ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/ev.pod
Revision: 1.461
Committed: Wed Jan 22 12:15:52 2020 UTC (4 years, 3 months ago) by root
Branch: MAIN
Changes since 1.460: +3 -0 lines
Log Message:
*** empty log message ***

File Contents

# User Rev Content
1 root 1.430 =encoding utf-8
2    
3 root 1.1 =head1 NAME
4    
5     libev - a high performance full-featured event loop written in C
6    
7     =head1 SYNOPSIS
8    
9 root 1.164 #include <ev.h>
10 root 1.1
11 root 1.105 =head2 EXAMPLE PROGRAM
12 root 1.54
13 root 1.164 // a single header file is required
14     #include <ev.h>
15 root 1.54
16 root 1.217 #include <stdio.h> // for puts
17    
18 root 1.164 // every watcher type has its own typedef'd struct
19 root 1.200 // with the name ev_TYPE
20 root 1.164 ev_io stdin_watcher;
21     ev_timer timeout_watcher;
22    
23     // all watcher callbacks have a similar signature
24     // this callback is called when data is readable on stdin
25     static void
26 root 1.198 stdin_cb (EV_P_ ev_io *w, int revents)
27 root 1.164 {
28     puts ("stdin ready");
29     // for one-shot events, one must manually stop the watcher
30     // with its corresponding stop function.
31     ev_io_stop (EV_A_ w);
32    
33 root 1.310 // this causes all nested ev_run's to stop iterating
34     ev_break (EV_A_ EVBREAK_ALL);
35 root 1.164 }
36    
37     // another callback, this time for a time-out
38     static void
39 root 1.198 timeout_cb (EV_P_ ev_timer *w, int revents)
40 root 1.164 {
41     puts ("timeout");
42 root 1.310 // this causes the innermost ev_run to stop iterating
43     ev_break (EV_A_ EVBREAK_ONE);
44 root 1.164 }
45    
46     int
47     main (void)
48     {
49     // use the default event loop unless you have special needs
50 root 1.322 struct ev_loop *loop = EV_DEFAULT;
51 root 1.164
52     // initialise an io watcher, then start it
53     // this one will watch for stdin to become readable
54     ev_io_init (&stdin_watcher, stdin_cb, /*STDIN_FILENO*/ 0, EV_READ);
55     ev_io_start (loop, &stdin_watcher);
56    
57     // initialise a timer watcher, then start it
58     // simple non-repeating 5.5 second timeout
59     ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
60     ev_timer_start (loop, &timeout_watcher);
61    
62     // now wait for events to arrive
63 root 1.310 ev_run (loop, 0);
64 root 1.164
65 root 1.362 // break was called, so exit
66 root 1.164 return 0;
67     }
68 root 1.53
69 root 1.236 =head1 ABOUT THIS DOCUMENT
70    
71     This document documents the libev software package.
72 root 1.1
73 root 1.135 The newest version of this document is also available as an html-formatted
74 root 1.69 web page you might find easier to navigate when reading it for the first
75 root 1.154 time: L<http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod>.
76 root 1.69
77 root 1.236 While this document tries to be as complete as possible in documenting
78     libev, its usage and the rationale behind its design, it is not a tutorial
79     on event-based programming, nor will it introduce event-based programming
80     with libev.
81    
82 sf-exg 1.298 Familiarity with event based programming techniques in general is assumed
83 root 1.236 throughout this document.
84    
85 root 1.334 =head1 WHAT TO READ WHEN IN A HURRY
86    
87     This manual tries to be very detailed, but unfortunately, this also makes
88     it very long. If you just want to know the basics of libev, I suggest
89 root 1.414 reading L</ANATOMY OF A WATCHER>, then the L</EXAMPLE PROGRAM> above and
90 root 1.413 look up the missing functions in L</GLOBAL FUNCTIONS> and the C<ev_io> and
91     C<ev_timer> sections in L</WATCHER TYPES>.
92 root 1.334
93 root 1.236 =head1 ABOUT LIBEV
94    
95 root 1.1 Libev is an event loop: you register interest in certain events (such as a
96 root 1.92 file descriptor being readable or a timeout occurring), and it will manage
97 root 1.4 these event sources and provide your program with events.
98 root 1.1
99     To do this, it must take more or less complete control over your process
100     (or thread) by executing the I<event loop> handler, and will then
101     communicate events via a callback mechanism.
102    
103     You register interest in certain events by registering so-called I<event
104     watchers>, which are relatively small C structures you initialise with the
105     details of the event, and then hand it over to libev by I<starting> the
106     watcher.
107    
108 root 1.105 =head2 FEATURES
109 root 1.1
110 root 1.447 Libev supports C<select>, C<poll>, the Linux-specific aio and C<epoll>
111     interfaces, the BSD-specific C<kqueue> and the Solaris-specific event port
112     mechanisms for file descriptor events (C<ev_io>), the Linux C<inotify>
113     interface (for C<ev_stat>), Linux eventfd/signalfd (for faster and cleaner
114 root 1.261 inter-thread wakeup (C<ev_async>)/signal handling (C<ev_signal>)) relative
115     timers (C<ev_timer>), absolute timers with customised rescheduling
116     (C<ev_periodic>), synchronous signals (C<ev_signal>), process status
117     change events (C<ev_child>), and event watchers dealing with the event
118     loop mechanism itself (C<ev_idle>, C<ev_embed>, C<ev_prepare> and
119     C<ev_check> watchers) as well as file watchers (C<ev_stat>) and even
120     limited support for fork events (C<ev_fork>).
121 root 1.54
122     It also is quite fast (see this
123     L<benchmark|http://libev.schmorp.de/bench.html> comparing it to libevent
124     for example).
125 root 1.1
126 root 1.105 =head2 CONVENTIONS
127 root 1.1
128 root 1.135 Libev is very configurable. In this manual the default (and most common)
129     configuration will be described, which supports multiple event loops. For
130     more info about various configuration options please have a look at
131     B<EMBED> section in this manual. If libev was configured without support
132     for multiple event loops, then all functions taking an initial argument of
133 root 1.274 name C<loop> (which is always of type C<struct ev_loop *>) will not have
134 root 1.135 this argument.
135 root 1.1
136 root 1.105 =head2 TIME REPRESENTATION
137 root 1.1
138 root 1.237 Libev represents time as a single floating point number, representing
139 root 1.317 the (fractional) number of seconds since the (POSIX) epoch (in practice
140 root 1.293 somewhere near the beginning of 1970, details are complicated, don't
141     ask). This type is called C<ev_tstamp>, which is what you should use
142     too. It usually aliases to the C<double> type in C. When you need to do
143     any calculations on it, you should treat it as some floating point value.
144    
145     Unlike the name component C<stamp> might indicate, it is also used for
146     time differences (e.g. delays) throughout libev.
147 root 1.34
148 root 1.160 =head1 ERROR HANDLING
149    
150     Libev knows three classes of errors: operating system errors, usage errors
151     and internal errors (bugs).
152    
153     When libev catches an operating system error it cannot handle (for example
154 root 1.161 a system call indicating a condition libev cannot fix), it calls the callback
155 root 1.160 set via C<ev_set_syserr_cb>, which is supposed to fix the problem or
156     abort. The default is to print a diagnostic message and to call C<abort
157     ()>.
158    
159     When libev detects a usage error such as a negative timer interval, then
160     it will print a diagnostic message and abort (via the C<assert> mechanism,
161     so C<NDEBUG> will disable this checking): these are programming errors in
162     the libev caller and need to be fixed there.
163    
164 root 1.455 Via the C<EV_FREQUENT> macro you can compile in and/or enable extensive
165     consistency checking code inside libev that can be used to check for
166     internal inconsistencies, suually caused by application bugs.
167    
168     Libev also has a few internal error-checking C<assert>ions. These do not
169     trigger under normal circumstances, as they indicate either a bug in libev
170     or worse.
171 root 1.160
172    
173 root 1.17 =head1 GLOBAL FUNCTIONS
174    
175 root 1.18 These functions can be called anytime, even before initialising the
176     library in any way.
177    
178 root 1.1 =over 4
179    
180     =item ev_tstamp ev_time ()
181    
182 root 1.26 Returns the current time as libev would use it. Please note that the
183     C<ev_now> function is usually faster and also often returns the timestamp
184 sf-exg 1.321 you actually want to know. Also interesting is the combination of
185 sf-exg 1.384 C<ev_now_update> and C<ev_now>.
186 root 1.1
187 root 1.97 =item ev_sleep (ev_tstamp interval)
188    
189 root 1.371 Sleep for the given interval: The current thread will be blocked
190     until either it is interrupted or the given time interval has
191     passed (approximately - it might return a bit earlier even if not
192     interrupted). Returns immediately if C<< interval <= 0 >>.
193    
194     Basically this is a sub-second-resolution C<sleep ()>.
195    
196     The range of the C<interval> is limited - libev only guarantees to work
197     with sleep times of up to one day (C<< interval <= 86400 >>).
198 root 1.97
199 root 1.1 =item int ev_version_major ()
200    
201     =item int ev_version_minor ()
202    
203 root 1.80 You can find out the major and minor ABI version numbers of the library
204 root 1.1 you linked against by calling the functions C<ev_version_major> and
205     C<ev_version_minor>. If you want, you can compare against the global
206     symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the
207     version of the library your program was compiled against.
208    
209 root 1.80 These version numbers refer to the ABI version of the library, not the
210     release version.
211 root 1.79
212 root 1.9 Usually, it's a good idea to terminate if the major versions mismatch,
213 root 1.79 as this indicates an incompatible change. Minor versions are usually
214 root 1.1 compatible to older versions, so a larger minor version alone is usually
215     not a problem.
216    
217 root 1.54 Example: Make sure we haven't accidentally been linked against the wrong
218 root 1.320 version (note, however, that this will not detect other ABI mismatches,
219     such as LFS or reentrancy).
220 root 1.34
221 root 1.164 assert (("libev version mismatch",
222     ev_version_major () == EV_VERSION_MAJOR
223     && ev_version_minor () >= EV_VERSION_MINOR));
224 root 1.34
225 root 1.31 =item unsigned int ev_supported_backends ()
226    
227     Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
228     value) compiled into this binary of libev (independent of their
229     availability on the system you are running on). See C<ev_default_loop> for
230     a description of the set values.
231    
232 root 1.34 Example: make sure we have the epoll method, because yeah this is cool and
233     a must have and can we have a torrent of it please!!!11
234    
235 root 1.164 assert (("sorry, no epoll, no sex",
236     ev_supported_backends () & EVBACKEND_EPOLL));
237 root 1.34
238 root 1.31 =item unsigned int ev_recommended_backends ()
239    
240 root 1.318 Return the set of all backends compiled into this binary of libev and
241     also recommended for this platform, meaning it will work for most file
242     descriptor types. This set is often smaller than the one returned by
243     C<ev_supported_backends>, as for example kqueue is broken on most BSDs
244     and will not be auto-detected unless you explicitly request it (assuming
245     you know what you are doing). This is the set of backends that libev will
246     probe for if you specify no backends explicitly.
247 root 1.31
248 root 1.35 =item unsigned int ev_embeddable_backends ()
249    
250     Returns the set of backends that are embeddable in other event loops. This
251 root 1.319 value is platform-specific but can include backends not available on the
252     current system. To find which embeddable backends might be supported on
253     the current system, you would need to look at C<ev_embeddable_backends ()
254     & ev_supported_backends ()>, likewise for recommended ones.
255 root 1.35
256     See the description of C<ev_embed> watchers for more info.
257    
258 root 1.401 =item ev_set_allocator (void *(*cb)(void *ptr, long size) throw ())
259 root 1.1
260 root 1.59 Sets the allocation function to use (the prototype is similar - the
261 root 1.145 semantics are identical to the C<realloc> C89/SuS/POSIX function). It is
262     used to allocate and free memory (no surprises here). If it returns zero
263     when memory needs to be allocated (C<size != 0>), the library might abort
264     or take some potentially destructive action.
265    
266     Since some systems (at least OpenBSD and Darwin) fail to implement
267     correct C<realloc> semantics, libev will use a wrapper around the system
268     C<realloc> and C<free> functions by default.
269 root 1.1
270     You could override this function in high-availability programs to, say,
271     free some memory if it cannot allocate memory, to use a special allocator,
272     or even to sleep a while and retry until some memory is available.
273    
274 root 1.446 Example: The following is the C<realloc> function that libev itself uses
275     which should work with C<realloc> and C<free> functions of all kinds and
276     is probably a good basis for your own implementation.
277    
278     static void *
279     ev_realloc_emul (void *ptr, long size) EV_NOEXCEPT
280     {
281     if (size)
282     return realloc (ptr, size);
283    
284     free (ptr);
285     return 0;
286     }
287    
288 root 1.54 Example: Replace the libev allocator with one that waits a bit and then
289 root 1.446 retries.
290 root 1.34
291     static void *
292 root 1.52 persistent_realloc (void *ptr, size_t size)
293 root 1.34 {
294 root 1.446 if (!size)
295     {
296     free (ptr);
297     return 0;
298     }
299    
300 root 1.34 for (;;)
301     {
302     void *newptr = realloc (ptr, size);
303    
304     if (newptr)
305     return newptr;
306    
307     sleep (60);
308     }
309     }
310    
311     ...
312     ev_set_allocator (persistent_realloc);
313    
314 root 1.401 =item ev_set_syserr_cb (void (*cb)(const char *msg) throw ())
315 root 1.1
316 root 1.161 Set the callback function to call on a retryable system call error (such
317 root 1.1 as failed select, poll, epoll_wait). The message is a printable string
318     indicating the system call or subsystem causing the problem. If this
319 root 1.161 callback is set, then libev will expect it to remedy the situation, no
320 root 1.7 matter what, when it returns. That is, libev will generally retry the
321 root 1.1 requested operation, or, if the condition doesn't go away, do bad stuff
322     (such as abort).
323    
324 root 1.54 Example: This is basically the same thing that libev does internally, too.
325 root 1.34
326     static void
327     fatal_error (const char *msg)
328     {
329     perror (msg);
330     abort ();
331     }
332    
333     ...
334     ev_set_syserr_cb (fatal_error);
335    
336 root 1.349 =item ev_feed_signal (int signum)
337    
338     This function can be used to "simulate" a signal receive. It is completely
339     safe to call this function at any time, from any context, including signal
340     handlers or random threads.
341    
342 sf-exg 1.350 Its main use is to customise signal handling in your process, especially
343 root 1.349 in the presence of threads. For example, you could block signals
344     by default in all threads (and specifying C<EVFLAG_NOSIGMASK> when
345     creating any loops), and in one thread, use C<sigwait> or any other
346     mechanism to wait for signals, then "deliver" them to libev by calling
347     C<ev_feed_signal>.
348    
349 root 1.1 =back
350    
351 root 1.322 =head1 FUNCTIONS CONTROLLING EVENT LOOPS
352 root 1.1
353 root 1.310 An event loop is described by a C<struct ev_loop *> (the C<struct> is
354 root 1.311 I<not> optional in this case unless libev 3 compatibility is disabled, as
355     libev 3 had an C<ev_loop> function colliding with the struct name).
356 root 1.200
357     The library knows two types of such loops, the I<default> loop, which
358 root 1.331 supports child process events, and dynamically created event loops which
359     do not.
360 root 1.1
361     =over 4
362    
363     =item struct ev_loop *ev_default_loop (unsigned int flags)
364    
365 root 1.322 This returns the "default" event loop object, which is what you should
366     normally use when you just need "the event loop". Event loop objects and
367     the C<flags> parameter are described in more detail in the entry for
368     C<ev_loop_new>.
369    
370     If the default loop is already initialised then this function simply
371     returns it (and ignores the flags. If that is troubling you, check
372     C<ev_backend ()> afterwards). Otherwise it will create it with the given
373     flags, which should almost always be C<0>, unless the caller is also the
374     one calling C<ev_run> or otherwise qualifies as "the main program".
375 root 1.1
376     If you don't know what event loop to use, use the one returned from this
377 root 1.322 function (or via the C<EV_DEFAULT> macro).
378 root 1.1
379 root 1.139 Note that this function is I<not> thread-safe, so if you want to use it
380 root 1.322 from multiple threads, you have to employ some kind of mutex (note also
381     that this case is unlikely, as loops cannot be shared easily between
382     threads anyway).
383    
384     The default loop is the only loop that can handle C<ev_child> watchers,
385     and to do this, it always registers a handler for C<SIGCHLD>. If this is
386     a problem for your application you can either create a dynamic loop with
387     C<ev_loop_new> which doesn't do that, or you can simply overwrite the
388     C<SIGCHLD> signal handler I<after> calling C<ev_default_init>.
389 root 1.139
390 root 1.322 Example: This is the most typical usage.
391    
392     if (!ev_default_loop (0))
393     fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
394    
395     Example: Restrict libev to the select and poll backends, and do not allow
396     environment settings to be taken into account:
397    
398     ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
399    
400     =item struct ev_loop *ev_loop_new (unsigned int flags)
401    
402     This will create and initialise a new event loop object. If the loop
403     could not be initialised, returns false.
404    
405 root 1.343 This function is thread-safe, and one common way to use libev with
406     threads is indeed to create one loop per thread, and using the default
407     loop in the "main" or "initial" thread.
408 root 1.118
409 root 1.1 The flags argument can be used to specify special behaviour or specific
410 root 1.33 backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
411 root 1.1
412 root 1.33 The following flags are supported:
413 root 1.1
414     =over 4
415    
416 root 1.10 =item C<EVFLAG_AUTO>
417 root 1.1
418 root 1.9 The default flags value. Use this if you have no clue (it's the right
419 root 1.1 thing, believe me).
420    
421 root 1.10 =item C<EVFLAG_NOENV>
422 root 1.1
423 root 1.161 If this flag bit is or'ed into the flag value (or the program runs setuid
424 root 1.8 or setgid) then libev will I<not> look at the environment variable
425     C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
426     override the flags completely if it is found in the environment. This is
427 root 1.427 useful to try out specific backends to test their performance, to work
428     around bugs, or to make libev threadsafe (accessing environment variables
429     cannot be done in a threadsafe way, but usually it works if no other
430     thread modifies them).
431 root 1.1
432 root 1.62 =item C<EVFLAG_FORKCHECK>
433    
434 root 1.291 Instead of calling C<ev_loop_fork> manually after a fork, you can also
435     make libev check for a fork in each iteration by enabling this flag.
436 root 1.62
437     This works by calling C<getpid ()> on every iteration of the loop,
438     and thus this might slow down your event loop if you do a lot of loop
439 ayin 1.65 iterations and little real work, but is usually not noticeable (on my
440 root 1.441 GNU/Linux system for example, C<getpid> is actually a simple 5-insn
441     sequence without a system call and thus I<very> fast, but my GNU/Linux
442     system also has C<pthread_atfork> which is even faster). (Update: glibc
443     versions 2.25 apparently removed the C<getpid> optimisation again).
444 root 1.62
445     The big advantage of this flag is that you can forget about fork (and
446 root 1.436 forget about forgetting to tell libev about forking, although you still
447     have to ignore C<SIGPIPE>) when you use this flag.
448 root 1.62
449 root 1.161 This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
450 root 1.62 environment variable.
451    
452 root 1.260 =item C<EVFLAG_NOINOTIFY>
453    
454     When this flag is specified, then libev will not attempt to use the
455 root 1.340 I<inotify> API for its C<ev_stat> watchers. Apart from debugging and
456 root 1.260 testing, this flag can be useful to conserve inotify file descriptors, as
457     otherwise each loop using C<ev_stat> watchers consumes one inotify handle.
458    
459 root 1.277 =item C<EVFLAG_SIGNALFD>
460 root 1.260
461 root 1.277 When this flag is specified, then libev will attempt to use the
462 root 1.340 I<signalfd> API for its C<ev_signal> (and C<ev_child>) watchers. This API
463 root 1.278 delivers signals synchronously, which makes it both faster and might make
464     it possible to get the queued signal data. It can also simplify signal
465     handling with threads, as long as you properly block signals in your
466     threads that are not interested in handling them.
467 root 1.277
468     Signalfd will not be used by default as this changes your signal mask, and
469     there are a lot of shoddy libraries and programs (glib's threadpool for
470     example) that can't properly initialise their signal masks.
471 root 1.260
472 root 1.349 =item C<EVFLAG_NOSIGMASK>
473    
474     When this flag is specified, then libev will avoid to modify the signal
475 sf-exg 1.374 mask. Specifically, this means you have to make sure signals are unblocked
476 root 1.349 when you want to receive them.
477    
478     This behaviour is useful when you want to do your own signal handling, or
479     want to handle signals only in specific threads and want to avoid libev
480     unblocking the signals.
481    
482 root 1.360 It's also required by POSIX in a threaded program, as libev calls
483     C<sigprocmask>, whose behaviour is officially unspecified.
484    
485 root 1.458 =item C<EVFLAG_NOTIMERFD>
486    
487     When this flag is specified, the libev will avoid using a C<timerfd> to
488     detect time jumps. It will still be able to detect time jumps, but takes
489     longer and has a lower accuracy in doing so, but saves a file descriptor
490     per loop.
491    
492     The current implementation only tries to use a C<timerfd> when the first
493     C<ev_periodic> watcher is started and falls back on other methods if it
494     cannot be created, but this behaviour might change in the future.
495 root 1.349
496 root 1.31 =item C<EVBACKEND_SELECT> (value 1, portable select backend)
497 root 1.1
498 root 1.29 This is your standard select(2) backend. Not I<completely> standard, as
499     libev tries to roll its own fd_set with no limits on the number of fds,
500     but if that fails, expect a fairly low limit on the number of fds when
501 root 1.102 using this backend. It doesn't scale too well (O(highest_fd)), but its
502     usually the fastest backend for a low number of (low-numbered :) fds.
503    
504     To get good performance out of this backend you need a high amount of
505 root 1.161 parallelism (most of the file descriptors should be busy). If you are
506 root 1.102 writing a server, you should C<accept ()> in a loop to accept as many
507     connections as possible during one iteration. You might also want to have
508     a look at C<ev_set_io_collect_interval ()> to increase the amount of
509 root 1.155 readiness notifications you get per iteration.
510 root 1.1
511 root 1.179 This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
512     C<writefds> set (and to work around Microsoft Windows bugs, also onto the
513     C<exceptfds> set on that platform).
514    
515 root 1.31 =item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
516 root 1.1
517 root 1.102 And this is your standard poll(2) backend. It's more complicated
518     than select, but handles sparse fds better and has no artificial
519     limit on the number of fds you can use (except it will slow down
520     considerably with a lot of inactive fds). It scales similarly to select,
521     i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
522     performance tips.
523 root 1.1
524 root 1.179 This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
525     C<EV_WRITE> to C<POLLOUT | POLLERR | POLLHUP>.
526    
527 root 1.31 =item C<EVBACKEND_EPOLL> (value 4, Linux)
528 root 1.1
529 root 1.454 Use the Linux-specific epoll(7) interface (for both pre- and post-2.6.9
530 root 1.272 kernels).
531    
532 root 1.368 For few fds, this backend is a bit little slower than poll and select, but
533     it scales phenomenally better. While poll and select usually scale like
534     O(total_fds) where total_fds is the total number of fds (or the highest
535     fd), epoll scales either O(1) or O(active_fds).
536 root 1.205
537 root 1.210 The epoll mechanism deserves honorable mention as the most misdesigned
538     of the more advanced event mechanisms: mere annoyances include silently
539     dropping file descriptors, requiring a system call per change per file
540 root 1.337 descriptor (and unnecessary guessing of parameters), problems with dup,
541 root 1.338 returning before the timeout value, resulting in additional iterations
542     (and only giving 5ms accuracy while select on the same platform gives
543     0.1ms) and so on. The biggest issue is fork races, however - if a program
544     forks then I<both> parent and child process have to recreate the epoll
545     set, which can take considerable time (one syscall per file descriptor)
546     and is of course hard to detect.
547 root 1.1
548 root 1.370 Epoll is also notoriously buggy - embedding epoll fds I<should> work,
549     but of course I<doesn't>, and epoll just loves to report events for
550     totally I<different> file descriptors (even already closed ones, so
551     one cannot even remove them from the set) than registered in the set
552     (especially on SMP systems). Libev tries to counter these spurious
553     notifications by employing an additional generation counter and comparing
554     that against the events to filter out spurious ones, recreating the set
555 sf-exg 1.374 when required. Epoll also erroneously rounds down timeouts, but gives you
556 root 1.370 no way to know when and by how much, so sometimes you have to busy-wait
557     because epoll returns immediately despite a nonzero timeout. And last
558 root 1.306 not least, it also refuses to work with some file descriptors which work
559     perfectly fine with C<select> (files, many character devices...).
560 root 1.204
561 root 1.370 Epoll is truly the train wreck among event poll mechanisms, a frankenpoll,
562     cobbled together in a hurry, no thought to design or interaction with
563     others. Oh, the pain, will it ever stop...
564 root 1.338
565 root 1.94 While stopping, setting and starting an I/O watcher in the same iteration
566 root 1.210 will result in some caching, there is still a system call per such
567     incident (because the same I<file descriptor> could point to a different
568     I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
569     file descriptors might not work very well if you register events for both
570     file descriptors.
571 root 1.29
572 root 1.102 Best performance from this backend is achieved by not unregistering all
573 root 1.183 watchers for a file descriptor until it has been closed, if possible,
574     i.e. keep at least one watcher active per fd at all times. Stopping and
575     starting a watcher (without re-setting it) also usually doesn't cause
576 root 1.206 extra overhead. A fork can both result in spurious notifications as well
577     as in libev having to destroy and recreate the epoll object, which can
578     take considerable time and thus should be avoided.
579 root 1.102
580 root 1.215 All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
581     faster than epoll for maybe up to a hundred file descriptors, depending on
582 root 1.214 the usage. So sad.
583 root 1.213
584 root 1.161 While nominally embeddable in other event loops, this feature is broken in
585 root 1.447 a lot of kernel revisions, but probably(!) works in current versions.
586    
587     This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
588     C<EVBACKEND_POLL>.
589    
590     =item C<EVBACKEND_LINUXAIO> (value 64, Linux)
591    
592 root 1.454 Use the Linux-specific Linux AIO (I<not> C<< aio(7) >> but C<<
593 root 1.452 io_submit(2) >>) event interface available in post-4.18 kernels (but libev
594     only tries to use it in 4.19+).
595    
596 root 1.454 This is another Linux train wreck of an event interface.
597 root 1.447
598     If this backend works for you (as of this writing, it was very
599 root 1.454 experimental), it is the best event interface available on Linux and might
600 root 1.448 be well worth enabling it - if it isn't available in your kernel this will
601     be detected and this backend will be skipped.
602    
603     This backend can batch oneshot requests and supports a user-space ring
604     buffer to receive events. It also doesn't suffer from most of the design
605 root 1.452 problems of epoll (such as not being able to remove event sources from
606     the epoll set), and generally sounds too good to be true. Because, this
607 root 1.454 being the Linux kernel, of course it suffers from a whole new set of
608 root 1.452 limitations, forcing you to fall back to epoll, inheriting all its design
609     issues.
610 root 1.447
611     For one, it is not easily embeddable (but probably could be done using
612 root 1.448 an event fd at some extra overhead). It also is subject to a system wide
613 root 1.454 limit that can be configured in F</proc/sys/fs/aio-max-nr>. If no AIO
614 root 1.452 requests are left, this backend will be skipped during initialisation, and
615     will switch to epoll when the loop is active.
616 root 1.448
617 root 1.452 Most problematic in practice, however, is that not all file descriptors
618 root 1.454 work with it. For example, in Linux 5.1, TCP sockets, pipes, event fds,
619     files, F</dev/null> and many others are supported, but ttys do not work
620 root 1.450 properly (a known bug that the kernel developers don't care about, see
621     L<https://lore.kernel.org/patchwork/patch/1047453/>), so this is not
622     (yet?) a generic event polling interface.
623 root 1.448
624 root 1.454 Overall, it seems the Linux developers just don't want it to have a
625 root 1.451 generic event handling mechanism other than C<select> or C<poll>.
626    
627 root 1.452 To work around all these problem, the current version of libev uses its
628     epoll backend as a fallback for file descriptor types that do not work. Or
629     falls back completely to epoll if the kernel acts up.
630 root 1.102
631 root 1.179 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
632     C<EVBACKEND_POLL>.
633    
634 root 1.31 =item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
635 root 1.29
636 root 1.453 Kqueue deserves special mention, as at the time this backend was
637     implemented, it was broken on all BSDs except NetBSD (usually it doesn't
638     work reliably with anything but sockets and pipes, except on Darwin,
639     where of course it's completely useless). Unlike epoll, however, whose
640     brokenness is by design, these kqueue bugs can be (and mostly have been)
641     fixed without API changes to existing programs. For this reason it's not
642     being "auto-detected" on all platforms unless you explicitly specify it
643     in the flags (i.e. using C<EVBACKEND_KQUEUE>) or libev was compiled on a
644     known-to-be-good (-enough) system like NetBSD.
645 root 1.29
646 root 1.100 You still can embed kqueue into a normal poll or select backend and use it
647     only for sockets (after having made sure that sockets work with kqueue on
648     the target platform). See C<ev_embed> watchers for more info.
649    
650 root 1.29 It scales in the same way as the epoll backend, but the interface to the
651 root 1.100 kernel is more efficient (which says nothing about its actual speed, of
652     course). While stopping, setting and starting an I/O watcher does never
653 root 1.161 cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
654 root 1.398 two event changes per incident. Support for C<fork ()> is very bad (you
655 root 1.454 might have to leak fds on fork, but it's more sane than epoll) and it
656 root 1.422 drops fds silently in similarly hard-to-detect cases.
657 root 1.29
658 root 1.102 This backend usually performs well under most conditions.
659    
660     While nominally embeddable in other event loops, this doesn't work
661     everywhere, so you might need to test for this. And since it is broken
662     almost everywhere, you should only use it when you have a lot of sockets
663     (for which it usually works), by embedding it into another event loop
664 root 1.223 (e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
665     also broken on OS X)) and, did I mention it, using it only for sockets.
666 root 1.102
667 root 1.179 This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
668     C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
669     C<NOTE_EOF>.
670    
671 root 1.31 =item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
672 root 1.29
673 root 1.102 This is not implemented yet (and might never be, unless you send me an
674     implementation). According to reports, C</dev/poll> only supports sockets
675     and is not embeddable, which would limit the usefulness of this backend
676     immensely.
677 root 1.29
678 root 1.31 =item C<EVBACKEND_PORT> (value 32, Solaris 10)
679 root 1.29
680 root 1.94 This uses the Solaris 10 event port mechanism. As with everything on Solaris,
681 root 1.29 it's really slow, but it still scales very well (O(active_fds)).
682    
683 root 1.102 While this backend scales well, it requires one system call per active
684     file descriptor per loop iteration. For small and medium numbers of file
685     descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
686     might perform better.
687    
688 root 1.351 On the positive side, this backend actually performed fully to
689     specification in all tests and is fully embeddable, which is a rare feat
690     among the OS-specific backends (I vastly prefer correctness over speed
691     hacks).
692    
693 root 1.352 On the negative side, the interface is I<bizarre> - so bizarre that
694     even sun itself gets it wrong in their code examples: The event polling
695 root 1.375 function sometimes returns events to the caller even though an error
696 root 1.353 occurred, but with no indication whether it has done so or not (yes, it's
697 root 1.375 even documented that way) - deadly for edge-triggered interfaces where you
698     absolutely have to know whether an event occurred or not because you have
699     to re-arm the watcher.
