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

Comparing libev/ev.pod (file contents):
Revision 1.86 by root, Tue Dec 18 01:20:33 2007 UTC vs.
Revision 1.114 by root, Mon Dec 31 01:31:30 2007 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
… …
115 115
116Returns 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
117C<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
118you actually want to know. 118you actually want to know.
119 119
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 ()>.
125
120=item int ev_version_major () 126=item int ev_version_major ()
121 127
122=item int ev_version_minor () 128=item int ev_version_minor ()
123 129
124You can find out the major and minor ABI version numbers of the library 130You can find out the major and minor ABI version numbers of the library
… …
300=item C<EVBACKEND_SELECT> (value 1, portable select backend) 306=item C<EVBACKEND_SELECT> (value 1, portable select backend)
301 307
302This is your standard select(2) backend. Not I<completely> standard, as 308This is your standard select(2) backend. Not I<completely> standard, as
303libev tries to roll its own fd_set with no limits on the number of fds, 309libev tries to roll its own fd_set with no limits on the number of fds,
304but if that fails, expect a fairly low limit on the number of fds when 310but if that fails, expect a fairly low limit on the number of fds when
305using this backend. It doesn't scale too well (O(highest_fd)), but its usually 311using this backend. It doesn't scale too well (O(highest_fd)), but its
306the fastest backend for a low number of fds. 312usually the fastest backend for a low number of (low-numbered :) fds.
313
314To get good performance out of this backend you need a high amount of
315parallelity (most of the file descriptors should be busy). If you are
316writing a server, you should C<accept ()> in a loop to accept as many
317connections as possible during one iteration. You might also want to have
318a look at C<ev_set_io_collect_interval ()> to increase the amount of
319readyness notifications you get per iteration.
307 320
308=item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows) 321=item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
309 322
310And this is your standard poll(2) backend. It's more complicated than 323And this is your standard poll(2) backend. It's more complicated
311select, but handles sparse fds better and has no artificial limit on the 324than select, but handles sparse fds better and has no artificial
312number of fds you can use (except it will slow down considerably with a 325limit on the number of fds you can use (except it will slow down
313lot of inactive fds). It scales similarly to select, i.e. O(total_fds). 326considerably with a lot of inactive fds). It scales similarly to select,
327i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
328performance tips.
314 329
315=item C<EVBACKEND_EPOLL> (value 4, Linux) 330=item C<EVBACKEND_EPOLL> (value 4, Linux)
316 331
317For few fds, this backend is a bit little slower than poll and select, 332For few fds, this backend is a bit little slower than poll and select,
318but it scales phenomenally better. While poll and select usually scale like 333but it scales phenomenally better. While poll and select usually scale
319O(total_fds) where n is the total number of fds (or the highest fd), epoll scales 334like O(total_fds) where n is the total number of fds (or the highest fd),
320either O(1) or O(active_fds). 335epoll scales either O(1) or O(active_fds). The epoll design has a number
336of shortcomings, such as silently dropping events in some hard-to-detect
337cases and rewiring a syscall per fd change, no fork support and bad
338support for dup.
321 339
322While stopping and starting an I/O watcher in the same iteration will 340While stopping, setting and starting an I/O watcher in the same iteration
323result in some caching, there is still a syscall per such incident 341will result in some caching, there is still a syscall per such incident
324(because the fd could point to a different file description now), so its 342(because the fd could point to a different file description now), so its
325best to avoid that. Also, dup()ed file descriptors might not work very 343best to avoid that. Also, C<dup ()>'ed file descriptors might not work
326well if you register events for both fds. 344very well if you register events for both fds.
327 345
328Please note that epoll sometimes generates spurious notifications, so you 346Please note that epoll sometimes generates spurious notifications, so you
329need to use non-blocking I/O or other means to avoid blocking when no data 347need to use non-blocking I/O or other means to avoid blocking when no data
330(or space) is available. 348(or space) is available.
331 349
350Best performance from this backend is achieved by not unregistering all
351watchers for a file descriptor until it has been closed, if possible, i.e.
352keep at least one watcher active per fd at all times.
353
354While nominally embeddeble in other event loops, this feature is broken in
355all kernel versions tested so far.
356
332=item C<EVBACKEND_KQUEUE> (value 8, most BSD clones) 357=item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
333 358
334Kqueue deserves special mention, as at the time of this writing, it 359Kqueue deserves special mention, as at the time of this writing, it
335was broken on all BSDs except NetBSD (usually it doesn't work with 360was broken on all BSDs except NetBSD (usually it doesn't work reliably
336anything but sockets and pipes, except on Darwin, where of course its 361with anything but sockets and pipes, except on Darwin, where of course
337completely useless). For this reason its not being "autodetected" 362it's completely useless). For this reason it's not being "autodetected"
338unless you explicitly specify it explicitly in the flags (i.e. using 363unless you explicitly specify it explicitly in the flags (i.e. using
339C<EVBACKEND_KQUEUE>). 364C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
365system like NetBSD.
366
367You still can embed kqueue into a normal poll or select backend and use it
368only for sockets (after having made sure that sockets work with kqueue on
369the target platform). See C<ev_embed> watchers for more info.
340 370
341It scales in the same way as the epoll backend, but the interface to the 371It scales in the same way as the epoll backend, but the interface to the
342kernel is more efficient (which says nothing about its actual speed, of 372kernel is more efficient (which says nothing about its actual speed, of
343course). While starting and stopping an I/O watcher does not cause an 373course). While stopping, setting and starting an I/O watcher does never
344extra syscall as with epoll, it still adds up to four event changes per 374cause an extra syscall as with C<EVBACKEND_EPOLL>, it still adds up to
345incident, so its best to avoid that. 375two event changes per incident, support for C<fork ()> is very bad and it
376drops fds silently in similarly hard-to-detect cases.
377
378This backend usually performs well under most conditions.
