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 1.338 by root, Sun Oct 31 21:16:26 2010 UTC

…		…
2		2
3	libev - a high performance full-featured event loop written in C	3	libev - a high performance full-featured event loop written in C
4		4
5	=head1 SYNOPSIS	5	=head1 SYNOPSIS
6		6
7	#include <ev.h>	7	#include <ev.h>
8		8
9	=head1 DESCRIPTION	9	=head2 EXAMPLE PROGRAM
		10
		11	// a single header file is required
		12	#include <ev.h>
		13
		14	#include <stdio.h> // for puts
		15
		16	// every watcher type has its own typedef'd struct
		17	// with the name ev_TYPE
		18	ev_io stdin_watcher;
		19	ev_timer timeout_watcher;
		20
		21	// all watcher callbacks have a similar signature
		22	// this callback is called when data is readable on stdin
		23	static void
		24	stdin_cb (EV_P_ ev_io *w, int revents)
		25	{
		26	puts ("stdin ready");
		27	// for one-shot events, one must manually stop the watcher
		28	// with its corresponding stop function.
		29	ev_io_stop (EV_A_ w);
		30
		31	// this causes all nested ev_run's to stop iterating
		32	ev_break (EV_A_ EVBREAK_ALL);
		33	}
		34
		35	// another callback, this time for a time-out
		36	static void
		37	timeout_cb (EV_P_ ev_timer *w, int revents)
		38	{
		39	puts ("timeout");
		40	// this causes the innermost ev_run to stop iterating
		41	ev_break (EV_A_ EVBREAK_ONE);
		42	}
		43
		44	int
		45	main (void)
		46	{
		47	// use the default event loop unless you have special needs
		48	struct ev_loop *loop = EV_DEFAULT;
		49
		50	// initialise an io watcher, then start it
		51	// this one will watch for stdin to become readable
		52	ev_io_init (&stdin_watcher, stdin_cb, /*STDIN_FILENO*/ 0, EV_READ);
		53	ev_io_start (loop, &stdin_watcher);
		54
		55	// initialise a timer watcher, then start it
		56	// simple non-repeating 5.5 second timeout
		57	ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
		58	ev_timer_start (loop, &timeout_watcher);
		59
		60	// now wait for events to arrive
		61	ev_run (loop, 0);
		62
		63	// unloop was called, so exit
		64	return 0;
		65	}
		66
		67	=head1 ABOUT THIS DOCUMENT
		68
		69	This document documents the libev software package.
		70
		71	The newest version of this document is also available as an html-formatted
		72	web page you might find easier to navigate when reading it for the first
		73	time: L<http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod>.
		74
		75	While this document tries to be as complete as possible in documenting
		76	libev, its usage and the rationale behind its design, it is not a tutorial
		77	on event-based programming, nor will it introduce event-based programming
		78	with libev.
		79
		80	Familiarity with event based programming techniques in general is assumed
		81	throughout this document.
		82
		83	=head1 WHAT TO READ WHEN IN A HURRY
		84
		85	This manual tries to be very detailed, but unfortunately, this also makes
		86	it very long. If you just want to know the basics of libev, I suggest
		87	reading L<ANATOMY OF A WATCHER>, then the L<EXAMPLE PROGRAM> above and
		88	look up the missing functions in L<GLOBAL FUNCTIONS> and the C<ev_io> and
		89	C<ev_timer> sections in L<WATCHER TYPES>.
		90
		91	=head1 ABOUT LIBEV
10		92
11	Libev is an event loop: you register interest in certain events (such as a	93	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	94	file descriptor being readable or a timeout occurring), and it will manage
13	these event sources and provide your program with events.	95	these event sources and provide your program with events.
14		96
15	To do this, it must take more or less complete control over your process	97	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	98	(or thread) by executing the I<event loop> handler, and will then
17	communicate events via a callback mechanism.	99	communicate events via a callback mechanism.
…		…
19	You register interest in certain events by registering so-called I<event	101	You register interest in certain events by registering so-called I<event
20	watchers>, which are relatively small C structures you initialise with the	102	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	103	details of the event, and then hand it over to libev by I<starting> the
22	watcher.	104	watcher.
23		105
24	=head1 FEATURES	106	=head2 FEATURES
25		107
26	Libev supports select, poll, the linux-specific epoll and the bsd-specific	108	Libev supports C<select>, C<poll>, the Linux-specific C<epoll>, the
27	kqueue mechanisms for file descriptor events, relative timers, absolute	109	BSD-specific C<kqueue> and the Solaris-specific event port mechanisms
28	timers with customised rescheduling, signal events, process status change	110	for file descriptor events (C<ev_io>), the Linux C<inotify> interface
		111	(for C<ev_stat>), Linux eventfd/signalfd (for faster and cleaner
		112	inter-thread wakeup (C<ev_async>)/signal handling (C<ev_signal>)) relative
		113	timers (C<ev_timer>), absolute timers with customised rescheduling
		114	(C<ev_periodic>), synchronous signals (C<ev_signal>), process status
29	events (related to SIGCHLD), and event watchers dealing with the event	115	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	116	loop mechanism itself (C<ev_idle>, C<ev_embed>, C<ev_prepare> and
		117	C<ev_check> watchers) as well as file watchers (C<ev_stat>) and even
		118	limited support for fork events (C<ev_fork>).
		119
		120	It also is quite fast (see this
31	fast (see this L<benchmark|http://libev.schmorp.de/bench.html> comparing	121	L<benchmark|http://libev.schmorp.de/bench.html> comparing it to libevent
32	it to libevent for example).	122	for example).
33		123
34	=head1 CONVENTIONS	124	=head2 CONVENTIONS
35		125
36	Libev is very configurable. In this manual the default configuration	126	Libev is very configurable. In this manual the default (and most common)
37	will be described, which supports multiple event loops. For more info	127	configuration will be described, which supports multiple event loops. For
38	about various configuration options please have a look at the file	128	more info about various configuration options please have a look at
39	F<README.embed> in the libev distribution. If libev was configured without	129	B<EMBED> section in this manual. If libev was configured without support
40	support for multiple event loops, then all functions taking an initial	130	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 *>)	131	name C<loop> (which is always of type C<struct ev_loop *>) will not have
42	will not have this argument.	132	this argument.
43		133
44	=head1 TIME AND OTHER GLOBAL FUNCTIONS	134	=head2 TIME REPRESENTATION
45		135
46	Libev represents time as a single floating point number, representing the	136	Libev represents time as a single floating point number, representing
47	(fractional) number of seconds since the (POSIX) epoch (somewhere near	137	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	138	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	139	ask). This type is called C<ev_tstamp>, which is what you should use
50	to the double type in C.	140	too. It usually aliases to the C<double> type in C. When you need to do
		141	any calculations on it, you should treat it as some floating point value.
		142
		143	Unlike the name component C<stamp> might indicate, it is also used for
		144	time differences (e.g. delays) throughout libev.
		145
		146	=head1 ERROR HANDLING
		147
		148	Libev knows three classes of errors: operating system errors, usage errors
		149	and internal errors (bugs).
		150
		151	When libev catches an operating system error it cannot handle (for example
		152	a system call indicating a condition libev cannot fix), it calls the callback
		153	set via C<ev_set_syserr_cb>, which is supposed to fix the problem or
		154	abort. The default is to print a diagnostic message and to call C<abort
		155	()>.
		156
		157	When libev detects a usage error such as a negative timer interval, then
		158	it will print a diagnostic message and abort (via the C<assert> mechanism,
		159	so C<NDEBUG> will disable this checking): these are programming errors in
		160	the libev caller and need to be fixed there.
		161
		162	Libev also has a few internal error-checking C<assert>ions, and also has
		163	extensive consistency checking code. These do not trigger under normal
		164	circumstances, as they indicate either a bug in libev or worse.
		165
		166
		167	=head1 GLOBAL FUNCTIONS
		168
		169	These functions can be called anytime, even before initialising the
		170	library in any way.
51		171
52	=over 4	172	=over 4
53		173
54	=item ev_tstamp ev_time ()	174	=item ev_tstamp ev_time ()
55		175
56	Returns the current time as libev would use it.	176	Returns the current time as libev would use it. Please note that the
		177	C<ev_now> function is usually faster and also often returns the timestamp
		178	you actually want to know. Also interesting is the combination of
		179	C<ev_update_now> and C<ev_now>.
		180
		181	=item ev_sleep (ev_tstamp interval)
		182
		183	Sleep for the given interval: The current thread will be blocked until
		184	either it is interrupted or the given time interval has passed. Basically
		185	this is a sub-second-resolution C<sleep ()>.
57		186
58	=item int ev_version_major ()	187	=item int ev_version_major ()
59		188
60	=item int ev_version_minor ()	189	=item int ev_version_minor ()
61		190
62	You can find out the major and minor version numbers of the library	191	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	192	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	193	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	194	symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the
66	version of the library your program was compiled against.	195	version of the library your program was compiled against.
67		196
		197	These version numbers refer to the ABI version of the library, not the
		198	release version.
		199
68	Usually, it's a good idea to terminate if the major versions mismatch,	200	Usually, it's a good idea to terminate if the major versions mismatch,
69	as this indicates an incompatible change. Minor versions are usually	201	as this indicates an incompatible change. Minor versions are usually
70	compatible to older versions, so a larger minor version alone is usually	202	compatible to older versions, so a larger minor version alone is usually
71	not a problem.	203	not a problem.
72		204
		205	Example: Make sure we haven't accidentally been linked against the wrong
		206	version (note, however, that this will not detect other ABI mismatches,
		207	such as LFS or reentrancy).
		208
		209	assert (("libev version mismatch",
		210	ev_version_major () == EV_VERSION_MAJOR
		211	&& ev_version_minor () >= EV_VERSION_MINOR));
		212
		213	=item unsigned int ev_supported_backends ()
		214
		215	Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
		216	value) compiled into this binary of libev (independent of their
		217	availability on the system you are running on). See C<ev_default_loop> for
		218	a description of the set values.
		219
		220	Example: make sure we have the epoll method, because yeah this is cool and
		221	a must have and can we have a torrent of it please!!!11
		222
		223	assert (("sorry, no epoll, no sex",
		224	ev_supported_backends () & EVBACKEND_EPOLL));
		225
		226	=item unsigned int ev_recommended_backends ()
		227
		228	Return the set of all backends compiled into this binary of libev and
		229	also recommended for this platform, meaning it will work for most file
		230	descriptor types. This set is often smaller than the one returned by
		231	C<ev_supported_backends>, as for example kqueue is broken on most BSDs
		232	and will not be auto-detected unless you explicitly request it (assuming
		233	you know what you are doing). This is the set of backends that libev will
		234	probe for if you specify no backends explicitly.
		235
		236	=item unsigned int ev_embeddable_backends ()
		237
		238	Returns the set of backends that are embeddable in other event loops. This
		239	value is platform-specific but can include backends not available on the
		240	current system. To find which embeddable backends might be supported on
		241	the current system, you would need to look at C<ev_embeddable_backends ()
		242	& ev_supported_backends ()>, likewise for recommended ones.
		243
		244	See the description of C<ev_embed> watchers for more info.
		245
73	=item ev_set_allocator (void *(*cb)(void *ptr, long size))	246	=item ev_set_allocator (void *(*cb)(void *ptr, long size)) [NOT REENTRANT]
74		247
75	Sets the allocation function to use (the prototype is similar to the	248	Sets the allocation function to use (the prototype is similar - the
76	realloc C function, the semantics are identical). It is used to allocate	249	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	250	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	251	when memory needs to be allocated (C<size != 0>), the library might abort
79	destructive action. The default is your system realloc function.	252	or take some potentially destructive action.
		253
		254	Since some systems (at least OpenBSD and Darwin) fail to implement
		255	correct C<realloc> semantics, libev will use a wrapper around the system
		256	C<realloc> and C<free> functions by default.
80		257
81	You could override this function in high-availability programs to, say,	258	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,	259	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.	260	or even to sleep a while and retry until some memory is available.
84		261
		262	Example: Replace the libev allocator with one that waits a bit and then
		263	retries (example requires a standards-compliant C<realloc>).
		264
		265	static void *
		266	persistent_realloc (void *ptr, size_t size)
		267	{
		268	for (;;)
		269	{
		270	void *newptr = realloc (ptr, size);
		271
		272	if (newptr)
		273	return newptr;
		274
		275	sleep (60);
		276	}
		277	}
		278
		279	...
		280	ev_set_allocator (persistent_realloc);
		281
85	=item ev_set_syserr_cb (void (*cb)(const char *msg));	282	=item ev_set_syserr_cb (void (*cb)(const char *msg)); [NOT REENTRANT]
86		283
87	Set the callback function to call on a retryable syscall error (such	284	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	285	as failed select, poll, epoll_wait). The message is a printable string
89	indicating the system call or subsystem causing the problem. If this	286	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	287	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	288	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	289	requested operation, or, if the condition doesn't go away, do bad stuff
93	(such as abort).	290	(such as abort).
94		291
		292	Example: This is basically the same thing that libev does internally, too.
		293
		294	static void
		295	fatal_error (const char *msg)
		296	{
		297	perror (msg);
		298	abort ();
		299	}
		300
		301	...
		302	ev_set_syserr_cb (fatal_error);
		303
95	=back	304	=back
96		305
97	=head1 FUNCTIONS CONTROLLING THE EVENT LOOP	306	=head1 FUNCTIONS CONTROLLING EVENT LOOPS
98		307
99	An event loop is described by a C<struct ev_loop *>. The library knows two	308	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	309	I<not> optional in this case unless libev 3 compatibility is disabled, as
101	events, and dynamically created loops which do not.	310	libev 3 had an C<ev_loop> function colliding with the struct name).
102		311
103	If you use threads, a common model is to run the default event loop	312	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	313	supports child process events, and dynamically created event loops which
105	create, you also create another event loop. Libev itself does no locking	314	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		315
110	=over 4	316	=over 4
111		317
112	=item struct ev_loop *ev_default_loop (unsigned int flags)	318	=item struct ev_loop *ev_default_loop (unsigned int flags)
113		319
114	This will initialise the default event loop if it hasn't been initialised	320	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	321	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	322	the C<flags> parameter are described in more detail in the entry for
117	flags).	323	C<ev_loop_new>.
		324
		325	If the default loop is already initialised then this function simply
		326	returns it (and ignores the flags. If that is troubling you, check
		327	C<ev_backend ()> afterwards). Otherwise it will create it with the given
		328	flags, which should almost always be C<0>, unless the caller is also the
		329	one calling C<ev_run> or otherwise qualifies as "the main program".
118		330
119	If you don't know what event loop to use, use the one returned from this	331	If you don't know what event loop to use, use the one returned from this
120	function.	332	function (or via the C<EV_DEFAULT> macro).
		333
		334	Note that this function is I<not> thread-safe, so if you want to use it
		335	from multiple threads, you have to employ some kind of mutex (note also
		336	that this case is unlikely, as loops cannot be shared easily between
		337	threads anyway).