700 root 1.352
701     Fortunately libev seems to be able to work around these idiocies.
702 root 1.117
703 root 1.179 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
704     C<EVBACKEND_POLL>.
705    
706 root 1.31 =item C<EVBACKEND_ALL>
707 root 1.29
708     Try all backends (even potentially broken ones that wouldn't be tried
709     with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
710 root 1.31 C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.
711 root 1.1
712 root 1.349 It is definitely not recommended to use this flag, use whatever
713     C<ev_recommended_backends ()> returns, or simply do not specify a backend
714     at all.
715    
716     =item C<EVBACKEND_MASK>
717    
718     Not a backend at all, but a mask to select all backend bits from a
719     C<flags> value, in case you want to mask out any backends from a flags
720     value (e.g. when modifying the C<LIBEV_FLAGS> environment variable).
721 root 1.102
722 root 1.1 =back
723    
724 root 1.260 If one or more of the backend flags are or'ed into the flags value,
725     then only these backends will be tried (in the reverse order as listed
726     here). If none are specified, all backends in C<ev_recommended_backends
727     ()> will be tried.
728 root 1.29
729 root 1.54 Example: Try to create a event loop that uses epoll and nothing else.
730 root 1.34
731 root 1.164 struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
732     if (!epoller)
733     fatal ("no epoll found here, maybe it hides under your chair");
734 root 1.34
735 root 1.323 Example: Use whatever libev has to offer, but make sure that kqueue is
736     used if available.
737    
738     struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_KQUEUE);
739    
740 root 1.447 Example: Similarly, on linux, you mgiht want to take advantage of the
741     linux aio backend if possible, but fall back to something else if that
742     isn't available.
743    
744     struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_LINUXAIO);
745    
746 root 1.322 =item ev_loop_destroy (loop)
747 root 1.1
748 root 1.322 Destroys an event loop object (frees all memory and kernel state
749     etc.). None of the active event watchers will be stopped in the normal
750     sense, so e.g. C<ev_is_active> might still return true. It is your
751     responsibility to either stop all watchers cleanly yourself I<before>
752     calling this function, or cope with the fact afterwards (which is usually
753     the easiest thing, you can just ignore the watchers and/or C<free ()> them
754     for example).
755 root 1.1
756 root 1.203 Note that certain global state, such as signal state (and installed signal
757     handlers), will not be freed by this function, and related watchers (such
758     as signal and child watchers) would need to be stopped manually.
759 root 1.87
760 root 1.322 This function is normally used on loop objects allocated by
761     C<ev_loop_new>, but it can also be used on the default loop returned by
762     C<ev_default_loop>, in which case it is not thread-safe.
763    
764     Note that it is not advisable to call this function on the default loop
765 root 1.340 except in the rare occasion where you really need to free its resources.
766 root 1.322 If you need dynamically allocated loops it is better to use C<ev_loop_new>
767     and C<ev_loop_destroy>.
768 root 1.87
769 root 1.322 =item ev_loop_fork (loop)
770 root 1.1
771 root 1.433 This function sets a flag that causes subsequent C<ev_run> iterations
772     to reinitialise the kernel state for backends that have one. Despite
773     the name, you can call it anytime you are allowed to start or stop
774     watchers (except inside an C<ev_prepare> callback), but it makes most
775     sense after forking, in the child process. You I<must> call it (or use
776     C<EVFLAG_FORKCHECK>) in the child before resuming or calling C<ev_run>.
777 root 1.119
778 root 1.437 In addition, if you want to reuse a loop (via this function or
779 root 1.436 C<EVFLAG_FORKCHECK>), you I<also> have to ignore C<SIGPIPE>.
780    
781 root 1.429 Again, you I<have> to call it on I<any> loop that you want to re-use after
782 root 1.291 a fork, I<even if you do not plan to use the loop in the parent>. This is
783     because some kernel interfaces *cough* I<kqueue> *cough* do funny things
784     during fork.
785    
786 root 1.119 On the other hand, you only need to call this function in the child
787 root 1.310 process if and only if you want to use the event loop in the child. If
788     you just fork+exec or create a new loop in the child, you don't have to
789     call it at all (in fact, C<epoll> is so badly broken that it makes a
790     difference, but libev will usually detect this case on its own and do a
791     costly reset of the backend).
792 root 1.1
793 root 1.9 The function itself is quite fast and it's usually not a problem to call
794 root 1.322 it just in case after a fork.
795 root 1.1
796 root 1.322 Example: Automate calling C<ev_loop_fork> on the default loop when
797     using pthreads.
798 root 1.1
799 root 1.322 static void
800     post_fork_child (void)
801     {
802     ev_loop_fork (EV_DEFAULT);
803     }
804 root 1.1
805 root 1.322 ...
806     pthread_atfork (0, 0, post_fork_child);
807 root 1.1
808 root 1.131 =item int ev_is_default_loop (loop)
809    
810 root 1.183 Returns true when the given loop is, in fact, the default loop, and false
811     otherwise.
812 root 1.131
813 root 1.291 =item unsigned int ev_iteration (loop)
814 root 1.66
815 root 1.310 Returns the current iteration count for the event loop, which is identical
816     to the number of times libev did poll for new events. It starts at C<0>
817     and happily wraps around with enough iterations.
818 root 1.66
819     This value can sometimes be useful as a generation counter of sorts (it
820     "ticks" the number of loop iterations), as it roughly corresponds with
821 root 1.291 C<ev_prepare> and C<ev_check> calls - and is incremented between the
822     prepare and check phases.
823 root 1.66
824 root 1.291 =item unsigned int ev_depth (loop)
825 root 1.247
826 root 1.310 Returns the number of times C<ev_run> was entered minus the number of
827 root 1.343 times C<ev_run> was exited normally, in other words, the recursion depth.
828 root 1.247
829 root 1.310 Outside C<ev_run>, this number is zero. In a callback, this number is
830     C<1>, unless C<ev_run> was invoked recursively (or from another thread),
831 root 1.247 in which case it is higher.
832    
833 root 1.343 Leaving C<ev_run> abnormally (setjmp/longjmp, cancelling the thread,
834     throwing an exception etc.), doesn't count as "exit" - consider this
835     as a hint to avoid such ungentleman-like behaviour unless it's really
836     convenient, in which case it is fully supported.
837 root 1.247
838 root 1.31 =item unsigned int ev_backend (loop)
839 root 1.1
840 root 1.31 Returns one of the C<EVBACKEND_*> flags indicating the event backend in
841 root 1.1 use.
842    
843 root 1.9 =item ev_tstamp ev_now (loop)
844 root 1.1
845     Returns the current "event loop time", which is the time the event loop
846 root 1.34 received events and started processing them. This timestamp does not
847     change as long as callbacks are being processed, and this is also the base
848     time used for relative timers. You can treat it as the timestamp of the
849 root 1.92 event occurring (or more correctly, libev finding out about it).
850 root 1.1
851 root 1.176 =item ev_now_update (loop)
852    
853     Establishes the current time by querying the kernel, updating the time
854     returned by C<ev_now ()> in the progress. This is a costly operation and
855 root 1.310 is usually done automatically within C<ev_run ()>.
856 root 1.176
857     This function is rarely useful, but when some event callback runs for a
858     very long time without entering the event loop, updating libev's idea of
859     the current time is a good idea.
860    
861 root 1.413 See also L</The special problem of time updates> in the C<ev_timer> section.
862 root 1.176
863 root 1.231 =item ev_suspend (loop)
864    
865     =item ev_resume (loop)
866    
867 root 1.310 These two functions suspend and resume an event loop, for use when the
868     loop is not used for a while and timeouts should not be processed.
869 root 1.231
870     A typical use case would be an interactive program such as a game: When
871     the user presses C<^Z> to suspend the game and resumes it an hour later it
872     would be best to handle timeouts as if no time had actually passed while
873     the program was suspended. This can be achieved by calling C<ev_suspend>
874     in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
875     C<ev_resume> directly afterwards to resume timer processing.
876    
877     Effectively, all C<ev_timer> watchers will be delayed by the time spend
878     between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
879     will be rescheduled (that is, they will lose any events that would have
880 sf-exg 1.298 occurred while suspended).
881 root 1.231
882     After calling C<ev_suspend> you B<must not> call I<any> function on the
883     given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
884     without a previous call to C<ev_suspend>.
885    
886     Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
887     event loop time (see C<ev_now_update>).
888    
889 root 1.399 =item bool ev_run (loop, int flags)
890 root 1.1
891     Finally, this is it, the event handler. This function usually is called
892 root 1.268 after you have initialised all your watchers and you want to start
893 root 1.310 handling events. It will ask the operating system for any new events, call
894 root 1.399 the watcher callbacks, and then repeat the whole process indefinitely: This
895 root 1.310 is why event loops are called I<loops>.
896 root 1.1
897 root 1.310 If the flags argument is specified as C<0>, it will keep handling events
898     until either no event watchers are active anymore or C<ev_break> was
899     called.
900 root 1.1
901 root 1.399 The return value is false if there are no more active watchers (which
902     usually means "all jobs done" or "deadlock"), and true in all other cases
903     (which usually means " you should call C<ev_run> again").
904    
905 root 1.310 Please note that an explicit C<ev_break> is usually better than
906 root 1.34 relying on all watchers to be stopped when deciding when a program has
907 root 1.183 finished (especially in interactive programs), but having a program
908     that automatically loops as long as it has to and no longer by virtue
909     of relying on its watchers stopping correctly, that is truly a thing of
910     beauty.
911 root 1.34
912 root 1.399 This function is I<mostly> exception-safe - you can break out of a
913     C<ev_run> call by calling C<longjmp> in a callback, throwing a C++
914 root 1.343 exception and so on. This does not decrement the C<ev_depth> value, nor
915     will it clear any outstanding C<EVBREAK_ONE> breaks.
916    
917 root 1.310 A flags value of C<EVRUN_NOWAIT> will look for new events, will handle
918     those events and any already outstanding ones, but will not wait and
919     block your process in case there are no events and will return after one
920     iteration of the loop. This is sometimes useful to poll and handle new
921     events while doing lengthy calculations, to keep the program responsive.
922 root 1.1
923 root 1.310 A flags value of C<EVRUN_ONCE> will look for new events (waiting if
924 root 1.183 necessary) and will handle those and any already outstanding ones. It
925     will block your process until at least one new event arrives (which could
926 root 1.208 be an event internal to libev itself, so there is no guarantee that a
927 root 1.183 user-registered callback will be called), and will return after one
928     iteration of the loop.
929    
930     This is useful if you are waiting for some external event in conjunction
931     with something not expressible using other libev watchers (i.e. "roll your
932 root 1.310 own C<ev_run>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
933 root 1.33 usually a better approach for this kind of thing.
934    
935 root 1.369 Here are the gory details of what C<ev_run> does (this is for your
936     understanding, not a guarantee that things will work exactly like this in
937     future versions):
938 root 1.33
939 root 1.310 - Increment loop depth.
940     - Reset the ev_break status.
941 root 1.77 - Before the first iteration, call any pending watchers.
942 root 1.310 LOOP:
943     - If EVFLAG_FORKCHECK was used, check for a fork.
944 root 1.171 - If a fork was detected (by any means), queue and call all fork watchers.
945 root 1.113 - Queue and call all prepare watchers.
946 root 1.310 - If ev_break was called, goto FINISH.
947 root 1.171 - If we have been forked, detach and recreate the kernel state
948     as to not disturb the other process.
949 root 1.33 - Update the kernel state with all outstanding changes.
950 root 1.171 - Update the "event loop time" (ev_now ()).
951 root 1.113 - Calculate for how long to sleep or block, if at all
952 root 1.310 (active idle watchers, EVRUN_NOWAIT or not having
953 root 1.113 any active watchers at all will result in not sleeping).
954     - Sleep if the I/O and timer collect interval say so.
955 root 1.310 - Increment loop iteration counter.
956 root 1.33 - Block the process, waiting for any events.
957     - Queue all outstanding I/O (fd) events.
958 root 1.171 - Update the "event loop time" (ev_now ()), and do time jump adjustments.
959 root 1.183 - Queue all expired timers.
960     - Queue all expired periodics.
961 root 1.310 - Queue all idle watchers with priority higher than that of pending events.
962 root 1.33 - Queue all check watchers.
963     - Call all queued watchers in reverse order (i.e. check watchers first).
964     Signals and child watchers are implemented as I/O watchers, and will
965     be handled here by queueing them when their watcher gets executed.
966 root 1.310 - If ev_break has been called, or EVRUN_ONCE or EVRUN_NOWAIT
967     were used, or there are no active watchers, goto FINISH, otherwise
968     continue with step LOOP.
969     FINISH:
970     - Reset the ev_break status iff it was EVBREAK_ONE.
971     - Decrement the loop depth.
972     - Return.
973 root 1.27
974 root 1.114 Example: Queue some jobs and then loop until no events are outstanding
975 root 1.34 anymore.
976    
977     ... queue jobs here, make sure they register event watchers as long
978     ... as they still have work to do (even an idle watcher will do..)
979 root 1.310 ev_run (my_loop, 0);
980 root 1.362 ... jobs done or somebody called break. yeah!
981 root 1.34
982 root 1.310 =item ev_break (loop, how)
983 root 1.1
984 root 1.310 Can be used to make a call to C<ev_run> return early (but only after it
985 root 1.9 has processed all outstanding events). The C<how> argument must be either
986 root 1.310 C<EVBREAK_ONE>, which will make the innermost C<ev_run> call return, or
987     C<EVBREAK_ALL>, which will make all nested C<ev_run> calls return.
988 root 1.1
989 root 1.343 This "break state" will be cleared on the next call to C<ev_run>.
990 root 1.115
991 root 1.343 It is safe to call C<ev_break> from outside any C<ev_run> calls, too, in
992     which case it will have no effect.
993 root 1.194
994 root 1.1 =item ev_ref (loop)
995    
996     =item ev_unref (loop)
997    
998 root 1.9 Ref/unref can be used to add or remove a reference count on the event
999     loop: Every watcher keeps one reference, and as long as the reference
1000 root 1.310 count is nonzero, C<ev_run> will not return on its own.
1001 root 1.183
1002 root 1.276 This is useful when you have a watcher that you never intend to
1003 root 1.310 unregister, but that nevertheless should not keep C<ev_run> from
1004 root 1.276 returning. In such a case, call C<ev_unref> after starting, and C<ev_ref>
1005     before stopping it.
1006 root 1.183
1007 root 1.229 As an example, libev itself uses this for its internal signal pipe: It
1008 root 1.310 is not visible to the libev user and should not keep C<ev_run> from
1009 root 1.229 exiting if no event watchers registered by it are active. It is also an
1010     excellent way to do this for generic recurring timers or from within
1011     third-party libraries. Just remember to I<unref after start> and I<ref
1012     before stop> (but only if the watcher wasn't active before, or was active
1013     before, respectively. Note also that libev might stop watchers itself
1014     (e.g. non-repeating timers) in which case you have to C<ev_ref>
1015     in the callback).
1016 root 1.1
1017 root 1.310 Example: Create a signal watcher, but keep it from keeping C<ev_run>
1018 root 1.34 running when nothing else is active.
1019    
1020 root 1.198 ev_signal exitsig;
1021 root 1.164 ev_signal_init (&exitsig, sig_cb, SIGINT);
1022     ev_signal_start (loop, &exitsig);
1023 sf-exg 1.348 ev_unref (loop);
1024 root 1.34
1025 root 1.54 Example: For some weird reason, unregister the above signal handler again.
1026 root 1.34
1027 root 1.164 ev_ref (loop);
1028     ev_signal_stop (loop, &exitsig);
1029 root 1.34
1030 root 1.97 =item ev_set_io_collect_interval (loop, ev_tstamp interval)
1031    
1032     =item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
1033    
1034     These advanced functions influence the time that libev will spend waiting
1035 root 1.171 for events. Both time intervals are by default C<0>, meaning that libev
1036     will try to invoke timer/periodic callbacks and I/O callbacks with minimum
1037     latency.
1038 root 1.97
1039     Setting these to a higher value (the C<interval> I<must> be >= C<0>)
1040 root 1.171 allows libev to delay invocation of I/O and timer/periodic callbacks
1041     to increase efficiency of loop iterations (or to increase power-saving
1042     opportunities).
1043 root 1.97
1044 root 1.183 The idea is that sometimes your program runs just fast enough to handle
1045     one (or very few) event(s) per loop iteration. While this makes the
1046     program responsive, it also wastes a lot of CPU time to poll for new
1047 root 1.97 events, especially with backends like C<select ()> which have a high
1048     overhead for the actual polling but can deliver many events at once.
1049    
1050     By setting a higher I<io collect interval> you allow libev to spend more
1051     time collecting I/O events, so you can handle more events per iteration,
1052     at the cost of increasing latency. Timeouts (both C<ev_periodic> and
1053 root 1.372 C<ev_timer>) will not be affected. Setting this to a non-null value will
1054 root 1.245 introduce an additional C<ev_sleep ()> call into most loop iterations. The
1055     sleep time ensures that libev will not poll for I/O events more often then
1056 root 1.373 once per this interval, on average (as long as the host time resolution is
1057     good enough).
1058 root 1.97
1059     Likewise, by setting a higher I<timeout collect interval> you allow libev
1060     to spend more time collecting timeouts, at the expense of increased
1061 root 1.183 latency/jitter/inexactness (the watcher callback will be called
1062     later). C<ev_io> watchers will not be affected. Setting this to a non-null
1063     value will not introduce any overhead in libev.
1064 root 1.97
1065 root 1.161 Many (busy) programs can usually benefit by setting the I/O collect
1066 root 1.98 interval to a value near C<0.1> or so, which is often enough for
1067     interactive servers (of course not for games), likewise for timeouts. It
1068     usually doesn't make much sense to set it to a lower value than C<0.01>,
1069 root 1.245 as this approaches the timing granularity of most systems. Note that if
1070     you do transactions with the outside world and you can't increase the
1071     parallelity, then this setting will limit your transaction rate (if you
1072     need to poll once per transaction and the I/O collect interval is 0.01,
1073 sf-exg 1.298 then you can't do more than 100 transactions per second).
1074 root 1.97
1075 root 1.171 Setting the I<timeout collect interval> can improve the opportunity for
1076     saving power, as the program will "bundle" timer callback invocations that
1077     are "near" in time together, by delaying some, thus reducing the number of
1078     times the process sleeps and wakes up again. Another useful technique to
1079     reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
1080     they fire on, say, one-second boundaries only.
1081    
1082 root 1.245 Example: we only need 0.1s timeout granularity, and we wish not to poll
1083     more often than 100 times per second:
1084    
1085     ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
1086     ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
1087    
1088 root 1.253 =item ev_invoke_pending (loop)
1089    
1090     This call will simply invoke all pending watchers while resetting their
1091 root 1.310 pending state. Normally, C<ev_run> does this automatically when required,
1092 root 1.313 but when overriding the invoke callback this call comes handy. This
1093     function can be invoked from a watcher - this can be useful for example
1094     when you want to do some lengthy calculation and want to pass further
1095     event handling to another thread (you still have to make sure only one
1096     thread executes within C<ev_invoke_pending> or C<ev_run> of course).
1097 root 1.253
1098 root 1.256 =item int ev_pending_count (loop)
1099    
1100     Returns the number of pending watchers - zero indicates that no watchers
1101     are pending.
1102    
1103 root 1.253 =item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
1104    
1105     This overrides the invoke pending functionality of the loop: Instead of
1106 root 1.310 invoking all pending watchers when there are any, C<ev_run> will call
1107 root 1.253 this callback instead. This is useful, for example, when you want to
1108     invoke the actual watchers inside another context (another thread etc.).
1109    
1110     If you want to reset the callback, use C<ev_invoke_pending> as new
1111     callback.
1112    
1113 root 1.401 =item ev_set_loop_release_cb (loop, void (*release)(EV_P) throw (), void (*acquire)(EV_P) throw ())
1114 root 1.253
1115     Sometimes you want to share the same loop between multiple threads. This
1116     can be done relatively simply by putting mutex_lock/unlock calls around
1117     each call to a libev function.
1118    
1119 root 1.310 However, C<ev_run> can run an indefinite time, so it is not feasible
1120     to wait for it to return. One way around this is to wake up the event
1121 root 1.385 loop via C<ev_break> and C<ev_async_send>, another way is to set these
1122 root 1.310 I<release> and I<acquire> callbacks on the loop.
1123 root 1.253
1124     When set, then C<release> will be called just before the thread is
1125     suspended waiting for new events, and C<acquire> is called just
1126     afterwards.
1127    
1128     Ideally, C<release> will just call your mutex_unlock function, and
1129     C<acquire> will just call the mutex_lock function again.
1130    
1131 root 1.254 While event loop modifications are allowed between invocations of
1132     C<release> and C<acquire> (that's their only purpose after all), no
1133     modifications done will affect the event loop, i.e. adding watchers will
1134     have no effect on the set of file descriptors being watched, or the time
1135 root 1.310 waited. Use an C<ev_async> watcher to wake up C<ev_run> when you want it
1136 root 1.254 to take note of any changes you made.
1137    
1138 root 1.310 In theory, threads executing C<ev_run> will be async-cancel safe between
1139 root 1.254 invocations of C<release> and C<acquire>.
1140    
1141     See also the locking example in the C<THREADS> section later in this
1142     document.
1143    
1144 root 1.253 =item ev_set_userdata (loop, void *data)
1145    
1146 root 1.344 =item void *ev_userdata (loop)
1147 root 1.253
1148     Set and retrieve a single C<void *> associated with a loop. When
1149     C<ev_set_userdata> has never been called, then C<ev_userdata> returns
1150 root 1.342 C<0>.
1151 root 1.253
1152     These two functions can be used to associate arbitrary data with a loop,
1153     and are intended solely for the C<invoke_pending_cb>, C<release> and
1154     C<acquire> callbacks described above, but of course can be (ab-)used for
1155     any other purpose as well.
1156    
1157 root 1.309 =item ev_verify (loop)
1158 root 1.159
1159     This function only does something when C<EV_VERIFY> support has been
1160 root 1.200 compiled in, which is the default for non-minimal builds. It tries to go
1161 root 1.183 through all internal structures and checks them for validity. If anything
1162     is found to be inconsistent, it will print an error message to standard
1163     error and call C<abort ()>.
1164 root 1.159
1165     This can be used to catch bugs inside libev itself: under normal
1166     circumstances, this function will never abort as of course libev keeps its
1167     data structures consistent.
1168    
1169 root 1.1 =back
1170    
1171 root 1.42
1172 root 1.1 =head1 ANATOMY OF A WATCHER
1173    
1174 root 1.200 In the following description, uppercase C<TYPE> in names stands for the
1175     watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
1176     watchers and C<ev_io_start> for I/O watchers.
1177    
1178 root 1.311 A watcher is an opaque structure that you allocate and register to record
1179     your interest in some event. To make a concrete example, imagine you want
1180     to wait for STDIN to become readable, you would create an C<ev_io> watcher
1181     for that:
1182 root 1.1
1183 root 1.198 static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
1184 root 1.164 {
1185     ev_io_stop (w);
1186 root 1.310 ev_break (loop, EVBREAK_ALL);
1187 root 1.164 }
1188    
1189     struct ev_loop *loop = ev_default_loop (0);
1190 root 1.200
1191 root 1.198 ev_io stdin_watcher;
1192 root 1.200
1193 root 1.164 ev_init (&stdin_watcher, my_cb);
1194     ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
1195     ev_io_start (loop, &stdin_watcher);
1196 root 1.200
1197 root 1.310 ev_run (loop, 0);
1198 root 1.1
1199     As you can see, you are responsible for allocating the memory for your
1200 root 1.200 watcher structures (and it is I<usually> a bad idea to do this on the
1201     stack).
1202    
1203     Each watcher has an associated watcher structure (called C<struct ev_TYPE>
1204     or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
1205 root 1.1
1206 root 1.311 Each watcher structure must be initialised by a call to C<ev_init (watcher
1207     *, callback)>, which expects a callback to be provided. This callback is
1208     invoked each time the event occurs (or, in the case of I/O watchers, each
1209     time the event loop detects that the file descriptor given is readable
1210     and/or writable).
1211 root 1.1
1212 root 1.200 Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
1213     macro to configure it, with arguments specific to the watcher type. There
1214     is also a macro to combine initialisation and setting in one call: C<<
1215     ev_TYPE_init (watcher *, callback, ...) >>.
1216 root 1.1
1217     To make the watcher actually watch out for events, you have to start it
1218 root 1.200 with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
1219 root 1.1 *) >>), and you can stop watching for events at any time by calling the
1220 root 1.200 corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
1221 root 1.1
1222     As long as your watcher is active (has been started but not stopped) you
1223 root 1.460 must not touch the values stored in it except when explicitly documented
1224     otherwise. Most specifically you must never reinitialise it or call its
1225     C<ev_TYPE_set> macro.
1226 root 1.1
1227     Each and every callback receives the event loop pointer as first, the
1228     registered watcher structure as second, and a bitset of received events as
1229     third argument.
1230    
1231 root 1.14 The received events usually include a single bit per event type received
1232 root 1.1 (you can receive multiple events at the same time). The possible bit masks
1233     are:
1234    
1235     =over 4
1236    
1237 root 1.10 =item C<EV_READ>
1238 root 1.1
1239 root 1.10 =item C<EV_WRITE>
1240 root 1.1
1241 root 1.10 The file descriptor in the C<ev_io> watcher has become readable and/or
1242 root 1.1 writable.
1243    
1244 root 1.289 =item C<EV_TIMER>
1245 root 1.1
1246 root 1.10 The C<ev_timer> watcher has timed out.
1247 root 1.1
1248 root 1.10 =item C<EV_PERIODIC>
1249 root 1.1
1250 root 1.10 The C<ev_periodic> watcher has timed out.
1251 root 1.1
1252 root 1.10 =item C<EV_SIGNAL>
1253 root 1.1
1254 root 1.10 The signal specified in the C<ev_signal> watcher has been received by a thread.
1255 root 1.1
1256 root 1.10 =item C<EV_CHILD>
1257 root 1.1
1258 root 1.10 The pid specified in the C<ev_child> watcher has received a status change.
1259 root 1.1
1260 root 1.48 =item C<EV_STAT>
1261    
1262     The path specified in the C<ev_stat> watcher changed its attributes somehow.
1263    
1264 root 1.10 =item C<EV_IDLE>
1265 root 1.1
1266 root 1.10 The C<ev_idle> watcher has determined that you have nothing better to do.
1267 root 1.1
1268 root 1.10 =item C<EV_PREPARE>
1269 root 1.1
1270 root 1.10 =item C<EV_CHECK>
1271 root 1.1
1272 root 1.405 All C<ev_prepare> watchers are invoked just I<before> C<ev_run> starts to
1273     gather new events, and all C<ev_check> watchers are queued (not invoked)
1274     just after C<ev_run> has gathered them, but before it queues any callbacks
1275     for any received events. That means C<ev_prepare> watchers are the last
1276     watchers invoked before the event loop sleeps or polls for new events, and
1277     C<ev_check> watchers will be invoked before any other watchers of the same
1278     or lower priority within an event loop iteration.
1279    
1280     Callbacks of both watcher types can start and stop as many watchers as
1281     they want, and all of them will be taken into account (for example, a
1282     C<ev_prepare> watcher might start an idle watcher to keep C<ev_run> from
1283     blocking).
1284 root 1.1
1285 root 1.50 =item C<EV_EMBED>
1286    
1287     The embedded event loop specified in the C<ev_embed> watcher needs attention.
1288    
1289     =item C<EV_FORK>
1290    
1291     The event loop has been resumed in the child process after fork (see
1292     C<ev_fork>).
1293    
1294 root 1.324 =item C<EV_CLEANUP>
1295    
1296 sf-exg 1.330 The event loop is about to be destroyed (see C<ev_cleanup>).
1297 root 1.324
1298 root 1.122 =item C<EV_ASYNC>
1299    
1300     The given async watcher has been asynchronously notified (see C<ev_async>).
1301    
1302 root 1.229 =item C<EV_CUSTOM>
1303    
1304     Not ever sent (or otherwise used) by libev itself, but can be freely used
1305     by libev users to signal watchers (e.g. via C<ev_feed_event>).
1306    
1307 root 1.10 =item C<EV_ERROR>
1308 root 1.1
1309 root 1.161 An unspecified error has occurred, the watcher has been stopped. This might
1310 root 1.1 happen because the watcher could not be properly started because libev
1311     ran out of memory, a file descriptor was found to be closed or any other
1312 root 1.197 problem. Libev considers these application bugs.
1313    
1314     You best act on it by reporting the problem and somehow coping with the
1315     watcher being stopped. Note that well-written programs should not receive
1316     an error ever, so when your watcher receives it, this usually indicates a
1317     bug in your program.
1318 root 1.1
1319 root 1.183 Libev will usually signal a few "dummy" events together with an error, for
1320     example it might indicate that a fd is readable or writable, and if your
1321     callbacks is well-written it can just attempt the operation and cope with
1322     the error from read() or write(). This will not work in multi-threaded
1323     programs, though, as the fd could already be closed and reused for another
1324     thing, so beware.
1325 root 1.1
1326     =back
1327    
1328 root 1.42 =head2 GENERIC WATCHER FUNCTIONS
1329 root 1.36
1330     =over 4
1331    
1332     =item C<ev_init> (ev_TYPE *watcher, callback)
1333    
1334     This macro initialises the generic portion of a watcher. The contents
1335     of the watcher object can be arbitrary (so C<malloc> will do). Only
1336     the generic parts of the watcher are initialised, you I<need> to call
1337     the type-specific C<ev_TYPE_set> macro afterwards to initialise the
1338     type-specific parts. For each type there is also a C<ev_TYPE_init> macro
1339     which rolls both calls into one.
1340    
1341     You can reinitialise a watcher at any time as long as it has been stopped
1342     (or never started) and there are no pending events outstanding.
1343    
1344 root 1.198 The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
1345 root 1.36 int revents)>.
1346    
1347 root 1.183 Example: Initialise an C<ev_io> watcher in two steps.
1348    
1349     ev_io w;
1350     ev_init (&w, my_cb);
1351     ev_io_set (&w, STDIN_FILENO, EV_READ);
1352    
1353 root 1.274 =item C<ev_TYPE_set> (ev_TYPE *watcher, [args])
1354 root 1.36
1355     This macro initialises the type-specific parts of a watcher. You need to
1356     call C<ev_init> at least once before you call this macro, but you can
1357     call C<ev_TYPE_set> any number of times. You must not, however, call this
1358     macro on a watcher that is active (it can be pending, however, which is a
1359     difference to the C<ev_init> macro).
1360    
1361     Although some watcher types do not have type-specific arguments
1362     (e.g. C<ev_prepare>) you still need to call its C<set> macro.
1363    
1364 root 1.183 See C<ev_init>, above, for an example.
1365    
1366 root 1.36 =item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
1367    
1368 root 1.161 This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
1369     calls into a single call. This is the most convenient method to initialise
1370 root 1.36 a watcher. The same limitations apply, of course.
1371    
1372 root 1.183 Example: Initialise and set an C<ev_io> watcher in one step.
