems where the resolution is higher, many filesystems still
1546only support whole seconds.
1547
1548That means that, if the time is the only thing that changes, you might
1549miss updates: on the first update, C<ev_stat> detects a change and calls
1550your callback, which does something. When there is another update within
1551the same second, C<ev_stat> will be unable to detect it.
1552
1553The solution to this is to delay acting on a change for a second (or till
1554the next second boundary), using a roughly one-second delay C<ev_timer>
1555(C<ev_timer_set (w, 0., 1.01); ev_timer_again (loop, w)>). The C<.01>
1556is added to work around small timing inconsistencies of some operating
1557systems.
1558
1377=head3 Watcher-Specific Functions and Data Members 1559=head3 Watcher-Specific Functions and Data Members
1378 1560
1379=over 4 1561=over 4
1380 1562
1381=item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval) 1563=item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
… …
1418=item const char *path [read-only] 1600=item const char *path [read-only]
1419 1601
1420The filesystem path that is being watched. 1602The filesystem path that is being watched.
1421 1603
1422=back 1604=back
1605
1606=head3 Examples
1423 1607
1424Example: Watch C</etc/passwd> for attribute changes. 1608Example: Watch C</etc/passwd> for attribute changes.
1425 1609
1426 static void 1610 static void
1427 passwd_cb (struct ev_loop *loop, ev_stat *w, int revents) 1611 passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
… …
1440 } 1624 }
1441 1625
1442 ... 1626 ...
1443 ev_stat passwd; 1627 ev_stat passwd;
1444 1628
1445 ev_stat_init (&passwd, passwd_cb, "/etc/passwd"); 1629 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
1446 ev_stat_start (loop, &passwd); 1630 ev_stat_start (loop, &passwd);
1631
1632Example: Like above, but additionally use a one-second delay so we do not
1633miss updates (however, frequent updates will delay processing, too, so
1634one might do the work both on C<ev_stat> callback invocation I<and> on
1635C<ev_timer> callback invocation).
1636
1637 static ev_stat passwd;
1638 static ev_timer timer;
1639
1640 static void
1641 timer_cb (EV_P_ ev_timer *w, int revents)
1642 {
1643 ev_timer_stop (EV_A_ w);
1644
1645 /* now it's one second after the most recent passwd change */
1646 }
1647
1648 static void
1649 stat_cb (EV_P_ ev_stat *w, int revents)
1650 {
1651 /* reset the one-second timer */
1652 ev_timer_again (EV_A_ &timer);
1653 }
1654
1655 ...
1656 ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
1657 ev_stat_start (loop, &passwd);
1658 ev_timer_init (&timer, timer_cb, 0., 1.01);
1447 1659
1448 1660
1449=head2 C<ev_idle> - when you've got nothing better to do... 1661=head2 C<ev_idle> - when you've got nothing better to do...
1450 1662
1451Idle watchers trigger events when no other events of the same or higher 1663Idle watchers trigger events when no other events of the same or higher
… …
1477kind. There is a C<ev_idle_set> macro, but using it is utterly pointless, 1689kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
1478believe me. 1690believe me.
1479 1691
1480=back 1692=back
1481 1693
1694=head3 Examples
1695
1482Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the 1696Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
1483callback, free it. Also, use no error checking, as usual. 1697callback, free it. Also, use no error checking, as usual.
1484 1698
1485 static void 1699 static void
1486 idle_cb (struct ev_loop *loop, struct ev_idle *w, int revents) 1700 idle_cb (struct ev_loop *loop, struct ev_idle *w, int revents)
1487 { 1701 {
1488 free (w); 1702 free (w);
1489 // now do something you wanted to do when the program has 1703 // now do something you wanted to do when the program has
1490 // no longer asnything immediate to do. 1704 // no longer anything immediate to do.
