ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/ev.pod
Revision: 1.370
Committed: Thu Jun 2 23:42:40 2011 UTC (12 years, 11 months ago) by root
Branch: MAIN
Changes since 1.369: +13 -10 lines
Log Message:
*** empty log message ***

File Contents

# User Rev Content
1 root 1.1 =head1 NAME
2    
3     libev - a high performance full-featured event loop written in C
4    
5     =head1 SYNOPSIS
6    
7 root 1.164 #include <ev.h>
8 root 1.1
9 root 1.105 =head2 EXAMPLE PROGRAM
10 root 1.54
11 root 1.164 // a single header file is required
12     #include <ev.h>
13 root 1.54
14 root 1.217 #include <stdio.h> // for puts
15    
16 root 1.164 // every watcher type has its own typedef'd struct
17 root 1.200 // with the name ev_TYPE
18 root 1.164 ev_io stdin_watcher;
19     ev_timer timeout_watcher;
20    
21     // all watcher callbacks have a similar signature
22     // this callback is called when data is readable on stdin
23     static void
24 root 1.198 stdin_cb (EV_P_ ev_io *w, int revents)
25 root 1.164 {
26     puts ("stdin ready");
27     // for one-shot events, one must manually stop the watcher
28     // with its corresponding stop function.
29     ev_io_stop (EV_A_ w);
30    
31 root 1.310 // this causes all nested ev_run's to stop iterating
32     ev_break (EV_A_ EVBREAK_ALL);
33 root 1.164 }
34    
35     // another callback, this time for a time-out
36     static void
37 root 1.198 timeout_cb (EV_P_ ev_timer *w, int revents)
38 root 1.164 {
39     puts ("timeout");
40 root 1.310 // this causes the innermost ev_run to stop iterating
41     ev_break (EV_A_ EVBREAK_ONE);
42 root 1.164 }
43    
44     int
45     main (void)
46     {
47     // use the default event loop unless you have special needs
48 root 1.322 struct ev_loop *loop = EV_DEFAULT;
49 root 1.164
50     // initialise an io watcher, then start it
51     // this one will watch for stdin to become readable
52     ev_io_init (&stdin_watcher, stdin_cb, /*STDIN_FILENO*/ 0, EV_READ);
53     ev_io_start (loop, &stdin_watcher);
54    
55     // initialise a timer watcher, then start it
56     // simple non-repeating 5.5 second timeout
57     ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
58     ev_timer_start (loop, &timeout_watcher);
59    
60     // now wait for events to arrive
61 root 1.310 ev_run (loop, 0);
62 root 1.164
63 root 1.362 // break was called, so exit
64 root 1.164 return 0;
65     }
66 root 1.53
67 root 1.236 =head1 ABOUT THIS DOCUMENT
68    
69     This document documents the libev software package.
70 root 1.1
71 root 1.135 The newest version of this document is also available as an html-formatted
72 root 1.69 web page you might find easier to navigate when reading it for the first
73 root 1.154 time: L<http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod>.
74 root 1.69
75 root 1.236 While this document tries to be as complete as possible in documenting
76     libev, its usage and the rationale behind its design, it is not a tutorial
77     on event-based programming, nor will it introduce event-based programming
78     with libev.
79    
80 sf-exg 1.298 Familiarity with event based programming techniques in general is assumed
81 root 1.236 throughout this document.
82    
83 root 1.334 =head1 WHAT TO READ WHEN IN A HURRY
84    
85     This manual tries to be very detailed, but unfortunately, this also makes
86     it very long. If you just want to know the basics of libev, I suggest
87     reading L<ANATOMY OF A WATCHER>, then the L<EXAMPLE PROGRAM> above and
88     look up the missing functions in L<GLOBAL FUNCTIONS> and the C<ev_io> and
89     C<ev_timer> sections in L<WATCHER TYPES>.
90    
91 root 1.236 =head1 ABOUT LIBEV
92    
93 root 1.1 Libev is an event loop: you register interest in certain events (such as a
94 root 1.92 file descriptor being readable or a timeout occurring), and it will manage
95 root 1.4 these event sources and provide your program with events.
96 root 1.1
97     To do this, it must take more or less complete control over your process
98     (or thread) by executing the I<event loop> handler, and will then
99     communicate events via a callback mechanism.
100    
101     You register interest in certain events by registering so-called I<event
102     watchers>, which are relatively small C structures you initialise with the
103     details of the event, and then hand it over to libev by I<starting> the
104     watcher.
105    
106 root 1.105 =head2 FEATURES
107 root 1.1
108 root 1.58 Libev supports C<select>, C<poll>, the Linux-specific C<epoll>, the
109     BSD-specific C<kqueue> and the Solaris-specific event port mechanisms
110     for file descriptor events (C<ev_io>), the Linux C<inotify> interface
111 root 1.261 (for C<ev_stat>), Linux eventfd/signalfd (for faster and cleaner
112     inter-thread wakeup (C<ev_async>)/signal handling (C<ev_signal>)) relative
113     timers (C<ev_timer>), absolute timers with customised rescheduling
114     (C<ev_periodic>), synchronous signals (C<ev_signal>), process status
115     change events (C<ev_child>), and event watchers dealing with the event
116     loop mechanism itself (C<ev_idle>, C<ev_embed>, C<ev_prepare> and
117     C<ev_check> watchers) as well as file watchers (C<ev_stat>) and even
118     limited support for fork events (C<ev_fork>).
119 root 1.54
120     It also is quite fast (see this
121     L<benchmark|http://libev.schmorp.de/bench.html> comparing it to libevent
122     for example).
123 root 1.1
124 root 1.105 =head2 CONVENTIONS
125 root 1.1
126 root 1.135 Libev is very configurable. In this manual the default (and most common)
127     configuration will be described, which supports multiple event loops. For
128     more info about various configuration options please have a look at
129     B<EMBED> section in this manual. If libev was configured without support
130     for multiple event loops, then all functions taking an initial argument of
131 root 1.274 name C<loop> (which is always of type C<struct ev_loop *>) will not have
132 root 1.135 this argument.
133 root 1.1
134 root 1.105 =head2 TIME REPRESENTATION
135 root 1.1
136 root 1.237 Libev represents time as a single floating point number, representing
137 root 1.317 the (fractional) number of seconds since the (POSIX) epoch (in practice
138 root 1.293 somewhere near the beginning of 1970, details are complicated, don't
139     ask). This type is called C<ev_tstamp>, which is what you should use
140     too. It usually aliases to the C<double> type in C. When you need to do
141     any calculations on it, you should treat it as some floating point value.
142    
143     Unlike the name component C<stamp> might indicate, it is also used for
144     time differences (e.g. delays) throughout libev.
145 root 1.34
146 root 1.160 =head1 ERROR HANDLING
147    
148     Libev knows three classes of errors: operating system errors, usage errors
149     and internal errors (bugs).
150    
151     When libev catches an operating system error it cannot handle (for example
152 root 1.161 a system call indicating a condition libev cannot fix), it calls the callback
153 root 1.160 set via C<ev_set_syserr_cb>, which is supposed to fix the problem or
154     abort. The default is to print a diagnostic message and to call C<abort
155     ()>.
156    
157     When libev detects a usage error such as a negative timer interval, then
158     it will print a diagnostic message and abort (via the C<assert> mechanism,
159     so C<NDEBUG> will disable this checking): these are programming errors in
160     the libev caller and need to be fixed there.
161    
162     Libev also has a few internal error-checking C<assert>ions, and also has
163     extensive consistency checking code. These do not trigger under normal
164     circumstances, as they indicate either a bug in libev or worse.
165    
166    
167 root 1.17 =head1 GLOBAL FUNCTIONS
168    
169 root 1.18 These functions can be called anytime, even before initialising the
170     library in any way.
171    
172 root 1.1 =over 4
173    
174     =item ev_tstamp ev_time ()
175    
176 root 1.26 Returns the current time as libev would use it. Please note that the
177     C<ev_now> function is usually faster and also often returns the timestamp
178 sf-exg 1.321 you actually want to know. Also interesting is the combination of
179 root 1.318 C<ev_update_now> and C<ev_now>.
180 root 1.1
181 root 1.97 =item ev_sleep (ev_tstamp interval)
182    
183     Sleep for the given interval: The current thread will be blocked until
184     either it is interrupted or the given time interval has passed. Basically
185 root 1.161 this is a sub-second-resolution C<sleep ()>.
186 root 1.97
187 root 1.1 =item int ev_version_major ()
188    
189     =item int ev_version_minor ()
190    
191 root 1.80 You can find out the major and minor ABI version numbers of the library
192 root 1.1 you linked against by calling the functions C<ev_version_major> and
193     C<ev_version_minor>. If you want, you can compare against the global
194     symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the
195     version of the library your program was compiled against.
196    
197 root 1.80 These version numbers refer to the ABI version of the library, not the
198     release version.
199 root 1.79
200 root 1.9 Usually, it's a good idea to terminate if the major versions mismatch,
201 root 1.79 as this indicates an incompatible change. Minor versions are usually
202 root 1.1 compatible to older versions, so a larger minor version alone is usually
203     not a problem.
204    
205 root 1.54 Example: Make sure we haven't accidentally been linked against the wrong
206 root 1.320 version (note, however, that this will not detect other ABI mismatches,
207     such as LFS or reentrancy).
208 root 1.34
209 root 1.164 assert (("libev version mismatch",
210     ev_version_major () == EV_VERSION_MAJOR
211     && ev_version_minor () >= EV_VERSION_MINOR));
212 root 1.34
213 root 1.31 =item unsigned int ev_supported_backends ()
214    
215     Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
216     value) compiled into this binary of libev (independent of their
217     availability on the system you are running on). See C<ev_default_loop> for
218     a description of the set values.
219    
220 root 1.34 Example: make sure we have the epoll method, because yeah this is cool and
221     a must have and can we have a torrent of it please!!!11
222    
223 root 1.164 assert (("sorry, no epoll, no sex",
224     ev_supported_backends () & EVBACKEND_EPOLL));
225 root 1.34
226 root 1.31 =item unsigned int ev_recommended_backends ()
227    
228 root 1.318 Return the set of all backends compiled into this binary of libev and
229     also recommended for this platform, meaning it will work for most file
230     descriptor types. This set is often smaller than the one returned by
231     C<ev_supported_backends>, as for example kqueue is broken on most BSDs
232     and will not be auto-detected unless you explicitly request it (assuming
233     you know what you are doing). This is the set of backends that libev will
234     probe for if you specify no backends explicitly.
235 root 1.31
236 root 1.35 =item unsigned int ev_embeddable_backends ()
237    
238     Returns the set of backends that are embeddable in other event loops. This
239 root 1.319 value is platform-specific but can include backends not available on the
240     current system. To find which embeddable backends might be supported on
241     the current system, you would need to look at C<ev_embeddable_backends ()
242     & ev_supported_backends ()>, likewise for recommended ones.
243 root 1.35
244     See the description of C<ev_embed> watchers for more info.
245    
246 root 1.339 =item ev_set_allocator (void *(*cb)(void *ptr, long size))
247 root 1.1
248 root 1.59 Sets the allocation function to use (the prototype is similar - the
249 root 1.145 semantics are identical to the C<realloc> C89/SuS/POSIX function). It is
250     used to allocate and free memory (no surprises here). If it returns zero
251     when memory needs to be allocated (C<size != 0>), the library might abort
252     or take some potentially destructive action.
253    
254     Since some systems (at least OpenBSD and Darwin) fail to implement
255     correct C<realloc> semantics, libev will use a wrapper around the system
256     C<realloc> and C<free> functions by default.
257 root 1.1
258     You could override this function in high-availability programs to, say,
259     free some memory if it cannot allocate memory, to use a special allocator,
260     or even to sleep a while and retry until some memory is available.
261    
262 root 1.54 Example: Replace the libev allocator with one that waits a bit and then
263 root 1.145 retries (example requires a standards-compliant C<realloc>).
264 root 1.34
265     static void *
266 root 1.52 persistent_realloc (void *ptr, size_t size)
267 root 1.34 {
268     for (;;)
269     {
270     void *newptr = realloc (ptr, size);
271    
272     if (newptr)
273     return newptr;
274    
275     sleep (60);
276     }
277     }
278    
279     ...
280     ev_set_allocator (persistent_realloc);
281    
282 root 1.339 =item ev_set_syserr_cb (void (*cb)(const char *msg))
283 root 1.1
284 root 1.161 Set the callback function to call on a retryable system call error (such
285 root 1.1 as failed select, poll, epoll_wait). The message is a printable string
286     indicating the system call or subsystem causing the problem. If this
287 root 1.161 callback is set, then libev will expect it to remedy the situation, no
288 root 1.7 matter what, when it returns. That is, libev will generally retry the
289 root 1.1 requested operation, or, if the condition doesn't go away, do bad stuff
290     (such as abort).
291    
292 root 1.54 Example: This is basically the same thing that libev does internally, too.
293 root 1.34
294     static void
295     fatal_error (const char *msg)
296     {
297     perror (msg);
298     abort ();
299     }
300    
301     ...
302     ev_set_syserr_cb (fatal_error);
303    
304 root 1.349 =item ev_feed_signal (int signum)
305    
306     This function can be used to "simulate" a signal receive. It is completely
307     safe to call this function at any time, from any context, including signal
308     handlers or random threads.
309    
310 sf-exg 1.350 Its main use is to customise signal handling in your process, especially
311 root 1.349 in the presence of threads. For example, you could block signals
312     by default in all threads (and specifying C<EVFLAG_NOSIGMASK> when
313     creating any loops), and in one thread, use C<sigwait> or any other
314     mechanism to wait for signals, then "deliver" them to libev by calling
315     C<ev_feed_signal>.
316    
317 root 1.1 =back
318    
319 root 1.322 =head1 FUNCTIONS CONTROLLING EVENT LOOPS
320 root 1.1
321 root 1.310 An event loop is described by a C<struct ev_loop *> (the C<struct> is
322 root 1.311 I<not> optional in this case unless libev 3 compatibility is disabled, as
323     libev 3 had an C<ev_loop> function colliding with the struct name).
324 root 1.200
325     The library knows two types of such loops, the I<default> loop, which
326 root 1.331 supports child process events, and dynamically created event loops which
327     do not.
328 root 1.1
329     =over 4
330    
331     =item struct ev_loop *ev_default_loop (unsigned int flags)
332    
333 root 1.322 This returns the "default" event loop object, which is what you should
334     normally use when you just need "the event loop". Event loop objects and
335     the C<flags> parameter are described in more detail in the entry for
336     C<ev_loop_new>.
337    
338     If the default loop is already initialised then this function simply
339     returns it (and ignores the flags. If that is troubling you, check
340     C<ev_backend ()> afterwards). Otherwise it will create it with the given
341     flags, which should almost always be C<0>, unless the caller is also the
342     one calling C<ev_run> or otherwise qualifies as "the main program".
343 root 1.1
344     If you don't know what event loop to use, use the one returned from this
345 root 1.322 function (or via the C<EV_DEFAULT> macro).
346 root 1.1
347 root 1.139 Note that this function is I<not> thread-safe, so if you want to use it
348 root 1.322 from multiple threads, you have to employ some kind of mutex (note also
349     that this case is unlikely, as loops cannot be shared easily between
350     threads anyway).
351    
352     The default loop is the only loop that can handle C<ev_child> watchers,
353     and to do this, it always registers a handler for C<SIGCHLD>. If this is
354     a problem for your application you can either create a dynamic loop with
355     C<ev_loop_new> which doesn't do that, or you can simply overwrite the
356     C<SIGCHLD> signal handler I<after> calling C<ev_default_init>.
357 root 1.139
358 root 1.322 Example: This is the most typical usage.
359    
360     if (!ev_default_loop (0))
361     fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
362    
363     Example: Restrict libev to the select and poll backends, and do not allow
364     environment settings to be taken into account:
365    
366     ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
367    
368     =item struct ev_loop *ev_loop_new (unsigned int flags)
369    
370     This will create and initialise a new event loop object. If the loop
371     could not be initialised, returns false.
372    
373 root 1.343 This function is thread-safe, and one common way to use libev with
374     threads is indeed to create one loop per thread, and using the default
375     loop in the "main" or "initial" thread.
376 root 1.118
377 root 1.1 The flags argument can be used to specify special behaviour or specific
378 root 1.33 backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
379 root 1.1
380 root 1.33 The following flags are supported:
381 root 1.1
382     =over 4
383    
384 root 1.10 =item C<EVFLAG_AUTO>
385 root 1.1
386 root 1.9 The default flags value. Use this if you have no clue (it's the right
387 root 1.1 thing, believe me).
388    
389 root 1.10 =item C<EVFLAG_NOENV>
390 root 1.1
391 root 1.161 If this flag bit is or'ed into the flag value (or the program runs setuid
392 root 1.8 or setgid) then libev will I<not> look at the environment variable
393     C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
394     override the flags completely if it is found in the environment. This is
395     useful to try out specific backends to test their performance, or to work
396     around bugs.
397 root 1.1
398 root 1.62 =item C<EVFLAG_FORKCHECK>
399    
400 root 1.291 Instead of calling C<ev_loop_fork> manually after a fork, you can also
401     make libev check for a fork in each iteration by enabling this flag.
402 root 1.62
403     This works by calling C<getpid ()> on every iteration of the loop,
404     and thus this might slow down your event loop if you do a lot of loop
405 ayin 1.65 iterations and little real work, but is usually not noticeable (on my
406 root 1.135 GNU/Linux system for example, C<getpid> is actually a simple 5-insn sequence
407 root 1.161 without a system call and thus I<very> fast, but my GNU/Linux system also has
408 root 1.62 C<pthread_atfork> which is even faster).
409    
410     The big advantage of this flag is that you can forget about fork (and
411     forget about forgetting to tell libev about forking) when you use this
412     flag.
413    
414 root 1.161 This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
415 root 1.62 environment variable.
416    
417 root 1.260 =item C<EVFLAG_NOINOTIFY>
418    
419     When this flag is specified, then libev will not attempt to use the
420 root 1.340 I<inotify> API for its C<ev_stat> watchers. Apart from debugging and
421 root 1.260 testing, this flag can be useful to conserve inotify file descriptors, as
422     otherwise each loop using C<ev_stat> watchers consumes one inotify handle.
423    
424 root 1.277 =item C<EVFLAG_SIGNALFD>
425 root 1.260
426 root 1.277 When this flag is specified, then libev will attempt to use the
427 root 1.340 I<signalfd> API for its C<ev_signal> (and C<ev_child>) watchers. This API
428 root 1.278 delivers signals synchronously, which makes it both faster and might make
429     it possible to get the queued signal data. It can also simplify signal
430     handling with threads, as long as you properly block signals in your
431     threads that are not interested in handling them.
432 root 1.277
433     Signalfd will not be used by default as this changes your signal mask, and
434     there are a lot of shoddy libraries and programs (glib's threadpool for
435     example) that can't properly initialise their signal masks.
436 root 1.260
437 root 1.349 =item C<EVFLAG_NOSIGMASK>
438    
439     When this flag is specified, then libev will avoid to modify the signal
440     mask. Specifically, this means you ahve to make sure signals are unblocked
441     when you want to receive them.
442    
443     This behaviour is useful when you want to do your own signal handling, or
444     want to handle signals only in specific threads and want to avoid libev
445     unblocking the signals.
446    
447 root 1.360 It's also required by POSIX in a threaded program, as libev calls
448     C<sigprocmask>, whose behaviour is officially unspecified.
449    
450 root 1.349 This flag's behaviour will become the default in future versions of libev.
451    
452 root 1.31 =item C<EVBACKEND_SELECT> (value 1, portable select backend)
453 root 1.1
454 root 1.29 This is your standard select(2) backend. Not I<completely> standard, as
455     libev tries to roll its own fd_set with no limits on the number of fds,
456     but if that fails, expect a fairly low limit on the number of fds when
457 root 1.102 using this backend. It doesn't scale too well (O(highest_fd)), but its
458     usually the fastest backend for a low number of (low-numbered :) fds.
459    
460     To get good performance out of this backend you need a high amount of
461 root 1.161 parallelism (most of the file descriptors should be busy). If you are
462 root 1.102 writing a server, you should C<accept ()> in a loop to accept as many
463     connections as possible during one iteration. You might also want to have
464     a look at C<ev_set_io_collect_interval ()> to increase the amount of
465 root 1.155 readiness notifications you get per iteration.
466 root 1.1
467 root 1.179 This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
468     C<writefds> set (and to work around Microsoft Windows bugs, also onto the
469     C<exceptfds> set on that platform).
470    
471 root 1.31 =item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
472 root 1.1
473 root 1.102 And this is your standard poll(2) backend. It's more complicated
474     than select, but handles sparse fds better and has no artificial
475     limit on the number of fds you can use (except it will slow down
476     considerably with a lot of inactive fds). It scales similarly to select,
477     i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
478     performance tips.
479 root 1.1
480 root 1.179 This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
481     C<EV_WRITE> to C<POLLOUT | POLLERR | POLLHUP>.
482    
483 root 1.31 =item C<EVBACKEND_EPOLL> (value 4, Linux)
484 root 1.1
485 root 1.272 Use the linux-specific epoll(7) interface (for both pre- and post-2.6.9
486     kernels).
487    
488 root 1.368 For few fds, this backend is a bit little slower than poll and select, but
489     it scales phenomenally better. While poll and select usually scale like
490     O(total_fds) where total_fds is the total number of fds (or the highest
491     fd), epoll scales either O(1) or O(active_fds).
492 root 1.205
493 root 1.210 The epoll mechanism deserves honorable mention as the most misdesigned
494     of the more advanced event mechanisms: mere annoyances include silently
495     dropping file descriptors, requiring a system call per change per file
496 root 1.337 descriptor (and unnecessary guessing of parameters), problems with dup,
497 root 1.338 returning before the timeout value, resulting in additional iterations
498     (and only giving 5ms accuracy while select on the same platform gives
499     0.1ms) and so on. The biggest issue is fork races, however - if a program
500     forks then I<both> parent and child process have to recreate the epoll
501     set, which can take considerable time (one syscall per file descriptor)
502     and is of course hard to detect.
503 root 1.1
504 root 1.370 Epoll is also notoriously buggy - embedding epoll fds I<should> work,
505     but of course I<doesn't>, and epoll just loves to report events for
506     totally I<different> file descriptors (even already closed ones, so
507     one cannot even remove them from the set) than registered in the set
508     (especially on SMP systems). Libev tries to counter these spurious
509     notifications by employing an additional generation counter and comparing
510     that against the events to filter out spurious ones, recreating the set
511     when required. Epoll also errornously rounds down timeouts, but gives you
512     no way to know when and by how much, so sometimes you have to busy-wait
513     because epoll returns immediately despite a nonzero timeout. And last
514 root 1.306 not least, it also refuses to work with some file descriptors which work
515     perfectly fine with C<select> (files, many character devices...).
516 root 1.204
517 root 1.370 Epoll is truly the train wreck among event poll mechanisms, a frankenpoll,
518     cobbled together in a hurry, no thought to design or interaction with
519     others. Oh, the pain, will it ever stop...
520 root 1.338
521 root 1.94 While stopping, setting and starting an I/O watcher in the same iteration
522 root 1.210 will result in some caching, there is still a system call per such
523     incident (because the same I<file descriptor> could point to a different
524     I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
525     file descriptors might not work very well if you register events for both
526     file descriptors.
527 root 1.29
528 root 1.102 Best performance from this backend is achieved by not unregistering all
529 root 1.183 watchers for a file descriptor until it has been closed, if possible,
530     i.e. keep at least one watcher active per fd at all times. Stopping and
531     starting a watcher (without re-setting it) also usually doesn't cause
532 root 1.206 extra overhead. A fork can both result in spurious notifications as well
533     as in libev having to destroy and recreate the epoll object, which can
534     take considerable time and thus should be avoided.
535 root 1.102
536 root 1.215 All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
537     faster than epoll for maybe up to a hundred file descriptors, depending on
538 root 1.214 the usage. So sad.
539 root 1.213
540 root 1.161 While nominally embeddable in other event loops, this feature is broken in
541 root 1.102 all kernel versions tested so far.
542    
543 root 1.179 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
544     C<EVBACKEND_POLL>.
545    
546 root 1.31 =item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
547 root 1.29
548 root 1.210 Kqueue deserves special mention, as at the time of this writing, it
549     was broken on all BSDs except NetBSD (usually it doesn't work reliably
550     with anything but sockets and pipes, except on Darwin, where of course
551     it's completely useless). Unlike epoll, however, whose brokenness
552     is by design, these kqueue bugs can (and eventually will) be fixed
553     without API changes to existing programs. For this reason it's not being
554     "auto-detected" unless you explicitly specify it in the flags (i.e. using
555     C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
556     system like NetBSD.
557 root 1.29
558 root 1.100 You still can embed kqueue into a normal poll or select backend and use it
559     only for sockets (after having made sure that sockets work with kqueue on
560     the target platform). See C<ev_embed> watchers for more info.
561    
562 root 1.29 It scales in the same way as the epoll backend, but the interface to the
563 root 1.100 kernel is more efficient (which says nothing about its actual speed, of
564     course). While stopping, setting and starting an I/O watcher does never
565 root 1.161 cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
566 root 1.206 two event changes per incident. Support for C<fork ()> is very bad (but
567     sane, unlike epoll) and it drops fds silently in similarly hard-to-detect
568     cases
569 root 1.29
570 root 1.102 This backend usually performs well under most conditions.
571    
572     While nominally embeddable in other event loops, this doesn't work
573     everywhere, so you might need to test for this. And since it is broken
574     almost everywhere, you should only use it when you have a lot of sockets
575     (for which it usually works), by embedding it into another event loop
576 root 1.223 (e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
577     also broken on OS X)) and, did I mention it, using it only for sockets.
578 root 1.102
579 root 1.179 This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
580     C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
581     C<NOTE_EOF>.
582    
583 root 1.31 =item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
584 root 1.29
585 root 1.102 This is not implemented yet (and might never be, unless you send me an
586     implementation). According to reports, C</dev/poll> only supports sockets
587     and is not embeddable, which would limit the usefulness of this backend
588     immensely.
589 root 1.29
590 root 1.31 =item C<EVBACKEND_PORT> (value 32, Solaris 10)
591 root 1.29
592 root 1.94 This uses the Solaris 10 event port mechanism. As with everything on Solaris,
593 root 1.29 it's really slow, but it still scales very well (O(active_fds)).
594    
595 root 1.102 While this backend scales well, it requires one system call per active
596     file descriptor per loop iteration. For small and medium numbers of file
597     descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
598     might perform better.
599    
600 root 1.351 On the positive side, this backend actually performed fully to
601     specification in all tests and is fully embeddable, which is a rare feat
602     among the OS-specific backends (I vastly prefer correctness over speed
603     hacks).
604    
605 root 1.352 On the negative side, the interface is I<bizarre> - so bizarre that
606     even sun itself gets it wrong in their code examples: The event polling
607 root 1.351 function sometimes returning events to the caller even though an error
608 root 1.353 occurred, but with no indication whether it has done so or not (yes, it's
609 root 1.352 even documented that way) - deadly for edge-triggered interfaces where
610 root 1.353 you absolutely have to know whether an event occurred or not because you
611 root 1.352 have to re-arm the watcher.
612    
613     Fortunately libev seems to be able to work around these idiocies.
614 root 1.117
615 root 1.179 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
616     C<EVBACKEND_POLL>.
617    
618 root 1.31 =item C<EVBACKEND_ALL>
619 root 1.29
620     Try all backends (even potentially broken ones that wouldn't be tried
621     with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
622 root 1.31 C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.
623 root 1.1
624 root 1.349 It is definitely not recommended to use this flag, use whatever
625     C<ev_recommended_backends ()> returns, or simply do not specify a backend
626     at all.
627    
628     =item C<EVBACKEND_MASK>
629    
630     Not a backend at all, but a mask to select all backend bits from a
631     C<flags> value, in case you want to mask out any backends from a flags
632     value (e.g. when modifying the C<LIBEV_FLAGS> environment variable).
633 root 1.102
634 root 1.1 =back
635    
636 root 1.260 If one or more of the backend flags are or'ed into the flags value,
637     then only these backends will be tried (in the reverse order as listed
638     here). If none are specified, all backends in C<ev_recommended_backends
639     ()> will be tried.
640 root 1.29
641 root 1.54 Example: Try to create a event loop that uses epoll and nothing else.