		338
		339	The default loop is the only loop that can handle C<ev_child> watchers,
		340	and to do this, it always registers a handler for C<SIGCHLD>. If this is
		341	a problem for your application you can either create a dynamic loop with
		342	C<ev_loop_new> which doesn't do that, or you can simply overwrite the
		343	C<SIGCHLD> signal handler I<after> calling C<ev_default_init>.
		344
		345	Example: This is the most typical usage.
		346
		347	if (!ev_default_loop (0))
		348	fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
		349
		350	Example: Restrict libev to the select and poll backends, and do not allow
		351	environment settings to be taken into account:
		352
		353	ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
		354
		355	=item struct ev_loop *ev_loop_new (unsigned int flags)
		356
		357	This will create and initialise a new event loop object. If the loop
		358	could not be initialised, returns false.
		359
		360	Note that this function I<is> thread-safe, and one common way to use
		361	libev with threads is indeed to create one loop per thread, and using the
		362	default loop in the "main" or "initial" thread.
121		363
122	The flags argument can be used to specify special behaviour or specific	364	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).	365	backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
124		366
125	It supports the following flags:	367	The following flags are supported:
126		368
127	=over 4	369	=over 4
128		370
129	=item C<EVFLAG_AUTO>	371	=item C<EVFLAG_AUTO>
130		372
131	The default flags value. Use this if you have no clue (it's the right	373	The default flags value. Use this if you have no clue (it's the right
132	thing, believe me).	374	thing, believe me).
133		375
134	=item C<EVFLAG_NOENV>	376	=item C<EVFLAG_NOENV>
135		377
136	If this flag bit is ored into the flag value (or the program runs setuid	378	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	379	or setgid) then libev will I<not> look at the environment variable
138	C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will	380	C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
139	override the flags completely if it is found in the environment. This is	381	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	382	useful to try out specific backends to test their performance, or to work
141	around bugs.	383	around bugs.
142		384
		385	=item C<EVFLAG_FORKCHECK>
		386
		387	Instead of calling C<ev_loop_fork> manually after a fork, you can also
		388	make libev check for a fork in each iteration by enabling this flag.
		389
		390	This works by calling C<getpid ()> on every iteration of the loop,
		391	and thus this might slow down your event loop if you do a lot of loop
		392	iterations and little real work, but is usually not noticeable (on my
		393	GNU/Linux system for example, C<getpid> is actually a simple 5-insn sequence
		394	without a system call and thus I<very> fast, but my GNU/Linux system also has
		395	C<pthread_atfork> which is even faster).
		396
		397	The big advantage of this flag is that you can forget about fork (and
		398	forget about forgetting to tell libev about forking) when you use this
		399	flag.
		400
		401	This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
		402	environment variable.
		403
		404	=item C<EVFLAG_NOINOTIFY>
		405
		406	When this flag is specified, then libev will not attempt to use the
		407	I<inotify> API for it's C<ev_stat> watchers. Apart from debugging and
		408	testing, this flag can be useful to conserve inotify file descriptors, as
		409	otherwise each loop using C<ev_stat> watchers consumes one inotify handle.
		410
		411	=item C<EVFLAG_SIGNALFD>
		412
		413	When this flag is specified, then libev will attempt to use the
		414	I<signalfd> API for it's C<ev_signal> (and C<ev_child>) watchers. This API
		415	delivers signals synchronously, which makes it both faster and might make
		416	it possible to get the queued signal data. It can also simplify signal
		417	handling with threads, as long as you properly block signals in your
		418	threads that are not interested in handling them.
		419
		420	Signalfd will not be used by default as this changes your signal mask, and
		421	there are a lot of shoddy libraries and programs (glib's threadpool for
		422	example) that can't properly initialise their signal masks.
		423
143	=item C<EVMETHOD_SELECT> (portable select backend)	424	=item C<EVBACKEND_SELECT> (value 1, portable select backend)
144		425
		426	This is your standard select(2) backend. Not I<completely> standard, as
		427	libev tries to roll its own fd_set with no limits on the number of fds,
		428	but if that fails, expect a fairly low limit on the number of fds when
		429	using this backend. It doesn't scale too well (O(highest_fd)), but its
		430	usually the fastest backend for a low number of (low-numbered :) fds.
		431
		432	To get good performance out of this backend you need a high amount of
		433	parallelism (most of the file descriptors should be busy). If you are
		434	writing a server, you should C<accept ()> in a loop to accept as many
		435	connections as possible during one iteration. You might also want to have
		436	a look at C<ev_set_io_collect_interval ()> to increase the amount of
		437	readiness notifications you get per iteration.
		438
		439	This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
		440	C<writefds> set (and to work around Microsoft Windows bugs, also onto the
		441	C<exceptfds> set on that platform).
		442
145	=item C<EVMETHOD_POLL> (poll backend, available everywhere except on windows)	443	=item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
146		444
147	=item C<EVMETHOD_EPOLL> (linux only)	445	And this is your standard poll(2) backend. It's more complicated
		446	than select, but handles sparse fds better and has no artificial
		447	limit on the number of fds you can use (except it will slow down
		448	considerably with a lot of inactive fds). It scales similarly to select,
		449	i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
		450	performance tips.
148		451
149	=item C<EVMETHOD_KQUEUE> (some bsds only)	452	This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
		453	C<EV_WRITE> to C<POLLOUT | POLLERR | POLLHUP>.
150		454
151	=item C<EVMETHOD_DEVPOLL> (solaris 8 only)	455	=item C<EVBACKEND_EPOLL> (value 4, Linux)
152		456
153	=item C<EVMETHOD_PORT> (solaris 10 only)	457	Use the linux-specific epoll(7) interface (for both pre- and post-2.6.9
		458	kernels).
154		459
155	If one or more of these are ored into the flags value, then only these	460	For few fds, this backend is a bit little slower than poll and select,
156	backends will be tried (in the reverse order as given here). If one are	461	but it scales phenomenally better. While poll and select usually scale
157	specified, any backend will do.	462	like O(total_fds) where n is the total number of fds (or the highest fd),
		463	epoll scales either O(1) or O(active_fds).
		464
		465	The epoll mechanism deserves honorable mention as the most misdesigned
		466	of the more advanced event mechanisms: mere annoyances include silently
		467	dropping file descriptors, requiring a system call per change per file
		468	descriptor (and unnecessary guessing of parameters), problems with dup,
		469	returning before the timeout value, resulting in additional iterations
		470	(and only giving 5ms accuracy while select on the same platform gives
		471	0.1ms) and so on. The biggest issue is fork races, however - if a program
		472	forks then I<both> parent and child process have to recreate the epoll
		473	set, which can take considerable time (one syscall per file descriptor)
		474	and is of course hard to detect.
		475
		476	Epoll is also notoriously buggy - embedding epoll fds I<should> work, but
		477	of course I<doesn't>, and epoll just loves to report events for totally
		478	I<different> file descriptors (even already closed ones, so one cannot
		479	even remove them from the set) than registered in the set (especially
		480	on SMP systems). Libev tries to counter these spurious notifications by
		481	employing an additional generation counter and comparing that against the
		482	events to filter out spurious ones, recreating the set when required. Last
		483	not least, it also refuses to work with some file descriptors which work
		484	perfectly fine with C<select> (files, many character devices...).
		485
		486	Epoll is truly the train wreck analog among event poll mechanisms.
		487
		488	While stopping, setting and starting an I/O watcher in the same iteration
		489	will result in some caching, there is still a system call per such
		490	incident (because the same I<file descriptor> could point to a different
		491	I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
		492	file descriptors might not work very well if you register events for both
		493	file descriptors.
		494
		495	Best performance from this backend is achieved by not unregistering all
		496	watchers for a file descriptor until it has been closed, if possible,
		497	i.e. keep at least one watcher active per fd at all times. Stopping and
		498	starting a watcher (without re-setting it) also usually doesn't cause
		499	extra overhead. A fork can both result in spurious notifications as well
		500	as in libev having to destroy and recreate the epoll object, which can
		501	take considerable time and thus should be avoided.
		502
		503	All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
		504	faster than epoll for maybe up to a hundred file descriptors, depending on
		505	the usage. So sad.
		506
		507	While nominally embeddable in other event loops, this feature is broken in
		508	all kernel versions tested so far.
		509
		510	This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
		511	C<EVBACKEND_POLL>.
		512
		513	=item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
		514
		515	Kqueue deserves special mention, as at the time of this writing, it
		516	was broken on all BSDs except NetBSD (usually it doesn't work reliably
		517	with anything but sockets and pipes, except on Darwin, where of course
		518	it's completely useless). Unlike epoll, however, whose brokenness
		519	is by design, these kqueue bugs can (and eventually will) be fixed
		520	without API changes to existing programs. For this reason it's not being
		521	"auto-detected" unless you explicitly specify it in the flags (i.e. using
		522	C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
		523	system like NetBSD.
		524
		525	You still can embed kqueue into a normal poll or select backend and use it
		526	only for sockets (after having made sure that sockets work with kqueue on
		527	the target platform). See C<ev_embed> watchers for more info.
		528
		529	It scales in the same way as the epoll backend, but the interface to the
		530	kernel is more efficient (which says nothing about its actual speed, of
		531	course). While stopping, setting and starting an I/O watcher does never
		532	cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
		533	two event changes per incident. Support for C<fork ()> is very bad (but
		534	sane, unlike epoll) and it drops fds silently in similarly hard-to-detect
		535	cases
		536
		537	This backend usually performs well under most conditions.
		538
		539	While nominally embeddable in other event loops, this doesn't work
		540	everywhere, so you might need to test for this. And since it is broken
		541	almost everywhere, you should only use it when you have a lot of sockets
		542	(for which it usually works), by embedding it into another event loop
		543	(e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
		544	also broken on OS X)) and, did I mention it, using it only for sockets.
		545
		546	This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
		547	C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
		548	C<NOTE_EOF>.
		549
		550	=item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
		551
		552	This is not implemented yet (and might never be, unless you send me an
		553	implementation). According to reports, C</dev/poll> only supports sockets
		554	and is not embeddable, which would limit the usefulness of this backend
		555	immensely.
		556
		557	=item C<EVBACKEND_PORT> (value 32, Solaris 10)
		558
		559	This uses the Solaris 10 event port mechanism. As with everything on Solaris,
		560	it's really slow, but it still scales very well (O(active_fds)).
		561
		562	Please note that Solaris event ports can deliver a lot of spurious
		563	notifications, so you need to use non-blocking I/O or other means to avoid
		564	blocking when no data (or space) is available.
		565
		566	While this backend scales well, it requires one system call per active
		567	file descriptor per loop iteration. For small and medium numbers of file
		568	descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
		569	might perform better.
		570
		571	On the positive side, with the exception of the spurious readiness
		572	notifications, this backend actually performed fully to specification
		573	in all tests and is fully embeddable, which is a rare feat among the
		574	OS-specific backends (I vastly prefer correctness over speed hacks).
		575
		576	This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
		577	C<EVBACKEND_POLL>.
		578
		579	=item C<EVBACKEND_ALL>
		580
		581	Try all backends (even potentially broken ones that wouldn't be tried
		582	with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
		583	C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.
		584
		585	It is definitely not recommended to use this flag.
158		586
159	=back	587	=back
160		588
161	=item struct ev_loop *ev_loop_new (unsigned int flags)	589	If one or more of the backend flags are or'ed into the flags value,
		590	then only these backends will be tried (in the reverse order as listed
		591	here). If none are specified, all backends in C<ev_recommended_backends
		592	()> will be tried.
162		593
163	Similar to C<ev_default_loop>, but always creates a new event loop that is	594	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		595
168	=item ev_default_destroy ()	596	struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
		597	if (!epoller)
		598	fatal ("no epoll found here, maybe it hides under your chair");
169		599
170	Destroys the default loop again (frees all memory and kernel state	600	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	601	used if available.
172	any way whatsoever, although you cannot rely on this :).	602
		603	struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_KQUEUE);
173		604
174	=item ev_loop_destroy (loop)	605	=item ev_loop_destroy (loop)
175		606
176	Like C<ev_default_destroy>, but destroys an event loop created by an	607	Destroys an event loop object (frees all memory and kernel state
177	earlier call to C<ev_loop_new>.	608	etc.). None of the active event watchers will be stopped in the normal
		609	sense, so e.g. C<ev_is_active> might still return true. It is your
		610	responsibility to either stop all watchers cleanly yourself I<before>
		611	calling this function, or cope with the fact afterwards (which is usually
		612	the easiest thing, you can just ignore the watchers and/or C<free ()> them
		613	for example).
178		614
179	=item ev_default_fork ()	615	Note that certain global state, such as signal state (and installed signal
		616	handlers), will not be freed by this function, and related watchers (such
		617	as signal and child watchers) would need to be stopped manually.
180		618
		619	This function is normally used on loop objects allocated by
		620	C<ev_loop_new>, but it can also be used on the default loop returned by
		621	C<ev_default_loop>, in which case it is not thread-safe.
		622
		623	Note that it is not advisable to call this function on the default loop
		624	except in the rare occasion where you really need to free it's resources.
		625	If you need dynamically allocated loops it is better to use C<ev_loop_new>
		626	and C<ev_loop_destroy>.
		627
		628	=item ev_loop_fork (loop)
		629
		630	This function sets a flag that causes subsequent C<ev_run> iterations to
181	This function reinitialises the kernel state for backends that have	631	reinitialise the kernel state for backends that have one. Despite the
182	one. Despite the name, you can call it anytime, but it makes most sense	632	name, you can call it anytime, but it makes most sense after forking, in
183	after forking, in either the parent or child process (or both, but that	633	the child process. You I<must> call it (or use C<EVFLAG_FORKCHECK>) in the
184	again makes little sense).	634	child before resuming or calling C<ev_run>.
185		635
186	You I<must> call this function after forking if and only if you want to	636	Again, you I<have> to call it on I<any> loop that you want to re-use after
187	use the event library in both processes. If you just fork+exec, you don't	637	a fork, I<even if you do not plan to use the loop in the parent>. This is
188	have to call it.	638	because some kernel interfaces *cough* I<kqueue> *cough* do funny things
		639	during fork.
		640
		641	On the other hand, you only need to call this function in the child
		642	process if and only if you want to use the event loop in the child. If
		643	you just fork+exec or create a new loop in the child, you don't have to
		644	call it at all (in fact, C<epoll> is so badly broken that it makes a
		645	difference, but libev will usually detect this case on its own and do a
		646	costly reset of the backend).
189		647
190	The function itself is quite fast and it's usually not a problem to call	648	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	649	it just in case after a fork.
192	quite nicely into a call to C<pthread_atfork>:
193		650
		651	Example: Automate calling C<ev_loop_fork> on the default loop when
		652	using pthreads.
		653
		654	static void
		655	post_fork_child (void)
		656	{
		657	ev_loop_fork (EV_DEFAULT);
		658	}
		659
		660	...
194	pthread_atfork (0, 0, ev_default_fork);	661	pthread_atfork (0, 0, post_fork_child);
195		662
196	=item ev_loop_fork (loop)	663	=item int ev_is_default_loop (loop)
197		664
198	Like C<ev_default_fork>, but acts on an event loop created by	665	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	666	otherwise.