1491 } 1705 }
1492 1706
1493 struct ev_idle *idle_watcher = malloc (sizeof (struct ev_idle)); 1707 struct ev_idle *idle_watcher = malloc (sizeof (struct ev_idle));
1494 ev_idle_init (idle_watcher, idle_cb); 1708 ev_idle_init (idle_watcher, idle_cb);
1495 ev_idle_start (loop, idle_cb); 1709 ev_idle_start (loop, idle_cb);
… …
1537 1751
1538It is recommended to give C<ev_check> watchers highest (C<EV_MAXPRI>) 1752It is recommended to give C<ev_check> watchers highest (C<EV_MAXPRI>)
1539priority, to ensure that they are being run before any other watchers 1753priority, to ensure that they are being run before any other watchers
1540after the poll. Also, C<ev_check> watchers (and C<ev_prepare> watchers, 1754after the poll. Also, C<ev_check> watchers (and C<ev_prepare> watchers,
1541too) should not activate ("feed") events into libev. While libev fully 1755too) should not activate ("feed") events into libev. While libev fully
1542supports this, they will be called before other C<ev_check> watchers did 1756supports this, they will be called before other C<ev_check> watchers
1543their job. As C<ev_check> watchers are often used to embed other event 1757did their job. As C<ev_check> watchers are often used to embed other
1544loops those other event loops might be in an unusable state until their 1758(non-libev) event loops those other event loops might be in an unusable
1545C<ev_check> watcher ran (always remind yourself to coexist peacefully with 1759state until their C<ev_check> watcher ran (always remind yourself to
1546others). 1760coexist peacefully with others).
1547 1761
1548=head3 Watcher-Specific Functions and Data Members 1762=head3 Watcher-Specific Functions and Data Members
1549 1763
1550=over 4 1764=over 4
1551 1765
… …
1556Initialises and configures the prepare or check watcher - they have no 1770Initialises and configures the prepare or check watcher - they have no
1557parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set> 1771parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
1558macros, but using them is utterly, utterly and completely pointless. 1772macros, but using them is utterly, utterly and completely pointless.
1559 1773
1560=back 1774=back
1775
1776=head3 Examples
1561 1777
1562There are a number of principal ways to embed other event loops or modules 1778There are a number of principal ways to embed other event loops or modules
1563into libev. Here are some ideas on how to include libadns into libev 1779into libev. Here are some ideas on how to include libadns into libev
1564(there is a Perl module named C<EV::ADNS> that does this, which you could 1780(there is a Perl module named C<EV::ADNS> that does this, which you could
1565use for an actually working example. Another Perl module named C<EV::Glib> 1781use for an actually working example. Another Perl module named C<EV::Glib>
… …
1734portable one. 1950portable one.
1735 1951
1736So when you want to use this feature you will always have to be prepared 1952So when you want to use this feature you will always have to be prepared
1737that you cannot get an embeddable loop. The recommended way to get around 1953that you cannot get an embeddable loop. The recommended way to get around
1738this is to have a separate variables for your embeddable loop, try to 1954this is to have a separate variables for your embeddable loop, try to
1739create it, and if that fails, use the normal loop for everything: 1955create it, and if that fails, use the normal loop for everything.
1956
1957=head3 Watcher-Specific Functions and Data Members
1958
1959=over 4
1960
1961=item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
1962
1963=item ev_embed_set (ev_embed *, callback, struct ev_loop *embedded_loop)
1964
1965Configures the watcher to embed the given loop, which must be
1966embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
1967invoked automatically, otherwise it is the responsibility of the callback
1968to invoke it (it will continue to be called until the sweep has been done,
1969if you do not want thta, you need to temporarily stop the embed watcher).
1970
1971=item ev_embed_sweep (loop, ev_embed *)
1972
1973Make a single, non-blocking sweep over the embedded loop. This works
1974similarly to C<ev_loop (embedded_loop, EVLOOP_NONBLOCK)>, but in the most
1975apropriate way for embedded loops.
1976
1977=item struct ev_loop *other [read-only]
1978
1979The embedded event loop.
1980
1981=back
1982
1983=head3 Examples
1984
1985Example: Try to get an embeddable event loop and embed it into the default
1986event loop. If that is not possible, use the default loop. The default
1987loop is stored in C<loop_hi>, while the mebeddable loop is stored in
1988C<loop_lo> (which is C<loop_hi> in the acse no embeddable loop can be
1989used).