379
380While nominally embeddable in other event loops, this doesn't work
381everywhere, so you might need to test for this. And since it is broken
382almost everywhere, you should only use it when you have a lot of sockets
383(for which it usually works), by embedding it into another event loop
384(e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>) and using it only for
385sockets.
346 386
347=item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8) 387=item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
348 388
349This is not implemented yet (and might never be). 389This is not implemented yet (and might never be, unless you send me an
390implementation). According to reports, C</dev/poll> only supports sockets
391and is not embeddable, which would limit the usefulness of this backend
392immensely.
350 393
351=item C<EVBACKEND_PORT> (value 32, Solaris 10) 394=item C<EVBACKEND_PORT> (value 32, Solaris 10)
352 395
353This uses the Solaris 10 port mechanism. As with everything on Solaris, 396This uses the Solaris 10 event port mechanism. As with everything on Solaris,
354it's really slow, but it still scales very well (O(active_fds)). 397it's really slow, but it still scales very well (O(active_fds)).
355 398
356Please note that solaris ports can result in a lot of spurious 399Please note that solaris event ports can deliver a lot of spurious
357notifications, so you need to use non-blocking I/O or other means to avoid 400notifications, so you need to use non-blocking I/O or other means to avoid
358blocking when no data (or space) is available. 401blocking when no data (or space) is available.
402
403While this backend scales well, it requires one system call per active
404file descriptor per loop iteration. For small and medium numbers of file
405descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
406might perform better.
359 407
360=item C<EVBACKEND_ALL> 408=item C<EVBACKEND_ALL>
361 409
362Try all backends (even potentially broken ones that wouldn't be tried 410Try all backends (even potentially broken ones that wouldn't be tried
363with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as 411with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
364C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>. 412C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.
413
414It is definitely not recommended to use this flag.
365 415
366=back 416=back
367 417
368If one or more of these are ored into the flags value, then only these 418If one or more of these are ored into the flags value, then only these
369backends will be tried (in the reverse order as given here). If none are 419backends will be tried (in the reverse order as given here). If none are
… …
404Destroys the default loop again (frees all memory and kernel state 454Destroys the default loop again (frees all memory and kernel state
405etc.). None of the active event watchers will be stopped in the normal 455etc.). None of the active event watchers will be stopped in the normal
406sense, so e.g. C<ev_is_active> might still return true. It is your 456sense, so e.g. C<ev_is_active> might still return true. It is your
407responsibility to either stop all watchers cleanly yoursef I<before> 457responsibility to either stop all watchers cleanly yoursef I<before>
408calling this function, or cope with the fact afterwards (which is usually 458calling this function, or cope with the fact afterwards (which is usually
409the easiest thing, youc na just ignore the watchers and/or C<free ()> them 459the easiest thing, you can just ignore the watchers and/or C<free ()> them
410for example). 460for example).
461
462Note that certain global state, such as signal state, will not be freed by
463this function, and related watchers (such as signal and child watchers)
464would need to be stopped manually.
465
466In general it is not advisable to call this function except in the
467rare occasion where you really need to free e.g. the signal handling
468pipe fds. If you need dynamically allocated loops it is better to use
469C<ev_loop_new> and C<ev_loop_destroy>).
411 470
412=item ev_loop_destroy (loop) 471=item ev_loop_destroy (loop)
413 472
414Like C<ev_default_destroy>, but destroys an event loop created by an 473Like C<ev_default_destroy>, but destroys an event loop created by an
415earlier call to C<ev_loop_new>. 474earlier call to C<ev_loop_new>.
… …
460 519
461Returns the current "event loop time", which is the time the event loop 520Returns the current "event loop time", which is the time the event loop
462received events and started processing them. This timestamp does not 521received events and started processing them. This timestamp does not
463change as long as callbacks are being processed, and this is also the base 522change as long as callbacks are being processed, and this is also the base
464time used for relative timers. You can treat it as the timestamp of the 523time used for relative timers. You can treat it as the timestamp of the
465event occuring (or more correctly, libev finding out about it). 524event occurring (or more correctly, libev finding out about it).
466 525
467=item ev_loop (loop, int flags) 526=item ev_loop (loop, int flags)
468 527
469Finally, this is it, the event handler. This function usually is called 528Finally, this is it, the event handler. This function usually is called
470after you initialised all your watchers and you want to start handling 529after you initialised all your watchers and you want to start handling
… …
492usually a better approach for this kind of thing. 551usually a better approach for this kind of thing.
493 552
494Here are the gory details of what C<ev_loop> does: 553Here are the gory details of what C<ev_loop> does:
495 554
496 - Before the first iteration, call any pending watchers. 555 - Before the first iteration, call any pending watchers.
497 * If there are no active watchers (reference count is zero), return. 556 * If EVFLAG_FORKCHECK was used, check for a fork.
498 - Queue all prepare watchers and then call all outstanding watchers. 557 - If a fork was detected, queue and call all fork watchers.
558 - Queue and call all prepare watchers.
499 - If we have been forked, recreate the kernel state. 559 - If we have been forked, recreate the kernel state.
500 - Update the kernel state with all outstanding changes. 560 - Update the kernel state with all outstanding changes.
501 - Update the "event loop time". 561 - Update the "event loop time".
502 - Calculate for how long to block. 562 - Calculate for how long to sleep or block, if at all
563 (active idle watchers, EVLOOP_NONBLOCK or not having
564 any active watchers at all will result in not sleeping).
565 - Sleep if the I/O and timer collect interval say so.
503 - Block the process, waiting for any events. 566 - Block the process, waiting for any events.
504 - Queue all outstanding I/O (fd) events. 567 - Queue all outstanding I/O (fd) events.
505 - Update the "event loop time" and do time jump handling. 568 - Update the "event loop time" and do time jump handling.
506 - Queue all outstanding timers. 569 - Queue all outstanding timers.
507 - Queue all outstanding periodics. 570 - Queue all outstanding periodics.
508 - If no events are pending now, queue all idle watchers. 571 - If no events are pending now, queue all idle watchers.
509 - Queue all check watchers. 572 - Queue all check watchers.