200	after fork, and how you do this is entirely your own problem.
201		667
		668	=item unsigned int ev_iteration (loop)
		669
		670	Returns the current iteration count for the event loop, which is identical
		671	to the number of times libev did poll for new events. It starts at C<0>
		672	and happily wraps around with enough iterations.
		673
		674	This value can sometimes be useful as a generation counter of sorts (it
		675	"ticks" the number of loop iterations), as it roughly corresponds with
		676	C<ev_prepare> and C<ev_check> calls - and is incremented between the
		677	prepare and check phases.
		678
202	=item unsigned int ev_method (loop)	679	=item unsigned int ev_depth (loop)
203		680
		681	Returns the number of times C<ev_run> was entered minus the number of
		682	times C<ev_run> was exited, in other words, the recursion depth.
		683
		684	Outside C<ev_run>, this number is zero. In a callback, this number is
		685	C<1>, unless C<ev_run> was invoked recursively (or from another thread),
		686	in which case it is higher.
		687
		688	Leaving C<ev_run> abnormally (setjmp/longjmp, cancelling the thread
		689	etc.), doesn't count as "exit" - consider this as a hint to avoid such
		690	ungentleman-like behaviour unless it's really convenient.
		691
		692	=item unsigned int ev_backend (loop)
		693
204	Returns one of the C<EVMETHOD_*> flags indicating the event backend in	694	Returns one of the C<EVBACKEND_*> flags indicating the event backend in
205	use.	695	use.
206		696
207	=item ev_tstamp ev_now (loop)	697	=item ev_tstamp ev_now (loop)
208		698
209	Returns the current "event loop time", which is the time the event loop	699	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	700	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	701	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	702	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).	703	event occurring (or more correctly, libev finding out about it).
214		704
		705	=item ev_now_update (loop)
		706
		707	Establishes the current time by querying the kernel, updating the time
		708	returned by C<ev_now ()> in the progress. This is a costly operation and
		709	is usually done automatically within C<ev_run ()>.
		710
		711	This function is rarely useful, but when some event callback runs for a
		712	very long time without entering the event loop, updating libev's idea of
		713	the current time is a good idea.
		714
		715	See also L<The special problem of time updates> in the C<ev_timer> section.
		716
		717	=item ev_suspend (loop)
		718
		719	=item ev_resume (loop)
		720
		721	These two functions suspend and resume an event loop, for use when the
		722	loop is not used for a while and timeouts should not be processed.
		723
		724	A typical use case would be an interactive program such as a game: When
		725	the user presses C<^Z> to suspend the game and resumes it an hour later it
		726	would be best to handle timeouts as if no time had actually passed while
		727	the program was suspended. This can be achieved by calling C<ev_suspend>
		728	in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
		729	C<ev_resume> directly afterwards to resume timer processing.
		730
		731	Effectively, all C<ev_timer> watchers will be delayed by the time spend
		732	between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
		733	will be rescheduled (that is, they will lose any events that would have
		734	occurred while suspended).
		735
		736	After calling C<ev_suspend> you B<must not> call I<any> function on the
		737	given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
		738	without a previous call to C<ev_suspend>.
		739
		740	Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
		741	event loop time (see C<ev_now_update>).
		742
215	=item ev_loop (loop, int flags)	743	=item ev_run (loop, int flags)
216		744
217	Finally, this is it, the event handler. This function usually is called	745	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	746	after you have initialised all your watchers and you want to start
219	events.	747	handling events. It will ask the operating system for any new events, call
		748	the watcher callbacks, an then repeat the whole process indefinitely: This
		749	is why event loops are called I<loops>.
220		750
221	If the flags argument is specified as 0, it will not return until either	751	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.	752	until either no event watchers are active anymore or C<ev_break> was
		753	called.
223		754
		755	Please note that an explicit C<ev_break> is usually better than
		756	relying on all watchers to be stopped when deciding when a program has
		757	finished (especially in interactive programs), but having a program
		758	that automatically loops as long as it has to and no longer by virtue
		759	of relying on its watchers stopping correctly, that is truly a thing of
		760	beauty.
		761
224	A flags value of C<EVLOOP_NONBLOCK> will look for new events, will handle	762	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	763	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.	764	block your process in case there are no events and will return after one
		765	iteration of the loop. This is sometimes useful to poll and handle new
		766	events while doing lengthy calculations, to keep the program responsive.
227		767
228	A flags value of C<EVLOOP_ONESHOT> will look for new events (waiting if	768	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	769	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	770	will block your process until at least one new event arrives (which could
		771	be an event internal to libev itself, so there is no guarantee that a
		772	user-registered callback will be called), and will return after one
231	one iteration of the loop.	773	iteration of the loop.
232		774
233	This flags value could be used to implement alternative looping	775	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	776	with something not expressible using other libev watchers (i.e. "roll your
235	more generic mechanism.	777	own C<ev_run>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
		778	usually a better approach for this kind of thing.
236		779
		780	Here are the gory details of what C<ev_run> does:
		781
		782	- Increment loop depth.
		783	- Reset the ev_break status.
		784	- Before the first iteration, call any pending watchers.
		785	LOOP:
		786	- If EVFLAG_FORKCHECK was used, check for a fork.
		787	- If a fork was detected (by any means), queue and call all fork watchers.
		788	- Queue and call all prepare watchers.
		789	- If ev_break was called, goto FINISH.
		790	- If we have been forked, detach and recreate the kernel state
		791	as to not disturb the other process.
		792	- Update the kernel state with all outstanding changes.
		793	- Update the "event loop time" (ev_now ()).
		794	- Calculate for how long to sleep or block, if at all
		795	(active idle watchers, EVRUN_NOWAIT or not having
		796	any active watchers at all will result in not sleeping).
		797	- Sleep if the I/O and timer collect interval say so.
		798	- Increment loop iteration counter.
		799	- Block the process, waiting for any events.
		800	- Queue all outstanding I/O (fd) events.
		801	- Update the "event loop time" (ev_now ()), and do time jump adjustments.
		802	- Queue all expired timers.
		803	- Queue all expired periodics.
		804	- Queue all idle watchers with priority higher than that of pending events.
		805	- Queue all check watchers.
		806	- Call all queued watchers in reverse order (i.e. check watchers first).
		807	Signals and child watchers are implemented as I/O watchers, and will
		808	be handled here by queueing them when their watcher gets executed.
		809	- If ev_break has been called, or EVRUN_ONCE or EVRUN_NOWAIT
		810	were used, or there are no active watchers, goto FINISH, otherwise
		811	continue with step LOOP.
		812	FINISH:
		813	- Reset the ev_break status iff it was EVBREAK_ONE.
		814	- Decrement the loop depth.
		815	- Return.
		816
		817	Example: Queue some jobs and then loop until no events are outstanding
		818	anymore.
		819
		820	... queue jobs here, make sure they register event watchers as long
		821	... as they still have work to do (even an idle watcher will do..)
		822	ev_run (my_loop, 0);
		823	... jobs done or somebody called unloop. yeah!
		824
237	=item ev_unloop (loop, how)	825	=item ev_break (loop, how)
238		826
239	Can be used to make a call to C<ev_loop> return early (but only after it	827	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	828	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	829	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.	830	C<EVBREAK_ALL>, which will make all nested C<ev_run> calls return.
		831
		832	This "break state" will be cleared when entering C<ev_run> again.
		833
		834	It is safe to call C<ev_break> from outside any C<ev_run> calls, too.
243		835
244	=item ev_ref (loop)	836	=item ev_ref (loop)
245		837
246	=item ev_unref (loop)	838	=item ev_unref (loop)
247		839
248	Ref/unref can be used to add or remove a reference count on the event	840	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	841	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	842	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	843
252	returning, ev_unref() after starting, and ev_ref() before stopping it. For	844	This is useful when you have a watcher that you never intend to
		845	unregister, but that nevertheless should not keep C<ev_run> from
		846	returning. In such a case, call C<ev_unref> after starting, and C<ev_ref>
		847	before stopping it.
		848
253	example, libev itself uses this for its internal signal pipe: It is not	849	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	850	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	851	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	852	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>.	853	third-party libraries. Just remember to I<unref after start> and I<ref
		854	before stop> (but only if the watcher wasn't active before, or was active
		855	before, respectively. Note also that libev might stop watchers itself
		856	(e.g. non-repeating timers) in which case you have to C<ev_ref>
		857	in the callback).
		858
		859	Example: Create a signal watcher, but keep it from keeping C<ev_run>
		860	running when nothing else is active.
		861
		862	ev_signal exitsig;
		863	ev_signal_init (&exitsig, sig_cb, SIGINT);
		864	ev_signal_start (loop, &exitsig);
		865	evf_unref (loop);
		866
		867	Example: For some weird reason, unregister the above signal handler again.
		868
		869	ev_ref (loop);
		870	ev_signal_stop (loop, &exitsig);
		871
		872	=item ev_set_io_collect_interval (loop, ev_tstamp interval)
		873
		874	=item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
		875
		876	These advanced functions influence the time that libev will spend waiting
		877	for events. Both time intervals are by default C<0>, meaning that libev
		878	will try to invoke timer/periodic callbacks and I/O callbacks with minimum
		879	latency.
		880
		881	Setting these to a higher value (the C<interval> I<must> be >= C<0>)
		882	allows libev to delay invocation of I/O and timer/periodic callbacks
		883	to increase efficiency of loop iterations (or to increase power-saving
		884	opportunities).
		885
		886	The idea is that sometimes your program runs just fast enough to handle
		887	one (or very few) event(s) per loop iteration. While this makes the
		888	program responsive, it also wastes a lot of CPU time to poll for new
		889	events, especially with backends like C<select ()> which have a high
		890	overhead for the actual polling but can deliver many events at once.
		891
		892	By setting a higher I<io collect interval> you allow libev to spend more
		893	time collecting I/O events, so you can handle more events per iteration,
		894	at the cost of increasing latency. Timeouts (both C<ev_periodic> and
		895	C<ev_timer>) will be not affected. Setting this to a non-null value will
		896	introduce an additional C<ev_sleep ()> call into most loop iterations. The
		897	sleep time ensures that libev will not poll for I/O events more often then
		898	once per this interval, on average.
		899
		900	Likewise, by setting a higher I<timeout collect interval> you allow libev
		901	to spend more time collecting timeouts, at the expense of increased
		902	latency/jitter/inexactness (the watcher callback will be called
		903	later). C<ev_io> watchers will not be affected. Setting this to a non-null
		904	value will not introduce any overhead in libev.
		905
		906	Many (busy) programs can usually benefit by setting the I/O collect
		907	interval to a value near C<0.1> or so, which is often enough for
		908	interactive servers (of course not for games), likewise for timeouts. It
		909	usually doesn't make much sense to set it to a lower value than C<0.01>,
		910	as this approaches the timing granularity of most systems. Note that if
		911	you do transactions with the outside world and you can't increase the
		912	parallelity, then this setting will limit your transaction rate (if you
		913	need to poll once per transaction and the I/O collect interval is 0.01,
		914	then you can't do more than 100 transactions per second).
		915
		916	Setting the I<timeout collect interval> can improve the opportunity for
		917	saving power, as the program will "bundle" timer callback invocations that
		918	are "near" in time together, by delaying some, thus reducing the number of
		919	times the process sleeps and wakes up again. Another useful technique to
		920	reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
		921	they fire on, say, one-second boundaries only.
		922
		923	Example: we only need 0.1s timeout granularity, and we wish not to poll
		924	more often than 100 times per second:
		925
		926	ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
		927	ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
		928
		929	=item ev_invoke_pending (loop)
		930
		931	This call will simply invoke all pending watchers while resetting their
		932	pending state. Normally, C<ev_run> does this automatically when required,
		933	but when overriding the invoke callback this call comes handy. This
		934	function can be invoked from a watcher - this can be useful for example
		935	when you want to do some lengthy calculation and want to pass further
		936	event handling to another thread (you still have to make sure only one
		937	thread executes within C<ev_invoke_pending> or C<ev_run> of course).
		938
		939	=item int ev_pending_count (loop)
		940
		941	Returns the number of pending watchers - zero indicates that no watchers
		942	are pending.
		943
		944	=item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
		945
		946	This overrides the invoke pending functionality of the loop: Instead of
		947	invoking all pending watchers when there are any, C<ev_run> will call
		948	this callback instead. This is useful, for example, when you want to
		949	invoke the actual watchers inside another context (another thread etc.).
		950
		951	If you want to reset the callback, use C<ev_invoke_pending> as new
		952	callback.
		953
		954	=item ev_set_loop_release_cb (loop, void (*release)(EV_P), void (*acquire)(EV_P))
		955
		956	Sometimes you want to share the same loop between multiple threads. This
		957	can be done relatively simply by putting mutex_lock/unlock calls around
		958	each call to a libev function.
		959
		960	However, C<ev_run> can run an indefinite time, so it is not feasible
		961	to wait for it to return. One way around this is to wake up the event
		962	loop via C<ev_break> and C<av_async_send>, another way is to set these
		963	I<release> and I<acquire> callbacks on the loop.
		964
		965	When set, then C<release> will be called just before the thread is
		966	suspended waiting for new events, and C<acquire> is called just
		967	afterwards.
		968
		969	Ideally, C<release> will just call your mutex_unlock function, and
		970	C<acquire> will just call the mutex_lock function again.
		971
		972	While event loop modifications are allowed between invocations of
		973	C<release> and C<acquire> (that's their only purpose after all), no
		974	modifications done will affect the event loop, i.e. adding watchers will
		975	have no effect on the set of file descriptors being watched, or the time
		976	waited. Use an C<ev_async> watcher to wake up C<ev_run> when you want it
		977	to take note of any changes you made.
		978
		979	In theory, threads executing C<ev_run> will be async-cancel safe between
		980	invocations of C<release> and C<acquire>.
		981
		982	See also the locking example in the C<THREADS> section later in this
		983	document.
		984
		985	=item ev_set_userdata (loop, void *data)
		986
		987	=item ev_userdata (loop)
		988
		989	Set and retrieve a single C<void *> associated with a loop. When
		990	C<ev_set_userdata> has never been called, then C<ev_userdata> returns
		991	C<0.>
		992
		993	These two functions can be used to associate arbitrary data with a loop,
		994	and are intended solely for the C<invoke_pending_cb>, C<release> and
		995	C<acquire> callbacks described above, but of course can be (ab-)used for
		996	any other purpose as well.
		997
		998	=item ev_verify (loop)
		999
		1000	This function only does something when C<EV_VERIFY> support has been
		1001	compiled in, which is the default for non-minimal builds. It tries to go
		1002	through all internal structures and checks them for validity. If anything
		1003	is found to be inconsistent, it will print an error message to standard
		1004	error and call C<abort ()>.
		1005
		1006	This can be used to catch bugs inside libev itself: under normal
		1007	circumstances, this function will never abort as of course libev keeps its
		1008	data structures consistent.