1740 1990
1741 struct ev_loop *loop_hi = ev_default_init (0); 1991 struct ev_loop *loop_hi = ev_default_init (0);
1742 struct ev_loop *loop_lo = 0; 1992 struct ev_loop *loop_lo = 0;
1743 struct ev_embed embed; 1993 struct ev_embed embed;
1744 1994
… …
1755 ev_embed_start (loop_hi, &embed); 2005 ev_embed_start (loop_hi, &embed);
1756 } 2006 }
1757 else 2007 else
1758 loop_lo = loop_hi; 2008 loop_lo = loop_hi;
1759 2009
1760=head3 Watcher-Specific Functions and Data Members 2010Example: Check if kqueue is available but not recommended and create
2011a kqueue backend for use with sockets (which usually work with any
2012kqueue implementation). Store the kqueue/socket-only event loop in
2013C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
1761 2014
1762=over 4 2015 struct ev_loop *loop = ev_default_init (0);
2016 struct ev_loop *loop_socket = 0;
2017 struct ev_embed embed;
2018
2019 if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
2020 if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
2021 {
2022 ev_embed_init (&embed, 0, loop_socket);
2023 ev_embed_start (loop, &embed);
2024 }
1763 2025
1764=item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop) 2026 if (!loop_socket)
2027 loop_socket = loop;
1765 2028
1766=item ev_embed_set (ev_embed *, callback, struct ev_loop *embedded_loop) 2029 // now use loop_socket for all sockets, and loop for everything else
1767
1768Configures the watcher to embed the given loop, which must be
1769embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
1770invoked automatically, otherwise it is the responsibility of the callback
1771to invoke it (it will continue to be called until the sweep has been done,
1772if you do not want thta, you need to temporarily stop the embed watcher).
1773
1774=item ev_embed_sweep (loop, ev_embed *)
1775
1776Make a single, non-blocking sweep over the embedded loop. This works
1777similarly to C<ev_loop (embedded_loop, EVLOOP_NONBLOCK)>, but in the most
1778apropriate way for embedded loops.
1779
1780=item struct ev_loop *loop [read-only]
1781
1782The embedded event loop.
1783
1784=back
1785 2030
1786 2031
1787=head2 C<ev_fork> - the audacity to resume the event loop after a fork 2032=head2 C<ev_fork> - the audacity to resume the event loop after a fork
1788 2033
1789Fork watchers are called when a C<fork ()> was detected (usually because 2034Fork watchers are called when a C<fork ()> was detected (usually because
… …
1805believe me. 2050believe me.
1806 2051
1807=back 2052=back
1808 2053
1809 2054
2055=head2 C<ev_async> - how to wake up another event loop
2056
2057In general, you cannot use an C<ev_loop> from multiple threads or other
2058asynchronous sources such as signal handlers (as opposed to multiple event
2059loops - those are of course safe to use in different threads).
2060
2061Sometimes, however, you need to wake up another event loop you do not
2062control, for example because it belongs to another thread. This is what
2063C<ev_async> watchers do: as long as the C<ev_async> watcher is active, you
2064can signal it by calling C<ev_async_send>, which is thread- and signal
2065safe.
2066
2067This functionality is very similar to C<ev_signal> watchers, as signals,
2068too, are asynchronous in nature, and signals, too, will be compressed
2069(i.e. the number of callback invocations may be less than the number of
2070C<ev_async_sent> calls).
2071
2072Unlike C<ev_signal> watchers, C<ev_async> works with any event loop, not
2073just the default loop.
2074
2075=head3 Queueing
2076
2077C<ev_async> does not support queueing of data in any way. The reason
2078is that the author does not know of a simple (or any) algorithm for a
2079multiple-writer-single-reader queue that works in all cases and doesn't
2080need elaborate support such as pthreads.
2081
2082That means that if you want to queue data, you have to provide your own
2083queue. And here is how you would implement locking:
2084
2085=over 4
2086
2087=item queueing from a signal handler context
2088
2089To implement race-free queueing, you simply add to the queue in the signal
2090handler but you block the signal handler in the watcher callback. Here is an example that does that for
2091some fictitiuous SIGUSR1 handler:
2092
2093 static ev_async mysig;
2094
2095 static void
2096 sigusr1_handler (void)
2097 {
2098 sometype data;
2099
2100 // no locking etc.