1373    
1374     ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
1375    
1376 root 1.274 =item C<ev_TYPE_start> (loop, ev_TYPE *watcher)
1377 root 1.36
1378     Starts (activates) the given watcher. Only active watchers will receive
1379     events. If the watcher is already active nothing will happen.
1380    
1381 root 1.183 Example: Start the C<ev_io> watcher that is being abused as example in this
1382     whole section.
1383    
1384     ev_io_start (EV_DEFAULT_UC, &w);
1385    
1386 root 1.274 =item C<ev_TYPE_stop> (loop, ev_TYPE *watcher)
1387 root 1.36
1388 root 1.195 Stops the given watcher if active, and clears the pending status (whether
1389     the watcher was active or not).
1390    
1391     It is possible that stopped watchers are pending - for example,
1392     non-repeating timers are being stopped when they become pending - but
1393     calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
1394     pending. If you want to free or reuse the memory used by the watcher it is
1395     therefore a good idea to always call its C<ev_TYPE_stop> function.
1396 root 1.36
1397     =item bool ev_is_active (ev_TYPE *watcher)
1398    
1399     Returns a true value iff the watcher is active (i.e. it has been started
1400     and not yet been stopped). As long as a watcher is active you must not modify
1401     it.
1402    
1403     =item bool ev_is_pending (ev_TYPE *watcher)
1404    
1405     Returns a true value iff the watcher is pending, (i.e. it has outstanding
1406     events but its callback has not yet been invoked). As long as a watcher
1407     is pending (but not active) you must not call an init function on it (but
1408 root 1.73 C<ev_TYPE_set> is safe), you must not change its priority, and you must
1409     make sure the watcher is available to libev (e.g. you cannot C<free ()>
1410     it).
1411 root 1.36
1412 root 1.55 =item callback ev_cb (ev_TYPE *watcher)
1413 root 1.36
1414     Returns the callback currently set on the watcher.
1415    
1416 root 1.408 =item ev_set_cb (ev_TYPE *watcher, callback)
1417 root 1.36
1418     Change the callback. You can change the callback at virtually any time
1419     (modulo threads).
1420    
1421 root 1.274 =item ev_set_priority (ev_TYPE *watcher, int priority)
1422 root 1.67
1423     =item int ev_priority (ev_TYPE *watcher)
1424    
1425     Set and query the priority of the watcher. The priority is a small
1426     integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
1427     (default: C<-2>). Pending watchers with higher priority will be invoked
1428     before watchers with lower priority, but priority will not keep watchers
1429     from being executed (except for C<ev_idle> watchers).
1430    
1431     If you need to suppress invocation when higher priority events are pending
1432     you need to look at C<ev_idle> watchers, which provide this functionality.
1433    
1434 root 1.73 You I<must not> change the priority of a watcher as long as it is active or
1435     pending.
1436    
1437 root 1.233 Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
1438     fine, as long as you do not mind that the priority value you query might
1439     or might not have been clamped to the valid range.
1440    
1441 root 1.67 The default priority used by watchers when no priority has been set is
1442     always C<0>, which is supposed to not be too high and not be too low :).
1443    
1444 root 1.413 See L</WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
1445 root 1.233 priorities.
1446 root 1.67
1447 root 1.74 =item ev_invoke (loop, ev_TYPE *watcher, int revents)
1448    
1449     Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
1450     C<loop> nor C<revents> need to be valid as long as the watcher callback
1451 root 1.183 can deal with that fact, as both are simply passed through to the
1452     callback.
1453 root 1.74
1454     =item int ev_clear_pending (loop, ev_TYPE *watcher)
1455    
1456 root 1.183 If the watcher is pending, this function clears its pending status and
1457     returns its C<revents> bitset (as if its callback was invoked). If the
1458 root 1.74 watcher isn't pending it does nothing and returns C<0>.
1459    
1460 root 1.183 Sometimes it can be useful to "poll" a watcher instead of waiting for its
1461     callback to be invoked, which can be accomplished with this function.
1462    
1463 root 1.274 =item ev_feed_event (loop, ev_TYPE *watcher, int revents)
1464 root 1.273
1465     Feeds the given event set into the event loop, as if the specified event
1466     had happened for the specified watcher (which must be a pointer to an
1467     initialised but not necessarily started event watcher). Obviously you must
1468     not free the watcher as long as it has pending events.
1469    
1470     Stopping the watcher, letting libev invoke it, or calling
1471     C<ev_clear_pending> will clear the pending event, even if the watcher was
1472     not started in the first place.
1473    
1474     See also C<ev_feed_fd_event> and C<ev_feed_signal_event> for related
1475     functions that do not need a watcher.
1476    
1477 root 1.36 =back
1478    
1479 root 1.413 See also the L</ASSOCIATING CUSTOM DATA WITH A WATCHER> and L</BUILDING YOUR
1480 root 1.357 OWN COMPOSITE WATCHERS> idioms.
1481 root 1.1
1482 root 1.335 =head2 WATCHER STATES
1483    
1484     There are various watcher states mentioned throughout this manual -
1485     active, pending and so on. In this section these states and the rules to
1486     transition between them will be described in more detail - and while these
1487     rules might look complicated, they usually do "the right thing".
1488    
1489     =over 4
1490    
1491 root 1.422 =item initialised
1492 root 1.335
1493 sf-exg 1.374 Before a watcher can be registered with the event loop it has to be
1494 root 1.335 initialised. This can be done with a call to C<ev_TYPE_init>, or calls to
1495     C<ev_init> followed by the watcher-specific C<ev_TYPE_set> function.
1496    
1497 root 1.361 In this state it is simply some block of memory that is suitable for
1498     use in an event loop. It can be moved around, freed, reused etc. at
1499     will - as long as you either keep the memory contents intact, or call
1500     C<ev_TYPE_init> again.
1501 root 1.335
1502     =item started/running/active
1503    
1504     Once a watcher has been started with a call to C<ev_TYPE_start> it becomes
1505     property of the event loop, and is actively waiting for events. While in
1506     this state it cannot be accessed (except in a few documented ways), moved,
1507     freed or anything else - the only legal thing is to keep a pointer to it,
1508     and call libev functions on it that are documented to work on active watchers.
1509    
1510     =item pending
1511    
1512     If a watcher is active and libev determines that an event it is interested
1513     in has occurred (such as a timer expiring), it will become pending. It will
1514     stay in this pending state until either it is stopped or its callback is
1515     about to be invoked, so it is not normally pending inside the watcher
1516     callback.
1517    
1518     The watcher might or might not be active while it is pending (for example,
1519     an expired non-repeating timer can be pending but no longer active). If it
1520     is stopped, it can be freely accessed (e.g. by calling C<ev_TYPE_set>),
1521     but it is still property of the event loop at this time, so cannot be
1522     moved, freed or reused. And if it is active the rules described in the
1523     previous item still apply.
1524    
1525     It is also possible to feed an event on a watcher that is not active (e.g.
1526     via C<ev_feed_event>), in which case it becomes pending without being
1527     active.
1528    
1529     =item stopped
1530    
1531     A watcher can be stopped implicitly by libev (in which case it might still
1532     be pending), or explicitly by calling its C<ev_TYPE_stop> function. The
1533     latter will clear any pending state the watcher might be in, regardless
1534     of whether it was active or not, so stopping a watcher explicitly before
1535     freeing it is often a good idea.
1536    
1537     While stopped (and not pending) the watcher is essentially in the
1538 root 1.361 initialised state, that is, it can be reused, moved, modified in any way
1539     you wish (but when you trash the memory block, you need to C<ev_TYPE_init>
1540     it again).
1541 root 1.335
1542     =back
1543    
1544 root 1.233 =head2 WATCHER PRIORITY MODELS
1545    
1546     Many event loops support I<watcher priorities>, which are usually small
1547     integers that influence the ordering of event callback invocation
1548     between watchers in some way, all else being equal.
1549    
1550 root 1.457 In libev, watcher priorities can be set using C<ev_set_priority>. See its
1551 root 1.233 description for the more technical details such as the actual priority
1552     range.
1553    
1554     There are two common ways how these these priorities are being interpreted
1555     by event loops:
1556    
1557     In the more common lock-out model, higher priorities "lock out" invocation
1558     of lower priority watchers, which means as long as higher priority
1559     watchers receive events, lower priority watchers are not being invoked.
1560    
1561     The less common only-for-ordering model uses priorities solely to order
1562     callback invocation within a single event loop iteration: Higher priority
1563     watchers are invoked before lower priority ones, but they all get invoked
1564     before polling for new events.
1565    
1566     Libev uses the second (only-for-ordering) model for all its watchers
1567     except for idle watchers (which use the lock-out model).
1568    
1569     The rationale behind this is that implementing the lock-out model for
1570     watchers is not well supported by most kernel interfaces, and most event
1571     libraries will just poll for the same events again and again as long as
1572     their callbacks have not been executed, which is very inefficient in the
1573     common case of one high-priority watcher locking out a mass of lower
1574     priority ones.
1575    
1576     Static (ordering) priorities are most useful when you have two or more
1577     watchers handling the same resource: a typical usage example is having an
1578     C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
1579     timeouts. Under load, data might be received while the program handles
1580     other jobs, but since timers normally get invoked first, the timeout
1581     handler will be executed before checking for data. In that case, giving
1582     the timer a lower priority than the I/O watcher ensures that I/O will be
1583     handled first even under adverse conditions (which is usually, but not
1584     always, what you want).
1585    
1586     Since idle watchers use the "lock-out" model, meaning that idle watchers
1587     will only be executed when no same or higher priority watchers have
1588     received events, they can be used to implement the "lock-out" model when
1589     required.
1590    
1591     For example, to emulate how many other event libraries handle priorities,
1592     you can associate an C<ev_idle> watcher to each such watcher, and in
1593     the normal watcher callback, you just start the idle watcher. The real
1594     processing is done in the idle watcher callback. This causes libev to
1595 sf-exg 1.298 continuously poll and process kernel event data for the watcher, but when
1596 root 1.233 the lock-out case is known to be rare (which in turn is rare :), this is
1597     workable.
1598    
1599     Usually, however, the lock-out model implemented that way will perform
1600     miserably under the type of load it was designed to handle. In that case,
1601     it might be preferable to stop the real watcher before starting the
1602     idle watcher, so the kernel will not have to process the event in case
1603     the actual processing will be delayed for considerable time.
1604    
1605     Here is an example of an I/O watcher that should run at a strictly lower
1606     priority than the default, and which should only process data when no
1607     other events are pending:
1608    
1609     ev_idle idle; // actual processing watcher
1610     ev_io io; // actual event watcher
1611    
1612     static void
1613     io_cb (EV_P_ ev_io *w, int revents)
1614     {
1615     // stop the I/O watcher, we received the event, but
1616     // are not yet ready to handle it.
1617     ev_io_stop (EV_A_ w);
1618    
1619 root 1.296 // start the idle watcher to handle the actual event.
1620 root 1.233 // it will not be executed as long as other watchers
1621     // with the default priority are receiving events.
1622     ev_idle_start (EV_A_ &idle);
1623     }
1624    
1625     static void
1626 root 1.242 idle_cb (EV_P_ ev_idle *w, int revents)
1627 root 1.233 {
1628     // actual processing
1629     read (STDIN_FILENO, ...);
1630    
1631     // have to start the I/O watcher again, as
1632     // we have handled the event
1633     ev_io_start (EV_P_ &io);
1634     }
1635    
1636     // initialisation
1637     ev_idle_init (&idle, idle_cb);
1638     ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
1639     ev_io_start (EV_DEFAULT_ &io);
1640    
1641     In the "real" world, it might also be beneficial to start a timer, so that
1642     low-priority connections can not be locked out forever under load. This
1643     enables your program to keep a lower latency for important connections
1644     during short periods of high load, while not completely locking out less
1645     important ones.
1646    
1647 root 1.1
1648     =head1 WATCHER TYPES
1649    
1650     This section describes each watcher in detail, but will not repeat
1651 root 1.48 information given in the last section. Any initialisation/set macros,
1652     functions and members specific to the watcher type are explained.
1653    
1654 root 1.459 Most members are additionally marked with either I<[read-only]>, meaning
1655     that, while the watcher is active, you can look at the member and expect
1656     some sensible content, but you must not modify it (you can modify it while
1657     the watcher is stopped to your hearts content), or I<[read-write]>, which
1658 root 1.48 means you can expect it to have some sensible content while the watcher
1659     is active, but you can also modify it. Modifying it may not do something
1660     sensible or take immediate effect (or do anything at all), but libev will
1661     not crash or malfunction in any way.
1662 root 1.1
1663 root 1.459 In any case, the documentation for each member will explain what the
1664     effects are, and if there are any additional access restrictions.
1665 root 1.34
1666 root 1.42 =head2 C<ev_io> - is this file descriptor readable or writable?
1667 root 1.1
1668 root 1.4 I/O watchers check whether a file descriptor is readable or writable
1669 root 1.42 in each iteration of the event loop, or, more precisely, when reading
1670     would not block the process and writing would at least be able to write
1671     some data. This behaviour is called level-triggering because you keep
1672     receiving events as long as the condition persists. Remember you can stop
1673     the watcher if you don't want to act on the event and neither want to
1674     receive future events.
1675 root 1.1
1676 root 1.23 In general you can register as many read and/or write event watchers per
1677 root 1.8 fd as you want (as long as you don't confuse yourself). Setting all file
1678     descriptors to non-blocking mode is also usually a good idea (but not
1679     required if you know what you are doing).
1680    
1681 root 1.42 Another thing you have to watch out for is that it is quite easy to
1682 root 1.354 receive "spurious" readiness notifications, that is, your callback might
1683 root 1.42 be called with C<EV_READ> but a subsequent C<read>(2) will actually block
1684 root 1.354 because there is no data. It is very easy to get into this situation even
1685     with a relatively standard program structure. Thus it is best to always
1686     use non-blocking I/O: An extra C<read>(2) returning C<EAGAIN> is far
1687     preferable to a program hanging until some data arrives.
1688 root 1.42
1689 root 1.183 If you cannot run the fd in non-blocking mode (for example you should
1690     not play around with an Xlib connection), then you have to separately
1691     re-test whether a file descriptor is really ready with a known-to-be good
1692 root 1.354 interface such as poll (fortunately in the case of Xlib, it already does
1693     this on its own, so its quite safe to use). Some people additionally
1694 root 1.183 use C<SIGALRM> and an interval timer, just to be sure you won't block
1695     indefinitely.
1696    
1697     But really, best use non-blocking mode.
1698 root 1.42
1699 root 1.81 =head3 The special problem of disappearing file descriptors
1700    
1701 root 1.447 Some backends (e.g. kqueue, epoll, linuxaio) need to be told about closing
1702     a file descriptor (either due to calling C<close> explicitly or any other
1703     means, such as C<dup2>). The reason is that you register interest in some
1704     file descriptor, but when it goes away, the operating system will silently
1705     drop this interest. If another file descriptor with the same number then
1706     is registered with libev, there is no efficient way to see that this is,
1707     in fact, a different file descriptor.
1708 root 1.81
1709     To avoid having to explicitly tell libev about such cases, libev follows
1710     the following policy: Each time C<ev_io_set> is being called, libev
1711     will assume that this is potentially a new file descriptor, otherwise
1712     it is assumed that the file descriptor stays the same. That means that
1713     you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
1714     descriptor even if the file descriptor number itself did not change.
1715    
1716     This is how one would do it normally anyway, the important point is that
1717     the libev application should not optimise around libev but should leave
1718     optimisations to libev.
1719    
1720 root 1.95 =head3 The special problem of dup'ed file descriptors
1721 root 1.94
1722     Some backends (e.g. epoll), cannot register events for file descriptors,
1723 root 1.103 but only events for the underlying file descriptions. That means when you
1724 root 1.109 have C<dup ()>'ed file descriptors or weirder constellations, and register
1725     events for them, only one file descriptor might actually receive events.
1726 root 1.94
1727 root 1.103 There is no workaround possible except not registering events
1728     for potentially C<dup ()>'ed file descriptors, or to resort to
1729 root 1.94 C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
1730    
1731 root 1.354 =head3 The special problem of files
1732    
1733     Many people try to use C<select> (or libev) on file descriptors
1734     representing files, and expect it to become ready when their program
1735     doesn't block on disk accesses (which can take a long time on their own).
1736    
1737     However, this cannot ever work in the "expected" way - you get a readiness
1738     notification as soon as the kernel knows whether and how much data is
1739     there, and in the case of open files, that's always the case, so you
1740     always get a readiness notification instantly, and your read (or possibly
1741     write) will still block on the disk I/O.
1742    
1743     Another way to view it is that in the case of sockets, pipes, character
1744     devices and so on, there is another party (the sender) that delivers data
1745 sf-exg 1.358 on its own, but in the case of files, there is no such thing: the disk
1746     will not send data on its own, simply because it doesn't know what you
1747 root 1.354 wish to read - you would first have to request some data.
1748    
1749     Since files are typically not-so-well supported by advanced notification
1750     mechanism, libev tries hard to emulate POSIX behaviour with respect
1751     to files, even though you should not use it. The reason for this is
1752     convenience: sometimes you want to watch STDIN or STDOUT, which is
1753     usually a tty, often a pipe, but also sometimes files or special devices
1754     (for example, C<epoll> on Linux works with F</dev/random> but not with
1755     F</dev/urandom>), and even though the file might better be served with
1756     asynchronous I/O instead of with non-blocking I/O, it is still useful when
1757     it "just works" instead of freezing.
1758    
1759     So avoid file descriptors pointing to files when you know it (e.g. use
1760     libeio), but use them when it is convenient, e.g. for STDIN/STDOUT, or
1761     when you rarely read from a file instead of from a socket, and want to
1762     reuse the same code path.
1763    
1764 root 1.94 =head3 The special problem of fork
1765    
1766 root 1.456 Some backends (epoll, kqueue, linuxaio, iouring) do not support C<fork ()>
1767 root 1.447 at all or exhibit useless behaviour. Libev fully supports fork, but needs
1768     to be told about it in the child if you want to continue to use it in the
1769     child.
1770 root 1.94
1771 root 1.354 To support fork in your child processes, you have to call C<ev_loop_fork
1772     ()> after a fork in the child, enable C<EVFLAG_FORKCHECK>, or resort to
1773     C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
1774 root 1.94
1775 root 1.138 =head3 The special problem of SIGPIPE
1776    
1777 root 1.183 While not really specific to libev, it is easy to forget about C<SIGPIPE>:
1778 root 1.174 when writing to a pipe whose other end has been closed, your program gets
1779 root 1.183 sent a SIGPIPE, which, by default, aborts your program. For most programs
1780 root 1.174 this is sensible behaviour, for daemons, this is usually undesirable.
1781 root 1.138
1782     So when you encounter spurious, unexplained daemon exits, make sure you
1783     ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
1784     somewhere, as that would have given you a big clue).
1785    
1786 root 1.284 =head3 The special problem of accept()ing when you can't
1787    
1788     Many implementations of the POSIX C<accept> function (for example,
1789 sf-exg 1.292 found in post-2004 Linux) have the peculiar behaviour of not removing a
1790 root 1.284 connection from the pending queue in all error cases.
1791    
1792     For example, larger servers often run out of file descriptors (because
1793     of resource limits), causing C<accept> to fail with C<ENFILE> but not
1794     rejecting the connection, leading to libev signalling readiness on
1795     the next iteration again (the connection still exists after all), and
1796     typically causing the program to loop at 100% CPU usage.
1797    
1798     Unfortunately, the set of errors that cause this issue differs between
1799     operating systems, there is usually little the app can do to remedy the
1800     situation, and no known thread-safe method of removing the connection to
1801     cope with overload is known (to me).
1802    
1803     One of the easiest ways to handle this situation is to just ignore it
1804     - when the program encounters an overload, it will just loop until the
1805     situation is over. While this is a form of busy waiting, no OS offers an
1806     event-based way to handle this situation, so it's the best one can do.
1807    
1808     A better way to handle the situation is to log any errors other than
1809     C<EAGAIN> and C<EWOULDBLOCK>, making sure not to flood the log with such
1810     messages, and continue as usual, which at least gives the user an idea of
1811     what could be wrong ("raise the ulimit!"). For extra points one could stop
1812     the C<ev_io> watcher on the listening fd "for a while", which reduces CPU
1813     usage.
1814    
1815     If your program is single-threaded, then you could also keep a dummy file
1816     descriptor for overload situations (e.g. by opening F</dev/null>), and
1817     when you run into C<ENFILE> or C<EMFILE>, close it, run C<accept>,
1818     close that fd, and create a new dummy fd. This will gracefully refuse
1819     clients under typical overload conditions.
1820    
1821     The last way to handle it is to simply log the error and C<exit>, as
1822     is often done with C<malloc> failures, but this results in an easy
1823     opportunity for a DoS attack.
1824 root 1.81
1825 root 1.82 =head3 Watcher-Specific Functions
1826    
1827 root 1.1 =over 4
1828    
1829     =item ev_io_init (ev_io *, callback, int fd, int events)
1830    
1831     =item ev_io_set (ev_io *, int fd, int events)
1832    
1833 root 1.42 Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
1834 root 1.183 receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
1835     C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
1836 root 1.32
1837 root 1.459 =item ev_io_modify (ev_io *, int events)
1838 root 1.48
1839 root 1.459 Similar to C<ev_io_set>, but only changes the event mask. Using this might
1840     be faster with some backends, as libev can assume that the C<fd> still
1841     refers to the same underlying file description, something it cannot do
1842     when using C<ev_io_set>.
1843 root 1.48
1844 root 1.459 =item int fd [no-modify]
1845 root 1.48
1846 root 1.459 The file descriptor being watched. While it can be read at any time, you
1847     must not modify this member even when the watcher is stopped - always use
1848     C<ev_io_set> for that.
1849    
1850     =item int events [no-modify]
1851    
1852 root 1.460 The set of events the fd is being watched for, among other flags. Remember
1853     that this is a bit set - to test for C<EV_READ>, use C<< w->events &
1854     EV_READ >>, and similarly for C<EV_WRITE>.
1855 root 1.459
1856     As with C<fd>, you must not modify this member even when the watcher is
1857     stopped, always use C<ev_io_set> or C<ev_io_modify> for that.
1858 root 1.48
1859 root 1.1 =back
1860    
1861 root 1.111 =head3 Examples
1862    
1863 root 1.54 Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
1864 root 1.34 readable, but only once. Since it is likely line-buffered, you could
1865 root 1.54 attempt to read a whole line in the callback.
1866 root 1.34
1867 root 1.164 static void
1868 root 1.198 stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
1869 root 1.164 {
1870     ev_io_stop (loop, w);
1871 root 1.183 .. read from stdin here (or from w->fd) and handle any I/O errors
1872 root 1.164 }
1873    
1874     ...
1875     struct ev_loop *loop = ev_default_init (0);
1876 root 1.198 ev_io stdin_readable;
1877 root 1.164 ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
1878     ev_io_start (loop, &stdin_readable);
1879 root 1.310 ev_run (loop, 0);
1880 root 1.34
1881    
1882 root 1.42 =head2 C<ev_timer> - relative and optionally repeating timeouts
1883 root 1.1
1884     Timer watchers are simple relative timers that generate an event after a
1885     given time, and optionally repeating in regular intervals after that.
1886    
1887     The timers are based on real time, that is, if you register an event that
1888 root 1.161 times out after an hour and you reset your system clock to January last
1889 root 1.183 year, it will still time out after (roughly) one hour. "Roughly" because
1890 root 1.28 detecting time jumps is hard, and some inaccuracies are unavoidable (the
1891 root 1.1 monotonic clock option helps a lot here).
1892    
1893 root 1.183 The callback is guaranteed to be invoked only I<after> its timeout has
1894 root 1.240 passed (not I<at>, so on systems with very low-resolution clocks this
1895 root 1.381 might introduce a small delay, see "the special problem of being too
1896     early", below). If multiple timers become ready during the same loop
1897     iteration then the ones with earlier time-out values are invoked before
1898     ones of the same priority with later time-out values (but this is no
1899     longer true when a callback calls C<ev_run> recursively).
1900 root 1.175
1901 root 1.198 =head3 Be smart about timeouts
1902    
1903 root 1.199 Many real-world problems involve some kind of timeout, usually for error
1904 root 1.198 recovery. A typical example is an HTTP request - if the other side hangs,
1905     you want to raise some error after a while.
1906    
1907 root 1.199 What follows are some ways to handle this problem, from obvious and
1908     inefficient to smart and efficient.
1909 root 1.198
1910 root 1.199 In the following, a 60 second activity timeout is assumed - a timeout that
1911     gets reset to 60 seconds each time there is activity (e.g. each time some
1912     data or other life sign was received).
1913 root 1.198
1914     =over 4
1915    
1916 root 1.199 =item 1. Use a timer and stop, reinitialise and start it on activity.
1917 root 1.198
1918     This is the most obvious, but not the most simple way: In the beginning,
1919     start the watcher:
1920    
1921     ev_timer_init (timer, callback, 60., 0.);
1922     ev_timer_start (loop, timer);
1923    
1924 root 1.199 Then, each time there is some activity, C<ev_timer_stop> it, initialise it
1925     and start it again:
1926 root 1.198
1927     ev_timer_stop (loop, timer);
1928     ev_timer_set (timer, 60., 0.);
1929     ev_timer_start (loop, timer);
1930    
1931 root 1.199 This is relatively simple to implement, but means that each time there is
1932     some activity, libev will first have to remove the timer from its internal
1933     data structure and then add it again. Libev tries to be fast, but it's
1934     still not a constant-time operation.
1935 root 1.198
1936     =item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
1937    
1938     This is the easiest way, and involves using C<ev_timer_again> instead of
1939     C<ev_timer_start>.
1940    
1941 root 1.199 To implement this, configure an C<ev_timer> with a C<repeat> value
1942     of C<60> and then call C<ev_timer_again> at start and each time you
1943     successfully read or write some data. If you go into an idle state where
1944     you do not expect data to travel on the socket, you can C<ev_timer_stop>
1945     the timer, and C<ev_timer_again> will automatically restart it if need be.
1946    
1947     That means you can ignore both the C<ev_timer_start> function and the
1948     C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
1949     member and C<ev_timer_again>.
1950 root 1.198
1951     At start:
1952    
1953 root 1.243 ev_init (timer, callback);
1954 root 1.199 timer->repeat = 60.;
1955 root 1.198 ev_timer_again (loop, timer);
1956    
1957 root 1.199 Each time there is some activity:
1958 root 1.198
1959     ev_timer_again (loop, timer);
1960    
1961 root 1.199 It is even possible to change the time-out on the fly, regardless of
1962     whether the watcher is active or not:
1963 root 1.198
1964     timer->repeat = 30.;
1965     ev_timer_again (loop, timer);
1966    
1967     This is slightly more efficient then stopping/starting the timer each time
1968     you want to modify its timeout value, as libev does not have to completely
1969 root 1.199 remove and re-insert the timer from/into its internal data structure.
1970    
1971     It is, however, even simpler than the "obvious" way to do it.
1972 root 1.198
1973     =item 3. Let the timer time out, but then re-arm it as required.
1974    
1975     This method is more tricky, but usually most efficient: Most timeouts are
1976 root 1.199 relatively long compared to the intervals between other activity - in
1977     our example, within 60 seconds, there are usually many I/O events with
1978     associated activity resets.
1979 root 1.198
1980     In this case, it would be more efficient to leave the C<ev_timer> alone,
1981     but remember the time of last activity, and check for a real timeout only
1982     within the callback:
1983    
1984 root 1.387 ev_tstamp timeout = 60.;
1985 root 1.198 ev_tstamp last_activity; // time of last activity
1986 root 1.387 ev_timer timer;
1987 root 1.198
1988     static void
1989     callback (EV_P_ ev_timer *w, int revents)
1990     {
1991 root 1.387 // calculate when the timeout would happen
1992     ev_tstamp after = last_activity - ev_now (EV_A) + timeout;
1993 root 1.198
1994 sf-exg 1.403 // if negative, it means we the timeout already occurred
1995 root 1.387 if (after < 0.)
1996 root 1.198 {
1997 sf-exg 1.298 // timeout occurred, take action
1998 root 1.198 }
1999     else
2000     {
2001 root 1.387 // callback was invoked, but there was some recent
2002 root 1.392 // activity. simply restart the timer to time out
2003 root 1.387 // after "after" seconds, which is the earliest time
2004     // the timeout can occur.
2005     ev_timer_set (w, after, 0.);
2006     ev_timer_start (EV_A_ w);
2007 root 1.198 }
2008     }
2009    
2010 root 1.387 To summarise the callback: first calculate in how many seconds the
2011     timeout will occur (by calculating the absolute time when it would occur,
2012     C<last_activity + timeout>, and subtracting the current time, C<ev_now
2013     (EV_A)> from that).
2014    
2015     If this value is negative, then we are already past the timeout, i.e. we
2016     timed out, and need to do whatever is needed in this case.
2017    
2018     Otherwise, we now the earliest time at which the timeout would trigger,
2019     and simply start the timer with this timeout value.
2020    
2021     In other words, each time the callback is invoked it will check whether
2022 sf-exg 1.403 the timeout occurred. If not, it will simply reschedule itself to check
2023 root 1.387 again at the earliest time it could time out. Rinse. Repeat.
2024 root 1.198
2025 root 1.199 This scheme causes more callback invocations (about one every 60 seconds
2026     minus half the average time between activity), but virtually no calls to
2027     libev to change the timeout.
2028    
2029 root 1.387 To start the machinery, simply initialise the watcher and set
2030     C<last_activity> to the current time (meaning there was some activity just
2031     now), then call the callback, which will "do the right thing" and start
2032     the timer:
2033    
2034     last_activity = ev_now (EV_A);
2035     ev_init (&timer, callback);
2036     callback (EV_A_ &timer, 0);
2037 root 1.198
2038 root 1.387 When there is some activity, simply store the current time in
2039 root 1.199 C<last_activity>, no libev calls at all:
2040 root 1.198
2041 root 1.387 if (activity detected)
2042     last_activity = ev_now (EV_A);
2043    
2044     When your timeout value changes, then the timeout can be changed by simply
2045     providing a new value, stopping the timer and calling the callback, which
2046 sf-exg 1.403 will again do the right thing (for example, time out immediately :).
2047 root 1.387
2048     timeout = new_value;
2049     ev_timer_stop (EV_A_ &timer);
2050     callback (EV_A_ &timer, 0);
2051 root 1.198
2052     This technique is slightly more complex, but in most cases where the
2053     time-out is unlikely to be triggered, much more efficient.
2054    
2055 root 1.200 =item 4. Wee, just use a double-linked list for your timeouts.
2056 root 1.199
2057 root 1.200 If there is not one request, but many thousands (millions...), all
2058     employing some kind of timeout with the same timeout value, then one can
2059     do even better:
2060 root 1.199
2061     When starting the timeout, calculate the timeout value and put the timeout
2062     at the I<end> of the list.
2063    
2064     Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
2065     the list is expected to fire (for example, using the technique #3).
2066    
2067     When there is some activity, remove the timer from the list, recalculate
2068     the timeout, append it to the end of the list again, and make sure to
2069     update the C<ev_timer> if it was taken from the beginning of the list.
2070    
2071     This way, one can manage an unlimited number of timeouts in O(1) time for
2072     starting, stopping and updating the timers, at the expense of a major
2073     complication, and having to use a constant timeout. The constant timeout
2074     ensures that the list stays sorted.
2075    
2076 root 1.198 =back
2077    
2078 root 1.200 So which method the best?