510 - Call all queued watchers in reverse order (i.e. check watchers first). 573 - Call all queued watchers in reverse order (i.e. check watchers first).
511 Signals and child watchers are implemented as I/O watchers, and will 574 Signals and child watchers are implemented as I/O watchers, and will
512 be handled here by queueing them when their watcher gets executed. 575 be handled here by queueing them when their watcher gets executed.
513 - If ev_unloop has been called or EVLOOP_ONESHOT or EVLOOP_NONBLOCK 576 - If ev_unloop has been called, or EVLOOP_ONESHOT or EVLOOP_NONBLOCK
514 were used, return, otherwise continue with step *. 577 were used, or there are no active watchers, return, otherwise
578 continue with step *.
515 579
516Example: Queue some jobs and then loop until no events are outsanding 580Example: Queue some jobs and then loop until no events are outstanding
517anymore. 581anymore.
518 582
519 ... queue jobs here, make sure they register event watchers as long 583 ... queue jobs here, make sure they register event watchers as long
520 ... as they still have work to do (even an idle watcher will do..) 584 ... as they still have work to do (even an idle watcher will do..)
521 ev_loop (my_loop, 0); 585 ev_loop (my_loop, 0);
… …
554Example: For some weird reason, unregister the above signal handler again. 618Example: For some weird reason, unregister the above signal handler again.
555 619
556 ev_ref (loop); 620 ev_ref (loop);
557 ev_signal_stop (loop, &exitsig); 621 ev_signal_stop (loop, &exitsig);
558 622
623=item ev_set_io_collect_interval (loop, ev_tstamp interval)
624
625=item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
626
627These advanced functions influence the time that libev will spend waiting
628for events. Both are by default C<0>, meaning that libev will try to
629invoke timer/periodic callbacks and I/O callbacks with minimum latency.
630
631Setting these to a higher value (the C<interval> I<must> be >= C<0>)
632allows libev to delay invocation of I/O and timer/periodic callbacks to
633increase efficiency of loop iterations.
634
635The background is that sometimes your program runs just fast enough to
636handle one (or very few) event(s) per loop iteration. While this makes
637the program responsive, it also wastes a lot of CPU time to poll for new
638events, especially with backends like C<select ()> which have a high
639overhead for the actual polling but can deliver many events at once.
640
641By setting a higher I<io collect interval> you allow libev to spend more
642time collecting I/O events, so you can handle more events per iteration,
643at the cost of increasing latency. Timeouts (both C<ev_periodic> and
644C<ev_timer>) will be not affected. Setting this to a non-null value will
645introduce an additional C<ev_sleep ()> call into most loop iterations.
646
647Likewise, by setting a higher I<timeout collect interval> you allow libev
648to spend more time collecting timeouts, at the expense of increased
649latency (the watcher callback will be called later). C<ev_io> watchers
650will not be affected. Setting this to a non-null value will not introduce
651any overhead in libev.
652
653Many (busy) programs can usually benefit by setting the io collect
654interval to a value near C<0.1> or so, which is often enough for
655interactive servers (of course not for games), likewise for timeouts. It
656usually doesn't make much sense to set it to a lower value than C<0.01>,
657as this approsaches the timing granularity of most systems.
658
559=back 659=back
560 660
561 661
562=head1 ANATOMY OF A WATCHER 662=head1 ANATOMY OF A WATCHER
563 663
… …
888In general you can register as many read and/or write event watchers per 988In general you can register as many read and/or write event watchers per
889fd as you want (as long as you don't confuse yourself). Setting all file 989fd as you want (as long as you don't confuse yourself). Setting all file
890descriptors to non-blocking mode is also usually a good idea (but not 990descriptors to non-blocking mode is also usually a good idea (but not
891required if you know what you are doing). 991required if you know what you are doing).
892 992
893You have to be careful with dup'ed file descriptors, though. Some backends
894(the linux epoll backend is a notable example) cannot handle dup'ed file
895descriptors correctly if you register interest in two or more fds pointing
896to the same underlying file/socket/etc. description (that is, they share
897the same underlying "file open").
898
899If you must do this, then force the use of a known-to-be-good backend 993If you must do this, then force the use of a known-to-be-good backend
900(at the time of this writing, this includes only C<EVBACKEND_SELECT> and 994(at the time of this writing, this includes only C<EVBACKEND_SELECT> and
901C<EVBACKEND_POLL>). 995C<EVBACKEND_POLL>).
902 996
903Another thing you have to watch out for is that it is quite easy to 997Another thing you have to watch out for is that it is quite easy to
… …
915such as poll (fortunately in our Xlib example, Xlib already does this on 1009such as poll (fortunately in our Xlib example, Xlib already does this on
916its own, so its quite safe to use). 1010its own, so its quite safe to use).
917 1011
918=head3 The special problem of disappearing file descriptors 1012=head3 The special problem of disappearing file descriptors
919 1013
920Some backends (e.g kqueue, epoll) need to be told about closing a file 1014Some backends (e.g. kqueue, epoll) need to be told about closing a file
921descriptor (either by calling C<close> explicitly or by any other means, 1015descriptor (either by calling C<close> explicitly or by any other means,
922such as C<dup>). The reason is that you register interest in some file 1016such as C<dup>). The reason is that you register interest in some file
923descriptor, but when it goes away, the operating system will silently drop 1017descriptor, but when it goes away, the operating system will silently drop
924this interest. If another file descriptor with the same number then is 1018this interest. If another file descriptor with the same number then is
925registered with libev, there is no efficient way to see that this is, in 1019registered with libev, there is no efficient way to see that this is, in
… …
934 1028
935This is how one would do it normally anyway, the important point is that 1029This is how one would do it normally anyway, the important point is that
936the libev application should not optimise around libev but should leave 1030the libev application should not optimise around libev but should leave
937optimisations to libev. 1031optimisations to libev.