2101 queue_put (data);
2102 ev_async_send (DEFAULT_ &mysig);
2103 }
2104
2105 static void
2106 mysig_cb (EV_P_ ev_async *w, int revents)
2107 {
2108 sometype data;
2109 sigset_t block, prev;
2110
2111 sigemptyset (&block);
2112 sigaddset (&block, SIGUSR1);
2113 sigprocmask (SIG_BLOCK, &block, &prev);
2114
2115 while (queue_get (&data))
2116 process (data);
2117
2118 if (sigismember (&prev, SIGUSR1)
2119 sigprocmask (SIG_UNBLOCK, &block, 0);
2120 }
2121
2122(Note: pthreads in theory requires you to use C<pthread_setmask>
2123instead of C<sigprocmask> when you use threads, but libev doesn't do it
2124either...).
2125
2126=item queueing from a thread context
2127
2128The strategy for threads is different, as you cannot (easily) block
2129threads but you can easily preempt them, so to queue safely you need to
2130emply a traditional mutex lock, such as in this pthread example:
2131
2132 static ev_async mysig;
2133 static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
2134
2135 static void
2136 otherthread (void)
2137 {
2138 // only need to lock the actual queueing operation
2139 pthread_mutex_lock (&mymutex);
2140 queue_put (data);
2141 pthread_mutex_unlock (&mymutex);
2142
2143 ev_async_send (DEFAULT_ &mysig);
2144 }
2145
2146 static void
2147 mysig_cb (EV_P_ ev_async *w, int revents)
2148 {
2149 pthread_mutex_lock (&mymutex);
2150
2151 while (queue_get (&data))
2152 process (data);
2153
2154 pthread_mutex_unlock (&mymutex);
2155 }
2156
2157=back
2158
2159
2160=head3 Watcher-Specific Functions and Data Members
2161
2162=over 4
2163
2164=item ev_async_init (ev_async *, callback)
2165
2166Initialises and configures the async watcher - it has no parameters of any
2167kind. There is a C<ev_asynd_set> macro, but using it is utterly pointless,
2168believe me.
2169
2170=item ev_async_send (loop, ev_async *)
2171
2172Sends/signals/activates the given C<ev_async> watcher, that is, feeds
2173an C<EV_ASYNC> event on the watcher into the event loop. Unlike
2174C<ev_feed_event>, this call is safe to do in other threads, signal or
2175similar contexts (see the dicusssion of C<EV_ATOMIC_T> in the embedding
2176section below on what exactly this means).
2177
2178This call incurs the overhead of a syscall only once per loop iteration,
2179so while the overhead might be noticable, it doesn't apply to repeated
2180calls to C<ev_async_send>.
2181
2182=back
2183
2184
1810=head1 OTHER FUNCTIONS 2185=head1 OTHER FUNCTIONS
1811 2186
1812There are some other functions of possible interest. Described. Here. Now. 2187There are some other functions of possible interest. Described. Here. Now.
1813 2188
1814=over 4 2189=over 4
… …
2041Example: Define a class with an IO and idle watcher, start one of them in 2416Example: Define a class with an IO and idle watcher, start one of them in
2042the constructor. 2417the constructor.
2043 2418
2044 class myclass 2419 class myclass
2045 { 2420 {
2046 ev_io io; void io_cb (ev::io &w, int revents); 2421 ev::io io; void io_cb (ev::io &w, int revents);
2047 ev_idle idle void idle_cb (ev::idle &w, int revents); 2422 ev:idle idle void idle_cb (ev::idle &w, int revents);
2048 2423
2049 myclass (); 2424 myclass (int fd)
2050 }
2051
2052 myclass::myclass (int fd)
2053 { 2425 {
2054 io .set <myclass, &myclass::io_cb > (this); 2426 io .set <myclass, &myclass::io_cb > (this);
2055 idle.set <myclass, &myclass::idle_cb> (this); 2427 idle.set <myclass, &myclass::idle_cb> (this);
2056 2428
2057 io.start (fd, ev::READ); 2429 io.start (fd, ev::READ);
2430 }
2058 } 2431 };
2059 2432
2060 2433
2061=head1 MACRO MAGIC 2434=head1 MACRO MAGIC
2062 2435
2063Libev can be compiled with a variety of options, the most fundamantal 2436Libev can be compiled with a variety of options, the most fundamantal
… …
2124Libev can (and often is) directly embedded into host 2497Libev can (and often is) directly embedded into host
2125applications. Examples of applications that embed it include the Deliantra 2498applications. Examples of applications that embed it include the Deliantra
2126Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe) 2499Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
2127and rxvt-unicode. 2500and rxvt-unicode.