258		1009
259	=back	1010	=back
260		1011
		1012
261	=head1 ANATOMY OF A WATCHER	1013	=head1 ANATOMY OF A WATCHER
262		1014
		1015	In the following description, uppercase C<TYPE> in names stands for the
		1016	watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
		1017	watchers and C<ev_io_start> for I/O watchers.
		1018
263	A watcher is a structure that you create and register to record your	1019	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	1020	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:	1021	to wait for STDIN to become readable, you would create an C<ev_io> watcher
		1022	for that:
266		1023
267	static void my_cb (struct ev_loop *loop, struct ev_io *w, int revents)	1024	static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
268	{	1025	{
269	ev_io_stop (w);	1026	ev_io_stop (w);
270	ev_unloop (loop, EVUNLOOP_ALL);	1027	ev_break (loop, EVBREAK_ALL);
271	}	1028	}
272		1029
273	struct ev_loop *loop = ev_default_loop (0);	1030	struct ev_loop *loop = ev_default_loop (0);
		1031
274	struct ev_io stdin_watcher;	1032	ev_io stdin_watcher;
		1033
275	ev_init (&stdin_watcher, my_cb);	1034	ev_init (&stdin_watcher, my_cb);
276	ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);	1035	ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
277	ev_io_start (loop, &stdin_watcher);	1036	ev_io_start (loop, &stdin_watcher);
		1037
278	ev_loop (loop, 0);	1038	ev_run (loop, 0);
279		1039
280	As you can see, you are responsible for allocating the memory for your	1040	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,	1041	watcher structures (and it is I<usually> a bad idea to do this on the
282	although this can sometimes be quite valid).	1042	stack).
283		1043
		1044	Each watcher has an associated watcher structure (called C<struct ev_TYPE>
		1045	or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
		1046
284	Each watcher structure must be initialised by a call to C<ev_init	1047	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	1048	*, 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	1049	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	1050	time the event loop detects that the file descriptor given is readable
288	is readable and/or writable).	1051	and/or writable).
289		1052
290	Each watcher type has its own C<< ev_<type>_set (watcher *, ...) >> macro	1053	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	1054	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	1055	is also a macro to combine initialisation and setting in one call: C<<
293	(watcher *, callback, ...) >>.	1056	ev_TYPE_init (watcher *, callback, ...) >>.
294		1057
295	To make the watcher actually watch out for events, you have to start it	1058	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	1059	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	1060	*) >>), and you can stop watching for events at any time by calling the
298	corresponding stop function (C<< ev_<type>_stop (loop, watcher *) >>.	1061	corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
299		1062
300	As long as your watcher is active (has been started but not stopped) you	1063	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	1064	must not touch the values stored in it. Most specifically you must never
302	reinitialise it or call its set method.	1065	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		1066
309	Each and every callback receives the event loop pointer as first, the	1067	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	1068	registered watcher structure as second, and a bitset of received events as
311	third argument.	1069	third argument.
312		1070
…		…
321	=item C<EV_WRITE>	1079	=item C<EV_WRITE>
322		1080
323	The file descriptor in the C<ev_io> watcher has become readable and/or	1081	The file descriptor in the C<ev_io> watcher has become readable and/or
324	writable.	1082	writable.
325		1083
326	=item C<EV_TIMEOUT>	1084	=item C<EV_TIMER>
327		1085
328	The C<ev_timer> watcher has timed out.	1086	The C<ev_timer> watcher has timed out.
329		1087
330	=item C<EV_PERIODIC>	1088	=item C<EV_PERIODIC>
331		1089
…		…
337		1095
338	=item C<EV_CHILD>	1096	=item C<EV_CHILD>
339		1097
340	The pid specified in the C<ev_child> watcher has received a status change.	1098	The pid specified in the C<ev_child> watcher has received a status change.
341		1099
		1100	=item C<EV_STAT>
		1101
		1102	The path specified in the C<ev_stat> watcher changed its attributes somehow.
		1103
342	=item C<EV_IDLE>	1104	=item C<EV_IDLE>
343		1105
344	The C<ev_idle> watcher has determined that you have nothing better to do.	1106	The C<ev_idle> watcher has determined that you have nothing better to do.
345		1107
346	=item C<EV_PREPARE>	1108	=item C<EV_PREPARE>
347		1109
348	=item C<EV_CHECK>	1110	=item C<EV_CHECK>
349		1111
350	All C<ev_prepare> watchers are invoked just I<before> C<ev_loop> starts	1112	All C<ev_prepare> watchers are invoked just I<before> C<ev_run> starts
351	to gather new events, and all C<ev_check> watchers are invoked just after	1113	to gather new events, and all C<ev_check> watchers are invoked just after
352	C<ev_loop> has gathered them, but before it invokes any callbacks for any	1114	C<ev_run> has gathered them, but before it invokes any callbacks for any
353	received events. Callbacks of both watcher types can start and stop as	1115	received events. Callbacks of both watcher types can start and stop as
354	many watchers as they want, and all of them will be taken into account	1116	many watchers as they want, and all of them will be taken into account
355	(for example, a C<ev_prepare> watcher might start an idle watcher to keep	1117	(for example, a C<ev_prepare> watcher might start an idle watcher to keep
356	C<ev_loop> from blocking).	1118	C<ev_run> from blocking).
		1119
		1120	=item C<EV_EMBED>
		1121
		1122	The embedded event loop specified in the C<ev_embed> watcher needs attention.
		1123
		1124	=item C<EV_FORK>
		1125
		1126	The event loop has been resumed in the child process after fork (see
		1127	C<ev_fork>).
		1128
		1129	=item C<EV_CLEANUP>
		1130
		1131	The event loop is about to be destroyed (see C<ev_cleanup>).
		1132
		1133	=item C<EV_ASYNC>
		1134
		1135	The given async watcher has been asynchronously notified (see C<ev_async>).
		1136
		1137	=item C<EV_CUSTOM>
		1138
		1139	Not ever sent (or otherwise used) by libev itself, but can be freely used
		1140	by libev users to signal watchers (e.g. via C<ev_feed_event>).
357		1141
358	=item C<EV_ERROR>	1142	=item C<EV_ERROR>
359		1143
360	An unspecified error has occured, the watcher has been stopped. This might	1144	An unspecified error has occurred, the watcher has been stopped. This might
361	happen because the watcher could not be properly started because libev	1145	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	1146	ran out of memory, a file descriptor was found to be closed or any other
		1147	problem. Libev considers these application bugs.
		1148
363	problem. You best act on it by reporting the problem and somehow coping	1149	You best act on it by reporting the problem and somehow coping with the
364	with the watcher being stopped.	1150	watcher being stopped. Note that well-written programs should not receive
		1151	an error ever, so when your watcher receives it, this usually indicates a
		1152	bug in your program.
365		1153
366	Libev will usually signal a few "dummy" events together with an error,	1154	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	1155	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	1156	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	1157	the error from read() or write(). This will not work in multi-threaded
370	programs, though, so beware.	1158	programs, though, as the fd could already be closed and reused for another
		1159	thing, so beware.
371		1160
372	=back	1161	=back
373		1162
		1163	=head2 GENERIC WATCHER FUNCTIONS
		1164
		1165	=over 4
		1166
		1167	=item C<ev_init> (ev_TYPE *watcher, callback)
		1168
		1169	This macro initialises the generic portion of a watcher. The contents
		1170	of the watcher object can be arbitrary (so C<malloc> will do). Only
		1171	the generic parts of the watcher are initialised, you I<need> to call
		1172	the type-specific C<ev_TYPE_set> macro afterwards to initialise the
		1173	type-specific parts. For each type there is also a C<ev_TYPE_init> macro
		1174	which rolls both calls into one.
		1175
		1176	You can reinitialise a watcher at any time as long as it has been stopped
		1177	(or never started) and there are no pending events outstanding.
		1178
		1179	The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
		1180	int revents)>.
		1181
		1182	Example: Initialise an C<ev_io> watcher in two steps.
		1183
		1184	ev_io w;
		1185	ev_init (&w, my_cb);
		1186	ev_io_set (&w, STDIN_FILENO, EV_READ);
		1187
		1188	=item C<ev_TYPE_set> (ev_TYPE *watcher, [args])
		1189
		1190	This macro initialises the type-specific parts of a watcher. You need to
		1191	call C<ev_init> at least once before you call this macro, but you can
		1192	call C<ev_TYPE_set> any number of times. You must not, however, call this
		1193	macro on a watcher that is active (it can be pending, however, which is a
		1194	difference to the C<ev_init> macro).
		1195
		1196	Although some watcher types do not have type-specific arguments
		1197	(e.g. C<ev_prepare>) you still need to call its C<set> macro.
		1198
		1199	See C<ev_init>, above, for an example.
		1200
		1201	=item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
		1202
		1203	This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
		1204	calls into a single call. This is the most convenient method to initialise
		1205	a watcher. The same limitations apply, of course.
		1206
		1207	Example: Initialise and set an C<ev_io> watcher in one step.
		1208
		1209	ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
		1210
		1211	=item C<ev_TYPE_start> (loop, ev_TYPE *watcher)
		1212
		1213	Starts (activates) the given watcher. Only active watchers will receive
		1214	events. If the watcher is already active nothing will happen.
		1215
		1216	Example: Start the C<ev_io> watcher that is being abused as example in this
		1217	whole section.
		1218
		1219	ev_io_start (EV_DEFAULT_UC, &w);
		1220
		1221	=item C<ev_TYPE_stop> (loop, ev_TYPE *watcher)
		1222
		1223	Stops the given watcher if active, and clears the pending status (whether
		1224	the watcher was active or not).
		1225
		1226	It is possible that stopped watchers are pending - for example,
		1227	non-repeating timers are being stopped when they become pending - but
		1228	calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
		1229	pending. If you want to free or reuse the memory used by the watcher it is
		1230	therefore a good idea to always call its C<ev_TYPE_stop> function.
		1231
		1232	=item bool ev_is_active (ev_TYPE *watcher)
		1233
		1234	Returns a true value iff the watcher is active (i.e. it has been started
		1235	and not yet been stopped). As long as a watcher is active you must not modify
		1236	it.
		1237
		1238	=item bool ev_is_pending (ev_TYPE *watcher)
		1239
		1240	Returns a true value iff the watcher is pending, (i.e. it has outstanding
		1241	events but its callback has not yet been invoked). As long as a watcher
		1242	is pending (but not active) you must not call an init function on it (but
		1243	C<ev_TYPE_set> is safe), you must not change its priority, and you must
		1244	make sure the watcher is available to libev (e.g. you cannot C<free ()>
		1245	it).
		1246
		1247	=item callback ev_cb (ev_TYPE *watcher)
		1248
		1249	Returns the callback currently set on the watcher.
		1250
		1251	=item ev_cb_set (ev_TYPE *watcher, callback)
		1252
		1253	Change the callback. You can change the callback at virtually any time
		1254	(modulo threads).
		1255
		1256	=item ev_set_priority (ev_TYPE *watcher, int priority)
		1257
		1258	=item int ev_priority (ev_TYPE *watcher)
		1259
		1260	Set and query the priority of the watcher. The priority is a small
		1261	integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
		1262	(default: C<-2>). Pending watchers with higher priority will be invoked
		1263	before watchers with lower priority, but priority will not keep watchers
		1264	from being executed (except for C<ev_idle> watchers).
		1265
		1266	If you need to suppress invocation when higher priority events are pending
		1267	you need to look at C<ev_idle> watchers, which provide this functionality.
		1268
		1269	You I<must not> change the priority of a watcher as long as it is active or
		1270	pending.
		1271
		1272	Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
		1273	fine, as long as you do not mind that the priority value you query might
		1274	or might not have been clamped to the valid range.
		1275
		1276	The default priority used by watchers when no priority has been set is
		1277	always C<0>, which is supposed to not be too high and not be too low :).
		1278
		1279	See L<WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
		1280	priorities.
		1281
		1282	=item ev_invoke (loop, ev_TYPE *watcher, int revents)
		1283
		1284	Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
		1285	C<loop> nor C<revents> need to be valid as long as the watcher callback
		1286	can deal with that fact, as both are simply passed through to the
		1287	callback.
		1288
		1289	=item int ev_clear_pending (loop, ev_TYPE *watcher)
		1290
		1291	If the watcher is pending, this function clears its pending status and
		1292	returns its C<revents> bitset (as if its callback was invoked). If the
		1293	watcher isn't pending it does nothing and returns C<0>.
		1294
		1295	Sometimes it can be useful to "poll" a watcher instead of waiting for its
		1296	callback to be invoked, which can be accomplished with this function.
		1297
		1298	=item ev_feed_event (loop, ev_TYPE *watcher, int revents)
		1299
		1300	Feeds the given event set into the event loop, as if the specified event
		1301	had happened for the specified watcher (which must be a pointer to an
		1302	initialised but not necessarily started event watcher). Obviously you must
		1303	not free the watcher as long as it has pending events.
		1304
		1305	Stopping the watcher, letting libev invoke it, or calling
		1306	C<ev_clear_pending> will clear the pending event, even if the watcher was
		1307	not started in the first place.
		1308
		1309	See also C<ev_feed_fd_event> and C<ev_feed_signal_event> for related
		1310	functions that do not need a watcher.
		1311
		1312	=back
		1313
374	=head2 ASSOCIATING CUSTOM DATA WITH A WATCHER	1314	=head2 ASSOCIATING CUSTOM DATA WITH A WATCHER
375		1315
376	Each watcher has, by default, a member C<void *data> that you can change	1316	Each watcher has, by default, a member C<void *data> that you can change
377	and read at any time, libev will completely ignore it. This can be used	1317	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	1318	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	1319	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	1320	member, you can also "subclass" the watcher type and provide your own
381	data:	1321	data:
382		1322
383	struct my_io	1323	struct my_io
384	{	1324	{
385	struct ev_io io;	1325	ev_io io;
386	int otherfd;	1326	int otherfd;
387	void *somedata;	1327	void *somedata;
388	struct whatever *mostinteresting;	1328	struct whatever *mostinteresting;
389	}	1329	};
		1330
		1331	...
		1332	struct my_io w;
		1333	ev_io_init (&w.io, my_cb, fd, EV_READ);
390		1334
391	And since your callback will be called with a pointer to the watcher, you	1335	And since your callback will be called with a pointer to the watcher, you
392	can cast it back to your own type:	1336	can cast it back to your own type:
393		1337
394	static void my_cb (struct ev_loop *loop, struct ev_io *w_, int revents)	1338	static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
395	{	1339	{
396	struct my_io *w = (struct my_io *)w_;	1340	struct my_io *w = (struct my_io *)w_;
397	...	1341	...
398	}	1342	}
399		1343
400	More interesting and less C-conformant ways of catsing your callback type	1344	More interesting and less C-conformant ways of casting your callback type
401	have been omitted....	1345	instead have been omitted.