642 root 1.34
643 root 1.164 struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
644     if (!epoller)
645     fatal ("no epoll found here, maybe it hides under your chair");
646 root 1.34
647 root 1.323 Example: Use whatever libev has to offer, but make sure that kqueue is
648     used if available.
649    
650     struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_KQUEUE);
651    
652 root 1.322 =item ev_loop_destroy (loop)
653 root 1.1
654 root 1.322 Destroys an event loop object (frees all memory and kernel state
655     etc.). None of the active event watchers will be stopped in the normal
656     sense, so e.g. C<ev_is_active> might still return true. It is your
657     responsibility to either stop all watchers cleanly yourself I<before>
658     calling this function, or cope with the fact afterwards (which is usually
659     the easiest thing, you can just ignore the watchers and/or C<free ()> them
660     for example).
661 root 1.1
662 root 1.203 Note that certain global state, such as signal state (and installed signal
663     handlers), will not be freed by this function, and related watchers (such
664     as signal and child watchers) would need to be stopped manually.
665 root 1.87
666 root 1.322 This function is normally used on loop objects allocated by
667     C<ev_loop_new>, but it can also be used on the default loop returned by
668     C<ev_default_loop>, in which case it is not thread-safe.
669    
670     Note that it is not advisable to call this function on the default loop
671 root 1.340 except in the rare occasion where you really need to free its resources.
672 root 1.322 If you need dynamically allocated loops it is better to use C<ev_loop_new>
673     and C<ev_loop_destroy>.
674 root 1.87
675 root 1.322 =item ev_loop_fork (loop)
676 root 1.1
677 root 1.322 This function sets a flag that causes subsequent C<ev_run> iterations to
678     reinitialise the kernel state for backends that have one. Despite the
679 root 1.119 name, you can call it anytime, but it makes most sense after forking, in
680 root 1.322 the child process. You I<must> call it (or use C<EVFLAG_FORKCHECK>) in the
681     child before resuming or calling C<ev_run>.
682 root 1.119
683 root 1.291 Again, you I<have> to call it on I<any> loop that you want to re-use after
684     a fork, I<even if you do not plan to use the loop in the parent>. This is
685     because some kernel interfaces *cough* I<kqueue> *cough* do funny things
686     during fork.
687    
688 root 1.119 On the other hand, you only need to call this function in the child
689 root 1.310 process if and only if you want to use the event loop in the child. If
690     you just fork+exec or create a new loop in the child, you don't have to
691     call it at all (in fact, C<epoll> is so badly broken that it makes a
692     difference, but libev will usually detect this case on its own and do a
693     costly reset of the backend).
694 root 1.1
695 root 1.9 The function itself is quite fast and it's usually not a problem to call
696 root 1.322 it just in case after a fork.
697 root 1.1
698 root 1.322 Example: Automate calling C<ev_loop_fork> on the default loop when
699     using pthreads.
700 root 1.1
701 root 1.322 static void
702     post_fork_child (void)
703     {
704     ev_loop_fork (EV_DEFAULT);
705     }
706 root 1.1
707 root 1.322 ...
708     pthread_atfork (0, 0, post_fork_child);
709 root 1.1
710 root 1.131 =item int ev_is_default_loop (loop)
711    
712 root 1.183 Returns true when the given loop is, in fact, the default loop, and false
713     otherwise.
714 root 1.131
715 root 1.291 =item unsigned int ev_iteration (loop)
716 root 1.66
717 root 1.310 Returns the current iteration count for the event loop, which is identical
718     to the number of times libev did poll for new events. It starts at C<0>
719     and happily wraps around with enough iterations.
720 root 1.66
721     This value can sometimes be useful as a generation counter of sorts (it
722     "ticks" the number of loop iterations), as it roughly corresponds with
723 root 1.291 C<ev_prepare> and C<ev_check> calls - and is incremented between the
724     prepare and check phases.
725 root 1.66
726 root 1.291 =item unsigned int ev_depth (loop)
727 root 1.247
728 root 1.310 Returns the number of times C<ev_run> was entered minus the number of
729 root 1.343 times C<ev_run> was exited normally, in other words, the recursion depth.
730 root 1.247
731 root 1.310 Outside C<ev_run>, this number is zero. In a callback, this number is
732     C<1>, unless C<ev_run> was invoked recursively (or from another thread),
733 root 1.247 in which case it is higher.
734    
735 root 1.343 Leaving C<ev_run> abnormally (setjmp/longjmp, cancelling the thread,
736     throwing an exception etc.), doesn't count as "exit" - consider this
737     as a hint to avoid such ungentleman-like behaviour unless it's really
738     convenient, in which case it is fully supported.
739 root 1.247
740 root 1.31 =item unsigned int ev_backend (loop)
741 root 1.1
742 root 1.31 Returns one of the C<EVBACKEND_*> flags indicating the event backend in
743 root 1.1 use.
744    
745 root 1.9 =item ev_tstamp ev_now (loop)
746 root 1.1
747     Returns the current "event loop time", which is the time the event loop
748 root 1.34 received events and started processing them. This timestamp does not
749     change as long as callbacks are being processed, and this is also the base
750     time used for relative timers. You can treat it as the timestamp of the
751 root 1.92 event occurring (or more correctly, libev finding out about it).
752 root 1.1
753 root 1.176 =item ev_now_update (loop)
754    
755     Establishes the current time by querying the kernel, updating the time
756     returned by C<ev_now ()> in the progress. This is a costly operation and
757 root 1.310 is usually done automatically within C<ev_run ()>.
758 root 1.176
759     This function is rarely useful, but when some event callback runs for a
760     very long time without entering the event loop, updating libev's idea of
761     the current time is a good idea.
762    
763 root 1.238 See also L<The special problem of time updates> in the C<ev_timer> section.
764 root 1.176
765 root 1.231 =item ev_suspend (loop)
766    
767     =item ev_resume (loop)
768    
769 root 1.310 These two functions suspend and resume an event loop, for use when the
770     loop is not used for a while and timeouts should not be processed.
771 root 1.231
772     A typical use case would be an interactive program such as a game: When
773     the user presses C<^Z> to suspend the game and resumes it an hour later it
774     would be best to handle timeouts as if no time had actually passed while
775     the program was suspended. This can be achieved by calling C<ev_suspend>
776     in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
777     C<ev_resume> directly afterwards to resume timer processing.
778    
779     Effectively, all C<ev_timer> watchers will be delayed by the time spend
780     between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
781     will be rescheduled (that is, they will lose any events that would have
782 sf-exg 1.298 occurred while suspended).
783 root 1.231
784     After calling C<ev_suspend> you B<must not> call I<any> function on the
785     given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
786     without a previous call to C<ev_suspend>.
787    
788     Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
789     event loop time (see C<ev_now_update>).
790    
791 root 1.310 =item ev_run (loop, int flags)
792 root 1.1
793     Finally, this is it, the event handler. This function usually is called
794 root 1.268 after you have initialised all your watchers and you want to start
795 root 1.310 handling events. It will ask the operating system for any new events, call
796     the watcher callbacks, an then repeat the whole process indefinitely: This
797     is why event loops are called I<loops>.
798 root 1.1
799 root 1.310 If the flags argument is specified as C<0>, it will keep handling events
800     until either no event watchers are active anymore or C<ev_break> was
801     called.
802 root 1.1
803 root 1.310 Please note that an explicit C<ev_break> is usually better than
804 root 1.34 relying on all watchers to be stopped when deciding when a program has
805 root 1.183 finished (especially in interactive programs), but having a program
806     that automatically loops as long as it has to and no longer by virtue
807     of relying on its watchers stopping correctly, that is truly a thing of
808     beauty.
809 root 1.34
810 root 1.343 This function is also I<mostly> exception-safe - you can break out of
811     a C<ev_run> call by calling C<longjmp> in a callback, throwing a C++
812     exception and so on. This does not decrement the C<ev_depth> value, nor
813     will it clear any outstanding C<EVBREAK_ONE> breaks.
814    
815 root 1.310 A flags value of C<EVRUN_NOWAIT> will look for new events, will handle
816     those events and any already outstanding ones, but will not wait and
817     block your process in case there are no events and will return after one
818     iteration of the loop. This is sometimes useful to poll and handle new
819     events while doing lengthy calculations, to keep the program responsive.
820 root 1.1
821 root 1.310 A flags value of C<EVRUN_ONCE> will look for new events (waiting if
822 root 1.183 necessary) and will handle those and any already outstanding ones. It
823     will block your process until at least one new event arrives (which could
824 root 1.208 be an event internal to libev itself, so there is no guarantee that a
825 root 1.183 user-registered callback will be called), and will return after one
826     iteration of the loop.
827    
828     This is useful if you are waiting for some external event in conjunction
829     with something not expressible using other libev watchers (i.e. "roll your
830 root 1.310 own C<ev_run>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
831 root 1.33 usually a better approach for this kind of thing.
832    
833 root 1.369 Here are the gory details of what C<ev_run> does (this is for your
834     understanding, not a guarantee that things will work exactly like this in
835     future versions):
836 root 1.33
837 root 1.310 - Increment loop depth.
838     - Reset the ev_break status.
839 root 1.77 - Before the first iteration, call any pending watchers.
840 root 1.310 LOOP:
841     - If EVFLAG_FORKCHECK was used, check for a fork.
842 root 1.171 - If a fork was detected (by any means), queue and call all fork watchers.
843 root 1.113 - Queue and call all prepare watchers.
844 root 1.310 - If ev_break was called, goto FINISH.
845 root 1.171 - If we have been forked, detach and recreate the kernel state
846     as to not disturb the other process.
847 root 1.33 - Update the kernel state with all outstanding changes.
848 root 1.171 - Update the "event loop time" (ev_now ()).
849 root 1.113 - Calculate for how long to sleep or block, if at all
850 root 1.310 (active idle watchers, EVRUN_NOWAIT or not having
851 root 1.113 any active watchers at all will result in not sleeping).
852     - Sleep if the I/O and timer collect interval say so.
853 root 1.310 - Increment loop iteration counter.
854 root 1.33 - Block the process, waiting for any events.
855     - Queue all outstanding I/O (fd) events.
856 root 1.171 - Update the "event loop time" (ev_now ()), and do time jump adjustments.
857 root 1.183 - Queue all expired timers.
858     - Queue all expired periodics.
859 root 1.310 - Queue all idle watchers with priority higher than that of pending events.
860 root 1.33 - Queue all check watchers.
861     - Call all queued watchers in reverse order (i.e. check watchers first).
862     Signals and child watchers are implemented as I/O watchers, and will
863     be handled here by queueing them when their watcher gets executed.
864 root 1.310 - If ev_break has been called, or EVRUN_ONCE or EVRUN_NOWAIT
865     were used, or there are no active watchers, goto FINISH, otherwise
866     continue with step LOOP.
867     FINISH:
868     - Reset the ev_break status iff it was EVBREAK_ONE.
869     - Decrement the loop depth.
870     - Return.
871 root 1.27
872 root 1.114 Example: Queue some jobs and then loop until no events are outstanding
873 root 1.34 anymore.
874    
875     ... queue jobs here, make sure they register event watchers as long
876     ... as they still have work to do (even an idle watcher will do..)
877 root 1.310 ev_run (my_loop, 0);
878 root 1.362 ... jobs done or somebody called break. yeah!
879 root 1.34
880 root 1.310 =item ev_break (loop, how)
881 root 1.1
882 root 1.310 Can be used to make a call to C<ev_run> return early (but only after it
883 root 1.9 has processed all outstanding events). The C<how> argument must be either
884 root 1.310 C<EVBREAK_ONE>, which will make the innermost C<ev_run> call return, or
885     C<EVBREAK_ALL>, which will make all nested C<ev_run> calls return.
886 root 1.1
887 root 1.343 This "break state" will be cleared on the next call to C<ev_run>.
888 root 1.115
889 root 1.343 It is safe to call C<ev_break> from outside any C<ev_run> calls, too, in
890     which case it will have no effect.
891 root 1.194
892 root 1.1 =item ev_ref (loop)
893    
894     =item ev_unref (loop)
895    
896 root 1.9 Ref/unref can be used to add or remove a reference count on the event
897     loop: Every watcher keeps one reference, and as long as the reference
898 root 1.310 count is nonzero, C<ev_run> will not return on its own.
899 root 1.183
900 root 1.276 This is useful when you have a watcher that you never intend to
901 root 1.310 unregister, but that nevertheless should not keep C<ev_run> from
902 root 1.276 returning. In such a case, call C<ev_unref> after starting, and C<ev_ref>
903     before stopping it.
904 root 1.183
905 root 1.229 As an example, libev itself uses this for its internal signal pipe: It
906 root 1.310 is not visible to the libev user and should not keep C<ev_run> from
907 root 1.229 exiting if no event watchers registered by it are active. It is also an
908     excellent way to do this for generic recurring timers or from within
909     third-party libraries. Just remember to I<unref after start> and I<ref
910     before stop> (but only if the watcher wasn't active before, or was active
911     before, respectively. Note also that libev might stop watchers itself
912     (e.g. non-repeating timers) in which case you have to C<ev_ref>
913     in the callback).
914 root 1.1
915 root 1.310 Example: Create a signal watcher, but keep it from keeping C<ev_run>
916 root 1.34 running when nothing else is active.
917    
918 root 1.198 ev_signal exitsig;
919 root 1.164 ev_signal_init (&exitsig, sig_cb, SIGINT);
920     ev_signal_start (loop, &exitsig);
921 sf-exg 1.348 ev_unref (loop);
922 root 1.34
923 root 1.54 Example: For some weird reason, unregister the above signal handler again.
924 root 1.34
925 root 1.164 ev_ref (loop);
926     ev_signal_stop (loop, &exitsig);
927 root 1.34
928 root 1.97 =item ev_set_io_collect_interval (loop, ev_tstamp interval)
929    
930     =item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
931    
932     These advanced functions influence the time that libev will spend waiting
933 root 1.171 for events. Both time intervals are by default C<0>, meaning that libev
934     will try to invoke timer/periodic callbacks and I/O callbacks with minimum
935     latency.
936 root 1.97
937     Setting these to a higher value (the C<interval> I<must> be >= C<0>)
938 root 1.171 allows libev to delay invocation of I/O and timer/periodic callbacks
939     to increase efficiency of loop iterations (or to increase power-saving
940     opportunities).
941 root 1.97
942 root 1.183 The idea is that sometimes your program runs just fast enough to handle
943     one (or very few) event(s) per loop iteration. While this makes the
944     program responsive, it also wastes a lot of CPU time to poll for new
945 root 1.97 events, especially with backends like C<select ()> which have a high
946     overhead for the actual polling but can deliver many events at once.
947    
948     By setting a higher I<io collect interval> you allow libev to spend more
949     time collecting I/O events, so you can handle more events per iteration,
950     at the cost of increasing latency. Timeouts (both C<ev_periodic> and
951 ayin 1.101 C<ev_timer>) will be not affected. Setting this to a non-null value will
952 root 1.245 introduce an additional C<ev_sleep ()> call into most loop iterations. The
953     sleep time ensures that libev will not poll for I/O events more often then
954     once per this interval, on average.
955 root 1.97
956     Likewise, by setting a higher I<timeout collect interval> you allow libev
957     to spend more time collecting timeouts, at the expense of increased
958 root 1.183 latency/jitter/inexactness (the watcher callback will be called
959     later). C<ev_io> watchers will not be affected. Setting this to a non-null
960     value will not introduce any overhead in libev.
961 root 1.97
962 root 1.161 Many (busy) programs can usually benefit by setting the I/O collect
963 root 1.98 interval to a value near C<0.1> or so, which is often enough for
964     interactive servers (of course not for games), likewise for timeouts. It
965     usually doesn't make much sense to set it to a lower value than C<0.01>,
966 root 1.245 as this approaches the timing granularity of most systems. Note that if
967     you do transactions with the outside world and you can't increase the
968     parallelity, then this setting will limit your transaction rate (if you
969     need to poll once per transaction and the I/O collect interval is 0.01,
970 sf-exg 1.298 then you can't do more than 100 transactions per second).
971 root 1.97
972 root 1.171 Setting the I<timeout collect interval> can improve the opportunity for
973     saving power, as the program will "bundle" timer callback invocations that
974     are "near" in time together, by delaying some, thus reducing the number of
975     times the process sleeps and wakes up again. Another useful technique to
976     reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
977     they fire on, say, one-second boundaries only.
978    
979 root 1.245 Example: we only need 0.1s timeout granularity, and we wish not to poll
980     more often than 100 times per second:
981    
982     ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
983     ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
984    
985 root 1.253 =item ev_invoke_pending (loop)
986    
987     This call will simply invoke all pending watchers while resetting their
988 root 1.310 pending state. Normally, C<ev_run> does this automatically when required,
989 root 1.313 but when overriding the invoke callback this call comes handy. This
990     function can be invoked from a watcher - this can be useful for example
991     when you want to do some lengthy calculation and want to pass further
992     event handling to another thread (you still have to make sure only one
993     thread executes within C<ev_invoke_pending> or C<ev_run> of course).
994 root 1.253
995 root 1.256 =item int ev_pending_count (loop)
996    
997     Returns the number of pending watchers - zero indicates that no watchers
998     are pending.
999    
1000 root 1.253 =item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
1001    
1002     This overrides the invoke pending functionality of the loop: Instead of
1003 root 1.310 invoking all pending watchers when there are any, C<ev_run> will call
1004 root 1.253 this callback instead. This is useful, for example, when you want to
1005     invoke the actual watchers inside another context (another thread etc.).
1006    
1007     If you want to reset the callback, use C<ev_invoke_pending> as new
1008     callback.
1009    
1010     =item ev_set_loop_release_cb (loop, void (*release)(EV_P), void (*acquire)(EV_P))
1011    
1012     Sometimes you want to share the same loop between multiple threads. This
1013     can be done relatively simply by putting mutex_lock/unlock calls around
1014     each call to a libev function.
1015    
1016 root 1.310 However, C<ev_run> can run an indefinite time, so it is not feasible
1017     to wait for it to return. One way around this is to wake up the event
1018     loop via C<ev_break> and C<av_async_send>, another way is to set these
1019     I<release> and I<acquire> callbacks on the loop.
1020 root 1.253
1021     When set, then C<release> will be called just before the thread is
1022     suspended waiting for new events, and C<acquire> is called just
1023     afterwards.
1024    
1025     Ideally, C<release> will just call your mutex_unlock function, and
1026     C<acquire> will just call the mutex_lock function again.
1027    
1028 root 1.254 While event loop modifications are allowed between invocations of
1029     C<release> and C<acquire> (that's their only purpose after all), no
1030     modifications done will affect the event loop, i.e. adding watchers will
1031     have no effect on the set of file descriptors being watched, or the time
1032 root 1.310 waited. Use an C<ev_async> watcher to wake up C<ev_run> when you want it
1033 root 1.254 to take note of any changes you made.
1034    
1035 root 1.310 In theory, threads executing C<ev_run> will be async-cancel safe between
1036 root 1.254 invocations of C<release> and C<acquire>.
1037    
1038     See also the locking example in the C<THREADS> section later in this
1039     document.
1040    
1041 root 1.253 =item ev_set_userdata (loop, void *data)
1042    
1043 root 1.344 =item void *ev_userdata (loop)
1044 root 1.253
1045     Set and retrieve a single C<void *> associated with a loop. When
1046     C<ev_set_userdata> has never been called, then C<ev_userdata> returns
1047 root 1.342 C<0>.
1048 root 1.253
1049     These two functions can be used to associate arbitrary data with a loop,
1050     and are intended solely for the C<invoke_pending_cb>, C<release> and
1051     C<acquire> callbacks described above, but of course can be (ab-)used for
1052     any other purpose as well.
1053    
1054 root 1.309 =item ev_verify (loop)
1055 root 1.159
1056     This function only does something when C<EV_VERIFY> support has been
1057 root 1.200 compiled in, which is the default for non-minimal builds. It tries to go
1058 root 1.183 through all internal structures and checks them for validity. If anything
1059     is found to be inconsistent, it will print an error message to standard
1060     error and call C<abort ()>.
1061 root 1.159
1062     This can be used to catch bugs inside libev itself: under normal
1063     circumstances, this function will never abort as of course libev keeps its
1064     data structures consistent.
1065    
1066 root 1.1 =back
1067    
1068 root 1.42
1069 root 1.1 =head1 ANATOMY OF A WATCHER
1070    
1071 root 1.200 In the following description, uppercase C<TYPE> in names stands for the
1072     watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
1073     watchers and C<ev_io_start> for I/O watchers.
1074    
1075 root 1.311 A watcher is an opaque structure that you allocate and register to record
1076     your interest in some event. To make a concrete example, imagine you want
1077     to wait for STDIN to become readable, you would create an C<ev_io> watcher
1078     for that:
1079 root 1.1
1080 root 1.198 static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
1081 root 1.164 {
1082     ev_io_stop (w);
1083 root 1.310 ev_break (loop, EVBREAK_ALL);
1084 root 1.164 }
1085    
1086     struct ev_loop *loop = ev_default_loop (0);
1087 root 1.200
1088 root 1.198 ev_io stdin_watcher;
1089 root 1.200
1090 root 1.164 ev_init (&stdin_watcher, my_cb);
1091     ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
1092     ev_io_start (loop, &stdin_watcher);
1093 root 1.200
1094 root 1.310 ev_run (loop, 0);
1095 root 1.1
1096     As you can see, you are responsible for allocating the memory for your
1097 root 1.200 watcher structures (and it is I<usually> a bad idea to do this on the
1098     stack).
1099    
1100     Each watcher has an associated watcher structure (called C<struct ev_TYPE>
1101     or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
1102 root 1.1
1103 root 1.311 Each watcher structure must be initialised by a call to C<ev_init (watcher
1104     *, callback)>, which expects a callback to be provided. This callback is
1105     invoked each time the event occurs (or, in the case of I/O watchers, each
1106     time the event loop detects that the file descriptor given is readable
1107     and/or writable).
1108 root 1.1
1109 root 1.200 Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
1110     macro to configure it, with arguments specific to the watcher type. There
1111     is also a macro to combine initialisation and setting in one call: C<<
1112     ev_TYPE_init (watcher *, callback, ...) >>.
1113 root 1.1
1114     To make the watcher actually watch out for events, you have to start it
1115 root 1.200 with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
1116 root 1.1 *) >>), and you can stop watching for events at any time by calling the
1117 root 1.200 corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
1118 root 1.1
1119     As long as your watcher is active (has been started but not stopped) you
1120     must not touch the values stored in it. Most specifically you must never
1121 root 1.200 reinitialise it or call its C<ev_TYPE_set> macro.
1122 root 1.1
1123     Each and every callback receives the event loop pointer as first, the
1124     registered watcher structure as second, and a bitset of received events as
1125     third argument.
1126    
1127 root 1.14 The received events usually include a single bit per event type received
1128 root 1.1 (you can receive multiple events at the same time). The possible bit masks
1129     are:
1130    
1131     =over 4
1132    
1133 root 1.10 =item C<EV_READ>
1134 root 1.1
1135 root 1.10 =item C<EV_WRITE>
1136 root 1.1
1137 root 1.10 The file descriptor in the C<ev_io> watcher has become readable and/or
1138 root 1.1 writable.
1139    
1140 root 1.289 =item C<EV_TIMER>
1141 root 1.1
1142 root 1.10 The C<ev_timer> watcher has timed out.
1143 root 1.1
1144 root 1.10 =item C<EV_PERIODIC>
1145 root 1.1
1146 root 1.10 The C<ev_periodic> watcher has timed out.
1147 root 1.1
1148 root 1.10 =item C<EV_SIGNAL>
1149 root 1.1
1150 root 1.10 The signal specified in the C<ev_signal> watcher has been received by a thread.
1151 root 1.1
1152 root 1.10 =item C<EV_CHILD>
1153 root 1.1
1154 root 1.10 The pid specified in the C<ev_child> watcher has received a status change.
1155 root 1.1
1156 root 1.48 =item C<EV_STAT>
1157    
1158     The path specified in the C<ev_stat> watcher changed its attributes somehow.
1159    
1160 root 1.10 =item C<EV_IDLE>
1161 root 1.1
1162 root 1.10 The C<ev_idle> watcher has determined that you have nothing better to do.
1163 root 1.1
1164 root 1.10 =item C<EV_PREPARE>
1165 root 1.1
1166 root 1.10 =item C<EV_CHECK>
1167 root 1.1
1168 root 1.310 All C<ev_prepare> watchers are invoked just I<before> C<ev_run> starts
1169 root 1.10 to gather new events, and all C<ev_check> watchers are invoked just after
1170 root 1.310 C<ev_run> has gathered them, but before it invokes any callbacks for any
1171 root 1.1 received events. Callbacks of both watcher types can start and stop as
1172     many watchers as they want, and all of them will be taken into account
1173 root 1.10 (for example, a C<ev_prepare> watcher might start an idle watcher to keep
1174 root 1.310 C<ev_run> from blocking).
1175 root 1.1
1176 root 1.50 =item C<EV_EMBED>
1177    
1178     The embedded event loop specified in the C<ev_embed> watcher needs attention.
1179    
1180     =item C<EV_FORK>
1181    
1182     The event loop has been resumed in the child process after fork (see
1183     C<ev_fork>).
1184    
1185 root 1.324 =item C<EV_CLEANUP>
1186    
1187 sf-exg 1.330 The event loop is about to be destroyed (see C<ev_cleanup>).
1188 root 1.324
1189 root 1.122 =item C<EV_ASYNC>
1190    
1191     The given async watcher has been asynchronously notified (see C<ev_async>).
1192    
1193 root 1.229 =item C<EV_CUSTOM>
1194    
1195     Not ever sent (or otherwise used) by libev itself, but can be freely used
1196     by libev users to signal watchers (e.g. via C<ev_feed_event>).
1197    
1198 root 1.10 =item C<EV_ERROR>
1199 root 1.1
1200 root 1.161 An unspecified error has occurred, the watcher has been stopped. This might
1201 root 1.1 happen because the watcher could not be properly started because libev
1202     ran out of memory, a file descriptor was found to be closed or any other
1203 root 1.197 problem. Libev considers these application bugs.
1204    
1205     You best act on it by reporting the problem and somehow coping with the
1206     watcher being stopped. Note that well-written programs should not receive
1207     an error ever, so when your watcher receives it, this usually indicates a
1208     bug in your program.
1209 root 1.1
1210 root 1.183 Libev will usually signal a few "dummy" events together with an error, for
1211     example it might indicate that a fd is readable or writable, and if your
1212     callbacks is well-written it can just attempt the operation and cope with
1213     the error from read() or write(). This will not work in multi-threaded
1214     programs, though, as the fd could already be closed and reused for another
1215     thing, so beware.
1216 root 1.1
1217     =back
1218    
1219 root 1.42 =head2 GENERIC WATCHER FUNCTIONS
1220 root 1.36
1221     =over 4
1222    
1223     =item C<ev_init> (ev_TYPE *watcher, callback)
1224    
1225     This macro initialises the generic portion of a watcher. The contents
1226     of the watcher object can be arbitrary (so C<malloc> will do). Only
1227     the generic parts of the watcher are initialised, you I<need> to call
1228     the type-specific C<ev_TYPE_set> macro afterwards to initialise the
1229     type-specific parts. For each type there is also a C<ev_TYPE_init> macro
1230     which rolls both calls into one.
1231    
1232     You can reinitialise a watcher at any time as long as it has been stopped
1233     (or never started) and there are no pending events outstanding.
1234    
1235 root 1.198 The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
1236 root 1.36 int revents)>.
1237    
1238 root 1.183 Example: Initialise an C<ev_io> watcher in two steps.
1239    
1240     ev_io w;
1241     ev_init (&w, my_cb);
1242     ev_io_set (&w, STDIN_FILENO, EV_READ);
1243    
1244 root 1.274 =item C<ev_TYPE_set> (ev_TYPE *watcher, [args])
1245 root 1.36
1246     This macro initialises the type-specific parts of a watcher. You need to
1247     call C<ev_init> at least once before you call this macro, but you can
1248     call C<ev_TYPE_set> any number of times. You must not, however, call this
1249     macro on a watcher that is active (it can be pending, however, which is a
1250     difference to the C<ev_init> macro).
1251    
1252     Although some watcher types do not have type-specific arguments
1253     (e.g. C<ev_prepare>) you still need to call its C<set> macro.
1254    
1255 root 1.183 See C<ev_init>, above, for an example.
1256    
1257 root 1.36 =item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
1258    
1259 root 1.161 This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
1260     calls into a single call. This is the most convenient method to initialise
1261 root 1.36 a watcher. The same limitations apply, of course.