2128 2501
2129The goal is to enable you to just copy the neecssary files into your 2502The goal is to enable you to just copy the necessary files into your
2130source directory without having to change even a single line in them, so 2503source directory without having to change even a single line in them, so
2131you can easily upgrade by simply copying (or having a checked-out copy of 2504you can easily upgrade by simply copying (or having a checked-out copy of
2132libev somewhere in your source tree). 2505libev somewhere in your source tree).
2133 2506
2134=head2 FILESETS 2507=head2 FILESETS
… …
2224 2597
2225If defined to be C<1>, libev will try to detect the availability of the 2598If defined to be C<1>, libev will try to detect the availability of the
2226monotonic clock option at both compiletime and runtime. Otherwise no use 2599monotonic clock option at both compiletime and runtime. Otherwise no use
2227of the monotonic clock option will be attempted. If you enable this, you 2600of the monotonic clock option will be attempted. If you enable this, you
2228usually have to link against librt or something similar. Enabling it when 2601usually have to link against librt or something similar. Enabling it when
2229the functionality isn't available is safe, though, althoguh you have 2602the functionality isn't available is safe, though, although you have
2230to make sure you link against any libraries where the C<clock_gettime> 2603to make sure you link against any libraries where the C<clock_gettime>
2231function is hiding in (often F<-lrt>). 2604function is hiding in (often F<-lrt>).
2232 2605
2233=item EV_USE_REALTIME 2606=item EV_USE_REALTIME
2234 2607
2235If defined to be C<1>, libev will try to detect the availability of the 2608If defined to be C<1>, libev will try to detect the availability of the
2236realtime clock option at compiletime (and assume its availability at 2609realtime clock option at compiletime (and assume its availability at
2237runtime if successful). Otherwise no use of the realtime clock option will 2610runtime if successful). Otherwise no use of the realtime clock option will
2238be attempted. This effectively replaces C<gettimeofday> by C<clock_get 2611be attempted. This effectively replaces C<gettimeofday> by C<clock_get
2239(CLOCK_REALTIME, ...)> and will not normally affect correctness. See tzhe note about libraries 2612(CLOCK_REALTIME, ...)> and will not normally affect correctness. See the
2240in the description of C<EV_USE_MONOTONIC>, though. 2613note about libraries in the description of C<EV_USE_MONOTONIC>, though.
2614
2615=item EV_USE_NANOSLEEP
2616
2617If defined to be C<1>, libev will assume that C<nanosleep ()> is available
2618and will use it for delays. Otherwise it will use C<select ()>.
2241 2619
2242=item EV_USE_SELECT 2620=item EV_USE_SELECT
2243 2621
2244If undefined or defined to be C<1>, libev will compile in support for the 2622If undefined or defined to be C<1>, libev will compile in support for the
2245C<select>(2) backend. No attempt at autodetection will be done: if no 2623C<select>(2) backend. No attempt at autodetection will be done: if no
… …
2263wants osf handles on win32 (this is the case when the select to 2641wants osf handles on win32 (this is the case when the select to
2264be used is the winsock select). This means that it will call 2642be used is the winsock select). This means that it will call
2265C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise, 2643C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
2266it is assumed that all these functions actually work on fds, even 2644it is assumed that all these functions actually work on fds, even
2267on win32. Should not be defined on non-win32 platforms. 2645on win32. Should not be defined on non-win32 platforms.
2646
2647=item EV_FD_TO_WIN32_HANDLE
2648
2649If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
2650file descriptors to socket handles. When not defining this symbol (the
2651default), then libev will call C<_get_osfhandle>, which is usually
2652correct. In some cases, programs use their own file descriptor management,
2653in which case they can provide this function to map fds to socket handles.
2268 2654
2269=item EV_USE_POLL 2655=item EV_USE_POLL
2270 2656
2271If defined to be C<1>, libev will compile in support for the C<poll>(2) 2657If defined to be C<1>, libev will compile in support for the C<poll>(2)
2272backend. Otherwise it will be enabled on non-win32 platforms. It 2658backend. Otherwise it will be enabled on non-win32 platforms. It
… …
2306 2692
2307If defined to be C<1>, libev will compile in support for the Linux inotify 2693If defined to be C<1>, libev will compile in support for the Linux inotify
2308interface to speed up C<ev_stat> watchers. Its actual availability will 2694interface to speed up C<ev_stat> watchers. Its actual availability will
2309be detected at runtime. 2695be detected at runtime.