		1346
		1347	Another common scenario is to use some data structure with multiple
		1348	embedded watchers:
		1349
		1350	struct my_biggy
		1351	{
		1352	int some_data;
		1353	ev_timer t1;
		1354	ev_timer t2;
		1355	}
		1356
		1357	In this case getting the pointer to C<my_biggy> is a bit more
		1358	complicated: Either you store the address of your C<my_biggy> struct
		1359	in the C<data> member of the watcher (for woozies), or you need to use
		1360	some pointer arithmetic using C<offsetof> inside your watchers (for real
		1361	programmers):
		1362
		1363	#include <stddef.h>
		1364
		1365	static void
		1366	t1_cb (EV_P_ ev_timer *w, int revents)
		1367	{
		1368	struct my_biggy big = (struct my_biggy *)
		1369	(((char *)w) - offsetof (struct my_biggy, t1));
		1370	}
		1371
		1372	static void
		1373	t2_cb (EV_P_ ev_timer *w, int revents)
		1374	{
		1375	struct my_biggy big = (struct my_biggy *)
		1376	(((char *)w) - offsetof (struct my_biggy, t2));
		1377	}
		1378
		1379	=head2 WATCHER STATES
		1380
		1381	There are various watcher states mentioned throughout this manual -
		1382	active, pending and so on. In this section these states and the rules to
		1383	transition between them will be described in more detail - and while these
		1384	rules might look complicated, they usually do "the right thing".
		1385
		1386	=over 4
		1387
		1388	=item initialiased
		1389
		1390	Before a watcher can be registered with the event looop it has to be
		1391	initialised. This can be done with a call to C<ev_TYPE_init>, or calls to
		1392	C<ev_init> followed by the watcher-specific C<ev_TYPE_set> function.
		1393
		1394	In this state it is simply some block of memory that is suitable for use
		1395	in an event loop. It can be moved around, freed, reused etc. at will.
		1396
		1397	=item started/running/active
		1398
		1399	Once a watcher has been started with a call to C<ev_TYPE_start> it becomes
		1400	property of the event loop, and is actively waiting for events. While in
		1401	this state it cannot be accessed (except in a few documented ways), moved,
		1402	freed or anything else - the only legal thing is to keep a pointer to it,
		1403	and call libev functions on it that are documented to work on active watchers.
		1404
		1405	=item pending
		1406
		1407	If a watcher is active and libev determines that an event it is interested
		1408	in has occurred (such as a timer expiring), it will become pending. It will
		1409	stay in this pending state until either it is stopped or its callback is
		1410	about to be invoked, so it is not normally pending inside the watcher
		1411	callback.
		1412
		1413	The watcher might or might not be active while it is pending (for example,
		1414	an expired non-repeating timer can be pending but no longer active). If it
		1415	is stopped, it can be freely accessed (e.g. by calling C<ev_TYPE_set>),
		1416	but it is still property of the event loop at this time, so cannot be
		1417	moved, freed or reused. And if it is active the rules described in the
		1418	previous item still apply.
		1419
		1420	It is also possible to feed an event on a watcher that is not active (e.g.
		1421	via C<ev_feed_event>), in which case it becomes pending without being
		1422	active.
		1423
		1424	=item stopped
		1425
		1426	A watcher can be stopped implicitly by libev (in which case it might still
		1427	be pending), or explicitly by calling its C<ev_TYPE_stop> function. The
		1428	latter will clear any pending state the watcher might be in, regardless
		1429	of whether it was active or not, so stopping a watcher explicitly before
		1430	freeing it is often a good idea.
		1431
		1432	While stopped (and not pending) the watcher is essentially in the
		1433	initialised state, that is it can be reused, moved, modified in any way
		1434	you wish.
		1435
		1436	=back
		1437
		1438	=head2 WATCHER PRIORITY MODELS
		1439
		1440	Many event loops support I<watcher priorities>, which are usually small
		1441	integers that influence the ordering of event callback invocation
		1442	between watchers in some way, all else being equal.
		1443
		1444	In libev, Watcher priorities can be set using C<ev_set_priority>. See its
		1445	description for the more technical details such as the actual priority
		1446	range.
		1447
		1448	There are two common ways how these these priorities are being interpreted
		1449	by event loops:
		1450
		1451	In the more common lock-out model, higher priorities "lock out" invocation
		1452	of lower priority watchers, which means as long as higher priority
		1453	watchers receive events, lower priority watchers are not being invoked.
		1454
		1455	The less common only-for-ordering model uses priorities solely to order
		1456	callback invocation within a single event loop iteration: Higher priority
		1457	watchers are invoked before lower priority ones, but they all get invoked
		1458	before polling for new events.
		1459
		1460	Libev uses the second (only-for-ordering) model for all its watchers
		1461	except for idle watchers (which use the lock-out model).
		1462
		1463	The rationale behind this is that implementing the lock-out model for
		1464	watchers is not well supported by most kernel interfaces, and most event
		1465	libraries will just poll for the same events again and again as long as
		1466	their callbacks have not been executed, which is very inefficient in the
		1467	common case of one high-priority watcher locking out a mass of lower
		1468	priority ones.
		1469
		1470	Static (ordering) priorities are most useful when you have two or more
		1471	watchers handling the same resource: a typical usage example is having an
		1472	C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
		1473	timeouts. Under load, data might be received while the program handles
		1474	other jobs, but since timers normally get invoked first, the timeout
		1475	handler will be executed before checking for data. In that case, giving
		1476	the timer a lower priority than the I/O watcher ensures that I/O will be
		1477	handled first even under adverse conditions (which is usually, but not
		1478	always, what you want).
		1479
		1480	Since idle watchers use the "lock-out" model, meaning that idle watchers
		1481	will only be executed when no same or higher priority watchers have
		1482	received events, they can be used to implement the "lock-out" model when
		1483	required.
		1484
		1485	For example, to emulate how many other event libraries handle priorities,
		1486	you can associate an C<ev_idle> watcher to each such watcher, and in
		1487	the normal watcher callback, you just start the idle watcher. The real
		1488	processing is done in the idle watcher callback. This causes libev to
		1489	continuously poll and process kernel event data for the watcher, but when
		1490	the lock-out case is known to be rare (which in turn is rare :), this is
		1491	workable.
		1492
		1493	Usually, however, the lock-out model implemented that way will perform
		1494	miserably under the type of load it was designed to handle. In that case,
		1495	it might be preferable to stop the real watcher before starting the
		1496	idle watcher, so the kernel will not have to process the event in case
		1497	the actual processing will be delayed for considerable time.
		1498
		1499	Here is an example of an I/O watcher that should run at a strictly lower
		1500	priority than the default, and which should only process data when no
		1501	other events are pending:
		1502
		1503	ev_idle idle; // actual processing watcher
		1504	ev_io io; // actual event watcher
		1505
		1506	static void
		1507	io_cb (EV_P_ ev_io *w, int revents)
		1508	{
		1509	// stop the I/O watcher, we received the event, but
		1510	// are not yet ready to handle it.
		1511	ev_io_stop (EV_A_ w);
		1512
		1513	// start the idle watcher to handle the actual event.
		1514	// it will not be executed as long as other watchers
		1515	// with the default priority are receiving events.
		1516	ev_idle_start (EV_A_ &idle);
		1517	}
		1518
		1519	static void
		1520	idle_cb (EV_P_ ev_idle *w, int revents)
		1521	{
		1522	// actual processing
		1523	read (STDIN_FILENO, ...);
		1524
		1525	// have to start the I/O watcher again, as
		1526	// we have handled the event
		1527	ev_io_start (EV_P_ &io);
		1528	}
		1529
		1530	// initialisation
		1531	ev_idle_init (&idle, idle_cb);
		1532	ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
		1533	ev_io_start (EV_DEFAULT_ &io);
		1534
		1535	In the "real" world, it might also be beneficial to start a timer, so that
		1536	low-priority connections can not be locked out forever under load. This
		1537	enables your program to keep a lower latency for important connections
		1538	during short periods of high load, while not completely locking out less
		1539	important ones.
402		1540
403		1541
404	=head1 WATCHER TYPES	1542	=head1 WATCHER TYPES
405		1543
406	This section describes each watcher in detail, but will not repeat	1544	This section describes each watcher in detail, but will not repeat
407	information given in the last section.	1545	information given in the last section. Any initialisation/set macros,
		1546	functions and members specific to the watcher type are explained.
408		1547
		1548	Members are additionally marked with either I<[read-only]>, meaning that,
		1549	while the watcher is active, you can look at the member and expect some
		1550	sensible content, but you must not modify it (you can modify it while the
		1551	watcher is stopped to your hearts content), or I<[read-write]>, which
		1552	means you can expect it to have some sensible content while the watcher
		1553	is active, but you can also modify it. Modifying it may not do something
		1554	sensible or take immediate effect (or do anything at all), but libev will
		1555	not crash or malfunction in any way.
		1556
		1557
409	=head2 C<ev_io> - is this file descriptor readable or writable	1558	=head2 C<ev_io> - is this file descriptor readable or writable?
410		1559
411	I/O watchers check whether a file descriptor is readable or writable	1560	I/O watchers check whether a file descriptor is readable or writable
412	in each iteration of the event loop (This behaviour is called	1561	in each iteration of the event loop, or, more precisely, when reading
413	level-triggering because you keep receiving events as long as the	1562	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	1563	some data. This behaviour is called level-triggering because you keep
415	act on the event and neither want to receive future events).	1564	receiving events as long as the condition persists. Remember you can stop
		1565	the watcher if you don't want to act on the event and neither want to
		1566	receive future events.
416		1567
417	In general you can register as many read and/or write event watchers oer	1568	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	1569	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	1570	descriptors to non-blocking mode is also usually a good idea (but not
420	required if you know what you are doing).	1571	required if you know what you are doing).
421		1572
422	You have to be careful with dup'ed file descriptors, though. Some backends	1573	If you cannot use non-blocking mode, then force the use of a
423	(the linux epoll backend is a notable example) cannot handle dup'ed file	1574	known-to-be-good backend (at the time of this writing, this includes only
424	descriptors correctly if you register interest in two or more fds pointing	1575	C<EVBACKEND_SELECT> and C<EVBACKEND_POLL>). The same applies to file
425	to the same file/socket etc. description.	1576	descriptors for which non-blocking operation makes no sense (such as
		1577	files) - libev doesn't guarantee any specific behaviour in that case.
426		1578
427	If you must do this, then force the use of a known-to-be-good backend	1579	Another thing you have to watch out for is that it is quite easy to
428	(at the time of this writing, this includes only EVMETHOD_SELECT and	1580	receive "spurious" readiness notifications, that is your callback might
429	EVMETHOD_POLL).	1581	be called with C<EV_READ> but a subsequent C<read>(2) will actually block
		1582	because there is no data. Not only are some backends known to create a
		1583	lot of those (for example Solaris ports), it is very easy to get into
		1584	this situation even with a relatively standard program structure. Thus
		1585	it is best to always use non-blocking I/O: An extra C<read>(2) returning
		1586	C<EAGAIN> is far preferable to a program hanging until some data arrives.
		1587
		1588	If you cannot run the fd in non-blocking mode (for example you should
		1589	not play around with an Xlib connection), then you have to separately
		1590	re-test whether a file descriptor is really ready with a known-to-be good
		1591	interface such as poll (fortunately in our Xlib example, Xlib already
		1592	does this on its own, so its quite safe to use). Some people additionally
		1593	use C<SIGALRM> and an interval timer, just to be sure you won't block
		1594	indefinitely.
		1595
		1596	But really, best use non-blocking mode.
		1597
		1598	=head3 The special problem of disappearing file descriptors
		1599
		1600	Some backends (e.g. kqueue, epoll) need to be told about closing a file
		1601	descriptor (either due to calling C<close> explicitly or any other means,
		1602	such as C<dup2>). The reason is that you register interest in some file
		1603	descriptor, but when it goes away, the operating system will silently drop
		1604	this interest. If another file descriptor with the same number then is
		1605	registered with libev, there is no efficient way to see that this is, in
		1606	fact, a different file descriptor.
		1607
		1608	To avoid having to explicitly tell libev about such cases, libev follows
		1609	the following policy: Each time C<ev_io_set> is being called, libev
		1610	will assume that this is potentially a new file descriptor, otherwise
		1611	it is assumed that the file descriptor stays the same. That means that
		1612	you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
		1613	descriptor even if the file descriptor number itself did not change.
		1614
		1615	This is how one would do it normally anyway, the important point is that
		1616	the libev application should not optimise around libev but should leave
		1617	optimisations to libev.
		1618
		1619	=head3 The special problem of dup'ed file descriptors
		1620
		1621	Some backends (e.g. epoll), cannot register events for file descriptors,
		1622	but only events for the underlying file descriptions. That means when you
		1623	have C<dup ()>'ed file descriptors or weirder constellations, and register
		1624	events for them, only one file descriptor might actually receive events.
		1625
		1626	There is no workaround possible except not registering events
		1627	for potentially C<dup ()>'ed file descriptors, or to resort to
		1628	C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
		1629
		1630	=head3 The special problem of fork
		1631
		1632	Some backends (epoll, kqueue) do not support C<fork ()> at all or exhibit
		1633	useless behaviour. Libev fully supports fork, but needs to be told about
		1634	it in the child.
		1635
		1636	To support fork in your programs, you either have to call
		1637	C<ev_default_fork ()> or C<ev_loop_fork ()> after a fork in the child,
		1638	enable C<EVFLAG_FORKCHECK>, or resort to C<EVBACKEND_SELECT> or
		1639	C<EVBACKEND_POLL>.
		1640
		1641	=head3 The special problem of SIGPIPE
		1642
		1643	While not really specific to libev, it is easy to forget about C<SIGPIPE>:
		1644	when writing to a pipe whose other end has been closed, your program gets
		1645	sent a SIGPIPE, which, by default, aborts your program. For most programs
		1646	this is sensible behaviour, for daemons, this is usually undesirable.
		1647
		1648	So when you encounter spurious, unexplained daemon exits, make sure you
		1649	ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
		1650	somewhere, as that would have given you a big clue).
		1651
		1652	=head3 The special problem of accept()ing when you can't
		1653
		1654	Many implementations of the POSIX C<accept> function (for example,
		1655	found in post-2004 Linux) have the peculiar behaviour of not removing a
		1656	connection from the pending queue in all error cases.
		1657
		1658	For example, larger servers often run out of file descriptors (because
		1659	of resource limits), causing C<accept> to fail with C<ENFILE> but not
		1660	rejecting the connection, leading to libev signalling readiness on
		1661	the next iteration again (the connection still exists after all), and
		1662	typically causing the program to loop at 100% CPU usage.
		1663
		1664	Unfortunately, the set of errors that cause this issue differs between
		1665	operating systems, there is usually little the app can do to remedy the
		1666	situation, and no known thread-safe method of removing the connection to
		1667	cope with overload is known (to me).
		1668
		1669	One of the easiest ways to handle this situation is to just ignore it
		1670	- when the program encounters an overload, it will just loop until the
		1671	situation is over. While this is a form of busy waiting, no OS offers an
		1672	event-based way to handle this situation, so it's the best one can do.
		1673
		1674	A better way to handle the situation is to log any errors other than
		1675	C<EAGAIN> and C<EWOULDBLOCK>, making sure not to flood the log with such
		1676	messages, and continue as usual, which at least gives the user an idea of
		1677	what could be wrong ("raise the ulimit!"). For extra points one could stop
		1678	the C<ev_io> watcher on the listening fd "for a while", which reduces CPU
		1679	usage.