1262    
1263 root 1.183 Example: Initialise and set an C<ev_io> watcher in one step.
1264    
1265     ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
1266    
1267 root 1.274 =item C<ev_TYPE_start> (loop, ev_TYPE *watcher)
1268 root 1.36
1269     Starts (activates) the given watcher. Only active watchers will receive
1270     events. If the watcher is already active nothing will happen.
1271    
1272 root 1.183 Example: Start the C<ev_io> watcher that is being abused as example in this
1273     whole section.
1274    
1275     ev_io_start (EV_DEFAULT_UC, &w);
1276    
1277 root 1.274 =item C<ev_TYPE_stop> (loop, ev_TYPE *watcher)
1278 root 1.36
1279 root 1.195 Stops the given watcher if active, and clears the pending status (whether
1280     the watcher was active or not).
1281    
1282     It is possible that stopped watchers are pending - for example,
1283     non-repeating timers are being stopped when they become pending - but
1284     calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
1285     pending. If you want to free or reuse the memory used by the watcher it is
1286     therefore a good idea to always call its C<ev_TYPE_stop> function.
1287 root 1.36
1288     =item bool ev_is_active (ev_TYPE *watcher)
1289    
1290     Returns a true value iff the watcher is active (i.e. it has been started
1291     and not yet been stopped). As long as a watcher is active you must not modify
1292     it.
1293    
1294     =item bool ev_is_pending (ev_TYPE *watcher)
1295    
1296     Returns a true value iff the watcher is pending, (i.e. it has outstanding
1297     events but its callback has not yet been invoked). As long as a watcher
1298     is pending (but not active) you must not call an init function on it (but
1299 root 1.73 C<ev_TYPE_set> is safe), you must not change its priority, and you must
1300     make sure the watcher is available to libev (e.g. you cannot C<free ()>
1301     it).
1302 root 1.36
1303 root 1.55 =item callback ev_cb (ev_TYPE *watcher)
1304 root 1.36
1305     Returns the callback currently set on the watcher.
1306    
1307     =item ev_cb_set (ev_TYPE *watcher, callback)
1308    
1309     Change the callback. You can change the callback at virtually any time
1310     (modulo threads).
1311    
1312 root 1.274 =item ev_set_priority (ev_TYPE *watcher, int priority)
1313 root 1.67
1314     =item int ev_priority (ev_TYPE *watcher)
1315    
1316     Set and query the priority of the watcher. The priority is a small
1317     integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
1318     (default: C<-2>). Pending watchers with higher priority will be invoked
1319     before watchers with lower priority, but priority will not keep watchers
1320     from being executed (except for C<ev_idle> watchers).
1321    
1322     If you need to suppress invocation when higher priority events are pending
1323     you need to look at C<ev_idle> watchers, which provide this functionality.
1324    
1325 root 1.73 You I<must not> change the priority of a watcher as long as it is active or
1326     pending.
1327    
1328 root 1.233 Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
1329     fine, as long as you do not mind that the priority value you query might
1330     or might not have been clamped to the valid range.
1331    
1332 root 1.67 The default priority used by watchers when no priority has been set is
1333     always C<0>, which is supposed to not be too high and not be too low :).
1334    
1335 root 1.235 See L<WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
1336 root 1.233 priorities.
1337 root 1.67
1338 root 1.74 =item ev_invoke (loop, ev_TYPE *watcher, int revents)
1339    
1340     Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
1341     C<loop> nor C<revents> need to be valid as long as the watcher callback
1342 root 1.183 can deal with that fact, as both are simply passed through to the
1343     callback.
1344 root 1.74
1345     =item int ev_clear_pending (loop, ev_TYPE *watcher)
1346    
1347 root 1.183 If the watcher is pending, this function clears its pending status and
1348     returns its C<revents> bitset (as if its callback was invoked). If the
1349 root 1.74 watcher isn't pending it does nothing and returns C<0>.
1350    
1351 root 1.183 Sometimes it can be useful to "poll" a watcher instead of waiting for its
1352     callback to be invoked, which can be accomplished with this function.
1353    
1354 root 1.274 =item ev_feed_event (loop, ev_TYPE *watcher, int revents)
1355 root 1.273
1356     Feeds the given event set into the event loop, as if the specified event
1357     had happened for the specified watcher (which must be a pointer to an
1358     initialised but not necessarily started event watcher). Obviously you must
1359     not free the watcher as long as it has pending events.
1360    
1361     Stopping the watcher, letting libev invoke it, or calling
1362     C<ev_clear_pending> will clear the pending event, even if the watcher was
1363     not started in the first place.
1364    
1365     See also C<ev_feed_fd_event> and C<ev_feed_signal_event> for related
1366     functions that do not need a watcher.
1367    
1368 root 1.36 =back
1369    
1370 root 1.357 See also the L<ASSOCIATING CUSTOM DATA WITH A WATCHER> and L<BUILDING YOUR
1371     OWN COMPOSITE WATCHERS> idioms.
1372 root 1.1
1373 root 1.335 =head2 WATCHER STATES
1374    
1375     There are various watcher states mentioned throughout this manual -
1376     active, pending and so on. In this section these states and the rules to
1377     transition between them will be described in more detail - and while these
1378     rules might look complicated, they usually do "the right thing".
1379    
1380     =over 4
1381    
1382     =item initialiased
1383    
1384     Before a watcher can be registered with the event looop it has to be
1385     initialised. This can be done with a call to C<ev_TYPE_init>, or calls to
1386     C<ev_init> followed by the watcher-specific C<ev_TYPE_set> function.
1387    
1388 root 1.361 In this state it is simply some block of memory that is suitable for
1389     use in an event loop. It can be moved around, freed, reused etc. at
1390     will - as long as you either keep the memory contents intact, or call
1391     C<ev_TYPE_init> again.
1392 root 1.335
1393     =item started/running/active
1394    
1395     Once a watcher has been started with a call to C<ev_TYPE_start> it becomes
1396     property of the event loop, and is actively waiting for events. While in
1397     this state it cannot be accessed (except in a few documented ways), moved,
1398     freed or anything else - the only legal thing is to keep a pointer to it,
1399     and call libev functions on it that are documented to work on active watchers.
1400    
1401     =item pending
1402    
1403     If a watcher is active and libev determines that an event it is interested
1404     in has occurred (such as a timer expiring), it will become pending. It will
1405     stay in this pending state until either it is stopped or its callback is
1406     about to be invoked, so it is not normally pending inside the watcher
1407     callback.
1408    
1409     The watcher might or might not be active while it is pending (for example,
1410     an expired non-repeating timer can be pending but no longer active). If it
1411     is stopped, it can be freely accessed (e.g. by calling C<ev_TYPE_set>),
1412     but it is still property of the event loop at this time, so cannot be
1413     moved, freed or reused. And if it is active the rules described in the
1414     previous item still apply.
1415    
1416     It is also possible to feed an event on a watcher that is not active (e.g.
1417     via C<ev_feed_event>), in which case it becomes pending without being
1418     active.
1419    
1420     =item stopped
1421    
1422     A watcher can be stopped implicitly by libev (in which case it might still
1423     be pending), or explicitly by calling its C<ev_TYPE_stop> function. The
1424     latter will clear any pending state the watcher might be in, regardless
1425     of whether it was active or not, so stopping a watcher explicitly before
1426     freeing it is often a good idea.
1427    
1428     While stopped (and not pending) the watcher is essentially in the
1429 root 1.361 initialised state, that is, it can be reused, moved, modified in any way
1430     you wish (but when you trash the memory block, you need to C<ev_TYPE_init>
1431     it again).
1432 root 1.335
1433     =back
1434    
1435 root 1.233 =head2 WATCHER PRIORITY MODELS
1436    
1437     Many event loops support I<watcher priorities>, which are usually small
1438     integers that influence the ordering of event callback invocation
1439     between watchers in some way, all else being equal.
1440    
1441     In libev, Watcher priorities can be set using C<ev_set_priority>. See its
1442     description for the more technical details such as the actual priority
1443     range.
1444    
1445     There are two common ways how these these priorities are being interpreted
1446     by event loops:
1447    
1448     In the more common lock-out model, higher priorities "lock out" invocation
1449     of lower priority watchers, which means as long as higher priority
1450     watchers receive events, lower priority watchers are not being invoked.
1451    
1452     The less common only-for-ordering model uses priorities solely to order
1453     callback invocation within a single event loop iteration: Higher priority
1454     watchers are invoked before lower priority ones, but they all get invoked
1455     before polling for new events.
1456    
1457     Libev uses the second (only-for-ordering) model for all its watchers
1458     except for idle watchers (which use the lock-out model).
1459    
1460     The rationale behind this is that implementing the lock-out model for
1461     watchers is not well supported by most kernel interfaces, and most event
1462     libraries will just poll for the same events again and again as long as
1463     their callbacks have not been executed, which is very inefficient in the
1464     common case of one high-priority watcher locking out a mass of lower
1465     priority ones.
1466    
1467     Static (ordering) priorities are most useful when you have two or more
1468     watchers handling the same resource: a typical usage example is having an
1469     C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
1470     timeouts. Under load, data might be received while the program handles
1471     other jobs, but since timers normally get invoked first, the timeout
1472     handler will be executed before checking for data. In that case, giving
1473     the timer a lower priority than the I/O watcher ensures that I/O will be
1474     handled first even under adverse conditions (which is usually, but not
1475     always, what you want).
1476    
1477     Since idle watchers use the "lock-out" model, meaning that idle watchers
1478     will only be executed when no same or higher priority watchers have
1479     received events, they can be used to implement the "lock-out" model when
1480     required.
1481    
1482     For example, to emulate how many other event libraries handle priorities,
1483     you can associate an C<ev_idle> watcher to each such watcher, and in
1484     the normal watcher callback, you just start the idle watcher. The real
1485     processing is done in the idle watcher callback. This causes libev to
1486 sf-exg 1.298 continuously poll and process kernel event data for the watcher, but when
1487 root 1.233 the lock-out case is known to be rare (which in turn is rare :), this is
1488     workable.
1489    
1490     Usually, however, the lock-out model implemented that way will perform
1491     miserably under the type of load it was designed to handle. In that case,
1492     it might be preferable to stop the real watcher before starting the
1493     idle watcher, so the kernel will not have to process the event in case
1494     the actual processing will be delayed for considerable time.
1495    
1496     Here is an example of an I/O watcher that should run at a strictly lower
1497     priority than the default, and which should only process data when no
1498     other events are pending:
1499    
1500     ev_idle idle; // actual processing watcher
1501     ev_io io; // actual event watcher
1502    
1503     static void
1504     io_cb (EV_P_ ev_io *w, int revents)
1505     {
1506     // stop the I/O watcher, we received the event, but
1507     // are not yet ready to handle it.
1508     ev_io_stop (EV_A_ w);
1509    
1510 root 1.296 // start the idle watcher to handle the actual event.
1511 root 1.233 // it will not be executed as long as other watchers
1512     // with the default priority are receiving events.
1513     ev_idle_start (EV_A_ &idle);
1514     }
1515    
1516     static void
1517 root 1.242 idle_cb (EV_P_ ev_idle *w, int revents)
1518 root 1.233 {
1519     // actual processing
1520     read (STDIN_FILENO, ...);
1521    
1522     // have to start the I/O watcher again, as
1523     // we have handled the event
1524     ev_io_start (EV_P_ &io);
1525     }
1526    
1527     // initialisation
1528     ev_idle_init (&idle, idle_cb);
1529     ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
1530     ev_io_start (EV_DEFAULT_ &io);
1531    
1532     In the "real" world, it might also be beneficial to start a timer, so that
1533     low-priority connections can not be locked out forever under load. This
1534     enables your program to keep a lower latency for important connections
1535     during short periods of high load, while not completely locking out less
1536     important ones.
1537    
1538 root 1.1
1539     =head1 WATCHER TYPES
1540    
1541     This section describes each watcher in detail, but will not repeat
1542 root 1.48 information given in the last section. Any initialisation/set macros,
1543     functions and members specific to the watcher type are explained.
1544    
1545     Members are additionally marked with either I<[read-only]>, meaning that,
1546     while the watcher is active, you can look at the member and expect some
1547     sensible content, but you must not modify it (you can modify it while the
1548     watcher is stopped to your hearts content), or I<[read-write]>, which
1549     means you can expect it to have some sensible content while the watcher
1550     is active, but you can also modify it. Modifying it may not do something
1551     sensible or take immediate effect (or do anything at all), but libev will
1552     not crash or malfunction in any way.
1553 root 1.1
1554 root 1.34
1555 root 1.42 =head2 C<ev_io> - is this file descriptor readable or writable?
1556 root 1.1
1557 root 1.4 I/O watchers check whether a file descriptor is readable or writable
1558 root 1.42 in each iteration of the event loop, or, more precisely, when reading
1559     would not block the process and writing would at least be able to write
1560     some data. This behaviour is called level-triggering because you keep
1561     receiving events as long as the condition persists. Remember you can stop
1562     the watcher if you don't want to act on the event and neither want to
1563     receive future events.
1564 root 1.1
1565 root 1.23 In general you can register as many read and/or write event watchers per
1566 root 1.8 fd as you want (as long as you don't confuse yourself). Setting all file
1567     descriptors to non-blocking mode is also usually a good idea (but not
1568     required if you know what you are doing).
1569    
1570 root 1.42 Another thing you have to watch out for is that it is quite easy to
1571 root 1.354 receive "spurious" readiness notifications, that is, your callback might
1572 root 1.42 be called with C<EV_READ> but a subsequent C<read>(2) will actually block
1573 root 1.354 because there is no data. It is very easy to get into this situation even
1574     with a relatively standard program structure. Thus it is best to always
1575     use non-blocking I/O: An extra C<read>(2) returning C<EAGAIN> is far
1576     preferable to a program hanging until some data arrives.
1577 root 1.42
1578 root 1.183 If you cannot run the fd in non-blocking mode (for example you should
1579     not play around with an Xlib connection), then you have to separately
1580     re-test whether a file descriptor is really ready with a known-to-be good
1581 root 1.354 interface such as poll (fortunately in the case of Xlib, it already does
1582     this on its own, so its quite safe to use). Some people additionally
1583 root 1.183 use C<SIGALRM> and an interval timer, just to be sure you won't block
1584     indefinitely.
1585    
1586     But really, best use non-blocking mode.
1587 root 1.42
1588 root 1.81 =head3 The special problem of disappearing file descriptors
1589    
1590 root 1.94 Some backends (e.g. kqueue, epoll) need to be told about closing a file
1591 root 1.183 descriptor (either due to calling C<close> explicitly or any other means,
1592     such as C<dup2>). The reason is that you register interest in some file
1593 root 1.81 descriptor, but when it goes away, the operating system will silently drop
1594     this interest. If another file descriptor with the same number then is
1595     registered with libev, there is no efficient way to see that this is, in
1596     fact, a different file descriptor.
1597    
1598     To avoid having to explicitly tell libev about such cases, libev follows
1599     the following policy: Each time C<ev_io_set> is being called, libev
1600     will assume that this is potentially a new file descriptor, otherwise
1601     it is assumed that the file descriptor stays the same. That means that
1602     you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
1603     descriptor even if the file descriptor number itself did not change.
1604    
1605     This is how one would do it normally anyway, the important point is that
1606     the libev application should not optimise around libev but should leave
1607     optimisations to libev.
1608    
1609 root 1.95 =head3 The special problem of dup'ed file descriptors
1610 root 1.94
1611     Some backends (e.g. epoll), cannot register events for file descriptors,
1612 root 1.103 but only events for the underlying file descriptions. That means when you
1613 root 1.109 have C<dup ()>'ed file descriptors or weirder constellations, and register
1614     events for them, only one file descriptor might actually receive events.
1615 root 1.94
1616 root 1.103 There is no workaround possible except not registering events
1617     for potentially C<dup ()>'ed file descriptors, or to resort to
1618 root 1.94 C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
1619    
1620 root 1.354 =head3 The special problem of files
1621    
1622     Many people try to use C<select> (or libev) on file descriptors
1623     representing files, and expect it to become ready when their program
1624     doesn't block on disk accesses (which can take a long time on their own).
1625    
1626     However, this cannot ever work in the "expected" way - you get a readiness
1627     notification as soon as the kernel knows whether and how much data is
1628     there, and in the case of open files, that's always the case, so you
1629     always get a readiness notification instantly, and your read (or possibly
1630     write) will still block on the disk I/O.
1631    
1632     Another way to view it is that in the case of sockets, pipes, character
1633     devices and so on, there is another party (the sender) that delivers data
1634 sf-exg 1.358 on its own, but in the case of files, there is no such thing: the disk
1635     will not send data on its own, simply because it doesn't know what you
1636 root 1.354 wish to read - you would first have to request some data.
1637    
1638     Since files are typically not-so-well supported by advanced notification
1639     mechanism, libev tries hard to emulate POSIX behaviour with respect
1640     to files, even though you should not use it. The reason for this is
1641     convenience: sometimes you want to watch STDIN or STDOUT, which is
1642     usually a tty, often a pipe, but also sometimes files or special devices
1643     (for example, C<epoll> on Linux works with F</dev/random> but not with
1644     F</dev/urandom>), and even though the file might better be served with
1645     asynchronous I/O instead of with non-blocking I/O, it is still useful when
1646     it "just works" instead of freezing.
1647    
1648     So avoid file descriptors pointing to files when you know it (e.g. use
1649     libeio), but use them when it is convenient, e.g. for STDIN/STDOUT, or
1650     when you rarely read from a file instead of from a socket, and want to
1651     reuse the same code path.
1652    
1653 root 1.94 =head3 The special problem of fork
1654    
1655     Some backends (epoll, kqueue) do not support C<fork ()> at all or exhibit
1656     useless behaviour. Libev fully supports fork, but needs to be told about
1657 root 1.354 it in the child if you want to continue to use it in the child.
1658 root 1.94
1659 root 1.354 To support fork in your child processes, you have to call C<ev_loop_fork
1660     ()> after a fork in the child, enable C<EVFLAG_FORKCHECK>, or resort to
1661     C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
1662 root 1.94
1663 root 1.138 =head3 The special problem of SIGPIPE
1664    
1665 root 1.183 While not really specific to libev, it is easy to forget about C<SIGPIPE>:
1666 root 1.174 when writing to a pipe whose other end has been closed, your program gets
1667 root 1.183 sent a SIGPIPE, which, by default, aborts your program. For most programs
1668 root 1.174 this is sensible behaviour, for daemons, this is usually undesirable.
1669 root 1.138
1670     So when you encounter spurious, unexplained daemon exits, make sure you
1671     ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
1672     somewhere, as that would have given you a big clue).
1673    
1674 root 1.284 =head3 The special problem of accept()ing when you can't
1675    
1676     Many implementations of the POSIX C<accept> function (for example,
1677 sf-exg 1.292 found in post-2004 Linux) have the peculiar behaviour of not removing a
1678 root 1.284 connection from the pending queue in all error cases.
1679    
1680     For example, larger servers often run out of file descriptors (because
1681     of resource limits), causing C<accept> to fail with C<ENFILE> but not
1682     rejecting the connection, leading to libev signalling readiness on
1683     the next iteration again (the connection still exists after all), and
1684     typically causing the program to loop at 100% CPU usage.
1685    
1686     Unfortunately, the set of errors that cause this issue differs between
1687     operating systems, there is usually little the app can do to remedy the
1688     situation, and no known thread-safe method of removing the connection to
1689     cope with overload is known (to me).
1690    
1691     One of the easiest ways to handle this situation is to just ignore it
1692     - when the program encounters an overload, it will just loop until the
1693     situation is over. While this is a form of busy waiting, no OS offers an
1694     event-based way to handle this situation, so it's the best one can do.
1695    
1696     A better way to handle the situation is to log any errors other than
1697     C<EAGAIN> and C<EWOULDBLOCK>, making sure not to flood the log with such
1698     messages, and continue as usual, which at least gives the user an idea of
1699     what could be wrong ("raise the ulimit!"). For extra points one could stop
1700     the C<ev_io> watcher on the listening fd "for a while", which reduces CPU
1701     usage.
1702    
1703     If your program is single-threaded, then you could also keep a dummy file
1704     descriptor for overload situations (e.g. by opening F</dev/null>), and
1705     when you run into C<ENFILE> or C<EMFILE>, close it, run C<accept>,
1706     close that fd, and create a new dummy fd. This will gracefully refuse
1707     clients under typical overload conditions.
1708    
1709     The last way to handle it is to simply log the error and C<exit>, as
1710     is often done with C<malloc> failures, but this results in an easy
1711     opportunity for a DoS attack.
1712 root 1.81
1713 root 1.82 =head3 Watcher-Specific Functions
1714    
1715 root 1.1 =over 4
1716    
1717     =item ev_io_init (ev_io *, callback, int fd, int events)
1718    
1719     =item ev_io_set (ev_io *, int fd, int events)
1720    
1721 root 1.42 Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
1722 root 1.183 receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
1723     C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
1724 root 1.32
1725 root 1.48 =item int fd [read-only]
1726    
1727     The file descriptor being watched.
1728    
1729     =item int events [read-only]
1730    
1731     The events being watched.
1732    
1733 root 1.1 =back
1734    
1735 root 1.111 =head3 Examples
1736    
1737 root 1.54 Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
1738 root 1.34 readable, but only once. Since it is likely line-buffered, you could
1739 root 1.54 attempt to read a whole line in the callback.
1740 root 1.34
1741 root 1.164 static void
1742 root 1.198 stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
1743 root 1.164 {
1744     ev_io_stop (loop, w);
1745 root 1.183 .. read from stdin here (or from w->fd) and handle any I/O errors
1746 root 1.164 }
1747    
1748     ...
1749     struct ev_loop *loop = ev_default_init (0);
1750 root 1.198 ev_io stdin_readable;
1751 root 1.164 ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
1752     ev_io_start (loop, &stdin_readable);
1753 root 1.310 ev_run (loop, 0);
1754 root 1.34
1755    
1756 root 1.42 =head2 C<ev_timer> - relative and optionally repeating timeouts
1757 root 1.1
1758     Timer watchers are simple relative timers that generate an event after a
1759     given time, and optionally repeating in regular intervals after that.
1760    
1761     The timers are based on real time, that is, if you register an event that
1762 root 1.161 times out after an hour and you reset your system clock to January last
1763 root 1.183 year, it will still time out after (roughly) one hour. "Roughly" because
1764 root 1.28 detecting time jumps is hard, and some inaccuracies are unavoidable (the
1765 root 1.1 monotonic clock option helps a lot here).
1766    
1767 root 1.183 The callback is guaranteed to be invoked only I<after> its timeout has
1768 root 1.240 passed (not I<at>, so on systems with very low-resolution clocks this
1769     might introduce a small delay). If multiple timers become ready during the
1770     same loop iteration then the ones with earlier time-out values are invoked
1771 root 1.248 before ones of the same priority with later time-out values (but this is
1772 root 1.310 no longer true when a callback calls C<ev_run> recursively).
1773 root 1.175
1774 root 1.198 =head3 Be smart about timeouts
1775    
1776 root 1.199 Many real-world problems involve some kind of timeout, usually for error
1777 root 1.198 recovery. A typical example is an HTTP request - if the other side hangs,
1778     you want to raise some error after a while.
1779    
1780 root 1.199 What follows are some ways to handle this problem, from obvious and
1781     inefficient to smart and efficient.
1782 root 1.198
1783 root 1.199 In the following, a 60 second activity timeout is assumed - a timeout that
1784     gets reset to 60 seconds each time there is activity (e.g. each time some
1785     data or other life sign was received).
1786 root 1.198
1787     =over 4
1788    
1789 root 1.199 =item 1. Use a timer and stop, reinitialise and start it on activity.
1790 root 1.198
1791     This is the most obvious, but not the most simple way: In the beginning,
1792     start the watcher:
1793    
1794     ev_timer_init (timer, callback, 60., 0.);
1795     ev_timer_start (loop, timer);
1796    
1797 root 1.199 Then, each time there is some activity, C<ev_timer_stop> it, initialise it
1798     and start it again:
1799 root 1.198
1800     ev_timer_stop (loop, timer);
1801     ev_timer_set (timer, 60., 0.);
1802     ev_timer_start (loop, timer);
1803    
1804 root 1.199 This is relatively simple to implement, but means that each time there is
1805     some activity, libev will first have to remove the timer from its internal
1806     data structure and then add it again. Libev tries to be fast, but it's
1807     still not a constant-time operation.
1808 root 1.198
1809     =item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
1810    
1811     This is the easiest way, and involves using C<ev_timer_again> instead of
1812     C<ev_timer_start>.
1813    
1814 root 1.199 To implement this, configure an C<ev_timer> with a C<repeat> value
1815     of C<60> and then call C<ev_timer_again> at start and each time you
1816     successfully read or write some data. If you go into an idle state where
1817     you do not expect data to travel on the socket, you can C<ev_timer_stop>
1818     the timer, and C<ev_timer_again> will automatically restart it if need be.
1819    
1820     That means you can ignore both the C<ev_timer_start> function and the
1821     C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
1822     member and C<ev_timer_again>.
1823 root 1.198
1824     At start:
1825    
1826 root 1.243 ev_init (timer, callback);
1827 root 1.199 timer->repeat = 60.;
1828 root 1.198 ev_timer_again (loop, timer);
1829    
1830 root 1.199 Each time there is some activity:
1831 root 1.198
1832     ev_timer_again (loop, timer);
1833    
1834 root 1.199 It is even possible to change the time-out on the fly, regardless of
1835     whether the watcher is active or not:
1836 root 1.198
1837     timer->repeat = 30.;
1838     ev_timer_again (loop, timer);
1839    
1840     This is slightly more efficient then stopping/starting the timer each time
1841     you want to modify its timeout value, as libev does not have to completely
1842 root 1.199 remove and re-insert the timer from/into its internal data structure.
1843    
1844     It is, however, even simpler than the "obvious" way to do it.
1845 root 1.198
1846     =item 3. Let the timer time out, but then re-arm it as required.
1847    
1848     This method is more tricky, but usually most efficient: Most timeouts are
1849 root 1.199 relatively long compared to the intervals between other activity - in
1850     our example, within 60 seconds, there are usually many I/O events with
1851     associated activity resets.
1852 root 1.198
1853     In this case, it would be more efficient to leave the C<ev_timer> alone,
1854     but remember the time of last activity, and check for a real timeout only
1855     within the callback:
1856    
1857     ev_tstamp last_activity; // time of last activity
1858    
1859     static void
1860     callback (EV_P_ ev_timer *w, int revents)
1861     {
1862 root 1.199 ev_tstamp now = ev_now (EV_A);
1863 root 1.198 ev_tstamp timeout = last_activity + 60.;
1864    
1865 root 1.199 // if last_activity + 60. is older than now, we did time out
1866 root 1.198 if (timeout < now)
1867     {
1868 sf-exg 1.298 // timeout occurred, take action
1869 root 1.198 }
1870     else
1871     {
1872     // callback was invoked, but there was some activity, re-arm
1873 root 1.199 // the watcher to fire in last_activity + 60, which is
1874     // guaranteed to be in the future, so "again" is positive:
1875 root 1.216 w->repeat = timeout - now;
1876 root 1.198 ev_timer_again (EV_A_ w);
1877     }
1878     }
1879    
1880 root 1.199 To summarise the callback: first calculate the real timeout (defined
1881     as "60 seconds after the last activity"), then check if that time has
1882     been reached, which means something I<did>, in fact, time out. Otherwise
1883     the callback was invoked too early (C<timeout> is in the future), so
1884     re-schedule the timer to fire at that future time, to see if maybe we have
1885     a timeout then.
1886 root 1.198
1887     Note how C<ev_timer_again> is used, taking advantage of the
1888     C<ev_timer_again> optimisation when the timer is already running.
1889    
1890 root 1.199 This scheme causes more callback invocations (about one every 60 seconds
1891     minus half the average time between activity), but virtually no calls to
1892     libev to change the timeout.
1893    
1894     To start the timer, simply initialise the watcher and set C<last_activity>
1895     to the current time (meaning we just have some activity :), then call the
1896     callback, which will "do the right thing" and start the timer:
1897 root 1.198
1898 root 1.243 ev_init (timer, callback);
1899 root 1.198 last_activity = ev_now (loop);
1900 root 1.289 callback (loop, timer, EV_TIMER);
1901 root 1.198
1902 root 1.199 And when there is some activity, simply store the current time in
1903     C<last_activity>, no libev calls at all:
1904 root 1.198
1905 root 1.295 last_activity = ev_now (loop);
1906 root 1.198
1907     This technique is slightly more complex, but in most cases where the
1908     time-out is unlikely to be triggered, much more efficient.