938 1032
1033=head3 The special problem of dup'ed file descriptors
1034
1035Some backends (e.g. epoll), cannot register events for file descriptors,
1036but only events for the underlying file descriptions. That means when you
1037have C<dup ()>'ed file descriptors or weirder constellations, and register
1038events for them, only one file descriptor might actually receive events.
1039
1040There is no workaround possible except not registering events
1041for potentially C<dup ()>'ed file descriptors, or to resort to
1042C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
1043
1044=head3 The special problem of fork
1045
1046Some backends (epoll, kqueue) do not support C<fork ()> at all or exhibit
1047useless behaviour. Libev fully supports fork, but needs to be told about
1048it in the child.
1049
1050To support fork in your programs, you either have to call
1051C<ev_default_fork ()> or C<ev_loop_fork ()> after a fork in the child,
1052enable C<EVFLAG_FORKCHECK>, or resort to C<EVBACKEND_SELECT> or
1053C<EVBACKEND_POLL>.
1054
939 1055
940=head3 Watcher-Specific Functions 1056=head3 Watcher-Specific Functions
941 1057
942=over 4 1058=over 4
943 1059
… …
956=item int events [read-only] 1072=item int events [read-only]
957 1073
958The events being watched. 1074The events being watched.
959 1075
960=back 1076=back
1077
1078=head3 Examples
961 1079
962Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well 1080Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
963readable, but only once. Since it is likely line-buffered, you could 1081readable, but only once. Since it is likely line-buffered, you could
964attempt to read a whole line in the callback. 1082attempt to read a whole line in the callback.
965 1083
… …
1063or C<ev_timer_again> is called and determines the next timeout (if any), 1181or C<ev_timer_again> is called and determines the next timeout (if any),
1064which is also when any modifications are taken into account. 1182which is also when any modifications are taken into account.
1065 1183
1066=back 1184=back
1067 1185
1186=head3 Examples
1187
1068Example: Create a timer that fires after 60 seconds. 1188Example: Create a timer that fires after 60 seconds.
1069 1189
1070 static void 1190 static void
1071 one_minute_cb (struct ev_loop *loop, struct ev_timer *w, int revents) 1191 one_minute_cb (struct ev_loop *loop, struct ev_timer *w, int revents)
1072 { 1192 {
… …
1229When active, contains the absolute time that the watcher is supposed to 1349When active, contains the absolute time that the watcher is supposed to
1230trigger next. 1350trigger next.
1231 1351
1232=back 1352=back
1233 1353
1354=head3 Examples
1355
1234Example: Call a callback every hour, or, more precisely, whenever the 1356Example: Call a callback every hour, or, more precisely, whenever the
1235system clock is divisible by 3600. The callback invocation times have 1357system clock is divisible by 3600. The callback invocation times have
1236potentially a lot of jittering, but good long-term stability. 1358potentially a lot of jittering, but good long-term stability.
1237 1359
1238 static void 1360 static void
… …
1329 1451
1330The process exit/trace status caused by C<rpid> (see your systems 1452The process exit/trace status caused by C<rpid> (see your systems
1331C<waitpid> and C<sys/wait.h> documentation for details). 1453C<waitpid> and C<sys/wait.h> documentation for details).
1332 1454
1333=back 1455=back
1456
1457=head3 Examples
1334 1458
1335Example: Try to exit cleanly on SIGINT and SIGTERM. 1459Example: Try to exit cleanly on SIGINT and SIGTERM.
1336 1460
1337 static void 1461 static void
1338 sigint_cb (struct ev_loop *loop, struct ev_signal *w, int revents) 1462 sigint_cb (struct ev_loop *loop, struct ev_signal *w, int revents)
… …
1379semantics of C<ev_stat> watchers, which means that libev sometimes needs 1503semantics of C<ev_stat> watchers, which means that libev sometimes needs
1380to fall back to regular polling again even with inotify, but changes are 1504to fall back to regular polling again even with inotify, but changes are
1381usually detected immediately, and if the file exists there will be no 1505usually detected immediately, and if the file exists there will be no
1382polling. 1506polling.
1383 1507
1508=head3 Inotify
1509
1510When C<inotify (7)> support has been compiled into libev (generally only
1511available on Linux) and present at runtime, it will be used to speed up
1512change detection where possible. The inotify descriptor will be created lazily
1513when the first C<ev_stat> watcher is being started.
1514
1515Inotify presense does not change the semantics of C<ev_stat> watchers
1516except that changes might be detected earlier, and in some cases, to avoid
1517making regular C<stat> calls. Even in the presense of inotify support
1518there are many cases where libev has to resort to regular C<stat> polling.
1519
1520(There is no support for kqueue, as apparently it cannot be used to
1521implement this functionality, due to the requirement of having a file
1522descriptor open on the object at all times).
1523
1524=head3 The special problem of stat time resolution
1525
1526The C<stat ()> syscall only supports full-second resolution portably, and
1527even on systems where the resolution is higher, many filesystems still
1528only support whole seconds.
1529
1530That means that, if the time is the only thing that changes, you might
1531miss updates: on the first update, C<ev_stat> detects a change and calls
1532your callback, which does something. When there is another update within
1533the same second, C<ev_stat> will be unable to detect it.
1534
1535The solution to this is to delay acting on a change for a second (or till
1536the next second boundary), using a roughly one-second delay C<ev_timer>
1537(C<ev_timer_set (w, 0., 1.01); ev_timer_again (loop, w)>). The C<.01>
1538is added to work around small timing inconsistencies of some operating
1539systems.
1540
1384=head3 Watcher-Specific Functions and Data Members 1541=head3 Watcher-Specific Functions and Data Members
1385 1542
1386=over 4 1543=over 4
1387 1544
1388=item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval) 1545=item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
… …
1425=item const char *path [read-only] 1582=item const char *path [read-only]
1426 1583
1427The filesystem path that is being watched. 1584The filesystem path that is being watched.
1428 1585
1429=back 1586=back
1587
1588=head3 Examples
1430 1589
1431Example: Watch C</etc/passwd> for attribute changes. 1590Example: Watch C</etc/passwd> for attribute changes.