		1680
		1681	If your program is single-threaded, then you could also keep a dummy file
		1682	descriptor for overload situations (e.g. by opening F</dev/null>), and
		1683	when you run into C<ENFILE> or C<EMFILE>, close it, run C<accept>,
		1684	close that fd, and create a new dummy fd. This will gracefully refuse
		1685	clients under typical overload conditions.
		1686
		1687	The last way to handle it is to simply log the error and C<exit>, as
		1688	is often done with C<malloc> failures, but this results in an easy
		1689	opportunity for a DoS attack.
		1690
		1691	=head3 Watcher-Specific Functions
430		1692
431	=over 4	1693	=over 4
432		1694
433	=item ev_io_init (ev_io *, callback, int fd, int events)	1695	=item ev_io_init (ev_io *, callback, int fd, int events)
434		1696
435	=item ev_io_set (ev_io *, int fd, int events)	1697	=item ev_io_set (ev_io *, int fd, int events)
436		1698
437	Configures an C<ev_io> watcher. The fd is the file descriptor to rceeive	1699	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 |	1700	receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
439	EV_WRITE> to receive the given events.	1701	C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
		1702
		1703	=item int fd [read-only]
		1704
		1705	The file descriptor being watched.
		1706
		1707	=item int events [read-only]
		1708
		1709	The events being watched.
440		1710
441	=back	1711	=back
442		1712
		1713	=head3 Examples
		1714
		1715	Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
		1716	readable, but only once. Since it is likely line-buffered, you could
		1717	attempt to read a whole line in the callback.
		1718
		1719	static void
		1720	stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
		1721	{
		1722	ev_io_stop (loop, w);
		1723	.. read from stdin here (or from w->fd) and handle any I/O errors
		1724	}
		1725
		1726	...
		1727	struct ev_loop *loop = ev_default_init (0);
		1728	ev_io stdin_readable;
		1729	ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
		1730	ev_io_start (loop, &stdin_readable);
		1731	ev_run (loop, 0);
		1732
		1733
443	=head2 C<ev_timer> - relative and optionally recurring timeouts	1734	=head2 C<ev_timer> - relative and optionally repeating timeouts
444		1735
445	Timer watchers are simple relative timers that generate an event after a	1736	Timer watchers are simple relative timers that generate an event after a
446	given time, and optionally repeating in regular intervals after that.	1737	given time, and optionally repeating in regular intervals after that.
447		1738
448	The timers are based on real time, that is, if you register an event that	1739	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	1740	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	1741	year, it will still time out after (roughly) one hour. "Roughly" because
451	detecting time jumps is hard, and soem inaccuracies are unavoidable (the	1742	detecting time jumps is hard, and some inaccuracies are unavoidable (the
452	monotonic clock option helps a lot here).	1743	monotonic clock option helps a lot here).
		1744
		1745	The callback is guaranteed to be invoked only I<after> its timeout has
		1746	passed (not I<at>, so on systems with very low-resolution clocks this
		1747	might introduce a small delay). If multiple timers become ready during the
		1748	same loop iteration then the ones with earlier time-out values are invoked
		1749	before ones of the same priority with later time-out values (but this is
		1750	no longer true when a callback calls C<ev_run> recursively).
		1751
		1752	=head3 Be smart about timeouts
		1753
		1754	Many real-world problems involve some kind of timeout, usually for error
		1755	recovery. A typical example is an HTTP request - if the other side hangs,
		1756	you want to raise some error after a while.
		1757
		1758	What follows are some ways to handle this problem, from obvious and
		1759	inefficient to smart and efficient.
		1760
		1761	In the following, a 60 second activity timeout is assumed - a timeout that
		1762	gets reset to 60 seconds each time there is activity (e.g. each time some
		1763	data or other life sign was received).
		1764
		1765	=over 4
		1766
		1767	=item 1. Use a timer and stop, reinitialise and start it on activity.
		1768
		1769	This is the most obvious, but not the most simple way: In the beginning,
		1770	start the watcher:
		1771
		1772	ev_timer_init (timer, callback, 60., 0.);
		1773	ev_timer_start (loop, timer);
		1774
		1775	Then, each time there is some activity, C<ev_timer_stop> it, initialise it
		1776	and start it again:
		1777
		1778	ev_timer_stop (loop, timer);
		1779	ev_timer_set (timer, 60., 0.);
		1780	ev_timer_start (loop, timer);
		1781
		1782	This is relatively simple to implement, but means that each time there is
		1783	some activity, libev will first have to remove the timer from its internal
		1784	data structure and then add it again. Libev tries to be fast, but it's
		1785	still not a constant-time operation.
		1786
		1787	=item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
		1788
		1789	This is the easiest way, and involves using C<ev_timer_again> instead of
		1790	C<ev_timer_start>.
		1791
		1792	To implement this, configure an C<ev_timer> with a C<repeat> value
		1793	of C<60> and then call C<ev_timer_again> at start and each time you
		1794	successfully read or write some data. If you go into an idle state where
		1795	you do not expect data to travel on the socket, you can C<ev_timer_stop>
		1796	the timer, and C<ev_timer_again> will automatically restart it if need be.
		1797
		1798	That means you can ignore both the C<ev_timer_start> function and the
		1799	C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
		1800	member and C<ev_timer_again>.
		1801
		1802	At start:
		1803
		1804	ev_init (timer, callback);
		1805	timer->repeat = 60.;
		1806	ev_timer_again (loop, timer);
		1807
		1808	Each time there is some activity:
		1809
		1810	ev_timer_again (loop, timer);
		1811
		1812	It is even possible to change the time-out on the fly, regardless of
		1813	whether the watcher is active or not:
		1814
		1815	timer->repeat = 30.;
		1816	ev_timer_again (loop, timer);
		1817
		1818	This is slightly more efficient then stopping/starting the timer each time
		1819	you want to modify its timeout value, as libev does not have to completely
		1820	remove and re-insert the timer from/into its internal data structure.
		1821
		1822	It is, however, even simpler than the "obvious" way to do it.
		1823
		1824	=item 3. Let the timer time out, but then re-arm it as required.
		1825
		1826	This method is more tricky, but usually most efficient: Most timeouts are
		1827	relatively long compared to the intervals between other activity - in
		1828	our example, within 60 seconds, there are usually many I/O events with
		1829	associated activity resets.
		1830
		1831	In this case, it would be more efficient to leave the C<ev_timer> alone,
		1832	but remember the time of last activity, and check for a real timeout only
		1833	within the callback:
		1834
		1835	ev_tstamp last_activity; // time of last activity
		1836
		1837	static void
		1838	callback (EV_P_ ev_timer *w, int revents)
		1839	{
		1840	ev_tstamp now = ev_now (EV_A);
		1841	ev_tstamp timeout = last_activity + 60.;
		1842
		1843	// if last_activity + 60. is older than now, we did time out
		1844	if (timeout < now)
		1845	{
		1846	// timeout occurred, take action
		1847	}
		1848	else
		1849	{
		1850	// callback was invoked, but there was some activity, re-arm
		1851	// the watcher to fire in last_activity + 60, which is
		1852	// guaranteed to be in the future, so "again" is positive:
		1853	w->repeat = timeout - now;
		1854	ev_timer_again (EV_A_ w);
		1855	}
		1856	}
		1857
		1858	To summarise the callback: first calculate the real timeout (defined
		1859	as "60 seconds after the last activity"), then check if that time has
		1860	been reached, which means something I<did>, in fact, time out. Otherwise
		1861	the callback was invoked too early (C<timeout> is in the future), so
		1862	re-schedule the timer to fire at that future time, to see if maybe we have
		1863	a timeout then.
		1864
		1865	Note how C<ev_timer_again> is used, taking advantage of the
		1866	C<ev_timer_again> optimisation when the timer is already running.
		1867
		1868	This scheme causes more callback invocations (about one every 60 seconds
		1869	minus half the average time between activity), but virtually no calls to
		1870	libev to change the timeout.
		1871
		1872	To start the timer, simply initialise the watcher and set C<last_activity>
		1873	to the current time (meaning we just have some activity :), then call the
		1874	callback, which will "do the right thing" and start the timer:
		1875
		1876	ev_init (timer, callback);
		1877	last_activity = ev_now (loop);
		1878	callback (loop, timer, EV_TIMER);
		1879
		1880	And when there is some activity, simply store the current time in
		1881	C<last_activity>, no libev calls at all:
		1882
		1883	last_activity = ev_now (loop);
		1884
		1885	This technique is slightly more complex, but in most cases where the
		1886	time-out is unlikely to be triggered, much more efficient.
		1887
		1888	Changing the timeout is trivial as well (if it isn't hard-coded in the
		1889	callback :) - just change the timeout and invoke the callback, which will
		1890	fix things for you.
		1891
		1892	=item 4. Wee, just use a double-linked list for your timeouts.
		1893
		1894	If there is not one request, but many thousands (millions...), all
		1895	employing some kind of timeout with the same timeout value, then one can
		1896	do even better:
		1897
		1898	When starting the timeout, calculate the timeout value and put the timeout
		1899	at the I<end> of the list.
		1900
		1901	Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
		1902	the list is expected to fire (for example, using the technique #3).
		1903
		1904	When there is some activity, remove the timer from the list, recalculate
		1905	the timeout, append it to the end of the list again, and make sure to
		1906	update the C<ev_timer> if it was taken from the beginning of the list.
		1907
		1908	This way, one can manage an unlimited number of timeouts in O(1) time for
		1909	starting, stopping and updating the timers, at the expense of a major
		1910	complication, and having to use a constant timeout. The constant timeout
		1911	ensures that the list stays sorted.
		1912
		1913	=back
		1914
		1915	So which method the best?
		1916
		1917	Method #2 is a simple no-brain-required solution that is adequate in most
		1918	situations. Method #3 requires a bit more thinking, but handles many cases
		1919	better, and isn't very complicated either. In most case, choosing either
		1920	one is fine, with #3 being better in typical situations.
		1921
		1922	Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
		1923	rather complicated, but extremely efficient, something that really pays
		1924	off after the first million or so of active timers, i.e. it's usually
		1925	overkill :)
		1926
		1927	=head3 The special problem of time updates
		1928
		1929	Establishing the current time is a costly operation (it usually takes at
		1930	least two system calls): EV therefore updates its idea of the current
		1931	time only before and after C<ev_run> collects new events, which causes a
		1932	growing difference between C<ev_now ()> and C<ev_time ()> when handling
		1933	lots of events in one iteration.
453		1934
454	The relative timeouts are calculated relative to the C<ev_now ()>	1935	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	1936	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	1937	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	1938	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:	1939	timeout on the current time, use something like this to adjust for this:
459		1940
460	ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);	1941	ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);
461		1942
		1943	If the event loop is suspended for a long time, you can also force an
		1944	update of the time returned by C<ev_now ()> by calling C<ev_now_update
		1945	()>.
		1946
		1947	=head3 The special problems of suspended animation
		1948
		1949	When you leave the server world it is quite customary to hit machines that
		1950	can suspend/hibernate - what happens to the clocks during such a suspend?
		1951
		1952	Some quick tests made with a Linux 2.6.28 indicate that a suspend freezes
		1953	all processes, while the clocks (C<times>, C<CLOCK_MONOTONIC>) continue
		1954	to run until the system is suspended, but they will not advance while the
		1955	system is suspended. That means, on resume, it will be as if the program
		1956	was frozen for a few seconds, but the suspend time will not be counted
		1957	towards C<ev_timer> when a monotonic clock source is used. The real time
		1958	clock advanced as expected, but if it is used as sole clocksource, then a
		1959	long suspend would be detected as a time jump by libev, and timers would
		1960	be adjusted accordingly.
		1961
		1962	I would not be surprised to see different behaviour in different between
		1963	operating systems, OS versions or even different hardware.
		1964
		1965	The other form of suspend (job control, or sending a SIGSTOP) will see a
		1966	time jump in the monotonic clocks and the realtime clock. If the program
		1967	is suspended for a very long time, and monotonic clock sources are in use,
		1968	then you can expect C<ev_timer>s to expire as the full suspension time
		1969	will be counted towards the timers. When no monotonic clock source is in
		1970	use, then libev will again assume a timejump and adjust accordingly.
		1971
		1972	It might be beneficial for this latter case to call C<ev_suspend>
		1973	and C<ev_resume> in code that handles C<SIGTSTP>, to at least get
		1974	deterministic behaviour in this case (you can do nothing against
		1975	C<SIGSTOP>).
		1976
		1977	=head3 Watcher-Specific Functions and Data Members
		1978
462	=over 4	1979	=over 4
463		1980
464	=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)	1981	=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
465		1982
466	=item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)	1983	=item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
467		1984
468	Configure the timer to trigger after C<after> seconds. If C<repeat> is	1985	Configure the timer to trigger after C<after> seconds. If C<repeat>
469	C<0.>, then it will automatically be stopped. If it is positive, then the	1986	is C<0.>, then it will automatically be stopped once the timeout is
470	timer will automatically be configured to trigger again C<repeat> seconds	1987	reached. If it is positive, then the timer will automatically be
471	later, again, and again, until stopped manually.	1988	configured to trigger again C<repeat> seconds later, again, and again,
		1989	until stopped manually.
472		1990
473	The timer itself will do a best-effort at avoiding drift, that is, if you	1991	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	1992	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	1993	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	1994	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.	1995	do stuff) the timer will not fire more than once per event loop iteration.
478		1996
479	=item ev_timer_again (loop)	1997	=item ev_timer_again (loop, ev_timer *)
480		1998
481	This will act as if the timer timed out and restart it again if it is	1999	This will act as if the timer timed out and restart it again if it is
482	repeating. The exact semantics are:	2000	repeating. The exact semantics are:
483		2001
		2002	If the timer is pending, its pending status is cleared.
		2003
484	If the timer is started but nonrepeating, stop it.	2004	If the timer is started but non-repeating, stop it (as if it timed out).
485		2005
486	If the timer is repeating, either start it if necessary (with the repeat	2006	If the timer is repeating, either start it if necessary (with the
487	value), or reset the running timer to the repeat value.	2007	C<repeat> value), or reset the running timer to the C<repeat> value.
488		2008
489	This sounds a bit complicated, but here is a useful and typical	2009	This sounds a bit complicated, see L<Be smart about timeouts>, above, for a
490	example: Imagine you have a tcp connection and you want a so-called idle	2010	usage example.
491	timeout, that is, you want to be called when there have been, say, 60	2011
492	seconds of inactivity on the socket. The easiest way to do this is to	2012	=item ev_tstamp ev_timer_remaining (loop, ev_timer *)
493	configure an C<ev_timer> with after=repeat=60 and calling ev_timer_again each	2013
494	time you successfully read or write some data. If you go into an idle	2014	Returns the remaining time until a timer fires. If the timer is active,
495	state where you do not expect data to travel on the socket, you can stop	2015	then this time is relative to the current event loop time, otherwise it's
496	the timer, and again will automatically restart it if need be.	2016	the timeout value currently configured.