1909    
1910 root 1.199 Changing the timeout is trivial as well (if it isn't hard-coded in the
1911     callback :) - just change the timeout and invoke the callback, which will
1912     fix things for you.
1913    
1914 root 1.200 =item 4. Wee, just use a double-linked list for your timeouts.
1915 root 1.199
1916 root 1.200 If there is not one request, but many thousands (millions...), all
1917     employing some kind of timeout with the same timeout value, then one can
1918     do even better:
1919 root 1.199
1920     When starting the timeout, calculate the timeout value and put the timeout
1921     at the I<end> of the list.
1922    
1923     Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
1924     the list is expected to fire (for example, using the technique #3).
1925    
1926     When there is some activity, remove the timer from the list, recalculate
1927     the timeout, append it to the end of the list again, and make sure to
1928     update the C<ev_timer> if it was taken from the beginning of the list.
1929    
1930     This way, one can manage an unlimited number of timeouts in O(1) time for
1931     starting, stopping and updating the timers, at the expense of a major
1932     complication, and having to use a constant timeout. The constant timeout
1933     ensures that the list stays sorted.
1934    
1935 root 1.198 =back
1936    
1937 root 1.200 So which method the best?
1938 root 1.199
1939 root 1.200 Method #2 is a simple no-brain-required solution that is adequate in most
1940     situations. Method #3 requires a bit more thinking, but handles many cases
1941     better, and isn't very complicated either. In most case, choosing either
1942     one is fine, with #3 being better in typical situations.
1943 root 1.199
1944     Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
1945     rather complicated, but extremely efficient, something that really pays
1946 root 1.200 off after the first million or so of active timers, i.e. it's usually
1947 root 1.199 overkill :)
1948    
1949 root 1.175 =head3 The special problem of time updates
1950    
1951 root 1.176 Establishing the current time is a costly operation (it usually takes at
1952     least two system calls): EV therefore updates its idea of the current
1953 root 1.310 time only before and after C<ev_run> collects new events, which causes a
1954 root 1.183 growing difference between C<ev_now ()> and C<ev_time ()> when handling
1955     lots of events in one iteration.
1956 root 1.175
1957 root 1.9 The relative timeouts are calculated relative to the C<ev_now ()>
1958     time. This is usually the right thing as this timestamp refers to the time
1959 root 1.28 of the event triggering whatever timeout you are modifying/starting. If
1960 root 1.175 you suspect event processing to be delayed and you I<need> to base the
1961     timeout on the current time, use something like this to adjust for this:
1962 root 1.9
1963     ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);
1964    
1965 root 1.177 If the event loop is suspended for a long time, you can also force an
1966 root 1.176 update of the time returned by C<ev_now ()> by calling C<ev_now_update
1967     ()>.
1968    
1969 root 1.257 =head3 The special problems of suspended animation
1970    
1971     When you leave the server world it is quite customary to hit machines that
1972     can suspend/hibernate - what happens to the clocks during such a suspend?
1973    
1974     Some quick tests made with a Linux 2.6.28 indicate that a suspend freezes
1975     all processes, while the clocks (C<times>, C<CLOCK_MONOTONIC>) continue
1976     to run until the system is suspended, but they will not advance while the
1977     system is suspended. That means, on resume, it will be as if the program
1978     was frozen for a few seconds, but the suspend time will not be counted
1979     towards C<ev_timer> when a monotonic clock source is used. The real time
1980     clock advanced as expected, but if it is used as sole clocksource, then a
1981     long suspend would be detected as a time jump by libev, and timers would
1982     be adjusted accordingly.
1983    
1984     I would not be surprised to see different behaviour in different between
1985     operating systems, OS versions or even different hardware.
1986    
1987     The other form of suspend (job control, or sending a SIGSTOP) will see a
1988     time jump in the monotonic clocks and the realtime clock. If the program
1989     is suspended for a very long time, and monotonic clock sources are in use,
1990     then you can expect C<ev_timer>s to expire as the full suspension time
1991     will be counted towards the timers. When no monotonic clock source is in
1992     use, then libev will again assume a timejump and adjust accordingly.
1993    
1994     It might be beneficial for this latter case to call C<ev_suspend>
1995     and C<ev_resume> in code that handles C<SIGTSTP>, to at least get
1996     deterministic behaviour in this case (you can do nothing against
1997     C<SIGSTOP>).
1998    
1999 root 1.82 =head3 Watcher-Specific Functions and Data Members
2000    
2001 root 1.1 =over 4
2002    
2003     =item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
2004    
2005     =item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
2006    
2007 root 1.157 Configure the timer to trigger after C<after> seconds. If C<repeat>
2008     is C<0.>, then it will automatically be stopped once the timeout is
2009     reached. If it is positive, then the timer will automatically be
2010     configured to trigger again C<repeat> seconds later, again, and again,
2011     until stopped manually.
2012    
2013     The timer itself will do a best-effort at avoiding drift, that is, if
2014     you configure a timer to trigger every 10 seconds, then it will normally
2015     trigger at exactly 10 second intervals. If, however, your program cannot
2016     keep up with the timer (because it takes longer than those 10 seconds to
2017     do stuff) the timer will not fire more than once per event loop iteration.
2018 root 1.1
2019 root 1.132 =item ev_timer_again (loop, ev_timer *)
2020 root 1.1
2021     This will act as if the timer timed out and restart it again if it is
2022     repeating. The exact semantics are:
2023    
2024 root 1.61 If the timer is pending, its pending status is cleared.
2025 root 1.1
2026 root 1.161 If the timer is started but non-repeating, stop it (as if it timed out).
2027 root 1.61
2028     If the timer is repeating, either start it if necessary (with the
2029     C<repeat> value), or reset the running timer to the C<repeat> value.
2030 root 1.1
2031 root 1.232 This sounds a bit complicated, see L<Be smart about timeouts>, above, for a
2032 root 1.198 usage example.
2033 root 1.183
2034 root 1.275 =item ev_tstamp ev_timer_remaining (loop, ev_timer *)
2035 root 1.258
2036     Returns the remaining time until a timer fires. If the timer is active,
2037     then this time is relative to the current event loop time, otherwise it's
2038     the timeout value currently configured.
2039    
2040     That is, after an C<ev_timer_set (w, 5, 7)>, C<ev_timer_remaining> returns
2041 sf-exg 1.280 C<5>. When the timer is started and one second passes, C<ev_timer_remaining>
2042 root 1.258 will return C<4>. When the timer expires and is restarted, it will return
2043     roughly C<7> (likely slightly less as callback invocation takes some time,
2044     too), and so on.
2045    
2046 root 1.48 =item ev_tstamp repeat [read-write]
2047    
2048     The current C<repeat> value. Will be used each time the watcher times out
2049 root 1.183 or C<ev_timer_again> is called, and determines the next timeout (if any),
2050 root 1.48 which is also when any modifications are taken into account.
2051 root 1.1
2052     =back
2053    
2054 root 1.111 =head3 Examples
2055    
2056 root 1.54 Example: Create a timer that fires after 60 seconds.
2057 root 1.34
2058 root 1.164 static void
2059 root 1.198 one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
2060 root 1.164 {
2061     .. one minute over, w is actually stopped right here
2062     }
2063    
2064 root 1.198 ev_timer mytimer;
2065 root 1.164 ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
2066     ev_timer_start (loop, &mytimer);
2067 root 1.34
2068 root 1.54 Example: Create a timeout timer that times out after 10 seconds of
2069 root 1.34 inactivity.
2070    
2071 root 1.164 static void
2072 root 1.198 timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
2073 root 1.164 {
2074     .. ten seconds without any activity
2075     }
2076    
2077 root 1.198 ev_timer mytimer;
2078 root 1.164 ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
2079     ev_timer_again (&mytimer); /* start timer */
2080 root 1.310 ev_run (loop, 0);
2081 root 1.164
2082     // and in some piece of code that gets executed on any "activity":
2083     // reset the timeout to start ticking again at 10 seconds
2084     ev_timer_again (&mytimer);
2085 root 1.34
2086    
2087 root 1.42 =head2 C<ev_periodic> - to cron or not to cron?
2088 root 1.1
2089     Periodic watchers are also timers of a kind, but they are very versatile
2090     (and unfortunately a bit complex).
2091    
2092 root 1.227 Unlike C<ev_timer>, periodic watchers are not based on real time (or
2093     relative time, the physical time that passes) but on wall clock time
2094     (absolute time, the thing you can read on your calender or clock). The
2095     difference is that wall clock time can run faster or slower than real
2096     time, and time jumps are not uncommon (e.g. when you adjust your
2097     wrist-watch).
2098    
2099     You can tell a periodic watcher to trigger after some specific point
2100     in time: for example, if you tell a periodic watcher to trigger "in 10
2101     seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
2102     not a delay) and then reset your system clock to January of the previous
2103     year, then it will take a year or more to trigger the event (unlike an
2104     C<ev_timer>, which would still trigger roughly 10 seconds after starting
2105     it, as it uses a relative timeout).
2106    
2107     C<ev_periodic> watchers can also be used to implement vastly more complex
2108     timers, such as triggering an event on each "midnight, local time", or
2109     other complicated rules. This cannot be done with C<ev_timer> watchers, as
2110     those cannot react to time jumps.
2111 root 1.1
2112 root 1.161 As with timers, the callback is guaranteed to be invoked only when the
2113 root 1.230 point in time where it is supposed to trigger has passed. If multiple
2114     timers become ready during the same loop iteration then the ones with
2115     earlier time-out values are invoked before ones with later time-out values
2116 root 1.310 (but this is no longer true when a callback calls C<ev_run> recursively).
2117 root 1.28
2118 root 1.82 =head3 Watcher-Specific Functions and Data Members
2119    
2120 root 1.1 =over 4
2121    
2122 root 1.227 =item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
2123 root 1.1
2124 root 1.227 =item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
2125 root 1.1
2126 root 1.227 Lots of arguments, let's sort it out... There are basically three modes of
2127 root 1.183 operation, and we will explain them from simplest to most complex:
2128 root 1.1
2129     =over 4
2130    
2131 root 1.227 =item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
2132 root 1.1
2133 root 1.161 In this configuration the watcher triggers an event after the wall clock
2134 root 1.227 time C<offset> has passed. It will not repeat and will not adjust when a
2135     time jump occurs, that is, if it is to be run at January 1st 2011 then it
2136     will be stopped and invoked when the system clock reaches or surpasses
2137     this point in time.
2138 root 1.1
2139 root 1.227 =item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
2140 root 1.1
2141     In this mode the watcher will always be scheduled to time out at the next
2142 root 1.227 C<offset + N * interval> time (for some integer N, which can also be
2143     negative) and then repeat, regardless of any time jumps. The C<offset>
2144     argument is merely an offset into the C<interval> periods.
2145 root 1.1
2146 root 1.183 This can be used to create timers that do not drift with respect to the
2147 root 1.227 system clock, for example, here is an C<ev_periodic> that triggers each
2148     hour, on the hour (with respect to UTC):
2149 root 1.1
2150     ev_periodic_set (&periodic, 0., 3600., 0);
2151    
2152     This doesn't mean there will always be 3600 seconds in between triggers,
2153 root 1.161 but only that the callback will be called when the system time shows a
2154 root 1.12 full hour (UTC), or more correctly, when the system time is evenly divisible
2155 root 1.1 by 3600.
2156    
2157     Another way to think about it (for the mathematically inclined) is that
2158 root 1.10 C<ev_periodic> will try to run the callback in this mode at the next possible
2159 root 1.227 time where C<time = offset (mod interval)>, regardless of any time jumps.
2160 root 1.1
2161 root 1.367 The C<interval> I<MUST> be positive, and for numerical stability, the
2162     interval value should be higher than C<1/8192> (which is around 100
2163     microseconds) and C<offset> should be higher than C<0> and should have
2164     at most a similar magnitude as the current time (say, within a factor of
2165     ten). Typical values for offset are, in fact, C<0> or something between
2166     C<0> and C<interval>, which is also the recommended range.
2167 root 1.78
2168 root 1.161 Note also that there is an upper limit to how often a timer can fire (CPU
2169 root 1.158 speed for example), so if C<interval> is very small then timing stability
2170 root 1.161 will of course deteriorate. Libev itself tries to be exact to be about one
2171 root 1.158 millisecond (if the OS supports it and the machine is fast enough).
2172    
2173 root 1.227 =item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
2174 root 1.1
2175 root 1.227 In this mode the values for C<interval> and C<offset> are both being
2176 root 1.1 ignored. Instead, each time the periodic watcher gets scheduled, the
2177     reschedule callback will be called with the watcher as first, and the
2178     current time as second argument.
2179    
2180 root 1.227 NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, ever,
2181     or make ANY other event loop modifications whatsoever, unless explicitly
2182     allowed by documentation here>.
2183 root 1.1
2184 root 1.157 If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
2185     it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
2186     only event loop modification you are allowed to do).
2187    
2188 root 1.198 The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
2189 root 1.157 *w, ev_tstamp now)>, e.g.:
2190 root 1.1
2191 root 1.198 static ev_tstamp
2192     my_rescheduler (ev_periodic *w, ev_tstamp now)
2193 root 1.1 {
2194     return now + 60.;
2195     }
2196    
2197     It must return the next time to trigger, based on the passed time value
2198     (that is, the lowest time value larger than to the second argument). It
2199     will usually be called just before the callback will be triggered, but
2200     might be called at other times, too.
2201    
2202 root 1.157 NOTE: I<< This callback must always return a time that is higher than or
2203     equal to the passed C<now> value >>.
2204 root 1.18
2205 root 1.1 This can be used to create very complex timers, such as a timer that
2206 root 1.157 triggers on "next midnight, local time". To do this, you would calculate the
2207 root 1.19 next midnight after C<now> and return the timestamp value for this. How
2208     you do this is, again, up to you (but it is not trivial, which is the main
2209     reason I omitted it as an example).
2210 root 1.1
2211     =back
2212    
2213     =item ev_periodic_again (loop, ev_periodic *)
2214    
2215     Simply stops and restarts the periodic watcher again. This is only useful
2216     when you changed some parameters or the reschedule callback would return
2217     a different time than the last time it was called (e.g. in a crond like
2218     program when the crontabs have changed).
2219    
2220 root 1.149 =item ev_tstamp ev_periodic_at (ev_periodic *)
2221    
2222 root 1.227 When active, returns the absolute time that the watcher is supposed
2223     to trigger next. This is not the same as the C<offset> argument to
2224     C<ev_periodic_set>, but indeed works even in interval and manual
2225     rescheduling modes.
2226 root 1.149
2227 root 1.78 =item ev_tstamp offset [read-write]
2228    
2229     When repeating, this contains the offset value, otherwise this is the
2230 root 1.227 absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
2231     although libev might modify this value for better numerical stability).
2232 root 1.78
2233     Can be modified any time, but changes only take effect when the periodic
2234     timer fires or C<ev_periodic_again> is being called.
2235    
2236 root 1.48 =item ev_tstamp interval [read-write]
2237    
2238     The current interval value. Can be modified any time, but changes only
2239     take effect when the periodic timer fires or C<ev_periodic_again> is being
2240     called.
2241    
2242 root 1.198 =item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
2243 root 1.48
2244     The current reschedule callback, or C<0>, if this functionality is
2245     switched off. Can be changed any time, but changes only take effect when
2246     the periodic timer fires or C<ev_periodic_again> is being called.
2247    
2248 root 1.1 =back
2249    
2250 root 1.111 =head3 Examples
2251    
2252 root 1.54 Example: Call a callback every hour, or, more precisely, whenever the
2253 root 1.183 system time is divisible by 3600. The callback invocation times have
2254 root 1.161 potentially a lot of jitter, but good long-term stability.
2255 root 1.34
2256 root 1.164 static void
2257 root 1.301 clock_cb (struct ev_loop *loop, ev_periodic *w, int revents)
2258 root 1.164 {
2259     ... its now a full hour (UTC, or TAI or whatever your clock follows)
2260     }
2261    
2262 root 1.198 ev_periodic hourly_tick;
2263 root 1.164 ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
2264     ev_periodic_start (loop, &hourly_tick);
2265 root 1.34
2266 root 1.54 Example: The same as above, but use a reschedule callback to do it:
2267 root 1.34
2268 root 1.164 #include <math.h>
2269 root 1.34
2270 root 1.164 static ev_tstamp
2271 root 1.198 my_scheduler_cb (ev_periodic *w, ev_tstamp now)
2272 root 1.164 {
2273 root 1.183 return now + (3600. - fmod (now, 3600.));
2274 root 1.164 }
2275 root 1.34
2276 root 1.164 ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
2277 root 1.34
2278 root 1.54 Example: Call a callback every hour, starting now:
2279 root 1.34
2280 root 1.198 ev_periodic hourly_tick;
2281 root 1.164 ev_periodic_init (&hourly_tick, clock_cb,
2282     fmod (ev_now (loop), 3600.), 3600., 0);
2283     ev_periodic_start (loop, &hourly_tick);
2284 root 1.34
2285    
2286 root 1.42 =head2 C<ev_signal> - signal me when a signal gets signalled!
2287 root 1.1
2288     Signal watchers will trigger an event when the process receives a specific
2289     signal one or more times. Even though signals are very asynchronous, libev
2290 root 1.340 will try its best to deliver signals synchronously, i.e. as part of the
2291 root 1.1 normal event processing, like any other event.
2292    
2293 root 1.260 If you want signals to be delivered truly asynchronously, just use
2294     C<sigaction> as you would do without libev and forget about sharing
2295     the signal. You can even use C<ev_async> from a signal handler to
2296     synchronously wake up an event loop.
2297    
2298     You can configure as many watchers as you like for the same signal, but
2299     only within the same loop, i.e. you can watch for C<SIGINT> in your
2300     default loop and for C<SIGIO> in another loop, but you cannot watch for
2301     C<SIGINT> in both the default loop and another loop at the same time. At
2302     the moment, C<SIGCHLD> is permanently tied to the default loop.
2303    
2304     When the first watcher gets started will libev actually register something
2305     with the kernel (thus it coexists with your own signal handlers as long as
2306     you don't register any with libev for the same signal).
2307 root 1.259
2308 root 1.135 If possible and supported, libev will install its handlers with
2309 root 1.259 C<SA_RESTART> (or equivalent) behaviour enabled, so system calls should
2310     not be unduly interrupted. If you have a problem with system calls getting
2311     interrupted by signals you can block all signals in an C<ev_check> watcher
2312     and unblock them in an C<ev_prepare> watcher.
2313 root 1.135
2314 root 1.277 =head3 The special problem of inheritance over fork/execve/pthread_create
2315 root 1.265
2316     Both the signal mask (C<sigprocmask>) and the signal disposition
2317     (C<sigaction>) are unspecified after starting a signal watcher (and after
2318     stopping it again), that is, libev might or might not block the signal,
2319 root 1.360 and might or might not set or restore the installed signal handler (but
2320     see C<EVFLAG_NOSIGMASK>).
2321 root 1.265
2322     While this does not matter for the signal disposition (libev never
2323     sets signals to C<SIG_IGN>, so handlers will be reset to C<SIG_DFL> on
2324     C<execve>), this matters for the signal mask: many programs do not expect
2325 root 1.266 certain signals to be blocked.
2326 root 1.265
2327     This means that before calling C<exec> (from the child) you should reset
2328     the signal mask to whatever "default" you expect (all clear is a good
2329     choice usually).
2330    
2331 root 1.267 The simplest way to ensure that the signal mask is reset in the child is
2332     to install a fork handler with C<pthread_atfork> that resets it. That will
2333     catch fork calls done by libraries (such as the libc) as well.
2334    
2335 root 1.277 In current versions of libev, the signal will not be blocked indefinitely
2336     unless you use the C<signalfd> API (C<EV_SIGNALFD>). While this reduces
2337     the window of opportunity for problems, it will not go away, as libev
2338     I<has> to modify the signal mask, at least temporarily.
2339    
2340 root 1.278 So I can't stress this enough: I<If you do not reset your signal mask when
2341     you expect it to be empty, you have a race condition in your code>. This
2342     is not a libev-specific thing, this is true for most event libraries.
2343 root 1.266
2344 root 1.349 =head3 The special problem of threads signal handling
2345    
2346     POSIX threads has problematic signal handling semantics, specifically,
2347     a lot of functionality (sigfd, sigwait etc.) only really works if all
2348     threads in a process block signals, which is hard to achieve.
2349    
2350     When you want to use sigwait (or mix libev signal handling with your own
2351     for the same signals), you can tackle this problem by globally blocking
2352     all signals before creating any threads (or creating them with a fully set
2353     sigprocmask) and also specifying the C<EVFLAG_NOSIGMASK> when creating
2354     loops. Then designate one thread as "signal receiver thread" which handles
2355     these signals. You can pass on any signals that libev might be interested
2356     in by calling C<ev_feed_signal>.
2357    
2358 root 1.82 =head3 Watcher-Specific Functions and Data Members
2359    
2360 root 1.1 =over 4
2361    
2362     =item ev_signal_init (ev_signal *, callback, int signum)
2363    
2364     =item ev_signal_set (ev_signal *, int signum)
2365    
2366     Configures the watcher to trigger on the given signal number (usually one
2367     of the C<SIGxxx> constants).
2368    
2369 root 1.48 =item int signum [read-only]
2370    
2371     The signal the watcher watches out for.
2372    
2373 root 1.1 =back
2374    
2375 root 1.132 =head3 Examples
2376    
2377 root 1.188 Example: Try to exit cleanly on SIGINT.
2378 root 1.132
2379 root 1.164 static void
2380 root 1.198 sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
2381 root 1.164 {
2382 root 1.310 ev_break (loop, EVBREAK_ALL);
2383 root 1.164 }
2384    
2385 root 1.198 ev_signal signal_watcher;
2386 root 1.164 ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
2387 root 1.188 ev_signal_start (loop, &signal_watcher);
2388 root 1.132
2389 root 1.35
2390 root 1.42 =head2 C<ev_child> - watch out for process status changes
2391 root 1.1
2392     Child watchers trigger when your process receives a SIGCHLD in response to
2393 root 1.183 some child status changes (most typically when a child of yours dies or
2394     exits). It is permissible to install a child watcher I<after> the child
2395     has been forked (which implies it might have already exited), as long
2396     as the event loop isn't entered (or is continued from a watcher), i.e.,
2397     forking and then immediately registering a watcher for the child is fine,
2398 root 1.244 but forking and registering a watcher a few event loop iterations later or
2399     in the next callback invocation is not.
2400 root 1.134
2401     Only the default event loop is capable of handling signals, and therefore
2402 root 1.161 you can only register child watchers in the default event loop.
2403 root 1.134
2404 root 1.248 Due to some design glitches inside libev, child watchers will always be
2405 root 1.249 handled at maximum priority (their priority is set to C<EV_MAXPRI> by
2406     libev)
2407 root 1.248
2408 root 1.134 =head3 Process Interaction
2409    
2410     Libev grabs C<SIGCHLD> as soon as the default event loop is
2411 root 1.259 initialised. This is necessary to guarantee proper behaviour even if the
2412     first child watcher is started after the child exits. The occurrence
2413 root 1.134 of C<SIGCHLD> is recorded asynchronously, but child reaping is done
2414     synchronously as part of the event loop processing. Libev always reaps all
2415     children, even ones not watched.
2416    
2417     =head3 Overriding the Built-In Processing
2418    
2419     Libev offers no special support for overriding the built-in child
2420     processing, but if your application collides with libev's default child
2421     handler, you can override it easily by installing your own handler for
2422     C<SIGCHLD> after initialising the default loop, and making sure the
2423     default loop never gets destroyed. You are encouraged, however, to use an
2424     event-based approach to child reaping and thus use libev's support for
2425     that, so other libev users can use C<ev_child> watchers freely.
2426 root 1.1
2427 root 1.173 =head3 Stopping the Child Watcher
2428    
2429     Currently, the child watcher never gets stopped, even when the
2430     child terminates, so normally one needs to stop the watcher in the
2431     callback. Future versions of libev might stop the watcher automatically
2432 root 1.259 when a child exit is detected (calling C<ev_child_stop> twice is not a
2433     problem).
2434 root 1.173
2435 root 1.82 =head3 Watcher-Specific Functions and Data Members
2436    
2437 root 1.1 =over 4
2438    
2439 root 1.120 =item ev_child_init (ev_child *, callback, int pid, int trace)
2440 root 1.1
2441 root 1.120 =item ev_child_set (ev_child *, int pid, int trace)
2442 root 1.1
2443     Configures the watcher to wait for status changes of process C<pid> (or
2444     I<any> process if C<pid> is specified as C<0>). The callback can look
2445     at the C<rstatus> member of the C<ev_child> watcher structure to see
2446 root 1.14 the status word (use the macros from C<sys/wait.h> and see your systems
2447     C<waitpid> documentation). The C<rpid> member contains the pid of the
2448 root 1.120 process causing the status change. C<trace> must be either C<0> (only
2449     activate the watcher when the process terminates) or C<1> (additionally
2450     activate the watcher when the process is stopped or continued).
2451 root 1.1
2452 root 1.48 =item int pid [read-only]
2453    
2454     The process id this watcher watches out for, or C<0>, meaning any process id.
2455    
2456     =item int rpid [read-write]
2457    
2458     The process id that detected a status change.
2459    
2460     =item int rstatus [read-write]
2461    
2462     The process exit/trace status caused by C<rpid> (see your systems
2463     C<waitpid> and C<sys/wait.h> documentation for details).
2464    
2465 root 1.1 =back
2466    
2467 root 1.134 =head3 Examples
2468    
2469     Example: C<fork()> a new process and install a child handler to wait for
2470     its completion.
2471    
2472 root 1.164 ev_child cw;
2473    
2474     static void
2475 root 1.198 child_cb (EV_P_ ev_child *w, int revents)
2476 root 1.164 {
2477     ev_child_stop (EV_A_ w);
2478     printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
2479     }
2480    
2481     pid_t pid = fork ();
2482 root 1.134
2483 root 1.164 if (pid < 0)
2484     // error
2485     else if (pid == 0)
2486     {
2487     // the forked child executes here
2488     exit (1);
2489     }
2490     else
2491     {
2492     ev_child_init (&cw, child_cb, pid, 0);
2493     ev_child_start (EV_DEFAULT_ &cw);
2494     }
2495 root 1.134
2496 root 1.34
2497 root 1.48 =head2 C<ev_stat> - did the file attributes just change?
2498    
2499 root 1.161 This watches a file system path for attribute changes. That is, it calls
2500 root 1.207 C<stat> on that path in regular intervals (or when the OS says it changed)
2501     and sees if it changed compared to the last time, invoking the callback if
2502     it did.
2503 root 1.48
2504     The path does not need to exist: changing from "path exists" to "path does
2505 root 1.211 not exist" is a status change like any other. The condition "path does not
2506     exist" (or more correctly "path cannot be stat'ed") is signified by the
2507     C<st_nlink> field being zero (which is otherwise always forced to be at
2508     least one) and all the other fields of the stat buffer having unspecified
2509     contents.
2510 root 1.48
2511 root 1.207 The path I<must not> end in a slash or contain special components such as
2512     C<.> or C<..>. The path I<should> be absolute: If it is relative and
2513     your working directory changes, then the behaviour is undefined.
2514    
2515     Since there is no portable change notification interface available, the
2516     portable implementation simply calls C<stat(2)> regularly on the path
2517     to see if it changed somehow. You can specify a recommended polling
2518     interval for this case. If you specify a polling interval of C<0> (highly
2519     recommended!) then a I<suitable, unspecified default> value will be used
2520     (which you can expect to be around five seconds, although this might
2521     change dynamically). Libev will also impose a minimum interval which is
2522 root 1.208 currently around C<0.1>, but that's usually overkill.
2523 root 1.48
2524     This watcher type is not meant for massive numbers of stat watchers,
2525     as even with OS-supported change notifications, this can be
2526     resource-intensive.
2527    
2528 root 1.183 At the time of this writing, the only OS-specific interface implemented
2529 root 1.211 is the Linux inotify interface (implementing kqueue support is left as an
2530     exercise for the reader. Note, however, that the author sees no way of
2531     implementing C<ev_stat> semantics with kqueue, except as a hint).