2310 2696
2697=item EV_ATOMIC_T
2698
2699Libev requires an integer type (suitable for storing C<0> or C<1>) whose
2700access is atomic with respect to other threads or signal contexts. No such
2701type is easily found in the C language, so you can provide your own type
2702that you know is safe for your purposes. It is used both for signal handler "locking"
2703as well as for signal and thread safety in C<ev_async> watchers.
2704
2705In the absense of this define, libev will use C<sig_atomic_t volatile>
2706(from F<signal.h>), which is usually good enough on most platforms.
2707
2311=item EV_H 2708=item EV_H
2312 2709
2313The name of the F<ev.h> header file used to include it. The default if 2710The name of the F<ev.h> header file used to include it. The default if
2314undefined is C<< <ev.h> >> in F<event.h> and C<"ev.h"> in F<ev.c>. This 2711undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
2315can be used to virtually rename the F<ev.h> header file in case of conflicts. 2712used to virtually rename the F<ev.h> header file in case of conflicts.
2316 2713
2317=item EV_CONFIG_H 2714=item EV_CONFIG_H
2318 2715
2319If C<EV_STANDALONE> isn't C<1>, this variable can be used to override 2716If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
2320F<ev.c>'s idea of where to find the F<config.h> file, similarly to 2717F<ev.c>'s idea of where to find the F<config.h> file, similarly to
2321C<EV_H>, above. 2718C<EV_H>, above.
2322 2719
2323=item EV_EVENT_H 2720=item EV_EVENT_H
2324 2721
2325Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea 2722Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
2326of how the F<event.h> header can be found. 2723of how the F<event.h> header can be found, the default is C<"event.h">.
2327 2724
2328=item EV_PROTOTYPES 2725=item EV_PROTOTYPES
2329 2726
2330If defined to be C<0>, then F<ev.h> will not define any function 2727If defined to be C<0>, then F<ev.h> will not define any function
2331prototypes, but still define all the structs and other symbols. This is 2728prototypes, but still define all the structs and other symbols. This is
… …
2382=item EV_FORK_ENABLE 2779=item EV_FORK_ENABLE
2383 2780
2384If undefined or defined to be C<1>, then fork watchers are supported. If 2781If undefined or defined to be C<1>, then fork watchers are supported. If
2385defined to be C<0>, then they are not. 2782defined to be C<0>, then they are not.
2386 2783
2784=item EV_ASYNC_ENABLE
2785
2786If undefined or defined to be C<1>, then async watchers are supported. If
2787defined to be C<0>, then they are not.
2788
2387=item EV_MINIMAL 2789=item EV_MINIMAL
2388 2790
2389If you need to shave off some kilobytes of code at the expense of some 2791If you need to shave off some kilobytes of code at the expense of some
2390speed, define this symbol to C<1>. Currently only used for gcc to override 2792speed, define this symbol to C<1>. Currently only used for gcc to override
2391some inlining decisions, saves roughly 30% codesize of amd64. 2793some inlining decisions, saves roughly 30% codesize of amd64.
… …
2397than enough. If you need to manage thousands of children you might want to 2799than enough. If you need to manage thousands of children you might want to
2398increase this value (I<must> be a power of two). 2800increase this value (I<must> be a power of two).
2399 2801
2400=item EV_INOTIFY_HASHSIZE 2802=item EV_INOTIFY_HASHSIZE
2401 2803
2402C<ev_staz> watchers use a small hash table to distribute workload by 2804C<ev_stat> watchers use a small hash table to distribute workload by
2403inotify watch id. The default size is C<16> (or C<1> with C<EV_MINIMAL>), 2805inotify watch id. The default size is C<16> (or C<1> with C<EV_MINIMAL>),
2404usually more than enough. If you need to manage thousands of C<ev_stat> 2806usually more than enough. If you need to manage thousands of C<ev_stat>
2405watchers you might want to increase this value (I<must> be a power of 2807watchers you might want to increase this value (I<must> be a power of
2406two). 2808two).