2079 root 1.199
2080 root 1.200 Method #2 is a simple no-brain-required solution that is adequate in most
2081     situations. Method #3 requires a bit more thinking, but handles many cases
2082     better, and isn't very complicated either. In most case, choosing either
2083     one is fine, with #3 being better in typical situations.
2084 root 1.199
2085     Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
2086     rather complicated, but extremely efficient, something that really pays
2087 root 1.200 off after the first million or so of active timers, i.e. it's usually
2088 root 1.199 overkill :)
2089    
2090 root 1.381 =head3 The special problem of being too early
2091    
2092     If you ask a timer to call your callback after three seconds, then
2093     you expect it to be invoked after three seconds - but of course, this
2094     cannot be guaranteed to infinite precision. Less obviously, it cannot be
2095     guaranteed to any precision by libev - imagine somebody suspending the
2096 root 1.386 process with a STOP signal for a few hours for example.
2097 root 1.381
2098     So, libev tries to invoke your callback as soon as possible I<after> the
2099 sf-exg 1.382 delay has occurred, but cannot guarantee this.
2100 root 1.381
2101     A less obvious failure mode is calling your callback too early: many event
2102     loops compare timestamps with a "elapsed delay >= requested delay", but
2103     this can cause your callback to be invoked much earlier than you would
2104     expect.
2105    
2106     To see why, imagine a system with a clock that only offers full second
2107     resolution (think windows if you can't come up with a broken enough OS
2108     yourself). If you schedule a one-second timer at the time 500.9, then the
2109     event loop will schedule your timeout to elapse at a system time of 500
2110     (500.9 truncated to the resolution) + 1, or 501.
2111    
2112     If an event library looks at the timeout 0.1s later, it will see "501 >=
2113     501" and invoke the callback 0.1s after it was started, even though a
2114     one-second delay was requested - this is being "too early", despite best
2115     intentions.
2116    
2117     This is the reason why libev will never invoke the callback if the elapsed
2118     delay equals the requested delay, but only when the elapsed delay is
2119     larger than the requested delay. In the example above, libev would only invoke
2120     the callback at system time 502, or 1.1s after the timer was started.
2121    
2122     So, while libev cannot guarantee that your callback will be invoked
2123     exactly when requested, it I<can> and I<does> guarantee that the requested
2124     delay has actually elapsed, or in other words, it always errs on the "too
2125     late" side of things.
2126    
2127 root 1.175 =head3 The special problem of time updates
2128    
2129 root 1.383 Establishing the current time is a costly operation (it usually takes
2130     at least one system call): EV therefore updates its idea of the current
2131 root 1.310 time only before and after C<ev_run> collects new events, which causes a
2132 root 1.183 growing difference between C<ev_now ()> and C<ev_time ()> when handling
2133     lots of events in one iteration.
2134 root 1.175
2135 root 1.9 The relative timeouts are calculated relative to the C<ev_now ()>
2136     time. This is usually the right thing as this timestamp refers to the time
2137 root 1.28 of the event triggering whatever timeout you are modifying/starting. If
2138 root 1.175 you suspect event processing to be delayed and you I<need> to base the
2139 root 1.434 timeout on the current time, use something like the following to adjust
2140     for it:
2141 root 1.9
2142 root 1.434 ev_timer_set (&timer, after + (ev_time () - ev_now ()), 0.);
2143 root 1.9
2144 root 1.177 If the event loop is suspended for a long time, you can also force an
2145 root 1.176 update of the time returned by C<ev_now ()> by calling C<ev_now_update
2146 root 1.434 ()>, although that will push the event time of all outstanding events
2147     further into the future.
2148 root 1.176
2149 root 1.383 =head3 The special problem of unsynchronised clocks
2150 root 1.381
2151     Modern systems have a variety of clocks - libev itself uses the normal
2152     "wall clock" clock and, if available, the monotonic clock (to avoid time
2153     jumps).
2154    
2155     Neither of these clocks is synchronised with each other or any other clock
2156     on the system, so C<ev_time ()> might return a considerably different time
2157     than C<gettimeofday ()> or C<time ()>. On a GNU/Linux system, for example,
2158     a call to C<gettimeofday> might return a second count that is one higher
2159     than a directly following call to C<time>.
2160    
2161     The moral of this is to only compare libev-related timestamps with
2162     C<ev_time ()> and C<ev_now ()>, at least if you want better precision than
2163 sf-exg 1.382 a second or so.
2164 root 1.381
2165     One more problem arises due to this lack of synchronisation: if libev uses
2166     the system monotonic clock and you compare timestamps from C<ev_time>
2167     or C<ev_now> from when you started your timer and when your callback is
2168     invoked, you will find that sometimes the callback is a bit "early".
2169    
2170     This is because C<ev_timer>s work in real time, not wall clock time, so
2171     libev makes sure your callback is not invoked before the delay happened,
2172     I<measured according to the real time>, not the system clock.
2173    
2174     If your timeouts are based on a physical timescale (e.g. "time out this
2175     connection after 100 seconds") then this shouldn't bother you as it is
2176     exactly the right behaviour.
2177    
2178     If you want to compare wall clock/system timestamps to your timers, then
2179     you need to use C<ev_periodic>s, as these are based on the wall clock
2180     time, where your comparisons will always generate correct results.
2181    
2182 root 1.257 =head3 The special problems of suspended animation
2183    
2184     When you leave the server world it is quite customary to hit machines that
2185     can suspend/hibernate - what happens to the clocks during such a suspend?
2186    
2187     Some quick tests made with a Linux 2.6.28 indicate that a suspend freezes
2188     all processes, while the clocks (C<times>, C<CLOCK_MONOTONIC>) continue
2189     to run until the system is suspended, but they will not advance while the
2190     system is suspended. That means, on resume, it will be as if the program
2191     was frozen for a few seconds, but the suspend time will not be counted
2192     towards C<ev_timer> when a monotonic clock source is used. The real time
2193     clock advanced as expected, but if it is used as sole clocksource, then a
2194     long suspend would be detected as a time jump by libev, and timers would
2195     be adjusted accordingly.
2196    
2197     I would not be surprised to see different behaviour in different between
2198     operating systems, OS versions or even different hardware.
2199    
2200     The other form of suspend (job control, or sending a SIGSTOP) will see a
2201     time jump in the monotonic clocks and the realtime clock. If the program
2202     is suspended for a very long time, and monotonic clock sources are in use,
2203     then you can expect C<ev_timer>s to expire as the full suspension time
2204     will be counted towards the timers. When no monotonic clock source is in
2205     use, then libev will again assume a timejump and adjust accordingly.
2206    
2207     It might be beneficial for this latter case to call C<ev_suspend>
2208     and C<ev_resume> in code that handles C<SIGTSTP>, to at least get
2209     deterministic behaviour in this case (you can do nothing against
2210     C<SIGSTOP>).
2211    
2212 root 1.82 =head3 Watcher-Specific Functions and Data Members
2213    
2214 root 1.1 =over 4
2215    
2216     =item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
2217    
2218     =item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
2219    
2220 root 1.443 Configure the timer to trigger after C<after> seconds (fractional and
2221     negative values are supported). If C<repeat> is C<0.>, then it will
2222     automatically be stopped once the timeout is reached. If it is positive,
2223     then the timer will automatically be configured to trigger again C<repeat>
2224     seconds later, again, and again, until stopped manually.
2225 root 1.157
2226     The timer itself will do a best-effort at avoiding drift, that is, if
2227     you configure a timer to trigger every 10 seconds, then it will normally
2228     trigger at exactly 10 second intervals. If, however, your program cannot
2229     keep up with the timer (because it takes longer than those 10 seconds to
2230     do stuff) the timer will not fire more than once per event loop iteration.
2231 root 1.1
2232 root 1.132 =item ev_timer_again (loop, ev_timer *)
2233 root 1.1
2234 root 1.394 This will act as if the timer timed out, and restarts it again if it is
2235     repeating. It basically works like calling C<ev_timer_stop>, updating the
2236     timeout to the C<repeat> value and calling C<ev_timer_start>.
2237 root 1.1
2238 root 1.395 The exact semantics are as in the following rules, all of which will be
2239 root 1.394 applied to the watcher:
2240 root 1.1
2241 root 1.394 =over 4
2242    
2243     =item If the timer is pending, the pending status is always cleared.
2244    
2245     =item If the timer is started but non-repeating, stop it (as if it timed
2246     out, without invoking it).
2247 root 1.61
2248 root 1.394 =item If the timer is repeating, make the C<repeat> value the new timeout
2249     and start the timer, if necessary.
2250    
2251     =back
2252 root 1.1
2253 root 1.413 This sounds a bit complicated, see L</Be smart about timeouts>, above, for a
2254 root 1.198 usage example.
2255 root 1.183
2256 root 1.275 =item ev_tstamp ev_timer_remaining (loop, ev_timer *)
2257 root 1.258
2258     Returns the remaining time until a timer fires. If the timer is active,
2259     then this time is relative to the current event loop time, otherwise it's
2260     the timeout value currently configured.
2261    
2262     That is, after an C<ev_timer_set (w, 5, 7)>, C<ev_timer_remaining> returns
2263 sf-exg 1.280 C<5>. When the timer is started and one second passes, C<ev_timer_remaining>
2264 root 1.258 will return C<4>. When the timer expires and is restarted, it will return
2265     roughly C<7> (likely slightly less as callback invocation takes some time,
2266     too), and so on.
2267    
2268 root 1.48 =item ev_tstamp repeat [read-write]
2269    
2270     The current C<repeat> value. Will be used each time the watcher times out
2271 root 1.183 or C<ev_timer_again> is called, and determines the next timeout (if any),
2272 root 1.48 which is also when any modifications are taken into account.
2273 root 1.1
2274     =back
2275    
2276 root 1.111 =head3 Examples
2277    
2278 root 1.54 Example: Create a timer that fires after 60 seconds.
2279 root 1.34
2280 root 1.164 static void
2281 root 1.198 one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
2282 root 1.164 {
2283     .. one minute over, w is actually stopped right here
2284     }
2285    
2286 root 1.198 ev_timer mytimer;
2287 root 1.164 ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
2288     ev_timer_start (loop, &mytimer);
2289 root 1.34
2290 root 1.54 Example: Create a timeout timer that times out after 10 seconds of
2291 root 1.34 inactivity.
2292    
2293 root 1.164 static void
2294 root 1.198 timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
2295 root 1.164 {
2296     .. ten seconds without any activity
2297     }
2298    
2299 root 1.198 ev_timer mytimer;
2300 root 1.164 ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
2301     ev_timer_again (&mytimer); /* start timer */
2302 root 1.310 ev_run (loop, 0);
2303 root 1.164
2304     // and in some piece of code that gets executed on any "activity":
2305     // reset the timeout to start ticking again at 10 seconds
2306     ev_timer_again (&mytimer);
2307 root 1.34
2308    
2309 root 1.42 =head2 C<ev_periodic> - to cron or not to cron?
2310 root 1.1
2311     Periodic watchers are also timers of a kind, but they are very versatile
2312     (and unfortunately a bit complex).
2313    
2314 root 1.227 Unlike C<ev_timer>, periodic watchers are not based on real time (or
2315     relative time, the physical time that passes) but on wall clock time
2316 root 1.438 (absolute time, the thing you can read on your calendar or clock). The
2317 root 1.227 difference is that wall clock time can run faster or slower than real
2318     time, and time jumps are not uncommon (e.g. when you adjust your
2319     wrist-watch).
2320    
2321     You can tell a periodic watcher to trigger after some specific point
2322     in time: for example, if you tell a periodic watcher to trigger "in 10
2323     seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
2324     not a delay) and then reset your system clock to January of the previous
2325     year, then it will take a year or more to trigger the event (unlike an
2326     C<ev_timer>, which would still trigger roughly 10 seconds after starting
2327     it, as it uses a relative timeout).
2328    
2329     C<ev_periodic> watchers can also be used to implement vastly more complex
2330     timers, such as triggering an event on each "midnight, local time", or
2331 root 1.444 other complicated rules. This cannot easily be done with C<ev_timer>
2332     watchers, as those cannot react to time jumps.
2333 root 1.1
2334 root 1.161 As with timers, the callback is guaranteed to be invoked only when the
2335 root 1.230 point in time where it is supposed to trigger has passed. If multiple
2336     timers become ready during the same loop iteration then the ones with
2337     earlier time-out values are invoked before ones with later time-out values
2338 root 1.310 (but this is no longer true when a callback calls C<ev_run> recursively).
2339 root 1.28
2340 root 1.82 =head3 Watcher-Specific Functions and Data Members
2341    
2342 root 1.1 =over 4
2343    
2344 root 1.227 =item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
2345 root 1.1
2346 root 1.227 =item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
2347 root 1.1
2348 root 1.227 Lots of arguments, let's sort it out... There are basically three modes of
2349 root 1.183 operation, and we will explain them from simplest to most complex:
2350 root 1.1
2351     =over 4
2352    
2353 root 1.227 =item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
2354 root 1.1
2355 root 1.161 In this configuration the watcher triggers an event after the wall clock
2356 root 1.227 time C<offset> has passed. It will not repeat and will not adjust when a
2357     time jump occurs, that is, if it is to be run at January 1st 2011 then it
2358     will be stopped and invoked when the system clock reaches or surpasses
2359     this point in time.
2360 root 1.1
2361 root 1.227 =item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
2362 root 1.1
2363     In this mode the watcher will always be scheduled to time out at the next
2364 root 1.227 C<offset + N * interval> time (for some integer N, which can also be
2365     negative) and then repeat, regardless of any time jumps. The C<offset>
2366     argument is merely an offset into the C<interval> periods.
2367 root 1.1
2368 root 1.183 This can be used to create timers that do not drift with respect to the
2369 root 1.227 system clock, for example, here is an C<ev_periodic> that triggers each
2370     hour, on the hour (with respect to UTC):
2371 root 1.1
2372     ev_periodic_set (&periodic, 0., 3600., 0);
2373    
2374     This doesn't mean there will always be 3600 seconds in between triggers,
2375 root 1.161 but only that the callback will be called when the system time shows a
2376 root 1.12 full hour (UTC), or more correctly, when the system time is evenly divisible
2377 root 1.1 by 3600.
2378    
2379     Another way to think about it (for the mathematically inclined) is that
2380 root 1.10 C<ev_periodic> will try to run the callback in this mode at the next possible
2381 root 1.227 time where C<time = offset (mod interval)>, regardless of any time jumps.
2382 root 1.1
2383 root 1.367 The C<interval> I<MUST> be positive, and for numerical stability, the
2384     interval value should be higher than C<1/8192> (which is around 100
2385     microseconds) and C<offset> should be higher than C<0> and should have
2386     at most a similar magnitude as the current time (say, within a factor of
2387     ten). Typical values for offset are, in fact, C<0> or something between
2388     C<0> and C<interval>, which is also the recommended range.
2389 root 1.78
2390 root 1.161 Note also that there is an upper limit to how often a timer can fire (CPU
2391 root 1.158 speed for example), so if C<interval> is very small then timing stability
2392 root 1.161 will of course deteriorate. Libev itself tries to be exact to be about one
2393 root 1.158 millisecond (if the OS supports it and the machine is fast enough).
2394    
2395 root 1.227 =item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
2396 root 1.1
2397 root 1.227 In this mode the values for C<interval> and C<offset> are both being
2398 root 1.1 ignored. Instead, each time the periodic watcher gets scheduled, the
2399     reschedule callback will be called with the watcher as first, and the
2400     current time as second argument.
2401    
2402 root 1.227 NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, ever,
2403     or make ANY other event loop modifications whatsoever, unless explicitly
2404     allowed by documentation here>.
2405 root 1.1
2406 root 1.157 If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
2407     it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
2408     only event loop modification you are allowed to do).
2409    
2410 root 1.198 The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
2411 root 1.157 *w, ev_tstamp now)>, e.g.:
2412 root 1.1
2413 root 1.198 static ev_tstamp
2414     my_rescheduler (ev_periodic *w, ev_tstamp now)
2415 root 1.1 {
2416     return now + 60.;
2417     }
2418    
2419     It must return the next time to trigger, based on the passed time value
2420     (that is, the lowest time value larger than to the second argument). It
2421     will usually be called just before the callback will be triggered, but
2422     might be called at other times, too.
2423    
2424 root 1.157 NOTE: I<< This callback must always return a time that is higher than or
2425     equal to the passed C<now> value >>.
2426 root 1.18
2427 root 1.1 This can be used to create very complex timers, such as a timer that
2428 root 1.444 triggers on "next midnight, local time". To do this, you would calculate
2429     the next midnight after C<now> and return the timestamp value for
2430     this. Here is a (completely untested, no error checking) example on how to
2431     do this:
2432    
2433     #include <time.h>
2434    
2435     static ev_tstamp
2436     my_rescheduler (ev_periodic *w, ev_tstamp now)
2437     {
2438     time_t tnow = (time_t)now;
2439     struct tm tm;
2440     localtime_r (&tnow, &tm);
2441    
2442     tm.tm_sec = tm.tm_min = tm.tm_hour = 0; // midnight current day
2443     ++tm.tm_mday; // midnight next day
2444    
2445     return mktime (&tm);
2446     }
2447    
2448     Note: this code might run into trouble on days that have more then two
2449     midnights (beginning and end).
2450 root 1.1
2451     =back
2452    
2453     =item ev_periodic_again (loop, ev_periodic *)
2454    
2455     Simply stops and restarts the periodic watcher again. This is only useful
2456     when you changed some parameters or the reschedule callback would return
2457     a different time than the last time it was called (e.g. in a crond like
2458     program when the crontabs have changed).
2459    
2460 root 1.149 =item ev_tstamp ev_periodic_at (ev_periodic *)
2461    
2462 root 1.227 When active, returns the absolute time that the watcher is supposed
2463     to trigger next. This is not the same as the C<offset> argument to
2464     C<ev_periodic_set>, but indeed works even in interval and manual
2465     rescheduling modes.
2466 root 1.149
2467 root 1.78 =item ev_tstamp offset [read-write]
2468    
2469     When repeating, this contains the offset value, otherwise this is the
2470 root 1.227 absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
2471     although libev might modify this value for better numerical stability).
2472 root 1.78
2473     Can be modified any time, but changes only take effect when the periodic
2474     timer fires or C<ev_periodic_again> is being called.
2475    
2476 root 1.48 =item ev_tstamp interval [read-write]
2477    
2478     The current interval value. Can be modified any time, but changes only
2479     take effect when the periodic timer fires or C<ev_periodic_again> is being
2480     called.
2481    
2482 root 1.198 =item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
2483 root 1.48
2484     The current reschedule callback, or C<0>, if this functionality is
2485     switched off. Can be changed any time, but changes only take effect when
2486     the periodic timer fires or C<ev_periodic_again> is being called.
2487    
2488 root 1.1 =back
2489    
2490 root 1.111 =head3 Examples
2491    
2492 root 1.54 Example: Call a callback every hour, or, more precisely, whenever the
2493 root 1.183 system time is divisible by 3600. The callback invocation times have
2494 root 1.161 potentially a lot of jitter, but good long-term stability.
2495 root 1.34
2496 root 1.164 static void
2497 root 1.301 clock_cb (struct ev_loop *loop, ev_periodic *w, int revents)
2498 root 1.164 {
2499     ... its now a full hour (UTC, or TAI or whatever your clock follows)
2500     }
2501    
2502 root 1.198 ev_periodic hourly_tick;
2503 root 1.164 ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
2504     ev_periodic_start (loop, &hourly_tick);
2505 root 1.34
2506 root 1.54 Example: The same as above, but use a reschedule callback to do it:
2507 root 1.34
2508 root 1.164 #include <math.h>
2509 root 1.34
2510 root 1.164 static ev_tstamp
2511 root 1.198 my_scheduler_cb (ev_periodic *w, ev_tstamp now)
2512 root 1.164 {
2513 root 1.183 return now + (3600. - fmod (now, 3600.));
2514 root 1.164 }
2515 root 1.34
2516 root 1.164 ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
2517 root 1.34
2518 root 1.54 Example: Call a callback every hour, starting now:
2519 root 1.34
2520 root 1.198 ev_periodic hourly_tick;
2521 root 1.164 ev_periodic_init (&hourly_tick, clock_cb,
2522     fmod (ev_now (loop), 3600.), 3600., 0);
2523     ev_periodic_start (loop, &hourly_tick);
2524 root 1.432
2525 root 1.34
2526 root 1.42 =head2 C<ev_signal> - signal me when a signal gets signalled!
2527 root 1.1
2528     Signal watchers will trigger an event when the process receives a specific
2529     signal one or more times. Even though signals are very asynchronous, libev
2530 root 1.340 will try its best to deliver signals synchronously, i.e. as part of the
2531 root 1.1 normal event processing, like any other event.
2532    
2533 root 1.260 If you want signals to be delivered truly asynchronously, just use
2534     C<sigaction> as you would do without libev and forget about sharing
2535     the signal. You can even use C<ev_async> from a signal handler to
2536     synchronously wake up an event loop.
2537    
2538     You can configure as many watchers as you like for the same signal, but
2539     only within the same loop, i.e. you can watch for C<SIGINT> in your
2540     default loop and for C<SIGIO> in another loop, but you cannot watch for
2541     C<SIGINT> in both the default loop and another loop at the same time. At
2542     the moment, C<SIGCHLD> is permanently tied to the default loop.
2543    
2544 root 1.431 Only after the first watcher for a signal is started will libev actually
2545     register something with the kernel. It thus coexists with your own signal
2546     handlers as long as you don't register any with libev for the same signal.
2547 root 1.259
2548 root 1.135 If possible and supported, libev will install its handlers with
2549 root 1.259 C<SA_RESTART> (or equivalent) behaviour enabled, so system calls should
2550     not be unduly interrupted. If you have a problem with system calls getting
2551     interrupted by signals you can block all signals in an C<ev_check> watcher
2552     and unblock them in an C<ev_prepare> watcher.
2553 root 1.135
2554 root 1.277 =head3 The special problem of inheritance over fork/execve/pthread_create
2555 root 1.265
2556     Both the signal mask (C<sigprocmask>) and the signal disposition
2557     (C<sigaction>) are unspecified after starting a signal watcher (and after
2558     stopping it again), that is, libev might or might not block the signal,
2559 root 1.360 and might or might not set or restore the installed signal handler (but
2560     see C<EVFLAG_NOSIGMASK>).
2561 root 1.265
2562     While this does not matter for the signal disposition (libev never
2563     sets signals to C<SIG_IGN>, so handlers will be reset to C<SIG_DFL> on
2564     C<execve>), this matters for the signal mask: many programs do not expect
2565 root 1.266 certain signals to be blocked.
2566 root 1.265
2567     This means that before calling C<exec> (from the child) you should reset
2568     the signal mask to whatever "default" you expect (all clear is a good
2569     choice usually).
2570    
2571 root 1.267 The simplest way to ensure that the signal mask is reset in the child is
2572     to install a fork handler with C<pthread_atfork> that resets it. That will
2573     catch fork calls done by libraries (such as the libc) as well.
2574    
2575 root 1.277 In current versions of libev, the signal will not be blocked indefinitely
2576     unless you use the C<signalfd> API (C<EV_SIGNALFD>). While this reduces
2577     the window of opportunity for problems, it will not go away, as libev
2578     I<has> to modify the signal mask, at least temporarily.
2579    
2580 root 1.278 So I can't stress this enough: I<If you do not reset your signal mask when
2581     you expect it to be empty, you have a race condition in your code>. This
2582     is not a libev-specific thing, this is true for most event libraries.
2583 root 1.266
2584 root 1.349 =head3 The special problem of threads signal handling
2585    
2586     POSIX threads has problematic signal handling semantics, specifically,
2587     a lot of functionality (sigfd, sigwait etc.) only really works if all
2588     threads in a process block signals, which is hard to achieve.
2589    
2590     When you want to use sigwait (or mix libev signal handling with your own
2591     for the same signals), you can tackle this problem by globally blocking
2592     all signals before creating any threads (or creating them with a fully set
2593     sigprocmask) and also specifying the C<EVFLAG_NOSIGMASK> when creating
2594     loops. Then designate one thread as "signal receiver thread" which handles
2595     these signals. You can pass on any signals that libev might be interested
2596     in by calling C<ev_feed_signal>.
2597    
2598 root 1.82 =head3 Watcher-Specific Functions and Data Members
2599    
2600 root 1.1 =over 4
2601    
2602     =item ev_signal_init (ev_signal *, callback, int signum)
2603    
2604     =item ev_signal_set (ev_signal *, int signum)
2605    
2606     Configures the watcher to trigger on the given signal number (usually one
2607     of the C<SIGxxx> constants).
2608    
2609 root 1.48 =item int signum [read-only]
2610    
2611     The signal the watcher watches out for.
2612    
2613 root 1.1 =back
2614    
2615 root 1.132 =head3 Examples
2616    
2617 root 1.188 Example: Try to exit cleanly on SIGINT.
2618 root 1.132
2619 root 1.164 static void
2620 root 1.198 sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
2621 root 1.164 {
2622 root 1.310 ev_break (loop, EVBREAK_ALL);
2623 root 1.164 }
2624    
2625 root 1.198 ev_signal signal_watcher;
2626 root 1.164 ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
2627 root 1.188 ev_signal_start (loop, &signal_watcher);
2628 root 1.132
2629 root 1.35
2630 root 1.42 =head2 C<ev_child> - watch out for process status changes
2631 root 1.1
2632     Child watchers trigger when your process receives a SIGCHLD in response to
2633 root 1.183 some child status changes (most typically when a child of yours dies or
2634     exits). It is permissible to install a child watcher I<after> the child
2635     has been forked (which implies it might have already exited), as long
2636     as the event loop isn't entered (or is continued from a watcher), i.e.,
2637     forking and then immediately registering a watcher for the child is fine,
2638 root 1.244 but forking and registering a watcher a few event loop iterations later or
2639     in the next callback invocation is not.
2640 root 1.134
2641     Only the default event loop is capable of handling signals, and therefore
2642 root 1.161 you can only register child watchers in the default event loop.
2643 root 1.134
2644 root 1.248 Due to some design glitches inside libev, child watchers will always be
2645 root 1.249 handled at maximum priority (their priority is set to C<EV_MAXPRI> by
2646     libev)
2647 root 1.248
2648 root 1.134 =head3 Process Interaction
2649    
2650     Libev grabs C<SIGCHLD> as soon as the default event loop is
2651 root 1.259 initialised. This is necessary to guarantee proper behaviour even if the
2652     first child watcher is started after the child exits. The occurrence
2653 root 1.134 of C<SIGCHLD> is recorded asynchronously, but child reaping is done
2654     synchronously as part of the event loop processing. Libev always reaps all
2655     children, even ones not watched.
2656    
2657     =head3 Overriding the Built-In Processing
2658    
2659     Libev offers no special support for overriding the built-in child
2660     processing, but if your application collides with libev's default child
2661     handler, you can override it easily by installing your own handler for
2662     C<SIGCHLD> after initialising the default loop, and making sure the
2663     default loop never gets destroyed. You are encouraged, however, to use an
2664     event-based approach to child reaping and thus use libev's support for
2665     that, so other libev users can use C<ev_child> watchers freely.
2666 root 1.1
2667 root 1.173 =head3 Stopping the Child Watcher
2668    
2669     Currently, the child watcher never gets stopped, even when the
2670     child terminates, so normally one needs to stop the watcher in the
2671     callback. Future versions of libev might stop the watcher automatically
2672 root 1.259 when a child exit is detected (calling C<ev_child_stop> twice is not a
2673     problem).
2674 root 1.173
2675 root 1.82 =head3 Watcher-Specific Functions and Data Members
2676    
2677 root 1.1 =over 4
2678    
2679 root 1.120 =item ev_child_init (ev_child *, callback, int pid, int trace)
2680 root 1.1
2681 root 1.120 =item ev_child_set (ev_child *, int pid, int trace)
2682 root 1.1
2683     Configures the watcher to wait for status changes of process C<pid> (or
2684     I<any> process if C<pid> is specified as C<0>). The callback can look
2685     at the C<rstatus> member of the C<ev_child> watcher structure to see
2686 root 1.14 the status word (use the macros from C<sys/wait.h> and see your systems
2687     C<waitpid> documentation). The C<rpid> member contains the pid of the
2688 root 1.120 process causing the status change. C<trace> must be either C<0> (only
2689     activate the watcher when the process terminates) or C<1> (additionally
2690     activate the watcher when the process is stopped or continued).
2691 root 1.1
2692 root 1.48 =item int pid [read-only]
2693    
2694     The process id this watcher watches out for, or C<0>, meaning any process id.
2695    
2696     =item int rpid [read-write]
2697    
2698     The process id that detected a status change.
2699    
2700     =item int rstatus [read-write]
2701    
2702     The process exit/trace status caused by C<rpid> (see your systems
2703     C<waitpid> and C<sys/wait.h> documentation for details).
2704    
2705 root 1.1 =back
2706    
2707 root 1.134 =head3 Examples
2708    
2709     Example: C<fork()> a new process and install a child handler to wait for
2710     its completion.
2711    
2712 root 1.164 ev_child cw;
2713    
2714     static void
2715 root 1.198 child_cb (EV_P_ ev_child *w, int revents)
2716 root 1.164 {
2717     ev_child_stop (EV_A_ w);
2718     printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
2719     }
2720    
2721     pid_t pid = fork ();
2722 root 1.134
2723 root 1.164 if (pid < 0)
2724     // error
2725     else if (pid == 0)
2726     {
2727     // the forked child executes here
2728     exit (1);
2729     }
2730     else
2731     {
2732     ev_child_init (&cw, child_cb, pid, 0);
2733     ev_child_start (EV_DEFAULT_ &cw);
2734     }
2735 root 1.134
2736 root 1.34
2737 root 1.48 =head2 C<ev_stat> - did the file attributes just change?
2738    
2739 root 1.161 This watches a file system path for attribute changes. That is, it calls
2740 root 1.207 C<stat> on that path in regular intervals (or when the OS says it changed)
2741 root 1.425 and sees if it changed compared to the last time, invoking the callback
2742     if it did. Starting the watcher C<stat>'s the file, so only changes that
2743     happen after the watcher has been started will be reported.
2744 root 1.48
2745     The path does not need to exist: changing from "path exists" to "path does
2746 root 1.211 not exist" is a status change like any other. The condition "path does not
2747     exist" (or more correctly "path cannot be stat'ed") is signified by the
2748     C<st_nlink> field being zero (which is otherwise always forced to be at
2749     least one) and all the other fields of the stat buffer having unspecified
2750     contents.
2751 root 1.48
2752 root 1.207 The path I<must not> end in a slash or contain special components such as
2753     C<.> or C<..>. The path I<should> be absolute: If it is relative and
2754     your working directory changes, then the behaviour is undefined.
2755    
2756     Since there is no portable change notification interface available, the
2757     portable implementation simply calls C<stat(2)> regularly on the path
2758     to see if it changed somehow. You can specify a recommended polling
2759     interval for this case. If you specify a polling interval of C<0> (highly
2760     recommended!) then a I<suitable, unspecified default> value will be used
2761     (which you can expect to be around five seconds, although this might
2762     change dynamically). Libev will also impose a minimum interval which is
2763 root 1.208 currently around C<0.1>, but that's usually overkill.
