ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/ev.pod
Revision: 1.311
Committed: Thu Oct 21 14:40:07 2010 UTC (13 years, 6 months ago) by root
Branch: MAIN
Changes since 1.310: +70 -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 1.220 struct ev_loop *loop = ev_default_loop (0);
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     // unloop was called, so exit
64     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     =head1 ABOUT LIBEV
84    
85 root 1.1 Libev is an event loop: you register interest in certain events (such as a
86 root 1.92 file descriptor being readable or a timeout occurring), and it will manage
87 root 1.4 these event sources and provide your program with events.
88 root 1.1
89     To do this, it must take more or less complete control over your process
90     (or thread) by executing the I<event loop> handler, and will then
91     communicate events via a callback mechanism.
92    
93     You register interest in certain events by registering so-called I<event
94     watchers>, which are relatively small C structures you initialise with the
95     details of the event, and then hand it over to libev by I<starting> the
96     watcher.
97    
98 root 1.105 =head2 FEATURES
99 root 1.1
100 root 1.58 Libev supports C<select>, C<poll>, the Linux-specific C<epoll>, the
101     BSD-specific C<kqueue> and the Solaris-specific event port mechanisms
102     for file descriptor events (C<ev_io>), the Linux C<inotify> interface
103 root 1.261 (for C<ev_stat>), Linux eventfd/signalfd (for faster and cleaner
104     inter-thread wakeup (C<ev_async>)/signal handling (C<ev_signal>)) relative
105     timers (C<ev_timer>), absolute timers with customised rescheduling
106     (C<ev_periodic>), synchronous signals (C<ev_signal>), process status
107     change events (C<ev_child>), and event watchers dealing with the event
108     loop mechanism itself (C<ev_idle>, C<ev_embed>, C<ev_prepare> and
109     C<ev_check> watchers) as well as file watchers (C<ev_stat>) and even
110     limited support for fork events (C<ev_fork>).
111 root 1.54
112     It also is quite fast (see this
113     L<benchmark|http://libev.schmorp.de/bench.html> comparing it to libevent
114     for example).
115 root 1.1
116 root 1.105 =head2 CONVENTIONS
117 root 1.1
118 root 1.135 Libev is very configurable. In this manual the default (and most common)
119     configuration will be described, which supports multiple event loops. For
120     more info about various configuration options please have a look at
121     B<EMBED> section in this manual. If libev was configured without support
122     for multiple event loops, then all functions taking an initial argument of
123 root 1.274 name C<loop> (which is always of type C<struct ev_loop *>) will not have
124 root 1.135 this argument.
125 root 1.1
126 root 1.105 =head2 TIME REPRESENTATION
127 root 1.1
128 root 1.237 Libev represents time as a single floating point number, representing
129 root 1.293 the (fractional) number of seconds since the (POSIX) epoch (in practise
130     somewhere near the beginning of 1970, details are complicated, don't
131     ask). This type is called C<ev_tstamp>, which is what you should use
132     too. It usually aliases to the C<double> type in C. When you need to do
133     any calculations on it, you should treat it as some floating point value.
134    
135     Unlike the name component C<stamp> might indicate, it is also used for
136     time differences (e.g. delays) throughout libev.
137 root 1.34
138 root 1.160 =head1 ERROR HANDLING
139    
140     Libev knows three classes of errors: operating system errors, usage errors
141     and internal errors (bugs).
142    
143     When libev catches an operating system error it cannot handle (for example
144 root 1.161 a system call indicating a condition libev cannot fix), it calls the callback
145 root 1.160 set via C<ev_set_syserr_cb>, which is supposed to fix the problem or
146     abort. The default is to print a diagnostic message and to call C<abort
147     ()>.
148    
149     When libev detects a usage error such as a negative timer interval, then
150     it will print a diagnostic message and abort (via the C<assert> mechanism,
151     so C<NDEBUG> will disable this checking): these are programming errors in
152     the libev caller and need to be fixed there.
153    
154     Libev also has a few internal error-checking C<assert>ions, and also has
155     extensive consistency checking code. These do not trigger under normal
156     circumstances, as they indicate either a bug in libev or worse.
157    
158    
159 root 1.17 =head1 GLOBAL FUNCTIONS
160    
161 root 1.18 These functions can be called anytime, even before initialising the
162     library in any way.
163    
164 root 1.1 =over 4
165    
166     =item ev_tstamp ev_time ()
167    
168 root 1.26 Returns the current time as libev would use it. Please note that the
169     C<ev_now> function is usually faster and also often returns the timestamp
170     you actually want to know.
171 root 1.1
172 root 1.97 =item ev_sleep (ev_tstamp interval)
173    
174     Sleep for the given interval: The current thread will be blocked until
175     either it is interrupted or the given time interval has passed. Basically
176 root 1.161 this is a sub-second-resolution C<sleep ()>.
177 root 1.97
178 root 1.1 =item int ev_version_major ()
179    
180     =item int ev_version_minor ()
181    
182 root 1.80 You can find out the major and minor ABI version numbers of the library
183 root 1.1 you linked against by calling the functions C<ev_version_major> and
184     C<ev_version_minor>. If you want, you can compare against the global
185     symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the
186     version of the library your program was compiled against.
187    
188 root 1.80 These version numbers refer to the ABI version of the library, not the
189     release version.
190 root 1.79
191 root 1.9 Usually, it's a good idea to terminate if the major versions mismatch,
192 root 1.79 as this indicates an incompatible change. Minor versions are usually
193 root 1.1 compatible to older versions, so a larger minor version alone is usually
194     not a problem.
195    
196 root 1.54 Example: Make sure we haven't accidentally been linked against the wrong
197 root 1.294 version (note, however, that this will not detect ABI mismatches :).
198 root 1.34
199 root 1.164 assert (("libev version mismatch",
200     ev_version_major () == EV_VERSION_MAJOR
201     && ev_version_minor () >= EV_VERSION_MINOR));
202 root 1.34
203 root 1.31 =item unsigned int ev_supported_backends ()
204    
205     Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
206     value) compiled into this binary of libev (independent of their
207     availability on the system you are running on). See C<ev_default_loop> for
208     a description of the set values.
209    
210 root 1.34 Example: make sure we have the epoll method, because yeah this is cool and
211     a must have and can we have a torrent of it please!!!11
212    
213 root 1.164 assert (("sorry, no epoll, no sex",
214     ev_supported_backends () & EVBACKEND_EPOLL));
215 root 1.34
216 root 1.31 =item unsigned int ev_recommended_backends ()
217    
218     Return the set of all backends compiled into this binary of libev and also
219     recommended for this platform. This set is often smaller than the one
220     returned by C<ev_supported_backends>, as for example kqueue is broken on
221 root 1.161 most BSDs and will not be auto-detected unless you explicitly request it
222 root 1.31 (assuming you know what you are doing). This is the set of backends that
223 root 1.33 libev will probe for if you specify no backends explicitly.
224 root 1.31
225 root 1.35 =item unsigned int ev_embeddable_backends ()
226    
227     Returns the set of backends that are embeddable in other event loops. This
228     is the theoretical, all-platform, value. To find which backends
229     might be supported on the current system, you would need to look at
230     C<ev_embeddable_backends () & ev_supported_backends ()>, likewise for
231     recommended ones.
232    
233     See the description of C<ev_embed> watchers for more info.
234    
235 root 1.186 =item ev_set_allocator (void *(*cb)(void *ptr, long size)) [NOT REENTRANT]
236 root 1.1
237 root 1.59 Sets the allocation function to use (the prototype is similar - the
238 root 1.145 semantics are identical to the C<realloc> C89/SuS/POSIX function). It is
239     used to allocate and free memory (no surprises here). If it returns zero
240     when memory needs to be allocated (C<size != 0>), the library might abort
241     or take some potentially destructive action.
242    
243     Since some systems (at least OpenBSD and Darwin) fail to implement
244     correct C<realloc> semantics, libev will use a wrapper around the system
245     C<realloc> and C<free> functions by default.
246 root 1.1
247     You could override this function in high-availability programs to, say,
248     free some memory if it cannot allocate memory, to use a special allocator,
249     or even to sleep a while and retry until some memory is available.
250    
251 root 1.54 Example: Replace the libev allocator with one that waits a bit and then
252 root 1.145 retries (example requires a standards-compliant C<realloc>).
253 root 1.34
254     static void *
255 root 1.52 persistent_realloc (void *ptr, size_t size)
256 root 1.34 {
257     for (;;)
258     {
259     void *newptr = realloc (ptr, size);
260    
261     if (newptr)
262     return newptr;
263    
264     sleep (60);
265     }
266     }
267    
268     ...
269     ev_set_allocator (persistent_realloc);
270    
271 root 1.186 =item ev_set_syserr_cb (void (*cb)(const char *msg)); [NOT REENTRANT]
272 root 1.1
273 root 1.161 Set the callback function to call on a retryable system call error (such
274 root 1.1 as failed select, poll, epoll_wait). The message is a printable string
275     indicating the system call or subsystem causing the problem. If this
276 root 1.161 callback is set, then libev will expect it to remedy the situation, no
277 root 1.7 matter what, when it returns. That is, libev will generally retry the
278 root 1.1 requested operation, or, if the condition doesn't go away, do bad stuff
279     (such as abort).
280    
281 root 1.54 Example: This is basically the same thing that libev does internally, too.
282 root 1.34
283     static void
284     fatal_error (const char *msg)
285     {
286     perror (msg);
287     abort ();
288     }
289    
290     ...
291     ev_set_syserr_cb (fatal_error);
292    
293 root 1.1 =back
294    
295     =head1 FUNCTIONS CONTROLLING THE EVENT LOOP
296    
297 root 1.310 An event loop is described by a C<struct ev_loop *> (the C<struct> is
298 root 1.311 I<not> optional in this case unless libev 3 compatibility is disabled, as
299     libev 3 had an C<ev_loop> function colliding with the struct name).
300 root 1.200
301     The library knows two types of such loops, the I<default> loop, which
302 root 1.310 supports signals and child events, and dynamically created event loops
303     which do not.
304 root 1.1
305     =over 4
306    
307     =item struct ev_loop *ev_default_loop (unsigned int flags)
308    
309     This will initialise the default event loop if it hasn't been initialised
310     yet and return it. If the default loop could not be initialised, returns
311     false. If it already was initialised it simply returns it (and ignores the
312 root 1.31 flags. If that is troubling you, check C<ev_backend ()> afterwards).
313 root 1.1
314     If you don't know what event loop to use, use the one returned from this
315     function.
316    
317 root 1.139 Note that this function is I<not> thread-safe, so if you want to use it
318     from multiple threads, you have to lock (note also that this is unlikely,
319 root 1.208 as loops cannot be shared easily between threads anyway).
320 root 1.139
321 root 1.118 The default loop is the only loop that can handle C<ev_signal> and
322     C<ev_child> watchers, and to do this, it always registers a handler
323 root 1.161 for C<SIGCHLD>. If this is a problem for your application you can either
324 root 1.118 create a dynamic loop with C<ev_loop_new> that doesn't do that, or you
325     can simply overwrite the C<SIGCHLD> signal handler I<after> calling
326     C<ev_default_init>.
327    
328 root 1.1 The flags argument can be used to specify special behaviour or specific
329 root 1.33 backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
330 root 1.1
331 root 1.33 The following flags are supported:
332 root 1.1
333     =over 4
334    
335 root 1.10 =item C<EVFLAG_AUTO>
336 root 1.1
337 root 1.9 The default flags value. Use this if you have no clue (it's the right
338 root 1.1 thing, believe me).
339    
340 root 1.10 =item C<EVFLAG_NOENV>
341 root 1.1
342 root 1.161 If this flag bit is or'ed into the flag value (or the program runs setuid
343 root 1.8 or setgid) then libev will I<not> look at the environment variable
344     C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
345     override the flags completely if it is found in the environment. This is
346     useful to try out specific backends to test their performance, or to work
347     around bugs.
348 root 1.1
349 root 1.62 =item C<EVFLAG_FORKCHECK>
350    
351 root 1.291 Instead of calling C<ev_loop_fork> manually after a fork, you can also
352     make libev check for a fork in each iteration by enabling this flag.
353 root 1.62
354     This works by calling C<getpid ()> on every iteration of the loop,
355     and thus this might slow down your event loop if you do a lot of loop
356 ayin 1.65 iterations and little real work, but is usually not noticeable (on my
357 root 1.135 GNU/Linux system for example, C<getpid> is actually a simple 5-insn sequence
358 root 1.161 without a system call and thus I<very> fast, but my GNU/Linux system also has
359 root 1.62 C<pthread_atfork> which is even faster).
360    
361     The big advantage of this flag is that you can forget about fork (and
362     forget about forgetting to tell libev about forking) when you use this
363     flag.
364    
365 root 1.161 This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
366 root 1.62 environment variable.
367    
368 root 1.260 =item C<EVFLAG_NOINOTIFY>
369    
370     When this flag is specified, then libev will not attempt to use the
371     I<inotify> API for it's C<ev_stat> watchers. Apart from debugging and
372     testing, this flag can be useful to conserve inotify file descriptors, as
373     otherwise each loop using C<ev_stat> watchers consumes one inotify handle.
374    
375 root 1.277 =item C<EVFLAG_SIGNALFD>
376 root 1.260
377 root 1.277 When this flag is specified, then libev will attempt to use the
378     I<signalfd> API for it's C<ev_signal> (and C<ev_child>) watchers. This API
379 root 1.278 delivers signals synchronously, which makes it both faster and might make
380     it possible to get the queued signal data. It can also simplify signal
381     handling with threads, as long as you properly block signals in your
382     threads that are not interested in handling them.
383 root 1.277
384     Signalfd will not be used by default as this changes your signal mask, and
385     there are a lot of shoddy libraries and programs (glib's threadpool for
386     example) that can't properly initialise their signal masks.
387 root 1.260
388 root 1.31 =item C<EVBACKEND_SELECT> (value 1, portable select backend)
389 root 1.1
390 root 1.29 This is your standard select(2) backend. Not I<completely> standard, as
391     libev tries to roll its own fd_set with no limits on the number of fds,
392     but if that fails, expect a fairly low limit on the number of fds when
393 root 1.102 using this backend. It doesn't scale too well (O(highest_fd)), but its
394     usually the fastest backend for a low number of (low-numbered :) fds.
395    
396     To get good performance out of this backend you need a high amount of
397 root 1.161 parallelism (most of the file descriptors should be busy). If you are
398 root 1.102 writing a server, you should C<accept ()> in a loop to accept as many
399     connections as possible during one iteration. You might also want to have
400     a look at C<ev_set_io_collect_interval ()> to increase the amount of
401 root 1.155 readiness notifications you get per iteration.
402 root 1.1
403 root 1.179 This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
404     C<writefds> set (and to work around Microsoft Windows bugs, also onto the
405     C<exceptfds> set on that platform).
406    
407 root 1.31 =item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
408 root 1.1
409 root 1.102 And this is your standard poll(2) backend. It's more complicated
410     than select, but handles sparse fds better and has no artificial
411     limit on the number of fds you can use (except it will slow down
412     considerably with a lot of inactive fds). It scales similarly to select,
413     i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
414     performance tips.
415 root 1.1
416 root 1.179 This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
417     C<EV_WRITE> to C<POLLOUT | POLLERR | POLLHUP>.
418    
419 root 1.31 =item C<EVBACKEND_EPOLL> (value 4, Linux)
420 root 1.1
421 root 1.272 Use the linux-specific epoll(7) interface (for both pre- and post-2.6.9
422     kernels).
423    
424 root 1.29 For few fds, this backend is a bit little slower than poll and select,
425 root 1.94 but it scales phenomenally better. While poll and select usually scale
426     like O(total_fds) where n is the total number of fds (or the highest fd),
427 root 1.205 epoll scales either O(1) or O(active_fds).
428    
429 root 1.210 The epoll mechanism deserves honorable mention as the most misdesigned
430     of the more advanced event mechanisms: mere annoyances include silently
431     dropping file descriptors, requiring a system call per change per file
432     descriptor (and unnecessary guessing of parameters), problems with dup and
433     so on. The biggest issue is fork races, however - if a program forks then
434     I<both> parent and child process have to recreate the epoll set, which can
435     take considerable time (one syscall per file descriptor) and is of course
436     hard to detect.
437 root 1.1
438 root 1.210 Epoll is also notoriously buggy - embedding epoll fds I<should> work, but
439     of course I<doesn't>, and epoll just loves to report events for totally
440 root 1.205 I<different> file descriptors (even already closed ones, so one cannot
441     even remove them from the set) than registered in the set (especially
442     on SMP systems). Libev tries to counter these spurious notifications by
443     employing an additional generation counter and comparing that against the
444 root 1.306 events to filter out spurious ones, recreating the set when required. Last
445     not least, it also refuses to work with some file descriptors which work
446     perfectly fine with C<select> (files, many character devices...).
447 root 1.204
448 root 1.94 While stopping, setting and starting an I/O watcher in the same iteration
449 root 1.210 will result in some caching, there is still a system call per such
450     incident (because the same I<file descriptor> could point to a different
451     I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
452     file descriptors might not work very well if you register events for both
453     file descriptors.
454 root 1.29
455 root 1.102 Best performance from this backend is achieved by not unregistering all
456 root 1.183 watchers for a file descriptor until it has been closed, if possible,
457     i.e. keep at least one watcher active per fd at all times. Stopping and
458     starting a watcher (without re-setting it) also usually doesn't cause
459 root 1.206 extra overhead. A fork can both result in spurious notifications as well
460     as in libev having to destroy and recreate the epoll object, which can
461     take considerable time and thus should be avoided.
462 root 1.102
463 root 1.215 All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
464     faster than epoll for maybe up to a hundred file descriptors, depending on
465 root 1.214 the usage. So sad.
466 root 1.213
467 root 1.161 While nominally embeddable in other event loops, this feature is broken in
468 root 1.102 all kernel versions tested so far.
469    
470 root 1.179 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
471     C<EVBACKEND_POLL>.
472    
473 root 1.31 =item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
474 root 1.29
475 root 1.210 Kqueue deserves special mention, as at the time of this writing, it
476     was broken on all BSDs except NetBSD (usually it doesn't work reliably
477     with anything but sockets and pipes, except on Darwin, where of course
478     it's completely useless). Unlike epoll, however, whose brokenness
479     is by design, these kqueue bugs can (and eventually will) be fixed
480     without API changes to existing programs. For this reason it's not being
481     "auto-detected" unless you explicitly specify it in the flags (i.e. using
482     C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
483     system like NetBSD.
484 root 1.29
485 root 1.100 You still can embed kqueue into a normal poll or select backend and use it
486     only for sockets (after having made sure that sockets work with kqueue on
487     the target platform). See C<ev_embed> watchers for more info.
488    
489 root 1.29 It scales in the same way as the epoll backend, but the interface to the
490 root 1.100 kernel is more efficient (which says nothing about its actual speed, of
491     course). While stopping, setting and starting an I/O watcher does never
492 root 1.161 cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
493 root 1.206 two event changes per incident. Support for C<fork ()> is very bad (but
494     sane, unlike epoll) and it drops fds silently in similarly hard-to-detect
495     cases
496 root 1.29
497 root 1.102 This backend usually performs well under most conditions.
498    
499     While nominally embeddable in other event loops, this doesn't work
500     everywhere, so you might need to test for this. And since it is broken
501     almost everywhere, you should only use it when you have a lot of sockets
502     (for which it usually works), by embedding it into another event loop
503 root 1.223 (e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
504     also broken on OS X)) and, did I mention it, using it only for sockets.
505 root 1.102
506 root 1.179 This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
507     C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
508     C<NOTE_EOF>.
509    
510 root 1.31 =item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
511 root 1.29
512 root 1.102 This is not implemented yet (and might never be, unless you send me an
513     implementation). According to reports, C</dev/poll> only supports sockets
514     and is not embeddable, which would limit the usefulness of this backend
515     immensely.
516 root 1.29
517 root 1.31 =item C<EVBACKEND_PORT> (value 32, Solaris 10)
518 root 1.29
519 root 1.94 This uses the Solaris 10 event port mechanism. As with everything on Solaris,
520 root 1.29 it's really slow, but it still scales very well (O(active_fds)).
521    
522 root 1.161 Please note that Solaris event ports can deliver a lot of spurious
523 root 1.32 notifications, so you need to use non-blocking I/O or other means to avoid
524     blocking when no data (or space) is available.
525    
526 root 1.102 While this backend scales well, it requires one system call per active
527     file descriptor per loop iteration. For small and medium numbers of file
528     descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
529     might perform better.
530    
531 root 1.183 On the positive side, with the exception of the spurious readiness
532     notifications, this backend actually performed fully to specification
533     in all tests and is fully embeddable, which is a rare feat among the
534 root 1.206 OS-specific backends (I vastly prefer correctness over speed hacks).
535 root 1.117
536 root 1.179 This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
537     C<EVBACKEND_POLL>.
538    
539 root 1.31 =item C<EVBACKEND_ALL>
540 root 1.29
541     Try all backends (even potentially broken ones that wouldn't be tried
542     with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
543 root 1.31 C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.
544 root 1.1
545 root 1.102 It is definitely not recommended to use this flag.
546    
547 root 1.1 =back
548    
549 root 1.260 If one or more of the backend flags are or'ed into the flags value,
550     then only these backends will be tried (in the reverse order as listed
551     here). If none are specified, all backends in C<ev_recommended_backends
552     ()> will be tried.
553 root 1.29
554 root 1.183 Example: This is the most typical usage.
555 root 1.33
556 root 1.164 if (!ev_default_loop (0))
557     fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
558 root 1.33
559 root 1.183 Example: Restrict libev to the select and poll backends, and do not allow
560 root 1.33 environment settings to be taken into account:
561    
562 root 1.164 ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
563 root 1.33
564 root 1.183 Example: Use whatever libev has to offer, but make sure that kqueue is
565     used if available (warning, breaks stuff, best use only with your own
566     private event loop and only if you know the OS supports your types of
567     fds):
568 root 1.33
569 root 1.164 ev_default_loop (ev_recommended_backends () | EVBACKEND_KQUEUE);
570 root 1.33
571 root 1.1 =item struct ev_loop *ev_loop_new (unsigned int flags)
572    
573     Similar to C<ev_default_loop>, but always creates a new event loop that is
574 root 1.290 always distinct from the default loop.
575 root 1.1
576 root 1.290 Note that this function I<is> thread-safe, and one common way to use
577 root 1.139 libev with threads is indeed to create one loop per thread, and using the
578     default loop in the "main" or "initial" thread.
579    
580 root 1.54 Example: Try to create a event loop that uses epoll and nothing else.
581 root 1.34
582 root 1.164 struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
583     if (!epoller)
584     fatal ("no epoll found here, maybe it hides under your chair");
585 root 1.34
586 root 1.1 =item ev_default_destroy ()
587    
588 root 1.290 Destroys the default loop (frees all memory and kernel state etc.). None
589     of the active event watchers will be stopped in the normal sense, so
590     e.g. C<ev_is_active> might still return true. It is your responsibility to
591     either stop all watchers cleanly yourself I<before> calling this function,
592     or cope with the fact afterwards (which is usually the easiest thing, you
593     can just ignore the watchers and/or C<free ()> them for example).
594 root 1.1
595 root 1.203 Note that certain global state, such as signal state (and installed signal
596     handlers), will not be freed by this function, and related watchers (such
597     as signal and child watchers) would need to be stopped manually.
598 root 1.87
599     In general it is not advisable to call this function except in the
600     rare occasion where you really need to free e.g. the signal handling
601     pipe fds. If you need dynamically allocated loops it is better to use
602 root 1.269 C<ev_loop_new> and C<ev_loop_destroy>.
603 root 1.87
604 root 1.1 =item ev_loop_destroy (loop)
605    
606     Like C<ev_default_destroy>, but destroys an event loop created by an
607     earlier call to C<ev_loop_new>.
608    
609     =item ev_default_fork ()
610    
611 root 1.310 This function sets a flag that causes subsequent C<ev_run> iterations
612 root 1.119 to reinitialise the kernel state for backends that have one. Despite the
613     name, you can call it anytime, but it makes most sense after forking, in
614     the child process (or both child and parent, but that again makes little
615     sense). You I<must> call it in the child before using any of the libev
616 root 1.310 functions, and it will only take effect at the next C<ev_run> iteration.
617 root 1.119
618 root 1.291 Again, you I<have> to call it on I<any> loop that you want to re-use after
619     a fork, I<even if you do not plan to use the loop in the parent>. This is
620     because some kernel interfaces *cough* I<kqueue> *cough* do funny things
621     during fork.
622    
623 root 1.119 On the other hand, you only need to call this function in the child
624 root 1.310 process if and only if you want to use the event loop in the child. If
625     you just fork+exec or create a new loop in the child, you don't have to
626     call it at all (in fact, C<epoll> is so badly broken that it makes a
627     difference, but libev will usually detect this case on its own and do a
628     costly reset of the backend).
629 root 1.1
630 root 1.9 The function itself is quite fast and it's usually not a problem to call
631 root 1.1 it just in case after a fork. To make this easy, the function will fit in
632     quite nicely into a call to C<pthread_atfork>:
633    
634     pthread_atfork (0, 0, ev_default_fork);
635    
636     =item ev_loop_fork (loop)
637    
638     Like C<ev_default_fork>, but acts on an event loop created by
639     C<ev_loop_new>. Yes, you have to call this on every allocated event loop
640 root 1.291 after fork that you want to re-use in the child, and how you keep track of
641     them is entirely your own problem.
642 root 1.1
643 root 1.131 =item int ev_is_default_loop (loop)
644    
645 root 1.183 Returns true when the given loop is, in fact, the default loop, and false
646     otherwise.
647 root 1.131
648 root 1.291 =item unsigned int ev_iteration (loop)
649 root 1.66
650 root 1.310 Returns the current iteration count for the event loop, which is identical
651     to the number of times libev did poll for new events. It starts at C<0>
652     and happily wraps around with enough iterations.
653 root 1.66
654     This value can sometimes be useful as a generation counter of sorts (it
655     "ticks" the number of loop iterations), as it roughly corresponds with
656 root 1.291 C<ev_prepare> and C<ev_check> calls - and is incremented between the
657     prepare and check phases.