		2017
		2018	That is, after an C<ev_timer_set (w, 5, 7)>, C<ev_timer_remaining> returns
		2019	C<5>. When the timer is started and one second passes, C<ev_timer_remaining>
		2020	will return C<4>. When the timer expires and is restarted, it will return
		2021	roughly C<7> (likely slightly less as callback invocation takes some time,
		2022	too), and so on.
		2023
		2024	=item ev_tstamp repeat [read-write]
		2025
		2026	The current C<repeat> value. Will be used each time the watcher times out
		2027	or C<ev_timer_again> is called, and determines the next timeout (if any),
		2028	which is also when any modifications are taken into account.
497		2029
498	=back	2030	=back
499		2031
		2032	=head3 Examples
		2033
		2034	Example: Create a timer that fires after 60 seconds.
		2035
		2036	static void
		2037	one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
		2038	{
		2039	.. one minute over, w is actually stopped right here
		2040	}
		2041
		2042	ev_timer mytimer;
		2043	ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
		2044	ev_timer_start (loop, &mytimer);
		2045
		2046	Example: Create a timeout timer that times out after 10 seconds of
		2047	inactivity.
		2048
		2049	static void
		2050	timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
		2051	{
		2052	.. ten seconds without any activity
		2053	}
		2054
		2055	ev_timer mytimer;
		2056	ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
		2057	ev_timer_again (&mytimer); /* start timer */
		2058	ev_run (loop, 0);
		2059
		2060	// and in some piece of code that gets executed on any "activity":
		2061	// reset the timeout to start ticking again at 10 seconds
		2062	ev_timer_again (&mytimer);
		2063
		2064
500	=head2 C<ev_periodic> - to cron or not to cron	2065	=head2 C<ev_periodic> - to cron or not to cron?
501		2066
502	Periodic watchers are also timers of a kind, but they are very versatile	2067	Periodic watchers are also timers of a kind, but they are very versatile
503	(and unfortunately a bit complex).	2068	(and unfortunately a bit complex).
504		2069
505	Unlike C<ev_timer>'s, they are not based on real time (or relative time)	2070	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	2071	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	2072	(absolute time, the thing you can read on your calender or clock). The
508	periodic watcher to trigger in 10 seconds (by specifiying e.g. c<ev_now ()	2073	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	2074	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	2075	wrist-watch).
511	roughly 10 seconds later and of course not if you reset your system time
512	again).
513		2076
514	They can also be used to implement vastly more complex timers, such as	2077	You can tell a periodic watcher to trigger after some specific point
		2078	in time: for example, if you tell a periodic watcher to trigger "in 10
		2079	seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
		2080	not a delay) and then reset your system clock to January of the previous
		2081	year, then it will take a year or more to trigger the event (unlike an
		2082	C<ev_timer>, which would still trigger roughly 10 seconds after starting
		2083	it, as it uses a relative timeout).
		2084
		2085	C<ev_periodic> watchers can also be used to implement vastly more complex
515	triggering an event on eahc midnight, local time.	2086	timers, such as triggering an event on each "midnight, local time", or
		2087	other complicated rules. This cannot be done with C<ev_timer> watchers, as
		2088	those cannot react to time jumps.
		2089
		2090	As with timers, the callback is guaranteed to be invoked only when the
		2091	point in time where it is supposed to trigger has passed. If multiple
		2092	timers become ready during the same loop iteration then the ones with
		2093	earlier time-out values are invoked before ones with later time-out values
		2094	(but this is no longer true when a callback calls C<ev_run> recursively).
		2095
		2096	=head3 Watcher-Specific Functions and Data Members
516		2097
517	=over 4	2098	=over 4
518		2099
519	=item ev_periodic_init (ev_periodic *, callback, ev_tstamp at, ev_tstamp interval, reschedule_cb)	2100	=item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
520		2101
521	=item ev_periodic_set (ev_periodic *, ev_tstamp after, ev_tstamp repeat, reschedule_cb)	2102	=item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
522		2103
523	Lots of arguments, lets sort it out... There are basically three modes of	2104	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:	2105	operation, and we will explain them from simplest to most complex:
525
526		2106
527	=over 4	2107	=over 4
528		2108
529	=item * absolute timer (interval = reschedule_cb = 0)	2109	=item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
530		2110
531	In this configuration the watcher triggers an event at the wallclock time	2111	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,	2112	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	2113	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.	2114	will be stopped and invoked when the system clock reaches or surpasses
		2115	this point in time.
535		2116
536	=item * non-repeating interval timer (interval > 0, reschedule_cb = 0)	2117	=item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
537		2118
538	In this mode the watcher will always be scheduled to time out at the next	2119	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	2120	C<offset + N * interval> time (for some integer N, which can also be
540	of any time jumps.	2121	negative) and then repeat, regardless of any time jumps. The C<offset>
		2122	argument is merely an offset into the C<interval> periods.
541		2123
542	This can be used to create timers that do not drift with respect to system	2124	This can be used to create timers that do not drift with respect to the
543	time:	2125	system clock, for example, here is an C<ev_periodic> that triggers each
		2126	hour, on the hour (with respect to UTC):
544		2127
545	ev_periodic_set (&periodic, 0., 3600., 0);	2128	ev_periodic_set (&periodic, 0., 3600., 0);
546		2129
547	This doesn't mean there will always be 3600 seconds in between triggers,	2130	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	2131	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	2132	full hour (UTC), or more correctly, when the system time is evenly divisible
550	by 3600.	2133	by 3600.
551		2134
552	Another way to think about it (for the mathematically inclined) is that	2135	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	2136	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.	2137	time where C<time = offset (mod interval)>, regardless of any time jumps.
555		2138
		2139	For numerical stability it is preferable that the C<offset> value is near
		2140	C<ev_now ()> (the current time), but there is no range requirement for
		2141	this value, and in fact is often specified as zero.
		2142
		2143	Note also that there is an upper limit to how often a timer can fire (CPU
		2144	speed for example), so if C<interval> is very small then timing stability
		2145	will of course deteriorate. Libev itself tries to be exact to be about one
		2146	millisecond (if the OS supports it and the machine is fast enough).
		2147
556	=item * manual reschedule mode (reschedule_cb = callback)	2148	=item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
557		2149
558	In this mode the values for C<interval> and C<at> are both being	2150	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	2151	ignored. Instead, each time the periodic watcher gets scheduled, the
560	reschedule callback will be called with the watcher as first, and the	2152	reschedule callback will be called with the watcher as first, and the
561	current time as second argument.	2153	current time as second argument.
562		2154
563	NOTE: I<This callback MUST NOT stop or destroy the periodic or any other	2155	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	2156	or make ANY other event loop modifications whatsoever, unless explicitly
		2157	allowed by documentation here>.
		2158
565	to stop it, return C<now + 1e30> (or so, fudge fudge) and stop it afterwards.	2159	If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
		2160	it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
		2161	only event loop modification you are allowed to do).
566		2162
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,	2163	The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
571	ev_tstamp now)>, e.g.:	2164	*w, ev_tstamp now)>, e.g.:
572		2165
		2166	static ev_tstamp
573	static ev_tstamp my_rescheduler (struct ev_periodic *w, ev_tstamp now)	2167	my_rescheduler (ev_periodic *w, ev_tstamp now)
574	{	2168	{
575	return now + 60.;	2169	return now + 60.;
576	}	2170	}
577		2171
578	It must return the next time to trigger, based on the passed time value	2172	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	2173	(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	2174	will usually be called just before the callback will be triggered, but
581	might be called at other times, too.	2175	might be called at other times, too.
582		2176
		2177	NOTE: I<< This callback must always return a time that is higher than or
		2178	equal to the passed C<now> value >>.
		2179
583	This can be used to create very complex timers, such as a timer that	2180	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	2181	triggers on "next midnight, local time". To do this, you would calculate the
585	next midnight after C<now> and return the timestamp value for this. How you do this	2182	next midnight after C<now> and return the timestamp value for this. How
586	is, again, up to you (but it is not trivial).	2183	you do this is, again, up to you (but it is not trivial, which is the main
		2184	reason I omitted it as an example).
587		2185
588	=back	2186	=back
589		2187
590	=item ev_periodic_again (loop, ev_periodic *)	2188	=item ev_periodic_again (loop, ev_periodic *)
591		2189
592	Simply stops and restarts the periodic watcher again. This is only useful	2190	Simply stops and restarts the periodic watcher again. This is only useful
593	when you changed some parameters or the reschedule callback would return	2191	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	2192	a different time than the last time it was called (e.g. in a crond like
595	program when the crontabs have changed).	2193	program when the crontabs have changed).
596		2194
		2195	=item ev_tstamp ev_periodic_at (ev_periodic *)
		2196
		2197	When active, returns the absolute time that the watcher is supposed
		2198	to trigger next. This is not the same as the C<offset> argument to
		2199	C<ev_periodic_set>, but indeed works even in interval and manual
		2200	rescheduling modes.
		2201
		2202	=item ev_tstamp offset [read-write]
		2203
		2204	When repeating, this contains the offset value, otherwise this is the
		2205	absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
		2206	although libev might modify this value for better numerical stability).
		2207
		2208	Can be modified any time, but changes only take effect when the periodic
		2209	timer fires or C<ev_periodic_again> is being called.
		2210
		2211	=item ev_tstamp interval [read-write]
		2212
		2213	The current interval value. Can be modified any time, but changes only
		2214	take effect when the periodic timer fires or C<ev_periodic_again> is being
		2215	called.
		2216
		2217	=item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
		2218
		2219	The current reschedule callback, or C<0>, if this functionality is
		2220	switched off. Can be changed any time, but changes only take effect when
		2221	the periodic timer fires or C<ev_periodic_again> is being called.
		2222
597	=back	2223	=back
598		2224
		2225	=head3 Examples
		2226
		2227	Example: Call a callback every hour, or, more precisely, whenever the
		2228	system time is divisible by 3600. The callback invocation times have
		2229	potentially a lot of jitter, but good long-term stability.
		2230
		2231	static void
		2232	clock_cb (struct ev_loop *loop, ev_periodic *w, int revents)
		2233	{
		2234	... its now a full hour (UTC, or TAI or whatever your clock follows)
		2235	}
		2236
		2237	ev_periodic hourly_tick;
		2238	ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
		2239	ev_periodic_start (loop, &hourly_tick);
		2240
		2241	Example: The same as above, but use a reschedule callback to do it:
		2242
		2243	#include <math.h>
		2244
		2245	static ev_tstamp
		2246	my_scheduler_cb (ev_periodic *w, ev_tstamp now)
		2247	{
		2248	return now + (3600. - fmod (now, 3600.));
		2249	}
		2250
		2251	ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
		2252
		2253	Example: Call a callback every hour, starting now:
		2254
		2255	ev_periodic hourly_tick;
		2256	ev_periodic_init (&hourly_tick, clock_cb,
		2257	fmod (ev_now (loop), 3600.), 3600., 0);
		2258	ev_periodic_start (loop, &hourly_tick);
		2259
		2260
599	=head2 C<ev_signal> - signal me when a signal gets signalled	2261	=head2 C<ev_signal> - signal me when a signal gets signalled!
600		2262
601	Signal watchers will trigger an event when the process receives a specific	2263	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	2264	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	2265	will try it's best to deliver signals synchronously, i.e. as part of the
604	normal event processing, like any other event.	2266	normal event processing, like any other event.
605		2267
		2268	If you want signals to be delivered truly asynchronously, just use
		2269	C<sigaction> as you would do without libev and forget about sharing
		2270	the signal. You can even use C<ev_async> from a signal handler to
		2271	synchronously wake up an event loop.
		2272
606	You can configure as many watchers as you like per signal. Only when the	2273	You can configure as many watchers as you like for the same signal, but
		2274	only within the same loop, i.e. you can watch for C<SIGINT> in your
		2275	default loop and for C<SIGIO> in another loop, but you cannot watch for
		2276	C<SIGINT> in both the default loop and another loop at the same time. At
		2277	the moment, C<SIGCHLD> is permanently tied to the default loop.
		2278
607	first watcher gets started will libev actually register a signal watcher	2279	When the first watcher gets started will libev actually register something
608	with the kernel (thus it coexists with your own signal handlers as long	2280	with the kernel (thus it coexists with your own signal handlers as long as
609	as you don't register any with libev). Similarly, when the last signal	2281	you don't register any with libev for the same signal).
610	watcher for a signal is stopped libev will reset the signal handler to	2282
611	SIG_DFL (regardless of what it was set to before).	2283	If possible and supported, libev will install its handlers with
		2284	C<SA_RESTART> (or equivalent) behaviour enabled, so system calls should
		2285	not be unduly interrupted. If you have a problem with system calls getting
		2286	interrupted by signals you can block all signals in an C<ev_check> watcher
		2287	and unblock them in an C<ev_prepare> watcher.
		2288
		2289	=head3 The special problem of inheritance over fork/execve/pthread_create
		2290
		2291	Both the signal mask (C<sigprocmask>) and the signal disposition
		2292	(C<sigaction>) are unspecified after starting a signal watcher (and after
		2293	stopping it again), that is, libev might or might not block the signal,
		2294	and might or might not set or restore the installed signal handler.
		2295
		2296	While this does not matter for the signal disposition (libev never
		2297	sets signals to C<SIG_IGN>, so handlers will be reset to C<SIG_DFL> on
		2298	C<execve>), this matters for the signal mask: many programs do not expect
		2299	certain signals to be blocked.
		2300
		2301	This means that before calling C<exec> (from the child) you should reset
		2302	the signal mask to whatever "default" you expect (all clear is a good
		2303	choice usually).
		2304
		2305	The simplest way to ensure that the signal mask is reset in the child is
		2306	to install a fork handler with C<pthread_atfork> that resets it. That will
		2307	catch fork calls done by libraries (such as the libc) as well.
		2308
		2309	In current versions of libev, the signal will not be blocked indefinitely
		2310	unless you use the C<signalfd> API (C<EV_SIGNALFD>). While this reduces
		2311	the window of opportunity for problems, it will not go away, as libev
		2312	I<has> to modify the signal mask, at least temporarily.
		2313
		2314	So I can't stress this enough: I<If you do not reset your signal mask when
		2315	you expect it to be empty, you have a race condition in your code>. This
		2316	is not a libev-specific thing, this is true for most event libraries.
		2317
		2318	=head3 Watcher-Specific Functions and Data Members
612		2319
613	=over 4	2320	=over 4
614		2321
615	=item ev_signal_init (ev_signal *, callback, int signum)	2322	=item ev_signal_init (ev_signal *, callback, int signum)
616		2323
617	=item ev_signal_set (ev_signal *, int signum)	2324	=item ev_signal_set (ev_signal *, int signum)
618		2325
619	Configures the watcher to trigger on the given signal number (usually one	2326	Configures the watcher to trigger on the given signal number (usually one
620	of the C<SIGxxx> constants).	2327	of the C<SIGxxx> constants).