2764 root 1.48
2765     This watcher type is not meant for massive numbers of stat watchers,
2766     as even with OS-supported change notifications, this can be
2767     resource-intensive.
2768    
2769 root 1.183 At the time of this writing, the only OS-specific interface implemented
2770 root 1.211 is the Linux inotify interface (implementing kqueue support is left as an
2771     exercise for the reader. Note, however, that the author sees no way of
2772     implementing C<ev_stat> semantics with kqueue, except as a hint).
2773 root 1.48
2774 root 1.137 =head3 ABI Issues (Largefile Support)
2775    
2776     Libev by default (unless the user overrides this) uses the default
2777 root 1.169 compilation environment, which means that on systems with large file
2778     support disabled by default, you get the 32 bit version of the stat
2779 root 1.137 structure. When using the library from programs that change the ABI to
2780     use 64 bit file offsets the programs will fail. In that case you have to
2781     compile libev with the same flags to get binary compatibility. This is
2782     obviously the case with any flags that change the ABI, but the problem is
2783 root 1.207 most noticeably displayed with ev_stat and large file support.
2784 root 1.169
2785     The solution for this is to lobby your distribution maker to make large
2786     file interfaces available by default (as e.g. FreeBSD does) and not
2787     optional. Libev cannot simply switch on large file support because it has
2788     to exchange stat structures with application programs compiled using the
2789     default compilation environment.
2790 root 1.137
2791 root 1.183 =head3 Inotify and Kqueue
2792 root 1.108
2793 root 1.211 When C<inotify (7)> support has been compiled into libev and present at
2794     runtime, it will be used to speed up change detection where possible. The
2795     inotify descriptor will be created lazily when the first C<ev_stat>
2796     watcher is being started.
2797 root 1.108
2798 root 1.147 Inotify presence does not change the semantics of C<ev_stat> watchers
2799 root 1.108 except that changes might be detected earlier, and in some cases, to avoid
2800 root 1.147 making regular C<stat> calls. Even in the presence of inotify support
2801 root 1.183 there are many cases where libev has to resort to regular C<stat> polling,
2802 root 1.211 but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
2803     many bugs), the path exists (i.e. stat succeeds), and the path resides on
2804     a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
2805     xfs are fully working) libev usually gets away without polling.
2806 root 1.108
2807 root 1.183 There is no support for kqueue, as apparently it cannot be used to
2808 root 1.108 implement this functionality, due to the requirement of having a file
2809 root 1.183 descriptor open on the object at all times, and detecting renames, unlinks
2810     etc. is difficult.
2811 root 1.108
2812 root 1.212 =head3 C<stat ()> is a synchronous operation
2813    
2814     Libev doesn't normally do any kind of I/O itself, and so is not blocking
2815     the process. The exception are C<ev_stat> watchers - those call C<stat
2816     ()>, which is a synchronous operation.
2817    
2818     For local paths, this usually doesn't matter: unless the system is very
2819     busy or the intervals between stat's are large, a stat call will be fast,
2820 root 1.222 as the path data is usually in memory already (except when starting the
2821 root 1.212 watcher).
2822    
2823     For networked file systems, calling C<stat ()> can block an indefinite
2824     time due to network issues, and even under good conditions, a stat call
2825     often takes multiple milliseconds.
2826    
2827     Therefore, it is best to avoid using C<ev_stat> watchers on networked
2828     paths, although this is fully supported by libev.
2829    
2830 root 1.107 =head3 The special problem of stat time resolution
2831    
2832 root 1.207 The C<stat ()> system call only supports full-second resolution portably,
2833     and even on systems where the resolution is higher, most file systems
2834     still only support whole seconds.
2835 root 1.107
2836 root 1.150 That means that, if the time is the only thing that changes, you can
2837     easily miss updates: on the first update, C<ev_stat> detects a change and
2838     calls your callback, which does something. When there is another update
2839 root 1.183 within the same second, C<ev_stat> will be unable to detect unless the
2840     stat data does change in other ways (e.g. file size).
2841 root 1.150
2842     The solution to this is to delay acting on a change for slightly more
2843 root 1.155 than a second (or till slightly after the next full second boundary), using
2844 root 1.150 a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
2845     ev_timer_again (loop, w)>).
2846    
2847     The C<.02> offset is added to work around small timing inconsistencies
2848     of some operating systems (where the second counter of the current time
2849     might be be delayed. One such system is the Linux kernel, where a call to
2850     C<gettimeofday> might return a timestamp with a full second later than
2851     a subsequent C<time> call - if the equivalent of C<time ()> is used to
2852     update file times then there will be a small window where the kernel uses
2853     the previous second to update file times but libev might already execute
2854     the timer callback).
2855 root 1.107
2856 root 1.82 =head3 Watcher-Specific Functions and Data Members
2857    
2858 root 1.48 =over 4
2859    
2860     =item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
2861    
2862     =item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
2863    
2864     Configures the watcher to wait for status changes of the given
2865     C<path>. The C<interval> is a hint on how quickly a change is expected to
2866     be detected and should normally be specified as C<0> to let libev choose
2867     a suitable value. The memory pointed to by C<path> must point to the same
2868     path for as long as the watcher is active.
2869    
2870 root 1.183 The callback will receive an C<EV_STAT> event when a change was detected,
2871     relative to the attributes at the time the watcher was started (or the
2872     last change was detected).
2873 root 1.48
2874 root 1.132 =item ev_stat_stat (loop, ev_stat *)
2875 root 1.48
2876     Updates the stat buffer immediately with new values. If you change the
2877 root 1.150 watched path in your callback, you could call this function to avoid
2878     detecting this change (while introducing a race condition if you are not
2879     the only one changing the path). Can also be useful simply to find out the
2880     new values.
2881 root 1.48
2882     =item ev_statdata attr [read-only]
2883    
2884 root 1.150 The most-recently detected attributes of the file. Although the type is
2885 root 1.48 C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
2886 root 1.150 suitable for your system, but you can only rely on the POSIX-standardised
2887     members to be present. If the C<st_nlink> member is C<0>, then there was
2888     some error while C<stat>ing the file.
2889 root 1.48
2890     =item ev_statdata prev [read-only]
2891    
2892     The previous attributes of the file. The callback gets invoked whenever
2893 root 1.150 C<prev> != C<attr>, or, more precisely, one or more of these members
2894     differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
2895     C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
2896 root 1.48
2897     =item ev_tstamp interval [read-only]
2898    
2899     The specified interval.
2900    
2901     =item const char *path [read-only]
2902    
2903 root 1.161 The file system path that is being watched.
2904 root 1.48
2905     =back
2906    
2907 root 1.108 =head3 Examples
2908    
2909 root 1.48 Example: Watch C</etc/passwd> for attribute changes.
2910    
2911 root 1.164 static void
2912     passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
2913     {
2914     /* /etc/passwd changed in some way */
2915     if (w->attr.st_nlink)
2916     {
2917     printf ("passwd current size %ld\n", (long)w->attr.st_size);
2918     printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
2919     printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
2920     }
2921     else
2922     /* you shalt not abuse printf for puts */
2923     puts ("wow, /etc/passwd is not there, expect problems. "
2924     "if this is windows, they already arrived\n");
2925     }
2926 root 1.48
2927 root 1.164 ...
2928     ev_stat passwd;
2929 root 1.48
2930 root 1.164 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
2931     ev_stat_start (loop, &passwd);
2932 root 1.107
2933     Example: Like above, but additionally use a one-second delay so we do not
2934     miss updates (however, frequent updates will delay processing, too, so
2935     one might do the work both on C<ev_stat> callback invocation I<and> on
2936     C<ev_timer> callback invocation).
2937    
2938 root 1.164 static ev_stat passwd;
2939     static ev_timer timer;
2940 root 1.107
2941 root 1.164 static void
2942     timer_cb (EV_P_ ev_timer *w, int revents)
2943     {
2944     ev_timer_stop (EV_A_ w);
2945    
2946     /* now it's one second after the most recent passwd change */
2947     }
2948    
2949     static void
2950     stat_cb (EV_P_ ev_stat *w, int revents)
2951     {
2952     /* reset the one-second timer */
2953     ev_timer_again (EV_A_ &timer);
2954     }
2955    
2956     ...
2957     ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
2958     ev_stat_start (loop, &passwd);
2959     ev_timer_init (&timer, timer_cb, 0., 1.02);
2960 root 1.48
2961    
2962 root 1.42 =head2 C<ev_idle> - when you've got nothing better to do...
2963 root 1.1
2964 root 1.67 Idle watchers trigger events when no other events of the same or higher
2965 root 1.183 priority are pending (prepare, check and other idle watchers do not count
2966     as receiving "events").
2967 root 1.67
2968     That is, as long as your process is busy handling sockets or timeouts
2969     (or even signals, imagine) of the same or higher priority it will not be
2970     triggered. But when your process is idle (or only lower-priority watchers
2971     are pending), the idle watchers are being called once per event loop
2972     iteration - until stopped, that is, or your process receives more events
2973     and becomes busy again with higher priority stuff.
2974 root 1.1
2975     The most noteworthy effect is that as long as any idle watchers are
2976     active, the process will not block when waiting for new events.
2977    
2978     Apart from keeping your process non-blocking (which is a useful
2979     effect on its own sometimes), idle watchers are a good place to do
2980     "pseudo-background processing", or delay processing stuff to after the
2981     event loop has handled all outstanding events.
2982    
2983 root 1.406 =head3 Abusing an C<ev_idle> watcher for its side-effect
2984    
2985     As long as there is at least one active idle watcher, libev will never
2986     sleep unnecessarily. Or in other words, it will loop as fast as possible.
2987     For this to work, the idle watcher doesn't need to be invoked at all - the
2988     lowest priority will do.
2989    
2990     This mode of operation can be useful together with an C<ev_check> watcher,
2991     to do something on each event loop iteration - for example to balance load
2992     between different connections.
2993    
2994 root 1.415 See L</Abusing an ev_check watcher for its side-effect> for a longer
2995 root 1.406 example.
2996    
2997 root 1.82 =head3 Watcher-Specific Functions and Data Members
2998    
2999 root 1.1 =over 4
3000    
3001 root 1.226 =item ev_idle_init (ev_idle *, callback)
3002 root 1.1
3003     Initialises and configures the idle watcher - it has no parameters of any
3004     kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
3005     believe me.
3006    
3007     =back
3008    
3009 root 1.111 =head3 Examples
3010    
3011 root 1.54 Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
3012     callback, free it. Also, use no error checking, as usual.
3013 root 1.34
3014 root 1.164 static void
3015 root 1.198 idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
3016 root 1.164 {
3017 root 1.407 // stop the watcher
3018     ev_idle_stop (loop, w);
3019    
3020     // now we can free it
3021 root 1.164 free (w);
3022 root 1.407
3023 root 1.164 // now do something you wanted to do when the program has
3024     // no longer anything immediate to do.
3025     }
3026    
3027 root 1.198 ev_idle *idle_watcher = malloc (sizeof (ev_idle));
3028 root 1.164 ev_idle_init (idle_watcher, idle_cb);
3029 root 1.242 ev_idle_start (loop, idle_watcher);
3030 root 1.34
3031    
3032 root 1.42 =head2 C<ev_prepare> and C<ev_check> - customise your event loop!
3033 root 1.1
3034 root 1.406 Prepare and check watchers are often (but not always) used in pairs:
3035 root 1.20 prepare watchers get invoked before the process blocks and check watchers
3036 root 1.14 afterwards.
3037 root 1.1
3038 root 1.433 You I<must not> call C<ev_run> (or similar functions that enter the
3039     current event loop) or C<ev_loop_fork> from either C<ev_prepare> or
3040     C<ev_check> watchers. Other loops than the current one are fine,
3041     however. The rationale behind this is that you do not need to check
3042     for recursion in those watchers, i.e. the sequence will always be
3043     C<ev_prepare>, blocking, C<ev_check> so if you have one watcher of each
3044     kind they will always be called in pairs bracketing the blocking call.
3045 root 1.45
3046 root 1.35 Their main purpose is to integrate other event mechanisms into libev and
3047 root 1.183 their use is somewhat advanced. They could be used, for example, to track
3048 root 1.35 variable changes, implement your own watchers, integrate net-snmp or a
3049 root 1.45 coroutine library and lots more. They are also occasionally useful if
3050     you cache some data and want to flush it before blocking (for example,
3051     in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
3052     watcher).
3053 root 1.1
3054 root 1.183 This is done by examining in each prepare call which file descriptors
3055     need to be watched by the other library, registering C<ev_io> watchers
3056     for them and starting an C<ev_timer> watcher for any timeouts (many
3057     libraries provide exactly this functionality). Then, in the check watcher,
3058     you check for any events that occurred (by checking the pending status
3059     of all watchers and stopping them) and call back into the library. The
3060     I/O and timer callbacks will never actually be called (but must be valid
3061     nevertheless, because you never know, you know?).
3062 root 1.1
3063 root 1.14 As another example, the Perl Coro module uses these hooks to integrate
3064 root 1.1 coroutines into libev programs, by yielding to other active coroutines
3065     during each prepare and only letting the process block if no coroutines
3066 root 1.20 are ready to run (it's actually more complicated: it only runs coroutines
3067     with priority higher than or equal to the event loop and one coroutine
3068     of lower priority, but only once, using idle watchers to keep the event
3069     loop from blocking if lower-priority coroutines are active, thus mapping
3070     low-priority coroutines to idle/background tasks).
3071 root 1.1
3072 root 1.406 When used for this purpose, it is recommended to give C<ev_check> watchers
3073     highest (C<EV_MAXPRI>) priority, to ensure that they are being run before
3074     any other watchers after the poll (this doesn't matter for C<ev_prepare>
3075     watchers).
3076 root 1.183
3077     Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
3078     activate ("feed") events into libev. While libev fully supports this, they
3079     might get executed before other C<ev_check> watchers did their job. As
3080     C<ev_check> watchers are often used to embed other (non-libev) event
3081     loops those other event loops might be in an unusable state until their
3082     C<ev_check> watcher ran (always remind yourself to coexist peacefully with
3083     others).
3084 root 1.77
3085 root 1.406 =head3 Abusing an C<ev_check> watcher for its side-effect
3086    
3087     C<ev_check> (and less often also C<ev_prepare>) watchers can also be
3088     useful because they are called once per event loop iteration. For
3089     example, if you want to handle a large number of connections fairly, you
3090     normally only do a bit of work for each active connection, and if there
3091     is more work to do, you wait for the next event loop iteration, so other
3092     connections have a chance of making progress.
3093    
3094     Using an C<ev_check> watcher is almost enough: it will be called on the
3095     next event loop iteration. However, that isn't as soon as possible -
3096     without external events, your C<ev_check> watcher will not be invoked.
3097    
3098     This is where C<ev_idle> watchers come in handy - all you need is a
3099     single global idle watcher that is active as long as you have one active
3100     C<ev_check> watcher. The C<ev_idle> watcher makes sure the event loop
3101     will not sleep, and the C<ev_check> watcher makes sure a callback gets
3102     invoked. Neither watcher alone can do that.
3103    
3104 root 1.82 =head3 Watcher-Specific Functions and Data Members
3105    
3106 root 1.1 =over 4
3107    
3108     =item ev_prepare_init (ev_prepare *, callback)
3109    
3110     =item ev_check_init (ev_check *, callback)
3111    
3112     Initialises and configures the prepare or check watcher - they have no
3113     parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
3114 root 1.183 macros, but using them is utterly, utterly, utterly and completely
3115     pointless.
3116 root 1.1
3117     =back
3118    
3119 root 1.111 =head3 Examples
3120    
3121 root 1.76 There are a number of principal ways to embed other event loops or modules
3122     into libev. Here are some ideas on how to include libadns into libev
3123     (there is a Perl module named C<EV::ADNS> that does this, which you could
3124 root 1.150 use as a working example. Another Perl module named C<EV::Glib> embeds a
3125     Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
3126     Glib event loop).
3127 root 1.76
3128     Method 1: Add IO watchers and a timeout watcher in a prepare handler,
3129     and in a check watcher, destroy them and call into libadns. What follows
3130     is pseudo-code only of course. This requires you to either use a low
3131     priority for the check watcher or use C<ev_clear_pending> explicitly, as
3132     the callbacks for the IO/timeout watchers might not have been called yet.
3133 root 1.45
3134 root 1.164 static ev_io iow [nfd];
3135     static ev_timer tw;
3136 root 1.45
3137 root 1.164 static void
3138 root 1.198 io_cb (struct ev_loop *loop, ev_io *w, int revents)
3139 root 1.164 {
3140     }
3141 root 1.45
3142 root 1.164 // create io watchers for each fd and a timer before blocking
3143     static void
3144 root 1.198 adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
3145 root 1.164 {
3146     int timeout = 3600000;
3147     struct pollfd fds [nfd];
3148     // actual code will need to loop here and realloc etc.
3149     adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
3150    
3151     /* the callback is illegal, but won't be called as we stop during check */
3152 root 1.243 ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
3153 root 1.164 ev_timer_start (loop, &tw);
3154    
3155     // create one ev_io per pollfd
3156     for (int i = 0; i < nfd; ++i)
3157     {
3158     ev_io_init (iow + i, io_cb, fds [i].fd,
3159     ((fds [i].events & POLLIN ? EV_READ : 0)
3160     | (fds [i].events & POLLOUT ? EV_WRITE : 0)));
3161    
3162     fds [i].revents = 0;
3163     ev_io_start (loop, iow + i);
3164     }
3165     }
3166    
3167     // stop all watchers after blocking
3168     static void
3169 root 1.198 adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
3170 root 1.164 {
3171     ev_timer_stop (loop, &tw);
3172    
3173     for (int i = 0; i < nfd; ++i)
3174     {
3175     // set the relevant poll flags
3176     // could also call adns_processreadable etc. here
3177     struct pollfd *fd = fds + i;
3178     int revents = ev_clear_pending (iow + i);
3179     if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
3180     if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
3181    
3182     // now stop the watcher
3183     ev_io_stop (loop, iow + i);
3184     }
3185    
3186     adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
3187     }
3188 root 1.34
3189 root 1.76 Method 2: This would be just like method 1, but you run C<adns_afterpoll>
3190     in the prepare watcher and would dispose of the check watcher.
3191    
3192     Method 3: If the module to be embedded supports explicit event
3193 root 1.161 notification (libadns does), you can also make use of the actual watcher
3194 root 1.76 callbacks, and only destroy/create the watchers in the prepare watcher.
3195    
3196 root 1.164 static void
3197     timer_cb (EV_P_ ev_timer *w, int revents)
3198     {
3199     adns_state ads = (adns_state)w->data;
3200     update_now (EV_A);
3201    
3202     adns_processtimeouts (ads, &tv_now);
3203     }
3204    
3205     static void
3206     io_cb (EV_P_ ev_io *w, int revents)
3207     {
3208     adns_state ads = (adns_state)w->data;
3209     update_now (EV_A);
3210    
3211     if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
3212     if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
3213     }
3214 root 1.76
3215 root 1.164 // do not ever call adns_afterpoll
3216 root 1.76
3217     Method 4: Do not use a prepare or check watcher because the module you
3218 root 1.183 want to embed is not flexible enough to support it. Instead, you can
3219     override their poll function. The drawback with this solution is that the
3220     main loop is now no longer controllable by EV. The C<Glib::EV> module uses
3221     this approach, effectively embedding EV as a client into the horrible
3222     libglib event loop.
3223 root 1.76
3224 root 1.164 static gint
3225     event_poll_func (GPollFD *fds, guint nfds, gint timeout)
3226     {
3227     int got_events = 0;
3228    
3229     for (n = 0; n < nfds; ++n)
3230     // create/start io watcher that sets the relevant bits in fds[n] and increment got_events
3231    
3232     if (timeout >= 0)
3233     // create/start timer
3234    
3235     // poll
3236 root 1.310 ev_run (EV_A_ 0);
3237 root 1.76
3238 root 1.164 // stop timer again
3239     if (timeout >= 0)
3240     ev_timer_stop (EV_A_ &to);
3241    
3242     // stop io watchers again - their callbacks should have set
3243     for (n = 0; n < nfds; ++n)
3244     ev_io_stop (EV_A_ iow [n]);
3245    
3246     return got_events;
3247     }
3248 root 1.76
3249 root 1.34
3250 root 1.42 =head2 C<ev_embed> - when one backend isn't enough...
3251 root 1.35
3252     This is a rather advanced watcher type that lets you embed one event loop
3253 root 1.36 into another (currently only C<ev_io> events are supported in the embedded
3254     loop, other types of watchers might be handled in a delayed or incorrect
3255 root 1.100 fashion and must not be used).
3256 root 1.35
3257     There are primarily two reasons you would want that: work around bugs and
3258     prioritise I/O.
3259    
3260     As an example for a bug workaround, the kqueue backend might only support
3261     sockets on some platform, so it is unusable as generic backend, but you
3262     still want to make use of it because you have many sockets and it scales
3263 root 1.183 so nicely. In this case, you would create a kqueue-based loop and embed
3264     it into your default loop (which might use e.g. poll). Overall operation
3265     will be a bit slower because first libev has to call C<poll> and then
3266     C<kevent>, but at least you can use both mechanisms for what they are
3267     best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
3268    
3269     As for prioritising I/O: under rare circumstances you have the case where
3270     some fds have to be watched and handled very quickly (with low latency),
3271     and even priorities and idle watchers might have too much overhead. In
3272     this case you would put all the high priority stuff in one loop and all
3273     the rest in a second one, and embed the second one in the first.
3274 root 1.35
3275 root 1.223 As long as the watcher is active, the callback will be invoked every
3276     time there might be events pending in the embedded loop. The callback
3277     must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
3278     sweep and invoke their callbacks (the callback doesn't need to invoke the
3279     C<ev_embed_sweep> function directly, it could also start an idle watcher
3280     to give the embedded loop strictly lower priority for example).
3281    
3282     You can also set the callback to C<0>, in which case the embed watcher
3283     will automatically execute the embedded loop sweep whenever necessary.
3284    
3285     Fork detection will be handled transparently while the C<ev_embed> watcher
3286     is active, i.e., the embedded loop will automatically be forked when the
3287     embedding loop forks. In other cases, the user is responsible for calling
3288     C<ev_loop_fork> on the embedded loop.
3289 root 1.35
3290 root 1.184 Unfortunately, not all backends are embeddable: only the ones returned by
3291 root 1.35 C<ev_embeddable_backends> are, which, unfortunately, does not include any
3292     portable one.
3293    
3294     So when you want to use this feature you will always have to be prepared
3295     that you cannot get an embeddable loop. The recommended way to get around
3296     this is to have a separate variables for your embeddable loop, try to
3297 root 1.111 create it, and if that fails, use the normal loop for everything.
3298 root 1.35
3299 root 1.187 =head3 C<ev_embed> and fork
3300    
3301     While the C<ev_embed> watcher is running, forks in the embedding loop will
3302     automatically be applied to the embedded loop as well, so no special
3303     fork handling is required in that case. When the watcher is not running,
3304     however, it is still the task of the libev user to call C<ev_loop_fork ()>
3305     as applicable.
3306    
3307 root 1.82 =head3 Watcher-Specific Functions and Data Members
3308    
3309 root 1.35 =over 4
3310    
3311 root 1.36 =item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
3312    
3313 root 1.424 =item ev_embed_set (ev_embed *, struct ev_loop *embedded_loop)
3314 root 1.36
3315     Configures the watcher to embed the given loop, which must be
3316     embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
3317     invoked automatically, otherwise it is the responsibility of the callback
3318     to invoke it (it will continue to be called until the sweep has been done,
3319 root 1.161 if you do not want that, you need to temporarily stop the embed watcher).
3320 root 1.35
3321 root 1.36 =item ev_embed_sweep (loop, ev_embed *)
3322 root 1.35
3323 root 1.36 Make a single, non-blocking sweep over the embedded loop. This works
3324 root 1.310 similarly to C<ev_run (embedded_loop, EVRUN_NOWAIT)>, but in the most
3325 root 1.161 appropriate way for embedded loops.
3326 root 1.35
3327 root 1.91 =item struct ev_loop *other [read-only]
3328 root 1.48
3329     The embedded event loop.
3330    
3331 root 1.35 =back
3332    
3333 root 1.111 =head3 Examples
3334    
3335     Example: Try to get an embeddable event loop and embed it into the default
3336     event loop. If that is not possible, use the default loop. The default
3337 root 1.161 loop is stored in C<loop_hi>, while the embeddable loop is stored in
3338     C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
3339 root 1.111 used).
3340    
3341 root 1.164 struct ev_loop *loop_hi = ev_default_init (0);
3342     struct ev_loop *loop_lo = 0;
3343 root 1.198 ev_embed embed;
3344 root 1.432
3345 root 1.164 // see if there is a chance of getting one that works
3346     // (remember that a flags value of 0 means autodetection)
3347     loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
3348     ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
3349     : 0;
3350    
3351     // if we got one, then embed it, otherwise default to loop_hi
3352     if (loop_lo)
3353     {
3354     ev_embed_init (&embed, 0, loop_lo);
3355     ev_embed_start (loop_hi, &embed);
3356     }
3357     else
3358     loop_lo = loop_hi;
3359 root 1.111
3360     Example: Check if kqueue is available but not recommended and create
3361     a kqueue backend for use with sockets (which usually work with any
3362     kqueue implementation). Store the kqueue/socket-only event loop in
3363     C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
3364    
3365 root 1.164 struct ev_loop *loop = ev_default_init (0);
3366     struct ev_loop *loop_socket = 0;
3367 root 1.198 ev_embed embed;
3368 root 1.432
3369 root 1.164 if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
3370     if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
3371     {
3372     ev_embed_init (&embed, 0, loop_socket);
3373     ev_embed_start (loop, &embed);
3374     }
3375 root 1.111
3376 root 1.164 if (!loop_socket)
3377     loop_socket = loop;
3378 root 1.111
3379 root 1.164 // now use loop_socket for all sockets, and loop for everything else
3380 root 1.111
3381 root 1.35
3382 root 1.50 =head2 C<ev_fork> - the audacity to resume the event loop after a fork
3383    
3384     Fork watchers are called when a C<fork ()> was detected (usually because
3385     whoever is a good citizen cared to tell libev about it by calling
3386 root 1.421 C<ev_loop_fork>). The invocation is done before the event loop blocks next
3387     and before C<ev_check> watchers are being called, and only in the child
3388     after the fork. If whoever good citizen calling C<ev_default_fork> cheats
3389     and calls it in the wrong process, the fork handlers will be invoked, too,
3390     of course.
3391 root 1.50
3392 root 1.238 =head3 The special problem of life after fork - how is it possible?
3393    
3394 root 1.433 Most uses of C<fork ()> consist of forking, then some simple calls to set
3395 root 1.238 up/change the process environment, followed by a call to C<exec()>. This
3396     sequence should be handled by libev without any problems.
3397    
3398     This changes when the application actually wants to do event handling
3399     in the child, or both parent in child, in effect "continuing" after the
3400     fork.
3401    
3402     The default mode of operation (for libev, with application help to detect
3403     forks) is to duplicate all the state in the child, as would be expected
3404     when I<either> the parent I<or> the child process continues.
3405    
3406     When both processes want to continue using libev, then this is usually the
3407     wrong result. In that case, usually one process (typically the parent) is
3408     supposed to continue with all watchers in place as before, while the other
3409     process typically wants to start fresh, i.e. without any active watchers.
3410    
3411     The cleanest and most efficient way to achieve that with libev is to
3412     simply create a new event loop, which of course will be "empty", and
3413     use that for new watchers. This has the advantage of not touching more
3414     memory than necessary, and thus avoiding the copy-on-write, and the
3415     disadvantage of having to use multiple event loops (which do not support
3416     signal watchers).
3417    
3418     When this is not possible, or you want to use the default loop for
3419     other reasons, then in the process that wants to start "fresh", call
3420 root 1.322 C<ev_loop_destroy (EV_DEFAULT)> followed by C<ev_default_loop (...)>.
3421     Destroying the default loop will "orphan" (not stop) all registered
3422     watchers, so you have to be careful not to execute code that modifies
3423     those watchers. Note also that in that case, you have to re-register any
3424     signal watchers.
3425 root 1.238
3426 root 1.83 =head3 Watcher-Specific Functions and Data Members
3427    
3428 root 1.50 =over 4
3429    
3430 root 1.325 =item ev_fork_init (ev_fork *, callback)
3431 root 1.50
3432     Initialises and configures the fork watcher - it has no parameters of any
3433     kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
3434 root 1.329 really.
3435 root 1.50
3436     =back
3437    
3438    
3439 root 1.324 =head2 C<ev_cleanup> - even the best things end
3440    
3441 root 1.328 Cleanup watchers are called just before the event loop is being destroyed
3442     by a call to C<ev_loop_destroy>.
3443 root 1.324
3444     While there is no guarantee that the event loop gets destroyed, cleanup
3445 root 1.326 watchers provide a convenient method to install cleanup hooks for your
3446 root 1.324 program, worker threads and so on - you just to make sure to destroy the
3447     loop when you want them to be invoked.
3448    
3449 root 1.327 Cleanup watchers are invoked in the same way as any other watcher. Unlike
3450     all other watchers, they do not keep a reference to the event loop (which
3451     makes a lot of sense if you think about it). Like all other watchers, you
3452     can call libev functions in the callback, except C<ev_cleanup_start>.
3453    
3454 root 1.324 =head3 Watcher-Specific Functions and Data Members
3455    
3456     =over 4
3457    
3458 root 1.325 =item ev_cleanup_init (ev_cleanup *, callback)
3459 root 1.324
3460     Initialises and configures the cleanup watcher - it has no parameters of
3461     any kind. There is a C<ev_cleanup_set> macro, but using it is utterly
3462 root 1.329 pointless, I assure you.
3463 root 1.324
3464     =back
3465    
3466     Example: Register an atexit handler to destroy the default loop, so any
3467     cleanup functions are called.
3468    
3469     static void
3470     program_exits (void)
3471     {
3472     ev_loop_destroy (EV_DEFAULT_UC);
3473     }
3474    
3475     ...
3476     atexit (program_exits);
3477    
3478    
3479 root 1.302 =head2 C<ev_async> - how to wake up an event loop
3480 root 1.122
3481 root 1.363 In general, you cannot use an C<ev_loop> from multiple threads or other
3482 root 1.122 asynchronous sources such as signal handlers (as opposed to multiple event
3483     loops - those are of course safe to use in different threads).
3484    
3485 root 1.302 Sometimes, however, you need to wake up an event loop you do not control,
3486     for example because it belongs to another thread. This is what C<ev_async>
3487     watchers do: as long as the C<ev_async> watcher is active, you can signal
3488     it by calling C<ev_async_send>, which is thread- and signal safe.
3489 root 1.122
3490     This functionality is very similar to C<ev_signal> watchers, as signals,
3491     too, are asynchronous in nature, and signals, too, will be compressed
3492     (i.e. the number of callback invocations may be less than the number of
3493 root 1.404 C<ev_async_send> calls). In fact, you could use signal watchers as a kind
3494 root 1.349 of "global async watchers" by using a watcher on an otherwise unused
3495     signal, and C<ev_feed_signal> to signal this watcher from another thread,
3496     even without knowing which loop owns the signal.
3497 root 1.122
3498 root 1.124 =head3 Queueing
3499    
3500     C<ev_async> does not support queueing of data in any way. The reason
3501     is that the author does not know of a simple (or any) algorithm for a
3502     multiple-writer-single-reader queue that works in all cases and doesn't
3503 root 1.274 need elaborate support such as pthreads or unportable memory access
3504     semantics.