1432 1591
1433 static void 1592 static void
1434 passwd_cb (struct ev_loop *loop, ev_stat *w, int revents) 1593 passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
… …
1447 } 1606 }
1448 1607
1449 ... 1608 ...
1450 ev_stat passwd; 1609 ev_stat passwd;
1451 1610
1452 ev_stat_init (&passwd, passwd_cb, "/etc/passwd"); 1611 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
1453 ev_stat_start (loop, &passwd); 1612 ev_stat_start (loop, &passwd);
1613
1614Example: Like above, but additionally use a one-second delay so we do not
1615miss updates (however, frequent updates will delay processing, too, so
1616one might do the work both on C<ev_stat> callback invocation I<and> on
1617C<ev_timer> callback invocation).
1618
1619 static ev_stat passwd;
1620 static ev_timer timer;
1621
1622 static void
1623 timer_cb (EV_P_ ev_timer *w, int revents)
1624 {
1625 ev_timer_stop (EV_A_ w);
1626
1627 /* now it's one second after the most recent passwd change */
1628 }
1629
1630 static void
1631 stat_cb (EV_P_ ev_stat *w, int revents)
1632 {
1633 /* reset the one-second timer */
1634 ev_timer_again (EV_A_ &timer);
1635 }
1636
1637 ...
1638 ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
1639 ev_stat_start (loop, &passwd);
1640 ev_timer_init (&timer, timer_cb, 0., 1.01);
1454 1641
1455 1642
1456=head2 C<ev_idle> - when you've got nothing better to do... 1643=head2 C<ev_idle> - when you've got nothing better to do...
1457 1644
1458Idle watchers trigger events when no other events of the same or higher 1645Idle watchers trigger events when no other events of the same or higher
… …
1483Initialises and configures the idle watcher - it has no parameters of any 1670Initialises and configures the idle watcher - it has no parameters of any
1484kind. There is a C<ev_idle_set> macro, but using it is utterly pointless, 1671kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
1485believe me. 1672believe me.
1486 1673
1487=back 1674=back
1675
1676=head3 Examples
1488 1677
1489Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the 1678Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
1490callback, free it. Also, use no error checking, as usual. 1679callback, free it. Also, use no error checking, as usual.
1491 1680
1492 static void 1681 static void
… …
1544 1733
1545It is recommended to give C<ev_check> watchers highest (C<EV_MAXPRI>) 1734It is recommended to give C<ev_check> watchers highest (C<EV_MAXPRI>)
1546priority, to ensure that they are being run before any other watchers 1735priority, to ensure that they are being run before any other watchers
1547after the poll. Also, C<ev_check> watchers (and C<ev_prepare> watchers, 1736after the poll. Also, C<ev_check> watchers (and C<ev_prepare> watchers,
1548too) should not activate ("feed") events into libev. While libev fully 1737too) should not activate ("feed") events into libev. While libev fully
1549supports this, they will be called before other C<ev_check> watchers did 1738supports this, they will be called before other C<ev_check> watchers
1550their job. As C<ev_check> watchers are often used to embed other event 1739did their job. As C<ev_check> watchers are often used to embed other
1551loops those other event loops might be in an unusable state until their 1740(non-libev) event loops those other event loops might be in an unusable
1552C<ev_check> watcher ran (always remind yourself to coexist peacefully with 1741state until their C<ev_check> watcher ran (always remind yourself to
1553others). 1742coexist peacefully with others).
1554 1743
1555=head3 Watcher-Specific Functions and Data Members 1744=head3 Watcher-Specific Functions and Data Members
1556 1745
1557=over 4 1746=over 4
1558 1747
… …
1563Initialises and configures the prepare or check watcher - they have no 1752Initialises and configures the prepare or check watcher - they have no
1564parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set> 1753parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
1565macros, but using them is utterly, utterly and completely pointless. 1754macros, but using them is utterly, utterly and completely pointless.
1566 1755
1567=back 1756=back
1757
1758=head3 Examples
1568 1759
1569There are a number of principal ways to embed other event loops or modules 1760There are a number of principal ways to embed other event loops or modules
1570into libev. Here are some ideas on how to include libadns into libev 1761into libev. Here are some ideas on how to include libadns into libev
1571(there is a Perl module named C<EV::ADNS> that does this, which you could 1762(there is a Perl module named C<EV::ADNS> that does this, which you could
1572use for an actually working example. Another Perl module named C<EV::Glib> 1763use for an actually working example. Another Perl module named C<EV::Glib>
… …
1741portable one. 1932portable one.
1742 1933
1743So when you want to use this feature you will always have to be prepared 1934So when you want to use this feature you will always have to be prepared
1744that you cannot get an embeddable loop. The recommended way to get around 1935that you cannot get an embeddable loop. The recommended way to get around
1745this is to have a separate variables for your embeddable loop, try to 1936this is to have a separate variables for your embeddable loop, try to
1746create it, and if that fails, use the normal loop for everything: 1937create it, and if that fails, use the normal loop for everything.
1938
1939=head3 Watcher-Specific Functions and Data Members
1940
1941=over 4
1942
1943=item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
1944
1945=item ev_embed_set (ev_embed *, callback, struct ev_loop *embedded_loop)
1946
1947Configures the watcher to embed the given loop, which must be
1948embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
1949invoked automatically, otherwise it is the responsibility of the callback
1950to invoke it (it will continue to be called until the sweep has been done,
1951if you do not want thta, you need to temporarily stop the embed watcher).
1952
1953=item ev_embed_sweep (loop, ev_embed *)
1954
1955Make a single, non-blocking sweep over the embedded loop. This works
1956similarly to C<ev_loop (embedded_loop, EVLOOP_NONBLOCK)>, but in the most
1957apropriate way for embedded loops.
1958
1959=item struct ev_loop *other [read-only]
1960
1961The embedded event loop.