2532 root 1.48
2533 root 1.137 =head3 ABI Issues (Largefile Support)
2534    
2535     Libev by default (unless the user overrides this) uses the default
2536 root 1.169 compilation environment, which means that on systems with large file
2537     support disabled by default, you get the 32 bit version of the stat
2538 root 1.137 structure. When using the library from programs that change the ABI to
2539     use 64 bit file offsets the programs will fail. In that case you have to
2540     compile libev with the same flags to get binary compatibility. This is
2541     obviously the case with any flags that change the ABI, but the problem is
2542 root 1.207 most noticeably displayed with ev_stat and large file support.
2543 root 1.169
2544     The solution for this is to lobby your distribution maker to make large
2545     file interfaces available by default (as e.g. FreeBSD does) and not
2546     optional. Libev cannot simply switch on large file support because it has
2547     to exchange stat structures with application programs compiled using the
2548     default compilation environment.
2549 root 1.137
2550 root 1.183 =head3 Inotify and Kqueue
2551 root 1.108
2552 root 1.211 When C<inotify (7)> support has been compiled into libev and present at
2553     runtime, it will be used to speed up change detection where possible. The
2554     inotify descriptor will be created lazily when the first C<ev_stat>
2555     watcher is being started.
2556 root 1.108
2557 root 1.147 Inotify presence does not change the semantics of C<ev_stat> watchers
2558 root 1.108 except that changes might be detected earlier, and in some cases, to avoid
2559 root 1.147 making regular C<stat> calls. Even in the presence of inotify support
2560 root 1.183 there are many cases where libev has to resort to regular C<stat> polling,
2561 root 1.211 but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
2562     many bugs), the path exists (i.e. stat succeeds), and the path resides on
2563     a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
2564     xfs are fully working) libev usually gets away without polling.
2565 root 1.108
2566 root 1.183 There is no support for kqueue, as apparently it cannot be used to
2567 root 1.108 implement this functionality, due to the requirement of having a file
2568 root 1.183 descriptor open on the object at all times, and detecting renames, unlinks
2569     etc. is difficult.
2570 root 1.108
2571 root 1.212 =head3 C<stat ()> is a synchronous operation
2572    
2573     Libev doesn't normally do any kind of I/O itself, and so is not blocking
2574     the process. The exception are C<ev_stat> watchers - those call C<stat
2575     ()>, which is a synchronous operation.
2576    
2577     For local paths, this usually doesn't matter: unless the system is very
2578     busy or the intervals between stat's are large, a stat call will be fast,
2579 root 1.222 as the path data is usually in memory already (except when starting the
2580 root 1.212 watcher).
2581    
2582     For networked file systems, calling C<stat ()> can block an indefinite
2583     time due to network issues, and even under good conditions, a stat call
2584     often takes multiple milliseconds.
2585    
2586     Therefore, it is best to avoid using C<ev_stat> watchers on networked
2587     paths, although this is fully supported by libev.
2588    
2589 root 1.107 =head3 The special problem of stat time resolution
2590    
2591 root 1.207 The C<stat ()> system call only supports full-second resolution portably,
2592     and even on systems where the resolution is higher, most file systems
2593     still only support whole seconds.
2594 root 1.107
2595 root 1.150 That means that, if the time is the only thing that changes, you can
2596     easily miss updates: on the first update, C<ev_stat> detects a change and
2597     calls your callback, which does something. When there is another update
2598 root 1.183 within the same second, C<ev_stat> will be unable to detect unless the
2599     stat data does change in other ways (e.g. file size).
2600 root 1.150
2601     The solution to this is to delay acting on a change for slightly more
2602 root 1.155 than a second (or till slightly after the next full second boundary), using
2603 root 1.150 a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
2604     ev_timer_again (loop, w)>).
2605    
2606     The C<.02> offset is added to work around small timing inconsistencies
2607     of some operating systems (where the second counter of the current time
2608     might be be delayed. One such system is the Linux kernel, where a call to
2609     C<gettimeofday> might return a timestamp with a full second later than
2610     a subsequent C<time> call - if the equivalent of C<time ()> is used to
2611     update file times then there will be a small window where the kernel uses
2612     the previous second to update file times but libev might already execute
2613     the timer callback).
2614 root 1.107
2615 root 1.82 =head3 Watcher-Specific Functions and Data Members
2616    
2617 root 1.48 =over 4
2618    
2619     =item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
2620    
2621     =item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
2622    
2623     Configures the watcher to wait for status changes of the given
2624     C<path>. The C<interval> is a hint on how quickly a change is expected to
2625     be detected and should normally be specified as C<0> to let libev choose
2626     a suitable value. The memory pointed to by C<path> must point to the same
2627     path for as long as the watcher is active.
2628    
2629 root 1.183 The callback will receive an C<EV_STAT> event when a change was detected,
2630     relative to the attributes at the time the watcher was started (or the
2631     last change was detected).
2632 root 1.48
2633 root 1.132 =item ev_stat_stat (loop, ev_stat *)
2634 root 1.48
2635     Updates the stat buffer immediately with new values. If you change the
2636 root 1.150 watched path in your callback, you could call this function to avoid
2637     detecting this change (while introducing a race condition if you are not
2638     the only one changing the path). Can also be useful simply to find out the
2639     new values.
2640 root 1.48
2641     =item ev_statdata attr [read-only]
2642    
2643 root 1.150 The most-recently detected attributes of the file. Although the type is
2644 root 1.48 C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
2645 root 1.150 suitable for your system, but you can only rely on the POSIX-standardised
2646     members to be present. If the C<st_nlink> member is C<0>, then there was
2647     some error while C<stat>ing the file.
2648 root 1.48
2649     =item ev_statdata prev [read-only]
2650    
2651     The previous attributes of the file. The callback gets invoked whenever
2652 root 1.150 C<prev> != C<attr>, or, more precisely, one or more of these members
2653     differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
2654     C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
2655 root 1.48
2656     =item ev_tstamp interval [read-only]
2657    
2658     The specified interval.
2659    
2660     =item const char *path [read-only]
2661    
2662 root 1.161 The file system path that is being watched.
2663 root 1.48
2664     =back
2665    
2666 root 1.108 =head3 Examples
2667    
2668 root 1.48 Example: Watch C</etc/passwd> for attribute changes.
2669    
2670 root 1.164 static void
2671     passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
2672     {
2673     /* /etc/passwd changed in some way */
2674     if (w->attr.st_nlink)
2675     {
2676     printf ("passwd current size %ld\n", (long)w->attr.st_size);
2677     printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
2678     printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
2679     }
2680     else
2681     /* you shalt not abuse printf for puts */
2682     puts ("wow, /etc/passwd is not there, expect problems. "
2683     "if this is windows, they already arrived\n");
2684     }
2685 root 1.48
2686 root 1.164 ...
2687     ev_stat passwd;
2688 root 1.48
2689 root 1.164 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
2690     ev_stat_start (loop, &passwd);
2691 root 1.107
2692     Example: Like above, but additionally use a one-second delay so we do not
2693     miss updates (however, frequent updates will delay processing, too, so
2694     one might do the work both on C<ev_stat> callback invocation I<and> on
2695     C<ev_timer> callback invocation).
2696    
2697 root 1.164 static ev_stat passwd;
2698     static ev_timer timer;
2699 root 1.107
2700 root 1.164 static void
2701     timer_cb (EV_P_ ev_timer *w, int revents)
2702     {
2703     ev_timer_stop (EV_A_ w);
2704    
2705     /* now it's one second after the most recent passwd change */
2706     }
2707    
2708     static void
2709     stat_cb (EV_P_ ev_stat *w, int revents)
2710     {
2711     /* reset the one-second timer */
2712     ev_timer_again (EV_A_ &timer);
2713     }
2714    
2715     ...
2716     ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
2717     ev_stat_start (loop, &passwd);
2718     ev_timer_init (&timer, timer_cb, 0., 1.02);
2719 root 1.48
2720    
2721 root 1.42 =head2 C<ev_idle> - when you've got nothing better to do...
2722 root 1.1
2723 root 1.67 Idle watchers trigger events when no other events of the same or higher
2724 root 1.183 priority are pending (prepare, check and other idle watchers do not count
2725     as receiving "events").
2726 root 1.67
2727     That is, as long as your process is busy handling sockets or timeouts
2728     (or even signals, imagine) of the same or higher priority it will not be
2729     triggered. But when your process is idle (or only lower-priority watchers
2730     are pending), the idle watchers are being called once per event loop
2731     iteration - until stopped, that is, or your process receives more events
2732     and becomes busy again with higher priority stuff.
2733 root 1.1
2734     The most noteworthy effect is that as long as any idle watchers are
2735     active, the process will not block when waiting for new events.
2736    
2737     Apart from keeping your process non-blocking (which is a useful
2738     effect on its own sometimes), idle watchers are a good place to do
2739     "pseudo-background processing", or delay processing stuff to after the
2740     event loop has handled all outstanding events.
2741    
2742 root 1.82 =head3 Watcher-Specific Functions and Data Members
2743    
2744 root 1.1 =over 4
2745    
2746 root 1.226 =item ev_idle_init (ev_idle *, callback)
2747 root 1.1
2748     Initialises and configures the idle watcher - it has no parameters of any
2749     kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
2750     believe me.
2751    
2752     =back
2753    
2754 root 1.111 =head3 Examples
2755    
2756 root 1.54 Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
2757     callback, free it. Also, use no error checking, as usual.
2758 root 1.34
2759 root 1.164 static void
2760 root 1.198 idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
2761 root 1.164 {
2762     free (w);
2763     // now do something you wanted to do when the program has
2764     // no longer anything immediate to do.
2765     }
2766    
2767 root 1.198 ev_idle *idle_watcher = malloc (sizeof (ev_idle));
2768 root 1.164 ev_idle_init (idle_watcher, idle_cb);
2769 root 1.242 ev_idle_start (loop, idle_watcher);
2770 root 1.34
2771    
2772 root 1.42 =head2 C<ev_prepare> and C<ev_check> - customise your event loop!
2773 root 1.1
2774 root 1.183 Prepare and check watchers are usually (but not always) used in pairs:
2775 root 1.20 prepare watchers get invoked before the process blocks and check watchers
2776 root 1.14 afterwards.
2777 root 1.1
2778 root 1.310 You I<must not> call C<ev_run> or similar functions that enter
2779 root 1.45 the current event loop from either C<ev_prepare> or C<ev_check>
2780     watchers. Other loops than the current one are fine, however. The
2781     rationale behind this is that you do not need to check for recursion in
2782     those watchers, i.e. the sequence will always be C<ev_prepare>, blocking,
2783     C<ev_check> so if you have one watcher of each kind they will always be
2784     called in pairs bracketing the blocking call.
2785    
2786 root 1.35 Their main purpose is to integrate other event mechanisms into libev and
2787 root 1.183 their use is somewhat advanced. They could be used, for example, to track
2788 root 1.35 variable changes, implement your own watchers, integrate net-snmp or a
2789 root 1.45 coroutine library and lots more. They are also occasionally useful if
2790     you cache some data and want to flush it before blocking (for example,
2791     in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
2792     watcher).
2793 root 1.1
2794 root 1.183 This is done by examining in each prepare call which file descriptors
2795     need to be watched by the other library, registering C<ev_io> watchers
2796     for them and starting an C<ev_timer> watcher for any timeouts (many
2797     libraries provide exactly this functionality). Then, in the check watcher,
2798     you check for any events that occurred (by checking the pending status
2799     of all watchers and stopping them) and call back into the library. The
2800     I/O and timer callbacks will never actually be called (but must be valid
2801     nevertheless, because you never know, you know?).
2802 root 1.1
2803 root 1.14 As another example, the Perl Coro module uses these hooks to integrate
2804 root 1.1 coroutines into libev programs, by yielding to other active coroutines
2805     during each prepare and only letting the process block if no coroutines
2806 root 1.20 are ready to run (it's actually more complicated: it only runs coroutines
2807     with priority higher than or equal to the event loop and one coroutine
2808     of lower priority, but only once, using idle watchers to keep the event
2809     loop from blocking if lower-priority coroutines are active, thus mapping
2810     low-priority coroutines to idle/background tasks).
2811 root 1.1
2812 root 1.77 It is recommended to give C<ev_check> watchers highest (C<EV_MAXPRI>)
2813     priority, to ensure that they are being run before any other watchers
2814 root 1.183 after the poll (this doesn't matter for C<ev_prepare> watchers).
2815    
2816     Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
2817     activate ("feed") events into libev. While libev fully supports this, they
2818     might get executed before other C<ev_check> watchers did their job. As
2819     C<ev_check> watchers are often used to embed other (non-libev) event
2820     loops those other event loops might be in an unusable state until their
2821     C<ev_check> watcher ran (always remind yourself to coexist peacefully with
2822     others).
2823 root 1.77
2824 root 1.82 =head3 Watcher-Specific Functions and Data Members
2825    
2826 root 1.1 =over 4
2827    
2828     =item ev_prepare_init (ev_prepare *, callback)
2829    
2830     =item ev_check_init (ev_check *, callback)
2831    
2832     Initialises and configures the prepare or check watcher - they have no
2833     parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
2834 root 1.183 macros, but using them is utterly, utterly, utterly and completely
2835     pointless.
2836 root 1.1
2837     =back
2838    
2839 root 1.111 =head3 Examples
2840    
2841 root 1.76 There are a number of principal ways to embed other event loops or modules
2842     into libev. Here are some ideas on how to include libadns into libev
2843     (there is a Perl module named C<EV::ADNS> that does this, which you could
2844 root 1.150 use as a working example. Another Perl module named C<EV::Glib> embeds a
2845     Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
2846     Glib event loop).
2847 root 1.76
2848     Method 1: Add IO watchers and a timeout watcher in a prepare handler,
2849     and in a check watcher, destroy them and call into libadns. What follows
2850     is pseudo-code only of course. This requires you to either use a low
2851     priority for the check watcher or use C<ev_clear_pending> explicitly, as
2852     the callbacks for the IO/timeout watchers might not have been called yet.
2853 root 1.45
2854 root 1.164 static ev_io iow [nfd];
2855     static ev_timer tw;
2856 root 1.45
2857 root 1.164 static void
2858 root 1.198 io_cb (struct ev_loop *loop, ev_io *w, int revents)
2859 root 1.164 {
2860     }
2861 root 1.45
2862 root 1.164 // create io watchers for each fd and a timer before blocking
2863     static void
2864 root 1.198 adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
2865 root 1.164 {
2866     int timeout = 3600000;
2867     struct pollfd fds [nfd];
2868     // actual code will need to loop here and realloc etc.
2869     adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
2870    
2871     /* the callback is illegal, but won't be called as we stop during check */
2872 root 1.243 ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
2873 root 1.164 ev_timer_start (loop, &tw);
2874    
2875     // create one ev_io per pollfd
2876     for (int i = 0; i < nfd; ++i)
2877     {
2878     ev_io_init (iow + i, io_cb, fds [i].fd,
2879     ((fds [i].events & POLLIN ? EV_READ : 0)
2880     | (fds [i].events & POLLOUT ? EV_WRITE : 0)));
2881    
2882     fds [i].revents = 0;
2883     ev_io_start (loop, iow + i);
2884     }
2885     }
2886    
2887     // stop all watchers after blocking
2888     static void
2889 root 1.198 adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
2890 root 1.164 {
2891     ev_timer_stop (loop, &tw);
2892    
2893     for (int i = 0; i < nfd; ++i)
2894     {
2895     // set the relevant poll flags
2896     // could also call adns_processreadable etc. here
2897     struct pollfd *fd = fds + i;
2898     int revents = ev_clear_pending (iow + i);
2899     if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
2900     if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
2901    
2902     // now stop the watcher
2903     ev_io_stop (loop, iow + i);
2904     }
2905    
2906     adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
2907     }
2908 root 1.34
2909 root 1.76 Method 2: This would be just like method 1, but you run C<adns_afterpoll>
2910     in the prepare watcher and would dispose of the check watcher.
2911    
2912     Method 3: If the module to be embedded supports explicit event
2913 root 1.161 notification (libadns does), you can also make use of the actual watcher
2914 root 1.76 callbacks, and only destroy/create the watchers in the prepare watcher.
2915    
2916 root 1.164 static void
2917     timer_cb (EV_P_ ev_timer *w, int revents)
2918     {
2919     adns_state ads = (adns_state)w->data;
2920     update_now (EV_A);
2921    
2922     adns_processtimeouts (ads, &tv_now);
2923     }
2924    
2925     static void
2926     io_cb (EV_P_ ev_io *w, int revents)
2927     {
2928     adns_state ads = (adns_state)w->data;
2929     update_now (EV_A);
2930    
2931     if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
2932     if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
2933     }
2934 root 1.76
2935 root 1.164 // do not ever call adns_afterpoll
2936 root 1.76
2937     Method 4: Do not use a prepare or check watcher because the module you
2938 root 1.183 want to embed is not flexible enough to support it. Instead, you can
2939     override their poll function. The drawback with this solution is that the
2940     main loop is now no longer controllable by EV. The C<Glib::EV> module uses
2941     this approach, effectively embedding EV as a client into the horrible
2942     libglib event loop.
2943 root 1.76
2944 root 1.164 static gint
2945     event_poll_func (GPollFD *fds, guint nfds, gint timeout)
2946     {
2947     int got_events = 0;
2948    
2949     for (n = 0; n < nfds; ++n)
2950     // create/start io watcher that sets the relevant bits in fds[n] and increment got_events
2951    
2952     if (timeout >= 0)
2953     // create/start timer
2954    
2955     // poll
2956 root 1.310 ev_run (EV_A_ 0);
2957 root 1.76
2958 root 1.164 // stop timer again
2959     if (timeout >= 0)
2960     ev_timer_stop (EV_A_ &to);
2961    
2962     // stop io watchers again - their callbacks should have set
2963     for (n = 0; n < nfds; ++n)
2964     ev_io_stop (EV_A_ iow [n]);
2965    
2966     return got_events;
2967     }
2968 root 1.76
2969 root 1.34
2970 root 1.42 =head2 C<ev_embed> - when one backend isn't enough...
2971 root 1.35
2972     This is a rather advanced watcher type that lets you embed one event loop
2973 root 1.36 into another (currently only C<ev_io> events are supported in the embedded
2974     loop, other types of watchers might be handled in a delayed or incorrect
2975 root 1.100 fashion and must not be used).
2976 root 1.35
2977     There are primarily two reasons you would want that: work around bugs and
2978     prioritise I/O.
2979    
2980     As an example for a bug workaround, the kqueue backend might only support
2981     sockets on some platform, so it is unusable as generic backend, but you
2982     still want to make use of it because you have many sockets and it scales
2983 root 1.183 so nicely. In this case, you would create a kqueue-based loop and embed
2984     it into your default loop (which might use e.g. poll). Overall operation
2985     will be a bit slower because first libev has to call C<poll> and then
2986     C<kevent>, but at least you can use both mechanisms for what they are
2987     best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
2988    
2989     As for prioritising I/O: under rare circumstances you have the case where
2990     some fds have to be watched and handled very quickly (with low latency),
2991     and even priorities and idle watchers might have too much overhead. In
2992     this case you would put all the high priority stuff in one loop and all
2993     the rest in a second one, and embed the second one in the first.
2994 root 1.35
2995 root 1.223 As long as the watcher is active, the callback will be invoked every
2996     time there might be events pending in the embedded loop. The callback
2997     must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
2998     sweep and invoke their callbacks (the callback doesn't need to invoke the
2999     C<ev_embed_sweep> function directly, it could also start an idle watcher
3000     to give the embedded loop strictly lower priority for example).
3001    
3002     You can also set the callback to C<0>, in which case the embed watcher
3003     will automatically execute the embedded loop sweep whenever necessary.
3004    
3005     Fork detection will be handled transparently while the C<ev_embed> watcher
3006     is active, i.e., the embedded loop will automatically be forked when the
3007     embedding loop forks. In other cases, the user is responsible for calling
3008     C<ev_loop_fork> on the embedded loop.
3009 root 1.35
3010 root 1.184 Unfortunately, not all backends are embeddable: only the ones returned by
3011 root 1.35 C<ev_embeddable_backends> are, which, unfortunately, does not include any
3012     portable one.
3013    
3014     So when you want to use this feature you will always have to be prepared
3015     that you cannot get an embeddable loop. The recommended way to get around
3016     this is to have a separate variables for your embeddable loop, try to
3017 root 1.111 create it, and if that fails, use the normal loop for everything.
3018 root 1.35
3019 root 1.187 =head3 C<ev_embed> and fork
3020    
3021     While the C<ev_embed> watcher is running, forks in the embedding loop will
3022     automatically be applied to the embedded loop as well, so no special
3023     fork handling is required in that case. When the watcher is not running,
3024     however, it is still the task of the libev user to call C<ev_loop_fork ()>
3025     as applicable.
3026    
3027 root 1.82 =head3 Watcher-Specific Functions and Data Members
3028    
3029 root 1.35 =over 4
3030    
3031 root 1.36 =item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
3032    
3033     =item ev_embed_set (ev_embed *, callback, struct ev_loop *embedded_loop)
3034    
3035     Configures the watcher to embed the given loop, which must be
3036     embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
3037     invoked automatically, otherwise it is the responsibility of the callback
3038     to invoke it (it will continue to be called until the sweep has been done,
3039 root 1.161 if you do not want that, you need to temporarily stop the embed watcher).
3040 root 1.35
3041 root 1.36 =item ev_embed_sweep (loop, ev_embed *)
3042 root 1.35
3043 root 1.36 Make a single, non-blocking sweep over the embedded loop. This works
3044 root 1.310 similarly to C<ev_run (embedded_loop, EVRUN_NOWAIT)>, but in the most
3045 root 1.161 appropriate way for embedded loops.
3046 root 1.35
3047 root 1.91 =item struct ev_loop *other [read-only]
3048 root 1.48
3049     The embedded event loop.
3050    
3051 root 1.35 =back
3052    
3053 root 1.111 =head3 Examples
3054    
3055     Example: Try to get an embeddable event loop and embed it into the default
3056     event loop. If that is not possible, use the default loop. The default
3057 root 1.161 loop is stored in C<loop_hi>, while the embeddable loop is stored in
3058     C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
3059 root 1.111 used).
3060    
3061 root 1.164 struct ev_loop *loop_hi = ev_default_init (0);
3062     struct ev_loop *loop_lo = 0;
3063 root 1.198 ev_embed embed;
3064 root 1.164
3065     // see if there is a chance of getting one that works
3066     // (remember that a flags value of 0 means autodetection)
3067     loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
3068     ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
3069     : 0;
3070    
3071     // if we got one, then embed it, otherwise default to loop_hi
3072     if (loop_lo)
3073     {
3074     ev_embed_init (&embed, 0, loop_lo);
3075     ev_embed_start (loop_hi, &embed);
3076     }
3077     else
3078     loop_lo = loop_hi;
3079 root 1.111
3080     Example: Check if kqueue is available but not recommended and create
3081     a kqueue backend for use with sockets (which usually work with any
3082     kqueue implementation). Store the kqueue/socket-only event loop in
3083     C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
3084    
3085 root 1.164 struct ev_loop *loop = ev_default_init (0);
3086     struct ev_loop *loop_socket = 0;
3087 root 1.198 ev_embed embed;
3088 root 1.164
3089     if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
3090     if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
3091     {
3092     ev_embed_init (&embed, 0, loop_socket);
3093     ev_embed_start (loop, &embed);
3094     }
3095 root 1.111
3096 root 1.164 if (!loop_socket)
3097     loop_socket = loop;
3098 root 1.111
3099 root 1.164 // now use loop_socket for all sockets, and loop for everything else
3100 root 1.111
3101 root 1.35
3102 root 1.50 =head2 C<ev_fork> - the audacity to resume the event loop after a fork
3103    
3104     Fork watchers are called when a C<fork ()> was detected (usually because
3105     whoever is a good citizen cared to tell libev about it by calling
3106     C<ev_default_fork> or C<ev_loop_fork>). The invocation is done before the
3107     event loop blocks next and before C<ev_check> watchers are being called,
3108     and only in the child after the fork. If whoever good citizen calling
3109     C<ev_default_fork> cheats and calls it in the wrong process, the fork
3110     handlers will be invoked, too, of course.
3111    
3112 root 1.238 =head3 The special problem of life after fork - how is it possible?
3113    
3114 root 1.302 Most uses of C<fork()> consist of forking, then some simple calls to set
3115 root 1.238 up/change the process environment, followed by a call to C<exec()>. This
3116     sequence should be handled by libev without any problems.
3117    
3118     This changes when the application actually wants to do event handling
3119     in the child, or both parent in child, in effect "continuing" after the
3120     fork.
3121    
3122     The default mode of operation (for libev, with application help to detect
3123     forks) is to duplicate all the state in the child, as would be expected
3124     when I<either> the parent I<or> the child process continues.
3125    
3126     When both processes want to continue using libev, then this is usually the
3127     wrong result. In that case, usually one process (typically the parent) is
3128     supposed to continue with all watchers in place as before, while the other
3129     process typically wants to start fresh, i.e. without any active watchers.
3130    
3131     The cleanest and most efficient way to achieve that with libev is to
3132     simply create a new event loop, which of course will be "empty", and
3133     use that for new watchers. This has the advantage of not touching more
3134     memory than necessary, and thus avoiding the copy-on-write, and the
3135     disadvantage of having to use multiple event loops (which do not support
3136     signal watchers).
3137    
3138     When this is not possible, or you want to use the default loop for
3139     other reasons, then in the process that wants to start "fresh", call
3140 root 1.322 C<ev_loop_destroy (EV_DEFAULT)> followed by C<ev_default_loop (...)>.
3141     Destroying the default loop will "orphan" (not stop) all registered
3142     watchers, so you have to be careful not to execute code that modifies
3143     those watchers. Note also that in that case, you have to re-register any
3144     signal watchers.
3145 root 1.238
3146 root 1.83 =head3 Watcher-Specific Functions and Data Members
3147    
3148 root 1.50 =over 4
3149    
3150 root 1.325 =item ev_fork_init (ev_fork *, callback)
3151 root 1.50
3152     Initialises and configures the fork watcher - it has no parameters of any
3153     kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
3154 root 1.329 really.
3155 root 1.50
3156     =back
3157    
3158    
3159 root 1.324 =head2 C<ev_cleanup> - even the best things end
3160    
3161 root 1.328 Cleanup watchers are called just before the event loop is being destroyed
3162     by a call to C<ev_loop_destroy>.
3163 root 1.324
3164     While there is no guarantee that the event loop gets destroyed, cleanup
3165 root 1.326 watchers provide a convenient method to install cleanup hooks for your
3166 root 1.324 program, worker threads and so on - you just to make sure to destroy the
3167     loop when you want them to be invoked.
3168    
3169 root 1.327 Cleanup watchers are invoked in the same way as any other watcher. Unlike
3170     all other watchers, they do not keep a reference to the event loop (which
3171     makes a lot of sense if you think about it). Like all other watchers, you
3172     can call libev functions in the callback, except C<ev_cleanup_start>.
3173    
3174 root 1.324 =head3 Watcher-Specific Functions and Data Members
3175    
3176     =over 4
3177    
3178 root 1.325 =item ev_cleanup_init (ev_cleanup *, callback)
3179 root 1.324
3180     Initialises and configures the cleanup watcher - it has no parameters of
3181     any kind. There is a C<ev_cleanup_set> macro, but using it is utterly
3182 root 1.329 pointless, I assure you.
3183 root 1.324
3184     =back
3185    
3186     Example: Register an atexit handler to destroy the default loop, so any
3187     cleanup functions are called.
3188    
3189     static void
3190     program_exits (void)
3191     {
3192     ev_loop_destroy (EV_DEFAULT_UC);
3193     }
3194    
3195     ...
3196     atexit (program_exits);
3197    
3198    
3199 root 1.302 =head2 C<ev_async> - how to wake up an event loop
3200 root 1.122
3201 root 1.363 In general, you cannot use an C<ev_loop> from multiple threads or other
3202 root 1.122 asynchronous sources such as signal handlers (as opposed to multiple event
3203     loops - those are of course safe to use in different threads).
3204    
3205 root 1.302 Sometimes, however, you need to wake up an event loop you do not control,
3206     for example because it belongs to another thread. This is what C<ev_async>
3207     watchers do: as long as the C<ev_async> watcher is active, you can signal
3208     it by calling C<ev_async_send>, which is thread- and signal safe.
3209 root 1.122
3210     This functionality is very similar to C<ev_signal> watchers, as signals,
3211     too, are asynchronous in nature, and signals, too, will be compressed
3212     (i.e. the number of callback invocations may be less than the number of
3213 root 1.349 C<ev_async_sent> calls). In fact, you could use signal watchers as a kind
3214     of "global async watchers" by using a watcher on an otherwise unused
3215     signal, and C<ev_feed_signal> to signal this watcher from another thread,
3216     even without knowing which loop owns the signal.
3217 root 1.122
3218     Unlike C<ev_signal> watchers, C<ev_async> works with any event loop, not
3219     just the default loop.
3220    
3221 root 1.124 =head3 Queueing
3222    
3223     C<ev_async> does not support queueing of data in any way. The reason
3224     is that the author does not know of a simple (or any) algorithm for a
3225     multiple-writer-single-reader queue that works in all cases and doesn't
3226 root 1.274 need elaborate support such as pthreads or unportable memory access
3227     semantics.
3228 root 1.124
3229     That means that if you want to queue data, you have to provide your own
3230 root 1.184 queue. But at least I can tell you how to implement locking around your
3231 root 1.130 queue:
3232 root 1.124
3233     =over 4
3234    
3235     =item queueing from a signal handler context
3236    
3237     To implement race-free queueing, you simply add to the queue in the signal
3238 root 1.191 handler but you block the signal handler in the watcher callback. Here is
3239     an example that does that for some fictitious SIGUSR1 handler:
3240 root 1.124
3241     static ev_async mysig;
3242    
3243     static void
3244     sigusr1_handler (void)
3245     {
3246     sometype data;
3247    
3248     // no locking etc.
3249     queue_put (data);
3250 root 1.133 ev_async_send (EV_DEFAULT_ &mysig);
3251 root 1.124 }
3252    
3253     static void
3254     mysig_cb (EV_P_ ev_async *w, int revents)
3255     {
3256     sometype data;
3257     sigset_t block, prev;
3258    
3259     sigemptyset (&block);
3260     sigaddset (&block, SIGUSR1);
3261     sigprocmask (SIG_BLOCK, &block, &prev);
3262    
3263     while (queue_get (&data))
3264     process (data);
3265    
3266     if (sigismember (&prev, SIGUSR1)
3267     sigprocmask (SIG_UNBLOCK, &block, 0);
3268     }
3269    
3270     (Note: pthreads in theory requires you to use C<pthread_setmask>
3271     instead of C<sigprocmask> when you use threads, but libev doesn't do it
3272     either...).
3273    
3274     =item queueing from a thread context
3275    
3276     The strategy for threads is different, as you cannot (easily) block
3277     threads but you can easily preempt them, so to queue safely you need to
3278 root 1.130 employ a traditional mutex lock, such as in this pthread example:
3279 root 1.124
3280     static ev_async mysig;
3281     static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
3282    
3283     static void
3284     otherthread (void)
3285     {
3286     // only need to lock the actual queueing operation
3287     pthread_mutex_lock (&mymutex);
3288     queue_put (data);
3289     pthread_mutex_unlock (&mymutex);
3290    
3291 root 1.133 ev_async_send (EV_DEFAULT_ &mysig);
3292 root 1.124 }
3293    
3294     static void
3295     mysig_cb (EV_P_ ev_async *w, int revents)
3296     {
3297     pthread_mutex_lock (&mymutex);
3298    
3299     while (queue_get (&data))
3300     process (data);
3301    
3302     pthread_mutex_unlock (&mymutex);
3303     }
3304    
3305     =back
3306    
3307    
3308 root 1.122 =head3 Watcher-Specific Functions and Data Members
3309    
3310     =over 4
3311    
3312     =item ev_async_init (ev_async *, callback)
3313    
3314     Initialises and configures the async watcher - it has no parameters of any
3315 root 1.208 kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
3316 root 1.184 trust me.
3317 root 1.122
3318     =item ev_async_send (loop, ev_async *)
3319    
3320     Sends/signals/activates the given C<ev_async> watcher, that is, feeds
3321 sf-exg 1.364 an C<EV_ASYNC> event on the watcher into the event loop, and instantly
3322 root 1.363 returns.
3323    
3324     Unlike C<ev_feed_event>, this call is safe to do from other threads,
3325     signal or similar contexts (see the discussion of C<EV_ATOMIC_T> in the
3326     embedding section below on what exactly this means).
3327 root 1.122
3328 root 1.227 Note that, as with other watchers in libev, multiple events might get
3329     compressed into a single callback invocation (another way to look at this
3330     is that C<ev_async> watchers are level-triggered, set on C<ev_async_send>,
3331     reset when the event loop detects that).
3332    
3333     This call incurs the overhead of a system call only once per event loop
3334     iteration, so while the overhead might be noticeable, it doesn't apply to
3335     repeated calls to C<ev_async_send> for the same event loop.
3336 root 1.122
3337 root 1.140 =item bool = ev_async_pending (ev_async *)
3338    
3339     Returns a non-zero value when C<ev_async_send> has been called on the
3340     watcher but the event has not yet been processed (or even noted) by the
3341     event loop.
3342    
3343     C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
3344     the loop iterates next and checks for the watcher to have become active,
3345     it will reset the flag again. C<ev_async_pending> can be used to very
3346 root 1.161 quickly check whether invoking the loop might be a good idea.
3347 root 1.140
3348 root 1.227 Not that this does I<not> check whether the watcher itself is pending,
3349     only whether it has been requested to make this watcher pending: there
3350     is a time window between the event loop checking and resetting the async
3351     notification, and the callback being invoked.
3352 root 1.140
3353 root 1.122 =back
3354    
3355    
3356 root 1.1 =head1 OTHER FUNCTIONS
3357    
3358 root 1.14 There are some other functions of possible interest. Described. Here. Now.
3359 root 1.1
3360     =over 4
3361    
3362     =item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)
3363    
3364     This function combines a simple timer and an I/O watcher, calls your
3365 root 1.192 callback on whichever event happens first and automatically stops both
3366 root 1.1 watchers. This is useful if you want to wait for a single event on an fd
3367 root 1.22 or timeout without having to allocate/configure/start/stop/free one or
3368 root 1.1 more watchers yourself.
3369    
3370 root 1.192 If C<fd> is less than 0, then no I/O watcher will be started and the
3371     C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
3372     the given C<fd> and C<events> set will be created and started.
3373 root 1.1
3374     If C<timeout> is less than 0, then no timeout watcher will be
3375 root 1.14 started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
3376 root 1.193 repeat = 0) will be started. C<0> is a valid timeout.
3377 root 1.14
3378 root 1.289 The callback has the type C<void (*cb)(int revents, void *arg)> and is
3379 root 1.21 passed an C<revents> set like normal event callbacks (a combination of
3380 root 1.289 C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMER>) and the C<arg>
3381 root 1.193 value passed to C<ev_once>. Note that it is possible to receive I<both>
3382     a timeout and an io event at the same time - you probably should give io
3383     events precedence.
3384    
3385     Example: wait up to ten seconds for data to appear on STDIN_FILENO.
3386 root 1.1
3387 root 1.164 static void stdin_ready (int revents, void *arg)
3388     {
3389 root 1.193 if (revents & EV_READ)
3390     /* stdin might have data for us, joy! */;
3391 root 1.289 else if (revents & EV_TIMER)
3392 root 1.164 /* doh, nothing entered */;
3393     }
3394 root 1.1
3395 root 1.164 ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
3396 root 1.1
3397 root 1.274 =item ev_feed_fd_event (loop, int fd, int revents)
3398 root 1.1
3399 root 1.14 Feed an event on the given fd, as if a file descriptor backend detected
3400     the given events it.
3401 root 1.1
3402 root 1.274 =item ev_feed_signal_event (loop, int signum)
3403 root 1.1
3404 root 1.349 Feed an event as if the given signal occurred. See also C<ev_feed_signal>,
3405     which is async-safe.
3406 root 1.1
3407     =back
3408    
3409 root 1.34
3410 root 1.345 =head1 COMMON OR USEFUL IDIOMS (OR BOTH)
3411    
3412     This section explains some common idioms that are not immediately
3413     obvious. Note that examples are sprinkled over the whole manual, and this
3414     section only contains stuff that wouldn't fit anywhere else.
3415    
3416 root 1.357 =head2 ASSOCIATING CUSTOM DATA WITH A WATCHER
3417    
3418     Each watcher has, by default, a C<void *data> member that you can read
3419     or modify at any time: libev will completely ignore it. This can be used
3420     to associate arbitrary data with your watcher. If you need more data and
3421     don't want to allocate memory separately and store a pointer to it in that
3422     data member, you can also "subclass" the watcher type and provide your own
3423     data:
3424    
3425     struct my_io
3426     {
3427     ev_io io;
3428     int otherfd;
3429     void *somedata;
3430     struct whatever *mostinteresting;
3431     };
3432    
3433     ...
3434     struct my_io w;
3435     ev_io_init (&w.io, my_cb, fd, EV_READ);
3436    
3437     And since your callback will be called with a pointer to the watcher, you
3438     can cast it back to your own type:
3439    
3440     static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
3441     {
3442     struct my_io *w = (struct my_io *)w_;
3443     ...
3444     }
3445    
3446     More interesting and less C-conformant ways of casting your callback
3447     function type instead have been omitted.
3448    
3449     =head2 BUILDING YOUR OWN COMPOSITE WATCHERS
3450    
3451     Another common scenario is to use some data structure with multiple
3452     embedded watchers, in effect creating your own watcher that combines
3453     multiple libev event sources into one "super-watcher":
3454    
3455     struct my_biggy
3456     {
3457     int some_data;
3458     ev_timer t1;
3459     ev_timer t2;
3460     }
3461    
3462     In this case getting the pointer to C<my_biggy> is a bit more
3463     complicated: Either you store the address of your C<my_biggy> struct in
3464     the C<data> member of the watcher (for woozies or C++ coders), or you need
3465     to use some pointer arithmetic using C<offsetof> inside your watchers (for
3466     real programmers):
3467    
3468     #include <stddef.h>
3469    
3470     static void
3471     t1_cb (EV_P_ ev_timer *w, int revents)
3472     {
3473     struct my_biggy big = (struct my_biggy *)
3474     (((char *)w) - offsetof (struct my_biggy, t1));
3475     }
3476    
3477     static void
3478     t2_cb (EV_P_ ev_timer *w, int revents)
3479     {
3480     struct my_biggy big = (struct my_biggy *)
3481     (((char *)w) - offsetof (struct my_biggy, t2));
3482     }
3483    
3484 root 1.355 =head2 MODEL/NESTED EVENT LOOP INVOCATIONS AND EXIT CONDITIONS
3485 root 1.345
3486     Often (especially in GUI toolkits) there are places where you have
3487     I<modal> interaction, which is most easily implemented by recursively
3488     invoking C<ev_run>.
3489    
3490     This brings the problem of exiting - a callback might want to finish the
3491     main C<ev_run> call, but not the nested one (e.g. user clicked "Quit", but
3492     a modal "Are you sure?" dialog is still waiting), or just the nested one
3493     and not the main one (e.g. user clocked "Ok" in a modal dialog), or some
3494     other combination: In these cases, C<ev_break> will not work alone.
3495    
3496     The solution is to maintain "break this loop" variable for each C<ev_run>
3497     invocation, and use a loop around C<ev_run> until the condition is
3498     triggered, using C<EVRUN_ONCE>:
3499    
3500     // main loop
3501     int exit_main_loop = 0;
3502    
3503     while (!exit_main_loop)
3504     ev_run (EV_DEFAULT_ EVRUN_ONCE);
3505    
3506     // in a model watcher
3507     int exit_nested_loop = 0;
3508    
3509     while (!exit_nested_loop)
3510     ev_run (EV_A_ EVRUN_ONCE);
3511    
3512     To exit from any of these loops, just set the corresponding exit variable:
3513    
3514     // exit modal loop
3515     exit_nested_loop = 1;
3516    
3517     // exit main program, after modal loop is finished
3518     exit_main_loop = 1;
3519    
3520     // exit both
3521     exit_main_loop = exit_nested_loop = 1;
3522    
3523 root 1.355 =head2 THREAD LOCKING EXAMPLE
3524 root 1.354
3525     Here is a fictitious example of how to run an event loop in a different
3526 root 1.359 thread from where callbacks are being invoked and watchers are
3527 root 1.354 created/added/removed.
3528    
3529     For a real-world example, see the C<EV::Loop::Async> perl module,
3530     which uses exactly this technique (which is suited for many high-level
3531     languages).
3532    
3533     The example uses a pthread mutex to protect the loop data, a condition
3534     variable to wait for callback invocations, an async watcher to notify the
3535     event loop thread and an unspecified mechanism to wake up the main thread.
3536    
3537     First, you need to associate some data with the event loop:
3538    
3539     typedef struct {
3540     mutex_t lock; /* global loop lock */
3541     ev_async async_w;
3542     thread_t tid;
3543     cond_t invoke_cv;
3544     } userdata;
3545    
3546     void prepare_loop (EV_P)
3547     {
3548     // for simplicity, we use a static userdata struct.
3549     static userdata u;
3550    
3551     ev_async_init (&u->async_w, async_cb);
3552     ev_async_start (EV_A_ &u->async_w);
3553    
3554     pthread_mutex_init (&u->lock, 0);
3555     pthread_cond_init (&u->invoke_cv, 0);
3556    
3557     // now associate this with the loop
3558     ev_set_userdata (EV_A_ u);
3559     ev_set_invoke_pending_cb (EV_A_ l_invoke);
3560     ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
3561    
3562 root 1.362 // then create the thread running ev_run
3563 root 1.354 pthread_create (&u->tid, 0, l_run, EV_A);
3564     }
3565    
3566     The callback for the C<ev_async> watcher does nothing: the watcher is used
3567     solely to wake up the event loop so it takes notice of any new watchers
3568     that might have been added:
3569    
3570     static void
3571     async_cb (EV_P_ ev_async *w, int revents)
3572     {
3573     // just used for the side effects
3574     }
3575    
3576     The C<l_release> and C<l_acquire> callbacks simply unlock/lock the mutex
3577     protecting the loop data, respectively.
3578    
3579     static void
3580     l_release (EV_P)
3581     {
3582     userdata *u = ev_userdata (EV_A);
3583     pthread_mutex_unlock (&u->lock);
3584     }
3585    
3586     static void
3587     l_acquire (EV_P)
3588     {
3589     userdata *u = ev_userdata (EV_A);
3590     pthread_mutex_lock (&u->lock);
3591     }
3592    
3593     The event loop thread first acquires the mutex, and then jumps straight
3594     into C<ev_run>:
3595    
3596     void *
3597     l_run (void *thr_arg)
3598     {
3599     struct ev_loop *loop = (struct ev_loop *)thr_arg;
3600    
3601     l_acquire (EV_A);
3602     pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
3603     ev_run (EV_A_ 0);
3604     l_release (EV_A);
3605    
3606     return 0;
3607     }
3608    
3609     Instead of invoking all pending watchers, the C<l_invoke> callback will
3610     signal the main thread via some unspecified mechanism (signals? pipe
3611     writes? C<Async::Interrupt>?) and then waits until all pending watchers
3612     have been called (in a while loop because a) spurious wakeups are possible
3613     and b) skipping inter-thread-communication when there are no pending
3614     watchers is very beneficial):
3615    
3616     static void
3617     l_invoke (EV_P)
3618     {
3619     userdata *u = ev_userdata (EV_A);
3620    
3621     while (ev_pending_count (EV_A))
3622     {
3623     wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
3624     pthread_cond_wait (&u->invoke_cv, &u->lock);
3625     }
3626     }
3627    
3628     Now, whenever the main thread gets told to invoke pending watchers, it
3629     will grab the lock, call C<ev_invoke_pending> and then signal the loop
3630     thread to continue:
3631    
3632     static void
3633     real_invoke_pending (EV_P)
3634     {
3635     userdata *u = ev_userdata (EV_A);
3636    
3637     pthread_mutex_lock (&u->lock);
3638     ev_invoke_pending (EV_A);
3639     pthread_cond_signal (&u->invoke_cv);
3640     pthread_mutex_unlock (&u->lock);
3641     }
3642    
3643     Whenever you want to start/stop a watcher or do other modifications to an
3644     event loop, you will now have to lock:
3645    
3646     ev_timer timeout_watcher;
3647     userdata *u = ev_userdata (EV_A);
3648    
3649     ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
3650    
3651     pthread_mutex_lock (&u->lock);
3652     ev_timer_start (EV_A_ &timeout_watcher);
3653     ev_async_send (EV_A_ &u->async_w);
3654     pthread_mutex_unlock (&u->lock);
3655    
3656     Note that sending the C<ev_async> watcher is required because otherwise
3657     an event loop currently blocking in the kernel will have no knowledge
3658     about the newly added timer. By waking up the loop it will pick up any new
3659     watchers in the next event loop iteration.
3660    
3661 root 1.357 =head2 THREADS, COROUTINES, CONTINUATIONS, QUEUES... INSTEAD OF CALLBACKS
3662    
3663     While the overhead of a callback that e.g. schedules a thread is small, it
3664     is still an overhead. If you embed libev, and your main usage is with some
3665     kind of threads or coroutines, you might want to customise libev so that
3666     doesn't need callbacks anymore.
3667    
3668     Imagine you have coroutines that you can switch to using a function
3669     C<switch_to (coro)>, that libev runs in a coroutine called C<libev_coro>
3670     and that due to some magic, the currently active coroutine is stored in a
3671     global called C<current_coro>. Then you can build your own "wait for libev
3672     event" primitive by changing C<EV_CB_DECLARE> and C<EV_CB_INVOKE> (note
3673     the differing C<;> conventions):
3674    
3675     #define EV_CB_DECLARE(type) struct my_coro *cb;
3676     #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
3677    
3678     That means instead of having a C callback function, you store the
3679     coroutine to switch to in each watcher, and instead of having libev call
3680     your callback, you instead have it switch to that coroutine.
3681    
3682     A coroutine might now wait for an event with a function called
3683     C<wait_for_event>. (the watcher needs to be started, as always, but it doesn't
3684     matter when, or whether the watcher is active or not when this function is
3685     called):
3686    
3687     void
3688     wait_for_event (ev_watcher *w)
3689     {
3690     ev_cb_set (w) = current_coro;
3691     switch_to (libev_coro);
3692     }
3693    
3694     That basically suspends the coroutine inside C<wait_for_event> and
3695     continues the libev coroutine, which, when appropriate, switches back to
3696     this or any other coroutine. I am sure if you sue this your own :)
3697    
3698     You can do similar tricks if you have, say, threads with an event queue -
3699     instead of storing a coroutine, you store the queue object and instead of
3700     switching to a coroutine, you push the watcher onto the queue and notify
3701     any waiters.
3702    
3703     To embed libev, see L<EMBEDDING>, but in short, it's easiest to create two
3704     files, F<my_ev.h> and F<my_ev.c> that include the respective libev files:
3705    
3706     // my_ev.h
3707     #define EV_CB_DECLARE(type) struct my_coro *cb;
3708     #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb);
3709     #include "../libev/ev.h"
3710    
3711     // my_ev.c
3712     #define EV_H "my_ev.h"
3713     #include "../libev/ev.c"
3714    
3715     And then use F<my_ev.h> when you would normally use F<ev.h>, and compile
3716     F<my_ev.c> into your project. When properly specifying include paths, you
3717     can even use F<ev.h> as header file name directly.
3718    
3719 root 1.345
3720 root 1.20 =head1 LIBEVENT EMULATION
3721    
3722 root 1.24 Libev offers a compatibility emulation layer for libevent. It cannot
3723     emulate the internals of libevent, so here are some usage hints:
3724    
3725     =over 4
3726    
3727 root 1.345 =item * Only the libevent-1.4.1-beta API is being emulated.
3728    
3729     This was the newest libevent version available when libev was implemented,
3730 sf-exg 1.347 and is still mostly unchanged in 2010.
3731 root 1.345
3732 root 1.24 =item * Use it by including <event.h>, as usual.
3733    
3734     =item * The following members are fully supported: ev_base, ev_callback,
3735     ev_arg, ev_fd, ev_res, ev_events.
3736    
3737     =item * Avoid using ev_flags and the EVLIST_*-macros, while it is
3738     maintained by libev, it does not work exactly the same way as in libevent (consider
3739     it a private API).
3740    
3741     =item * Priorities are not currently supported. Initialising priorities
3742     will fail and all watchers will have the same priority, even though there
3743     is an ev_pri field.
3744    
3745 root 1.146 =item * In libevent, the last base created gets the signals, in libev, the
3746 root 1.341 base that registered the signal gets the signals.
3747 root 1.146
3748 root 1.24 =item * Other members are not supported.
3749    
3750     =item * The libev emulation is I<not> ABI compatible to libevent, you need
3751     to use the libev header file and library.
3752    
3753     =back
3754 root 1.20
3755     =head1 C++ SUPPORT
3756    
3757 root 1.38 Libev comes with some simplistic wrapper classes for C++ that mainly allow
3758 root 1.161 you to use some convenience methods to start/stop watchers and also change
3759 root 1.38 the callback model to a model using method callbacks on objects.
3760    
3761     To use it,
3762    
3763 root 1.164 #include <ev++.h>
3764 root 1.38
3765 root 1.71 This automatically includes F<ev.h> and puts all of its definitions (many
3766     of them macros) into the global namespace. All C++ specific things are
3767     put into the C<ev> namespace. It should support all the same embedding
3768     options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
3769    
3770 root 1.72 Care has been taken to keep the overhead low. The only data member the C++
3771     classes add (compared to plain C-style watchers) is the event loop pointer
3772     that the watcher is associated with (or no additional members at all if
3773     you disable C<EV_MULTIPLICITY> when embedding libev).
3774 root 1.71
3775 root 1.346 Currently, functions, static and non-static member functions and classes
3776     with C<operator ()> can be used as callbacks. Other types should be easy
3777     to add as long as they only need one additional pointer for context. If
3778     you need support for other types of functors please contact the author
3779     (preferably after implementing it).
3780 root 1.38
3781     Here is a list of things available in the C<ev> namespace:
3782    
3783     =over 4
3784    
3785     =item C<ev::READ>, C<ev::WRITE> etc.
3786    
3787     These are just enum values with the same values as the C<EV_READ> etc.
3788     macros from F<ev.h>.
3789    
3790     =item C<ev::tstamp>, C<ev::now>
3791    
3792     Aliases to the same types/functions as with the C<ev_> prefix.
3793    
3794     =item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
3795    
3796     For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
3797     the same name in the C<ev> namespace, with the exception of C<ev_signal>
3798     which is called C<ev::sig> to avoid clashes with the C<signal> macro
3799     defines by many implementations.
3800    
3801     All of those classes have these methods:
3802    
3803     =over 4
3804    
3805 root 1.71 =item ev::TYPE::TYPE ()
3806 root 1.38
3807 root 1.274 =item ev::TYPE::TYPE (loop)
3808 root 1.38
3809     =item ev::TYPE::~TYPE
3810    
3811 root 1.71 The constructor (optionally) takes an event loop to associate the watcher
3812     with. If it is omitted, it will use C<EV_DEFAULT>.
3813    
3814     The constructor calls C<ev_init> for you, which means you have to call the
3815     C<set> method before starting it.
3816    
3817     It will not set a callback, however: You have to call the templated C<set>
3818     method to set a callback before you can start the watcher.
3819    
3820     (The reason why you have to use a method is a limitation in C++ which does
3821     not allow explicit template arguments for constructors).
3822 root 1.38
3823     The destructor automatically stops the watcher if it is active.
3824    
3825 root 1.71 =item w->set<class, &class::method> (object *)
3826    
3827     This method sets the callback method to call. The method has to have a
3828     signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
3829     first argument and the C<revents> as second. The object must be given as
3830     parameter and is stored in the C<data> member of the watcher.
3831    
3832     This method synthesizes efficient thunking code to call your method from
3833     the C callback that libev requires. If your compiler can inline your
3834     callback (i.e. it is visible to it at the place of the C<set> call and
3835     your compiler is good :), then the method will be fully inlined into the
3836     thunking function, making it as fast as a direct C callback.
3837    
3838     Example: simple class declaration and watcher initialisation
3839    
3840 root 1.164 struct myclass
3841     {
3842     void io_cb (ev::io &w, int revents) { }
3843     }
3844    
3845     myclass obj;
3846     ev::io iow;
3847     iow.set <myclass, &myclass::io_cb> (&obj);
3848 root 1.71
3849 root 1.221 =item w->set (object *)
3850    
3851     This is a variation of a method callback - leaving out the method to call
3852     will default the method to C<operator ()>, which makes it possible to use
3853     functor objects without having to manually specify the C<operator ()> all
3854     the time. Incidentally, you can then also leave out the template argument
3855     list.
3856    
3857     The C<operator ()> method prototype must be C<void operator ()(watcher &w,
3858     int revents)>.
3859    
3860     See the method-C<set> above for more details.
3861    
3862     Example: use a functor object as callback.
3863    
3864     struct myfunctor
3865     {
3866     void operator() (ev::io &w, int revents)
3867     {
3868     ...
3869     }
3870     }
3871    
3872     myfunctor f;
3873    
3874     ev::io w;
3875     w.set (&f);
3876    
3877 root 1.75 =item w->set<function> (void *data = 0)
3878 root 1.71
3879     Also sets a callback, but uses a static method or plain function as
3880     callback. The optional C<data> argument will be stored in the watcher's
3881     C<data> member and is free for you to use.
3882    
3883 root 1.75 The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
3884    
3885 root 1.71 See the method-C<set> above for more details.
3886    
3887 root 1.184 Example: Use a plain function as callback.
3888 root 1.75
3889 root 1.164 static void io_cb (ev::io &w, int revents) { }
3890     iow.set <io_cb> ();
3891 root 1.75
3892 root 1.274 =item w->set (loop)
3893 root 1.38
3894     Associates a different C<struct ev_loop> with this watcher. You can only
3895     do this when the watcher is inactive (and not pending either).
3896    
3897 root 1.161 =item w->set ([arguments])
3898 root 1.38
3899 root 1.307 Basically the same as C<ev_TYPE_set>, with the same arguments. Either this
3900     method or a suitable start method must be called at least once. Unlike the
3901     C counterpart, an active watcher gets automatically stopped and restarted
3902     when reconfiguring it with this method.
3903 root 1.38
3904     =item w->start ()
3905    
3906 root 1.71 Starts the watcher. Note that there is no C<loop> argument, as the
3907     constructor already stores the event loop.
3908 root 1.38
3909 root 1.307 =item w->start ([arguments])
3910    
3911     Instead of calling C<set> and C<start> methods separately, it is often
3912     convenient to wrap them in one call. Uses the same type of arguments as
3913     the configure C<set> method of the watcher.
3914    
3915 root 1.38 =item w->stop ()
3916    
3917     Stops the watcher if it is active. Again, no C<loop> argument.
3918    
3919 root 1.84 =item w->again () (C<ev::timer>, C<ev::periodic> only)
3920 root 1.38
3921     For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
3922     C<ev_TYPE_again> function.
3923    
3924 root 1.84 =item w->sweep () (C<ev::embed> only)
3925 root 1.38
3926     Invokes C<ev_embed_sweep>.
3927    
3928 root 1.84 =item w->update () (C<ev::stat> only)
3929 root 1.49
3930     Invokes C<ev_stat_stat>.
3931    
3932 root 1.38 =back
3933    
3934     =back
3935    
3936 root 1.307 Example: Define a class with two I/O and idle watchers, start the I/O
3937     watchers in the constructor.
3938 root 1.38
3939 root 1.164 class myclass
3940     {
3941 root 1.184 ev::io io ; void io_cb (ev::io &w, int revents);
3942 root 1.307 ev::io2 io2 ; void io2_cb (ev::io &w, int revents);
3943 root 1.184 ev::idle idle; void idle_cb (ev::idle &w, int revents);
3944 root 1.164
3945     myclass (int fd)
3946     {
3947     io .set <myclass, &myclass::io_cb > (this);
3948 root 1.307 io2 .set <myclass, &myclass::io2_cb > (this);
3949 root 1.164 idle.set <myclass, &myclass::idle_cb> (this);
3950    
3951 root 1.307 io.set (fd, ev::WRITE); // configure the watcher
3952     io.start (); // start it whenever convenient
3953    
3954     io2.start (fd, ev::READ); // set + start in one call
3955 root 1.164 }
3956     };
3957 root 1.20
3958 root 1.50
3959 root 1.136 =head1 OTHER LANGUAGE BINDINGS
3960    
3961     Libev does not offer other language bindings itself, but bindings for a
3962 root 1.161 number of languages exist in the form of third-party packages. If you know
3963 root 1.136 any interesting language binding in addition to the ones listed here, drop
3964     me a note.
3965    
3966     =over 4
3967    
3968     =item Perl
3969    
3970     The EV module implements the full libev API and is actually used to test
3971     libev. EV is developed together with libev. Apart from the EV core module,
3972     there are additional modules that implement libev-compatible interfaces
3973 root 1.184 to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
3974     C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
3975     and C<EV::Glib>).
3976 root 1.136
3977 root 1.166 It can be found and installed via CPAN, its homepage is at
3978 root 1.136 L<http://software.schmorp.de/pkg/EV>.
3979    
3980 root 1.166 =item Python
3981    
3982     Python bindings can be found at L<http://code.google.com/p/pyev/>. It
3983 root 1.228 seems to be quite complete and well-documented.
3984 root 1.166
3985 root 1.136 =item Ruby
3986    
3987     Tony Arcieri has written a ruby extension that offers access to a subset
3988 root 1.161 of the libev API and adds file handle abstractions, asynchronous DNS and
3989 root 1.136 more on top of it. It can be found via gem servers. Its homepage is at
3990     L<http://rev.rubyforge.org/>.
3991    
3992 root 1.218 Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
3993     makes rev work even on mingw.
3994    
3995 root 1.228 =item Haskell
3996    
3997     A haskell binding to libev is available at
3998     L<http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
3999    
4000 root 1.136 =item D
4001    
4002     Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
4003 root 1.172 be found at L<http://proj.llucax.com.ar/wiki/evd>.
4004 root 1.136
4005 root 1.201 =item Ocaml
4006    
4007     Erkki Seppala has written Ocaml bindings for libev, to be found at
4008     L<http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
4009    
4010 root 1.263 =item Lua
4011    
4012 root 1.279 Brian Maher has written a partial interface to libev for lua (at the
4013     time of this writing, only C<ev_io> and C<ev_timer>), to be found at
4014 root 1.263 L<http://github.com/brimworks/lua-ev>.
4015    
4016 root 1.136 =back
4017    
4018    
4019 root 1.50 =head1 MACRO MAGIC
4020    
4021 root 1.161 Libev can be compiled with a variety of options, the most fundamental
4022 root 1.84 of which is C<EV_MULTIPLICITY>. This option determines whether (most)
4023     functions and callbacks have an initial C<struct ev_loop *> argument.
4024 root 1.50
4025     To make it easier to write programs that cope with either variant, the
4026     following macros are defined:
4027    
4028     =over 4
4029    
4030     =item C<EV_A>, C<EV_A_>
4031    
4032     This provides the loop I<argument> for functions, if one is required ("ev
4033     loop argument"). The C<EV_A> form is used when this is the sole argument,
4034     C<EV_A_> is used when other arguments are following. Example:
4035    
4036 root 1.164 ev_unref (EV_A);
4037     ev_timer_add (EV_A_ watcher);
4038 root 1.310 ev_run (EV_A_ 0);
4039 root 1.50
4040     It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
4041     which is often provided by the following macro.
4042    
4043     =item C<EV_P>, C<EV_P_>
4044    
4045     This provides the loop I<parameter> for functions, if one is required ("ev
4046     loop parameter"). The C<EV_P> form is used when this is the sole parameter,
4047     C<EV_P_> is used when other parameters are following. Example:
4048    
4049 root 1.164 // this is how ev_unref is being declared
4050     static void ev_unref (EV_P);
4051 root 1.50
4052 root 1.164 // this is how you can declare your typical callback
4053     static void cb (EV_P_ ev_timer *w, int revents)
4054 root 1.50
4055     It declares a parameter C<loop> of type C<struct ev_loop *>, quite
4056     suitable for use with C<EV_A>.
4057    
4058     =item C<EV_DEFAULT>, C<EV_DEFAULT_>
4059    
4060     Similar to the other two macros, this gives you the value of the default
4061     loop, if multiple loops are supported ("ev loop default").
4062    
4063 root 1.143 =item C<EV_DEFAULT_UC>, C<EV_DEFAULT_UC_>
4064    
4065     Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
4066     default loop has been initialised (C<UC> == unchecked). Their behaviour
4067     is undefined when the default loop has not been initialised by a previous
4068     execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
4069    
4070     It is often prudent to use C<EV_DEFAULT> when initialising the first
4071     watcher in a function but use C<EV_DEFAULT_UC> afterwards.
4072    
4073 root 1.50 =back
4074    
4075 root 1.63 Example: Declare and initialise a check watcher, utilising the above
4076 root 1.68 macros so it will work regardless of whether multiple loops are supported
4077 root 1.63 or not.
4078 root 1.50
4079 root 1.164 static void
4080     check_cb (EV_P_ ev_timer *w, int revents)
4081     {
4082     ev_check_stop (EV_A_ w);
4083     }
4084    
4085     ev_check check;
4086     ev_check_init (&check, check_cb);
4087     ev_check_start (EV_DEFAULT_ &check);
4088 root 1.310 ev_run (EV_DEFAULT_ 0);
4089 root 1.50
4090 root 1.39 =head1 EMBEDDING
4091    
4092     Libev can (and often is) directly embedded into host
4093     applications. Examples of applications that embed it include the Deliantra
4094     Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
4095     and rxvt-unicode.
4096    
4097 root 1.91 The goal is to enable you to just copy the necessary files into your
4098 root 1.39 source directory without having to change even a single line in them, so
4099     you can easily upgrade by simply copying (or having a checked-out copy of
4100     libev somewhere in your source tree).
4101    
4102     =head2 FILESETS
4103    
4104     Depending on what features you need you need to include one or more sets of files
4105 root 1.161 in your application.
4106 root 1.39
4107     =head3 CORE EVENT LOOP
4108    
4109     To include only the libev core (all the C<ev_*> functions), with manual
4110     configuration (no autoconf):
4111    
4112 root 1.164 #define EV_STANDALONE 1
4113     #include "ev.c"
4114 root 1.39
4115     This will automatically include F<ev.h>, too, and should be done in a
4116     single C source file only to provide the function implementations. To use
4117     it, do the same for F<ev.h> in all files wishing to use this API (best
4118     done by writing a wrapper around F<ev.h> that you can include instead and
4119     where you can put other configuration options):
4120    
4121 root 1.164 #define EV_STANDALONE 1
4122     #include "ev.h"
4123 root 1.39
4124     Both header files and implementation files can be compiled with a C++
4125 root 1.208 compiler (at least, that's a stated goal, and breakage will be treated
4126 root 1.39 as a bug).
4127    
4128     You need the following files in your source tree, or in a directory
4129     in your include path (e.g. in libev/ when using -Ilibev):
4130    
4131 root 1.164 ev.h
4132     ev.c
4133     ev_vars.h
4134     ev_wrap.h
4135    
4136     ev_win32.c required on win32 platforms only
4137    
4138     ev_select.c only when select backend is enabled (which is enabled by default)
4139     ev_poll.c only when poll backend is enabled (disabled by default)
4140     ev_epoll.c only when the epoll backend is enabled (disabled by default)
4141     ev_kqueue.c only when the kqueue backend is enabled (disabled by default)
4142     ev_port.c only when the solaris port backend is enabled (disabled by default)
4143 root 1.39
4144     F<ev.c> includes the backend files directly when enabled, so you only need
4145 root 1.43 to compile this single file.
4146 root 1.39
4147     =head3 LIBEVENT COMPATIBILITY API
4148    
4149     To include the libevent compatibility API, also include:
4150    
4151 root 1.164 #include "event.c"
4152 root 1.39
4153     in the file including F<ev.c>, and:
4154    
4155 root 1.164 #include "event.h"
4156 root 1.39
4157     in the files that want to use the libevent API. This also includes F<ev.h>.
4158    
4159     You need the following additional files for this:
4160    
4161 root 1.164 event.h
4162     event.c
4163 root 1.39
4164     =head3 AUTOCONF SUPPORT
4165    
4166 root 1.161 Instead of using C<EV_STANDALONE=1> and providing your configuration in
4167 root 1.39 whatever way you want, you can also C<m4_include([libev.m4])> in your
4168 root 1.43 F<configure.ac> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
4169     include F<config.h> and configure itself accordingly.
4170 root 1.39
4171     For this of course you need the m4 file:
4172    
4173 root 1.164 libev.m4
4174 root 1.39
4175     =head2 PREPROCESSOR SYMBOLS/MACROS
4176    
4177 root 1.142 Libev can be configured via a variety of preprocessor symbols you have to
4178 root 1.281 define before including (or compiling) any of its files. The default in
4179     the absence of autoconf is documented for every option.
4180    
4181     Symbols marked with "(h)" do not change the ABI, and can have different
4182     values when compiling libev vs. including F<ev.h>, so it is permissible
4183 sf-exg 1.292 to redefine them before including F<ev.h> without breaking compatibility
4184 root 1.281 to a compiled library. All other symbols change the ABI, which means all
4185     users of libev and the libev code itself must be compiled with compatible
4186     settings.
4187 root 1.39
4188     =over 4
4189    
4190 root 1.310 =item EV_COMPAT3 (h)
4191    
4192     Backwards compatibility is a major concern for libev. This is why this
4193     release of libev comes with wrappers for the functions and symbols that
4194     have been renamed between libev version 3 and 4.
4195    
4196     You can disable these wrappers (to test compatibility with future
4197     versions) by defining C<EV_COMPAT3> to C<0> when compiling your
4198     sources. This has the additional advantage that you can drop the C<struct>
4199     from C<struct ev_loop> declarations, as libev will provide an C<ev_loop>
4200     typedef in that case.
4201    
4202     In some future version, the default for C<EV_COMPAT3> will become C<0>,
4203     and in some even more future version the compatibility code will be
4204     removed completely.
4205    
4206 root 1.281 =item EV_STANDALONE (h)
4207 root 1.39
4208     Must always be C<1> if you do not use autoconf configuration, which
4209     keeps libev from including F<config.h>, and it also defines dummy
4210     implementations for some libevent functions (such as logging, which is not
4211     supported). It will also not define any of the structs usually found in
4212     F<event.h> that are not directly supported by the libev core alone.
4213    
4214 root 1.262 In standalone mode, libev will still try to automatically deduce the
4215 root 1.218 configuration, but has to be more conservative.
4216    
4217 root 1.367 =item EV_USE_FLOOR
4218    
4219     If defined to be C<1>, libev will use the C<floor ()> function for its
4220     periodic reschedule calculations, otherwise libev will fall back on a
4221     portable (slower) implementation. If you enable this, you usually have to
4222     link against libm or something equivalent. Enabling this when the C<floor>
4223     function is not available will fail, so the safe default is to not enable
4224     this.
4225    
4226 root 1.39 =item EV_USE_MONOTONIC
4227    
4228     If defined to be C<1>, libev will try to detect the availability of the
4229 root 1.218 monotonic clock option at both compile time and runtime. Otherwise no
4230     use of the monotonic clock option will be attempted. If you enable this,
4231     you usually have to link against librt or something similar. Enabling it
4232     when the functionality isn't available is safe, though, although you have
4233 root 1.39 to make sure you link against any libraries where the C<clock_gettime>
4234 root 1.218 function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
4235 root 1.39
4236     =item EV_USE_REALTIME
4237    
4238     If defined to be C<1>, libev will try to detect the availability of the
4239 root 1.224 real-time clock option at compile time (and assume its availability
4240     at runtime if successful). Otherwise no use of the real-time clock
4241     option will be attempted. This effectively replaces C<gettimeofday>
4242     by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
4243     correctness. See the note about libraries in the description of
4244     C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
4245     C<EV_USE_CLOCK_SYSCALL>.
4246 root 1.39
4247 root 1.218 =item EV_USE_CLOCK_SYSCALL
4248    
4249     If defined to be C<1>, libev will try to use a direct syscall instead
4250     of calling the system-provided C<clock_gettime> function. This option
4251     exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
4252     unconditionally pulls in C<libpthread>, slowing down single-threaded
4253 root 1.219 programs needlessly. Using a direct syscall is slightly slower (in
4254     theory), because no optimised vdso implementation can be used, but avoids
4255     the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
4256     higher, as it simplifies linking (no need for C<-lrt>).
4257 root 1.218
4258 root 1.97 =item EV_USE_NANOSLEEP
4259    
4260     If defined to be C<1>, libev will assume that C<nanosleep ()> is available
4261     and will use it for delays. Otherwise it will use C<select ()>.
4262    
4263 root 1.142 =item EV_USE_EVENTFD
4264    
4265     If defined to be C<1>, then libev will assume that C<eventfd ()> is
4266     available and will probe for kernel support at runtime. This will improve
4267     C<ev_signal> and C<ev_async> performance and reduce resource consumption.
4268     If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
4269     2.7 or newer, otherwise disabled.
4270    
4271 root 1.39 =item EV_USE_SELECT
4272    
4273     If undefined or defined to be C<1>, libev will compile in support for the
4274 root 1.161 C<select>(2) backend. No attempt at auto-detection will be done: if no
4275 root 1.39 other method takes over, select will be it. Otherwise the select backend
4276     will not be compiled in.
4277    
4278     =item EV_SELECT_USE_FD_SET
4279    
4280     If defined to C<1>, then the select backend will use the system C<fd_set>
4281     structure. This is useful if libev doesn't compile due to a missing
4282 root 1.218 C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
4283     on exotic systems. This usually limits the range of file descriptors to
4284     some low limit such as 1024 or might have other limitations (winsocket
4285     only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
4286     configures the maximum size of the C<fd_set>.
4287 root 1.39
4288     =item EV_SELECT_IS_WINSOCKET
4289    
4290     When defined to C<1>, the select backend will assume that
4291     select/socket/connect etc. don't understand file descriptors but
4292     wants osf handles on win32 (this is the case when the select to
4293     be used is the winsock select). This means that it will call
4294     C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
4295     it is assumed that all these functions actually work on fds, even
4296     on win32. Should not be defined on non-win32 platforms.
4297    
4298 root 1.264 =item EV_FD_TO_WIN32_HANDLE(fd)
4299 root 1.112
4300     If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
4301     file descriptors to socket handles. When not defining this symbol (the
4302     default), then libev will call C<_get_osfhandle>, which is usually
4303     correct. In some cases, programs use their own file descriptor management,
4304     in which case they can provide this function to map fds to socket handles.
4305    
4306 root 1.264 =item EV_WIN32_HANDLE_TO_FD(handle)
4307    
4308     If C<EV_SELECT_IS_WINSOCKET> then libev maps handles to file descriptors
4309     using the standard C<_open_osfhandle> function. For programs implementing
4310     their own fd to handle mapping, overwriting this function makes it easier
4311     to do so. This can be done by defining this macro to an appropriate value.
4312    
4313     =item EV_WIN32_CLOSE_FD(fd)
4314    
4315     If programs implement their own fd to handle mapping on win32, then this
4316     macro can be used to override the C<close> function, useful to unregister
4317     file descriptors again. Note that the replacement function has to close
4318     the underlying OS handle.
4319    
4320 root 1.39 =item EV_USE_POLL
4321    
4322     If defined to be C<1>, libev will compile in support for the C<poll>(2)
4323     backend. Otherwise it will be enabled on non-win32 platforms. It
4324     takes precedence over select.
4325    
4326     =item EV_USE_EPOLL
4327    
4328     If defined to be C<1>, libev will compile in support for the Linux
4329     C<epoll>(7) backend. Its availability will be detected at runtime,
4330 root 1.142 otherwise another method will be used as fallback. This is the preferred
4331     backend for GNU/Linux systems. If undefined, it will be enabled if the
4332     headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
4333 root 1.39
4334     =item EV_USE_KQUEUE
4335    
4336     If defined to be C<1>, libev will compile in support for the BSD style
4337     C<kqueue>(2) backend. Its actual availability will be detected at runtime,
4338     otherwise another method will be used as fallback. This is the preferred
4339     backend for BSD and BSD-like systems, although on most BSDs kqueue only
4340     supports some types of fds correctly (the only platform we found that
4341     supports ptys for example was NetBSD), so kqueue might be compiled in, but
4342     not be used unless explicitly requested. The best way to use it is to find
4343 root 1.41 out whether kqueue supports your type of fd properly and use an embedded
4344 root 1.39 kqueue loop.
4345    
4346     =item EV_USE_PORT
4347    
4348     If defined to be C<1>, libev will compile in support for the Solaris
4349     10 port style backend. Its availability will be detected at runtime,
4350     otherwise another method will be used as fallback. This is the preferred
4351     backend for Solaris 10 systems.
4352    
4353     =item EV_USE_DEVPOLL
4354    
4355 root 1.161 Reserved for future expansion, works like the USE symbols above.
4356 root 1.39
4357 root 1.56 =item EV_USE_INOTIFY
4358    
4359     If defined to be C<1>, libev will compile in support for the Linux inotify
4360     interface to speed up C<ev_stat> watchers. Its actual availability will
4361 root 1.142 be detected at runtime. If undefined, it will be enabled if the headers
4362     indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
4363 root 1.56
4364 root 1.123 =item EV_ATOMIC_T
4365    
4366     Libev requires an integer type (suitable for storing C<0> or C<1>) whose
4367 root 1.126 access is atomic with respect to other threads or signal contexts. No such
4368     type is easily found in the C language, so you can provide your own type
4369 root 1.127 that you know is safe for your purposes. It is used both for signal handler "locking"
4370     as well as for signal and thread safety in C<ev_async> watchers.
4371 root 1.123
4372 root 1.161 In the absence of this define, libev will use C<sig_atomic_t volatile>
4373 root 1.126 (from F<signal.h>), which is usually good enough on most platforms.
4374 root 1.123
4375 root 1.281 =item EV_H (h)
4376 root 1.39
4377     The name of the F<ev.h> header file used to include it. The default if
4378 root 1.118 undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
4379     used to virtually rename the F<ev.h> header file in case of conflicts.
4380 root 1.39
4381 root 1.281 =item EV_CONFIG_H (h)
4382 root 1.39
4383     If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
4384     F<ev.c>'s idea of where to find the F<config.h> file, similarly to
4385     C<EV_H>, above.
4386    
4387 root 1.281 =item EV_EVENT_H (h)
4388 root 1.39
4389     Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
4390 root 1.118 of how the F<event.h> header can be found, the default is C<"event.h">.
4391 root 1.39
4392 root 1.281 =item EV_PROTOTYPES (h)
4393 root 1.39
4394     If defined to be C<0>, then F<ev.h> will not define any function
4395     prototypes, but still define all the structs and other symbols. This is
4396     occasionally useful if you want to provide your own wrapper functions
4397     around libev functions.
4398    
4399     =item EV_MULTIPLICITY
4400    
4401     If undefined or defined to C<1>, then all event-loop-specific functions
4402     will have the C<struct ev_loop *> as first argument, and you can create
4403     additional independent event loops. Otherwise there will be no support
4404     for multiple event loops and there is no first event loop pointer
4405     argument. Instead, all functions act on the single default loop.
4406    
4407 root 1.69 =item EV_MINPRI
4408    
4409     =item EV_MAXPRI
4410    
4411     The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
4412     C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
4413     provide for more priorities by overriding those symbols (usually defined
4414     to be C<-2> and C<2>, respectively).
4415    
4416     When doing priority-based operations, libev usually has to linearly search
4417     all the priorities, so having many of them (hundreds) uses a lot of space
4418     and time, so using the defaults of five priorities (-2 .. +2) is usually
4419     fine.
4420    
4421 root 1.184 If your embedding application does not need any priorities, defining these
4422     both to C<0> will save some memory and CPU.
4423 root 1.69
4424 root 1.283 =item EV_PERIODIC_ENABLE, EV_IDLE_ENABLE, EV_EMBED_ENABLE, EV_STAT_ENABLE,
4425     EV_PREPARE_ENABLE, EV_CHECK_ENABLE, EV_FORK_ENABLE, EV_SIGNAL_ENABLE,
4426     EV_ASYNC_ENABLE, EV_CHILD_ENABLE.
4427    
4428     If undefined or defined to be C<1> (and the platform supports it), then
4429     the respective watcher type is supported. If defined to be C<0>, then it
4430 sf-exg 1.299 is not. Disabling watcher types mainly saves code size.
4431 root 1.282
4432 root 1.285 =item EV_FEATURES
4433 root 1.47
4434     If you need to shave off some kilobytes of code at the expense of some
4435 root 1.285 speed (but with the full API), you can define this symbol to request
4436     certain subsets of functionality. The default is to enable all features
4437     that can be enabled on the platform.
4438    
4439     A typical way to use this symbol is to define it to C<0> (or to a bitset
4440     with some broad features you want) and then selectively re-enable
4441     additional parts you want, for example if you want everything minimal,
4442     but multiple event loop support, async and child watchers and the poll
4443     backend, use this:
4444    
4445     #define EV_FEATURES 0
4446     #define EV_MULTIPLICITY 1
4447     #define EV_USE_POLL 1
4448     #define EV_CHILD_ENABLE 1
4449     #define EV_ASYNC_ENABLE 1
4450    
4451     The actual value is a bitset, it can be a combination of the following
4452     values:
4453    
4454     =over 4
4455    
4456     =item C<1> - faster/larger code
4457    
4458     Use larger code to speed up some operations.
4459    
4460 sf-exg 1.299 Currently this is used to override some inlining decisions (enlarging the
4461     code size by roughly 30% on amd64).
4462 root 1.285
4463 root 1.286 When optimising for size, use of compiler flags such as C<-Os> with
4464 sf-exg 1.299 gcc is recommended, as well as C<-DNDEBUG>, as libev contains a number of
4465 root 1.286 assertions.
4466 root 1.285
4467     =item C<2> - faster/larger data structures
4468    
4469     Replaces the small 2-heap for timer management by a faster 4-heap, larger
4470 sf-exg 1.299 hash table sizes and so on. This will usually further increase code size
4471 root 1.285 and can additionally have an effect on the size of data structures at
4472     runtime.
4473    
4474     =item C<4> - full API configuration
4475    
4476     This enables priorities (sets C<EV_MAXPRI>=2 and C<EV_MINPRI>=-2), and
4477     enables multiplicity (C<EV_MULTIPLICITY>=1).
4478    
4479 root 1.287 =item C<8> - full API
4480    
4481     This enables a lot of the "lesser used" API functions. See C<ev.h> for
4482     details on which parts of the API are still available without this
4483 root 1.285 feature, and do not complain if this subset changes over time.
4484    
4485 root 1.287 =item C<16> - enable all optional watcher types
4486 root 1.285
4487     Enables all optional watcher types. If you want to selectively enable
4488     only some watcher types other than I/O and timers (e.g. prepare,
4489     embed, async, child...) you can enable them manually by defining
4490     C<EV_watchertype_ENABLE> to C<1> instead.
4491    
4492 root 1.287 =item C<32> - enable all backends
4493 root 1.285
4494     This enables all backends - without this feature, you need to enable at
4495     least one backend manually (C<EV_USE_SELECT> is a good choice).
4496    
4497 root 1.287 =item C<64> - enable OS-specific "helper" APIs
4498 root 1.285
4499     Enable inotify, eventfd, signalfd and similar OS-specific helper APIs by
4500     default.
4501    
4502     =back
4503    
4504     Compiling with C<gcc -Os -DEV_STANDALONE -DEV_USE_EPOLL=1 -DEV_FEATURES=0>
4505 root 1.288 reduces the compiled size of libev from 24.7Kb code/2.8Kb data to 6.5Kb
4506     code/0.3Kb data on my GNU/Linux amd64 system, while still giving you I/O
4507     watchers, timers and monotonic clock support.
4508 root 1.285
4509     With an intelligent-enough linker (gcc+binutils are intelligent enough
4510     when you use C<-Wl,--gc-sections -ffunction-sections>) functions unused by
4511     your program might be left out as well - a binary starting a timer and an
4512     I/O watcher then might come out at only 5Kb.
4513 root 1.282
4514 root 1.281 =item EV_AVOID_STDIO
4515    
4516     If this is set to C<1> at compiletime, then libev will avoid using stdio
4517 sf-exg 1.299 functions (printf, scanf, perror etc.). This will increase the code size
4518 root 1.281 somewhat, but if your program doesn't otherwise depend on stdio and your
4519     libc allows it, this avoids linking in the stdio library which is quite
4520     big.
4521    
4522     Note that error messages might become less precise when this option is
4523     enabled.
4524    
4525 root 1.260 =item EV_NSIG
4526    
4527     The highest supported signal number, +1 (or, the number of
4528     signals): Normally, libev tries to deduce the maximum number of signals
4529     automatically, but sometimes this fails, in which case it can be
4530     specified. Also, using a lower number than detected (C<32> should be
4531 sf-exg 1.298 good for about any system in existence) can save some memory, as libev
4532 root 1.260 statically allocates some 12-24 bytes per signal number.
4533    
4534 root 1.51 =item EV_PID_HASHSIZE
4535    
4536     C<ev_child> watchers use a small hash table to distribute workload by
4537 root 1.285 pid. The default size is C<16> (or C<1> with C<EV_FEATURES> disabled),
4538     usually more than enough. If you need to manage thousands of children you
4539     might want to increase this value (I<must> be a power of two).
4540 root 1.56
4541     =item EV_INOTIFY_HASHSIZE
4542    
4543 root 1.104 C<ev_stat> watchers use a small hash table to distribute workload by
4544 root 1.285 inotify watch id. The default size is C<16> (or C<1> with C<EV_FEATURES>
4545     disabled), usually more than enough. If you need to manage thousands of
4546     C<ev_stat> watchers you might want to increase this value (I<must> be a
4547     power of two).
4548 root 1.51
4549 root 1.153 =item EV_USE_4HEAP
4550    
4551     Heaps are not very cache-efficient. To improve the cache-efficiency of the
4552 root 1.184 timer and periodics heaps, libev uses a 4-heap when this symbol is defined
4553     to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
4554     faster performance with many (thousands) of watchers.
4555 root 1.153
4556 root 1.285 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4557     will be C<0>.
4558 root 1.153
4559     =item EV_HEAP_CACHE_AT
4560    
4561     Heaps are not very cache-efficient. To improve the cache-efficiency of the
4562 root 1.184 timer and periodics heaps, libev can cache the timestamp (I<at>) within
4563 root 1.153 the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
4564     which uses 8-12 bytes more per watcher and a few hundred bytes more code,
4565 root 1.155 but avoids random read accesses on heap changes. This improves performance
4566 root 1.184 noticeably with many (hundreds) of watchers.
4567 root 1.153
4568 root 1.285 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4569     will be C<0>.
4570 root 1.153
4571 root 1.159 =item EV_VERIFY
4572    
4573 root 1.309 Controls how much internal verification (see C<ev_verify ()>) will
4574 root 1.159 be done: If set to C<0>, no internal verification code will be compiled
4575     in. If set to C<1>, then verification code will be compiled in, but not
4576     called. If set to C<2>, then the internal verification code will be
4577     called once per loop, which can slow down libev. If set to C<3>, then the
4578     verification code will be called very frequently, which will slow down
4579     libev considerably.
4580    
4581 root 1.285 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4582     will be C<0>.
4583 root 1.159
4584 root 1.39 =item EV_COMMON
4585    
4586     By default, all watchers have a C<void *data> member. By redefining
4587 sf-exg 1.300 this macro to something else you can include more and other types of
4588 root 1.39 members. You have to define it each time you include one of the files,
4589     though, and it must be identical each time.
4590    
4591     For example, the perl EV module uses something like this:
4592    
4593 root 1.164 #define EV_COMMON \
4594     SV *self; /* contains this struct */ \
4595     SV *cb_sv, *fh /* note no trailing ";" */
4596 root 1.39
4597 root 1.44 =item EV_CB_DECLARE (type)
4598 root 1.39
4599 root 1.44 =item EV_CB_INVOKE (watcher, revents)
4600 root 1.39
4601 root 1.44 =item ev_set_cb (ev, cb)
4602 root 1.39
4603     Can be used to change the callback member declaration in each watcher,
4604     and the way callbacks are invoked and set. Must expand to a struct member
4605 root 1.93 definition and a statement, respectively. See the F<ev.h> header file for
4606 root 1.39 their default definitions. One possible use for overriding these is to
4607 root 1.44 avoid the C<struct ev_loop *> as first argument in all cases, or to use
4608     method calls instead of plain function calls in C++.
4609 root 1.39
4610 root 1.185 =back
4611    
4612 root 1.89 =head2 EXPORTED API SYMBOLS
4613    
4614 root 1.161 If you need to re-export the API (e.g. via a DLL) and you need a list of
4615 root 1.89 exported symbols, you can use the provided F<Symbol.*> files which list
4616     all public symbols, one per line:
4617    
4618 root 1.164 Symbols.ev for libev proper
4619     Symbols.event for the libevent emulation
4620 root 1.89
4621     This can also be used to rename all public symbols to avoid clashes with
4622     multiple versions of libev linked together (which is obviously bad in
4623 root 1.161 itself, but sometimes it is inconvenient to avoid this).
4624 root 1.89
4625 root 1.92 A sed command like this will create wrapper C<#define>'s that you need to
4626 root 1.89 include before including F<ev.h>:
4627    
4628     <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
4629    
4630     This would create a file F<wrap.h> which essentially looks like this:
4631    
4632     #define ev_backend myprefix_ev_backend
4633     #define ev_check_start myprefix_ev_check_start
4634     #define ev_check_stop myprefix_ev_check_stop
4635     ...
4636    
4637 root 1.39 =head2 EXAMPLES
4638    
4639     For a real-world example of a program the includes libev
4640     verbatim, you can have a look at the EV perl module
4641     (L<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
4642     the F<libev/> subdirectory and includes them in the F<EV/EVAPI.h> (public
4643     interface) and F<EV.xs> (implementation) files. Only the F<EV.xs> file
4644     will be compiled. It is pretty complex because it provides its own header
4645     file.
4646    
4647     The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
4648 root 1.63 that everybody includes and which overrides some configure choices:
4649 root 1.39
4650 root 1.287 #define EV_FEATURES 8
4651 root 1.285 #define EV_USE_SELECT 1
4652 root 1.287 #define EV_PREPARE_ENABLE 1
4653     #define EV_IDLE_ENABLE 1
4654     #define EV_SIGNAL_ENABLE 1
4655     #define EV_CHILD_ENABLE 1
4656     #define EV_USE_STDEXCEPT 0
4657 root 1.164 #define EV_CONFIG_H <config.h>
4658 root 1.39
4659 root 1.164 #include "ev++.h"
4660 root 1.39
4661     And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
4662    
4663 root 1.164 #include "ev_cpp.h"
4664     #include "ev.c"
4665 root 1.39
4666 root 1.356 =head1 INTERACTION WITH OTHER PROGRAMS, LIBRARIES OR THE ENVIRONMENT
4667 root 1.46
4668 root 1.189 =head2 THREADS AND COROUTINES
4669 root 1.144
4670 root 1.189 =head3 THREADS
4671 root 1.144
4672 root 1.186 All libev functions are reentrant and thread-safe unless explicitly
4673 root 1.191 documented otherwise, but libev implements no locking itself. This means
4674     that you can use as many loops as you want in parallel, as long as there
4675     are no concurrent calls into any libev function with the same loop
4676     parameter (C<ev_default_*> calls have an implicit default loop parameter,
4677     of course): libev guarantees that different event loops share no data
4678 root 1.186 structures that need any locking.
4679 root 1.180
4680     Or to put it differently: calls with different loop parameters can be done
4681     concurrently from multiple threads, calls with the same loop parameter
4682     must be done serially (but can be done from different threads, as long as
4683     only one thread ever is inside a call at any point in time, e.g. by using
4684     a mutex per loop).
4685    
4686     Specifically to support threads (and signal handlers), libev implements
4687     so-called C<ev_async> watchers, which allow some limited form of
4688 root 1.186 concurrency on the same event loop, namely waking it up "from the
4689     outside".
4690 root 1.144
4691 root 1.170 If you want to know which design (one loop, locking, or multiple loops
4692     without or something else still) is best for your problem, then I cannot
4693 root 1.186 help you, but here is some generic advice:
4694 root 1.144
4695     =over 4
4696    
4697     =item * most applications have a main thread: use the default libev loop
4698 root 1.161 in that thread, or create a separate thread running only the default loop.
4699 root 1.144
4700     This helps integrating other libraries or software modules that use libev
4701     themselves and don't care/know about threading.
4702    
4703     =item * one loop per thread is usually a good model.
4704    
4705     Doing this is almost never wrong, sometimes a better-performance model
4706     exists, but it is always a good start.
4707    
4708     =item * other models exist, such as the leader/follower pattern, where one
4709 root 1.161 loop is handed through multiple threads in a kind of round-robin fashion.
4710 root 1.144
4711 root 1.161 Choosing a model is hard - look around, learn, know that usually you can do
4712 root 1.144 better than you currently do :-)
4713    
4714     =item * often you need to talk to some other thread which blocks in the
4715 root 1.182 event loop.
4716 root 1.144
4717 root 1.182 C<ev_async> watchers can be used to wake them up from other threads safely
4718     (or from signal contexts...).
4719    
4720     An example use would be to communicate signals or other events that only
4721     work in the default loop by registering the signal watcher with the
4722     default loop and triggering an C<ev_async> watcher from the default loop
4723     watcher callback into the event loop interested in the signal.
4724 root 1.180
4725 root 1.144 =back
4726    
4727 root 1.355 See also L<THREAD LOCKING EXAMPLE>.
4728 root 1.254
4729 root 1.189 =head3 COROUTINES
4730 root 1.144
4731 root 1.191 Libev is very accommodating to coroutines ("cooperative threads"):
4732     libev fully supports nesting calls to its functions from different
4733 root 1.310 coroutines (e.g. you can call C<ev_run> on the same loop from two
4734 root 1.255 different coroutines, and switch freely between both coroutines running
4735     the loop, as long as you don't confuse yourself). The only exception is
4736     that you must not do this from C<ev_periodic> reschedule callbacks.
4737 root 1.144
4738 root 1.181 Care has been taken to ensure that libev does not keep local state inside
4739 root 1.310 C<ev_run>, and other calls do not usually allow for coroutine switches as
4740 root 1.208 they do not call any callbacks.
4741 root 1.144
4742 root 1.189 =head2 COMPILER WARNINGS
4743    
4744     Depending on your compiler and compiler settings, you might get no or a
4745     lot of warnings when compiling libev code. Some people are apparently
4746     scared by this.
4747    
4748     However, these are unavoidable for many reasons. For one, each compiler
4749     has different warnings, and each user has different tastes regarding
4750     warning options. "Warn-free" code therefore cannot be a goal except when
4751     targeting a specific compiler and compiler-version.
4752    
4753     Another reason is that some compiler warnings require elaborate
4754     workarounds, or other changes to the code that make it less clear and less
4755     maintainable.
4756    
4757     And of course, some compiler warnings are just plain stupid, or simply
4758     wrong (because they don't actually warn about the condition their message
4759     seems to warn about). For example, certain older gcc versions had some
4760 sf-exg 1.300 warnings that resulted in an extreme number of false positives. These have
4761 root 1.189 been fixed, but some people still insist on making code warn-free with
4762     such buggy versions.
4763    
4764     While libev is written to generate as few warnings as possible,
4765     "warn-free" code is not a goal, and it is recommended not to build libev
4766     with any compiler warnings enabled unless you are prepared to cope with
4767     them (e.g. by ignoring them). Remember that warnings are just that:
4768     warnings, not errors, or proof of bugs.
4769    
4770    
4771 root 1.190 =head2 VALGRIND
4772 root 1.189
4773     Valgrind has a special section here because it is a popular tool that is
4774     highly useful. Unfortunately, valgrind reports are very hard to interpret.
4775    
4776     If you think you found a bug (memory leak, uninitialised data access etc.)
4777     in libev, then check twice: If valgrind reports something like:
4778    
4779     ==2274== definitely lost: 0 bytes in 0 blocks.
4780     ==2274== possibly lost: 0 bytes in 0 blocks.
4781     ==2274== still reachable: 256 bytes in 1 blocks.
4782    
4783     Then there is no memory leak, just as memory accounted to global variables
4784 root 1.208 is not a memleak - the memory is still being referenced, and didn't leak.
4785 root 1.189
4786     Similarly, under some circumstances, valgrind might report kernel bugs
4787     as if it were a bug in libev (e.g. in realloc or in the poll backend,
4788     although an acceptable workaround has been found here), or it might be
4789     confused.
4790    
4791     Keep in mind that valgrind is a very good tool, but only a tool. Don't
4792     make it into some kind of religion.
4793    
4794     If you are unsure about something, feel free to contact the mailing list
4795     with the full valgrind report and an explanation on why you think this
4796     is a bug in libev (best check the archives, too :). However, don't be
4797     annoyed when you get a brisk "this is no bug" answer and take the chance
4798     of learning how to interpret valgrind properly.
4799    
4800     If you need, for some reason, empty reports from valgrind for your project
4801     I suggest using suppression lists.
4802    
4803    
4804 root 1.190 =head1 PORTABILITY NOTES
4805 root 1.189
4806 root 1.302 =head2 GNU/LINUX 32 BIT LIMITATIONS
4807    
4808     GNU/Linux is the only common platform that supports 64 bit file/large file
4809 root 1.303 interfaces but I<disables> them by default.
4810 root 1.302
4811     That means that libev compiled in the default environment doesn't support
4812 root 1.303 files larger than 2GiB or so, which mainly affects C<ev_stat> watchers.
4813 root 1.302
4814     Unfortunately, many programs try to work around this GNU/Linux issue
4815     by enabling the large file API, which makes them incompatible with the
4816     standard libev compiled for their system.
4817    
4818     Likewise, libev cannot enable the large file API itself as this would
4819     suddenly make it incompatible to the default compile time environment,
4820     i.e. all programs not using special compile switches.
4821    
4822     =head2 OS/X AND DARWIN BUGS
4823    
4824     The whole thing is a bug if you ask me - basically any system interface
4825 root 1.303 you touch is broken, whether it is locales, poll, kqueue or even the
4826 root 1.302 OpenGL drivers.
4827    
4828 root 1.303 =head3 C<kqueue> is buggy
4829 root 1.302
4830     The kqueue syscall is broken in all known versions - most versions support
4831     only sockets, many support pipes.
4832    
4833 root 1.314 Libev tries to work around this by not using C<kqueue> by default on this
4834     rotten platform, but of course you can still ask for it when creating a
4835     loop - embedding a socket-only kqueue loop into a select-based one is
4836     probably going to work well.
4837 root 1.304
4838 root 1.303 =head3 C<poll> is buggy
4839 root 1.302
4840     Instead of fixing C<kqueue>, Apple replaced their (working) C<poll>
4841     implementation by something calling C<kqueue> internally around the 10.5.6
4842     release, so now C<kqueue> I<and> C<poll> are broken.
4843    
4844 root 1.304 Libev tries to work around this by not using C<poll> by default on
4845     this rotten platform, but of course you can still ask for it when creating
4846     a loop.
4847 root 1.302
4848 root 1.303 =head3 C<select> is buggy
4849 root 1.302
4850     All that's left is C<select>, and of course Apple found a way to fuck this
4851     one up as well: On OS/X, C<select> actively limits the number of file
4852 root 1.305 descriptors you can pass in to 1024 - your program suddenly crashes when
4853 root 1.302 you use more.
4854    
4855     There is an undocumented "workaround" for this - defining
4856     C<_DARWIN_UNLIMITED_SELECT>, which libev tries to use, so select I<should>
4857     work on OS/X.
4858    
4859     =head2 SOLARIS PROBLEMS AND WORKAROUNDS
4860    
4861 root 1.303 =head3 C<errno> reentrancy
4862 root 1.302
4863     The default compile environment on Solaris is unfortunately so
4864     thread-unsafe that you can't even use components/libraries compiled
4865 root 1.314 without C<-D_REENTRANT> in a threaded program, which, of course, isn't
4866     defined by default. A valid, if stupid, implementation choice.
4867 root 1.302
4868     If you want to use libev in threaded environments you have to make sure
4869     it's compiled with C<_REENTRANT> defined.
4870    
4871 root 1.303 =head3 Event port backend
4872 root 1.302
4873 root 1.314 The scalable event interface for Solaris is called "event
4874     ports". Unfortunately, this mechanism is very buggy in all major
4875     releases. If you run into high CPU usage, your program freezes or you get
4876     a large number of spurious wakeups, make sure you have all the relevant
4877     and latest kernel patches applied. No, I don't know which ones, but there
4878     are multiple ones to apply, and afterwards, event ports actually work
4879     great.
4880 root 1.302
4881 root 1.305 If you can't get it to work, you can try running the program by setting
4882     the environment variable C<LIBEV_FLAGS=3> to only allow C<poll> and
4883     C<select> backends.
4884 root 1.302
4885     =head2 AIX POLL BUG
4886    
4887     AIX unfortunately has a broken C<poll.h> header. Libev works around
4888     this by trying to avoid the poll backend altogether (i.e. it's not even
4889     compiled in), which normally isn't a big problem as C<select> works fine
4890 root 1.314 with large bitsets on AIX, and AIX is dead anyway.
4891 root 1.302
4892 root 1.189 =head2 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
4893 root 1.112
4894 root 1.303 =head3 General issues
4895    
4896 root 1.112 Win32 doesn't support any of the standards (e.g. POSIX) that libev
4897     requires, and its I/O model is fundamentally incompatible with the POSIX
4898     model. Libev still offers limited functionality on this platform in
4899     the form of the C<EVBACKEND_SELECT> backend, and only supports socket
4900     descriptors. This only applies when using Win32 natively, not when using
4901 root 1.303 e.g. cygwin. Actually, it only applies to the microsofts own compilers,
4902     as every compielr comes with a slightly differently broken/incompatible
4903     environment.
4904 root 1.112
4905 root 1.150 Lifting these limitations would basically require the full
4906 root 1.303 re-implementation of the I/O system. If you are into this kind of thing,
4907     then note that glib does exactly that for you in a very portable way (note
4908     also that glib is the slowest event library known to man).
4909 root 1.150
4910 root 1.112 There is no supported compilation method available on windows except
4911     embedding it into other applications.
4912    
4913 root 1.241 Sensible signal handling is officially unsupported by Microsoft - libev
4914     tries its best, but under most conditions, signals will simply not work.
4915    
4916 root 1.162 Not a libev limitation but worth mentioning: windows apparently doesn't
4917     accept large writes: instead of resulting in a partial write, windows will
4918     either accept everything or return C<ENOBUFS> if the buffer is too large,
4919     so make sure you only write small amounts into your sockets (less than a
4920 root 1.184 megabyte seems safe, but this apparently depends on the amount of memory
4921 root 1.162 available).
4922    
4923 root 1.150 Due to the many, low, and arbitrary limits on the win32 platform and
4924     the abysmal performance of winsockets, using a large number of sockets
4925     is not recommended (and not reasonable). If your program needs to use
4926     more than a hundred or so sockets, then likely it needs to use a totally
4927 root 1.155 different implementation for windows, as libev offers the POSIX readiness
4928 root 1.150 notification model, which cannot be implemented efficiently on windows
4929 root 1.241 (due to Microsoft monopoly games).
4930 root 1.112
4931 root 1.167 A typical way to use libev under windows is to embed it (see the embedding
4932     section for details) and use the following F<evwrap.h> header file instead
4933     of F<ev.h>:
4934    
4935     #define EV_STANDALONE /* keeps ev from requiring config.h */
4936     #define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
4937    
4938     #include "ev.h"
4939    
4940     And compile the following F<evwrap.c> file into your project (make sure
4941 root 1.184 you do I<not> compile the F<ev.c> or any other embedded source files!):
4942 root 1.167
4943     #include "evwrap.h"
4944     #include "ev.c"
4945    
4946 root 1.303 =head3 The winsocket C<select> function
4947 root 1.112
4948 root 1.160 The winsocket C<select> function doesn't follow POSIX in that it
4949     requires socket I<handles> and not socket I<file descriptors> (it is
4950     also extremely buggy). This makes select very inefficient, and also
4951 root 1.167 requires a mapping from file descriptors to socket handles (the Microsoft
4952     C runtime provides the function C<_open_osfhandle> for this). See the
4953 root 1.160 discussion of the C<EV_SELECT_USE_FD_SET>, C<EV_SELECT_IS_WINSOCKET> and
4954     C<EV_FD_TO_WIN32_HANDLE> preprocessor symbols for more info.
4955 root 1.112
4956 root 1.161 The configuration for a "naked" win32 using the Microsoft runtime
4957 root 1.112 libraries and raw winsocket select is:
4958    
4959 root 1.164 #define EV_USE_SELECT 1
4960     #define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
4961 root 1.112
4962     Note that winsockets handling of fd sets is O(n), so you can easily get a
4963     complexity in the O(n²) range when using win32.
4964    
4965 root 1.303 =head3 Limited number of file descriptors
4966 root 1.112
4967 root 1.150 Windows has numerous arbitrary (and low) limits on things.
4968    
4969     Early versions of winsocket's select only supported waiting for a maximum
4970     of C<64> handles (probably owning to the fact that all windows kernels
4971 root 1.161 can only wait for C<64> things at the same time internally; Microsoft
4972 root 1.150 recommends spawning a chain of threads and wait for 63 handles and the
4973 root 1.241 previous thread in each. Sounds great!).
4974 root 1.112
4975     Newer versions support more handles, but you need to define C<FD_SETSIZE>
4976     to some high number (e.g. C<2048>) before compiling the winsocket select
4977 root 1.241 call (which might be in libev or elsewhere, for example, perl and many
4978     other interpreters do their own select emulation on windows).
4979 root 1.112
4980 root 1.161 Another limit is the number of file descriptors in the Microsoft runtime
4981 root 1.241 libraries, which by default is C<64> (there must be a hidden I<64>
4982     fetish or something like this inside Microsoft). You can increase this
4983     by calling C<_setmaxstdio>, which can increase this limit to C<2048>
4984     (another arbitrary limit), but is broken in many versions of the Microsoft
4985     runtime libraries. This might get you to about C<512> or C<2048> sockets
4986     (depending on windows version and/or the phase of the moon). To get more,
4987     you need to wrap all I/O functions and provide your own fd management, but
4988     the cost of calling select (O(n²)) will likely make this unworkable.
4989 root 1.112
4990 root 1.189 =head2 PORTABILITY REQUIREMENTS
4991 root 1.112
4992 root 1.189 In addition to a working ISO-C implementation and of course the
4993     backend-specific APIs, libev relies on a few additional extensions:
4994 root 1.148
4995     =over 4
4996    
4997 root 1.165 =item C<void (*)(ev_watcher_type *, int revents)> must have compatible
4998     calling conventions regardless of C<ev_watcher_type *>.
4999    
5000     Libev assumes not only that all watcher pointers have the same internal
5001     structure (guaranteed by POSIX but not by ISO C for example), but it also
5002     assumes that the same (machine) code can be used to call any watcher
5003     callback: The watcher callbacks have different type signatures, but libev
5004     calls them using an C<ev_watcher *> internally.
5005    
5006 root 1.333 =item pointer accesses must be thread-atomic
5007    
5008     Accessing a pointer value must be atomic, it must both be readable and
5009     writable in one piece - this is the case on all current architectures.
5010    
5011 root 1.148 =item C<sig_atomic_t volatile> must be thread-atomic as well
5012    
5013     The type C<sig_atomic_t volatile> (or whatever is defined as
5014 root 1.184 C<EV_ATOMIC_T>) must be atomic with respect to accesses from different
5015 root 1.148 threads. This is not part of the specification for C<sig_atomic_t>, but is
5016     believed to be sufficiently portable.
5017    
5018     =item C<sigprocmask> must work in a threaded environment
5019    
5020     Libev uses C<sigprocmask> to temporarily block signals. This is not
5021     allowed in a threaded program (C<pthread_sigmask> has to be used). Typical
5022     pthread implementations will either allow C<sigprocmask> in the "main
5023     thread" or will block signals process-wide, both behaviours would
5024     be compatible with libev. Interaction between C<sigprocmask> and
5025     C<pthread_sigmask> could complicate things, however.
5026    
5027     The most portable way to handle signals is to block signals in all threads
5028     except the initial one, and run the default loop in the initial thread as
5029     well.
5030    
5031 root 1.150 =item C<long> must be large enough for common memory allocation sizes
5032    
5033 root 1.189 To improve portability and simplify its API, libev uses C<long> internally
5034     instead of C<size_t> when allocating its data structures. On non-POSIX
5035     systems (Microsoft...) this might be unexpectedly low, but is still at
5036     least 31 bits everywhere, which is enough for hundreds of millions of
5037     watchers.
5038 root 1.150
5039     =item C<double> must hold a time value in seconds with enough accuracy
5040    
5041 root 1.151 The type C<double> is used to represent timestamps. It is required to
5042 root 1.308 have at least 51 bits of mantissa (and 9 bits of exponent), which is
5043     good enough for at least into the year 4000 with millisecond accuracy
5044     (the design goal for libev). This requirement is overfulfilled by
5045     implementations using IEEE 754, which is basically all existing ones. With
5046     IEEE 754 doubles, you get microsecond accuracy until at least 2200.
5047 root 1.150
5048 root 1.148 =back
5049    
5050     If you know of other additional requirements drop me a note.
5051    
5052    
5053 root 1.191 =head1 ALGORITHMIC COMPLEXITIES
5054    
5055     In this section the complexities of (many of) the algorithms used inside
5056     libev will be documented. For complexity discussions about backends see
5057     the documentation for C<ev_default_init>.
5058    
5059     All of the following are about amortised time: If an array needs to be
5060     extended, libev needs to realloc and move the whole array, but this
5061     happens asymptotically rarer with higher number of elements, so O(1) might
5062     mean that libev does a lengthy realloc operation in rare cases, but on
5063     average it is much faster and asymptotically approaches constant time.
5064    
5065     =over 4
5066    
5067     =item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
5068    
5069     This means that, when you have a watcher that triggers in one hour and
5070     there are 100 watchers that would trigger before that, then inserting will
5071     have to skip roughly seven (C<ld 100>) of these watchers.
5072    
5073     =item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
5074    
5075     That means that changing a timer costs less than removing/adding them,
5076     as only the relative motion in the event queue has to be paid for.
5077    
5078     =item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
5079    
5080     These just add the watcher into an array or at the head of a list.
5081    
5082     =item Stopping check/prepare/idle/fork/async watchers: O(1)
5083    
5084     =item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
5085    
5086     These watchers are stored in lists, so they need to be walked to find the
5087     correct watcher to remove. The lists are usually short (you don't usually
5088     have many watchers waiting for the same fd or signal: one is typical, two
5089     is rare).
5090    
5091     =item Finding the next timer in each loop iteration: O(1)
5092    
5093     By virtue of using a binary or 4-heap, the next timer is always found at a
5094     fixed position in the storage array.
5095    
5096     =item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
5097    
5098     A change means an I/O watcher gets started or stopped, which requires
5099     libev to recalculate its status (and possibly tell the kernel, depending
5100     on backend and whether C<ev_io_set> was used).
5101    
5102     =item Activating one watcher (putting it into the pending state): O(1)
5103    
5104     =item Priority handling: O(number_of_priorities)
5105    
5106     Priorities are implemented by allocating some space for each
5107     priority. When doing priority-based operations, libev usually has to
5108     linearly search all the priorities, but starting/stopping and activating
5109     watchers becomes O(1) with respect to priority handling.
5110    
5111     =item Sending an ev_async: O(1)
5112    
5113     =item Processing ev_async_send: O(number_of_async_watchers)
5114    
5115     =item Processing signals: O(max_signal_number)
5116    
5117     Sending involves a system call I<iff> there were no other C<ev_async_send>
5118     calls in the current loop iteration. Checking for async and signal events
5119     involves iterating over all running async watchers or all signal numbers.
5120    
5121     =back
5122    
5123    
5124 root 1.291 =head1 PORTING FROM LIBEV 3.X TO 4.X
5125 root 1.289
5126 root 1.332 The major version 4 introduced some incompatible changes to the API.
5127 root 1.289
5128 root 1.332 At the moment, the C<ev.h> header file provides compatibility definitions
5129     for all changes, so most programs should still compile. The compatibility
5130     layer might be removed in later versions of libev, so better update to the
5131     new API early than late.
5132 root 1.291
5133 root 1.289 =over 4
5134    
5135 root 1.332 =item C<EV_COMPAT3> backwards compatibility mechanism
5136    
5137     The backward compatibility mechanism can be controlled by
5138     C<EV_COMPAT3>. See L<PREPROCESSOR SYMBOLS/MACROS> in the L<EMBEDDING>
5139     section.
5140    
5141 root 1.322 =item C<ev_default_destroy> and C<ev_default_fork> have been removed
5142    
5143     These calls can be replaced easily by their C<ev_loop_xxx> counterparts:
5144    
5145 root 1.325 ev_loop_destroy (EV_DEFAULT_UC);
5146 root 1.322 ev_loop_fork (EV_DEFAULT);
5147    
5148 root 1.310 =item function/symbol renames
5149    
5150     A number of functions and symbols have been renamed:
5151    
5152     ev_loop => ev_run
5153     EVLOOP_NONBLOCK => EVRUN_NOWAIT
5154     EVLOOP_ONESHOT => EVRUN_ONCE
5155    
5156     ev_unloop => ev_break
5157     EVUNLOOP_CANCEL => EVBREAK_CANCEL
5158     EVUNLOOP_ONE => EVBREAK_ONE
5159     EVUNLOOP_ALL => EVBREAK_ALL
5160 root 1.291
5161 root 1.310 EV_TIMEOUT => EV_TIMER
5162 root 1.291
5163 root 1.310 ev_loop_count => ev_iteration
5164     ev_loop_depth => ev_depth
5165     ev_loop_verify => ev_verify
5166 root 1.291
5167     Most functions working on C<struct ev_loop> objects don't have an
5168 root 1.310 C<ev_loop_> prefix, so it was removed; C<ev_loop>, C<ev_unloop> and
5169     associated constants have been renamed to not collide with the C<struct
5170     ev_loop> anymore and C<EV_TIMER> now follows the same naming scheme
5171     as all other watcher types. Note that C<ev_loop_fork> is still called
5172     C<ev_loop_fork> because it would otherwise clash with the C<ev_fork>
5173     typedef.
5174    
5175 root 1.289 =item C<EV_MINIMAL> mechanism replaced by C<EV_FEATURES>
5176    
5177     The preprocessor symbol C<EV_MINIMAL> has been replaced by a different
5178     mechanism, C<EV_FEATURES>. Programs using C<EV_MINIMAL> usually compile
5179     and work, but the library code will of course be larger.
5180    
5181     =back
5182    
5183    
5184 root 1.234 =head1 GLOSSARY
5185    
5186     =over 4
5187    
5188     =item active
5189    
5190 root 1.315 A watcher is active as long as it has been started and not yet stopped.
5191     See L<WATCHER STATES> for details.
5192 root 1.234
5193     =item application
5194    
5195     In this document, an application is whatever is using libev.
5196    
5197 root 1.316 =item backend
5198    
5199     The part of the code dealing with the operating system interfaces.
5200    
5201 root 1.234 =item callback
5202    
5203     The address of a function that is called when some event has been
5204     detected. Callbacks are being passed the event loop, the watcher that
5205     received the event, and the actual event bitset.
5206    
5207 root 1.315 =item callback/watcher invocation
5208 root 1.234
5209     The act of calling the callback associated with a watcher.
5210    
5211     =item event
5212    
5213     A change of state of some external event, such as data now being available
5214     for reading on a file descriptor, time having passed or simply not having
5215     any other events happening anymore.
5216    
5217     In libev, events are represented as single bits (such as C<EV_READ> or
5218 root 1.289 C<EV_TIMER>).
5219 root 1.234
5220     =item event library
5221    
5222     A software package implementing an event model and loop.
5223    
5224     =item event loop
5225    
5226     An entity that handles and processes external events and converts them
5227     into callback invocations.
5228    
5229     =item event model
5230    
5231     The model used to describe how an event loop handles and processes
5232     watchers and events.
5233    
5234     =item pending
5235    
5236 root 1.315 A watcher is pending as soon as the corresponding event has been
5237     detected. See L<WATCHER STATES> for details.
5238 root 1.234
5239     =item real time
5240    
5241     The physical time that is observed. It is apparently strictly monotonic :)
5242    
5243     =item wall-clock time
5244    
5245     The time and date as shown on clocks. Unlike real time, it can actually
5246 sf-exg 1.366 be wrong and jump forwards and backwards, e.g. when you adjust your
5247 root 1.234 clock.
5248    
5249     =item watcher
5250    
5251     A data structure that describes interest in certain events. Watchers need
5252     to be started (attached to an event loop) before they can receive events.
5253    
5254     =back
5255    
5256 root 1.1 =head1 AUTHOR
5257    
5258 root 1.333 Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael
5259 root 1.365 Magnusson and Emanuele Giaquinta, and minor corrections by many others.
5260 root 1.1