658 root 1.66
659 root 1.291 =item unsigned int ev_depth (loop)
660 root 1.247
661 root 1.310 Returns the number of times C<ev_run> was entered minus the number of
662     times C<ev_run> was exited, in other words, the recursion depth.
663 root 1.247
664 root 1.310 Outside C<ev_run>, this number is zero. In a callback, this number is
665     C<1>, unless C<ev_run> was invoked recursively (or from another thread),
666 root 1.247 in which case it is higher.
667    
668 root 1.310 Leaving C<ev_run> abnormally (setjmp/longjmp, cancelling the thread
669 root 1.291 etc.), doesn't count as "exit" - consider this as a hint to avoid such
670 root 1.310 ungentleman-like behaviour unless it's really convenient.
671 root 1.247
672 root 1.31 =item unsigned int ev_backend (loop)
673 root 1.1
674 root 1.31 Returns one of the C<EVBACKEND_*> flags indicating the event backend in
675 root 1.1 use.
676    
677 root 1.9 =item ev_tstamp ev_now (loop)
678 root 1.1
679     Returns the current "event loop time", which is the time the event loop
680 root 1.34 received events and started processing them. This timestamp does not
681     change as long as callbacks are being processed, and this is also the base
682     time used for relative timers. You can treat it as the timestamp of the
683 root 1.92 event occurring (or more correctly, libev finding out about it).
684 root 1.1
685 root 1.176 =item ev_now_update (loop)
686    
687     Establishes the current time by querying the kernel, updating the time
688     returned by C<ev_now ()> in the progress. This is a costly operation and
689 root 1.310 is usually done automatically within C<ev_run ()>.
690 root 1.176
691     This function is rarely useful, but when some event callback runs for a
692     very long time without entering the event loop, updating libev's idea of
693     the current time is a good idea.
694    
695 root 1.238 See also L<The special problem of time updates> in the C<ev_timer> section.
696 root 1.176
697 root 1.231 =item ev_suspend (loop)
698    
699     =item ev_resume (loop)
700    
701 root 1.310 These two functions suspend and resume an event loop, for use when the
702     loop is not used for a while and timeouts should not be processed.
703 root 1.231
704     A typical use case would be an interactive program such as a game: When
705     the user presses C<^Z> to suspend the game and resumes it an hour later it
706     would be best to handle timeouts as if no time had actually passed while
707     the program was suspended. This can be achieved by calling C<ev_suspend>
708     in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
709     C<ev_resume> directly afterwards to resume timer processing.
710    
711     Effectively, all C<ev_timer> watchers will be delayed by the time spend
712     between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
713     will be rescheduled (that is, they will lose any events that would have
714 sf-exg 1.298 occurred while suspended).
715 root 1.231
716     After calling C<ev_suspend> you B<must not> call I<any> function on the
717     given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
718     without a previous call to C<ev_suspend>.
719    
720     Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
721     event loop time (see C<ev_now_update>).
722    
723 root 1.310 =item ev_run (loop, int flags)
724 root 1.1
725     Finally, this is it, the event handler. This function usually is called
726 root 1.268 after you have initialised all your watchers and you want to start
727 root 1.310 handling events. It will ask the operating system for any new events, call
728     the watcher callbacks, an then repeat the whole process indefinitely: This
729     is why event loops are called I<loops>.
730 root 1.1
731 root 1.310 If the flags argument is specified as C<0>, it will keep handling events
732     until either no event watchers are active anymore or C<ev_break> was
733     called.
734 root 1.1
735 root 1.310 Please note that an explicit C<ev_break> is usually better than
736 root 1.34 relying on all watchers to be stopped when deciding when a program has
737 root 1.183 finished (especially in interactive programs), but having a program
738     that automatically loops as long as it has to and no longer by virtue
739     of relying on its watchers stopping correctly, that is truly a thing of
740     beauty.
741 root 1.34
742 root 1.310 A flags value of C<EVRUN_NOWAIT> will look for new events, will handle
743     those events and any already outstanding ones, but will not wait and
744     block your process in case there are no events and will return after one
745     iteration of the loop. This is sometimes useful to poll and handle new
746     events while doing lengthy calculations, to keep the program responsive.
747 root 1.1
748 root 1.310 A flags value of C<EVRUN_ONCE> will look for new events (waiting if
749 root 1.183 necessary) and will handle those and any already outstanding ones. It
750     will block your process until at least one new event arrives (which could
751 root 1.208 be an event internal to libev itself, so there is no guarantee that a
752 root 1.183 user-registered callback will be called), and will return after one
753     iteration of the loop.
754    
755     This is useful if you are waiting for some external event in conjunction
756     with something not expressible using other libev watchers (i.e. "roll your
757 root 1.310 own C<ev_run>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
758 root 1.33 usually a better approach for this kind of thing.
759    
760 root 1.310 Here are the gory details of what C<ev_run> does:
761 root 1.33
762 root 1.310 - Increment loop depth.
763     - Reset the ev_break status.
764 root 1.77 - Before the first iteration, call any pending watchers.
765 root 1.310 LOOP:
766     - If EVFLAG_FORKCHECK was used, check for a fork.
767 root 1.171 - If a fork was detected (by any means), queue and call all fork watchers.
768 root 1.113 - Queue and call all prepare watchers.
769 root 1.310 - If ev_break was called, goto FINISH.
770 root 1.171 - If we have been forked, detach and recreate the kernel state
771     as to not disturb the other process.
772 root 1.33 - Update the kernel state with all outstanding changes.
773 root 1.171 - Update the "event loop time" (ev_now ()).
774 root 1.113 - Calculate for how long to sleep or block, if at all
775 root 1.310 (active idle watchers, EVRUN_NOWAIT or not having
776 root 1.113 any active watchers at all will result in not sleeping).
777     - Sleep if the I/O and timer collect interval say so.
778 root 1.310 - Increment loop iteration counter.
779 root 1.33 - Block the process, waiting for any events.
780     - Queue all outstanding I/O (fd) events.
781 root 1.171 - Update the "event loop time" (ev_now ()), and do time jump adjustments.
782 root 1.183 - Queue all expired timers.
783     - Queue all expired periodics.
784 root 1.310 - Queue all idle watchers with priority higher than that of pending events.
785 root 1.33 - Queue all check watchers.
786     - Call all queued watchers in reverse order (i.e. check watchers first).
787     Signals and child watchers are implemented as I/O watchers, and will
788     be handled here by queueing them when their watcher gets executed.
789 root 1.310 - If ev_break has been called, or EVRUN_ONCE or EVRUN_NOWAIT
790     were used, or there are no active watchers, goto FINISH, otherwise
791     continue with step LOOP.
792     FINISH:
793     - Reset the ev_break status iff it was EVBREAK_ONE.
794     - Decrement the loop depth.
795     - Return.
796 root 1.27
797 root 1.114 Example: Queue some jobs and then loop until no events are outstanding
798 root 1.34 anymore.
799    
800     ... queue jobs here, make sure they register event watchers as long
801     ... as they still have work to do (even an idle watcher will do..)
802 root 1.310 ev_run (my_loop, 0);
803 root 1.171 ... jobs done or somebody called unloop. yeah!
804 root 1.34
805 root 1.310 =item ev_break (loop, how)
806 root 1.1
807 root 1.310 Can be used to make a call to C<ev_run> return early (but only after it
808 root 1.9 has processed all outstanding events). The C<how> argument must be either
809 root 1.310 C<EVBREAK_ONE>, which will make the innermost C<ev_run> call return, or
810     C<EVBREAK_ALL>, which will make all nested C<ev_run> calls return.
811 root 1.1
812 root 1.310 This "unloop state" will be cleared when entering C<ev_run> again.
813 root 1.115
814 root 1.310 It is safe to call C<ev_break> from outside any C<ev_run> calls. ##TODO##
815 root 1.194
816 root 1.1 =item ev_ref (loop)
817    
818     =item ev_unref (loop)
819    
820 root 1.9 Ref/unref can be used to add or remove a reference count on the event
821     loop: Every watcher keeps one reference, and as long as the reference
822 root 1.310 count is nonzero, C<ev_run> will not return on its own.
823 root 1.183
824 root 1.276 This is useful when you have a watcher that you never intend to
825 root 1.310 unregister, but that nevertheless should not keep C<ev_run> from
826 root 1.276 returning. In such a case, call C<ev_unref> after starting, and C<ev_ref>
827     before stopping it.
828 root 1.183
829 root 1.229 As an example, libev itself uses this for its internal signal pipe: It
830 root 1.310 is not visible to the libev user and should not keep C<ev_run> from
831 root 1.229 exiting if no event watchers registered by it are active. It is also an
832     excellent way to do this for generic recurring timers or from within
833     third-party libraries. Just remember to I<unref after start> and I<ref
834     before stop> (but only if the watcher wasn't active before, or was active
835     before, respectively. Note also that libev might stop watchers itself
836     (e.g. non-repeating timers) in which case you have to C<ev_ref>
837     in the callback).
838 root 1.1
839 root 1.310 Example: Create a signal watcher, but keep it from keeping C<ev_run>
840 root 1.34 running when nothing else is active.
841    
842 root 1.198 ev_signal exitsig;
843 root 1.164 ev_signal_init (&exitsig, sig_cb, SIGINT);
844     ev_signal_start (loop, &exitsig);
845     evf_unref (loop);
846 root 1.34
847 root 1.54 Example: For some weird reason, unregister the above signal handler again.
848 root 1.34
849 root 1.164 ev_ref (loop);
850     ev_signal_stop (loop, &exitsig);
851 root 1.34
852 root 1.97 =item ev_set_io_collect_interval (loop, ev_tstamp interval)
853    
854     =item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
855    
856     These advanced functions influence the time that libev will spend waiting
857 root 1.171 for events. Both time intervals are by default C<0>, meaning that libev
858     will try to invoke timer/periodic callbacks and I/O callbacks with minimum
859     latency.
860 root 1.97
861     Setting these to a higher value (the C<interval> I<must> be >= C<0>)
862 root 1.171 allows libev to delay invocation of I/O and timer/periodic callbacks
863     to increase efficiency of loop iterations (or to increase power-saving
864     opportunities).
865 root 1.97
866 root 1.183 The idea is that sometimes your program runs just fast enough to handle
867     one (or very few) event(s) per loop iteration. While this makes the
868     program responsive, it also wastes a lot of CPU time to poll for new
869 root 1.97 events, especially with backends like C<select ()> which have a high
870     overhead for the actual polling but can deliver many events at once.
871    
872     By setting a higher I<io collect interval> you allow libev to spend more
873     time collecting I/O events, so you can handle more events per iteration,
874     at the cost of increasing latency. Timeouts (both C<ev_periodic> and
875 ayin 1.101 C<ev_timer>) will be not affected. Setting this to a non-null value will
876 root 1.245 introduce an additional C<ev_sleep ()> call into most loop iterations. The
877     sleep time ensures that libev will not poll for I/O events more often then
878     once per this interval, on average.
879 root 1.97
880     Likewise, by setting a higher I<timeout collect interval> you allow libev
881     to spend more time collecting timeouts, at the expense of increased
882 root 1.183 latency/jitter/inexactness (the watcher callback will be called
883     later). C<ev_io> watchers will not be affected. Setting this to a non-null
884     value will not introduce any overhead in libev.
885 root 1.97
886 root 1.161 Many (busy) programs can usually benefit by setting the I/O collect
887 root 1.98 interval to a value near C<0.1> or so, which is often enough for
888     interactive servers (of course not for games), likewise for timeouts. It
889     usually doesn't make much sense to set it to a lower value than C<0.01>,
890 root 1.245 as this approaches the timing granularity of most systems. Note that if
891     you do transactions with the outside world and you can't increase the
892     parallelity, then this setting will limit your transaction rate (if you
893     need to poll once per transaction and the I/O collect interval is 0.01,
894 sf-exg 1.298 then you can't do more than 100 transactions per second).
895 root 1.97
896 root 1.171 Setting the I<timeout collect interval> can improve the opportunity for
897     saving power, as the program will "bundle" timer callback invocations that
898     are "near" in time together, by delaying some, thus reducing the number of
899     times the process sleeps and wakes up again. Another useful technique to
900     reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
901     they fire on, say, one-second boundaries only.
902    
903 root 1.245 Example: we only need 0.1s timeout granularity, and we wish not to poll
904     more often than 100 times per second:
905    
906     ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
907     ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
908    
909 root 1.253 =item ev_invoke_pending (loop)
910    
911     This call will simply invoke all pending watchers while resetting their
912 root 1.310 pending state. Normally, C<ev_run> does this automatically when required,
913 root 1.253 but when overriding the invoke callback this call comes handy.
914    
915 root 1.256 =item int ev_pending_count (loop)
916    
917     Returns the number of pending watchers - zero indicates that no watchers
918     are pending.
919    
920 root 1.253 =item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
921    
922     This overrides the invoke pending functionality of the loop: Instead of
923 root 1.310 invoking all pending watchers when there are any, C<ev_run> will call
924 root 1.253 this callback instead. This is useful, for example, when you want to
925     invoke the actual watchers inside another context (another thread etc.).
926    
927     If you want to reset the callback, use C<ev_invoke_pending> as new
928     callback.
929    
930     =item ev_set_loop_release_cb (loop, void (*release)(EV_P), void (*acquire)(EV_P))
931    
932     Sometimes you want to share the same loop between multiple threads. This
933     can be done relatively simply by putting mutex_lock/unlock calls around
934     each call to a libev function.
935    
936 root 1.310 However, C<ev_run> can run an indefinite time, so it is not feasible
937     to wait for it to return. One way around this is to wake up the event
938     loop via C<ev_break> and C<av_async_send>, another way is to set these
939     I<release> and I<acquire> callbacks on the loop.
940 root 1.253
941     When set, then C<release> will be called just before the thread is
942     suspended waiting for new events, and C<acquire> is called just
943     afterwards.
944    
945     Ideally, C<release> will just call your mutex_unlock function, and
946     C<acquire> will just call the mutex_lock function again.
947    
948 root 1.254 While event loop modifications are allowed between invocations of
949     C<release> and C<acquire> (that's their only purpose after all), no
950     modifications done will affect the event loop, i.e. adding watchers will
951     have no effect on the set of file descriptors being watched, or the time
952 root 1.310 waited. Use an C<ev_async> watcher to wake up C<ev_run> when you want it
953 root 1.254 to take note of any changes you made.
954    
955 root 1.310 In theory, threads executing C<ev_run> will be async-cancel safe between
956 root 1.254 invocations of C<release> and C<acquire>.
957    
958     See also the locking example in the C<THREADS> section later in this
959     document.
960    
961 root 1.253 =item ev_set_userdata (loop, void *data)
962    
963     =item ev_userdata (loop)
964    
965     Set and retrieve a single C<void *> associated with a loop. When
966     C<ev_set_userdata> has never been called, then C<ev_userdata> returns
967     C<0.>
968    
969     These two functions can be used to associate arbitrary data with a loop,
970     and are intended solely for the C<invoke_pending_cb>, C<release> and
971     C<acquire> callbacks described above, but of course can be (ab-)used for
972     any other purpose as well.
973    
974 root 1.309 =item ev_verify (loop)
975 root 1.159
976     This function only does something when C<EV_VERIFY> support has been
977 root 1.200 compiled in, which is the default for non-minimal builds. It tries to go
978 root 1.183 through all internal structures and checks them for validity. If anything
979     is found to be inconsistent, it will print an error message to standard
980     error and call C<abort ()>.
981 root 1.159
982     This can be used to catch bugs inside libev itself: under normal
983     circumstances, this function will never abort as of course libev keeps its
984     data structures consistent.
985    
986 root 1.1 =back
987    
988 root 1.42
989 root 1.1 =head1 ANATOMY OF A WATCHER
990    
991 root 1.200 In the following description, uppercase C<TYPE> in names stands for the
992     watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
993     watchers and C<ev_io_start> for I/O watchers.
994    
995 root 1.311 A watcher is an opaque structure that you allocate and register to record
996     your interest in some event. To make a concrete example, imagine you want
997     to wait for STDIN to become readable, you would create an C<ev_io> watcher
998     for that:
999 root 1.1
1000 root 1.198 static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
1001 root 1.164 {
1002     ev_io_stop (w);
1003 root 1.310 ev_break (loop, EVBREAK_ALL);
1004 root 1.164 }
1005    
1006     struct ev_loop *loop = ev_default_loop (0);
1007 root 1.200
1008 root 1.198 ev_io stdin_watcher;
1009 root 1.200
1010 root 1.164 ev_init (&stdin_watcher, my_cb);
1011     ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
1012     ev_io_start (loop, &stdin_watcher);
1013 root 1.200
1014 root 1.310 ev_run (loop, 0);
1015 root 1.1
1016     As you can see, you are responsible for allocating the memory for your
1017 root 1.200 watcher structures (and it is I<usually> a bad idea to do this on the
1018     stack).
1019    
1020     Each watcher has an associated watcher structure (called C<struct ev_TYPE>
1021     or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
1022 root 1.1
1023 root 1.311 Each watcher structure must be initialised by a call to C<ev_init (watcher
1024     *, callback)>, which expects a callback to be provided. This callback is
1025     invoked each time the event occurs (or, in the case of I/O watchers, each
1026     time the event loop detects that the file descriptor given is readable
1027     and/or writable).
1028 root 1.1
1029 root 1.200 Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
1030     macro to configure it, with arguments specific to the watcher type. There
1031     is also a macro to combine initialisation and setting in one call: C<<
1032     ev_TYPE_init (watcher *, callback, ...) >>.
1033 root 1.1
1034     To make the watcher actually watch out for events, you have to start it
1035 root 1.200 with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
1036 root 1.1 *) >>), and you can stop watching for events at any time by calling the
1037 root 1.200 corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
1038 root 1.1
1039     As long as your watcher is active (has been started but not stopped) you
1040     must not touch the values stored in it. Most specifically you must never
1041 root 1.200 reinitialise it or call its C<ev_TYPE_set> macro.
1042 root 1.1
1043     Each and every callback receives the event loop pointer as first, the
1044     registered watcher structure as second, and a bitset of received events as
1045     third argument.
1046    
1047 root 1.14 The received events usually include a single bit per event type received
1048 root 1.1 (you can receive multiple events at the same time). The possible bit masks
1049     are:
1050    
1051     =over 4
1052    
1053 root 1.10 =item C<EV_READ>
1054 root 1.1
1055 root 1.10 =item C<EV_WRITE>
1056 root 1.1
1057 root 1.10 The file descriptor in the C<ev_io> watcher has become readable and/or
1058 root 1.1 writable.
1059    
1060 root 1.289 =item C<EV_TIMER>
1061 root 1.1
1062 root 1.10 The C<ev_timer> watcher has timed out.
1063 root 1.1
1064 root 1.10 =item C<EV_PERIODIC>
1065 root 1.1
1066 root 1.10 The C<ev_periodic> watcher has timed out.
1067 root 1.1
1068 root 1.10 =item C<EV_SIGNAL>
1069 root 1.1
1070 root 1.10 The signal specified in the C<ev_signal> watcher has been received by a thread.
1071 root 1.1
1072 root 1.10 =item C<EV_CHILD>
1073 root 1.1
1074 root 1.10 The pid specified in the C<ev_child> watcher has received a status change.
1075 root 1.1
1076 root 1.48 =item C<EV_STAT>
1077    
1078     The path specified in the C<ev_stat> watcher changed its attributes somehow.
1079    
1080 root 1.10 =item C<EV_IDLE>
1081 root 1.1
1082 root 1.10 The C<ev_idle> watcher has determined that you have nothing better to do.
1083 root 1.1
1084 root 1.10 =item C<EV_PREPARE>
1085 root 1.1
1086 root 1.10 =item C<EV_CHECK>
1087 root 1.1
1088 root 1.310 All C<ev_prepare> watchers are invoked just I<before> C<ev_run> starts
1089 root 1.10 to gather new events, and all C<ev_check> watchers are invoked just after
1090 root 1.310 C<ev_run> has gathered them, but before it invokes any callbacks for any
1091 root 1.1 received events. Callbacks of both watcher types can start and stop as
1092     many watchers as they want, and all of them will be taken into account
1093 root 1.10 (for example, a C<ev_prepare> watcher might start an idle watcher to keep
1094 root 1.310 C<ev_run> from blocking).
1095 root 1.1
1096 root 1.50 =item C<EV_EMBED>
1097    
1098     The embedded event loop specified in the C<ev_embed> watcher needs attention.
1099    
1100     =item C<EV_FORK>
1101    
1102     The event loop has been resumed in the child process after fork (see
1103     C<ev_fork>).
1104    
1105 root 1.122 =item C<EV_ASYNC>
1106    
1107     The given async watcher has been asynchronously notified (see C<ev_async>).
1108    
1109 root 1.229 =item C<EV_CUSTOM>
1110    
1111     Not ever sent (or otherwise used) by libev itself, but can be freely used
1112     by libev users to signal watchers (e.g. via C<ev_feed_event>).
1113    
1114 root 1.10 =item C<EV_ERROR>
1115 root 1.1
1116 root 1.161 An unspecified error has occurred, the watcher has been stopped. This might
1117 root 1.1 happen because the watcher could not be properly started because libev
1118     ran out of memory, a file descriptor was found to be closed or any other
1119 root 1.197 problem. Libev considers these application bugs.
1120    
1121     You best act on it by reporting the problem and somehow coping with the
1122     watcher being stopped. Note that well-written programs should not receive
1123     an error ever, so when your watcher receives it, this usually indicates a
1124     bug in your program.
1125 root 1.1
1126 root 1.183 Libev will usually signal a few "dummy" events together with an error, for
1127     example it might indicate that a fd is readable or writable, and if your
1128     callbacks is well-written it can just attempt the operation and cope with
1129     the error from read() or write(). This will not work in multi-threaded
1130     programs, though, as the fd could already be closed and reused for another
1131     thing, so beware.
1132 root 1.1
1133     =back
1134    
1135 root 1.311 =head2 WATCHER STATES
1136    
1137     There are various watcher states mentioned throughout this manual -
1138     active, pending and so on. In this section these states and the rules to
1139     transition between them will be described in more detail - and while these
1140     rules might look complicated, they usually do "the right thing".
1141    
1142     =over 4
1143    
1144     =item initialiased
1145    
1146     Before a watcher can be registered with the event looop it has to be
1147     initialised. This can be done with a call to C<ev_TYPE_init>, or calls to
1148     C<ev_init> followed by the watcher-specific C<ev_TYPE_set> function.
1149    
1150     In this state it is simply some block of memory that is suitable for use
1151     in an event loop. It can be moved around, freed, reused etc. at will.
1152    
1153     =item started/running/active
1154    
1155     Once a watcher has been started with a call to C<ev_TYPE_start> it becomes
1156     property of the event loop, and is actively waiting for events. While in
1157     this state it cannot be accessed (except in a few documented ways), moved,
1158     freed or anything else - the only legal thing is to keep a pointer to it,
1159     and call libev functions on it that are documented to work on active watchers.
1160    
1161     =item pending
1162    
1163     If a watcher is active and libev determines that an event it is interested
1164     in has occured (such as a timer expiring), it will become pending. It will
1165     stay in this pending state until either it is stopped or its callback is
1166     about to be invoked, so it is not normally pending inside the watcher
1167     callback.
1168    
1169     The watcher might or might not be active while it is pending (for example,
1170     an expired non-repeating timer can be pending but no longer active). If it
1171     is stopped, it can be freely accessed (e.g. by calling C<ev_TYPE_set>),
1172     but it is still property of the event loop at this time, so cannot be
1173     moved, freed or reused. And if it is active the rules described in the
1174     previous item still apply.
1175    
1176     It is also possible to feed an event on a watcher that is not active (e.g.
1177     via C<ev_feed_event>), in which case it becomes pending without being
1178     active.
1179    
1180     =item stopped
1181    
1182     A watcher can be stopped implicitly by libev (in which case it might still
1183     be pending), or explicitly by calling its C<ev_TYPE_stop> function. The
1184     latter will clear any pending state the watcher might be in, regardless
1185     of whether it was active or not, so stopping a watcher explicitly before
1186     freeing it is often a good idea.
1187    
1188     While stopped (and not pending) the watcher is essentially in the
1189     initialised state, that is it can be reused, moved, modified in any way
1190     you wish.
1191    
1192     =back
1193    
1194 root 1.42 =head2 GENERIC WATCHER FUNCTIONS
1195 root 1.36
1196     =over 4
1197    
1198     =item C<ev_init> (ev_TYPE *watcher, callback)
1199    
1200     This macro initialises the generic portion of a watcher. The contents
1201     of the watcher object can be arbitrary (so C<malloc> will do). Only
1202     the generic parts of the watcher are initialised, you I<need> to call
1203     the type-specific C<ev_TYPE_set> macro afterwards to initialise the
1204     type-specific parts. For each type there is also a C<ev_TYPE_init> macro
1205     which rolls both calls into one.
1206    
1207     You can reinitialise a watcher at any time as long as it has been stopped
1208     (or never started) and there are no pending events outstanding.
1209    
1210 root 1.198 The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
1211 root 1.36 int revents)>.
1212    
1213 root 1.183 Example: Initialise an C<ev_io> watcher in two steps.
1214    
1215     ev_io w;
1216     ev_init (&w, my_cb);
1217     ev_io_set (&w, STDIN_FILENO, EV_READ);
1218    
1219 root 1.274 =item C<ev_TYPE_set> (ev_TYPE *watcher, [args])
1220 root 1.36
1221     This macro initialises the type-specific parts of a watcher. You need to
1222     call C<ev_init> at least once before you call this macro, but you can
1223     call C<ev_TYPE_set> any number of times. You must not, however, call this
1224     macro on a watcher that is active (it can be pending, however, which is a
1225     difference to the C<ev_init> macro).
1226    
1227     Although some watcher types do not have type-specific arguments
1228     (e.g. C<ev_prepare>) you still need to call its C<set> macro.
1229    
1230 root 1.183 See C<ev_init>, above, for an example.
1231    
1232 root 1.36 =item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
1233    
1234 root 1.161 This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
1235     calls into a single call. This is the most convenient method to initialise
1236 root 1.36 a watcher. The same limitations apply, of course.
1237    
1238 root 1.183 Example: Initialise and set an C<ev_io> watcher in one step.
1239    
1240     ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
1241    
1242 root 1.274 =item C<ev_TYPE_start> (loop, ev_TYPE *watcher)
1243 root 1.36
1244     Starts (activates) the given watcher. Only active watchers will receive
1245     events. If the watcher is already active nothing will happen.
1246    
1247 root 1.183 Example: Start the C<ev_io> watcher that is being abused as example in this
1248     whole section.
1249    
1250     ev_io_start (EV_DEFAULT_UC, &w);
1251    
1252 root 1.274 =item C<ev_TYPE_stop> (loop, ev_TYPE *watcher)
1253 root 1.36
1254 root 1.195 Stops the given watcher if active, and clears the pending status (whether
1255     the watcher was active or not).
1256    
1257     It is possible that stopped watchers are pending - for example,
1258     non-repeating timers are being stopped when they become pending - but
1259     calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
1260     pending. If you want to free or reuse the memory used by the watcher it is
1261     therefore a good idea to always call its C<ev_TYPE_stop> function.
1262 root 1.36
1263     =item bool ev_is_active (ev_TYPE *watcher)
1264    
1265     Returns a true value iff the watcher is active (i.e. it has been started
1266     and not yet been stopped). As long as a watcher is active you must not modify
1267     it.
1268    
1269     =item bool ev_is_pending (ev_TYPE *watcher)
1270    
1271     Returns a true value iff the watcher is pending, (i.e. it has outstanding
1272     events but its callback has not yet been invoked). As long as a watcher
1273     is pending (but not active) you must not call an init function on it (but
1274 root 1.73 C<ev_TYPE_set> is safe), you must not change its priority, and you must
1275     make sure the watcher is available to libev (e.g. you cannot C<free ()>
1276     it).
1277 root 1.36
1278 root 1.55 =item callback ev_cb (ev_TYPE *watcher)
1279 root 1.36
1280     Returns the callback currently set on the watcher.
1281    
1282     =item ev_cb_set (ev_TYPE *watcher, callback)
1283    
1284     Change the callback. You can change the callback at virtually any time
1285     (modulo threads).
1286    
1287 root 1.274 =item ev_set_priority (ev_TYPE *watcher, int priority)
1288 root 1.67
1289     =item int ev_priority (ev_TYPE *watcher)
1290    
1291     Set and query the priority of the watcher. The priority is a small
1292     integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
1293     (default: C<-2>). Pending watchers with higher priority will be invoked
1294     before watchers with lower priority, but priority will not keep watchers
1295     from being executed (except for C<ev_idle> watchers).
1296    
1297     If you need to suppress invocation when higher priority events are pending
1298     you need to look at C<ev_idle> watchers, which provide this functionality.
1299    
1300 root 1.73 You I<must not> change the priority of a watcher as long as it is active or
1301     pending.
1302    
1303 root 1.233 Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
1304     fine, as long as you do not mind that the priority value you query might
1305     or might not have been clamped to the valid range.
1306    
1307 root 1.67 The default priority used by watchers when no priority has been set is
1308     always C<0>, which is supposed to not be too high and not be too low :).
1309    
1310 root 1.235 See L<WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
1311 root 1.233 priorities.
1312 root 1.67
1313 root 1.74 =item ev_invoke (loop, ev_TYPE *watcher, int revents)
1314    
1315     Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
1316     C<loop> nor C<revents> need to be valid as long as the watcher callback
1317 root 1.183 can deal with that fact, as both are simply passed through to the
1318     callback.
1319 root 1.74
1320     =item int ev_clear_pending (loop, ev_TYPE *watcher)
1321    
1322 root 1.183 If the watcher is pending, this function clears its pending status and
1323     returns its C<revents> bitset (as if its callback was invoked). If the
1324 root 1.74 watcher isn't pending it does nothing and returns C<0>.
1325    
1326 root 1.183 Sometimes it can be useful to "poll" a watcher instead of waiting for its
1327     callback to be invoked, which can be accomplished with this function.
1328    
1329 root 1.274 =item ev_feed_event (loop, ev_TYPE *watcher, int revents)
1330 root 1.273
1331     Feeds the given event set into the event loop, as if the specified event
1332     had happened for the specified watcher (which must be a pointer to an
1333     initialised but not necessarily started event watcher). Obviously you must
1334     not free the watcher as long as it has pending events.
1335    
1336     Stopping the watcher, letting libev invoke it, or calling
1337     C<ev_clear_pending> will clear the pending event, even if the watcher was
1338     not started in the first place.
1339    
1340     See also C<ev_feed_fd_event> and C<ev_feed_signal_event> for related
1341     functions that do not need a watcher.
1342    
1343 root 1.36 =back
1344    
1345    
1346 root 1.1 =head2 ASSOCIATING CUSTOM DATA WITH A WATCHER
1347    
1348     Each watcher has, by default, a member C<void *data> that you can change
1349 root 1.183 and read at any time: libev will completely ignore it. This can be used
1350 root 1.1 to associate arbitrary data with your watcher. If you need more data and
1351     don't want to allocate memory and store a pointer to it in that data
1352     member, you can also "subclass" the watcher type and provide your own
1353     data:
1354    
1355 root 1.164 struct my_io
1356     {
1357 root 1.198 ev_io io;
1358 root 1.164 int otherfd;
1359     void *somedata;
1360     struct whatever *mostinteresting;
1361 root 1.178 };
1362    
1363     ...
1364     struct my_io w;
1365     ev_io_init (&w.io, my_cb, fd, EV_READ);
1366 root 1.1
1367     And since your callback will be called with a pointer to the watcher, you
1368     can cast it back to your own type:
1369    
1370 root 1.198 static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
1371 root 1.164 {
1372     struct my_io *w = (struct my_io *)w_;
1373     ...
1374     }
1375 root 1.1
1376 root 1.55 More interesting and less C-conformant ways of casting your callback type
1377     instead have been omitted.
1378    
1379 root 1.178 Another common scenario is to use some data structure with multiple
1380     embedded watchers:
1381 root 1.55
1382 root 1.164 struct my_biggy
1383     {
1384     int some_data;
1385     ev_timer t1;
1386     ev_timer t2;
1387     }
1388 root 1.55
1389 root 1.178 In this case getting the pointer to C<my_biggy> is a bit more
1390     complicated: Either you store the address of your C<my_biggy> struct
1391 root 1.183 in the C<data> member of the watcher (for woozies), or you need to use
1392     some pointer arithmetic using C<offsetof> inside your watchers (for real
1393     programmers):
1394 root 1.55
1395 root 1.164 #include <stddef.h>
1396    
1397     static void
1398 root 1.198 t1_cb (EV_P_ ev_timer *w, int revents)
1399 root 1.164 {
1400 root 1.242 struct my_biggy big = (struct my_biggy *)
1401 root 1.164 (((char *)w) - offsetof (struct my_biggy, t1));
1402     }
1403 root 1.55
1404 root 1.164 static void
1405 root 1.198 t2_cb (EV_P_ ev_timer *w, int revents)
1406 root 1.164 {
1407 root 1.242 struct my_biggy big = (struct my_biggy *)
1408 root 1.164 (((char *)w) - offsetof (struct my_biggy, t2));
1409     }
1410 root 1.1
1411 root 1.233 =head2 WATCHER PRIORITY MODELS
1412    
1413     Many event loops support I<watcher priorities>, which are usually small
1414     integers that influence the ordering of event callback invocation
1415     between watchers in some way, all else being equal.
1416    
1417     In libev, Watcher priorities can be set using C<ev_set_priority>. See its
1418     description for the more technical details such as the actual priority
1419     range.
1420    
1421     There are two common ways how these these priorities are being interpreted
1422     by event loops:
1423    
1424     In the more common lock-out model, higher priorities "lock out" invocation
1425     of lower priority watchers, which means as long as higher priority
1426     watchers receive events, lower priority watchers are not being invoked.
1427    
1428     The less common only-for-ordering model uses priorities solely to order
1429     callback invocation within a single event loop iteration: Higher priority
1430     watchers are invoked before lower priority ones, but they all get invoked
1431     before polling for new events.
1432    
1433     Libev uses the second (only-for-ordering) model for all its watchers
1434     except for idle watchers (which use the lock-out model).
1435    
1436     The rationale behind this is that implementing the lock-out model for
1437     watchers is not well supported by most kernel interfaces, and most event
1438     libraries will just poll for the same events again and again as long as
1439     their callbacks have not been executed, which is very inefficient in the
1440     common case of one high-priority watcher locking out a mass of lower
1441     priority ones.
1442    
1443     Static (ordering) priorities are most useful when you have two or more
1444     watchers handling the same resource: a typical usage example is having an
1445     C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
1446     timeouts. Under load, data might be received while the program handles
1447     other jobs, but since timers normally get invoked first, the timeout
1448     handler will be executed before checking for data. In that case, giving
1449     the timer a lower priority than the I/O watcher ensures that I/O will be
1450     handled first even under adverse conditions (which is usually, but not
1451     always, what you want).
1452    
1453     Since idle watchers use the "lock-out" model, meaning that idle watchers
1454     will only be executed when no same or higher priority watchers have
1455     received events, they can be used to implement the "lock-out" model when
1456     required.
1457    
1458     For example, to emulate how many other event libraries handle priorities,
1459     you can associate an C<ev_idle> watcher to each such watcher, and in
1460     the normal watcher callback, you just start the idle watcher. The real
1461     processing is done in the idle watcher callback. This causes libev to
1462 sf-exg 1.298 continuously poll and process kernel event data for the watcher, but when
1463 root 1.233 the lock-out case is known to be rare (which in turn is rare :), this is
1464     workable.
1465    
1466     Usually, however, the lock-out model implemented that way will perform
1467     miserably under the type of load it was designed to handle. In that case,
1468     it might be preferable to stop the real watcher before starting the
1469     idle watcher, so the kernel will not have to process the event in case
1470     the actual processing will be delayed for considerable time.
1471    
1472     Here is an example of an I/O watcher that should run at a strictly lower
1473     priority than the default, and which should only process data when no
1474     other events are pending:
1475    
1476     ev_idle idle; // actual processing watcher
1477     ev_io io; // actual event watcher
1478    
1479     static void
1480     io_cb (EV_P_ ev_io *w, int revents)
1481     {
1482     // stop the I/O watcher, we received the event, but
1483     // are not yet ready to handle it.
1484     ev_io_stop (EV_A_ w);
1485    
1486 root 1.296 // start the idle watcher to handle the actual event.
1487 root 1.233 // it will not be executed as long as other watchers
1488     // with the default priority are receiving events.
1489     ev_idle_start (EV_A_ &idle);
1490     }
1491    
1492     static void
1493 root 1.242 idle_cb (EV_P_ ev_idle *w, int revents)
1494 root 1.233 {
1495     // actual processing
1496     read (STDIN_FILENO, ...);
1497    
1498     // have to start the I/O watcher again, as
1499     // we have handled the event
1500     ev_io_start (EV_P_ &io);
1501     }
1502    
1503     // initialisation
1504     ev_idle_init (&idle, idle_cb);
1505     ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
1506     ev_io_start (EV_DEFAULT_ &io);
1507    
1508     In the "real" world, it might also be beneficial to start a timer, so that
1509     low-priority connections can not be locked out forever under load. This
1510     enables your program to keep a lower latency for important connections
1511     during short periods of high load, while not completely locking out less
1512     important ones.
1513    
1514 root 1.1
1515     =head1 WATCHER TYPES
1516    
1517     This section describes each watcher in detail, but will not repeat
1518 root 1.48 information given in the last section. Any initialisation/set macros,
1519     functions and members specific to the watcher type are explained.
1520    
1521     Members are additionally marked with either I<[read-only]>, meaning that,
1522     while the watcher is active, you can look at the member and expect some
1523     sensible content, but you must not modify it (you can modify it while the
1524     watcher is stopped to your hearts content), or I<[read-write]>, which
1525     means you can expect it to have some sensible content while the watcher
1526     is active, but you can also modify it. Modifying it may not do something
1527     sensible or take immediate effect (or do anything at all), but libev will
1528     not crash or malfunction in any way.
1529 root 1.1
1530 root 1.34
1531 root 1.42 =head2 C<ev_io> - is this file descriptor readable or writable?
1532 root 1.1
1533 root 1.4 I/O watchers check whether a file descriptor is readable or writable
1534 root 1.42 in each iteration of the event loop, or, more precisely, when reading
1535     would not block the process and writing would at least be able to write
1536     some data. This behaviour is called level-triggering because you keep
1537     receiving events as long as the condition persists. Remember you can stop
1538     the watcher if you don't want to act on the event and neither want to
1539     receive future events.
1540 root 1.1
1541 root 1.23 In general you can register as many read and/or write event watchers per
1542 root 1.8 fd as you want (as long as you don't confuse yourself). Setting all file
1543     descriptors to non-blocking mode is also usually a good idea (but not
1544     required if you know what you are doing).
1545    
1546 root 1.183 If you cannot use non-blocking mode, then force the use of a
1547     known-to-be-good backend (at the time of this writing, this includes only
1548 root 1.239 C<EVBACKEND_SELECT> and C<EVBACKEND_POLL>). The same applies to file
1549     descriptors for which non-blocking operation makes no sense (such as
1550 sf-exg 1.298 files) - libev doesn't guarantee any specific behaviour in that case.
1551 root 1.8
1552 root 1.42 Another thing you have to watch out for is that it is quite easy to
1553 root 1.155 receive "spurious" readiness notifications, that is your callback might
1554 root 1.42 be called with C<EV_READ> but a subsequent C<read>(2) will actually block
1555     because there is no data. Not only are some backends known to create a
1556 root 1.161 lot of those (for example Solaris ports), it is very easy to get into
1557 root 1.42 this situation even with a relatively standard program structure. Thus
1558     it is best to always use non-blocking I/O: An extra C<read>(2) returning
1559     C<EAGAIN> is far preferable to a program hanging until some data arrives.
1560    
1561 root 1.183 If you cannot run the fd in non-blocking mode (for example you should
1562     not play around with an Xlib connection), then you have to separately
1563     re-test whether a file descriptor is really ready with a known-to-be good
1564     interface such as poll (fortunately in our Xlib example, Xlib already
1565     does this on its own, so its quite safe to use). Some people additionally
1566     use C<SIGALRM> and an interval timer, just to be sure you won't block
1567     indefinitely.
1568    
1569     But really, best use non-blocking mode.
1570 root 1.42
1571 root 1.81 =head3 The special problem of disappearing file descriptors
1572    
1573 root 1.94 Some backends (e.g. kqueue, epoll) need to be told about closing a file
1574 root 1.183 descriptor (either due to calling C<close> explicitly or any other means,
1575     such as C<dup2>). The reason is that you register interest in some file
1576 root 1.81 descriptor, but when it goes away, the operating system will silently drop
1577     this interest. If another file descriptor with the same number then is
1578     registered with libev, there is no efficient way to see that this is, in
1579     fact, a different file descriptor.
1580    
1581     To avoid having to explicitly tell libev about such cases, libev follows
1582     the following policy: Each time C<ev_io_set> is being called, libev
1583     will assume that this is potentially a new file descriptor, otherwise
1584     it is assumed that the file descriptor stays the same. That means that
1585     you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
1586     descriptor even if the file descriptor number itself did not change.
1587    
1588     This is how one would do it normally anyway, the important point is that
1589     the libev application should not optimise around libev but should leave
1590     optimisations to libev.
1591    
1592 root 1.95 =head3 The special problem of dup'ed file descriptors
1593 root 1.94
1594     Some backends (e.g. epoll), cannot register events for file descriptors,
1595 root 1.103 but only events for the underlying file descriptions. That means when you
1596 root 1.109 have C<dup ()>'ed file descriptors or weirder constellations, and register
1597     events for them, only one file descriptor might actually receive events.
1598 root 1.94
1599 root 1.103 There is no workaround possible except not registering events
1600     for potentially C<dup ()>'ed file descriptors, or to resort to
1601 root 1.94 C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
1602    
1603     =head3 The special problem of fork
1604    
1605     Some backends (epoll, kqueue) do not support C<fork ()> at all or exhibit
1606     useless behaviour. Libev fully supports fork, but needs to be told about
1607     it in the child.
1608    
1609     To support fork in your programs, you either have to call
1610     C<ev_default_fork ()> or C<ev_loop_fork ()> after a fork in the child,
1611     enable C<EVFLAG_FORKCHECK>, or resort to C<EVBACKEND_SELECT> or
1612     C<EVBACKEND_POLL>.
1613    
1614 root 1.138 =head3 The special problem of SIGPIPE
1615    
1616 root 1.183 While not really specific to libev, it is easy to forget about C<SIGPIPE>:
1617 root 1.174 when writing to a pipe whose other end has been closed, your program gets
1618 root 1.183 sent a SIGPIPE, which, by default, aborts your program. For most programs
1619 root 1.174 this is sensible behaviour, for daemons, this is usually undesirable.
1620 root 1.138
1621     So when you encounter spurious, unexplained daemon exits, make sure you
1622     ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
1623     somewhere, as that would have given you a big clue).
1624    
1625 root 1.284 =head3 The special problem of accept()ing when you can't
1626    
1627     Many implementations of the POSIX C<accept> function (for example,
1628 sf-exg 1.292 found in post-2004 Linux) have the peculiar behaviour of not removing a
1629 root 1.284 connection from the pending queue in all error cases.
1630    
1631     For example, larger servers often run out of file descriptors (because
1632     of resource limits), causing C<accept> to fail with C<ENFILE> but not
1633     rejecting the connection, leading to libev signalling readiness on
1634     the next iteration again (the connection still exists after all), and
1635     typically causing the program to loop at 100% CPU usage.
1636    
1637     Unfortunately, the set of errors that cause this issue differs between
1638     operating systems, there is usually little the app can do to remedy the
1639     situation, and no known thread-safe method of removing the connection to
1640     cope with overload is known (to me).
1641    
1642     One of the easiest ways to handle this situation is to just ignore it
1643     - when the program encounters an overload, it will just loop until the
1644     situation is over. While this is a form of busy waiting, no OS offers an
1645     event-based way to handle this situation, so it's the best one can do.
1646    
1647     A better way to handle the situation is to log any errors other than
1648     C<EAGAIN> and C<EWOULDBLOCK>, making sure not to flood the log with such
1649     messages, and continue as usual, which at least gives the user an idea of
1650     what could be wrong ("raise the ulimit!"). For extra points one could stop
1651     the C<ev_io> watcher on the listening fd "for a while", which reduces CPU
1652     usage.
1653    
1654     If your program is single-threaded, then you could also keep a dummy file
1655     descriptor for overload situations (e.g. by opening F</dev/null>), and
1656     when you run into C<ENFILE> or C<EMFILE>, close it, run C<accept>,
1657     close that fd, and create a new dummy fd. This will gracefully refuse
1658     clients under typical overload conditions.
1659    
1660     The last way to handle it is to simply log the error and C<exit>, as
1661     is often done with C<malloc> failures, but this results in an easy
1662     opportunity for a DoS attack.
1663 root 1.81
1664 root 1.82 =head3 Watcher-Specific Functions
1665    
1666 root 1.1 =over 4
1667    
1668     =item ev_io_init (ev_io *, callback, int fd, int events)
1669    
1670     =item ev_io_set (ev_io *, int fd, int events)
1671    
1672 root 1.42 Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
1673 root 1.183 receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
1674     C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
1675 root 1.32
1676 root 1.48 =item int fd [read-only]
1677    
1678     The file descriptor being watched.
1679    
1680     =item int events [read-only]
1681    
1682     The events being watched.
1683    
1684 root 1.1 =back
1685    
1686 root 1.111 =head3 Examples
1687    
1688 root 1.54 Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
1689 root 1.34 readable, but only once. Since it is likely line-buffered, you could
1690 root 1.54 attempt to read a whole line in the callback.
1691 root 1.34
1692 root 1.164 static void
1693 root 1.198 stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
1694 root 1.164 {
1695     ev_io_stop (loop, w);
1696 root 1.183 .. read from stdin here (or from w->fd) and handle any I/O errors
1697 root 1.164 }
1698    
1699     ...
1700     struct ev_loop *loop = ev_default_init (0);
1701 root 1.198 ev_io stdin_readable;
1702 root 1.164 ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
1703     ev_io_start (loop, &stdin_readable);
1704 root 1.310 ev_run (loop, 0);
1705 root 1.34
1706    
1707 root 1.42 =head2 C<ev_timer> - relative and optionally repeating timeouts
1708 root 1.1
1709     Timer watchers are simple relative timers that generate an event after a
1710     given time, and optionally repeating in regular intervals after that.
1711    
1712     The timers are based on real time, that is, if you register an event that
1713 root 1.161 times out after an hour and you reset your system clock to January last
1714 root 1.183 year, it will still time out after (roughly) one hour. "Roughly" because
1715 root 1.28 detecting time jumps is hard, and some inaccuracies are unavoidable (the
1716 root 1.1 monotonic clock option helps a lot here).
1717    
1718 root 1.183 The callback is guaranteed to be invoked only I<after> its timeout has
1719 root 1.240 passed (not I<at>, so on systems with very low-resolution clocks this
1720     might introduce a small delay). If multiple timers become ready during the
1721     same loop iteration then the ones with earlier time-out values are invoked
1722 root 1.248 before ones of the same priority with later time-out values (but this is
1723 root 1.310 no longer true when a callback calls C<ev_run> recursively).
1724 root 1.175
1725 root 1.198 =head3 Be smart about timeouts
1726    
1727 root 1.199 Many real-world problems involve some kind of timeout, usually for error
1728 root 1.198 recovery. A typical example is an HTTP request - if the other side hangs,
1729     you want to raise some error after a while.
1730    
1731 root 1.199 What follows are some ways to handle this problem, from obvious and
1732     inefficient to smart and efficient.
1733 root 1.198
1734 root 1.199 In the following, a 60 second activity timeout is assumed - a timeout that
1735     gets reset to 60 seconds each time there is activity (e.g. each time some
1736     data or other life sign was received).
1737 root 1.198
1738     =over 4
1739    
1740 root 1.199 =item 1. Use a timer and stop, reinitialise and start it on activity.
1741 root 1.198
1742     This is the most obvious, but not the most simple way: In the beginning,
1743     start the watcher:
1744    
1745     ev_timer_init (timer, callback, 60., 0.);
1746     ev_timer_start (loop, timer);
1747    
1748 root 1.199 Then, each time there is some activity, C<ev_timer_stop> it, initialise it
1749     and start it again:
1750 root 1.198
1751     ev_timer_stop (loop, timer);
1752     ev_timer_set (timer, 60., 0.);
1753     ev_timer_start (loop, timer);
1754    
1755 root 1.199 This is relatively simple to implement, but means that each time there is
1756     some activity, libev will first have to remove the timer from its internal
1757     data structure and then add it again. Libev tries to be fast, but it's
1758     still not a constant-time operation.
1759 root 1.198
1760     =item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
1761    
1762     This is the easiest way, and involves using C<ev_timer_again> instead of
1763     C<ev_timer_start>.
1764    
1765 root 1.199 To implement this, configure an C<ev_timer> with a C<repeat> value
1766     of C<60> and then call C<ev_timer_again> at start and each time you
1767     successfully read or write some data. If you go into an idle state where
1768     you do not expect data to travel on the socket, you can C<ev_timer_stop>
1769     the timer, and C<ev_timer_again> will automatically restart it if need be.
1770    
1771     That means you can ignore both the C<ev_timer_start> function and the
1772     C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
1773     member and C<ev_timer_again>.
1774 root 1.198
1775     At start:
1776    
1777 root 1.243 ev_init (timer, callback);
1778 root 1.199 timer->repeat = 60.;
1779 root 1.198 ev_timer_again (loop, timer);
1780    
1781 root 1.199 Each time there is some activity:
1782 root 1.198
1783     ev_timer_again (loop, timer);
1784    
1785 root 1.199 It is even possible to change the time-out on the fly, regardless of
1786     whether the watcher is active or not:
1787 root 1.198
1788     timer->repeat = 30.;
1789     ev_timer_again (loop, timer);
1790    
1791     This is slightly more efficient then stopping/starting the timer each time
1792     you want to modify its timeout value, as libev does not have to completely
1793 root 1.199 remove and re-insert the timer from/into its internal data structure.
1794    
1795     It is, however, even simpler than the "obvious" way to do it.
1796 root 1.198
1797     =item 3. Let the timer time out, but then re-arm it as required.
1798    
1799     This method is more tricky, but usually most efficient: Most timeouts are
1800 root 1.199 relatively long compared to the intervals between other activity - in
1801     our example, within 60 seconds, there are usually many I/O events with
1802     associated activity resets.
1803 root 1.198
1804     In this case, it would be more efficient to leave the C<ev_timer> alone,
1805     but remember the time of last activity, and check for a real timeout only
1806     within the callback:
1807    
1808     ev_tstamp last_activity; // time of last activity
1809    
1810     static void
1811     callback (EV_P_ ev_timer *w, int revents)
1812     {
1813 root 1.199 ev_tstamp now = ev_now (EV_A);
1814 root 1.198 ev_tstamp timeout = last_activity + 60.;
1815    
1816 root 1.199 // if last_activity + 60. is older than now, we did time out
1817 root 1.198 if (timeout < now)
1818     {
1819 sf-exg 1.298 // timeout occurred, take action
1820 root 1.198 }
1821     else
1822     {
1823     // callback was invoked, but there was some activity, re-arm
1824 root 1.199 // the watcher to fire in last_activity + 60, which is
1825     // guaranteed to be in the future, so "again" is positive:
1826 root 1.216 w->repeat = timeout - now;
1827 root 1.198 ev_timer_again (EV_A_ w);
1828     }
1829     }
1830    
1831 root 1.199 To summarise the callback: first calculate the real timeout (defined
1832     as "60 seconds after the last activity"), then check if that time has
1833     been reached, which means something I<did>, in fact, time out. Otherwise
1834     the callback was invoked too early (C<timeout> is in the future), so
1835     re-schedule the timer to fire at that future time, to see if maybe we have
1836     a timeout then.
1837 root 1.198
1838     Note how C<ev_timer_again> is used, taking advantage of the
1839     C<ev_timer_again> optimisation when the timer is already running.
1840    
1841 root 1.199 This scheme causes more callback invocations (about one every 60 seconds
1842     minus half the average time between activity), but virtually no calls to
1843     libev to change the timeout.
1844    
1845     To start the timer, simply initialise the watcher and set C<last_activity>
1846     to the current time (meaning we just have some activity :), then call the
1847     callback, which will "do the right thing" and start the timer:
1848 root 1.198
1849 root 1.243 ev_init (timer, callback);
1850 root 1.198 last_activity = ev_now (loop);
1851 root 1.289 callback (loop, timer, EV_TIMER);
1852 root 1.198
1853 root 1.199 And when there is some activity, simply store the current time in
1854     C<last_activity>, no libev calls at all:
1855 root 1.198
1856 root 1.295 last_activity = ev_now (loop);
1857 root 1.198
1858     This technique is slightly more complex, but in most cases where the
1859     time-out is unlikely to be triggered, much more efficient.
1860    
1861 root 1.199 Changing the timeout is trivial as well (if it isn't hard-coded in the
1862     callback :) - just change the timeout and invoke the callback, which will
1863     fix things for you.
1864    
1865 root 1.200 =item 4. Wee, just use a double-linked list for your timeouts.
1866 root 1.199
1867 root 1.200 If there is not one request, but many thousands (millions...), all
1868     employing some kind of timeout with the same timeout value, then one can
1869     do even better:
1870 root 1.199
1871     When starting the timeout, calculate the timeout value and put the timeout
1872     at the I<end> of the list.
1873    
1874     Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
1875     the list is expected to fire (for example, using the technique #3).
1876    
1877     When there is some activity, remove the timer from the list, recalculate
1878     the timeout, append it to the end of the list again, and make sure to
1879     update the C<ev_timer> if it was taken from the beginning of the list.
1880    
1881     This way, one can manage an unlimited number of timeouts in O(1) time for
1882     starting, stopping and updating the timers, at the expense of a major
1883     complication, and having to use a constant timeout. The constant timeout
1884     ensures that the list stays sorted.
1885    
1886 root 1.198 =back
1887    
1888 root 1.200 So which method the best?
1889 root 1.199
1890 root 1.200 Method #2 is a simple no-brain-required solution that is adequate in most
1891     situations. Method #3 requires a bit more thinking, but handles many cases
1892     better, and isn't very complicated either. In most case, choosing either
1893     one is fine, with #3 being better in typical situations.
1894 root 1.199
1895     Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
1896     rather complicated, but extremely efficient, something that really pays
1897 root 1.200 off after the first million or so of active timers, i.e. it's usually
1898 root 1.199 overkill :)
1899    
1900 root 1.175 =head3 The special problem of time updates
1901    
1902 root 1.176 Establishing the current time is a costly operation (it usually takes at
1903     least two system calls): EV therefore updates its idea of the current
1904 root 1.310 time only before and after C<ev_run> collects new events, which causes a
1905 root 1.183 growing difference between C<ev_now ()> and C<ev_time ()> when handling
1906     lots of events in one iteration.
1907 root 1.175
1908 root 1.9 The relative timeouts are calculated relative to the C<ev_now ()>
1909     time. This is usually the right thing as this timestamp refers to the time
1910 root 1.28 of the event triggering whatever timeout you are modifying/starting. If
1911 root 1.175 you suspect event processing to be delayed and you I<need> to base the
1912     timeout on the current time, use something like this to adjust for this:
1913 root 1.9
1914     ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);
1915    
1916 root 1.177 If the event loop is suspended for a long time, you can also force an
1917 root 1.176 update of the time returned by C<ev_now ()> by calling C<ev_now_update
1918     ()>.
1919    
1920 root 1.257 =head3 The special problems of suspended animation
1921    
1922     When you leave the server world it is quite customary to hit machines that
1923     can suspend/hibernate - what happens to the clocks during such a suspend?
1924    
1925     Some quick tests made with a Linux 2.6.28 indicate that a suspend freezes
1926     all processes, while the clocks (C<times>, C<CLOCK_MONOTONIC>) continue
1927     to run until the system is suspended, but they will not advance while the
1928     system is suspended. That means, on resume, it will be as if the program
1929     was frozen for a few seconds, but the suspend time will not be counted
1930     towards C<ev_timer> when a monotonic clock source is used. The real time
1931     clock advanced as expected, but if it is used as sole clocksource, then a
1932     long suspend would be detected as a time jump by libev, and timers would
1933     be adjusted accordingly.
1934    
1935     I would not be surprised to see different behaviour in different between
1936     operating systems, OS versions or even different hardware.
1937    
1938     The other form of suspend (job control, or sending a SIGSTOP) will see a
1939     time jump in the monotonic clocks and the realtime clock. If the program
1940     is suspended for a very long time, and monotonic clock sources are in use,
1941     then you can expect C<ev_timer>s to expire as the full suspension time
1942     will be counted towards the timers. When no monotonic clock source is in
1943     use, then libev will again assume a timejump and adjust accordingly.
1944    
1945     It might be beneficial for this latter case to call C<ev_suspend>
1946     and C<ev_resume> in code that handles C<SIGTSTP>, to at least get
1947     deterministic behaviour in this case (you can do nothing against
1948     C<SIGSTOP>).
1949    
1950 root 1.82 =head3 Watcher-Specific Functions and Data Members
1951    
1952 root 1.1 =over 4
1953    
1954     =item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
1955    
1956     =item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
1957    
1958 root 1.157 Configure the timer to trigger after C<after> seconds. If C<repeat>
1959     is C<0.>, then it will automatically be stopped once the timeout is
1960     reached. If it is positive, then the timer will automatically be
1961     configured to trigger again C<repeat> seconds later, again, and again,
1962     until stopped manually.
1963    
1964     The timer itself will do a best-effort at avoiding drift, that is, if
1965     you configure a timer to trigger every 10 seconds, then it will normally
1966     trigger at exactly 10 second intervals. If, however, your program cannot
1967     keep up with the timer (because it takes longer than those 10 seconds to
1968     do stuff) the timer will not fire more than once per event loop iteration.
1969 root 1.1
1970 root 1.132 =item ev_timer_again (loop, ev_timer *)
1971 root 1.1
1972     This will act as if the timer timed out and restart it again if it is
1973     repeating. The exact semantics are:
1974    
1975 root 1.61 If the timer is pending, its pending status is cleared.
1976 root 1.1
1977 root 1.161 If the timer is started but non-repeating, stop it (as if it timed out).
1978 root 1.61
1979     If the timer is repeating, either start it if necessary (with the
1980     C<repeat> value), or reset the running timer to the C<repeat> value.
1981 root 1.1
1982 root 1.232 This sounds a bit complicated, see L<Be smart about timeouts>, above, for a
1983 root 1.198 usage example.
1984 root 1.183
1985 root 1.275 =item ev_tstamp ev_timer_remaining (loop, ev_timer *)
1986 root 1.258
1987     Returns the remaining time until a timer fires. If the timer is active,
1988     then this time is relative to the current event loop time, otherwise it's
1989     the timeout value currently configured.
1990    
1991     That is, after an C<ev_timer_set (w, 5, 7)>, C<ev_timer_remaining> returns
1992 sf-exg 1.280 C<5>. When the timer is started and one second passes, C<ev_timer_remaining>
1993 root 1.258 will return C<4>. When the timer expires and is restarted, it will return
1994     roughly C<7> (likely slightly less as callback invocation takes some time,
1995     too), and so on.
1996    
1997 root 1.48 =item ev_tstamp repeat [read-write]
1998    
1999     The current C<repeat> value. Will be used each time the watcher times out
2000 root 1.183 or C<ev_timer_again> is called, and determines the next timeout (if any),
2001 root 1.48 which is also when any modifications are taken into account.
2002 root 1.1
2003     =back
2004    
2005 root 1.111 =head3 Examples
2006    
2007 root 1.54 Example: Create a timer that fires after 60 seconds.
2008 root 1.34
2009 root 1.164 static void
2010 root 1.198 one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
2011 root 1.164 {
2012     .. one minute over, w is actually stopped right here
2013     }
2014    
2015 root 1.198 ev_timer mytimer;
2016 root 1.164 ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
2017     ev_timer_start (loop, &mytimer);
2018 root 1.34
2019 root 1.54 Example: Create a timeout timer that times out after 10 seconds of
2020 root 1.34 inactivity.
2021    
2022 root 1.164 static void
2023 root 1.198 timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
2024 root 1.164 {
2025     .. ten seconds without any activity
2026     }
2027    
2028 root 1.198 ev_timer mytimer;
2029 root 1.164 ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
2030     ev_timer_again (&mytimer); /* start timer */
2031 root 1.310 ev_run (loop, 0);
2032 root 1.164
2033     // and in some piece of code that gets executed on any "activity":
2034     // reset the timeout to start ticking again at 10 seconds
2035     ev_timer_again (&mytimer);
2036 root 1.34
2037    
2038 root 1.42 =head2 C<ev_periodic> - to cron or not to cron?
2039 root 1.1
2040     Periodic watchers are also timers of a kind, but they are very versatile
2041     (and unfortunately a bit complex).
2042    
2043 root 1.227 Unlike C<ev_timer>, periodic watchers are not based on real time (or
2044     relative time, the physical time that passes) but on wall clock time
2045     (absolute time, the thing you can read on your calender or clock). The
2046     difference is that wall clock time can run faster or slower than real
2047     time, and time jumps are not uncommon (e.g. when you adjust your
2048     wrist-watch).
2049    
2050     You can tell a periodic watcher to trigger after some specific point
2051     in time: for example, if you tell a periodic watcher to trigger "in 10
2052     seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
2053     not a delay) and then reset your system clock to January of the previous
2054     year, then it will take a year or more to trigger the event (unlike an
2055     C<ev_timer>, which would still trigger roughly 10 seconds after starting
2056     it, as it uses a relative timeout).
2057    
2058     C<ev_periodic> watchers can also be used to implement vastly more complex
2059     timers, such as triggering an event on each "midnight, local time", or
2060     other complicated rules. This cannot be done with C<ev_timer> watchers, as
2061     those cannot react to time jumps.
2062 root 1.1
2063 root 1.161 As with timers, the callback is guaranteed to be invoked only when the
2064 root 1.230 point in time where it is supposed to trigger has passed. If multiple
2065     timers become ready during the same loop iteration then the ones with
2066     earlier time-out values are invoked before ones with later time-out values
2067 root 1.310 (but this is no longer true when a callback calls C<ev_run> recursively).
2068 root 1.28
2069 root 1.82 =head3 Watcher-Specific Functions and Data Members
2070    
2071 root 1.1 =over 4
2072    
2073 root 1.227 =item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
2074 root 1.1
2075 root 1.227 =item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
2076 root 1.1
2077 root 1.227 Lots of arguments, let's sort it out... There are basically three modes of
2078 root 1.183 operation, and we will explain them from simplest to most complex:
2079 root 1.1
2080     =over 4
2081    
2082 root 1.227 =item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
2083 root 1.1
2084 root 1.161 In this configuration the watcher triggers an event after the wall clock
2085 root 1.227 time C<offset> has passed. It will not repeat and will not adjust when a
2086     time jump occurs, that is, if it is to be run at January 1st 2011 then it
2087     will be stopped and invoked when the system clock reaches or surpasses
2088     this point in time.
2089 root 1.1
2090 root 1.227 =item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
2091 root 1.1
2092     In this mode the watcher will always be scheduled to time out at the next
2093 root 1.227 C<offset + N * interval> time (for some integer N, which can also be
2094     negative) and then repeat, regardless of any time jumps. The C<offset>
2095     argument is merely an offset into the C<interval> periods.
2096 root 1.1
2097 root 1.183 This can be used to create timers that do not drift with respect to the
2098 root 1.227 system clock, for example, here is an C<ev_periodic> that triggers each
2099     hour, on the hour (with respect to UTC):
2100 root 1.1
2101     ev_periodic_set (&periodic, 0., 3600., 0);
2102    
2103     This doesn't mean there will always be 3600 seconds in between triggers,
2104 root 1.161 but only that the callback will be called when the system time shows a
2105 root 1.12 full hour (UTC), or more correctly, when the system time is evenly divisible
2106 root 1.1 by 3600.
2107    
2108     Another way to think about it (for the mathematically inclined) is that
2109 root 1.10 C<ev_periodic> will try to run the callback in this mode at the next possible
2110 root 1.227 time where C<time = offset (mod interval)>, regardless of any time jumps.
2111 root 1.1
2112 root 1.227 For numerical stability it is preferable that the C<offset> value is near
2113 root 1.78 C<ev_now ()> (the current time), but there is no range requirement for
2114 root 1.157 this value, and in fact is often specified as zero.
2115 root 1.78
2116 root 1.161 Note also that there is an upper limit to how often a timer can fire (CPU
2117 root 1.158 speed for example), so if C<interval> is very small then timing stability
2118 root 1.161 will of course deteriorate. Libev itself tries to be exact to be about one
2119 root 1.158 millisecond (if the OS supports it and the machine is fast enough).
2120    
2121 root 1.227 =item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
2122 root 1.1
2123 root 1.227 In this mode the values for C<interval> and C<offset> are both being
2124 root 1.1 ignored. Instead, each time the periodic watcher gets scheduled, the
2125     reschedule callback will be called with the watcher as first, and the
2126     current time as second argument.
2127    
2128 root 1.227 NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, ever,
2129     or make ANY other event loop modifications whatsoever, unless explicitly
2130     allowed by documentation here>.
2131 root 1.1
2132 root 1.157 If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
2133     it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
2134     only event loop modification you are allowed to do).
2135    
2136 root 1.198 The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
2137 root 1.157 *w, ev_tstamp now)>, e.g.:
2138 root 1.1
2139 root 1.198 static ev_tstamp
2140     my_rescheduler (ev_periodic *w, ev_tstamp now)
2141 root 1.1 {
2142     return now + 60.;
2143     }
2144    
2145     It must return the next time to trigger, based on the passed time value
2146     (that is, the lowest time value larger than to the second argument). It
2147     will usually be called just before the callback will be triggered, but
2148     might be called at other times, too.
2149    
2150 root 1.157 NOTE: I<< This callback must always return a time that is higher than or
2151     equal to the passed C<now> value >>.
2152 root 1.18
2153 root 1.1 This can be used to create very complex timers, such as a timer that
2154 root 1.157 triggers on "next midnight, local time". To do this, you would calculate the
2155 root 1.19 next midnight after C<now> and return the timestamp value for this. How
2156     you do this is, again, up to you (but it is not trivial, which is the main
2157     reason I omitted it as an example).
2158 root 1.1
2159     =back
2160    
2161     =item ev_periodic_again (loop, ev_periodic *)
2162    
2163     Simply stops and restarts the periodic watcher again. This is only useful
2164     when you changed some parameters or the reschedule callback would return
2165     a different time than the last time it was called (e.g. in a crond like
2166     program when the crontabs have changed).
2167    
2168 root 1.149 =item ev_tstamp ev_periodic_at (ev_periodic *)
2169    
2170 root 1.227 When active, returns the absolute time that the watcher is supposed
2171     to trigger next. This is not the same as the C<offset> argument to
2172     C<ev_periodic_set>, but indeed works even in interval and manual
2173     rescheduling modes.
2174 root 1.149
2175 root 1.78 =item ev_tstamp offset [read-write]
2176    
2177     When repeating, this contains the offset value, otherwise this is the
2178 root 1.227 absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
2179     although libev might modify this value for better numerical stability).
2180 root 1.78
2181     Can be modified any time, but changes only take effect when the periodic
2182     timer fires or C<ev_periodic_again> is being called.
2183    
2184 root 1.48 =item ev_tstamp interval [read-write]
2185    
2186     The current interval value. Can be modified any time, but changes only
2187     take effect when the periodic timer fires or C<ev_periodic_again> is being
2188     called.
2189    
2190 root 1.198 =item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
2191 root 1.48
2192     The current reschedule callback, or C<0>, if this functionality is
2193     switched off. Can be changed any time, but changes only take effect when
2194     the periodic timer fires or C<ev_periodic_again> is being called.
2195    
2196 root 1.1 =back
2197    
2198 root 1.111 =head3 Examples
2199    
2200 root 1.54 Example: Call a callback every hour, or, more precisely, whenever the
2201 root 1.183 system time is divisible by 3600. The callback invocation times have
2202 root 1.161 potentially a lot of jitter, but good long-term stability.
2203 root 1.34
2204 root 1.164 static void
2205 root 1.301 clock_cb (struct ev_loop *loop, ev_periodic *w, int revents)
2206 root 1.164 {
2207     ... its now a full hour (UTC, or TAI or whatever your clock follows)
2208     }
2209    
2210 root 1.198 ev_periodic hourly_tick;
2211 root 1.164 ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
2212     ev_periodic_start (loop, &hourly_tick);
2213 root 1.34
2214 root 1.54 Example: The same as above, but use a reschedule callback to do it:
2215 root 1.34
2216 root 1.164 #include <math.h>
2217 root 1.34
2218 root 1.164 static ev_tstamp
2219 root 1.198 my_scheduler_cb (ev_periodic *w, ev_tstamp now)
2220 root 1.164 {
2221 root 1.183 return now + (3600. - fmod (now, 3600.));
2222 root 1.164 }
2223 root 1.34
2224 root 1.164 ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
2225 root 1.34
2226 root 1.54 Example: Call a callback every hour, starting now:
2227 root 1.34
2228 root 1.198 ev_periodic hourly_tick;
2229 root 1.164 ev_periodic_init (&hourly_tick, clock_cb,
2230     fmod (ev_now (loop), 3600.), 3600., 0);
2231     ev_periodic_start (loop, &hourly_tick);
2232 root 1.34
2233    
2234 root 1.42 =head2 C<ev_signal> - signal me when a signal gets signalled!
2235 root 1.1
2236     Signal watchers will trigger an event when the process receives a specific
2237     signal one or more times. Even though signals are very asynchronous, libev
2238 root 1.9 will try it's best to deliver signals synchronously, i.e. as part of the
2239 root 1.1 normal event processing, like any other event.
2240    
2241 root 1.260 If you want signals to be delivered truly asynchronously, just use
2242     C<sigaction> as you would do without libev and forget about sharing
2243     the signal. You can even use C<ev_async> from a signal handler to
2244     synchronously wake up an event loop.
2245    
2246     You can configure as many watchers as you like for the same signal, but
2247     only within the same loop, i.e. you can watch for C<SIGINT> in your
2248     default loop and for C<SIGIO> in another loop, but you cannot watch for
2249     C<SIGINT> in both the default loop and another loop at the same time. At
2250     the moment, C<SIGCHLD> is permanently tied to the default loop.
2251    
2252     When the first watcher gets started will libev actually register something
2253     with the kernel (thus it coexists with your own signal handlers as long as
2254     you don't register any with libev for the same signal).
2255 root 1.259
2256 root 1.135 If possible and supported, libev will install its handlers with
2257 root 1.259 C<SA_RESTART> (or equivalent) behaviour enabled, so system calls should
2258     not be unduly interrupted. If you have a problem with system calls getting
2259     interrupted by signals you can block all signals in an C<ev_check> watcher
2260     and unblock them in an C<ev_prepare> watcher.
2261 root 1.135
2262 root 1.277 =head3 The special problem of inheritance over fork/execve/pthread_create
2263 root 1.265
2264     Both the signal mask (C<sigprocmask>) and the signal disposition
2265     (C<sigaction>) are unspecified after starting a signal watcher (and after
2266     stopping it again), that is, libev might or might not block the signal,
2267     and might or might not set or restore the installed signal handler.
2268    
2269     While this does not matter for the signal disposition (libev never
2270     sets signals to C<SIG_IGN>, so handlers will be reset to C<SIG_DFL> on
2271     C<execve>), this matters for the signal mask: many programs do not expect
2272 root 1.266 certain signals to be blocked.
2273 root 1.265
2274     This means that before calling C<exec> (from the child) you should reset
2275     the signal mask to whatever "default" you expect (all clear is a good
2276     choice usually).
2277    
2278 root 1.267 The simplest way to ensure that the signal mask is reset in the child is
2279     to install a fork handler with C<pthread_atfork> that resets it. That will
2280     catch fork calls done by libraries (such as the libc) as well.
2281    
2282 root 1.277 In current versions of libev, the signal will not be blocked indefinitely
2283     unless you use the C<signalfd> API (C<EV_SIGNALFD>). While this reduces
2284     the window of opportunity for problems, it will not go away, as libev
2285     I<has> to modify the signal mask, at least temporarily.
2286    
2287 root 1.278 So I can't stress this enough: I<If you do not reset your signal mask when
2288     you expect it to be empty, you have a race condition in your code>. This
2289     is not a libev-specific thing, this is true for most event libraries.
2290 root 1.266
2291 root 1.82 =head3 Watcher-Specific Functions and Data Members
2292    
2293 root 1.1 =over 4
2294    
2295     =item ev_signal_init (ev_signal *, callback, int signum)
2296    
2297     =item ev_signal_set (ev_signal *, int signum)
2298    
2299     Configures the watcher to trigger on the given signal number (usually one
2300     of the C<SIGxxx> constants).
2301    
2302 root 1.48 =item int signum [read-only]
2303    
2304     The signal the watcher watches out for.
2305    
2306 root 1.1 =back
2307    
2308 root 1.132 =head3 Examples
2309    
2310 root 1.188 Example: Try to exit cleanly on SIGINT.
2311 root 1.132
2312 root 1.164 static void
2313 root 1.198 sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
2314 root 1.164 {
2315 root 1.310 ev_break (loop, EVBREAK_ALL);
2316 root 1.164 }
2317    
2318 root 1.198 ev_signal signal_watcher;
2319 root 1.164 ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
2320 root 1.188 ev_signal_start (loop, &signal_watcher);
2321 root 1.132
2322 root 1.35
2323 root 1.42 =head2 C<ev_child> - watch out for process status changes
2324 root 1.1
2325     Child watchers trigger when your process receives a SIGCHLD in response to
2326 root 1.183 some child status changes (most typically when a child of yours dies or
2327     exits). It is permissible to install a child watcher I<after> the child
2328     has been forked (which implies it might have already exited), as long
2329     as the event loop isn't entered (or is continued from a watcher), i.e.,
2330     forking and then immediately registering a watcher for the child is fine,
2331 root 1.244 but forking and registering a watcher a few event loop iterations later or
2332     in the next callback invocation is not.
2333 root 1.134
2334     Only the default event loop is capable of handling signals, and therefore
2335 root 1.161 you can only register child watchers in the default event loop.
2336 root 1.134
2337 root 1.248 Due to some design glitches inside libev, child watchers will always be
2338 root 1.249 handled at maximum priority (their priority is set to C<EV_MAXPRI> by
2339     libev)
2340 root 1.248
2341 root 1.134 =head3 Process Interaction
2342    
2343     Libev grabs C<SIGCHLD> as soon as the default event loop is
2344 root 1.259 initialised. This is necessary to guarantee proper behaviour even if the
2345     first child watcher is started after the child exits. The occurrence
2346 root 1.134 of C<SIGCHLD> is recorded asynchronously, but child reaping is done
2347     synchronously as part of the event loop processing. Libev always reaps all
2348     children, even ones not watched.
2349    
2350     =head3 Overriding the Built-In Processing
2351    
2352     Libev offers no special support for overriding the built-in child
2353     processing, but if your application collides with libev's default child
2354     handler, you can override it easily by installing your own handler for
2355     C<SIGCHLD> after initialising the default loop, and making sure the
2356     default loop never gets destroyed. You are encouraged, however, to use an
2357     event-based approach to child reaping and thus use libev's support for
2358     that, so other libev users can use C<ev_child> watchers freely.
2359 root 1.1
2360 root 1.173 =head3 Stopping the Child Watcher
2361    
2362     Currently, the child watcher never gets stopped, even when the
2363     child terminates, so normally one needs to stop the watcher in the
2364     callback. Future versions of libev might stop the watcher automatically
2365 root 1.259 when a child exit is detected (calling C<ev_child_stop> twice is not a
2366     problem).
2367 root 1.173
2368 root 1.82 =head3 Watcher-Specific Functions and Data Members
2369    
2370 root 1.1 =over 4
2371    
2372 root 1.120 =item ev_child_init (ev_child *, callback, int pid, int trace)
2373 root 1.1
2374 root 1.120 =item ev_child_set (ev_child *, int pid, int trace)
2375 root 1.1
2376     Configures the watcher to wait for status changes of process C<pid> (or
2377     I<any> process if C<pid> is specified as C<0>). The callback can look
2378     at the C<rstatus> member of the C<ev_child> watcher structure to see
2379 root 1.14 the status word (use the macros from C<sys/wait.h> and see your systems
2380     C<waitpid> documentation). The C<rpid> member contains the pid of the
2381 root 1.120 process causing the status change. C<trace> must be either C<0> (only
2382     activate the watcher when the process terminates) or C<1> (additionally
2383     activate the watcher when the process is stopped or continued).
2384 root 1.1
2385 root 1.48 =item int pid [read-only]
2386    
2387     The process id this watcher watches out for, or C<0>, meaning any process id.
2388    
2389     =item int rpid [read-write]
2390    
2391     The process id that detected a status change.
2392    
2393     =item int rstatus [read-write]
2394    
2395     The process exit/trace status caused by C<rpid> (see your systems
2396     C<waitpid> and C<sys/wait.h> documentation for details).
2397    
2398 root 1.1 =back
2399    
2400 root 1.134 =head3 Examples
2401    
2402     Example: C<fork()> a new process and install a child handler to wait for
2403     its completion.
2404    
2405 root 1.164 ev_child cw;
2406    
2407     static void
2408 root 1.198 child_cb (EV_P_ ev_child *w, int revents)
2409 root 1.164 {
2410     ev_child_stop (EV_A_ w);
2411     printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
2412     }
2413    
2414     pid_t pid = fork ();
2415 root 1.134
2416 root 1.164 if (pid < 0)
2417     // error
2418     else if (pid == 0)
2419     {
2420     // the forked child executes here
2421     exit (1);
2422     }
2423     else
2424     {
2425     ev_child_init (&cw, child_cb, pid, 0);
2426     ev_child_start (EV_DEFAULT_ &cw);
2427     }
2428 root 1.134
2429 root 1.34
2430 root 1.48 =head2 C<ev_stat> - did the file attributes just change?
2431    
2432 root 1.161 This watches a file system path for attribute changes. That is, it calls
2433 root 1.207 C<stat> on that path in regular intervals (or when the OS says it changed)
2434     and sees if it changed compared to the last time, invoking the callback if
2435     it did.
2436 root 1.48
2437     The path does not need to exist: changing from "path exists" to "path does
2438 root 1.211 not exist" is a status change like any other. The condition "path does not
2439     exist" (or more correctly "path cannot be stat'ed") is signified by the
2440     C<st_nlink> field being zero (which is otherwise always forced to be at
2441     least one) and all the other fields of the stat buffer having unspecified
2442     contents.
2443 root 1.48
2444 root 1.207 The path I<must not> end in a slash or contain special components such as
2445     C<.> or C<..>. The path I<should> be absolute: If it is relative and
2446     your working directory changes, then the behaviour is undefined.
2447    
2448     Since there is no portable change notification interface available, the
2449     portable implementation simply calls C<stat(2)> regularly on the path
2450     to see if it changed somehow. You can specify a recommended polling
2451     interval for this case. If you specify a polling interval of C<0> (highly
2452     recommended!) then a I<suitable, unspecified default> value will be used
2453     (which you can expect to be around five seconds, although this might
2454     change dynamically). Libev will also impose a minimum interval which is
2455 root 1.208 currently around C<0.1>, but that's usually overkill.
2456 root 1.48
2457     This watcher type is not meant for massive numbers of stat watchers,
2458     as even with OS-supported change notifications, this can be
2459     resource-intensive.
2460    
2461 root 1.183 At the time of this writing, the only OS-specific interface implemented
2462 root 1.211 is the Linux inotify interface (implementing kqueue support is left as an
2463     exercise for the reader. Note, however, that the author sees no way of
2464     implementing C<ev_stat> semantics with kqueue, except as a hint).
2465 root 1.48
2466 root 1.137 =head3 ABI Issues (Largefile Support)
2467    
2468     Libev by default (unless the user overrides this) uses the default
2469 root 1.169 compilation environment, which means that on systems with large file
2470     support disabled by default, you get the 32 bit version of the stat
2471 root 1.137 structure. When using the library from programs that change the ABI to
2472     use 64 bit file offsets the programs will fail. In that case you have to
2473     compile libev with the same flags to get binary compatibility. This is
2474     obviously the case with any flags that change the ABI, but the problem is
2475 root 1.207 most noticeably displayed with ev_stat and large file support.
2476 root 1.169
2477     The solution for this is to lobby your distribution maker to make large
2478     file interfaces available by default (as e.g. FreeBSD does) and not
2479     optional. Libev cannot simply switch on large file support because it has
2480     to exchange stat structures with application programs compiled using the
2481     default compilation environment.
2482 root 1.137
2483 root 1.183 =head3 Inotify and Kqueue
2484 root 1.108
2485 root 1.211 When C<inotify (7)> support has been compiled into libev and present at
2486     runtime, it will be used to speed up change detection where possible. The
2487     inotify descriptor will be created lazily when the first C<ev_stat>
2488     watcher is being started.
2489 root 1.108
2490 root 1.147 Inotify presence does not change the semantics of C<ev_stat> watchers
2491 root 1.108 except that changes might be detected earlier, and in some cases, to avoid
2492 root 1.147 making regular C<stat> calls. Even in the presence of inotify support
2493 root 1.183 there are many cases where libev has to resort to regular C<stat> polling,
2494 root 1.211 but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
2495     many bugs), the path exists (i.e. stat succeeds), and the path resides on
2496     a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
2497     xfs are fully working) libev usually gets away without polling.
2498 root 1.108
2499 root 1.183 There is no support for kqueue, as apparently it cannot be used to
2500 root 1.108 implement this functionality, due to the requirement of having a file
2501 root 1.183 descriptor open on the object at all times, and detecting renames, unlinks
2502     etc. is difficult.
2503 root 1.108
2504 root 1.212 =head3 C<stat ()> is a synchronous operation
2505    
2506     Libev doesn't normally do any kind of I/O itself, and so is not blocking
2507     the process. The exception are C<ev_stat> watchers - those call C<stat
2508     ()>, which is a synchronous operation.
2509    
2510     For local paths, this usually doesn't matter: unless the system is very
2511     busy or the intervals between stat's are large, a stat call will be fast,
2512 root 1.222 as the path data is usually in memory already (except when starting the
2513 root 1.212 watcher).
2514    
2515     For networked file systems, calling C<stat ()> can block an indefinite
2516     time due to network issues, and even under good conditions, a stat call
2517     often takes multiple milliseconds.
2518    
2519     Therefore, it is best to avoid using C<ev_stat> watchers on networked
2520     paths, although this is fully supported by libev.
2521    
2522 root 1.107 =head3 The special problem of stat time resolution
2523    
2524 root 1.207 The C<stat ()> system call only supports full-second resolution portably,
2525     and even on systems where the resolution is higher, most file systems
2526     still only support whole seconds.
2527 root 1.107
2528 root 1.150 That means that, if the time is the only thing that changes, you can
2529     easily miss updates: on the first update, C<ev_stat> detects a change and
2530     calls your callback, which does something. When there is another update
2531 root 1.183 within the same second, C<ev_stat> will be unable to detect unless the
2532     stat data does change in other ways (e.g. file size).
2533 root 1.150
2534     The solution to this is to delay acting on a change for slightly more
2535 root 1.155 than a second (or till slightly after the next full second boundary), using
2536 root 1.150 a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
2537     ev_timer_again (loop, w)>).
2538    
2539     The C<.02> offset is added to work around small timing inconsistencies
2540     of some operating systems (where the second counter of the current time
2541     might be be delayed. One such system is the Linux kernel, where a call to
2542     C<gettimeofday> might return a timestamp with a full second later than
2543     a subsequent C<time> call - if the equivalent of C<time ()> is used to
2544     update file times then there will be a small window where the kernel uses
2545     the previous second to update file times but libev might already execute
2546     the timer callback).
2547 root 1.107
2548 root 1.82 =head3 Watcher-Specific Functions and Data Members
2549    
2550 root 1.48 =over 4
2551    
2552     =item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
2553    
2554     =item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
2555    
2556     Configures the watcher to wait for status changes of the given
2557     C<path>. The C<interval> is a hint on how quickly a change is expected to
2558     be detected and should normally be specified as C<0> to let libev choose
2559     a suitable value. The memory pointed to by C<path> must point to the same
2560     path for as long as the watcher is active.
2561    
2562 root 1.183 The callback will receive an C<EV_STAT> event when a change was detected,
2563     relative to the attributes at the time the watcher was started (or the
2564     last change was detected).
2565 root 1.48
2566 root 1.132 =item ev_stat_stat (loop, ev_stat *)
2567 root 1.48
2568     Updates the stat buffer immediately with new values. If you change the
2569 root 1.150 watched path in your callback, you could call this function to avoid
2570     detecting this change (while introducing a race condition if you are not
2571     the only one changing the path). Can also be useful simply to find out the
2572     new values.
2573 root 1.48
2574     =item ev_statdata attr [read-only]
2575    
2576 root 1.150 The most-recently detected attributes of the file. Although the type is
2577 root 1.48 C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
2578 root 1.150 suitable for your system, but you can only rely on the POSIX-standardised
2579     members to be present. If the C<st_nlink> member is C<0>, then there was
2580     some error while C<stat>ing the file.
2581 root 1.48
2582     =item ev_statdata prev [read-only]
2583    
2584     The previous attributes of the file. The callback gets invoked whenever
2585 root 1.150 C<prev> != C<attr>, or, more precisely, one or more of these members
2586     differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
2587     C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
2588 root 1.48
2589     =item ev_tstamp interval [read-only]
2590    
2591     The specified interval.
2592    
2593     =item const char *path [read-only]
2594    
2595 root 1.161 The file system path that is being watched.
2596 root 1.48
2597     =back
2598    
2599 root 1.108 =head3 Examples
2600    
2601 root 1.48 Example: Watch C</etc/passwd> for attribute changes.
2602    
2603 root 1.164 static void
2604     passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
2605     {
2606     /* /etc/passwd changed in some way */
2607     if (w->attr.st_nlink)
2608     {
2609     printf ("passwd current size %ld\n", (long)w->attr.st_size);
2610     printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
2611     printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
2612     }
2613     else
2614     /* you shalt not abuse printf for puts */
2615     puts ("wow, /etc/passwd is not there, expect problems. "
2616     "if this is windows, they already arrived\n");
2617     }
2618 root 1.48
2619 root 1.164 ...
2620     ev_stat passwd;
2621 root 1.48
2622 root 1.164 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
2623     ev_stat_start (loop, &passwd);
2624 root 1.107
2625     Example: Like above, but additionally use a one-second delay so we do not
2626     miss updates (however, frequent updates will delay processing, too, so
2627     one might do the work both on C<ev_stat> callback invocation I<and> on
2628     C<ev_timer> callback invocation).
2629    
2630 root 1.164 static ev_stat passwd;
2631     static ev_timer timer;
2632 root 1.107
2633 root 1.164 static void
2634     timer_cb (EV_P_ ev_timer *w, int revents)
2635     {
2636     ev_timer_stop (EV_A_ w);
2637    
2638     /* now it's one second after the most recent passwd change */
2639     }
2640    
2641     static void
2642     stat_cb (EV_P_ ev_stat *w, int revents)
2643     {
2644     /* reset the one-second timer */
2645     ev_timer_again (EV_A_ &timer);
2646     }
2647    
2648     ...
2649     ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
2650     ev_stat_start (loop, &passwd);
2651     ev_timer_init (&timer, timer_cb, 0., 1.02);
2652 root 1.48
2653    
2654 root 1.42 =head2 C<ev_idle> - when you've got nothing better to do...
2655 root 1.1
2656 root 1.67 Idle watchers trigger events when no other events of the same or higher
2657 root 1.183 priority are pending (prepare, check and other idle watchers do not count
2658     as receiving "events").
2659 root 1.67
2660     That is, as long as your process is busy handling sockets or timeouts
2661     (or even signals, imagine) of the same or higher priority it will not be
2662     triggered. But when your process is idle (or only lower-priority watchers
2663     are pending), the idle watchers are being called once per event loop
2664     iteration - until stopped, that is, or your process receives more events
2665     and becomes busy again with higher priority stuff.
2666 root 1.1
2667     The most noteworthy effect is that as long as any idle watchers are
2668     active, the process will not block when waiting for new events.
2669    
2670     Apart from keeping your process non-blocking (which is a useful
2671     effect on its own sometimes), idle watchers are a good place to do
2672     "pseudo-background processing", or delay processing stuff to after the
2673     event loop has handled all outstanding events.
2674    
2675 root 1.82 =head3 Watcher-Specific Functions and Data Members
2676    
2677 root 1.1 =over 4
2678    
2679 root 1.226 =item ev_idle_init (ev_idle *, callback)
2680 root 1.1
2681     Initialises and configures the idle watcher - it has no parameters of any
2682     kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
2683     believe me.
2684    
2685     =back
2686    
2687 root 1.111 =head3 Examples
2688    
2689 root 1.54 Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
2690     callback, free it. Also, use no error checking, as usual.
2691 root 1.34
2692 root 1.164 static void
2693 root 1.198 idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
2694 root 1.164 {
2695     free (w);
2696     // now do something you wanted to do when the program has
2697     // no longer anything immediate to do.
2698     }
2699    
2700 root 1.198 ev_idle *idle_watcher = malloc (sizeof (ev_idle));
2701 root 1.164 ev_idle_init (idle_watcher, idle_cb);
2702 root 1.242 ev_idle_start (loop, idle_watcher);
2703 root 1.34
2704    
2705 root 1.42 =head2 C<ev_prepare> and C<ev_check> - customise your event loop!
2706 root 1.1
2707 root 1.183 Prepare and check watchers are usually (but not always) used in pairs:
2708 root 1.20 prepare watchers get invoked before the process blocks and check watchers
2709 root 1.14 afterwards.
2710 root 1.1
2711 root 1.310 You I<must not> call C<ev_run> or similar functions that enter
2712 root 1.45 the current event loop from either C<ev_prepare> or C<ev_check>
2713     watchers. Other loops than the current one are fine, however. The
2714     rationale behind this is that you do not need to check for recursion in
2715     those watchers, i.e. the sequence will always be C<ev_prepare>, blocking,
2716     C<ev_check> so if you have one watcher of each kind they will always be
2717     called in pairs bracketing the blocking call.
2718    
2719 root 1.35 Their main purpose is to integrate other event mechanisms into libev and
2720 root 1.183 their use is somewhat advanced. They could be used, for example, to track
2721 root 1.35 variable changes, implement your own watchers, integrate net-snmp or a
2722 root 1.45 coroutine library and lots more. They are also occasionally useful if
2723     you cache some data and want to flush it before blocking (for example,
2724     in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
2725     watcher).
2726 root 1.1
2727 root 1.183 This is done by examining in each prepare call which file descriptors
2728     need to be watched by the other library, registering C<ev_io> watchers
2729     for them and starting an C<ev_timer> watcher for any timeouts (many
2730     libraries provide exactly this functionality). Then, in the check watcher,
2731     you check for any events that occurred (by checking the pending status
2732     of all watchers and stopping them) and call back into the library. The
2733     I/O and timer callbacks will never actually be called (but must be valid
2734     nevertheless, because you never know, you know?).
2735 root 1.1
2736 root 1.14 As another example, the Perl Coro module uses these hooks to integrate
2737 root 1.1 coroutines into libev programs, by yielding to other active coroutines
2738     during each prepare and only letting the process block if no coroutines
2739 root 1.20 are ready to run (it's actually more complicated: it only runs coroutines
2740     with priority higher than or equal to the event loop and one coroutine
2741     of lower priority, but only once, using idle watchers to keep the event
2742     loop from blocking if lower-priority coroutines are active, thus mapping
2743     low-priority coroutines to idle/background tasks).
2744 root 1.1
2745 root 1.77 It is recommended to give C<ev_check> watchers highest (C<EV_MAXPRI>)
2746     priority, to ensure that they are being run before any other watchers
2747 root 1.183 after the poll (this doesn't matter for C<ev_prepare> watchers).
2748    
2749     Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
2750     activate ("feed") events into libev. While libev fully supports this, they
2751     might get executed before other C<ev_check> watchers did their job. As
2752     C<ev_check> watchers are often used to embed other (non-libev) event
2753     loops those other event loops might be in an unusable state until their
2754     C<ev_check> watcher ran (always remind yourself to coexist peacefully with
2755     others).
2756 root 1.77
2757 root 1.82 =head3 Watcher-Specific Functions and Data Members
2758    
2759 root 1.1 =over 4
2760    
2761     =item ev_prepare_init (ev_prepare *, callback)
2762    
2763     =item ev_check_init (ev_check *, callback)
2764    
2765     Initialises and configures the prepare or check watcher - they have no
2766     parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
2767 root 1.183 macros, but using them is utterly, utterly, utterly and completely
2768     pointless.
2769 root 1.1
2770     =back
2771    
2772 root 1.111 =head3 Examples
2773    
2774 root 1.76 There are a number of principal ways to embed other event loops or modules
2775     into libev. Here are some ideas on how to include libadns into libev
2776     (there is a Perl module named C<EV::ADNS> that does this, which you could
2777 root 1.150 use as a working example. Another Perl module named C<EV::Glib> embeds a
2778     Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
2779     Glib event loop).
2780 root 1.76
2781     Method 1: Add IO watchers and a timeout watcher in a prepare handler,
2782     and in a check watcher, destroy them and call into libadns. What follows
2783     is pseudo-code only of course. This requires you to either use a low
2784     priority for the check watcher or use C<ev_clear_pending> explicitly, as
2785     the callbacks for the IO/timeout watchers might not have been called yet.
2786 root 1.45
2787 root 1.164 static ev_io iow [nfd];
2788     static ev_timer tw;
2789 root 1.45
2790 root 1.164 static void
2791 root 1.198 io_cb (struct ev_loop *loop, ev_io *w, int revents)
2792 root 1.164 {
2793     }
2794 root 1.45
2795 root 1.164 // create io watchers for each fd and a timer before blocking
2796     static void
2797 root 1.198 adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
2798 root 1.164 {
2799     int timeout = 3600000;
2800     struct pollfd fds [nfd];
2801     // actual code will need to loop here and realloc etc.
2802     adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
2803    
2804     /* the callback is illegal, but won't be called as we stop during check */
2805 root 1.243 ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
2806 root 1.164 ev_timer_start (loop, &tw);
2807    
2808     // create one ev_io per pollfd
2809     for (int i = 0; i < nfd; ++i)
2810     {
2811     ev_io_init (iow + i, io_cb, fds [i].fd,
2812     ((fds [i].events & POLLIN ? EV_READ : 0)
2813     | (fds [i].events & POLLOUT ? EV_WRITE : 0)));
2814    
2815     fds [i].revents = 0;
2816     ev_io_start (loop, iow + i);
2817     }
2818     }
2819    
2820     // stop all watchers after blocking
2821     static void
2822 root 1.198 adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
2823 root 1.164 {
2824     ev_timer_stop (loop, &tw);
2825    
2826     for (int i = 0; i < nfd; ++i)
2827     {
2828     // set the relevant poll flags
2829     // could also call adns_processreadable etc. here
2830     struct pollfd *fd = fds + i;
2831     int revents = ev_clear_pending (iow + i);
2832     if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
2833     if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
2834    
2835     // now stop the watcher
2836     ev_io_stop (loop, iow + i);
2837     }
2838    
2839     adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
2840     }
2841 root 1.34
2842 root 1.76 Method 2: This would be just like method 1, but you run C<adns_afterpoll>
2843     in the prepare watcher and would dispose of the check watcher.
2844    
2845     Method 3: If the module to be embedded supports explicit event
2846 root 1.161 notification (libadns does), you can also make use of the actual watcher
2847 root 1.76 callbacks, and only destroy/create the watchers in the prepare watcher.
2848    
2849 root 1.164 static void
2850     timer_cb (EV_P_ ev_timer *w, int revents)
2851     {
2852     adns_state ads = (adns_state)w->data;
2853     update_now (EV_A);
2854    
2855     adns_processtimeouts (ads, &tv_now);
2856     }
2857    
2858     static void
2859     io_cb (EV_P_ ev_io *w, int revents)
2860     {
2861     adns_state ads = (adns_state)w->data;
2862     update_now (EV_A);
2863    
2864     if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
2865     if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
2866     }
2867 root 1.76
2868 root 1.164 // do not ever call adns_afterpoll
2869 root 1.76
2870     Method 4: Do not use a prepare or check watcher because the module you
2871 root 1.183 want to embed is not flexible enough to support it. Instead, you can
2872     override their poll function. The drawback with this solution is that the
2873     main loop is now no longer controllable by EV. The C<Glib::EV> module uses
2874     this approach, effectively embedding EV as a client into the horrible
2875     libglib event loop.
2876 root 1.76
2877 root 1.164 static gint
2878     event_poll_func (GPollFD *fds, guint nfds, gint timeout)
2879     {
2880     int got_events = 0;
2881    
2882     for (n = 0; n < nfds; ++n)
2883     // create/start io watcher that sets the relevant bits in fds[n] and increment got_events
2884    
2885     if (timeout >= 0)
2886     // create/start timer
2887    
2888     // poll
2889 root 1.310 ev_run (EV_A_ 0);
2890 root 1.76
2891 root 1.164 // stop timer again
2892     if (timeout >= 0)
2893     ev_timer_stop (EV_A_ &to);
2894    
2895     // stop io watchers again - their callbacks should have set
2896     for (n = 0; n < nfds; ++n)
2897     ev_io_stop (EV_A_ iow [n]);
2898    
2899     return got_events;
2900     }
2901 root 1.76
2902 root 1.34
2903 root 1.42 =head2 C<ev_embed> - when one backend isn't enough...
2904 root 1.35
2905     This is a rather advanced watcher type that lets you embed one event loop
2906 root 1.36 into another (currently only C<ev_io> events are supported in the embedded
2907     loop, other types of watchers might be handled in a delayed or incorrect
2908 root 1.100 fashion and must not be used).
2909 root 1.35
2910     There are primarily two reasons you would want that: work around bugs and
2911     prioritise I/O.
2912    
2913     As an example for a bug workaround, the kqueue backend might only support
2914     sockets on some platform, so it is unusable as generic backend, but you
2915     still want to make use of it because you have many sockets and it scales
2916 root 1.183 so nicely. In this case, you would create a kqueue-based loop and embed
2917     it into your default loop (which might use e.g. poll). Overall operation
2918     will be a bit slower because first libev has to call C<poll> and then
2919     C<kevent>, but at least you can use both mechanisms for what they are
2920     best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
2921    
2922     As for prioritising I/O: under rare circumstances you have the case where
2923     some fds have to be watched and handled very quickly (with low latency),
2924     and even priorities and idle watchers might have too much overhead. In
2925     this case you would put all the high priority stuff in one loop and all
2926     the rest in a second one, and embed the second one in the first.
2927 root 1.35
2928 root 1.223 As long as the watcher is active, the callback will be invoked every
2929     time there might be events pending in the embedded loop. The callback
2930     must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
2931     sweep and invoke their callbacks (the callback doesn't need to invoke the
2932     C<ev_embed_sweep> function directly, it could also start an idle watcher
2933     to give the embedded loop strictly lower priority for example).
2934    
2935     You can also set the callback to C<0>, in which case the embed watcher
2936     will automatically execute the embedded loop sweep whenever necessary.
2937    
2938     Fork detection will be handled transparently while the C<ev_embed> watcher
2939     is active, i.e., the embedded loop will automatically be forked when the
2940     embedding loop forks. In other cases, the user is responsible for calling
2941     C<ev_loop_fork> on the embedded loop.
2942 root 1.35
2943 root 1.184 Unfortunately, not all backends are embeddable: only the ones returned by
2944 root 1.35 C<ev_embeddable_backends> are, which, unfortunately, does not include any
2945     portable one.
2946    
2947     So when you want to use this feature you will always have to be prepared
2948     that you cannot get an embeddable loop. The recommended way to get around
2949     this is to have a separate variables for your embeddable loop, try to
2950 root 1.111 create it, and if that fails, use the normal loop for everything.
2951 root 1.35
2952 root 1.187 =head3 C<ev_embed> and fork
2953    
2954     While the C<ev_embed> watcher is running, forks in the embedding loop will
2955     automatically be applied to the embedded loop as well, so no special
2956     fork handling is required in that case. When the watcher is not running,
2957     however, it is still the task of the libev user to call C<ev_loop_fork ()>
2958     as applicable.
2959    
2960 root 1.82 =head3 Watcher-Specific Functions and Data Members
2961    
2962 root 1.35 =over 4
2963    
2964 root 1.36 =item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
2965    
2966     =item ev_embed_set (ev_embed *, callback, struct ev_loop *embedded_loop)
2967    
2968     Configures the watcher to embed the given loop, which must be
2969     embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
2970     invoked automatically, otherwise it is the responsibility of the callback
2971     to invoke it (it will continue to be called until the sweep has been done,
2972 root 1.161 if you do not want that, you need to temporarily stop the embed watcher).
2973 root 1.35
2974 root 1.36 =item ev_embed_sweep (loop, ev_embed *)
2975 root 1.35
2976 root 1.36 Make a single, non-blocking sweep over the embedded loop. This works
2977 root 1.310 similarly to C<ev_run (embedded_loop, EVRUN_NOWAIT)>, but in the most
2978 root 1.161 appropriate way for embedded loops.
2979 root 1.35
2980 root 1.91 =item struct ev_loop *other [read-only]
2981 root 1.48
2982     The embedded event loop.
2983    
2984 root 1.35 =back
2985    
2986 root 1.111 =head3 Examples
2987    
2988     Example: Try to get an embeddable event loop and embed it into the default
2989     event loop. If that is not possible, use the default loop. The default
2990 root 1.161 loop is stored in C<loop_hi>, while the embeddable loop is stored in
2991     C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
2992 root 1.111 used).
2993    
2994 root 1.164 struct ev_loop *loop_hi = ev_default_init (0);
2995     struct ev_loop *loop_lo = 0;
2996 root 1.198 ev_embed embed;
2997 root 1.164
2998     // see if there is a chance of getting one that works
2999     // (remember that a flags value of 0 means autodetection)
3000     loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
3001     ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
3002     : 0;
3003    
3004     // if we got one, then embed it, otherwise default to loop_hi
3005     if (loop_lo)
3006     {
3007     ev_embed_init (&embed, 0, loop_lo);
3008     ev_embed_start (loop_hi, &embed);
3009     }
3010     else
3011     loop_lo = loop_hi;
3012 root 1.111
3013     Example: Check if kqueue is available but not recommended and create
3014     a kqueue backend for use with sockets (which usually work with any
3015     kqueue implementation). Store the kqueue/socket-only event loop in
3016     C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
3017    
3018 root 1.164 struct ev_loop *loop = ev_default_init (0);
3019     struct ev_loop *loop_socket = 0;
3020 root 1.198 ev_embed embed;
3021 root 1.164
3022     if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
3023     if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
3024     {
3025     ev_embed_init (&embed, 0, loop_socket);
3026     ev_embed_start (loop, &embed);
3027     }
3028 root 1.111
3029 root 1.164 if (!loop_socket)
3030     loop_socket = loop;
3031 root 1.111
3032 root 1.164 // now use loop_socket for all sockets, and loop for everything else
3033 root 1.111
3034 root 1.35
3035 root 1.50 =head2 C<ev_fork> - the audacity to resume the event loop after a fork
3036    
3037     Fork watchers are called when a C<fork ()> was detected (usually because
3038     whoever is a good citizen cared to tell libev about it by calling
3039     C<ev_default_fork> or C<ev_loop_fork>). The invocation is done before the
3040     event loop blocks next and before C<ev_check> watchers are being called,
3041     and only in the child after the fork. If whoever good citizen calling
3042     C<ev_default_fork> cheats and calls it in the wrong process, the fork
3043     handlers will be invoked, too, of course.
3044    
3045 root 1.238 =head3 The special problem of life after fork - how is it possible?
3046    
3047 root 1.302 Most uses of C<fork()> consist of forking, then some simple calls to set
3048 root 1.238 up/change the process environment, followed by a call to C<exec()>. This
3049     sequence should be handled by libev without any problems.
3050    
3051     This changes when the application actually wants to do event handling
3052     in the child, or both parent in child, in effect "continuing" after the
3053     fork.
3054    
3055     The default mode of operation (for libev, with application help to detect
3056     forks) is to duplicate all the state in the child, as would be expected
3057     when I<either> the parent I<or> the child process continues.
3058    
3059     When both processes want to continue using libev, then this is usually the
3060     wrong result. In that case, usually one process (typically the parent) is
3061     supposed to continue with all watchers in place as before, while the other
3062     process typically wants to start fresh, i.e. without any active watchers.
3063    
3064     The cleanest and most efficient way to achieve that with libev is to
3065     simply create a new event loop, which of course will be "empty", and
3066     use that for new watchers. This has the advantage of not touching more
3067     memory than necessary, and thus avoiding the copy-on-write, and the
3068     disadvantage of having to use multiple event loops (which do not support
3069     signal watchers).
3070    
3071     When this is not possible, or you want to use the default loop for
3072     other reasons, then in the process that wants to start "fresh", call
3073     C<ev_default_destroy ()> followed by C<ev_default_loop (...)>. Destroying
3074     the default loop will "orphan" (not stop) all registered watchers, so you
3075     have to be careful not to execute code that modifies those watchers. Note
3076     also that in that case, you have to re-register any signal watchers.
3077    
3078 root 1.83 =head3 Watcher-Specific Functions and Data Members
3079    
3080 root 1.50 =over 4
3081    
3082     =item ev_fork_init (ev_signal *, callback)
3083    
3084     Initialises and configures the fork watcher - it has no parameters of any
3085     kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
3086     believe me.
3087    
3088     =back
3089    
3090    
3091 root 1.302 =head2 C<ev_async> - how to wake up an event loop
3092 root 1.122
3093 root 1.310 In general, you cannot use an C<ev_run> from multiple threads or other
3094 root 1.122 asynchronous sources such as signal handlers (as opposed to multiple event
3095     loops - those are of course safe to use in different threads).
3096    
3097 root 1.302 Sometimes, however, you need to wake up an event loop you do not control,
3098     for example because it belongs to another thread. This is what C<ev_async>
3099     watchers do: as long as the C<ev_async> watcher is active, you can signal
3100     it by calling C<ev_async_send>, which is thread- and signal safe.
3101 root 1.122
3102     This functionality is very similar to C<ev_signal> watchers, as signals,
3103     too, are asynchronous in nature, and signals, too, will be compressed
3104     (i.e. the number of callback invocations may be less than the number of
3105     C<ev_async_sent> calls).
3106    
3107     Unlike C<ev_signal> watchers, C<ev_async> works with any event loop, not
3108     just the default loop.
3109    
3110 root 1.124 =head3 Queueing
3111    
3112     C<ev_async> does not support queueing of data in any way. The reason
3113     is that the author does not know of a simple (or any) algorithm for a
3114     multiple-writer-single-reader queue that works in all cases and doesn't
3115 root 1.274 need elaborate support such as pthreads or unportable memory access
3116     semantics.
3117 root 1.124
3118     That means that if you want to queue data, you have to provide your own
3119 root 1.184 queue. But at least I can tell you how to implement locking around your
3120 root 1.130 queue:
3121 root 1.124
3122     =over 4
3123    
3124     =item queueing from a signal handler context
3125    
3126     To implement race-free queueing, you simply add to the queue in the signal
3127 root 1.191 handler but you block the signal handler in the watcher callback. Here is
3128     an example that does that for some fictitious SIGUSR1 handler:
3129 root 1.124
3130     static ev_async mysig;
3131    
3132     static void
3133     sigusr1_handler (void)
3134     {
3135     sometype data;
3136    
3137     // no locking etc.
3138     queue_put (data);
3139 root 1.133 ev_async_send (EV_DEFAULT_ &mysig);
3140 root 1.124 }
3141    
3142     static void
3143     mysig_cb (EV_P_ ev_async *w, int revents)
3144     {
3145     sometype data;
3146     sigset_t block, prev;
3147    
3148     sigemptyset (&block);
3149     sigaddset (&block, SIGUSR1);
3150     sigprocmask (SIG_BLOCK, &block, &prev);
3151    
3152     while (queue_get (&data))
3153     process (data);
3154    
3155     if (sigismember (&prev, SIGUSR1)
3156     sigprocmask (SIG_UNBLOCK, &block, 0);
3157     }
3158    
3159     (Note: pthreads in theory requires you to use C<pthread_setmask>
3160     instead of C<sigprocmask> when you use threads, but libev doesn't do it
3161     either...).
3162    
3163     =item queueing from a thread context
3164    
3165     The strategy for threads is different, as you cannot (easily) block
3166     threads but you can easily preempt them, so to queue safely you need to
3167 root 1.130 employ a traditional mutex lock, such as in this pthread example:
3168 root 1.124
3169     static ev_async mysig;
3170     static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
3171    
3172     static void
3173     otherthread (void)
3174     {
3175     // only need to lock the actual queueing operation
3176     pthread_mutex_lock (&mymutex);
3177     queue_put (data);
3178     pthread_mutex_unlock (&mymutex);
3179    
3180 root 1.133 ev_async_send (EV_DEFAULT_ &mysig);
3181 root 1.124 }
3182    
3183     static void
3184     mysig_cb (EV_P_ ev_async *w, int revents)
3185     {
3186     pthread_mutex_lock (&mymutex);
3187    
3188     while (queue_get (&data))
3189     process (data);
3190    
3191     pthread_mutex_unlock (&mymutex);
3192     }
3193    
3194     =back
3195    
3196    
3197 root 1.122 =head3 Watcher-Specific Functions and Data Members
3198    
3199     =over 4
3200    
3201     =item ev_async_init (ev_async *, callback)
3202    
3203     Initialises and configures the async watcher - it has no parameters of any
3204 root 1.208 kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
3205 root 1.184 trust me.
3206 root 1.122
3207     =item ev_async_send (loop, ev_async *)
3208    
3209     Sends/signals/activates the given C<ev_async> watcher, that is, feeds
3210     an C<EV_ASYNC> event on the watcher into the event loop. Unlike
3211 root 1.184 C<ev_feed_event>, this call is safe to do from other threads, signal or
3212 root 1.161 similar contexts (see the discussion of C<EV_ATOMIC_T> in the embedding
3213 root 1.122 section below on what exactly this means).
3214    
3215 root 1.227 Note that, as with other watchers in libev, multiple events might get
3216     compressed into a single callback invocation (another way to look at this
3217     is that C<ev_async> watchers are level-triggered, set on C<ev_async_send>,
3218     reset when the event loop detects that).
3219    
3220     This call incurs the overhead of a system call only once per event loop
3221     iteration, so while the overhead might be noticeable, it doesn't apply to
3222     repeated calls to C<ev_async_send> for the same event loop.
3223 root 1.122
3224 root 1.140 =item bool = ev_async_pending (ev_async *)
3225    
3226     Returns a non-zero value when C<ev_async_send> has been called on the
3227     watcher but the event has not yet been processed (or even noted) by the
3228     event loop.
3229    
3230     C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
3231     the loop iterates next and checks for the watcher to have become active,
3232     it will reset the flag again. C<ev_async_pending> can be used to very
3233 root 1.161 quickly check whether invoking the loop might be a good idea.
3234 root 1.140
3235 root 1.227 Not that this does I<not> check whether the watcher itself is pending,
3236     only whether it has been requested to make this watcher pending: there
3237     is a time window between the event loop checking and resetting the async
3238     notification, and the callback being invoked.
3239 root 1.140
3240 root 1.122 =back
3241    
3242    
3243 root 1.1 =head1 OTHER FUNCTIONS
3244    
3245 root 1.14 There are some other functions of possible interest. Described. Here. Now.
3246 root 1.1
3247     =over 4
3248    
3249     =item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)
3250    
3251     This function combines a simple timer and an I/O watcher, calls your
3252 root 1.192 callback on whichever event happens first and automatically stops both
3253 root 1.1 watchers. This is useful if you want to wait for a single event on an fd
3254 root 1.22 or timeout without having to allocate/configure/start/stop/free one or
3255 root 1.1 more watchers yourself.
3256    
3257 root 1.192 If C<fd> is less than 0, then no I/O watcher will be started and the
3258     C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
3259     the given C<fd> and C<events> set will be created and started.
3260 root 1.1
3261     If C<timeout> is less than 0, then no timeout watcher will be
3262 root 1.14 started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
3263 root 1.193 repeat = 0) will be started. C<0> is a valid timeout.
3264 root 1.14
3265 root 1.289 The callback has the type C<void (*cb)(int revents, void *arg)> and is
3266 root 1.21 passed an C<revents> set like normal event callbacks (a combination of
3267 root 1.289 C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMER>) and the C<arg>
3268 root 1.193 value passed to C<ev_once>. Note that it is possible to receive I<both>
3269     a timeout and an io event at the same time - you probably should give io
3270     events precedence.
3271    
3272     Example: wait up to ten seconds for data to appear on STDIN_FILENO.
3273 root 1.1
3274 root 1.164 static void stdin_ready (int revents, void *arg)
3275     {
3276 root 1.193 if (revents & EV_READ)
3277     /* stdin might have data for us, joy! */;
3278 root 1.289 else if (revents & EV_TIMER)
3279 root 1.164 /* doh, nothing entered */;
3280     }
3281 root 1.1
3282 root 1.164 ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
3283 root 1.1
3284 root 1.274 =item ev_feed_fd_event (loop, int fd, int revents)
3285 root 1.1
3286 root 1.14 Feed an event on the given fd, as if a file descriptor backend detected
3287     the given events it.
3288 root 1.1
3289 root 1.274 =item ev_feed_signal_event (loop, int signum)
3290 root 1.1
3291 root 1.161 Feed an event as if the given signal occurred (C<loop> must be the default
3292 root 1.36 loop!).
3293 root 1.1
3294     =back
3295    
3296 root 1.34
3297 root 1.20 =head1 LIBEVENT EMULATION
3298    
3299 root 1.24 Libev offers a compatibility emulation layer for libevent. It cannot
3300     emulate the internals of libevent, so here are some usage hints:
3301    
3302     =over 4
3303    
3304     =item * Use it by including <event.h>, as usual.
3305    
3306     =item * The following members are fully supported: ev_base, ev_callback,
3307     ev_arg, ev_fd, ev_res, ev_events.
3308    
3309     =item * Avoid using ev_flags and the EVLIST_*-macros, while it is
3310     maintained by libev, it does not work exactly the same way as in libevent (consider
3311     it a private API).
3312    
3313     =item * Priorities are not currently supported. Initialising priorities
3314     will fail and all watchers will have the same priority, even though there
3315     is an ev_pri field.
3316    
3317 root 1.146 =item * In libevent, the last base created gets the signals, in libev, the
3318     first base created (== the default loop) gets the signals.
3319    
3320 root 1.24 =item * Other members are not supported.
3321    
3322     =item * The libev emulation is I<not> ABI compatible to libevent, you need
3323     to use the libev header file and library.
3324    
3325     =back
3326 root 1.20
3327     =head1 C++ SUPPORT
3328    
3329 root 1.38 Libev comes with some simplistic wrapper classes for C++ that mainly allow
3330 root 1.161 you to use some convenience methods to start/stop watchers and also change
3331 root 1.38 the callback model to a model using method callbacks on objects.
3332    
3333     To use it,
3334    
3335 root 1.164 #include <ev++.h>
3336 root 1.38
3337 root 1.71 This automatically includes F<ev.h> and puts all of its definitions (many
3338     of them macros) into the global namespace. All C++ specific things are
3339     put into the C<ev> namespace. It should support all the same embedding
3340     options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
3341    
3342 root 1.72 Care has been taken to keep the overhead low. The only data member the C++
3343     classes add (compared to plain C-style watchers) is the event loop pointer
3344     that the watcher is associated with (or no additional members at all if
3345     you disable C<EV_MULTIPLICITY> when embedding libev).
3346 root 1.71
3347 root 1.72 Currently, functions, and static and non-static member functions can be
3348 root 1.71 used as callbacks. Other types should be easy to add as long as they only
3349     need one additional pointer for context. If you need support for other
3350     types of functors please contact the author (preferably after implementing
3351     it).
3352 root 1.38
3353     Here is a list of things available in the C<ev> namespace:
3354    
3355     =over 4
3356    
3357     =item C<ev::READ>, C<ev::WRITE> etc.
3358    
3359     These are just enum values with the same values as the C<EV_READ> etc.
3360     macros from F<ev.h>.
3361    
3362     =item C<ev::tstamp>, C<ev::now>
3363    
3364     Aliases to the same types/functions as with the C<ev_> prefix.
3365    
3366     =item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
3367    
3368     For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
3369     the same name in the C<ev> namespace, with the exception of C<ev_signal>
3370     which is called C<ev::sig> to avoid clashes with the C<signal> macro
3371     defines by many implementations.
3372    
3373     All of those classes have these methods:
3374    
3375     =over 4
3376    
3377 root 1.71 =item ev::TYPE::TYPE ()
3378 root 1.38
3379 root 1.274 =item ev::TYPE::TYPE (loop)
3380 root 1.38
3381     =item ev::TYPE::~TYPE
3382    
3383 root 1.71 The constructor (optionally) takes an event loop to associate the watcher
3384     with. If it is omitted, it will use C<EV_DEFAULT>.
3385    
3386     The constructor calls C<ev_init> for you, which means you have to call the
3387     C<set> method before starting it.
3388    
3389     It will not set a callback, however: You have to call the templated C<set>
3390     method to set a callback before you can start the watcher.
3391    
3392     (The reason why you have to use a method is a limitation in C++ which does
3393     not allow explicit template arguments for constructors).
3394 root 1.38
3395     The destructor automatically stops the watcher if it is active.
3396    
3397 root 1.71 =item w->set<class, &class::method> (object *)
3398    
3399     This method sets the callback method to call. The method has to have a
3400     signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
3401     first argument and the C<revents> as second. The object must be given as
3402     parameter and is stored in the C<data> member of the watcher.
3403    
3404     This method synthesizes efficient thunking code to call your method from
3405     the C callback that libev requires. If your compiler can inline your
3406     callback (i.e. it is visible to it at the place of the C<set> call and
3407     your compiler is good :), then the method will be fully inlined into the
3408     thunking function, making it as fast as a direct C callback.
3409    
3410     Example: simple class declaration and watcher initialisation
3411    
3412 root 1.164 struct myclass
3413     {
3414     void io_cb (ev::io &w, int revents) { }
3415     }
3416    
3417     myclass obj;
3418     ev::io iow;
3419     iow.set <myclass, &myclass::io_cb> (&obj);
3420 root 1.71
3421 root 1.221 =item w->set (object *)
3422    
3423     This is a variation of a method callback - leaving out the method to call
3424     will default the method to C<operator ()>, which makes it possible to use
3425     functor objects without having to manually specify the C<operator ()> all
3426     the time. Incidentally, you can then also leave out the template argument
3427     list.
3428    
3429     The C<operator ()> method prototype must be C<void operator ()(watcher &w,
3430     int revents)>.
3431    
3432     See the method-C<set> above for more details.
3433    
3434     Example: use a functor object as callback.
3435    
3436     struct myfunctor
3437     {
3438     void operator() (ev::io &w, int revents)
3439     {
3440     ...
3441     }
3442     }
3443    
3444     myfunctor f;
3445    
3446     ev::io w;
3447     w.set (&f);
3448    
3449 root 1.75 =item w->set<function> (void *data = 0)
3450 root 1.71
3451     Also sets a callback, but uses a static method or plain function as
3452     callback. The optional C<data> argument will be stored in the watcher's
3453     C<data> member and is free for you to use.
3454    
3455 root 1.75 The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
3456    
3457 root 1.71 See the method-C<set> above for more details.
3458    
3459 root 1.184 Example: Use a plain function as callback.
3460 root 1.75
3461 root 1.164 static void io_cb (ev::io &w, int revents) { }
3462     iow.set <io_cb> ();
3463 root 1.75
3464 root 1.274 =item w->set (loop)
3465 root 1.38
3466     Associates a different C<struct ev_loop> with this watcher. You can only
3467     do this when the watcher is inactive (and not pending either).
3468    
3469 root 1.161 =item w->set ([arguments])
3470 root 1.38
3471 root 1.307 Basically the same as C<ev_TYPE_set>, with the same arguments. Either this
3472     method or a suitable start method must be called at least once. Unlike the
3473     C counterpart, an active watcher gets automatically stopped and restarted
3474     when reconfiguring it with this method.
3475 root 1.38
3476     =item w->start ()
3477    
3478 root 1.71 Starts the watcher. Note that there is no C<loop> argument, as the
3479     constructor already stores the event loop.
3480 root 1.38
3481 root 1.307 =item w->start ([arguments])
3482    
3483     Instead of calling C<set> and C<start> methods separately, it is often
3484     convenient to wrap them in one call. Uses the same type of arguments as
3485     the configure C<set> method of the watcher.
3486    
3487 root 1.38 =item w->stop ()
3488    
3489     Stops the watcher if it is active. Again, no C<loop> argument.
3490    
3491 root 1.84 =item w->again () (C<ev::timer>, C<ev::periodic> only)
3492 root 1.38
3493     For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
3494     C<ev_TYPE_again> function.
3495    
3496 root 1.84 =item w->sweep () (C<ev::embed> only)
3497 root 1.38
3498     Invokes C<ev_embed_sweep>.
3499    
3500 root 1.84 =item w->update () (C<ev::stat> only)
3501 root 1.49
3502     Invokes C<ev_stat_stat>.
3503    
3504 root 1.38 =back
3505    
3506     =back
3507    
3508 root 1.307 Example: Define a class with two I/O and idle watchers, start the I/O
3509     watchers in the constructor.
3510 root 1.38
3511 root 1.164 class myclass
3512     {
3513 root 1.184 ev::io io ; void io_cb (ev::io &w, int revents);
3514 root 1.307 ev::io2 io2 ; void io2_cb (ev::io &w, int revents);
3515 root 1.184 ev::idle idle; void idle_cb (ev::idle &w, int revents);
3516 root 1.164
3517     myclass (int fd)
3518     {
3519     io .set <myclass, &myclass::io_cb > (this);
3520 root 1.307 io2 .set <myclass, &myclass::io2_cb > (this);
3521 root 1.164 idle.set <myclass, &myclass::idle_cb> (this);
3522    
3523 root 1.307 io.set (fd, ev::WRITE); // configure the watcher
3524     io.start (); // start it whenever convenient
3525    
3526     io2.start (fd, ev::READ); // set + start in one call
3527 root 1.164 }
3528     };
3529 root 1.20
3530 root 1.50
3531 root 1.136 =head1 OTHER LANGUAGE BINDINGS
3532    
3533     Libev does not offer other language bindings itself, but bindings for a
3534 root 1.161 number of languages exist in the form of third-party packages. If you know
3535 root 1.136 any interesting language binding in addition to the ones listed here, drop
3536     me a note.
3537    
3538     =over 4
3539    
3540     =item Perl
3541    
3542     The EV module implements the full libev API and is actually used to test
3543     libev. EV is developed together with libev. Apart from the EV core module,
3544     there are additional modules that implement libev-compatible interfaces
3545 root 1.184 to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
3546     C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
3547     and C<EV::Glib>).
3548 root 1.136
3549 root 1.166 It can be found and installed via CPAN, its homepage is at
3550 root 1.136 L<http://software.schmorp.de/pkg/EV>.
3551    
3552 root 1.166 =item Python
3553    
3554     Python bindings can be found at L<http://code.google.com/p/pyev/>. It
3555 root 1.228 seems to be quite complete and well-documented.
3556 root 1.166
3557 root 1.136 =item Ruby
3558    
3559     Tony Arcieri has written a ruby extension that offers access to a subset
3560 root 1.161 of the libev API and adds file handle abstractions, asynchronous DNS and
3561 root 1.136 more on top of it. It can be found via gem servers. Its homepage is at
3562     L<http://rev.rubyforge.org/>.
3563    
3564 root 1.218 Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
3565     makes rev work even on mingw.
3566    
3567 root 1.228 =item Haskell
3568    
3569     A haskell binding to libev is available at
3570     L<http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
3571    
3572 root 1.136 =item D
3573    
3574     Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
3575 root 1.172 be found at L<http://proj.llucax.com.ar/wiki/evd>.
3576 root 1.136
3577 root 1.201 =item Ocaml
3578    
3579     Erkki Seppala has written Ocaml bindings for libev, to be found at
3580     L<http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
3581    
3582 root 1.263 =item Lua
3583    
3584 root 1.279 Brian Maher has written a partial interface to libev for lua (at the
3585     time of this writing, only C<ev_io> and C<ev_timer>), to be found at
3586 root 1.263 L<http://github.com/brimworks/lua-ev>.
3587    
3588 root 1.136 =back
3589    
3590    
3591 root 1.50 =head1 MACRO MAGIC
3592    
3593 root 1.161 Libev can be compiled with a variety of options, the most fundamental
3594 root 1.84 of which is C<EV_MULTIPLICITY>. This option determines whether (most)
3595     functions and callbacks have an initial C<struct ev_loop *> argument.
3596 root 1.50
3597     To make it easier to write programs that cope with either variant, the
3598     following macros are defined:
3599    
3600     =over 4
3601    
3602     =item C<EV_A>, C<EV_A_>
3603    
3604     This provides the loop I<argument> for functions, if one is required ("ev
3605     loop argument"). The C<EV_A> form is used when this is the sole argument,
3606     C<EV_A_> is used when other arguments are following. Example:
3607    
3608 root 1.164 ev_unref (EV_A);
3609     ev_timer_add (EV_A_ watcher);
3610 root 1.310 ev_run (EV_A_ 0);
3611 root 1.50
3612     It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
3613     which is often provided by the following macro.
3614    
3615     =item C<EV_P>, C<EV_P_>
3616    
3617     This provides the loop I<parameter> for functions, if one is required ("ev
3618     loop parameter"). The C<EV_P> form is used when this is the sole parameter,
3619     C<EV_P_> is used when other parameters are following. Example:
3620    
3621 root 1.164 // this is how ev_unref is being declared
3622     static void ev_unref (EV_P);
3623 root 1.50
3624 root 1.164 // this is how you can declare your typical callback
3625     static void cb (EV_P_ ev_timer *w, int revents)
3626 root 1.50
3627     It declares a parameter C<loop> of type C<struct ev_loop *>, quite
3628     suitable for use with C<EV_A>.
3629    
3630     =item C<EV_DEFAULT>, C<EV_DEFAULT_>
3631    
3632     Similar to the other two macros, this gives you the value of the default
3633     loop, if multiple loops are supported ("ev loop default").
3634    
3635 root 1.143 =item C<EV_DEFAULT_UC>, C<EV_DEFAULT_UC_>
3636    
3637     Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
3638     default loop has been initialised (C<UC> == unchecked). Their behaviour
3639     is undefined when the default loop has not been initialised by a previous
3640     execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
3641    
3642     It is often prudent to use C<EV_DEFAULT> when initialising the first
3643     watcher in a function but use C<EV_DEFAULT_UC> afterwards.
3644    
3645 root 1.50 =back
3646    
3647 root 1.63 Example: Declare and initialise a check watcher, utilising the above
3648 root 1.68 macros so it will work regardless of whether multiple loops are supported
3649 root 1.63 or not.
3650 root 1.50
3651 root 1.164 static void
3652     check_cb (EV_P_ ev_timer *w, int revents)
3653     {
3654     ev_check_stop (EV_A_ w);
3655     }
3656    
3657     ev_check check;
3658     ev_check_init (&check, check_cb);
3659     ev_check_start (EV_DEFAULT_ &check);
3660 root 1.310 ev_run (EV_DEFAULT_ 0);
3661 root 1.50
3662 root 1.39 =head1 EMBEDDING
3663    
3664     Libev can (and often is) directly embedded into host
3665     applications. Examples of applications that embed it include the Deliantra
3666     Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
3667     and rxvt-unicode.
3668    
3669 root 1.91 The goal is to enable you to just copy the necessary files into your
3670 root 1.39 source directory without having to change even a single line in them, so
3671     you can easily upgrade by simply copying (or having a checked-out copy of
3672     libev somewhere in your source tree).
3673    
3674     =head2 FILESETS
3675    
3676     Depending on what features you need you need to include one or more sets of files
3677 root 1.161 in your application.
3678 root 1.39
3679     =head3 CORE EVENT LOOP
3680    
3681     To include only the libev core (all the C<ev_*> functions), with manual
3682     configuration (no autoconf):
3683    
3684 root 1.164 #define EV_STANDALONE 1
3685     #include "ev.c"
3686 root 1.39
3687     This will automatically include F<ev.h>, too, and should be done in a
3688     single C source file only to provide the function implementations. To use
3689     it, do the same for F<ev.h> in all files wishing to use this API (best
3690     done by writing a wrapper around F<ev.h> that you can include instead and
3691     where you can put other configuration options):
3692    
3693 root 1.164 #define EV_STANDALONE 1
3694     #include "ev.h"
3695 root 1.39
3696     Both header files and implementation files can be compiled with a C++
3697 root 1.208 compiler (at least, that's a stated goal, and breakage will be treated
3698 root 1.39 as a bug).
3699    
3700     You need the following files in your source tree, or in a directory
3701     in your include path (e.g. in libev/ when using -Ilibev):
3702    
3703 root 1.164 ev.h
3704     ev.c
3705     ev_vars.h
3706     ev_wrap.h
3707    
3708     ev_win32.c required on win32 platforms only
3709    
3710     ev_select.c only when select backend is enabled (which is enabled by default)
3711     ev_poll.c only when poll backend is enabled (disabled by default)
3712     ev_epoll.c only when the epoll backend is enabled (disabled by default)
3713     ev_kqueue.c only when the kqueue backend is enabled (disabled by default)
3714     ev_port.c only when the solaris port backend is enabled (disabled by default)
3715 root 1.39
3716     F<ev.c> includes the backend files directly when enabled, so you only need
3717 root 1.43 to compile this single file.
3718 root 1.39
3719     =head3 LIBEVENT COMPATIBILITY API
3720    
3721     To include the libevent compatibility API, also include:
3722    
3723 root 1.164 #include "event.c"
3724 root 1.39
3725     in the file including F<ev.c>, and:
3726    
3727 root 1.164 #include "event.h"
3728 root 1.39
3729     in the files that want to use the libevent API. This also includes F<ev.h>.
3730    
3731     You need the following additional files for this:
3732    
3733 root 1.164 event.h
3734     event.c
3735 root 1.39
3736     =head3 AUTOCONF SUPPORT
3737    
3738 root 1.161 Instead of using C<EV_STANDALONE=1> and providing your configuration in
3739 root 1.39 whatever way you want, you can also C<m4_include([libev.m4])> in your
3740 root 1.43 F<configure.ac> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
3741     include F<config.h> and configure itself accordingly.
3742 root 1.39
3743     For this of course you need the m4 file:
3744    
3745 root 1.164 libev.m4
3746 root 1.39
3747     =head2 PREPROCESSOR SYMBOLS/MACROS
3748    
3749 root 1.142 Libev can be configured via a variety of preprocessor symbols you have to
3750 root 1.281 define before including (or compiling) any of its files. The default in
3751     the absence of autoconf is documented for every option.
3752    
3753     Symbols marked with "(h)" do not change the ABI, and can have different
3754     values when compiling libev vs. including F<ev.h>, so it is permissible
3755 sf-exg 1.292 to redefine them before including F<ev.h> without breaking compatibility
3756 root 1.281 to a compiled library. All other symbols change the ABI, which means all
3757     users of libev and the libev code itself must be compiled with compatible
3758     settings.
3759 root 1.39
3760     =over 4
3761    
3762 root 1.310 =item EV_COMPAT3 (h)
3763    
3764     Backwards compatibility is a major concern for libev. This is why this
3765     release of libev comes with wrappers for the functions and symbols that
3766     have been renamed between libev version 3 and 4.
3767    
3768     You can disable these wrappers (to test compatibility with future
3769     versions) by defining C<EV_COMPAT3> to C<0> when compiling your
3770     sources. This has the additional advantage that you can drop the C<struct>
3771     from C<struct ev_loop> declarations, as libev will provide an C<ev_loop>
3772     typedef in that case.
3773    
3774     In some future version, the default for C<EV_COMPAT3> will become C<0>,
3775     and in some even more future version the compatibility code will be
3776     removed completely.
3777    
3778 root 1.281 =item EV_STANDALONE (h)
3779 root 1.39
3780     Must always be C<1> if you do not use autoconf configuration, which
3781     keeps libev from including F<config.h>, and it also defines dummy
3782     implementations for some libevent functions (such as logging, which is not
3783     supported). It will also not define any of the structs usually found in
3784     F<event.h> that are not directly supported by the libev core alone.
3785    
3786 root 1.262 In standalone mode, libev will still try to automatically deduce the
3787 root 1.218 configuration, but has to be more conservative.
3788    
3789 root 1.39 =item EV_USE_MONOTONIC
3790    
3791     If defined to be C<1>, libev will try to detect the availability of the
3792 root 1.218 monotonic clock option at both compile time and runtime. Otherwise no
3793     use of the monotonic clock option will be attempted. If you enable this,
3794     you usually have to link against librt or something similar. Enabling it
3795     when the functionality isn't available is safe, though, although you have
3796 root 1.39 to make sure you link against any libraries where the C<clock_gettime>
3797 root 1.218 function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
3798 root 1.39
3799     =item EV_USE_REALTIME
3800    
3801     If defined to be C<1>, libev will try to detect the availability of the
3802 root 1.224 real-time clock option at compile time (and assume its availability
3803     at runtime if successful). Otherwise no use of the real-time clock
3804     option will be attempted. This effectively replaces C<gettimeofday>
3805     by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
3806     correctness. See the note about libraries in the description of
3807     C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
3808     C<EV_USE_CLOCK_SYSCALL>.
3809 root 1.39
3810 root 1.218 =item EV_USE_CLOCK_SYSCALL
3811    
3812     If defined to be C<1>, libev will try to use a direct syscall instead
3813     of calling the system-provided C<clock_gettime> function. This option
3814     exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
3815     unconditionally pulls in C<libpthread>, slowing down single-threaded
3816 root 1.219 programs needlessly. Using a direct syscall is slightly slower (in
3817     theory), because no optimised vdso implementation can be used, but avoids
3818     the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
3819     higher, as it simplifies linking (no need for C<-lrt>).
3820 root 1.218
3821 root 1.97 =item EV_USE_NANOSLEEP
3822    
3823     If defined to be C<1>, libev will assume that C<nanosleep ()> is available
3824     and will use it for delays. Otherwise it will use C<select ()>.
3825    
3826 root 1.142 =item EV_USE_EVENTFD
3827    
3828     If defined to be C<1>, then libev will assume that C<eventfd ()> is
3829     available and will probe for kernel support at runtime. This will improve
3830     C<ev_signal> and C<ev_async> performance and reduce resource consumption.
3831     If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
3832     2.7 or newer, otherwise disabled.
3833    
3834 root 1.39 =item EV_USE_SELECT
3835    
3836     If undefined or defined to be C<1>, libev will compile in support for the
3837 root 1.161 C<select>(2) backend. No attempt at auto-detection will be done: if no
3838 root 1.39 other method takes over, select will be it. Otherwise the select backend
3839     will not be compiled in.
3840    
3841     =item EV_SELECT_USE_FD_SET
3842    
3843     If defined to C<1>, then the select backend will use the system C<fd_set>
3844     structure. This is useful if libev doesn't compile due to a missing
3845 root 1.218 C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
3846     on exotic systems. This usually limits the range of file descriptors to
3847     some low limit such as 1024 or might have other limitations (winsocket
3848     only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
3849     configures the maximum size of the C<fd_set>.
3850 root 1.39
3851     =item EV_SELECT_IS_WINSOCKET
3852    
3853     When defined to C<1>, the select backend will assume that
3854     select/socket/connect etc. don't understand file descriptors but
3855     wants osf handles on win32 (this is the case when the select to
3856     be used is the winsock select). This means that it will call
3857     C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
3858     it is assumed that all these functions actually work on fds, even
3859     on win32. Should not be defined on non-win32 platforms.
3860    
3861 root 1.264 =item EV_FD_TO_WIN32_HANDLE(fd)
3862 root 1.112
3863     If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
3864     file descriptors to socket handles. When not defining this symbol (the
3865     default), then libev will call C<_get_osfhandle>, which is usually
3866     correct. In some cases, programs use their own file descriptor management,
3867     in which case they can provide this function to map fds to socket handles.
3868    
3869 root 1.264 =item EV_WIN32_HANDLE_TO_FD(handle)
3870    
3871     If C<EV_SELECT_IS_WINSOCKET> then libev maps handles to file descriptors
3872     using the standard C<_open_osfhandle> function. For programs implementing
3873     their own fd to handle mapping, overwriting this function makes it easier
3874     to do so. This can be done by defining this macro to an appropriate value.
3875    
3876     =item EV_WIN32_CLOSE_FD(fd)
3877    
3878     If programs implement their own fd to handle mapping on win32, then this
3879     macro can be used to override the C<close> function, useful to unregister
3880     file descriptors again. Note that the replacement function has to close
3881     the underlying OS handle.
3882    
3883 root 1.39 =item EV_USE_POLL
3884    
3885     If defined to be C<1>, libev will compile in support for the C<poll>(2)
3886     backend. Otherwise it will be enabled on non-win32 platforms. It
3887     takes precedence over select.
3888    
3889     =item EV_USE_EPOLL
3890    
3891     If defined to be C<1>, libev will compile in support for the Linux
3892     C<epoll>(7) backend. Its availability will be detected at runtime,
3893 root 1.142 otherwise another method will be used as fallback. This is the preferred
3894     backend for GNU/Linux systems. If undefined, it will be enabled if the
3895     headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
3896 root 1.39
3897     =item EV_USE_KQUEUE
3898    
3899     If defined to be C<1>, libev will compile in support for the BSD style
3900     C<kqueue>(2) backend. Its actual availability will be detected at runtime,
3901     otherwise another method will be used as fallback. This is the preferred
3902     backend for BSD and BSD-like systems, although on most BSDs kqueue only
3903     supports some types of fds correctly (the only platform we found that
3904     supports ptys for example was NetBSD), so kqueue might be compiled in, but
3905     not be used unless explicitly requested. The best way to use it is to find
3906 root 1.41 out whether kqueue supports your type of fd properly and use an embedded
3907 root 1.39 kqueue loop.
3908    
3909     =item EV_USE_PORT
3910    
3911     If defined to be C<1>, libev will compile in support for the Solaris
3912     10 port style backend. Its availability will be detected at runtime,
3913     otherwise another method will be used as fallback. This is the preferred
3914     backend for Solaris 10 systems.
3915    
3916     =item EV_USE_DEVPOLL
3917    
3918 root 1.161 Reserved for future expansion, works like the USE symbols above.
3919 root 1.39
3920 root 1.56 =item EV_USE_INOTIFY
3921    
3922     If defined to be C<1>, libev will compile in support for the Linux inotify
3923     interface to speed up C<ev_stat> watchers. Its actual availability will
3924 root 1.142 be detected at runtime. If undefined, it will be enabled if the headers
3925     indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
3926 root 1.56
3927 root 1.123 =item EV_ATOMIC_T
3928    
3929     Libev requires an integer type (suitable for storing C<0> or C<1>) whose
3930 root 1.126 access is atomic with respect to other threads or signal contexts. No such
3931     type is easily found in the C language, so you can provide your own type
3932 root 1.127 that you know is safe for your purposes. It is used both for signal handler "locking"
3933     as well as for signal and thread safety in C<ev_async> watchers.
3934 root 1.123
3935 root 1.161 In the absence of this define, libev will use C<sig_atomic_t volatile>
3936 root 1.126 (from F<signal.h>), which is usually good enough on most platforms.
3937 root 1.123
3938 root 1.281 =item EV_H (h)
3939 root 1.39
3940     The name of the F<ev.h> header file used to include it. The default if
3941 root 1.118 undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
3942     used to virtually rename the F<ev.h> header file in case of conflicts.
3943 root 1.39
3944 root 1.281 =item EV_CONFIG_H (h)
3945 root 1.39
3946     If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
3947     F<ev.c>'s idea of where to find the F<config.h> file, similarly to
3948     C<EV_H>, above.
3949    
3950 root 1.281 =item EV_EVENT_H (h)
3951 root 1.39
3952     Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
3953 root 1.118 of how the F<event.h> header can be found, the default is C<"event.h">.
3954 root 1.39
3955 root 1.281 =item EV_PROTOTYPES (h)
3956 root 1.39
3957     If defined to be C<0>, then F<ev.h> will not define any function
3958     prototypes, but still define all the structs and other symbols. This is
3959     occasionally useful if you want to provide your own wrapper functions
3960     around libev functions.
3961    
3962     =item EV_MULTIPLICITY
3963    
3964     If undefined or defined to C<1>, then all event-loop-specific functions
3965     will have the C<struct ev_loop *> as first argument, and you can create
3966     additional independent event loops. Otherwise there will be no support
3967     for multiple event loops and there is no first event loop pointer
3968     argument. Instead, all functions act on the single default loop.
3969    
3970 root 1.69 =item EV_MINPRI
3971    
3972     =item EV_MAXPRI
3973    
3974     The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
3975     C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
3976     provide for more priorities by overriding those symbols (usually defined
3977     to be C<-2> and C<2>, respectively).
3978    
3979     When doing priority-based operations, libev usually has to linearly search
3980     all the priorities, so having many of them (hundreds) uses a lot of space
3981     and time, so using the defaults of five priorities (-2 .. +2) is usually
3982     fine.
3983    
3984 root 1.184 If your embedding application does not need any priorities, defining these
3985     both to C<0> will save some memory and CPU.
3986 root 1.69
3987 root 1.283 =item EV_PERIODIC_ENABLE, EV_IDLE_ENABLE, EV_EMBED_ENABLE, EV_STAT_ENABLE,
3988     EV_PREPARE_ENABLE, EV_CHECK_ENABLE, EV_FORK_ENABLE, EV_SIGNAL_ENABLE,
3989     EV_ASYNC_ENABLE, EV_CHILD_ENABLE.
3990    
3991     If undefined or defined to be C<1> (and the platform supports it), then
3992     the respective watcher type is supported. If defined to be C<0>, then it
3993 sf-exg 1.299 is not. Disabling watcher types mainly saves code size.
3994 root 1.282
3995 root 1.285 =item EV_FEATURES
3996 root 1.47
3997     If you need to shave off some kilobytes of code at the expense of some
3998 root 1.285 speed (but with the full API), you can define this symbol to request
3999     certain subsets of functionality. The default is to enable all features
4000     that can be enabled on the platform.
4001    
4002     A typical way to use this symbol is to define it to C<0> (or to a bitset
4003     with some broad features you want) and then selectively re-enable
4004     additional parts you want, for example if you want everything minimal,
4005     but multiple event loop support, async and child watchers and the poll
4006     backend, use this:
4007    
4008     #define EV_FEATURES 0
4009     #define EV_MULTIPLICITY 1
4010     #define EV_USE_POLL 1
4011     #define EV_CHILD_ENABLE 1
4012     #define EV_ASYNC_ENABLE 1
4013    
4014     The actual value is a bitset, it can be a combination of the following
4015     values:
4016    
4017     =over 4
4018    
4019     =item C<1> - faster/larger code
4020    
4021     Use larger code to speed up some operations.
4022    
4023 sf-exg 1.299 Currently this is used to override some inlining decisions (enlarging the
4024     code size by roughly 30% on amd64).
4025 root 1.285
4026 root 1.286 When optimising for size, use of compiler flags such as C<-Os> with
4027 sf-exg 1.299 gcc is recommended, as well as C<-DNDEBUG>, as libev contains a number of
4028 root 1.286 assertions.
4029 root 1.285
4030     =item C<2> - faster/larger data structures
4031    
4032     Replaces the small 2-heap for timer management by a faster 4-heap, larger
4033 sf-exg 1.299 hash table sizes and so on. This will usually further increase code size
4034 root 1.285 and can additionally have an effect on the size of data structures at
4035     runtime.
4036    
4037     =item C<4> - full API configuration
4038    
4039     This enables priorities (sets C<EV_MAXPRI>=2 and C<EV_MINPRI>=-2), and
4040     enables multiplicity (C<EV_MULTIPLICITY>=1).
4041    
4042 root 1.287 =item C<8> - full API
4043    
4044     This enables a lot of the "lesser used" API functions. See C<ev.h> for
4045     details on which parts of the API are still available without this
4046 root 1.285 feature, and do not complain if this subset changes over time.
4047    
4048 root 1.287 =item C<16> - enable all optional watcher types
4049 root 1.285
4050     Enables all optional watcher types. If you want to selectively enable
4051     only some watcher types other than I/O and timers (e.g. prepare,
4052     embed, async, child...) you can enable them manually by defining
4053     C<EV_watchertype_ENABLE> to C<1> instead.
4054    
4055 root 1.287 =item C<32> - enable all backends
4056 root 1.285
4057     This enables all backends - without this feature, you need to enable at
4058     least one backend manually (C<EV_USE_SELECT> is a good choice).
4059    
4060 root 1.287 =item C<64> - enable OS-specific "helper" APIs
4061 root 1.285
4062     Enable inotify, eventfd, signalfd and similar OS-specific helper APIs by
4063     default.
4064    
4065     =back
4066    
4067     Compiling with C<gcc -Os -DEV_STANDALONE -DEV_USE_EPOLL=1 -DEV_FEATURES=0>
4068 root 1.288 reduces the compiled size of libev from 24.7Kb code/2.8Kb data to 6.5Kb
4069     code/0.3Kb data on my GNU/Linux amd64 system, while still giving you I/O
4070     watchers, timers and monotonic clock support.
4071 root 1.285
4072     With an intelligent-enough linker (gcc+binutils are intelligent enough
4073     when you use C<-Wl,--gc-sections -ffunction-sections>) functions unused by
4074     your program might be left out as well - a binary starting a timer and an
4075     I/O watcher then might come out at only 5Kb.
4076 root 1.282
4077 root 1.281 =item EV_AVOID_STDIO
4078    
4079     If this is set to C<1> at compiletime, then libev will avoid using stdio
4080 sf-exg 1.299 functions (printf, scanf, perror etc.). This will increase the code size
4081 root 1.281 somewhat, but if your program doesn't otherwise depend on stdio and your
4082     libc allows it, this avoids linking in the stdio library which is quite
4083     big.
4084    
4085     Note that error messages might become less precise when this option is
4086     enabled.
4087    
4088 root 1.260 =item EV_NSIG
4089    
4090     The highest supported signal number, +1 (or, the number of
4091     signals): Normally, libev tries to deduce the maximum number of signals
4092     automatically, but sometimes this fails, in which case it can be
4093     specified. Also, using a lower number than detected (C<32> should be
4094 sf-exg 1.298 good for about any system in existence) can save some memory, as libev
4095 root 1.260 statically allocates some 12-24 bytes per signal number.
4096    
4097 root 1.51 =item EV_PID_HASHSIZE
4098    
4099     C<ev_child> watchers use a small hash table to distribute workload by
4100 root 1.285 pid. The default size is C<16> (or C<1> with C<EV_FEATURES> disabled),
4101     usually more than enough. If you need to manage thousands of children you
4102     might want to increase this value (I<must> be a power of two).
4103 root 1.56
4104     =item EV_INOTIFY_HASHSIZE
4105    
4106 root 1.104 C<ev_stat> watchers use a small hash table to distribute workload by
4107 root 1.285 inotify watch id. The default size is C<16> (or C<1> with C<EV_FEATURES>
4108     disabled), usually more than enough. If you need to manage thousands of
4109     C<ev_stat> watchers you might want to increase this value (I<must> be a
4110     power of two).
4111 root 1.51
4112 root 1.153 =item EV_USE_4HEAP
4113    
4114     Heaps are not very cache-efficient. To improve the cache-efficiency of the
4115 root 1.184 timer and periodics heaps, libev uses a 4-heap when this symbol is defined
4116     to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
4117     faster performance with many (thousands) of watchers.
4118 root 1.153
4119 root 1.285 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4120     will be C<0>.
4121 root 1.153
4122     =item EV_HEAP_CACHE_AT
4123    
4124     Heaps are not very cache-efficient. To improve the cache-efficiency of the
4125 root 1.184 timer and periodics heaps, libev can cache the timestamp (I<at>) within
4126 root 1.153 the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
4127     which uses 8-12 bytes more per watcher and a few hundred bytes more code,
4128 root 1.155 but avoids random read accesses on heap changes. This improves performance
4129 root 1.184 noticeably with many (hundreds) of watchers.
4130 root 1.153
4131 root 1.285 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4132     will be C<0>.
4133 root 1.153
4134 root 1.159 =item EV_VERIFY
4135    
4136 root 1.309 Controls how much internal verification (see C<ev_verify ()>) will
4137 root 1.159 be done: If set to C<0>, no internal verification code will be compiled
4138     in. If set to C<1>, then verification code will be compiled in, but not
4139     called. If set to C<2>, then the internal verification code will be
4140     called once per loop, which can slow down libev. If set to C<3>, then the
4141     verification code will be called very frequently, which will slow down
4142     libev considerably.
4143    
4144 root 1.285 The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
4145     will be C<0>.
4146 root 1.159
4147 root 1.39 =item EV_COMMON
4148    
4149     By default, all watchers have a C<void *data> member. By redefining
4150 sf-exg 1.300 this macro to something else you can include more and other types of
4151 root 1.39 members. You have to define it each time you include one of the files,
4152     though, and it must be identical each time.
4153    
4154     For example, the perl EV module uses something like this:
4155    
4156 root 1.164 #define EV_COMMON \
4157     SV *self; /* contains this struct */ \
4158     SV *cb_sv, *fh /* note no trailing ";" */
4159 root 1.39
4160 root 1.44 =item EV_CB_DECLARE (type)
4161 root 1.39
4162 root 1.44 =item EV_CB_INVOKE (watcher, revents)
4163 root 1.39
4164 root 1.44 =item ev_set_cb (ev, cb)
4165 root 1.39
4166     Can be used to change the callback member declaration in each watcher,
4167     and the way callbacks are invoked and set. Must expand to a struct member
4168 root 1.93 definition and a statement, respectively. See the F<ev.h> header file for
4169 root 1.39 their default definitions. One possible use for overriding these is to
4170 root 1.44 avoid the C<struct ev_loop *> as first argument in all cases, or to use
4171     method calls instead of plain function calls in C++.
4172 root 1.39
4173 root 1.185 =back
4174    
4175 root 1.89 =head2 EXPORTED API SYMBOLS
4176    
4177 root 1.161 If you need to re-export the API (e.g. via a DLL) and you need a list of
4178 root 1.89 exported symbols, you can use the provided F<Symbol.*> files which list
4179     all public symbols, one per line:
4180    
4181 root 1.164 Symbols.ev for libev proper
4182     Symbols.event for the libevent emulation
4183 root 1.89
4184     This can also be used to rename all public symbols to avoid clashes with
4185     multiple versions of libev linked together (which is obviously bad in
4186 root 1.161 itself, but sometimes it is inconvenient to avoid this).
4187 root 1.89
4188 root 1.92 A sed command like this will create wrapper C<#define>'s that you need to
4189 root 1.89 include before including F<ev.h>:
4190    
4191     <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
4192    
4193     This would create a file F<wrap.h> which essentially looks like this:
4194    
4195     #define ev_backend myprefix_ev_backend
4196     #define ev_check_start myprefix_ev_check_start
4197     #define ev_check_stop myprefix_ev_check_stop
4198     ...
4199    
4200 root 1.39 =head2 EXAMPLES
4201    
4202     For a real-world example of a program the includes libev
4203     verbatim, you can have a look at the EV perl module
4204     (L<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
4205     the F<libev/> subdirectory and includes them in the F<EV/EVAPI.h> (public
4206     interface) and F<EV.xs> (implementation) files. Only the F<EV.xs> file
4207     will be compiled. It is pretty complex because it provides its own header
4208     file.
4209    
4210     The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
4211 root 1.63 that everybody includes and which overrides some configure choices:
4212 root 1.39
4213 root 1.287 #define EV_FEATURES 8
4214 root 1.285 #define EV_USE_SELECT 1
4215 root 1.287 #define EV_PREPARE_ENABLE 1
4216     #define EV_IDLE_ENABLE 1
4217     #define EV_SIGNAL_ENABLE 1
4218     #define EV_CHILD_ENABLE 1
4219     #define EV_USE_STDEXCEPT 0
4220 root 1.164 #define EV_CONFIG_H <config.h>
4221 root 1.39
4222 root 1.164 #include "ev++.h"
4223 root 1.39
4224     And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
4225    
4226 root 1.164 #include "ev_cpp.h"
4227     #include "ev.c"
4228 root 1.39
4229 root 1.189 =head1 INTERACTION WITH OTHER PROGRAMS OR LIBRARIES
4230 root 1.46
4231 root 1.189 =head2 THREADS AND COROUTINES
4232 root 1.144
4233 root 1.189 =head3 THREADS
4234 root 1.144
4235 root 1.186 All libev functions are reentrant and thread-safe unless explicitly
4236 root 1.191 documented otherwise, but libev implements no locking itself. This means
4237     that you can use as many loops as you want in parallel, as long as there
4238     are no concurrent calls into any libev function with the same loop
4239     parameter (C<ev_default_*> calls have an implicit default loop parameter,
4240     of course): libev guarantees that different event loops share no data
4241 root 1.186 structures that need any locking.
4242 root 1.180
4243     Or to put it differently: calls with different loop parameters can be done
4244     concurrently from multiple threads, calls with the same loop parameter
4245     must be done serially (but can be done from different threads, as long as
4246     only one thread ever is inside a call at any point in time, e.g. by using
4247     a mutex per loop).
4248    
4249     Specifically to support threads (and signal handlers), libev implements
4250     so-called C<ev_async> watchers, which allow some limited form of
4251 root 1.186 concurrency on the same event loop, namely waking it up "from the
4252     outside".
4253 root 1.144
4254 root 1.170 If you want to know which design (one loop, locking, or multiple loops
4255     without or something else still) is best for your problem, then I cannot
4256 root 1.186 help you, but here is some generic advice:
4257 root 1.144
4258     =over 4
4259    
4260     =item * most applications have a main thread: use the default libev loop
4261 root 1.161 in that thread, or create a separate thread running only the default loop.
4262 root 1.144
4263     This helps integrating other libraries or software modules that use libev
4264     themselves and don't care/know about threading.
4265    
4266     =item * one loop per thread is usually a good model.
4267    
4268     Doing this is almost never wrong, sometimes a better-performance model
4269     exists, but it is always a good start.
4270    
4271     =item * other models exist, such as the leader/follower pattern, where one
4272 root 1.161 loop is handed through multiple threads in a kind of round-robin fashion.
4273 root 1.144
4274 root 1.161 Choosing a model is hard - look around, learn, know that usually you can do
4275 root 1.144 better than you currently do :-)
4276    
4277     =item * often you need to talk to some other thread which blocks in the
4278 root 1.182 event loop.
4279 root 1.144
4280 root 1.182 C<ev_async> watchers can be used to wake them up from other threads safely
4281     (or from signal contexts...).
4282    
4283     An example use would be to communicate signals or other events that only
4284     work in the default loop by registering the signal watcher with the
4285     default loop and triggering an C<ev_async> watcher from the default loop
4286     watcher callback into the event loop interested in the signal.
4287 root 1.180
4288 root 1.144 =back
4289    
4290 root 1.253 =head4 THREAD LOCKING EXAMPLE
4291    
4292 root 1.254 Here is a fictitious example of how to run an event loop in a different
4293     thread than where callbacks are being invoked and watchers are
4294     created/added/removed.
4295    
4296     For a real-world example, see the C<EV::Loop::Async> perl module,
4297     which uses exactly this technique (which is suited for many high-level
4298     languages).
4299    
4300     The example uses a pthread mutex to protect the loop data, a condition
4301     variable to wait for callback invocations, an async watcher to notify the
4302     event loop thread and an unspecified mechanism to wake up the main thread.
4303    
4304     First, you need to associate some data with the event loop:
4305    
4306     typedef struct {
4307     mutex_t lock; /* global loop lock */
4308     ev_async async_w;
4309     thread_t tid;
4310     cond_t invoke_cv;
4311     } userdata;
4312    
4313     void prepare_loop (EV_P)
4314     {
4315     // for simplicity, we use a static userdata struct.
4316     static userdata u;
4317    
4318     ev_async_init (&u->async_w, async_cb);
4319     ev_async_start (EV_A_ &u->async_w);
4320    
4321     pthread_mutex_init (&u->lock, 0);
4322     pthread_cond_init (&u->invoke_cv, 0);
4323    
4324     // now associate this with the loop
4325     ev_set_userdata (EV_A_ u);
4326     ev_set_invoke_pending_cb (EV_A_ l_invoke);
4327     ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
4328    
4329     // then create the thread running ev_loop
4330     pthread_create (&u->tid, 0, l_run, EV_A);
4331     }
4332    
4333     The callback for the C<ev_async> watcher does nothing: the watcher is used
4334     solely to wake up the event loop so it takes notice of any new watchers
4335     that might have been added:
4336    
4337     static void
4338     async_cb (EV_P_ ev_async *w, int revents)
4339     {
4340     // just used for the side effects
4341     }
4342    
4343     The C<l_release> and C<l_acquire> callbacks simply unlock/lock the mutex
4344     protecting the loop data, respectively.
4345    
4346     static void
4347     l_release (EV_P)
4348     {
4349 root 1.255 userdata *u = ev_userdata (EV_A);
4350 root 1.254 pthread_mutex_unlock (&u->lock);
4351     }
4352    
4353     static void
4354     l_acquire (EV_P)
4355     {
4356 root 1.255 userdata *u = ev_userdata (EV_A);
4357 root 1.254 pthread_mutex_lock (&u->lock);
4358     }
4359    
4360     The event loop thread first acquires the mutex, and then jumps straight
4361 root 1.310 into C<ev_run>:
4362 root 1.254
4363     void *
4364     l_run (void *thr_arg)
4365     {
4366     struct ev_loop *loop = (struct ev_loop *)thr_arg;
4367    
4368     l_acquire (EV_A);
4369     pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
4370 root 1.310 ev_run (EV_A_ 0);
4371 root 1.254 l_release (EV_A);
4372    
4373     return 0;
4374     }
4375    
4376     Instead of invoking all pending watchers, the C<l_invoke> callback will
4377     signal the main thread via some unspecified mechanism (signals? pipe
4378     writes? C<Async::Interrupt>?) and then waits until all pending watchers
4379 root 1.256 have been called (in a while loop because a) spurious wakeups are possible
4380     and b) skipping inter-thread-communication when there are no pending
4381     watchers is very beneficial):
4382 root 1.254
4383     static void
4384     l_invoke (EV_P)
4385     {
4386 root 1.255 userdata *u = ev_userdata (EV_A);
4387 root 1.254
4388 root 1.256 while (ev_pending_count (EV_A))
4389     {
4390     wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
4391     pthread_cond_wait (&u->invoke_cv, &u->lock);
4392     }
4393 root 1.254 }
4394    
4395     Now, whenever the main thread gets told to invoke pending watchers, it
4396     will grab the lock, call C<ev_invoke_pending> and then signal the loop
4397     thread to continue:
4398    
4399     static void
4400     real_invoke_pending (EV_P)
4401     {
4402 root 1.255 userdata *u = ev_userdata (EV_A);
4403 root 1.254
4404     pthread_mutex_lock (&u->lock);
4405     ev_invoke_pending (EV_A);
4406     pthread_cond_signal (&u->invoke_cv);
4407     pthread_mutex_unlock (&u->lock);
4408     }
4409    
4410     Whenever you want to start/stop a watcher or do other modifications to an
4411     event loop, you will now have to lock:
4412    
4413     ev_timer timeout_watcher;
4414 root 1.255 userdata *u = ev_userdata (EV_A);
4415 root 1.254
4416     ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
4417    
4418     pthread_mutex_lock (&u->lock);
4419     ev_timer_start (EV_A_ &timeout_watcher);
4420     ev_async_send (EV_A_ &u->async_w);
4421     pthread_mutex_unlock (&u->lock);
4422    
4423     Note that sending the C<ev_async> watcher is required because otherwise
4424     an event loop currently blocking in the kernel will have no knowledge
4425     about the newly added timer. By waking up the loop it will pick up any new
4426     watchers in the next event loop iteration.
4427    
4428 root 1.189 =head3 COROUTINES
4429 root 1.144
4430 root 1.191 Libev is very accommodating to coroutines ("cooperative threads"):
4431     libev fully supports nesting calls to its functions from different
4432 root 1.310 coroutines (e.g. you can call C<ev_run> on the same loop from two
4433 root 1.255 different coroutines, and switch freely between both coroutines running
4434     the loop, as long as you don't confuse yourself). The only exception is
4435     that you must not do this from C<ev_periodic> reschedule callbacks.
4436 root 1.144
4437 root 1.181 Care has been taken to ensure that libev does not keep local state inside
4438 root 1.310 C<ev_run>, and other calls do not usually allow for coroutine switches as
4439 root 1.208 they do not call any callbacks.
4440 root 1.144
4441 root 1.189 =head2 COMPILER WARNINGS
4442    
4443     Depending on your compiler and compiler settings, you might get no or a
4444     lot of warnings when compiling libev code. Some people are apparently
4445     scared by this.
4446    
4447     However, these are unavoidable for many reasons. For one, each compiler
4448     has different warnings, and each user has different tastes regarding
4449     warning options. "Warn-free" code therefore cannot be a goal except when
4450     targeting a specific compiler and compiler-version.
4451    
4452     Another reason is that some compiler warnings require elaborate
4453     workarounds, or other changes to the code that make it less clear and less
4454     maintainable.
4455    
4456     And of course, some compiler warnings are just plain stupid, or simply
4457     wrong (because they don't actually warn about the condition their message
4458     seems to warn about). For example, certain older gcc versions had some
4459 sf-exg 1.300 warnings that resulted in an extreme number of false positives. These have
4460 root 1.189 been fixed, but some people still insist on making code warn-free with
4461     such buggy versions.
4462    
4463     While libev is written to generate as few warnings as possible,
4464     "warn-free" code is not a goal, and it is recommended not to build libev
4465     with any compiler warnings enabled unless you are prepared to cope with
4466     them (e.g. by ignoring them). Remember that warnings are just that:
4467     warnings, not errors, or proof of bugs.
4468    
4469    
4470 root 1.190 =head2 VALGRIND
4471 root 1.189
4472     Valgrind has a special section here because it is a popular tool that is
4473     highly useful. Unfortunately, valgrind reports are very hard to interpret.
4474    
4475     If you think you found a bug (memory leak, uninitialised data access etc.)
4476     in libev, then check twice: If valgrind reports something like:
4477    
4478     ==2274== definitely lost: 0 bytes in 0 blocks.
4479     ==2274== possibly lost: 0 bytes in 0 blocks.
4480     ==2274== still reachable: 256 bytes in 1 blocks.
4481    
4482     Then there is no memory leak, just as memory accounted to global variables
4483 root 1.208 is not a memleak - the memory is still being referenced, and didn't leak.
4484 root 1.189
4485     Similarly, under some circumstances, valgrind might report kernel bugs
4486     as if it were a bug in libev (e.g. in realloc or in the poll backend,
4487     although an acceptable workaround has been found here), or it might be
4488     confused.
4489    
4490     Keep in mind that valgrind is a very good tool, but only a tool. Don't
4491     make it into some kind of religion.
4492    
4493     If you are unsure about something, feel free to contact the mailing list
4494     with the full valgrind report and an explanation on why you think this
4495     is a bug in libev (best check the archives, too :). However, don't be
4496     annoyed when you get a brisk "this is no bug" answer and take the chance
4497     of learning how to interpret valgrind properly.
4498    
4499     If you need, for some reason, empty reports from valgrind for your project
4500     I suggest using suppression lists.
4501    
4502    
4503 root 1.190 =head1 PORTABILITY NOTES
4504 root 1.189
4505 root 1.302 =head2 GNU/LINUX 32 BIT LIMITATIONS
4506    
4507     GNU/Linux is the only common platform that supports 64 bit file/large file
4508 root 1.303 interfaces but I<disables> them by default.
4509 root 1.302
4510     That means that libev compiled in the default environment doesn't support
4511 root 1.303 files larger than 2GiB or so, which mainly affects C<ev_stat> watchers.
4512 root 1.302
4513     Unfortunately, many programs try to work around this GNU/Linux issue
4514     by enabling the large file API, which makes them incompatible with the
4515     standard libev compiled for their system.
4516    
4517     Likewise, libev cannot enable the large file API itself as this would
4518     suddenly make it incompatible to the default compile time environment,
4519     i.e. all programs not using special compile switches.
4520    
4521     =head2 OS/X AND DARWIN BUGS
4522    
4523     The whole thing is a bug if you ask me - basically any system interface
4524 root 1.303 you touch is broken, whether it is locales, poll, kqueue or even the
4525 root 1.302 OpenGL drivers.
4526    
4527 root 1.303 =head3 C<kqueue> is buggy
4528 root 1.302
4529     The kqueue syscall is broken in all known versions - most versions support
4530     only sockets, many support pipes.
4531    
4532 root 1.304 Libev tries to work around this by not using C<kqueue> by default on
4533     this rotten platform, but of course you can still ask for it when creating
4534     a loop.
4535    
4536 root 1.303 =head3 C<poll> is buggy
4537 root 1.302
4538     Instead of fixing C<kqueue>, Apple replaced their (working) C<poll>
4539     implementation by something calling C<kqueue> internally around the 10.5.6
4540     release, so now C<kqueue> I<and> C<poll> are broken.
4541    
4542 root 1.304 Libev tries to work around this by not using C<poll> by default on
4543     this rotten platform, but of course you can still ask for it when creating
4544     a loop.
4545 root 1.302
4546 root 1.303 =head3 C<select> is buggy
4547 root 1.302
4548     All that's left is C<select>, and of course Apple found a way to fuck this
4549     one up as well: On OS/X, C<select> actively limits the number of file
4550 root 1.305 descriptors you can pass in to 1024 - your program suddenly crashes when
4551 root 1.302 you use more.
4552    
4553     There is an undocumented "workaround" for this - defining
4554     C<_DARWIN_UNLIMITED_SELECT>, which libev tries to use, so select I<should>
4555     work on OS/X.
4556    
4557     =head2 SOLARIS PROBLEMS AND WORKAROUNDS
4558    
4559 root 1.303 =head3 C<errno> reentrancy
4560 root 1.302
4561     The default compile environment on Solaris is unfortunately so
4562     thread-unsafe that you can't even use components/libraries compiled
4563     without C<-D_REENTRANT> (as long as they use C<errno>), which, of course,
4564     isn't defined by default.
4565    
4566     If you want to use libev in threaded environments you have to make sure
4567     it's compiled with C<_REENTRANT> defined.
4568    
4569 root 1.303 =head3 Event port backend
4570 root 1.302
4571     The scalable event interface for Solaris is called "event ports". Unfortunately,
4572     this mechanism is very buggy. If you run into high CPU usage, your program
4573     freezes or you get a large number of spurious wakeups, make sure you have
4574     all the relevant and latest kernel patches applied. No, I don't know which
4575     ones, but there are multiple ones.
4576    
4577 root 1.305 If you can't get it to work, you can try running the program by setting
4578     the environment variable C<LIBEV_FLAGS=3> to only allow C<poll> and
4579     C<select> backends.
4580 root 1.302
4581     =head2 AIX POLL BUG
4582    
4583     AIX unfortunately has a broken C<poll.h> header. Libev works around
4584     this by trying to avoid the poll backend altogether (i.e. it's not even
4585     compiled in), which normally isn't a big problem as C<select> works fine
4586     with large bitsets, and AIX is dead anyway.
4587    
4588 root 1.189 =head2 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
4589 root 1.112
4590 root 1.303 =head3 General issues
4591    
4592 root 1.112 Win32 doesn't support any of the standards (e.g. POSIX) that libev
4593     requires, and its I/O model is fundamentally incompatible with the POSIX
4594     model. Libev still offers limited functionality on this platform in
4595     the form of the C<EVBACKEND_SELECT> backend, and only supports socket
4596     descriptors. This only applies when using Win32 natively, not when using
4597 root 1.303 e.g. cygwin. Actually, it only applies to the microsofts own compilers,
4598     as every compielr comes with a slightly differently broken/incompatible
4599     environment.
4600 root 1.112
4601 root 1.150 Lifting these limitations would basically require the full
4602 root 1.303 re-implementation of the I/O system. If you are into this kind of thing,
4603     then note that glib does exactly that for you in a very portable way (note
4604     also that glib is the slowest event library known to man).
4605 root 1.150
4606 root 1.112 There is no supported compilation method available on windows except
4607     embedding it into other applications.
4608    
4609 root 1.241 Sensible signal handling is officially unsupported by Microsoft - libev
4610     tries its best, but under most conditions, signals will simply not work.
4611    
4612 root 1.162 Not a libev limitation but worth mentioning: windows apparently doesn't
4613     accept large writes: instead of resulting in a partial write, windows will
4614     either accept everything or return C<ENOBUFS> if the buffer is too large,
4615     so make sure you only write small amounts into your sockets (less than a
4616 root 1.184 megabyte seems safe, but this apparently depends on the amount of memory
4617 root 1.162 available).
4618    
4619 root 1.150 Due to the many, low, and arbitrary limits on the win32 platform and
4620     the abysmal performance of winsockets, using a large number of sockets
4621     is not recommended (and not reasonable). If your program needs to use
4622     more than a hundred or so sockets, then likely it needs to use a totally
4623 root 1.155 different implementation for windows, as libev offers the POSIX readiness
4624 root 1.150 notification model, which cannot be implemented efficiently on windows
4625 root 1.241 (due to Microsoft monopoly games).
4626 root 1.112
4627 root 1.167 A typical way to use libev under windows is to embed it (see the embedding
4628     section for details) and use the following F<evwrap.h> header file instead
4629     of F<ev.h>:
4630    
4631     #define EV_STANDALONE /* keeps ev from requiring config.h */
4632     #define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
4633    
4634     #include "ev.h"
4635    
4636     And compile the following F<evwrap.c> file into your project (make sure
4637 root 1.184 you do I<not> compile the F<ev.c> or any other embedded source files!):
4638 root 1.167
4639     #include "evwrap.h"
4640     #include "ev.c"
4641    
4642 root 1.303 =head3 The winsocket C<select> function
4643 root 1.112
4644 root 1.160 The winsocket C<select> function doesn't follow POSIX in that it
4645     requires socket I<handles> and not socket I<file descriptors> (it is
4646     also extremely buggy). This makes select very inefficient, and also
4647 root 1.167 requires a mapping from file descriptors to socket handles (the Microsoft
4648     C runtime provides the function C<_open_osfhandle> for this). See the
4649 root 1.160 discussion of the C<EV_SELECT_USE_FD_SET>, C<EV_SELECT_IS_WINSOCKET> and
4650     C<EV_FD_TO_WIN32_HANDLE> preprocessor symbols for more info.
4651 root 1.112
4652 root 1.161 The configuration for a "naked" win32 using the Microsoft runtime
4653 root 1.112 libraries and raw winsocket select is:
4654    
4655 root 1.164 #define EV_USE_SELECT 1
4656     #define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
4657 root 1.112
4658     Note that winsockets handling of fd sets is O(n), so you can easily get a
4659     complexity in the O(n²) range when using win32.
4660    
4661 root 1.303 =head3 Limited number of file descriptors
4662 root 1.112
4663 root 1.150 Windows has numerous arbitrary (and low) limits on things.
4664    
4665     Early versions of winsocket's select only supported waiting for a maximum
4666     of C<64> handles (probably owning to the fact that all windows kernels
4667 root 1.161 can only wait for C<64> things at the same time internally; Microsoft
4668 root 1.150 recommends spawning a chain of threads and wait for 63 handles and the
4669 root 1.241 previous thread in each. Sounds great!).
4670 root 1.112
4671     Newer versions support more handles, but you need to define C<FD_SETSIZE>
4672     to some high number (e.g. C<2048>) before compiling the winsocket select
4673 root 1.241 call (which might be in libev or elsewhere, for example, perl and many
4674     other interpreters do their own select emulation on windows).
4675 root 1.112
4676 root 1.161 Another limit is the number of file descriptors in the Microsoft runtime
4677 root 1.241 libraries, which by default is C<64> (there must be a hidden I<64>
4678     fetish or something like this inside Microsoft). You can increase this
4679     by calling C<_setmaxstdio>, which can increase this limit to C<2048>
4680     (another arbitrary limit), but is broken in many versions of the Microsoft
4681     runtime libraries. This might get you to about C<512> or C<2048> sockets
4682     (depending on windows version and/or the phase of the moon). To get more,
4683     you need to wrap all I/O functions and provide your own fd management, but
4684     the cost of calling select (O(n²)) will likely make this unworkable.
4685 root 1.112
4686 root 1.189 =head2 PORTABILITY REQUIREMENTS
4687 root 1.112
4688 root 1.189 In addition to a working ISO-C implementation and of course the
4689     backend-specific APIs, libev relies on a few additional extensions:
4690 root 1.148
4691     =over 4
4692    
4693 root 1.165 =item C<void (*)(ev_watcher_type *, int revents)> must have compatible
4694     calling conventions regardless of C<ev_watcher_type *>.
4695    
4696     Libev assumes not only that all watcher pointers have the same internal
4697     structure (guaranteed by POSIX but not by ISO C for example), but it also
4698     assumes that the same (machine) code can be used to call any watcher
4699     callback: The watcher callbacks have different type signatures, but libev
4700     calls them using an C<ev_watcher *> internally.
4701    
4702 root 1.148 =item C<sig_atomic_t volatile> must be thread-atomic as well
4703    
4704     The type C<sig_atomic_t volatile> (or whatever is defined as
4705 root 1.184 C<EV_ATOMIC_T>) must be atomic with respect to accesses from different
4706 root 1.148 threads. This is not part of the specification for C<sig_atomic_t>, but is
4707     believed to be sufficiently portable.
4708    
4709     =item C<sigprocmask> must work in a threaded environment
4710    
4711     Libev uses C<sigprocmask> to temporarily block signals. This is not
4712     allowed in a threaded program (C<pthread_sigmask> has to be used). Typical
4713     pthread implementations will either allow C<sigprocmask> in the "main
4714     thread" or will block signals process-wide, both behaviours would
4715     be compatible with libev. Interaction between C<sigprocmask> and
4716     C<pthread_sigmask> could complicate things, however.
4717    
4718     The most portable way to handle signals is to block signals in all threads
4719     except the initial one, and run the default loop in the initial thread as
4720     well.
4721    
4722 root 1.150 =item C<long> must be large enough for common memory allocation sizes
4723    
4724 root 1.189 To improve portability and simplify its API, libev uses C<long> internally
4725     instead of C<size_t> when allocating its data structures. On non-POSIX
4726     systems (Microsoft...) this might be unexpectedly low, but is still at
4727     least 31 bits everywhere, which is enough for hundreds of millions of
4728     watchers.
4729 root 1.150
4730     =item C<double> must hold a time value in seconds with enough accuracy
4731    
4732 root 1.151 The type C<double> is used to represent timestamps. It is required to
4733 root 1.308 have at least 51 bits of mantissa (and 9 bits of exponent), which is
4734     good enough for at least into the year 4000 with millisecond accuracy
4735     (the design goal for libev). This requirement is overfulfilled by
4736     implementations using IEEE 754, which is basically all existing ones. With
4737     IEEE 754 doubles, you get microsecond accuracy until at least 2200.
4738 root 1.150
4739 root 1.148 =back
4740    
4741     If you know of other additional requirements drop me a note.
4742    
4743    
4744 root 1.191 =head1 ALGORITHMIC COMPLEXITIES
4745    
4746     In this section the complexities of (many of) the algorithms used inside
4747     libev will be documented. For complexity discussions about backends see
4748     the documentation for C<ev_default_init>.
4749    
4750     All of the following are about amortised time: If an array needs to be
4751     extended, libev needs to realloc and move the whole array, but this
4752     happens asymptotically rarer with higher number of elements, so O(1) might
4753     mean that libev does a lengthy realloc operation in rare cases, but on
4754     average it is much faster and asymptotically approaches constant time.
4755    
4756     =over 4
4757    
4758     =item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
4759    
4760     This means that, when you have a watcher that triggers in one hour and
4761     there are 100 watchers that would trigger before that, then inserting will
4762     have to skip roughly seven (C<ld 100>) of these watchers.
4763    
4764     =item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
4765    
4766     That means that changing a timer costs less than removing/adding them,
4767     as only the relative motion in the event queue has to be paid for.
4768    
4769     =item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
4770    
4771     These just add the watcher into an array or at the head of a list.
4772    
4773     =item Stopping check/prepare/idle/fork/async watchers: O(1)
4774    
4775     =item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
4776    
4777     These watchers are stored in lists, so they need to be walked to find the
4778     correct watcher to remove. The lists are usually short (you don't usually
4779     have many watchers waiting for the same fd or signal: one is typical, two
4780     is rare).
4781    
4782     =item Finding the next timer in each loop iteration: O(1)
4783    
4784     By virtue of using a binary or 4-heap, the next timer is always found at a
4785     fixed position in the storage array.
4786    
4787     =item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
4788    
4789     A change means an I/O watcher gets started or stopped, which requires
4790     libev to recalculate its status (and possibly tell the kernel, depending
4791     on backend and whether C<ev_io_set> was used).
4792    
4793     =item Activating one watcher (putting it into the pending state): O(1)
4794    
4795     =item Priority handling: O(number_of_priorities)
4796    
4797     Priorities are implemented by allocating some space for each
4798     priority. When doing priority-based operations, libev usually has to
4799     linearly search all the priorities, but starting/stopping and activating
4800     watchers becomes O(1) with respect to priority handling.
4801    
4802     =item Sending an ev_async: O(1)
4803    
4804     =item Processing ev_async_send: O(number_of_async_watchers)
4805    
4806     =item Processing signals: O(max_signal_number)
4807    
4808     Sending involves a system call I<iff> there were no other C<ev_async_send>
4809     calls in the current loop iteration. Checking for async and signal events
4810     involves iterating over all running async watchers or all signal numbers.
4811    
4812     =back
4813    
4814    
4815 root 1.291 =head1 PORTING FROM LIBEV 3.X TO 4.X
4816 root 1.289
4817     The major version 4 introduced some minor incompatible changes to the API.
4818    
4819 root 1.291 At the moment, the C<ev.h> header file tries to implement superficial
4820     compatibility, so most programs should still compile. Those might be
4821     removed in later versions of libev, so better update early than late.
4822    
4823 root 1.289 =over 4
4824    
4825 root 1.310 =item function/symbol renames
4826    
4827     A number of functions and symbols have been renamed:
4828    
4829     ev_loop => ev_run
4830     EVLOOP_NONBLOCK => EVRUN_NOWAIT
4831     EVLOOP_ONESHOT => EVRUN_ONCE
4832    
4833     ev_unloop => ev_break
4834     EVUNLOOP_CANCEL => EVBREAK_CANCEL
4835     EVUNLOOP_ONE => EVBREAK_ONE
4836     EVUNLOOP_ALL => EVBREAK_ALL
4837 root 1.291
4838 root 1.310 EV_TIMEOUT => EV_TIMER
4839 root 1.291
4840 root 1.310 ev_loop_count => ev_iteration
4841     ev_loop_depth => ev_depth
4842     ev_loop_verify => ev_verify
4843 root 1.291
4844     Most functions working on C<struct ev_loop> objects don't have an
4845 root 1.310 C<ev_loop_> prefix, so it was removed; C<ev_loop>, C<ev_unloop> and
4846     associated constants have been renamed to not collide with the C<struct
4847     ev_loop> anymore and C<EV_TIMER> now follows the same naming scheme
4848     as all other watcher types. Note that C<ev_loop_fork> is still called
4849     C<ev_loop_fork> because it would otherwise clash with the C<ev_fork>
4850     typedef.
4851    
4852     =item C<EV_COMPAT3> backwards compatibility mechanism
4853    
4854     The backward compatibility mechanism can be controlled by
4855     C<EV_COMPAT3>. See L<PREPROCESSOR SYMBOLS/MACROS> in the L<EMBEDDING>
4856     section.
4857 root 1.289
4858     =item C<EV_MINIMAL> mechanism replaced by C<EV_FEATURES>
4859    
4860     The preprocessor symbol C<EV_MINIMAL> has been replaced by a different
4861     mechanism, C<EV_FEATURES>. Programs using C<EV_MINIMAL> usually compile
4862     and work, but the library code will of course be larger.
4863    
4864     =back
4865    
4866    
4867 root 1.234 =head1 GLOSSARY
4868    
4869     =over 4
4870    
4871     =item active
4872    
4873     A watcher is active as long as it has been started (has been attached to
4874     an event loop) but not yet stopped (disassociated from the event loop).
4875    
4876     =item application
4877    
4878     In this document, an application is whatever is using libev.
4879    
4880     =item callback
4881    
4882     The address of a function that is called when some event has been
4883     detected. Callbacks are being passed the event loop, the watcher that
4884     received the event, and the actual event bitset.
4885    
4886     =item callback invocation
4887    
4888     The act of calling the callback associated with a watcher.
4889    
4890     =item event
4891    
4892     A change of state of some external event, such as data now being available
4893     for reading on a file descriptor, time having passed or simply not having
4894     any other events happening anymore.
4895    
4896     In libev, events are represented as single bits (such as C<EV_READ> or
4897 root 1.289 C<EV_TIMER>).
4898 root 1.234
4899     =item event library
4900    
4901     A software package implementing an event model and loop.
4902    
4903     =item event loop
4904    
4905     An entity that handles and processes external events and converts them
4906     into callback invocations.
4907    
4908     =item event model
4909    
4910     The model used to describe how an event loop handles and processes
4911     watchers and events.
4912    
4913     =item pending
4914    
4915     A watcher is pending as soon as the corresponding event has been detected,
4916     and stops being pending as soon as the watcher will be invoked or its
4917     pending status is explicitly cleared by the application.
4918    
4919     A watcher can be pending, but not active. Stopping a watcher also clears
4920     its pending status.
4921    
4922     =item real time
4923    
4924     The physical time that is observed. It is apparently strictly monotonic :)
4925    
4926     =item wall-clock time
4927    
4928     The time and date as shown on clocks. Unlike real time, it can actually
4929     be wrong and jump forwards and backwards, e.g. when the you adjust your
4930     clock.
4931    
4932     =item watcher
4933    
4934     A data structure that describes interest in certain events. Watchers need
4935     to be started (attached to an event loop) before they can receive events.
4936    
4937     =item watcher invocation
4938    
4939     The act of calling the callback associated with a watcher.
4940    
4941     =back
4942    
4943 root 1.1 =head1 AUTHOR
4944    
4945 root 1.209 Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael Magnusson.
4946 root 1.1