1962
1963=back
1964
1965=head3 Examples
1966
1967Example: Try to get an embeddable event loop and embed it into the default
1968event loop. If that is not possible, use the default loop. The default
1969loop is stored in C<loop_hi>, while the mebeddable loop is stored in
1970C<loop_lo> (which is C<loop_hi> in the acse no embeddable loop can be
1971used).
1747 1972
1748 struct ev_loop *loop_hi = ev_default_init (0); 1973 struct ev_loop *loop_hi = ev_default_init (0);
1749 struct ev_loop *loop_lo = 0; 1974 struct ev_loop *loop_lo = 0;
1750 struct ev_embed embed; 1975 struct ev_embed embed;
1751 1976
… …
1762 ev_embed_start (loop_hi, &embed); 1987 ev_embed_start (loop_hi, &embed);
1763 } 1988 }
1764 else 1989 else
1765 loop_lo = loop_hi; 1990 loop_lo = loop_hi;
1766 1991
1767=head3 Watcher-Specific Functions and Data Members 1992Example: Check if kqueue is available but not recommended and create
1993a kqueue backend for use with sockets (which usually work with any
1994kqueue implementation). Store the kqueue/socket-only event loop in
1995C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
1768 1996
1769=over 4 1997 struct ev_loop *loop = ev_default_init (0);
1998 struct ev_loop *loop_socket = 0;
1999 struct ev_embed embed;
2000
2001 if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
2002 if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
2003 {
2004 ev_embed_init (&embed, 0, loop_socket);
2005 ev_embed_start (loop, &embed);
2006 }
1770 2007
1771=item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop) 2008 if (!loop_socket)
2009 loop_socket = loop;
1772 2010
1773=item ev_embed_set (ev_embed *, callback, struct ev_loop *embedded_loop) 2011 // now use loop_socket for all sockets, and loop for everything else
1774
1775Configures the watcher to embed the given loop, which must be
1776embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
1777invoked automatically, otherwise it is the responsibility of the callback
1778to invoke it (it will continue to be called until the sweep has been done,
1779if you do not want thta, you need to temporarily stop the embed watcher).
1780
1781=item ev_embed_sweep (loop, ev_embed *)
1782
1783Make a single, non-blocking sweep over the embedded loop. This works
1784similarly to C<ev_loop (embedded_loop, EVLOOP_NONBLOCK)>, but in the most
1785apropriate way for embedded loops.
1786
1787=item struct ev_loop *loop [read-only]
1788
1789The embedded event loop.
1790
1791=back
1792 2012
1793 2013
1794=head2 C<ev_fork> - the audacity to resume the event loop after a fork 2014=head2 C<ev_fork> - the audacity to resume the event loop after a fork
1795 2015
1796Fork watchers are called when a C<fork ()> was detected (usually because 2016Fork watchers are called when a C<fork ()> was detected (usually because
… …
2131Libev can (and often is) directly embedded into host 2351Libev can (and often is) directly embedded into host
2132applications. Examples of applications that embed it include the Deliantra 2352applications. Examples of applications that embed it include the Deliantra
2133Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe) 2353Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
2134and rxvt-unicode. 2354and rxvt-unicode.
2135 2355
2136The goal is to enable you to just copy the neecssary files into your 2356The goal is to enable you to just copy the necessary files into your
2137source directory without having to change even a single line in them, so 2357source directory without having to change even a single line in them, so
2138you can easily upgrade by simply copying (or having a checked-out copy of 2358you can easily upgrade by simply copying (or having a checked-out copy of
2139libev somewhere in your source tree). 2359libev somewhere in your source tree).
2140 2360
2141=head2 FILESETS 2361=head2 FILESETS
… …
2231 2451
2232If defined to be C<1>, libev will try to detect the availability of the 2452If defined to be C<1>, libev will try to detect the availability of the
2233monotonic clock option at both compiletime and runtime. Otherwise no use 2453monotonic clock option at both compiletime and runtime. Otherwise no use
2234of the monotonic clock option will be attempted. If you enable this, you 2454of the monotonic clock option will be attempted. If you enable this, you
2235usually have to link against librt or something similar. Enabling it when 2455usually have to link against librt or something similar. Enabling it when
2236the functionality isn't available is safe, though, althoguh you have 2456the functionality isn't available is safe, though, although you have
2237to make sure you link against any libraries where the C<clock_gettime> 2457to make sure you link against any libraries where the C<clock_gettime>
2238function is hiding in (often F<-lrt>). 2458function is hiding in (often F<-lrt>).
2239 2459
2240=item EV_USE_REALTIME 2460=item EV_USE_REALTIME
2241 2461
2242If defined to be C<1>, libev will try to detect the availability of the 2462If defined to be C<1>, libev will try to detect the availability of the
2243realtime clock option at compiletime (and assume its availability at 2463realtime clock option at compiletime (and assume its availability at
2244runtime if successful). Otherwise no use of the realtime clock option will 2464runtime if successful). Otherwise no use of the realtime clock option will
2245be attempted. This effectively replaces C<gettimeofday> by C<clock_get 2465be attempted. This effectively replaces C<gettimeofday> by C<clock_get
2246(CLOCK_REALTIME, ...)> and will not normally affect correctness. See tzhe note about libraries 2466(CLOCK_REALTIME, ...)> and will not normally affect correctness. See the
2247in the description of C<EV_USE_MONOTONIC>, though. 2467note about libraries in the description of C<EV_USE_MONOTONIC>, though.
2468
2469=item EV_USE_NANOSLEEP
2470
2471If defined to be C<1>, libev will assume that C<nanosleep ()> is available
2472and will use it for delays. Otherwise it will use C<select ()>.
2248 2473
2249=item EV_USE_SELECT 2474=item EV_USE_SELECT
2250 2475
2251If undefined or defined to be C<1>, libev will compile in support for the 2476If undefined or defined to be C<1>, libev will compile in support for the
2252C<select>(2) backend. No attempt at autodetection will be done: if no 2477C<select>(2) backend. No attempt at autodetection will be done: if no
… …
2270wants osf handles on win32 (this is the case when the select to 2495wants osf handles on win32 (this is the case when the select to
2271be used is the winsock select). This means that it will call 2496be used is the winsock select). This means that it will call
2272C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise, 2497C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
2273it is assumed that all these functions actually work on fds, even 2498it is assumed that all these functions actually work on fds, even
2274on win32. Should not be defined on non-win32 platforms. 2499on win32. Should not be defined on non-win32 platforms.
2500
2501=item EV_FD_TO_WIN32_HANDLE
2502
2503If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
2504file descriptors to socket handles. When not defining this symbol (the
2505default), then libev will call C<_get_osfhandle>, which is usually
2506correct. In some cases, programs use their own file descriptor management,
2507in which case they can provide this function to map fds to socket handles.
2275 2508
2276=item EV_USE_POLL 2509=item EV_USE_POLL
2277 2510
2278If defined to be C<1>, libev will compile in support for the C<poll>(2) 2511If defined to be C<1>, libev will compile in support for the C<poll>(2)
2279backend. Otherwise it will be enabled on non-win32 platforms. It 2512backend. Otherwise it will be enabled on non-win32 platforms. It
… …
2316be detected at runtime. 2549be detected at runtime.
2317 2550
2318=item EV_H 2551=item EV_H
2319 2552
2320The name of the F<ev.h> header file used to include it. The default if 2553The name of the F<ev.h> header file used to include it. The default if
2321undefined is C<< <ev.h> >> in F<event.h> and C<"ev.h"> in F<ev.c>. This 2554undefined is C<"ev.h"> in F<event.h> and F<ev.c>. This can be used to
2322can be used to virtually rename the F<ev.h> header file in case of conflicts. 2555virtually rename the F<ev.h> header file in case of conflicts.
2323 2556
2324=item EV_CONFIG_H 2557=item EV_CONFIG_H
2325 2558
2326If C<EV_STANDALONE> isn't C<1>, this variable can be used to override 2559If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
2327F<ev.c>'s idea of where to find the F<config.h> file, similarly to 2560F<ev.c>'s idea of where to find the F<config.h> file, similarly to
2328C<EV_H>, above. 2561C<EV_H>, above.
2329 2562
2330=item EV_EVENT_H 2563=item EV_EVENT_H
2331 2564
2332Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea 2565Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
2333of how the F<event.h> header can be found. 2566of how the F<event.h> header can be found, the dfeault is C<"event.h">.
2334 2567
2335=item EV_PROTOTYPES 2568=item EV_PROTOTYPES
2336 2569
2337If defined to be C<0>, then F<ev.h> will not define any function 2570If defined to be C<0>, then F<ev.h> will not define any function
2338prototypes, but still define all the structs and other symbols. This is 2571prototypes, but still define all the structs and other symbols. This is
… …
2404than enough. If you need to manage thousands of children you might want to 2637than enough. If you need to manage thousands of children you might want to
2405increase this value (I<must> be a power of two). 2638increase this value (I<must> be a power of two).
2406 2639
2407=item EV_INOTIFY_HASHSIZE 2640=item EV_INOTIFY_HASHSIZE
2408 2641
2409C<ev_staz> watchers use a small hash table to distribute workload by 2642C<ev_stat> watchers use a small hash table to distribute workload by
2410inotify watch id. The default size is C<16> (or C<1> with C<EV_MINIMAL>), 2643inotify watch id. The default size is C<16> (or C<1> with C<EV_MINIMAL>),
2411usually more than enough. If you need to manage thousands of C<ev_stat> 2644usually more than enough. If you need to manage thousands of C<ev_stat>
2412watchers you might want to increase this value (I<must> be a power of 2645watchers you might want to increase this value (I<must> be a power of
2413two). 2646two).
2414 2647
… …
2431 2664
2432=item ev_set_cb (ev, cb) 2665=item ev_set_cb (ev, cb)
2433 2666
2434Can be used to change the callback member declaration in each watcher, 2667Can be used to change the callback member declaration in each watcher,
2435and the way callbacks are invoked and set. Must expand to a struct member 2668and the way callbacks are invoked and set. Must expand to a struct member
2436definition and a statement, respectively. See the F<ev.v> header file for 2669definition and a statement, respectively. See the F<ev.h> header file for
2437their default definitions. One possible use for overriding these is to 2670their default definitions. One possible use for overriding these is to
2438avoid the C<struct ev_loop *> as first argument in all cases, or to use 2671avoid the C<struct ev_loop *> as first argument in all cases, or to use
2439method calls instead of plain function calls in C++. 2672method calls instead of plain function calls in C++.
2673
2674=head2 EXPORTED API SYMBOLS
2675
2676If you need to re-export the API (e.g. via a dll) and you need a list of
2677exported symbols, you can use the provided F<Symbol.*> files which list
2678all public symbols, one per line:
2679
2680 Symbols.ev for libev proper
2681 Symbols.event for the libevent emulation
2682
2683This can also be used to rename all public symbols to avoid clashes with
2684multiple versions of libev linked together (which is obviously bad in
2685itself, but sometimes it is inconvinient to avoid this).
2686
2687A sed command like this will create wrapper C<#define>'s that you need to
2688include before including F<ev.h>:
2689
2690 <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
2691
2692This would create a file F<wrap.h> which essentially looks like this:
2693
2694 #define ev_backend myprefix_ev_backend
2695 #define ev_check_start myprefix_ev_check_start
2696 #define ev_check_stop myprefix_ev_check_stop
2697 ...
2440 2698
2441=head2 EXAMPLES 2699=head2 EXAMPLES
2442 2700
2443For a real-world example of a program the includes libev 2701For a real-world example of a program the includes libev
2444verbatim, you can have a look at the EV perl module 2702verbatim, you can have a look at the EV perl module
… …
2485 2743
2486=item Starting and stopping timer/periodic watchers: O(log skipped_other_timers) 2744=item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
2487 2745
2488This means that, when you have a watcher that triggers in one hour and 2746This means that, when you have a watcher that triggers in one hour and
2489there are 100 watchers that would trigger before that then inserting will 2747there are 100 watchers that would trigger before that then inserting will
2490have to skip those 100 watchers. 2748have to skip roughly seven (C<ld 100>) of these watchers.