3505 root 1.124
3506     That means that if you want to queue data, you have to provide your own
3507 root 1.184 queue. But at least I can tell you how to implement locking around your
3508 root 1.130 queue:
3509 root 1.124
3510     =over 4
3511    
3512     =item queueing from a signal handler context
3513    
3514     To implement race-free queueing, you simply add to the queue in the signal
3515 root 1.191 handler but you block the signal handler in the watcher callback. Here is
3516     an example that does that for some fictitious SIGUSR1 handler:
3517 root 1.124
3518     static ev_async mysig;
3519    
3520     static void
3521     sigusr1_handler (void)
3522     {
3523     sometype data;
3524    
3525     // no locking etc.
3526     queue_put (data);
3527 root 1.133 ev_async_send (EV_DEFAULT_ &mysig);
3528 root 1.124 }
3529    
3530     static void
3531     mysig_cb (EV_P_ ev_async *w, int revents)
3532     {
3533     sometype data;
3534     sigset_t block, prev;
3535    
3536     sigemptyset (&block);
3537     sigaddset (&block, SIGUSR1);
3538     sigprocmask (SIG_BLOCK, &block, &prev);
3539    
3540     while (queue_get (&data))
3541     process (data);
3542    
3543     if (sigismember (&prev, SIGUSR1)
3544     sigprocmask (SIG_UNBLOCK, &block, 0);
3545     }
3546    
3547     (Note: pthreads in theory requires you to use C<pthread_setmask>
3548     instead of C<sigprocmask> when you use threads, but libev doesn't do it
3549     either...).
3550    
3551     =item queueing from a thread context
3552    
3553     The strategy for threads is different, as you cannot (easily) block
3554     threads but you can easily preempt them, so to queue safely you need to
3555 root 1.130 employ a traditional mutex lock, such as in this pthread example:
3556 root 1.124
3557     static ev_async mysig;
3558     static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
3559    
3560     static void
3561     otherthread (void)
3562     {
3563     // only need to lock the actual queueing operation
3564     pthread_mutex_lock (&mymutex);
3565     queue_put (data);
3566     pthread_mutex_unlock (&mymutex);
3567    
3568 root 1.133 ev_async_send (EV_DEFAULT_ &mysig);
3569 root 1.124 }
3570    
3571     static void
3572     mysig_cb (EV_P_ ev_async *w, int revents)
3573     {
3574     pthread_mutex_lock (&mymutex);
3575    
3576     while (queue_get (&data))
3577     process (data);
3578    
3579     pthread_mutex_unlock (&mymutex);
3580     }
3581    
3582     =back
3583    
3584    
3585 root 1.122 =head3 Watcher-Specific Functions and Data Members
3586    
3587     =over 4
3588    
3589     =item ev_async_init (ev_async *, callback)
3590    
3591     Initialises and configures the async watcher - it has no parameters of any
3592 root 1.208 kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
3593 root 1.184 trust me.
3594 root 1.122
3595     =item ev_async_send (loop, ev_async *)
3596    
3597     Sends/signals/activates the given C<ev_async> watcher, that is, feeds
3598 sf-exg 1.364 an C<EV_ASYNC> event on the watcher into the event loop, and instantly
3599 root 1.363 returns.
3600    
3601     Unlike C<ev_feed_event>, this call is safe to do from other threads,
3602     signal or similar contexts (see the discussion of C<EV_ATOMIC_T> in the
3603     embedding section below on what exactly this means).
3604 root 1.122
3605 root 1.227 Note that, as with other watchers in libev, multiple events might get
3606 root 1.375 compressed into a single callback invocation (another way to look at
3607     this is that C<ev_async> watchers are level-triggered: they are set on
3608     C<ev_async_send>, reset when the event loop detects that).
3609    
3610     This call incurs the overhead of at most one extra system call per event
3611     loop iteration, if the event loop is blocked, and no syscall at all if
3612     the event loop (or your program) is processing events. That means that
3613     repeated calls are basically free (there is no need to avoid calls for
3614     performance reasons) and that the overhead becomes smaller (typically
3615     zero) under load.
3616 root 1.122
3617 root 1.140 =item bool = ev_async_pending (ev_async *)
3618    
3619     Returns a non-zero value when C<ev_async_send> has been called on the
3620     watcher but the event has not yet been processed (or even noted) by the
3621     event loop.
3622    
3623     C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
3624     the loop iterates next and checks for the watcher to have become active,
3625     it will reset the flag again. C<ev_async_pending> can be used to very
3626 root 1.161 quickly check whether invoking the loop might be a good idea.
3627 root 1.140
3628 root 1.227 Not that this does I<not> check whether the watcher itself is pending,
3629     only whether it has been requested to make this watcher pending: there
3630     is a time window between the event loop checking and resetting the async
3631     notification, and the callback being invoked.
3632 root 1.140
3633 root 1.122 =back
3634    
3635    
3636 root 1.1 =head1 OTHER FUNCTIONS
3637    
3638 root 1.14 There are some other functions of possible interest. Described. Here. Now.
3639 root 1.1
3640     =over 4
3641    
3642 root 1.442 =item ev_once (loop, int fd, int events, ev_tstamp timeout, callback, arg)
3643 root 1.1
3644     This function combines a simple timer and an I/O watcher, calls your
3645 root 1.192 callback on whichever event happens first and automatically stops both
3646 root 1.1 watchers. This is useful if you want to wait for a single event on an fd
3647 root 1.22 or timeout without having to allocate/configure/start/stop/free one or
3648 root 1.1 more watchers yourself.
3649    
3650 root 1.192 If C<fd> is less than 0, then no I/O watcher will be started and the
3651     C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
3652     the given C<fd> and C<events> set will be created and started.
3653 root 1.1
3654     If C<timeout> is less than 0, then no timeout watcher will be
3655 root 1.14 started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
3656 root 1.193 repeat = 0) will be started. C<0> is a valid timeout.
3657 root 1.14
3658 root 1.289 The callback has the type C<void (*cb)(int revents, void *arg)> and is
3659 root 1.21 passed an C<revents> set like normal event callbacks (a combination of
3660 root 1.289 C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMER>) and the C<arg>
3661 root 1.193 value passed to C<ev_once>. Note that it is possible to receive I<both>
3662     a timeout and an io event at the same time - you probably should give io
3663     events precedence.
3664    
3665     Example: wait up to ten seconds for data to appear on STDIN_FILENO.
3666 root 1.1
3667 root 1.164 static void stdin_ready (int revents, void *arg)
3668     {
3669 root 1.193 if (revents & EV_READ)
3670     /* stdin might have data for us, joy! */;
3671 root 1.289 else if (revents & EV_TIMER)
3672 root 1.164 /* doh, nothing entered */;
3673     }
3674 root 1.1
3675 root 1.164 ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
3676 root 1.1
3677 root 1.274 =item ev_feed_fd_event (loop, int fd, int revents)
3678 root 1.1
3679 root 1.14 Feed an event on the given fd, as if a file descriptor backend detected
3680 root 1.386 the given events.
3681 root 1.1
3682 root 1.274 =item ev_feed_signal_event (loop, int signum)
3683 root 1.1
3684 root 1.349 Feed an event as if the given signal occurred. See also C<ev_feed_signal>,
3685     which is async-safe.
3686 root 1.1
3687     =back
3688    
3689 root 1.34
3690 root 1.345 =head1 COMMON OR USEFUL IDIOMS (OR BOTH)
3691    
3692     This section explains some common idioms that are not immediately
3693     obvious. Note that examples are sprinkled over the whole manual, and this
3694     section only contains stuff that wouldn't fit anywhere else.
3695    
3696 root 1.357 =head2 ASSOCIATING CUSTOM DATA WITH A WATCHER
3697    
3698     Each watcher has, by default, a C<void *data> member that you can read
3699     or modify at any time: libev will completely ignore it. This can be used
3700     to associate arbitrary data with your watcher. If you need more data and
3701     don't want to allocate memory separately and store a pointer to it in that
3702     data member, you can also "subclass" the watcher type and provide your own
3703     data:
3704    
3705     struct my_io
3706     {
3707     ev_io io;
3708     int otherfd;
3709     void *somedata;
3710     struct whatever *mostinteresting;
3711     };
3712    
3713     ...
3714     struct my_io w;
3715     ev_io_init (&w.io, my_cb, fd, EV_READ);
3716    
3717     And since your callback will be called with a pointer to the watcher, you
3718     can cast it back to your own type:
3719    
3720     static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
3721     {
3722     struct my_io *w = (struct my_io *)w_;
3723     ...
3724     }
3725    
3726     More interesting and less C-conformant ways of casting your callback
3727     function type instead have been omitted.
3728    
3729     =head2 BUILDING YOUR OWN COMPOSITE WATCHERS
3730    
3731     Another common scenario is to use some data structure with multiple
3732     embedded watchers, in effect creating your own watcher that combines
3733     multiple libev event sources into one "super-watcher":
3734    
3735     struct my_biggy
3736     {
3737     int some_data;
3738     ev_timer t1;
3739     ev_timer t2;
3740     }
3741    
3742     In this case getting the pointer to C<my_biggy> is a bit more
3743     complicated: Either you store the address of your C<my_biggy> struct in
3744     the C<data> member of the watcher (for woozies or C++ coders), or you need
3745     to use some pointer arithmetic using C<offsetof> inside your watchers (for
3746     real programmers):
3747    
3748     #include <stddef.h>
3749    
3750     static void
3751     t1_cb (EV_P_ ev_timer *w, int revents)
3752     {
3753     struct my_biggy big = (struct my_biggy *)
3754     (((char *)w) - offsetof (struct my_biggy, t1));
3755     }
3756    
3757     static void
3758     t2_cb (EV_P_ ev_timer *w, int revents)
3759     {
3760     struct my_biggy big = (struct my_biggy *)
3761     (((char *)w) - offsetof (struct my_biggy, t2));
3762     }
3763    
3764 root 1.387 =head2 AVOIDING FINISHING BEFORE RETURNING
3765    
3766     Often you have structures like this in event-based programs:
3767    
3768     callback ()
3769     {
3770     free (request);
3771     }
3772    
3773     request = start_new_request (..., callback);
3774    
3775     The intent is to start some "lengthy" operation. The C<request> could be
3776     used to cancel the operation, or do other things with it.
3777    
3778     It's not uncommon to have code paths in C<start_new_request> that
3779     immediately invoke the callback, for example, to report errors. Or you add
3780     some caching layer that finds that it can skip the lengthy aspects of the
3781     operation and simply invoke the callback with the result.
3782    
3783     The problem here is that this will happen I<before> C<start_new_request>
3784     has returned, so C<request> is not set.
3785    
3786     Even if you pass the request by some safer means to the callback, you
3787     might want to do something to the request after starting it, such as
3788     canceling it, which probably isn't working so well when the callback has
3789     already been invoked.
3790    
3791     A common way around all these issues is to make sure that
3792     C<start_new_request> I<always> returns before the callback is invoked. If
3793     C<start_new_request> immediately knows the result, it can artificially
3794 root 1.423 delay invoking the callback by using a C<prepare> or C<idle> watcher for
3795     example, or more sneakily, by reusing an existing (stopped) watcher and
3796     pushing it into the pending queue:
3797 root 1.387
3798     ev_set_cb (watcher, callback);
3799     ev_feed_event (EV_A_ watcher, 0);
3800    
3801     This way, C<start_new_request> can safely return before the callback is
3802     invoked, while not delaying callback invocation too much.
3803    
3804 root 1.355 =head2 MODEL/NESTED EVENT LOOP INVOCATIONS AND EXIT CONDITIONS
3805 root 1.345
3806     Often (especially in GUI toolkits) there are places where you have
3807     I<modal> interaction, which is most easily implemented by recursively
3808     invoking C<ev_run>.
3809    
3810     This brings the problem of exiting - a callback might want to finish the
3811     main C<ev_run> call, but not the nested one (e.g. user clicked "Quit", but
3812     a modal "Are you sure?" dialog is still waiting), or just the nested one
3813     and not the main one (e.g. user clocked "Ok" in a modal dialog), or some
3814 root 1.423 other combination: In these cases, a simple C<ev_break> will not work.
3815 root 1.345
3816     The solution is to maintain "break this loop" variable for each C<ev_run>
3817     invocation, and use a loop around C<ev_run> until the condition is
3818     triggered, using C<EVRUN_ONCE>:
3819    
3820     // main loop
3821     int exit_main_loop = 0;
3822    
3823     while (!exit_main_loop)
3824     ev_run (EV_DEFAULT_ EVRUN_ONCE);
3825    
3826 sf-exg 1.389 // in a modal watcher
3827 root 1.345 int exit_nested_loop = 0;
3828    
3829     while (!exit_nested_loop)
3830     ev_run (EV_A_ EVRUN_ONCE);
3831    
3832     To exit from any of these loops, just set the corresponding exit variable:
3833    
3834     // exit modal loop
3835     exit_nested_loop = 1;
3836    
3837     // exit main program, after modal loop is finished
3838     exit_main_loop = 1;
3839    
3840     // exit both
3841     exit_main_loop = exit_nested_loop = 1;
3842    
3843 root 1.355 =head2 THREAD LOCKING EXAMPLE
3844 root 1.354
3845     Here is a fictitious example of how to run an event loop in a different
3846 root 1.359 thread from where callbacks are being invoked and watchers are
3847 root 1.354 created/added/removed.
3848    
3849     For a real-world example, see the C<EV::Loop::Async> perl module,
3850     which uses exactly this technique (which is suited for many high-level
3851     languages).
3852    
3853     The example uses a pthread mutex to protect the loop data, a condition
3854     variable to wait for callback invocations, an async watcher to notify the
3855     event loop thread and an unspecified mechanism to wake up the main thread.
3856    
3857     First, you need to associate some data with the event loop:
3858    
3859     typedef struct {
3860     mutex_t lock; /* global loop lock */
3861     ev_async async_w;
3862     thread_t tid;
3863     cond_t invoke_cv;
3864     } userdata;
3865    
3866     void prepare_loop (EV_P)
3867     {
3868     // for simplicity, we use a static userdata struct.
3869     static userdata u;
3870    
3871     ev_async_init (&u->async_w, async_cb);
3872     ev_async_start (EV_A_ &u->async_w);
3873    
3874     pthread_mutex_init (&u->lock, 0);
3875     pthread_cond_init (&u->invoke_cv, 0);
3876    
3877     // now associate this with the loop
3878     ev_set_userdata (EV_A_ u);
3879     ev_set_invoke_pending_cb (EV_A_ l_invoke);
3880     ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
3881    
3882 root 1.362 // then create the thread running ev_run
3883 root 1.354 pthread_create (&u->tid, 0, l_run, EV_A);
3884     }
3885    
3886     The callback for the C<ev_async> watcher does nothing: the watcher is used
3887     solely to wake up the event loop so it takes notice of any new watchers
3888     that might have been added:
3889    
3890     static void
3891     async_cb (EV_P_ ev_async *w, int revents)
3892     {
3893     // just used for the side effects
3894     }
3895    
3896     The C<l_release> and C<l_acquire> callbacks simply unlock/lock the mutex
3897     protecting the loop data, respectively.
3898    
3899     static void
3900     l_release (EV_P)
3901     {
3902     userdata *u = ev_userdata (EV_A);
3903     pthread_mutex_unlock (&u->lock);
3904     }
3905    
3906     static void
3907     l_acquire (EV_P)
3908     {
3909     userdata *u = ev_userdata (EV_A);
3910     pthread_mutex_lock (&u->lock);
3911     }
3912    
3913     The event loop thread first acquires the mutex, and then jumps straight
3914     into C<ev_run>:
3915    
3916     void *
3917     l_run (void *thr_arg)
3918     {
3919     struct ev_loop *loop = (struct ev_loop *)thr_arg;
3920    
3921     l_acquire (EV_A);
3922     pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
3923     ev_run (EV_A_ 0);
3924     l_release (EV_A);
3925    
3926     return 0;
3927     }
3928    
3929     Instead of invoking all pending watchers, the C<l_invoke> callback will
3930     signal the main thread via some unspecified mechanism (signals? pipe
3931     writes? C<Async::Interrupt>?) and then waits until all pending watchers
3932     have been called (in a while loop because a) spurious wakeups are possible
3933     and b) skipping inter-thread-communication when there are no pending
3934     watchers is very beneficial):
3935    
3936     static void
3937     l_invoke (EV_P)
3938     {
3939     userdata *u = ev_userdata (EV_A);
3940    
3941     while (ev_pending_count (EV_A))
3942     {
3943     wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
3944     pthread_cond_wait (&u->invoke_cv, &u->lock);
3945     }
3946     }
3947    
3948     Now, whenever the main thread gets told to invoke pending watchers, it
3949     will grab the lock, call C<ev_invoke_pending> and then signal the loop
3950     thread to continue:
3951    
3952     static void
3953     real_invoke_pending (EV_P)
3954     {
3955     userdata *u = ev_userdata (EV_A);
3956    
3957     pthread_mutex_lock (&u->lock);
3958     ev_invoke_pending (EV_A);
3959     pthread_cond_signal (&u->invoke_cv);
3960     pthread_mutex_unlock (&u->lock);
3961     }
3962    
3963     Whenever you want to start/stop a watcher or do other modifications to an
3964     event loop, you will now have to lock:
3965    
3966     ev_timer timeout_watcher;
3967     userdata *u = ev_userdata (EV_A);
3968    
3969     ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
3970    
3971     pthread_mutex_lock (&u->lock);
3972     ev_timer_start (EV_A_ &timeout_watcher);
3973     ev_async_send (EV_A_ &u->async_w);
3974     pthread_mutex_unlock (&u->lock);
3975    
3976     Note that sending the C<ev_async> watcher is required because otherwise
3977     an event loop currently blocking in the kernel will have no knowledge
3978     about the newly added timer. By waking up the loop it will pick up any new
3979     watchers in the next event loop iteration.
3980    
3981 root 1.357 =head2 THREADS, COROUTINES, CONTINUATIONS, QUEUES... INSTEAD OF CALLBACKS
3982    
3983     While the overhead of a callback that e.g. schedules a thread is small, it
3984     is still an overhead. If you embed libev, and your main usage is with some
3985     kind of threads or coroutines, you might want to customise libev so that
3986     doesn't need callbacks anymore.
3987    
3988     Imagine you have coroutines that you can switch to using a function
3989     C<switch_to (coro)>, that libev runs in a coroutine called C<libev_coro>
3990     and that due to some magic, the currently active coroutine is stored in a
3991     global called C<current_coro>. Then you can build your own "wait for libev
3992     event" primitive by changing C<EV_CB_DECLARE> and C<EV_CB_INVOKE> (note
3993     the differing C<;> conventions):
3994    
3995     #define EV_CB_DECLARE(type) struct my_coro *cb;
3996     #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
3997    
3998     That means instead of having a C callback function, you store the
3999     coroutine to switch to in each watcher, and instead of having libev call
4000     your callback, you instead have it switch to that coroutine.
4001    
4002     A coroutine might now wait for an event with a function called
4003     C<wait_for_event>. (the watcher needs to be started, as always, but it doesn't
4004     matter when, or whether the watcher is active or not when this function is
4005     called):
4006    
4007     void
4008     wait_for_event (ev_watcher *w)
4009     {
4010 root 1.409 ev_set_cb (w, current_coro);
4011 root 1.357 switch_to (libev_coro);
4012     }
4013    
4014     That basically suspends the coroutine inside C<wait_for_event> and
4015     continues the libev coroutine, which, when appropriate, switches back to
4016 root 1.390 this or any other coroutine.
4017 root 1.357
4018     You can do similar tricks if you have, say, threads with an event queue -
4019     instead of storing a coroutine, you store the queue object and instead of
4020     switching to a coroutine, you push the watcher onto the queue and notify
4021     any waiters.
4022    
4023 root 1.413 To embed libev, see L</EMBEDDING>, but in short, it's easiest to create two
4024 root 1.357 files, F<my_ev.h> and F<my_ev.c> that include the respective libev files:
4025    
4026     // my_ev.h
4027     #define EV_CB_DECLARE(type) struct my_coro *cb;
4028 root 1.435 #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
4029 root 1.357 #include "../libev/ev.h"
4030    
4031     // my_ev.c
4032     #define EV_H "my_ev.h"
4033     #include "../libev/ev.c"
4034    
4035     And then use F<my_ev.h> when you would normally use F<ev.h>, and compile
4036     F<my_ev.c> into your project. When properly specifying include paths, you
4037     can even use F<ev.h> as header file name directly.
4038    
4039 root 1.345
4040 root 1.20 =head1 LIBEVENT EMULATION
4041    
4042 root 1.24 Libev offers a compatibility emulation layer for libevent. It cannot
4043     emulate the internals of libevent, so here are some usage hints:
4044    
4045     =over 4
4046    
4047 root 1.345 =item * Only the libevent-1.4.1-beta API is being emulated.
4048    
4049     This was the newest libevent version available when libev was implemented,
4050 sf-exg 1.347 and is still mostly unchanged in 2010.
4051 root 1.345
4052 root 1.24 =item * Use it by including <event.h>, as usual.
4053    
4054     =item * The following members are fully supported: ev_base, ev_callback,
4055     ev_arg, ev_fd, ev_res, ev_events.
4056    
4057     =item * Avoid using ev_flags and the EVLIST_*-macros, while it is
4058     maintained by libev, it does not work exactly the same way as in libevent (consider
4059     it a private API).
4060    
4061     =item * Priorities are not currently supported. Initialising priorities
4062     will fail and all watchers will have the same priority, even though there
4063     is an ev_pri field.
4064    
4065 root 1.146 =item * In libevent, the last base created gets the signals, in libev, the
4066 root 1.341 base that registered the signal gets the signals.
4067 root 1.146
4068 root 1.24 =item * Other members are not supported.
4069    
4070     =item * The libev emulation is I<not> ABI compatible to libevent, you need
4071     to use the libev header file and library.
4072    
4073     =back
4074 root 1.20
4075     =head1 C++ SUPPORT
4076    
4077 root 1.401 =head2 C API
4078    
4079     The normal C API should work fine when used from C++: both ev.h and the
4080     libev sources can be compiled as C++. Therefore, code that uses the C API
4081     will work fine.
4082    
4083     Proper exception specifications might have to be added to callbacks passed
4084 root 1.445 to libev: exceptions may be thrown only from watcher callbacks, all other
4085     callbacks (allocator, syserr, loop acquire/release and periodic reschedule
4086     callbacks) must not throw exceptions, and might need a C<noexcept>
4087     specification. If you have code that needs to be compiled as both C and
4088     C++ you can use the C<EV_NOEXCEPT> macro for this:
4089 root 1.401
4090     static void
4091 root 1.445 fatal_error (const char *msg) EV_NOEXCEPT
4092 root 1.401 {
4093     perror (msg);
4094     abort ();
4095     }
4096    
4097     ...
4098     ev_set_syserr_cb (fatal_error);
4099    
4100     The only API functions that can currently throw exceptions are C<ev_run>,
4101 sf-exg 1.403 C<ev_invoke>, C<ev_invoke_pending> and C<ev_loop_destroy> (the latter
4102 root 1.402 because it runs cleanup watchers).
4103 root 1.401
4104     Throwing exceptions in watcher callbacks is only supported if libev itself
4105     is compiled with a C++ compiler or your C and C++ environments allow
4106     throwing exceptions through C libraries (most do).
4107    
4108     =head2 C++ API
4109    
4110 root 1.38 Libev comes with some simplistic wrapper classes for C++ that mainly allow
4111 root 1.161 you to use some convenience methods to start/stop watchers and also change
4112 root 1.38 the callback model to a model using method callbacks on objects.
4113    
4114     To use it,
4115 root 1.428
4116 root 1.164 #include <ev++.h>
4117 root 1.38
4118 root 1.71 This automatically includes F<ev.h> and puts all of its definitions (many
4119     of them macros) into the global namespace. All C++ specific things are
4120     put into the C<ev> namespace. It should support all the same embedding
4121     options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
4122    
4123 root 1.72 Care has been taken to keep the overhead low. The only data member the C++
4124     classes add (compared to plain C-style watchers) is the event loop pointer
4125     that the watcher is associated with (or no additional members at all if
4126     you disable C<EV_MULTIPLICITY> when embedding libev).
4127 root 1.71
4128 root 1.346 Currently, functions, static and non-static member functions and classes
4129     with C<operator ()> can be used as callbacks. Other types should be easy
4130     to add as long as they only need one additional pointer for context. If
4131     you need support for other types of functors please contact the author
4132     (preferably after implementing it).
4133 root 1.38
4134 root 1.397 For all this to work, your C++ compiler either has to use the same calling
4135     conventions as your C compiler (for static member functions), or you have
4136     to embed libev and compile libev itself as C++.
4137    
4138 root 1.38 Here is a list of things available in the C<ev> namespace:
4139    
4140     =over 4
4141    
4142     =item C<ev::READ>, C<ev::WRITE> etc.
4143    
4144     These are just enum values with the same values as the C<EV_READ> etc.
4145     macros from F<ev.h>.
4146    
4147     =item C<ev::tstamp>, C<ev::now>
4148    
4149     Aliases to the same types/functions as with the C<ev_> prefix.
4150    
4151     =item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
4152    
4153     For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
4154     the same name in the C<ev> namespace, with the exception of C<ev_signal>
4155     which is called C<ev::sig> to avoid clashes with the C<signal> macro
4156 root 1.391 defined by many implementations.
4157 root 1.38
4158     All of those classes have these methods:
4159    
4160     =over 4
4161    
4162 root 1.71 =item ev::TYPE::TYPE ()
4163 root 1.38
4164 root 1.274 =item ev::TYPE::TYPE (loop)
4165 root 1.38
4166     =item ev::TYPE::~TYPE
4167    
4168 root 1.71 The constructor (optionally) takes an event loop to associate the watcher
4169     with. If it is omitted, it will use C<EV_DEFAULT>.
4170    
4171     The constructor calls C<ev_init> for you, which means you have to call the
4172     C<set> method before starting it.
4173    
4174     It will not set a callback, however: You have to call the templated C<set>
4175     method to set a callback before you can start the watcher.
4176    
4177     (The reason why you have to use a method is a limitation in C++ which does
4178     not allow explicit template arguments for constructors).
4179 root 1.38
4180     The destructor automatically stops the watcher if it is active.
4181    
4182 root 1.71 =item w->set<class, &class::method> (object *)
4183    
4184     This method sets the callback method to call. The method has to have a
4185     signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
4186     first argument and the C<revents> as second. The object must be given as
4187     parameter and is stored in the C<data> member of the watcher.
4188    
4189     This method synthesizes efficient thunking code to call your method from
4190     the C callback that libev requires. If your compiler can inline your
4191     callback (i.e. it is visible to it at the place of the C<set> call and
4192     your compiler is good :), then the method will be fully inlined into the
4193     thunking function, making it as fast as a direct C callback.
4194    
4195     Example: simple class declaration and watcher initialisation
4196    
4197 root 1.164 struct myclass
4198     {
4199     void io_cb (ev::io &w, int revents) { }
4200     }
4201    
4202     myclass obj;
4203     ev::io iow;
4204     iow.set <myclass, &myclass::io_cb> (&obj);
4205 root 1.71
4206 root 1.221 =item w->set (object *)
4207    
4208     This is a variation of a method callback - leaving out the method to call
4209     will default the method to C<operator ()>, which makes it possible to use
4210     functor objects without having to manually specify the C<operator ()> all
4211     the time. Incidentally, you can then also leave out the template argument
4212     list.
4213    
4214     The C<operator ()> method prototype must be C<void operator ()(watcher &w,
4215     int revents)>.
4216    
4217     See the method-C<set> above for more details.
4218    
4219     Example: use a functor object as callback.
4220    
4221     struct myfunctor
4222     {
4223     void operator() (ev::io &w, int revents)
4224     {
4225     ...
4226     }
4227     }
4228 root 1.432
4229 root 1.221 myfunctor f;
4230    
4231     ev::io w;
4232     w.set (&f);
4233    
4234 root 1.75 =item w->set<function> (void *data = 0)
4235 root 1.71
4236     Also sets a callback, but uses a static method or plain function as
4237     callback. The optional C<data> argument will be stored in the watcher's
4238     C<data> member and is free for you to use.
4239    
4240 root 1.75 The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
4241    
4242 root 1.71 See the method-C<set> above for more details.
4243    
4244 root 1.184 Example: Use a plain function as callback.
4245 root 1.75
4246 root 1.164 static void io_cb (ev::io &w, int revents) { }
4247     iow.set <io_cb> ();
4248 root 1.75
4249 root 1.274 =item w->set (loop)
4250 root 1.38
4251     Associates a different C<struct ev_loop> with this watcher. You can only
4252     do this when the watcher is inactive (and not pending either).
4253    
4254 root 1.161 =item w->set ([arguments])
4255 root 1.38
4256 root 1.419 Basically the same as C<ev_TYPE_set> (except for C<ev::embed> watchers>),
4257     with the same arguments. Either this method or a suitable start method
4258     must be called at least once. Unlike the C counterpart, an active watcher
4259     gets automatically stopped and restarted when reconfiguring it with this
4260     method.
4261    
4262     For C<ev::embed> watchers this method is called C<set_embed>, to avoid
4263     clashing with the C<set (loop)> method.
4264 root 1.38
4265 root 1.461 For C<ev::io> watchers there is an additional C<set> method that acepts a
4266     new event mask only, and internally calls C<ev_io_modfify>.
4267    
4268 root 1.38 =item w->start ()
4269    
4270 root 1.71 Starts the watcher. Note that there is no C<loop> argument, as the
4271     constructor already stores the event loop.
4272 root 1.38
4273 root 1.307 =item w->start ([arguments])
4274    
4275     Instead of calling C<set> and C<start> methods separately, it is often
4276     convenient to wrap them in one call. Uses the same type of arguments as
4277     the configure C<set> method of the watcher.
4278    
4279 root 1.38 =item w->stop ()
4280    
4281     Stops the watcher if it is active. Again, no C<loop> argument.
4282    
4283 root 1.84 =item w->again () (C<ev::timer>, C<ev::periodic> only)
4284 root 1.38
4285     For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
4286     C<ev_TYPE_again> function.
4287    
4288 root 1.84 =item w->sweep () (C<ev::embed> only)
4289 root 1.38
4290     Invokes C<ev_embed_sweep>.
4291    
4292 root 1.84 =item w->update () (C<ev::stat> only)
4293 root 1.49
4294     Invokes C<ev_stat_stat>.
4295    
4296 root 1.38 =back
4297    
4298     =back
4299    
4300 root 1.307 Example: Define a class with two I/O and idle watchers, start the I/O
4301     watchers in the constructor.
4302 root 1.38
4303 root 1.164 class myclass
4304     {
4305 root 1.184 ev::io io ; void io_cb (ev::io &w, int revents);
4306 root 1.377 ev::io io2 ; void io2_cb (ev::io &w, int revents);
4307 root 1.184 ev::idle idle; void idle_cb (ev::idle &w, int revents);
4308 root 1.164
4309     myclass (int fd)
4310     {
4311     io .set <myclass, &myclass::io_cb > (this);
4312 root 1.307 io2 .set <myclass, &myclass::io2_cb > (this);
4313 root 1.164 idle.set <myclass, &myclass::idle_cb> (this);
4314    
4315 root 1.307 io.set (fd, ev::WRITE); // configure the watcher
4316     io.start (); // start it whenever convenient
4317    
4318     io2.start (fd, ev::READ); // set + start in one call
4319 root 1.164 }
4320     };
4321 root 1.20
4322 root 1.50
4323 root 1.136 =head1 OTHER LANGUAGE BINDINGS
4324    
4325     Libev does not offer other language bindings itself, but bindings for a
4326 root 1.161 number of languages exist in the form of third-party packages. If you know
4327 root 1.136 any interesting language binding in addition to the ones listed here, drop
4328     me a note.
4329    
4330     =over 4
4331    
4332     =item Perl
4333    
4334     The EV module implements the full libev API and is actually used to test
4335     libev. EV is developed together with libev. Apart from the EV core module,
4336     there are additional modules that implement libev-compatible interfaces
4337 root 1.184 to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
4338     C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
4339     and C<EV::Glib>).
4340 root 1.136
4341 root 1.166 It can be found and installed via CPAN, its homepage is at
4342 root 1.136 L<http://software.schmorp.de/pkg/EV>.
4343    
4344 root 1.166 =item Python
4345    
4346     Python bindings can be found at L<http://code.google.com/p/pyev/>. It
4347 root 1.228 seems to be quite complete and well-documented.
4348 root 1.166
4349 root 1.136 =item Ruby
4350    
4351     Tony Arcieri has written a ruby extension that offers access to a subset
4352 root 1.161 of the libev API and adds file handle abstractions, asynchronous DNS and
4353 root 1.136 more on top of it. It can be found via gem servers. Its homepage is at
4354     L<http://rev.rubyforge.org/>.
4355    
4356 root 1.218 Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
4357     makes rev work even on mingw.
4358    
4359 root 1.228 =item Haskell
4360    
4361     A haskell binding to libev is available at
4362     L<http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
4363    
4364 root 1.136 =item D
4365    
4366     Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
4367 sf-exg 1.378 be found at L<http://www.llucax.com.ar/proj/ev.d/index.html>.
4368 root 1.136
4369 root 1.201 =item Ocaml
4370    
4371     Erkki Seppala has written Ocaml bindings for libev, to be found at
4372     L<http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
4373    
4374 root 1.263 =item Lua
4375    
4376 root 1.279 Brian Maher has written a partial interface to libev for lua (at the
4377     time of this writing, only C<ev_io> and C<ev_timer>), to be found at
4378 root 1.263 L<http://github.com/brimworks/lua-ev>.
4379    
4380 root 1.416 =item Javascript
4381    
4382     Node.js (L<http://nodejs.org>) uses libev as the underlying event library.
4383    
4384     =item Others
4385    
4386     There are others, and I stopped counting.
4387    
4388 root 1.136 =back
4389    
4390    
4391 root 1.50 =head1 MACRO MAGIC
4392    
4393 root 1.161 Libev can be compiled with a variety of options, the most fundamental
4394 root 1.84 of which is C<EV_MULTIPLICITY>. This option determines whether (most)
4395     functions and callbacks have an initial C<struct ev_loop *> argument.
4396 root 1.50
4397     To make it easier to write programs that cope with either variant, the
4398     following macros are defined:
4399    
4400     =over 4
4401    
4402     =item C<EV_A>, C<EV_A_>
4403    
4404     This provides the loop I<argument> for functions, if one is required ("ev
4405     loop argument"). The C<EV_A> form is used when this is the sole argument,
4406     C<EV_A_> is used when other arguments are following. Example:
4407    
4408 root 1.164 ev_unref (EV_A);
4409     ev_timer_add (EV_A_ watcher);
4410 root 1.310 ev_run (EV_A_ 0);
4411 root 1.50
4412     It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
4413     which is often provided by the following macro.
4414    
4415     =item C<EV_P>, C<EV_P_>
4416    
4417     This provides the loop I<parameter> for functions, if one is required ("ev
4418     loop parameter"). The C<EV_P> form is used when this is the sole parameter,
4419     C<EV_P_> is used when other parameters are following. Example:
4420    
4421 root 1.164 // this is how ev_unref is being declared
4422     static void ev_unref (EV_P);
4423 root 1.50
4424 root 1.164 // this is how you can declare your typical callback
4425     static void cb (EV_P_ ev_timer *w, int revents)
4426 root 1.50
4427     It declares a parameter C<loop> of type C<struct ev_loop *>, quite
4428     suitable for use with C<EV_A>.
4429    
4430     =item C<EV_DEFAULT>, C<EV_DEFAULT_>
4431    
4432     Similar to the other two macros, this gives you the value of the default
4433 root 1.380 loop, if multiple loops are supported ("ev loop default"). The default loop
4434     will be initialised if it isn't already initialised.
4435    
4436     For non-multiplicity builds, these macros do nothing, so you always have
4437     to initialise the loop somewhere.
4438 root 1.50
4439 root 1.143 =item C<EV_DEFAULT_UC>, C<EV_DEFAULT_UC_>
4440    
4441     Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
4442     default loop has been initialised (C<UC> == unchecked). Their behaviour
4443     is undefined when the default loop has not been initialised by a previous
4444     execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
4445    
4446     It is often prudent to use C<EV_DEFAULT> when initialising the first
4447     watcher in a function but use C<EV_DEFAULT_UC> afterwards.
4448    
4449 root 1.50 =back
4450    
4451 root 1.63 Example: Declare and initialise a check watcher, utilising the above
4452 root 1.68 macros so it will work regardless of whether multiple loops are supported
4453 root 1.63 or not.
4454 root 1.50
4455 root 1.164 static void
4456     check_cb (EV_P_ ev_timer *w, int revents)
4457     {
4458     ev_check_stop (EV_A_ w);
4459     }
4460    
4461     ev_check check;
4462     ev_check_init (&check, check_cb);
4463     ev_check_start (EV_DEFAULT_ &check);
4464 root 1.310 ev_run (EV_DEFAULT_ 0);
4465 root 1.50
4466 root 1.39 =head1 EMBEDDING
4467    
4468     Libev can (and often is) directly embedded into host
4469     applications. Examples of applications that embed it include the Deliantra
4470     Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
4471     and rxvt-unicode.
4472    
4473 root 1.91 The goal is to enable you to just copy the necessary files into your
4474 root 1.39 source directory without having to change even a single line in them, so
4475     you can easily upgrade by simply copying (or having a checked-out copy of
4476     libev somewhere in your source tree).
4477    
4478     =head2 FILESETS
4479    
4480     Depending on what features you need you need to include one or more sets of files
4481 root 1.161 in your application.
4482 root 1.39
4483     =head3 CORE EVENT LOOP
4484    
4485     To include only the libev core (all the C<ev_*> functions), with manual
4486     configuration (no autoconf):
4487    
4488 root 1.164 #define EV_STANDALONE 1
4489     #include "ev.c"
4490 root 1.39
4491     This will automatically include F<ev.h>, too, and should be done in a
4492     single C source file only to provide the function implementations. To use
4493     it, do the same for F<ev.h> in all files wishing to use this API (best
4494     done by writing a wrapper around F<ev.h> that you can include instead and
4495     where you can put other configuration options):
4496    
4497 root 1.164 #define EV_STANDALONE 1
4498     #include "ev.h"
4499 root 1.39
4500     Both header files and implementation files can be compiled with a C++
4501 root 1.208 compiler (at least, that's a stated goal, and breakage will be treated
4502 root 1.39 as a bug).
4503    
4504     You need the following files in your source tree, or in a directory
4505     in your include path (e.g. in libev/ when using -Ilibev):
4506    
4507 root 1.164 ev.h
4508     ev.c
4509     ev_vars.h
4510     ev_wrap.h
4511    
4512     ev_win32.c required on win32 platforms only
4513    
4514 root 1.440 ev_select.c only when select backend is enabled
4515     ev_poll.c only when poll backend is enabled
4516     ev_epoll.c only when the epoll backend is enabled
4517 root 1.447 ev_linuxaio.c only when the linux aio backend is enabled
4518 root 1.456 ev_iouring.c only when the linux io_uring backend is enabled
4519 root 1.440 ev_kqueue.c only when the kqueue backend is enabled
4520     ev_port.c only when the solaris port backend is enabled
4521 root 1.39
4522     F<ev.c> includes the backend files directly when enabled, so you only need
4523 root 1.43 to compile this single file.
4524 root 1.39
4525     =head3 LIBEVENT COMPATIBILITY API
4526    
4527     To include the libevent compatibility API, also include:
4528    
4529 root 1.164 #include "event.c"
4530 root 1.39
4531     in the file including F<ev.c>, and:
4532    
4533 root 1.164 #include "event.h"
4534 root 1.39
4535     in the files that want to use the libevent API. This also includes F<ev.h>.
4536    
4537     You need the following additional files for this:
4538    
4539 root 1.164 event.h
4540     event.c
4541 root 1.39
4542     =head3 AUTOCONF SUPPORT
4543    
4544 root 1.161 Instead of using C<EV_STANDALONE=1> and providing your configuration in
4545 root 1.39 whatever way you want, you can also C<m4_include([libev.m4])> in your
4546 root 1.43 F<configure.ac> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
4547     include F<config.h> and configure itself accordingly.
4548 root 1.39
4549     For this of course you need the m4 file:
4550    
4551 root 1.164 libev.m4
4552 root 1.39
4553     =head2 PREPROCESSOR SYMBOLS/MACROS
4554    
4555 root 1.142 Libev can be configured via a variety of preprocessor symbols you have to
4556 root 1.281 define before including (or compiling) any of its files. The default in
4557     the absence of autoconf is documented for every option.
4558    
4559     Symbols marked with "(h)" do not change the ABI, and can have different
4560     values when compiling libev vs. including F<ev.h>, so it is permissible
4561 sf-exg 1.292 to redefine them before including F<ev.h> without breaking compatibility
4562 root 1.281 to a compiled library. All other symbols change the ABI, which means all
4563     users of libev and the libev code itself must be compiled with compatible
4564     settings.
4565 root 1.39
4566     =over 4
4567    
4568 root 1.310 =item EV_COMPAT3 (h)
4569    
4570     Backwards compatibility is a major concern for libev. This is why this
4571     release of libev comes with wrappers for the functions and symbols that
4572     have been renamed between libev version 3 and 4.
4573    
4574     You can disable these wrappers (to test compatibility with future
4575     versions) by defining C<EV_COMPAT3> to C<0> when compiling your
4576     sources. This has the additional advantage that you can drop the C<struct>
4577     from C<struct ev_loop> declarations, as libev will provide an C<ev_loop>
4578     typedef in that case.
4579    
4580     In some future version, the default for C<EV_COMPAT3> will become C<0>,
4581     and in some even more future version the compatibility code will be
4582     removed completely.
4583    
4584 root 1.281 =item EV_STANDALONE (h)
4585 root 1.39
4586     Must always be C<1> if you do not use autoconf configuration, which
4587     keeps libev from including F<config.h>, and it also defines dummy
4588     implementations for some libevent functions (such as logging, which is not
4589     supported). It will also not define any of the structs usually found in
4590     F<event.h> that are not directly supported by the libev core alone.
4591    
4592 root 1.262 In standalone mode, libev will still try to automatically deduce the
4593 root 1.218 configuration, but has to be more conservative.
4594    
4595 root 1.367 =item EV_USE_FLOOR
4596    
4597     If defined to be C<1>, libev will use the C<floor ()> function for its
4598     periodic reschedule calculations, otherwise libev will fall back on a
4599     portable (slower) implementation. If you enable this, you usually have to
4600     link against libm or something equivalent. Enabling this when the C<floor>
4601     function is not available will fail, so the safe default is to not enable
4602     this.
4603    
4604 root 1.39 =item EV_USE_MONOTONIC
4605    
4606     If defined to be C<1>, libev will try to detect the availability of the
4607 root 1.218 monotonic clock option at both compile time and runtime. Otherwise no
4608     use of the monotonic clock option will be attempted. If you enable this,
4609     you usually have to link against librt or something similar. Enabling it
4610     when the functionality isn't available is safe, though, although you have
4611 root 1.39 to make sure you link against any libraries where the C<clock_gettime>
4612 root 1.218 function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
4613 root 1.39
4614     =item EV_USE_REALTIME
4615    
4616     If defined to be C<1>, libev will try to detect the availability of the
4617 root 1.224 real-time clock option at compile time (and assume its availability
4618     at runtime if successful). Otherwise no use of the real-time clock
4619     option will be attempted. This effectively replaces C<gettimeofday>
4620     by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
4621     correctness. See the note about libraries in the description of
4622     C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
4623     C<EV_USE_CLOCK_SYSCALL>.
4624 root 1.39
4625 root 1.218 =item EV_USE_CLOCK_SYSCALL
4626    
4627     If defined to be C<1>, libev will try to use a direct syscall instead
4628     of calling the system-provided C<clock_gettime> function. This option
4629     exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
4630     unconditionally pulls in C<libpthread>, slowing down single-threaded
4631 root 1.219 programs needlessly. Using a direct syscall is slightly slower (in
4632     theory), because no optimised vdso implementation can be used, but avoids
4633     the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
4634     higher, as it simplifies linking (no need for C<-lrt>).
4635 root 1.218
4636 root 1.97 =item EV_USE_NANOSLEEP
4637    
4638     If defined to be C<1>, libev will assume that C<nanosleep ()> is available
4639     and will use it for delays. Otherwise it will use C<select ()>.
4640    
4641 root 1.142 =item EV_USE_EVENTFD
4642    
4643     If defined to be C<1>, then libev will assume that C<eventfd ()> is
4644     available and will probe for kernel support at runtime. This will improve
4645     C<ev_signal> and C<ev_async> performance and reduce resource consumption.
4646     If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
4647     2.7 or newer, otherwise disabled.
4648    
4649 root 1.458 =item EV_USE_SIGNALFD
4650    
4651     If defined to be C<1>, then libev will assume that C<signalfd ()> is
4652     available and will probe for kernel support at runtime. This enables
4653     the use of EVFLAG_SIGNALFD for faster and simpler signal handling. If
4654     undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
4655     2.7 or newer, otherwise disabled.
4656    
4657     =item EV_USE_TIMERFD
4658    
4659     If defined to be C<1>, then libev will assume that C<timerfd ()> is
4660     available and will probe for kernel support at runtime. This allows
4661     libev to detect time jumps accurately. If undefined, it will be enabled
4662     if the headers indicate GNU/Linux + Glibc 2.8 or newer and define
4663     C<TFD_TIMER_CANCEL_ON_SET>, otherwise disabled.
4664    
4665     =item EV_USE_EVENTFD
4666    
4667     If defined to be C<1>, then libev will assume that C<eventfd ()> is
4668     available and will probe for kernel support at runtime. This will improve
4669     C<ev_signal> and C<ev_async> performance and reduce resource consumption.
4670     If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
4671     2.7 or newer, otherwise disabled.
4672    
4673 root 1.39 =item EV_USE_SELECT
4674    
4675     If undefined or defined to be C<1>, libev will compile in support for the
4676 root 1.161 C<select>(2) backend. No attempt at auto-detection will be done: if no
4677 root 1.39 other method takes over, select will be it. Otherwise the select backend
4678     will not be compiled in.
4679    
4680     =item EV_SELECT_USE_FD_SET
4681    
4682     If defined to C<1>, then the select backend will use the system C<fd_set>
4683     structure. This is useful if libev doesn't compile due to a missing
4684 root 1.218 C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
4685     on exotic systems. This usually limits the range of file descriptors to
4686     some low limit such as 1024 or might have other limitations (winsocket
4687     only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
4688     configures the maximum size of the C<fd_set>.
4689 root 1.39
4690     =item EV_SELECT_IS_WINSOCKET
4691    
4692     When defined to C<1>, the select backend will assume that
4693     select/socket/connect etc. don't understand file descriptors but
4694     wants osf handles on win32 (this is the case when the select to
4695     be used is the winsock select). This means that it will call
4696     C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
4697     it is assumed that all these functions actually work on fds, even
4698     on win32. Should not be defined on non-win32 platforms.
4699    
4700 root 1.264 =item EV_FD_TO_WIN32_HANDLE(fd)
4701 root 1.112
4702     If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
4703     file descriptors to socket handles. When not defining this symbol (the
4704     default), then libev will call C<_get_osfhandle>, which is usually
4705     correct. In some cases, programs use their own file descriptor management,
4706     in which case they can provide this function to map fds to socket handles.
4707    
4708 root 1.264 =item EV_WIN32_HANDLE_TO_FD(handle)
4709    
4710     If C<EV_SELECT_IS_WINSOCKET> then libev maps handles to file descriptors
4711     using the standard C<_open_osfhandle> function. For programs implementing
4712     their own fd to handle mapping, overwriting this function makes it easier
4713     to do so. This can be done by defining this macro to an appropriate value.
4714    
4715     =item EV_WIN32_CLOSE_FD(fd)
4716    
4717     If programs implement their own fd to handle mapping on win32, then this
4718     macro can be used to override the C<close> function, useful to unregister
4719     file descriptors again. Note that the replacement function has to close
4720     the underlying OS handle.
4721    
4722 root 1.417 =item EV_USE_WSASOCKET
4723    
4724     If defined to be C<1>, libev will use C<WSASocket> to create its internal
4725     communication socket, which works better in some environments. Otherwise,
4726     the normal C<socket> function will be used, which works better in other
4727 sf-exg 1.418 environments.
4728 root 1.417
4729 root 1.39 =item EV_USE_POLL
4730    
4731     If defined to be C<1>, libev will compile in support for the C<poll>(2)
4732     backend. Otherwise it will be enabled on non-win32 platforms. It
4733     takes precedence over select.
4734    
4735     =item EV_USE_EPOLL
4736    
4737     If defined to be C<1>, libev will compile in support for the Linux
4738     C<epoll>(7) backend. Its availability will be detected at runtime,
4739 root 1.142 otherwise another method will be used as fallback. This is the preferred
4740     backend for GNU/Linux systems. If undefined, it will be enabled if the
4741     headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
4742 root 1.39
4743 root 1.447 =item EV_USE_LINUXAIO
4744    
4745 root 1.456 If defined to be C<1>, libev will compile in support for the Linux aio
4746     backend (C<EV_USE_EPOLL> must also be enabled). If undefined, it will be
4747     enabled on linux, otherwise disabled.
4748    
4749     =item EV_USE_IOURING
4750    
4751 root 1.447 If defined to be C<1>, libev will compile in support for the Linux
4752 root 1.456 io_uring backend (C<EV_USE_EPOLL> must also be enabled). Due to it's
4753     current limitations it has to be requested explicitly. If undefined, it
4754     will be enabled on linux, otherwise disabled.
4755 root 1.447
4756 root 1.39 =item EV_USE_KQUEUE
4757    
4758     If defined to be C<1>, libev will compile in support for the BSD style
4759     C<kqueue>(2) backend. Its actual availability will be detected at runtime,
4760     otherwise another method will be used as fallback. This is the preferred
4761     backend for BSD and BSD-like systems, although on most BSDs kqueue only
4762     supports some types of fds correctly (the only platform we found that
4763     supports ptys for example was NetBSD), so kqueue might be compiled in, but
4764     not be used unless explicitly requested. The best way to use it is to find
4765 root 1.41 out whether kqueue supports your type of fd properly and use an embedded
4766 root 1.39 kqueue loop.
4767    
4768     =item EV_USE_PORT
4769    
4770     If defined to be C<1>, libev will compile in support for the Solaris
4771     10 port style backend. Its availability will be detected at runtime,
4772     otherwise another method will be used as fallback. This is the preferred
4773     backend for Solaris 10 systems.
4774    
4775     =item EV_USE_DEVPOLL
4776    
4777 root 1.161 Reserved for future expansion, works like the USE symbols above.
4778 root 1.39
4779 root 1.56 =item EV_USE_INOTIFY
4780    
4781     If defined to be C<1>, libev will compile in support for the Linux inotify
4782     interface to speed up C<ev_stat> watchers. Its actual availability will
4783 root 1.142 be detected at runtime. If undefined, it will be enabled if the headers
4784     indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
4785 root 1.56
4786 root 1.396 =item EV_NO_SMP
4787    
4788     If defined to be C<1>, libev will assume that memory is always coherent
4789     between threads, that is, threads can be used, but threads never run on
4790     different cpus (or different cpu cores). This reduces dependencies
4791     and makes libev faster.
4792    
4793     =item EV_NO_THREADS
4794    
4795 root 1.426 If defined to be C<1>, libev will assume that it will never be called from
4796     different threads (that includes signal handlers), which is a stronger
4797     assumption than C<EV_NO_SMP>, above. This reduces dependencies and makes
4798     libev faster.
4799 root 1.396
4800 root 1.123 =item EV_ATOMIC_T
4801    
4802     Libev requires an integer type (suitable for storing C<0> or C<1>) whose
4803 root 1.420 access is atomic with respect to other threads or signal contexts. No
4804     such type is easily found in the C language, so you can provide your own
4805     type that you know is safe for your purposes. It is used both for signal
4806     handler "locking" as well as for signal and thread safety in C<ev_async>
4807     watchers.
4808 root 1.123
4809 root 1.161 In the absence of this define, libev will use C<sig_atomic_t volatile>
4810 root 1.420 (from F<signal.h>), which is usually good enough on most platforms.
4811 root 1.123
4812 root 1.281 =item EV_H (h)
4813 root 1.39
4814     The name of the F<ev.h> header file used to include it. The default if
4815 root 1.118 undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
4816     used to virtually rename the F<ev.h> header file in case of conflicts.
4817 root 1.39
4818 root 1.281 =item EV_CONFIG_H (h)
4819 root 1.39
4820     If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
4821     F<ev.c>'s idea of where to find the F<config.h> file, similarly to
4822     C<EV_H>, above.
4823    
4824 root 1.281 =item EV_EVENT_H (h)
4825 root 1.39
4826     Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
4827 root 1.118 of how the F<event.h> header can be found, the default is C<"event.h">.
4828 root 1.39
4829 root 1.281 =item EV_PROTOTYPES (h)
4830 root 1.39
4831     If defined to be C<0>, then F<ev.h> will not define any function
4832     prototypes, but still define all the structs and other symbols. This is
4833     occasionally useful if you want to provide your own wrapper functions
4834     around libev functions.
4835    
4836     =item EV_MULTIPLICITY
4837    
4838     If undefined or defined to C<1>, then all event-loop-specific functions
4839     will have the C<struct ev_loop *> as first argument, and you can create
4840     additional independent event loops. Otherwise there will be no support
4841     for multiple event loops and there is no first event loop pointer
4842     argument. Instead, all functions act on the single default loop.
4843    
4844 root 1.380 Note that C<EV_DEFAULT> and C<EV_DEFAULT_> will no longer provide a
4845     default loop when multiplicity is switched off - you always have to
4846     initialise the loop manually in this case.
4847    
4848 root 1.69 =item EV_MINPRI
4849    
4850     =item EV_MAXPRI
4851    
4852     The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
4853     C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
4854     provide for more priorities by overriding those symbols (usually defined
4855     to be C<-2> and C<2>, respectively).
4856    
4857     When doing priority-based operations, libev usually has to linearly search
4858     all the priorities, so having many of them (hundreds) uses a lot of space
4859     and time, so using the defaults of five priorities (-2 .. +2) is usually
4860     fine.
4861    
4862 root 1.184 If your embedding application does not need any priorities, defining these
4863     both to C<0> will save some memory and CPU.
4864 root 1.69
4865 root 1.283 =item EV_PERIODIC_ENABLE, EV_IDLE_ENABLE, EV_EMBED_ENABLE, EV_STAT_ENABLE,
4866     EV_PREPARE_ENABLE, EV_CHECK_ENABLE, EV_FORK_ENABLE, EV_SIGNAL_ENABLE,
4867     EV_ASYNC_ENABLE, EV_CHILD_ENABLE.
4868    
4869     If undefined or defined to be C<1> (and the platform supports it), then
4870     the respective watcher type is supported. If defined to be C<0>, then it
4871 sf-exg 1.299 is not. Disabling watcher types mainly saves code size.
4872 root 1.282
4873 root 1.285 =item EV_FEATURES
4874 root 1.47
4875     If you need to shave off some kilobytes of code at the expense of some
4876 root 1.285 speed (but with the full API), you can define this symbol to request
4877     certain subsets of functionality. The default is to enable all features
4878     that can be enabled on the platform.
4879    
4880     A typical way to use this symbol is to define it to C<0> (or to a bitset
4881     with some broad features you want) and then selectively re-enable
4882     additional parts you want, for example if you want everything minimal,
4883     but multiple event loop support, async and child watchers and the poll
4884     backend, use this:
4885    
4886     #define EV_FEATURES 0
4887     #define EV_MULTIPLICITY 1
4888     #define EV_USE_POLL 1
4889     #define EV_CHILD_ENABLE 1
4890     #define EV_ASYNC_ENABLE 1
4891    
4892     The actual value is a bitset, it can be a combination of the following
4893 root 1.400 values (by default, all of these are enabled):
4894 root 1.285
4895     =over 4
4896    
4897     =item C<1> - faster/larger code
4898    
4899     Use larger code to speed up some operations.
4900    
4901 sf-exg 1.299 Currently this is used to override some inlining decisions (enlarging the
4902     code size by roughly 30% on amd64).
4903 root 1.285
4904 root 1.286 When optimising for size, use of compiler flags such as C<-Os> with
4905 sf-exg 1.299 gcc is recommended, as well as C<-DNDEBUG>, as libev contains a number of
4906 root 1.286 assertions.
4907 root 1.285
4908 root 1.400 The default is off when C<__OPTIMIZE_SIZE__> is defined by your compiler
4909     (e.g. gcc with C<-Os>).
4910    
4911 root 1.285 =item C<2> - faster/larger data structures
4912    
4913     Replaces the small 2-heap for timer management by a faster 4-heap, larger
4914 sf-exg 1.299 hash table sizes and so on. This will usually further increase code size
4915 root 1.285 and can additionally have an effect on the size of data structures at
4916     runtime.
4917    
4918 root 1.400 The default is off when C<__OPTIMIZE_SIZE__> is defined by your compiler
4919     (e.g. gcc with C<-Os>).
4920    
4921 root 1.285 =item C<4> - full API configuration
4922    
4923     This enables priorities (sets C<EV_MAXPRI>=2 and C<EV_MINPRI>=-2), and
4924     enables multiplicity (C<EV_MULTIPLICITY>=1).
4925    
4926 root 1.287 =item C<8> - full API
4927    
4928     This enables a lot of the "lesser used" API functions. See C<ev.h> for
4929     details on which parts of the API are still available without this
4930 root 1.285 feature, and do not complain if this subset changes over time.
4931    
4932 root 1.287 =item C<16> - enable all optional watcher types
4933 root 1.285
4934     Enables all optional watcher types. If you want to selectively enable
4935     only some watcher types other than I/O and timers (e.g. prepare,
4936     embed, async, child...) you can enable them manually by defining
4937     C<EV_watchertype_ENABLE> to C<1> instead.
4938    
4939 root 1.287 =item C<32> - enable all backends
4940 root 1.285
4941     This enables all backends - without this feature, you need to enable at
4942     least one backend manually (C<EV_USE_SELECT> is a good choice).
4943    
4944 root 1.287 =item C<64> - enable OS-specific "helper" APIs
4945 root 1.285
4946     Enable inotify, eventfd, signalfd and similar OS-specific helper APIs by
4947     default.
4948    
4949     =back
4950    
4951     Compiling with C<gcc -Os -DEV_STANDALONE -DEV_USE_EPOLL=1 -DEV_FEATURES=0>
4952 root 1.288 reduces the compiled size of libev from 24.7Kb code/2.8Kb data to 6.5Kb
4953     code/0.3Kb data on my GNU/Linux amd64 system, while still giving you I/O
4954     watchers, timers and monotonic clock support.
4955 root 1.285
4956     With an intelligent-enough linker (gcc+binutils are intelligent enough
4957     when you use C<-Wl,--gc-sections -ffunction-sections>) functions unused by
4958     your program might be left out as well - a binary starting a timer and an
4959     I/O watcher then might come out at only 5Kb.
4960 root 1.282
4961 root 1.388 =item EV_API_STATIC
4962    
4963     If this symbol is defined (by default it is not), then all identifiers
4964     will have static linkage. This means that libev will not export any
4965     identifiers, and you cannot link against libev anymore. This can be useful
4966     when you embed libev, only want to use libev functions in a single file,
4967     and do not want its identifiers to be visible.
4968    
4969     To use this, define C<EV_API_STATIC> and include F<ev.c> in the file that
4970     wants to use libev.
4971    
4972 root 1.393 This option only works when libev is compiled with a C compiler, as C++
4973     doesn't support the required declaration syntax.
4974    
4975 root 1.281 =item EV_AVOID_STDIO
4976    
4977     If this is set to C<1> at compiletime, then libev will avoid using stdio
4978 sf-exg 1.299 functions (printf, scanf, perror etc.). This will increase the code size
4979 root 1.281 somewhat, but if your program doesn't otherwise depend on stdio and your
4980     libc allows it, this avoids linking in the stdio library which is quite
4981     big.
4982    
4983     Note that error messages might become less precise when this option is
4984     enabled.
4985    
4986 root 1.260 =item EV_NSIG
4987    
4988     The highest supported signal number, +1 (or, the number of
4989     signals): Normally, libev tries to deduce the maximum number of signals
4990     automatically, but sometimes this fails, in which case it can be
4991     specified. Also, using a lower number than detected (C<32> should be
4992 sf-exg 1.298 good for about any system in existence) can save some memory, as libev
4993 root 1.260 statically allocates some 12-24 bytes per signal number.
4994    
4995 root 1.51 =item EV_PID_HASHSIZE
4996    
4997     C<ev_child> watchers use a small hash table to distribute workload by
4998 root 1.285 pid. The default size is C<16> (or C<1> with C<EV_FEATURES> disabled),
4999     usually more than enough. If you need to manage thousands of children you
5000     might want to increase this value (I<must> be a power of two).
5001 root 1.56
5002     =item EV_INOTIFY_HASHSIZE
5003    
5004 root 1.104 C<ev_stat> watchers use a small hash table to distribute workload by
5005 root 1.285 inotify watch id. The default size is C<16> (or C<1> with C<EV_FEATURES>
5006     disabled), usually more than enough. If you need to manage thousands of
5007     C<ev_stat> watchers you might want to increase this value (I<must> be a
5008     power of two).
5009 root 1.51
5010 root 1.153 =item EV_USE_4HEAP
5011    
5012     Heaps are not very cache-efficient. To improve the cache-efficiency of the
5013 root 1.184 timer and periodics heaps, libev uses a 4-heap when this symbol is defined
5014     to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
5015     faster performance with many (thousands) of watchers.
5016 root 1.153
5017 root 1.285 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
5018     will be C<0>.
5019 root 1.153
5020     =item EV_HEAP_CACHE_AT
5021    
5022     Heaps are not very cache-efficient. To improve the cache-efficiency of the
5023 root 1.184 timer and periodics heaps, libev can cache the timestamp (I<at>) within
5024 root 1.153 the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
5025     which uses 8-12 bytes more per watcher and a few hundred bytes more code,
5026 root 1.155 but avoids random read accesses on heap changes. This improves performance
5027 root 1.184 noticeably with many (hundreds) of watchers.
5028 root 1.153
5029 root 1.285 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
5030     will be C<0>.