2407 2809
… …
2424 2826
2425=item ev_set_cb (ev, cb) 2827=item ev_set_cb (ev, cb)
2426 2828
2427Can be used to change the callback member declaration in each watcher, 2829Can be used to change the callback member declaration in each watcher,
2428and the way callbacks are invoked and set. Must expand to a struct member 2830and the way callbacks are invoked and set. Must expand to a struct member
2429definition and a statement, respectively. See the F<ev.v> header file for 2831definition and a statement, respectively. See the F<ev.h> header file for
2430their default definitions. One possible use for overriding these is to 2832their default definitions. One possible use for overriding these is to
2431avoid the C<struct ev_loop *> as first argument in all cases, or to use 2833avoid the C<struct ev_loop *> as first argument in all cases, or to use
2432method calls instead of plain function calls in C++. 2834method calls instead of plain function calls in C++.
2835
2836=head2 EXPORTED API SYMBOLS
2837
2838If you need to re-export the API (e.g. via a dll) and you need a list of
2839exported symbols, you can use the provided F<Symbol.*> files which list
2840all public symbols, one per line:
2841
2842 Symbols.ev for libev proper
2843 Symbols.event for the libevent emulation
2844
2845This can also be used to rename all public symbols to avoid clashes with
2846multiple versions of libev linked together (which is obviously bad in
2847itself, but sometimes it is inconvinient to avoid this).
2848
2849A sed command like this will create wrapper C<#define>'s that you need to
2850include before including F<ev.h>:
2851
2852 <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
2853
2854This would create a file F<wrap.h> which essentially looks like this:
2855
2856 #define ev_backend myprefix_ev_backend
2857 #define ev_check_start myprefix_ev_check_start
2858 #define ev_check_stop myprefix_ev_check_stop
2859 ...
2433 2860
2434=head2 EXAMPLES 2861=head2 EXAMPLES
2435 2862
2436For a real-world example of a program the includes libev 2863For a real-world example of a program the includes libev
2437verbatim, you can have a look at the EV perl module 2864verbatim, you can have a look at the EV perl module
… …
2478 2905
2479=item Starting and stopping timer/periodic watchers: O(log skipped_other_timers) 2906=item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
2480 2907
2481This means that, when you have a watcher that triggers in one hour and 2908This means that, when you have a watcher that triggers in one hour and
2482there are 100 watchers that would trigger before that then inserting will 2909there are 100 watchers that would trigger before that then inserting will
2483have to skip those 100 watchers. 2910have to skip roughly seven (C<ld 100>) of these watchers.
2484 2911
2485=item Changing timer/periodic watchers (by autorepeat, again): O(log skipped_other_timers) 2912=item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
2486 2913
2487That means that for changing a timer costs less than removing/adding them 2914That means that changing a timer costs less than removing/adding them
2488as only the relative motion in the event queue has to be paid for. 2915as only the relative motion in the event queue has to be paid for.
2489 2916
2490=item Starting io/check/prepare/idle/signal/child watchers: O(1) 2917=item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
2491 2918
2492These just add the watcher into an array or at the head of a list. 2919These just add the watcher into an array or at the head of a list.
2920
2493=item Stopping check/prepare/idle watchers: O(1) 2921=item Stopping check/prepare/idle/fork/async watchers: O(1)
2494 2922
2495=item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE)) 2923=item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
2496 2924
2497These watchers are stored in lists then need to be walked to find the 2925These watchers are stored in lists then need to be walked to find the
2498correct watcher to remove. The lists are usually short (you don't usually 2926correct watcher to remove. The lists are usually short (you don't usually
2499have many watchers waiting for the same fd or signal). 2927have many watchers waiting for the same fd or signal).
2500 2928
2501=item Finding the next timer per loop iteration: O(1) 2929=item Finding the next timer in each loop iteration: O(1)
2930
2931By virtue of using a binary heap, the next timer is always found at the
2932beginning of the storage array.
2502 2933
2503=item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd) 2934=item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
2504 2935
2505A change means an I/O watcher gets started or stopped, which requires 2936A change means an I/O watcher gets started or stopped, which requires
2506libev to recalculate its status (and possibly tell the kernel). 2937libev to recalculate its status (and possibly tell the kernel, depending
2938on backend and wether C<ev_io_set> was used).