2491 2749
2492=item Changing timer/periodic watchers (by autorepeat, again): O(log skipped_other_timers) 2750=item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
2493 2751
2494That means that for changing a timer costs less than removing/adding them 2752That means that changing a timer costs less than removing/adding them
2495as only the relative motion in the event queue has to be paid for. 2753as only the relative motion in the event queue has to be paid for.
2496 2754
2497=item Starting io/check/prepare/idle/signal/child watchers: O(1) 2755=item Starting io/check/prepare/idle/signal/child watchers: O(1)
2498 2756
2499These just add the watcher into an array or at the head of a list. 2757These just add the watcher into an array or at the head of a list.
2758
2500=item Stopping check/prepare/idle watchers: O(1) 2759=item Stopping check/prepare/idle watchers: O(1)
2501 2760
2502=item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE)) 2761=item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
2503 2762
2504These watchers are stored in lists then need to be walked to find the 2763These watchers are stored in lists then need to be walked to find the
2505correct watcher to remove. The lists are usually short (you don't usually 2764correct watcher to remove. The lists are usually short (you don't usually
2506have many watchers waiting for the same fd or signal). 2765have many watchers waiting for the same fd or signal).
2507 2766
2508=item Finding the next timer per loop iteration: O(1) 2767=item Finding the next timer in each loop iteration: O(1)
2768
2769By virtue of using a binary heap, the next timer is always found at the
2770beginning of the storage array.
2509 2771
2510=item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd) 2772=item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
2511 2773
2512A change means an I/O watcher gets started or stopped, which requires 2774A change means an I/O watcher gets started or stopped, which requires
2513libev to recalculate its status (and possibly tell the kernel). 2775libev to recalculate its status (and possibly tell the kernel, depending
2776on backend and wether C<ev_io_set> was used).
2514 2777
2515=item Activating one watcher: O(1) 2778=item Activating one watcher (putting it into the pending state): O(1)
2516 2779
2517=item Priority handling: O(number_of_priorities) 2780=item Priority handling: O(number_of_priorities)
2518 2781
2519Priorities are implemented by allocating some space for each 2782Priorities are implemented by allocating some space for each
2520priority. When doing priority-based operations, libev usually has to 2783priority. When doing priority-based operations, libev usually has to
2521linearly search all the priorities. 2784linearly search all the priorities, but starting/stopping and activating
2785watchers becomes O(1) w.r.t. prioritiy handling.
2522 2786
2523=back 2787=back
2524 2788
2525 2789
2790=head1 Win32 platform limitations and workarounds
2791
2792Win32 doesn't support any of the standards (e.g. POSIX) that libev
2793requires, and its I/O model is fundamentally incompatible with the POSIX
2794model. Libev still offers limited functionality on this platform in
2795the form of the C<EVBACKEND_SELECT> backend, and only supports socket
2796descriptors. This only applies when using Win32 natively, not when using
2797e.g. cygwin.
2798
2799There is no supported compilation method available on windows except
2800embedding it into other applications.
2801
2802Due to the many, low, and arbitrary limits on the win32 platform and the
2803abysmal performance of winsockets, using a large number of sockets is not
2804recommended (and not reasonable). If your program needs to use more than
2805a hundred or so sockets, then likely it needs to use a totally different
2806implementation for windows, as libev offers the POSIX model, which cannot
2807be implemented efficiently on windows (microsoft monopoly games).
2808
2809=over 4
2810
2811=item The winsocket select function
2812
2813The winsocket C<select> function doesn't follow POSIX in that it requires
2814socket I<handles> and not socket I<file descriptors>. This makes select
2815very inefficient, and also requires a mapping from file descriptors
2816to socket handles. See the discussion of the C<EV_SELECT_USE_FD_SET>,
2817C<EV_SELECT_IS_WINSOCKET> and C<EV_FD_TO_WIN32_HANDLE> preprocessor
2818symbols for more info.
2819
2820The configuration for a "naked" win32 using the microsoft runtime
2821libraries and raw winsocket select is:
2822
2823 #define EV_USE_SELECT 1
2824 #define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
2825
2826Note that winsockets handling of fd sets is O(n), so you can easily get a
2827complexity in the O(n²) range when using win32.
2828
2829=item Limited number of file descriptors
2830
2831Windows has numerous arbitrary (and low) limits on things. Early versions
2832of winsocket's select only supported waiting for a max. of C<64> handles
2833(probably owning to the fact that all windows kernels can only wait for
2834C<64> things at the same time internally; microsoft recommends spawning a
2835chain of threads and wait for 63 handles and the previous thread in each).
2836
2837Newer versions support more handles, but you need to define C<FD_SETSIZE>
2838to some high number (e.g. C<2048>) before compiling the winsocket select
2839call (which might be in libev or elsewhere, for example, perl does its own
2840select emulation on windows).
2841
2842Another limit is the number of file descriptors in the microsoft runtime
2843libraries, which by default is C<64> (there must be a hidden I<64> fetish
2844or something like this inside microsoft). You can increase this by calling
2845C<_setmaxstdio>, which can increase this limit to C<2048> (another
2846arbitrary limit), but is broken in many versions of the microsoft runtime
2847libraries.
2848
2849This might get you to about C<512> or C<2048> sockets (depending on
2850windows version and/or the phase of the moon). To get more, you need to
2851wrap all I/O functions and provide your own fd management, but the cost of
2852calling select (O(n²)) will likely make this unworkable.
2853
2854=back
2855
2856
2526=head1 AUTHOR 2857=head1 AUTHOR
2527 2858
2528Marc Lehmann <libev@schmorp.de>. 2859Marc Lehmann <libev@schmorp.de>.
2529 2860

Diff Legend

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