5031 root 1.153
5032 root 1.159 =item EV_VERIFY
5033    
5034 root 1.309 Controls how much internal verification (see C<ev_verify ()>) will
5035 root 1.159 be done: If set to C<0>, no internal verification code will be compiled
5036     in. If set to C<1>, then verification code will be compiled in, but not
5037     called. If set to C<2>, then the internal verification code will be
5038     called once per loop, which can slow down libev. If set to C<3>, then the
5039     verification code will be called very frequently, which will slow down
5040     libev considerably.
5041    
5042 root 1.455 Verification errors are reported via C's C<assert> mechanism, so if you
5043     disable that (e.g. by defining C<NDEBUG>) then no errors will be reported.
5044    
5045 root 1.285 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
5046     will be C<0>.
5047 root 1.159
5048 root 1.39 =item EV_COMMON
5049    
5050     By default, all watchers have a C<void *data> member. By redefining
5051 sf-exg 1.300 this macro to something else you can include more and other types of
5052 root 1.39 members. You have to define it each time you include one of the files,
5053     though, and it must be identical each time.
5054    
5055     For example, the perl EV module uses something like this:
5056    
5057 root 1.164 #define EV_COMMON \
5058     SV *self; /* contains this struct */ \
5059     SV *cb_sv, *fh /* note no trailing ";" */
5060 root 1.39
5061 root 1.44 =item EV_CB_DECLARE (type)
5062 root 1.39
5063 root 1.44 =item EV_CB_INVOKE (watcher, revents)
5064 root 1.39
5065 root 1.44 =item ev_set_cb (ev, cb)
5066 root 1.39
5067     Can be used to change the callback member declaration in each watcher,
5068     and the way callbacks are invoked and set. Must expand to a struct member
5069 root 1.93 definition and a statement, respectively. See the F<ev.h> header file for
5070 root 1.39 their default definitions. One possible use for overriding these is to
5071 root 1.44 avoid the C<struct ev_loop *> as first argument in all cases, or to use
5072     method calls instead of plain function calls in C++.
5073 root 1.39
5074 root 1.185 =back
5075    
5076 root 1.89 =head2 EXPORTED API SYMBOLS
5077    
5078 root 1.161 If you need to re-export the API (e.g. via a DLL) and you need a list of
5079 root 1.89 exported symbols, you can use the provided F<Symbol.*> files which list
5080     all public symbols, one per line:
5081    
5082 root 1.164 Symbols.ev for libev proper
5083     Symbols.event for the libevent emulation
5084 root 1.89
5085     This can also be used to rename all public symbols to avoid clashes with
5086     multiple versions of libev linked together (which is obviously bad in
5087 root 1.161 itself, but sometimes it is inconvenient to avoid this).
5088 root 1.89
5089 root 1.92 A sed command like this will create wrapper C<#define>'s that you need to
5090 root 1.89 include before including F<ev.h>:
5091    
5092     <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
5093    
5094     This would create a file F<wrap.h> which essentially looks like this:
5095    
5096     #define ev_backend myprefix_ev_backend
5097     #define ev_check_start myprefix_ev_check_start
5098     #define ev_check_stop myprefix_ev_check_stop
5099     ...
5100    
5101 root 1.39 =head2 EXAMPLES
5102    
5103     For a real-world example of a program the includes libev
5104     verbatim, you can have a look at the EV perl module
5105     (L<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
5106     the F<libev/> subdirectory and includes them in the F<EV/EVAPI.h> (public
5107     interface) and F<EV.xs> (implementation) files. Only the F<EV.xs> file
5108     will be compiled. It is pretty complex because it provides its own header
5109     file.
5110    
5111     The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
5112 root 1.63 that everybody includes and which overrides some configure choices:
5113 root 1.39
5114 root 1.287 #define EV_FEATURES 8
5115 root 1.285 #define EV_USE_SELECT 1
5116 root 1.287 #define EV_PREPARE_ENABLE 1
5117     #define EV_IDLE_ENABLE 1
5118     #define EV_SIGNAL_ENABLE 1
5119     #define EV_CHILD_ENABLE 1
5120     #define EV_USE_STDEXCEPT 0
5121 root 1.164 #define EV_CONFIG_H <config.h>
5122 root 1.39
5123 root 1.164 #include "ev++.h"
5124 root 1.39
5125     And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
5126    
5127 root 1.164 #include "ev_cpp.h"
5128     #include "ev.c"
5129 root 1.39
5130 root 1.356 =head1 INTERACTION WITH OTHER PROGRAMS, LIBRARIES OR THE ENVIRONMENT
5131 root 1.46
5132 root 1.189 =head2 THREADS AND COROUTINES
5133 root 1.144
5134 root 1.189 =head3 THREADS
5135 root 1.144
5136 root 1.186 All libev functions are reentrant and thread-safe unless explicitly
5137 root 1.191 documented otherwise, but libev implements no locking itself. This means
5138     that you can use as many loops as you want in parallel, as long as there
5139     are no concurrent calls into any libev function with the same loop
5140     parameter (C<ev_default_*> calls have an implicit default loop parameter,
5141     of course): libev guarantees that different event loops share no data
5142 root 1.186 structures that need any locking.
5143 root 1.180
5144     Or to put it differently: calls with different loop parameters can be done
5145     concurrently from multiple threads, calls with the same loop parameter
5146     must be done serially (but can be done from different threads, as long as
5147     only one thread ever is inside a call at any point in time, e.g. by using
5148     a mutex per loop).
5149    
5150     Specifically to support threads (and signal handlers), libev implements
5151     so-called C<ev_async> watchers, which allow some limited form of
5152 root 1.186 concurrency on the same event loop, namely waking it up "from the
5153     outside".
5154 root 1.144
5155 root 1.170 If you want to know which design (one loop, locking, or multiple loops
5156     without or something else still) is best for your problem, then I cannot
5157 root 1.186 help you, but here is some generic advice:
5158 root 1.144
5159     =over 4
5160    
5161     =item * most applications have a main thread: use the default libev loop
5162 root 1.161 in that thread, or create a separate thread running only the default loop.
5163 root 1.144
5164     This helps integrating other libraries or software modules that use libev
5165     themselves and don't care/know about threading.
5166    
5167     =item * one loop per thread is usually a good model.
5168    
5169     Doing this is almost never wrong, sometimes a better-performance model
5170     exists, but it is always a good start.
5171    
5172     =item * other models exist, such as the leader/follower pattern, where one
5173 root 1.161 loop is handed through multiple threads in a kind of round-robin fashion.
5174 root 1.144
5175 root 1.161 Choosing a model is hard - look around, learn, know that usually you can do
5176 root 1.144 better than you currently do :-)
5177    
5178     =item * often you need to talk to some other thread which blocks in the
5179 root 1.182 event loop.
5180 root 1.144
5181 root 1.182 C<ev_async> watchers can be used to wake them up from other threads safely
5182     (or from signal contexts...).
5183    
5184     An example use would be to communicate signals or other events that only
5185     work in the default loop by registering the signal watcher with the
5186     default loop and triggering an C<ev_async> watcher from the default loop
5187     watcher callback into the event loop interested in the signal.
5188 root 1.180
5189 root 1.144 =back
5190    
5191 root 1.413 See also L</THREAD LOCKING EXAMPLE>.
5192 root 1.254
5193 root 1.189 =head3 COROUTINES
5194 root 1.144
5195 root 1.191 Libev is very accommodating to coroutines ("cooperative threads"):
5196     libev fully supports nesting calls to its functions from different
5197 root 1.310 coroutines (e.g. you can call C<ev_run> on the same loop from two
5198 root 1.255 different coroutines, and switch freely between both coroutines running
5199     the loop, as long as you don't confuse yourself). The only exception is
5200     that you must not do this from C<ev_periodic> reschedule callbacks.
5201 root 1.144
5202 root 1.181 Care has been taken to ensure that libev does not keep local state inside
5203 root 1.310 C<ev_run>, and other calls do not usually allow for coroutine switches as
5204 root 1.208 they do not call any callbacks.
5205 root 1.144
5206 root 1.189 =head2 COMPILER WARNINGS
5207    
5208     Depending on your compiler and compiler settings, you might get no or a
5209     lot of warnings when compiling libev code. Some people are apparently
5210     scared by this.
5211    
5212     However, these are unavoidable for many reasons. For one, each compiler
5213     has different warnings, and each user has different tastes regarding
5214     warning options. "Warn-free" code therefore cannot be a goal except when
5215     targeting a specific compiler and compiler-version.
5216    
5217     Another reason is that some compiler warnings require elaborate
5218     workarounds, or other changes to the code that make it less clear and less
5219     maintainable.
5220    
5221     And of course, some compiler warnings are just plain stupid, or simply
5222     wrong (because they don't actually warn about the condition their message
5223     seems to warn about). For example, certain older gcc versions had some
5224 sf-exg 1.300 warnings that resulted in an extreme number of false positives. These have
5225 root 1.189 been fixed, but some people still insist on making code warn-free with
5226     such buggy versions.
5227    
5228     While libev is written to generate as few warnings as possible,
5229     "warn-free" code is not a goal, and it is recommended not to build libev
5230     with any compiler warnings enabled unless you are prepared to cope with
5231     them (e.g. by ignoring them). Remember that warnings are just that:
5232     warnings, not errors, or proof of bugs.
5233    
5234    
5235 root 1.190 =head2 VALGRIND
5236 root 1.189
5237     Valgrind has a special section here because it is a popular tool that is
5238     highly useful. Unfortunately, valgrind reports are very hard to interpret.
5239    
5240     If you think you found a bug (memory leak, uninitialised data access etc.)
5241     in libev, then check twice: If valgrind reports something like:
5242    
5243     ==2274== definitely lost: 0 bytes in 0 blocks.
5244     ==2274== possibly lost: 0 bytes in 0 blocks.
5245     ==2274== still reachable: 256 bytes in 1 blocks.
5246    
5247     Then there is no memory leak, just as memory accounted to global variables
5248 root 1.208 is not a memleak - the memory is still being referenced, and didn't leak.
5249 root 1.189
5250     Similarly, under some circumstances, valgrind might report kernel bugs
5251     as if it were a bug in libev (e.g. in realloc or in the poll backend,
5252     although an acceptable workaround has been found here), or it might be
5253     confused.
5254    
5255     Keep in mind that valgrind is a very good tool, but only a tool. Don't
5256     make it into some kind of religion.
5257    
5258     If you are unsure about something, feel free to contact the mailing list
5259     with the full valgrind report and an explanation on why you think this
5260     is a bug in libev (best check the archives, too :). However, don't be
5261     annoyed when you get a brisk "this is no bug" answer and take the chance
5262     of learning how to interpret valgrind properly.
5263    
5264     If you need, for some reason, empty reports from valgrind for your project
5265     I suggest using suppression lists.
5266    
5267    
5268 root 1.190 =head1 PORTABILITY NOTES
5269 root 1.189
5270 root 1.302 =head2 GNU/LINUX 32 BIT LIMITATIONS
5271    
5272     GNU/Linux is the only common platform that supports 64 bit file/large file
5273 root 1.303 interfaces but I<disables> them by default.
5274 root 1.302
5275     That means that libev compiled in the default environment doesn't support
5276 root 1.303 files larger than 2GiB or so, which mainly affects C<ev_stat> watchers.
5277 root 1.302
5278     Unfortunately, many programs try to work around this GNU/Linux issue
5279     by enabling the large file API, which makes them incompatible with the
5280     standard libev compiled for their system.
5281    
5282     Likewise, libev cannot enable the large file API itself as this would
5283     suddenly make it incompatible to the default compile time environment,
5284     i.e. all programs not using special compile switches.
5285    
5286     =head2 OS/X AND DARWIN BUGS
5287    
5288     The whole thing is a bug if you ask me - basically any system interface
5289 root 1.303 you touch is broken, whether it is locales, poll, kqueue or even the
5290 root 1.302 OpenGL drivers.
5291    
5292 root 1.303 =head3 C<kqueue> is buggy
5293 root 1.302
5294     The kqueue syscall is broken in all known versions - most versions support
5295     only sockets, many support pipes.
5296    
5297 root 1.314 Libev tries to work around this by not using C<kqueue> by default on this
5298     rotten platform, but of course you can still ask for it when creating a
5299     loop - embedding a socket-only kqueue loop into a select-based one is
5300     probably going to work well.
5301 root 1.304
5302 root 1.303 =head3 C<poll> is buggy
5303 root 1.302
5304     Instead of fixing C<kqueue>, Apple replaced their (working) C<poll>
5305     implementation by something calling C<kqueue> internally around the 10.5.6
5306     release, so now C<kqueue> I<and> C<poll> are broken.
5307    
5308 root 1.304 Libev tries to work around this by not using C<poll> by default on
5309     this rotten platform, but of course you can still ask for it when creating
5310     a loop.
5311 root 1.302
5312 root 1.303 =head3 C<select> is buggy
5313 root 1.302
5314     All that's left is C<select>, and of course Apple found a way to fuck this
5315     one up as well: On OS/X, C<select> actively limits the number of file
5316 root 1.305 descriptors you can pass in to 1024 - your program suddenly crashes when
5317 root 1.302 you use more.
5318    
5319     There is an undocumented "workaround" for this - defining
5320     C<_DARWIN_UNLIMITED_SELECT>, which libev tries to use, so select I<should>
5321     work on OS/X.
5322    
5323     =head2 SOLARIS PROBLEMS AND WORKAROUNDS
5324    
5325 root 1.303 =head3 C<errno> reentrancy
5326 root 1.302
5327     The default compile environment on Solaris is unfortunately so
5328     thread-unsafe that you can't even use components/libraries compiled
5329 root 1.314 without C<-D_REENTRANT> in a threaded program, which, of course, isn't
5330     defined by default. A valid, if stupid, implementation choice.
5331 root 1.302
5332     If you want to use libev in threaded environments you have to make sure
5333     it's compiled with C<_REENTRANT> defined.
5334    
5335 root 1.303 =head3 Event port backend
5336 root 1.302
5337 root 1.314 The scalable event interface for Solaris is called "event
5338     ports". Unfortunately, this mechanism is very buggy in all major
5339     releases. If you run into high CPU usage, your program freezes or you get
5340     a large number of spurious wakeups, make sure you have all the relevant
5341     and latest kernel patches applied. No, I don't know which ones, but there
5342     are multiple ones to apply, and afterwards, event ports actually work
5343     great.
5344 root 1.302
5345 root 1.305 If you can't get it to work, you can try running the program by setting
5346     the environment variable C<LIBEV_FLAGS=3> to only allow C<poll> and
5347     C<select> backends.
5348 root 1.302
5349     =head2 AIX POLL BUG
5350    
5351     AIX unfortunately has a broken C<poll.h> header. Libev works around
5352     this by trying to avoid the poll backend altogether (i.e. it's not even
5353     compiled in), which normally isn't a big problem as C<select> works fine
5354 root 1.314 with large bitsets on AIX, and AIX is dead anyway.
5355 root 1.302
5356 root 1.189 =head2 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
5357 root 1.112
5358 root 1.303 =head3 General issues
5359    
5360 root 1.112 Win32 doesn't support any of the standards (e.g. POSIX) that libev
5361     requires, and its I/O model is fundamentally incompatible with the POSIX
5362     model. Libev still offers limited functionality on this platform in
5363     the form of the C<EVBACKEND_SELECT> backend, and only supports socket
5364     descriptors. This only applies when using Win32 natively, not when using
5365 root 1.303 e.g. cygwin. Actually, it only applies to the microsofts own compilers,
5366 sf-exg 1.374 as every compiler comes with a slightly differently broken/incompatible
5367 root 1.303 environment.
5368 root 1.112
5369 root 1.150 Lifting these limitations would basically require the full
5370 root 1.303 re-implementation of the I/O system. If you are into this kind of thing,
5371     then note that glib does exactly that for you in a very portable way (note
5372     also that glib is the slowest event library known to man).
5373 root 1.150
5374 root 1.112 There is no supported compilation method available on windows except
5375     embedding it into other applications.
5376    
5377 root 1.241 Sensible signal handling is officially unsupported by Microsoft - libev
5378     tries its best, but under most conditions, signals will simply not work.
5379    
5380 root 1.162 Not a libev limitation but worth mentioning: windows apparently doesn't
5381     accept large writes: instead of resulting in a partial write, windows will
5382     either accept everything or return C<ENOBUFS> if the buffer is too large,
5383     so make sure you only write small amounts into your sockets (less than a
5384 root 1.184 megabyte seems safe, but this apparently depends on the amount of memory
5385 root 1.162 available).
5386    
5387 root 1.150 Due to the many, low, and arbitrary limits on the win32 platform and
5388     the abysmal performance of winsockets, using a large number of sockets
5389     is not recommended (and not reasonable). If your program needs to use
5390     more than a hundred or so sockets, then likely it needs to use a totally
5391 root 1.155 different implementation for windows, as libev offers the POSIX readiness
5392 root 1.150 notification model, which cannot be implemented efficiently on windows
5393 root 1.241 (due to Microsoft monopoly games).
5394 root 1.112
5395 root 1.167 A typical way to use libev under windows is to embed it (see the embedding
5396     section for details) and use the following F<evwrap.h> header file instead
5397     of F<ev.h>:
5398    
5399     #define EV_STANDALONE /* keeps ev from requiring config.h */
5400     #define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
5401    
5402     #include "ev.h"
5403    
5404     And compile the following F<evwrap.c> file into your project (make sure
5405 root 1.184 you do I<not> compile the F<ev.c> or any other embedded source files!):
5406 root 1.167
5407     #include "evwrap.h"
5408     #include "ev.c"
5409    
5410 root 1.303 =head3 The winsocket C<select> function
5411 root 1.112
5412 root 1.160 The winsocket C<select> function doesn't follow POSIX in that it
5413     requires socket I<handles> and not socket I<file descriptors> (it is
5414     also extremely buggy). This makes select very inefficient, and also
5415 root 1.167 requires a mapping from file descriptors to socket handles (the Microsoft
5416     C runtime provides the function C<_open_osfhandle> for this). See the
5417 root 1.160 discussion of the C<EV_SELECT_USE_FD_SET>, C<EV_SELECT_IS_WINSOCKET> and
5418     C<EV_FD_TO_WIN32_HANDLE> preprocessor symbols for more info.
5419 root 1.112
5420 root 1.161 The configuration for a "naked" win32 using the Microsoft runtime
5421 root 1.112 libraries and raw winsocket select is:
5422    
5423 root 1.164 #define EV_USE_SELECT 1
5424     #define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
5425 root 1.112
5426     Note that winsockets handling of fd sets is O(n), so you can easily get a
5427     complexity in the O(n²) range when using win32.
5428    
5429 root 1.303 =head3 Limited number of file descriptors
5430 root 1.112
5431 root 1.150 Windows has numerous arbitrary (and low) limits on things.
5432    
5433     Early versions of winsocket's select only supported waiting for a maximum
5434     of C<64> handles (probably owning to the fact that all windows kernels
5435 root 1.161 can only wait for C<64> things at the same time internally; Microsoft
5436 root 1.150 recommends spawning a chain of threads and wait for 63 handles and the
5437 root 1.241 previous thread in each. Sounds great!).
5438 root 1.112
5439     Newer versions support more handles, but you need to define C<FD_SETSIZE>
5440     to some high number (e.g. C<2048>) before compiling the winsocket select
5441 root 1.241 call (which might be in libev or elsewhere, for example, perl and many
5442     other interpreters do their own select emulation on windows).
5443 root 1.112
5444 root 1.161 Another limit is the number of file descriptors in the Microsoft runtime
5445 root 1.241 libraries, which by default is C<64> (there must be a hidden I<64>
5446     fetish or something like this inside Microsoft). You can increase this
5447     by calling C<_setmaxstdio>, which can increase this limit to C<2048>
5448     (another arbitrary limit), but is broken in many versions of the Microsoft
5449     runtime libraries. This might get you to about C<512> or C<2048> sockets
5450     (depending on windows version and/or the phase of the moon). To get more,
5451     you need to wrap all I/O functions and provide your own fd management, but
5452     the cost of calling select (O(n²)) will likely make this unworkable.
5453 root 1.112
5454 root 1.189 =head2 PORTABILITY REQUIREMENTS
5455 root 1.112
5456 root 1.189 In addition to a working ISO-C implementation and of course the
5457     backend-specific APIs, libev relies on a few additional extensions:
5458 root 1.148
5459     =over 4
5460    
5461 root 1.165 =item C<void (*)(ev_watcher_type *, int revents)> must have compatible
5462     calling conventions regardless of C<ev_watcher_type *>.
5463    
5464     Libev assumes not only that all watcher pointers have the same internal
5465     structure (guaranteed by POSIX but not by ISO C for example), but it also
5466     assumes that the same (machine) code can be used to call any watcher
5467     callback: The watcher callbacks have different type signatures, but libev
5468     calls them using an C<ev_watcher *> internally.
5469    
5470 root 1.439 =item null pointers and integer zero are represented by 0 bytes
5471    
5472     Libev uses C<memset> to initialise structs and arrays to C<0> bytes, and
5473     relies on this setting pointers and integers to null.
5474    
5475 root 1.333 =item pointer accesses must be thread-atomic
5476    
5477     Accessing a pointer value must be atomic, it must both be readable and
5478     writable in one piece - this is the case on all current architectures.
5479    
5480 root 1.148 =item C<sig_atomic_t volatile> must be thread-atomic as well
5481    
5482     The type C<sig_atomic_t volatile> (or whatever is defined as
5483 root 1.184 C<EV_ATOMIC_T>) must be atomic with respect to accesses from different
5484 root 1.148 threads. This is not part of the specification for C<sig_atomic_t>, but is
5485     believed to be sufficiently portable.
5486    
5487     =item C<sigprocmask> must work in a threaded environment
5488    
5489     Libev uses C<sigprocmask> to temporarily block signals. This is not
5490     allowed in a threaded program (C<pthread_sigmask> has to be used). Typical
5491     pthread implementations will either allow C<sigprocmask> in the "main
5492     thread" or will block signals process-wide, both behaviours would
5493     be compatible with libev. Interaction between C<sigprocmask> and
5494     C<pthread_sigmask> could complicate things, however.
5495    
5496     The most portable way to handle signals is to block signals in all threads
5497 root 1.421 except the initial one, and run the signal handling loop in the initial
5498     thread as well.
5499 root 1.148
5500 root 1.150 =item C<long> must be large enough for common memory allocation sizes
5501    
5502 root 1.189 To improve portability and simplify its API, libev uses C<long> internally
5503     instead of C<size_t> when allocating its data structures. On non-POSIX
5504     systems (Microsoft...) this might be unexpectedly low, but is still at
5505     least 31 bits everywhere, which is enough for hundreds of millions of
5506     watchers.
5507 root 1.150
5508     =item C<double> must hold a time value in seconds with enough accuracy
5509    
5510 root 1.151 The type C<double> is used to represent timestamps. It is required to
5511 root 1.308 have at least 51 bits of mantissa (and 9 bits of exponent), which is
5512     good enough for at least into the year 4000 with millisecond accuracy
5513     (the design goal for libev). This requirement is overfulfilled by
5514 root 1.376 implementations using IEEE 754, which is basically all existing ones.
5515    
5516     With IEEE 754 doubles, you get microsecond accuracy until at least the
5517 sf-exg 1.382 year 2255 (and millisecond accuracy till the year 287396 - by then, libev
5518 root 1.376 is either obsolete or somebody patched it to use C<long double> or
5519     something like that, just kidding).
5520 root 1.150
5521 root 1.148 =back
5522    
5523     If you know of other additional requirements drop me a note.
5524    
5525    
5526 root 1.191 =head1 ALGORITHMIC COMPLEXITIES
5527    
5528     In this section the complexities of (many of) the algorithms used inside
5529     libev will be documented. For complexity discussions about backends see
5530     the documentation for C<ev_default_init>.
5531    
5532     All of the following are about amortised time: If an array needs to be
5533     extended, libev needs to realloc and move the whole array, but this
5534     happens asymptotically rarer with higher number of elements, so O(1) might
5535     mean that libev does a lengthy realloc operation in rare cases, but on
5536     average it is much faster and asymptotically approaches constant time.
5537    
5538     =over 4
5539    
5540     =item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
5541    
5542     This means that, when you have a watcher that triggers in one hour and
5543     there are 100 watchers that would trigger before that, then inserting will
5544     have to skip roughly seven (C<ld 100>) of these watchers.
5545    
5546     =item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
5547    
5548     That means that changing a timer costs less than removing/adding them,
5549     as only the relative motion in the event queue has to be paid for.
5550    
5551     =item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
5552    
5553     These just add the watcher into an array or at the head of a list.
5554    
5555     =item Stopping check/prepare/idle/fork/async watchers: O(1)
5556    
5557     =item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
5558    
5559     These watchers are stored in lists, so they need to be walked to find the
5560     correct watcher to remove. The lists are usually short (you don't usually
5561     have many watchers waiting for the same fd or signal: one is typical, two
5562     is rare).
5563    
5564     =item Finding the next timer in each loop iteration: O(1)
5565    
5566     By virtue of using a binary or 4-heap, the next timer is always found at a
5567     fixed position in the storage array.
5568    
5569     =item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
5570    
5571     A change means an I/O watcher gets started or stopped, which requires
5572     libev to recalculate its status (and possibly tell the kernel, depending
5573     on backend and whether C<ev_io_set> was used).
5574    
5575     =item Activating one watcher (putting it into the pending state): O(1)
5576    
5577     =item Priority handling: O(number_of_priorities)
5578    
5579     Priorities are implemented by allocating some space for each
5580     priority. When doing priority-based operations, libev usually has to
5581     linearly search all the priorities, but starting/stopping and activating
5582     watchers becomes O(1) with respect to priority handling.
5583    
5584     =item Sending an ev_async: O(1)
5585    
5586     =item Processing ev_async_send: O(number_of_async_watchers)
5587    
5588     =item Processing signals: O(max_signal_number)
5589    
5590     Sending involves a system call I<iff> there were no other C<ev_async_send>
5591 root 1.375 calls in the current loop iteration and the loop is currently
5592     blocked. Checking for async and signal events involves iterating over all
5593     running async watchers or all signal numbers.
5594 root 1.191
5595     =back
5596    
5597    
5598 root 1.291 =head1 PORTING FROM LIBEV 3.X TO 4.X
5599 root 1.289
5600 root 1.332 The major version 4 introduced some incompatible changes to the API.
5601 root 1.289
5602 root 1.332 At the moment, the C<ev.h> header file provides compatibility definitions
5603     for all changes, so most programs should still compile. The compatibility
5604     layer might be removed in later versions of libev, so better update to the
5605     new API early than late.
5606 root 1.291
5607 root 1.289 =over 4
5608    
5609 root 1.332 =item C<EV_COMPAT3> backwards compatibility mechanism
5610    
5611     The backward compatibility mechanism can be controlled by
5612 root 1.428 C<EV_COMPAT3>. See L</"PREPROCESSOR SYMBOLS/MACROS"> in the L</EMBEDDING>
5613 root 1.332 section.
5614    
5615 root 1.322 =item C<ev_default_destroy> and C<ev_default_fork> have been removed
5616    
5617     These calls can be replaced easily by their C<ev_loop_xxx> counterparts:
5618    
5619 root 1.325 ev_loop_destroy (EV_DEFAULT_UC);
5620 root 1.322 ev_loop_fork (EV_DEFAULT);
5621    
5622 root 1.310 =item function/symbol renames
5623    
5624     A number of functions and symbols have been renamed:
5625    
5626     ev_loop => ev_run
5627     EVLOOP_NONBLOCK => EVRUN_NOWAIT
5628     EVLOOP_ONESHOT => EVRUN_ONCE
5629    
5630     ev_unloop => ev_break
5631     EVUNLOOP_CANCEL => EVBREAK_CANCEL
5632     EVUNLOOP_ONE => EVBREAK_ONE
5633     EVUNLOOP_ALL => EVBREAK_ALL
5634 root 1.291
5635 root 1.310 EV_TIMEOUT => EV_TIMER
5636 root 1.291
5637 root 1.310 ev_loop_count => ev_iteration
5638     ev_loop_depth => ev_depth
5639     ev_loop_verify => ev_verify
5640 root 1.291
5641     Most functions working on C<struct ev_loop> objects don't have an
5642 root 1.310 C<ev_loop_> prefix, so it was removed; C<ev_loop>, C<ev_unloop> and
5643     associated constants have been renamed to not collide with the C<struct
5644     ev_loop> anymore and C<EV_TIMER> now follows the same naming scheme
5645     as all other watcher types. Note that C<ev_loop_fork> is still called
5646     C<ev_loop_fork> because it would otherwise clash with the C<ev_fork>
5647     typedef.
5648    
5649 root 1.289 =item C<EV_MINIMAL> mechanism replaced by C<EV_FEATURES>
5650    
5651     The preprocessor symbol C<EV_MINIMAL> has been replaced by a different
5652     mechanism, C<EV_FEATURES>. Programs using C<EV_MINIMAL> usually compile
5653     and work, but the library code will of course be larger.
5654    
5655     =back
5656    
5657    
5658 root 1.234 =head1 GLOSSARY
5659    
5660     =over 4
5661    
5662     =item active
5663    
5664 root 1.315 A watcher is active as long as it has been started and not yet stopped.
5665 root 1.413 See L</WATCHER STATES> for details.
5666 root 1.234
5667     =item application
5668    
5669     In this document, an application is whatever is using libev.
5670    
5671 root 1.316 =item backend
5672    
5673     The part of the code dealing with the operating system interfaces.
5674    
5675 root 1.234 =item callback
5676    
5677     The address of a function that is called when some event has been
5678     detected. Callbacks are being passed the event loop, the watcher that
5679     received the event, and the actual event bitset.
5680    
5681 root 1.315 =item callback/watcher invocation
5682 root 1.234
5683     The act of calling the callback associated with a watcher.
5684    
5685     =item event
5686    
5687     A change of state of some external event, such as data now being available
5688     for reading on a file descriptor, time having passed or simply not having
5689     any other events happening anymore.
5690    
5691     In libev, events are represented as single bits (such as C<EV_READ> or
5692 root 1.289 C<EV_TIMER>).
5693 root 1.234
5694     =item event library
5695    
5696     A software package implementing an event model and loop.
5697    
5698     =item event loop
5699    
5700     An entity that handles and processes external events and converts them
5701     into callback invocations.
5702    
5703     =item event model
5704    
5705     The model used to describe how an event loop handles and processes
5706     watchers and events.
5707    
5708     =item pending
5709    
5710 root 1.315 A watcher is pending as soon as the corresponding event has been
5711 root 1.413 detected. See L</WATCHER STATES> for details.
5712 root 1.234
5713     =item real time
5714    
5715     The physical time that is observed. It is apparently strictly monotonic :)
5716    
5717     =item wall-clock time
5718    
5719     The time and date as shown on clocks. Unlike real time, it can actually
5720 sf-exg 1.366 be wrong and jump forwards and backwards, e.g. when you adjust your
5721 root 1.234 clock.
5722    
5723     =item watcher
5724    
5725     A data structure that describes interest in certain events. Watchers need
5726     to be started (attached to an event loop) before they can receive events.
5727    
5728     =back
5729    
5730 root 1.1 =head1 AUTHOR
5731    
5732 root 1.333 Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael
5733 root 1.365 Magnusson and Emanuele Giaquinta, and minor corrections by many others.
5734 root 1.1