2507 2939
2508=item Activating one watcher: O(1) 2940=item Activating one watcher (putting it into the pending state): O(1)
2509 2941
2510=item Priority handling: O(number_of_priorities) 2942=item Priority handling: O(number_of_priorities)
2511 2943
2512Priorities are implemented by allocating some space for each 2944Priorities are implemented by allocating some space for each
2513priority. When doing priority-based operations, libev usually has to 2945priority. When doing priority-based operations, libev usually has to
2514linearly search all the priorities. 2946linearly search all the priorities, but starting/stopping and activating
2947watchers becomes O(1) w.r.t. prioritiy handling.
2948
2949=item Sending an ev_async: O(1)
2950
2951=item Processing ev_async_send: O(number_of_async_watchers)
2952
2953=item Processing signals: O(max_signal_number)
2954
2955Sending involves a syscall I<iff> there were no other C<ev_async_send>
2956calls in the current loop iteration. Checking for async and signal events
2957involves iterating over all running async watchers or all signal numbers.
2515 2958
2516=back 2959=back
2517 2960
2518 2961
2962=head1 Win32 platform limitations and workarounds
2963
2964Win32 doesn't support any of the standards (e.g. POSIX) that libev
2965requires, and its I/O model is fundamentally incompatible with the POSIX
2966model. Libev still offers limited functionality on this platform in
2967the form of the C<EVBACKEND_SELECT> backend, and only supports socket
2968descriptors. This only applies when using Win32 natively, not when using
2969e.g. cygwin.
2970
2971There is no supported compilation method available on windows except
2972embedding it into other applications.
2973
2974Due to the many, low, and arbitrary limits on the win32 platform and the
2975abysmal performance of winsockets, using a large number of sockets is not
2976recommended (and not reasonable). If your program needs to use more than
2977a hundred or so sockets, then likely it needs to use a totally different
2978implementation for windows, as libev offers the POSIX model, which cannot
2979be implemented efficiently on windows (microsoft monopoly games).
2980
2981=over 4
2982
2983=item The winsocket select function
2984
2985The winsocket C<select> function doesn't follow POSIX in that it requires
2986socket I<handles> and not socket I<file descriptors>. This makes select
2987very inefficient, and also requires a mapping from file descriptors
2988to socket handles. See the discussion of the C<EV_SELECT_USE_FD_SET>,
2989C<EV_SELECT_IS_WINSOCKET> and C<EV_FD_TO_WIN32_HANDLE> preprocessor
2990symbols for more info.
2991
2992The configuration for a "naked" win32 using the microsoft runtime
2993libraries and raw winsocket select is:
2994
2995 #define EV_USE_SELECT 1
2996 #define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
2997
2998Note that winsockets handling of fd sets is O(n), so you can easily get a
2999complexity in the O(n²) range when using win32.
3000
3001=item Limited number of file descriptors
3002
3003Windows has numerous arbitrary (and low) limits on things. Early versions
3004of winsocket's select only supported waiting for a max. of C<64> handles
3005(probably owning to the fact that all windows kernels can only wait for
3006C<64> things at the same time internally; microsoft recommends spawning a
3007chain of threads and wait for 63 handles and the previous thread in each).
3008
3009Newer versions support more handles, but you need to define C<FD_SETSIZE>
3010to some high number (e.g. C<2048>) before compiling the winsocket select
3011call (which might be in libev or elsewhere, for example, perl does its own
3012select emulation on windows).
3013
3014Another limit is the number of file descriptors in the microsoft runtime
3015libraries, which by default is C<64> (there must be a hidden I<64> fetish
3016or something like this inside microsoft). You can increase this by calling
3017C<_setmaxstdio>, which can increase this limit to C<2048> (another
3018arbitrary limit), but is broken in many versions of the microsoft runtime
3019libraries.
3020
3021This might get you to about C<512> or C<2048> sockets (depending on
3022windows version and/or the phase of the moon). To get more, you need to
3023wrap all I/O functions and provide your own fd management, but the cost of
3024calling select (O(n²)) will likely make this unworkable.
3025
3026=back
3027
3028
2519=head1 AUTHOR 3029=head1 AUTHOR
2520 3030
2521Marc Lehmann <libev@schmorp.de>. 3031Marc Lehmann <libev@schmorp.de>.
2522 3032

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines