ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libev/ev.pod

Comparing libev/ev.pod (file contents):
Revision 1.16 by root, Mon Nov 12 08:47:14 2007 UTC vs.
Revision 1.450 by root, Mon Jun 24 00:04:26 2019 UTC

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

Diff Legend

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