621		2328
		2329	=item int signum [read-only]
		2330
		2331	The signal the watcher watches out for.
		2332
622	=back	2333	=back
623		2334
		2335	=head3 Examples
		2336
		2337	Example: Try to exit cleanly on SIGINT.
		2338
		2339	static void
		2340	sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
		2341	{
		2342	ev_break (loop, EVBREAK_ALL);
		2343	}
		2344
		2345	ev_signal signal_watcher;
		2346	ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
		2347	ev_signal_start (loop, &signal_watcher);
		2348
		2349
624	=head2 C<ev_child> - wait for pid status changes	2350	=head2 C<ev_child> - watch out for process status changes
625		2351
626	Child watchers trigger when your process receives a SIGCHLD in response to	2352	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).	2353	some child status changes (most typically when a child of yours dies or
		2354	exits). It is permissible to install a child watcher I<after> the child
		2355	has been forked (which implies it might have already exited), as long
		2356	as the event loop isn't entered (or is continued from a watcher), i.e.,
		2357	forking and then immediately registering a watcher for the child is fine,
		2358	but forking and registering a watcher a few event loop iterations later or
		2359	in the next callback invocation is not.
		2360
		2361	Only the default event loop is capable of handling signals, and therefore
		2362	you can only register child watchers in the default event loop.
		2363
		2364	Due to some design glitches inside libev, child watchers will always be
		2365	handled at maximum priority (their priority is set to C<EV_MAXPRI> by
		2366	libev)
		2367
		2368	=head3 Process Interaction
		2369
		2370	Libev grabs C<SIGCHLD> as soon as the default event loop is
		2371	initialised. This is necessary to guarantee proper behaviour even if the
		2372	first child watcher is started after the child exits. The occurrence
		2373	of C<SIGCHLD> is recorded asynchronously, but child reaping is done
		2374	synchronously as part of the event loop processing. Libev always reaps all
		2375	children, even ones not watched.
		2376
		2377	=head3 Overriding the Built-In Processing
		2378
		2379	Libev offers no special support for overriding the built-in child
		2380	processing, but if your application collides with libev's default child
		2381	handler, you can override it easily by installing your own handler for
		2382	C<SIGCHLD> after initialising the default loop, and making sure the
		2383	default loop never gets destroyed. You are encouraged, however, to use an
		2384	event-based approach to child reaping and thus use libev's support for
		2385	that, so other libev users can use C<ev_child> watchers freely.
		2386
		2387	=head3 Stopping the Child Watcher
		2388
		2389	Currently, the child watcher never gets stopped, even when the
		2390	child terminates, so normally one needs to stop the watcher in the
		2391	callback. Future versions of libev might stop the watcher automatically
		2392	when a child exit is detected (calling C<ev_child_stop> twice is not a
		2393	problem).
		2394
		2395	=head3 Watcher-Specific Functions and Data Members
628		2396
629	=over 4	2397	=over 4
630		2398
631	=item ev_child_init (ev_child *, callback, int pid)	2399	=item ev_child_init (ev_child *, callback, int pid, int trace)
632		2400
633	=item ev_child_set (ev_child *, int pid)	2401	=item ev_child_set (ev_child *, int pid, int trace)
634		2402
635	Configures the watcher to wait for status changes of process C<pid> (or	2403	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	2404	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	2405	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	2406	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	2407	C<waitpid> documentation). The C<rpid> member contains the pid of the
640	process causing the status change.	2408	process causing the status change. C<trace> must be either C<0> (only
		2409	activate the watcher when the process terminates) or C<1> (additionally
		2410	activate the watcher when the process is stopped or continued).
		2411
		2412	=item int pid [read-only]
		2413
		2414	The process id this watcher watches out for, or C<0>, meaning any process id.
		2415
		2416	=item int rpid [read-write]
		2417
		2418	The process id that detected a status change.
		2419
		2420	=item int rstatus [read-write]
		2421
		2422	The process exit/trace status caused by C<rpid> (see your systems
		2423	C<waitpid> and C<sys/wait.h> documentation for details).
641		2424
642	=back	2425	=back
643		2426
		2427	=head3 Examples
		2428
		2429	Example: C<fork()> a new process and install a child handler to wait for
		2430	its completion.
		2431
		2432	ev_child cw;
		2433
		2434	static void
		2435	child_cb (EV_P_ ev_child *w, int revents)
		2436	{
		2437	ev_child_stop (EV_A_ w);
		2438	printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
		2439	}
		2440
		2441	pid_t pid = fork ();
		2442
		2443	if (pid < 0)
		2444	// error
		2445	else if (pid == 0)
		2446	{
		2447	// the forked child executes here
		2448	exit (1);
		2449	}
		2450	else
		2451	{
		2452	ev_child_init (&cw, child_cb, pid, 0);
		2453	ev_child_start (EV_DEFAULT_ &cw);
		2454	}
		2455
		2456
		2457	=head2 C<ev_stat> - did the file attributes just change?
		2458
		2459	This watches a file system path for attribute changes. That is, it calls
		2460	C<stat> on that path in regular intervals (or when the OS says it changed)
		2461	and sees if it changed compared to the last time, invoking the callback if
		2462	it did.
		2463
		2464	The path does not need to exist: changing from "path exists" to "path does
		2465	not exist" is a status change like any other. The condition "path does not
		2466	exist" (or more correctly "path cannot be stat'ed") is signified by the
		2467	C<st_nlink> field being zero (which is otherwise always forced to be at
		2468	least one) and all the other fields of the stat buffer having unspecified
		2469	contents.
		2470
		2471	The path I<must not> end in a slash or contain special components such as
		2472	C<.> or C<..>. The path I<should> be absolute: If it is relative and
		2473	your working directory changes, then the behaviour is undefined.
		2474
		2475	Since there is no portable change notification interface available, the
		2476	portable implementation simply calls C<stat(2)> regularly on the path
		2477	to see if it changed somehow. You can specify a recommended polling
		2478	interval for this case. If you specify a polling interval of C<0> (highly
		2479	recommended!) then a I<suitable, unspecified default> value will be used
		2480	(which you can expect to be around five seconds, although this might
		2481	change dynamically). Libev will also impose a minimum interval which is
		2482	currently around C<0.1>, but that's usually overkill.
		2483
		2484	This watcher type is not meant for massive numbers of stat watchers,
		2485	as even with OS-supported change notifications, this can be
		2486	resource-intensive.
		2487
		2488	At the time of this writing, the only OS-specific interface implemented
		2489	is the Linux inotify interface (implementing kqueue support is left as an
		2490	exercise for the reader. Note, however, that the author sees no way of
		2491	implementing C<ev_stat> semantics with kqueue, except as a hint).
		2492
		2493	=head3 ABI Issues (Largefile Support)
		2494
		2495	Libev by default (unless the user overrides this) uses the default
		2496	compilation environment, which means that on systems with large file
		2497	support disabled by default, you get the 32 bit version of the stat
		2498	structure. When using the library from programs that change the ABI to
		2499	use 64 bit file offsets the programs will fail. In that case you have to
		2500	compile libev with the same flags to get binary compatibility. This is
		2501	obviously the case with any flags that change the ABI, but the problem is
		2502	most noticeably displayed with ev_stat and large file support.
		2503
		2504	The solution for this is to lobby your distribution maker to make large
		2505	file interfaces available by default (as e.g. FreeBSD does) and not
		2506	optional. Libev cannot simply switch on large file support because it has
		2507	to exchange stat structures with application programs compiled using the
		2508	default compilation environment.
		2509
		2510	=head3 Inotify and Kqueue
		2511
		2512	When C<inotify (7)> support has been compiled into libev and present at
		2513	runtime, it will be used to speed up change detection where possible. The
		2514	inotify descriptor will be created lazily when the first C<ev_stat>
		2515	watcher is being started.
		2516
		2517	Inotify presence does not change the semantics of C<ev_stat> watchers
		2518	except that changes might be detected earlier, and in some cases, to avoid
		2519	making regular C<stat> calls. Even in the presence of inotify support
		2520	there are many cases where libev has to resort to regular C<stat> polling,
		2521	but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
		2522	many bugs), the path exists (i.e. stat succeeds), and the path resides on
		2523	a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
		2524	xfs are fully working) libev usually gets away without polling.
		2525
		2526	There is no support for kqueue, as apparently it cannot be used to
		2527	implement this functionality, due to the requirement of having a file
		2528	descriptor open on the object at all times, and detecting renames, unlinks
		2529	etc. is difficult.
		2530
		2531	=head3 C<stat ()> is a synchronous operation
		2532
		2533	Libev doesn't normally do any kind of I/O itself, and so is not blocking
		2534	the process. The exception are C<ev_stat> watchers - those call C<stat
		2535	()>, which is a synchronous operation.
		2536
		2537	For local paths, this usually doesn't matter: unless the system is very
		2538	busy or the intervals between stat's are large, a stat call will be fast,
		2539	as the path data is usually in memory already (except when starting the
		2540	watcher).
		2541
		2542	For networked file systems, calling C<stat ()> can block an indefinite
		2543	time due to network issues, and even under good conditions, a stat call
		2544	often takes multiple milliseconds.
		2545
		2546	Therefore, it is best to avoid using C<ev_stat> watchers on networked
		2547	paths, although this is fully supported by libev.
		2548
		2549	=head3 The special problem of stat time resolution
		2550
		2551	The C<stat ()> system call only supports full-second resolution portably,
		2552	and even on systems where the resolution is higher, most file systems
		2553	still only support whole seconds.
		2554
		2555	That means that, if the time is the only thing that changes, you can
		2556	easily miss updates: on the first update, C<ev_stat> detects a change and
		2557	calls your callback, which does something. When there is another update
		2558	within the same second, C<ev_stat> will be unable to detect unless the
		2559	stat data does change in other ways (e.g. file size).
		2560
		2561	The solution to this is to delay acting on a change for slightly more
		2562	than a second (or till slightly after the next full second boundary), using
		2563	a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
		2564	ev_timer_again (loop, w)>).
		2565
		2566	The C<.02> offset is added to work around small timing inconsistencies
		2567	of some operating systems (where the second counter of the current time
		2568	might be be delayed. One such system is the Linux kernel, where a call to
		2569	C<gettimeofday> might return a timestamp with a full second later than
		2570	a subsequent C<time> call - if the equivalent of C<time ()> is used to
		2571	update file times then there will be a small window where the kernel uses
		2572	the previous second to update file times but libev might already execute
		2573	the timer callback).
		2574
		2575	=head3 Watcher-Specific Functions and Data Members
		2576
		2577	=over 4
		2578
		2579	=item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
		2580
		2581	=item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
		2582
		2583	Configures the watcher to wait for status changes of the given
		2584	C<path>. The C<interval> is a hint on how quickly a change is expected to
		2585	be detected and should normally be specified as C<0> to let libev choose
		2586	a suitable value. The memory pointed to by C<path> must point to the same
		2587	path for as long as the watcher is active.
		2588
		2589	The callback will receive an C<EV_STAT> event when a change was detected,
		2590	relative to the attributes at the time the watcher was started (or the
		2591	last change was detected).
		2592
		2593	=item ev_stat_stat (loop, ev_stat *)
		2594
		2595	Updates the stat buffer immediately with new values. If you change the
		2596	watched path in your callback, you could call this function to avoid
		2597	detecting this change (while introducing a race condition if you are not
		2598	the only one changing the path). Can also be useful simply to find out the
		2599	new values.
		2600
		2601	=item ev_statdata attr [read-only]
		2602
		2603	The most-recently detected attributes of the file. Although the type is
		2604	C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
		2605	suitable for your system, but you can only rely on the POSIX-standardised
		2606	members to be present. If the C<st_nlink> member is C<0>, then there was
		2607	some error while C<stat>ing the file.
		2608
		2609	=item ev_statdata prev [read-only]
		2610
		2611	The previous attributes of the file. The callback gets invoked whenever
		2612	C<prev> != C<attr>, or, more precisely, one or more of these members
		2613	differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
		2614	C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
		2615
		2616	=item ev_tstamp interval [read-only]
		2617
		2618	The specified interval.
		2619
		2620	=item const char *path [read-only]
		2621
		2622	The file system path that is being watched.
		2623
		2624	=back
		2625
		2626	=head3 Examples
		2627
		2628	Example: Watch C</etc/passwd> for attribute changes.
		2629
		2630	static void
		2631	passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
		2632	{
		2633	/* /etc/passwd changed in some way */
		2634	if (w->attr.st_nlink)
		2635	{
		2636	printf ("passwd current size %ld\n", (long)w->attr.st_size);
		2637	printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
		2638	printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
		2639	}
		2640	else
		2641	/* you shalt not abuse printf for puts */
		2642	puts ("wow, /etc/passwd is not there, expect problems. "
		2643	"if this is windows, they already arrived\n");
		2644	}
		2645
		2646	...
		2647	ev_stat passwd;
		2648
		2649	ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
		2650	ev_stat_start (loop, &passwd);
		2651
		2652	Example: Like above, but additionally use a one-second delay so we do not
		2653	miss updates (however, frequent updates will delay processing, too, so
		2654	one might do the work both on C<ev_stat> callback invocation I<and> on
		2655	C<ev_timer> callback invocation).
		2656
		2657	static ev_stat passwd;
		2658	static ev_timer timer;
		2659
		2660	static void
		2661	timer_cb (EV_P_ ev_timer *w, int revents)
		2662	{
		2663	ev_timer_stop (EV_A_ w);
		2664
		2665	/* now it's one second after the most recent passwd change */
		2666	}
		2667
		2668	static void
		2669	stat_cb (EV_P_ ev_stat *w, int revents)
		2670	{
		2671	/* reset the one-second timer */
		2672	ev_timer_again (EV_A_ &timer);
		2673	}
		2674
		2675	...
		2676	ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
		2677	ev_stat_start (loop, &passwd);
		2678	ev_timer_init (&timer, timer_cb, 0., 1.02);
		2679
		2680
644	=head2 C<ev_idle> - when you've got nothing better to do	2681	=head2 C<ev_idle> - when you've got nothing better to do...
645		2682
646	Idle watchers trigger events when there are no other events are pending	2683	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	2684	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,	2685	as receiving "events").
649	imagine) it will not be triggered. But when your process is idle all idle	2686
650	watchers are being called again and again, once per event loop iteration -	2687	That is, as long as your process is busy handling sockets or timeouts
		2688	(or even signals, imagine) of the same or higher priority it will not be
		2689	triggered. But when your process is idle (or only lower-priority watchers
		2690	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	2691	iteration - until stopped, that is, or your process receives more events
652	busy.	2692	and becomes busy again with higher priority stuff.
653		2693
654	The most noteworthy effect is that as long as any idle watchers are	2694	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.	2695	active, the process will not block when waiting for new events.
656		2696
657	Apart from keeping your process non-blocking (which is a useful	2697	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	2698	effect on its own sometimes), idle watchers are a good place to do
659	"pseudo-background processing", or delay processing stuff to after the	2699	"pseudo-background processing", or delay processing stuff to after the
660	event loop has handled all outstanding events.	2700	event loop has handled all outstanding events.
661		2701
		2702	=head3 Watcher-Specific Functions and Data Members
		2703
662	=over 4	2704	=over 4
663		2705
664	=item ev_idle_init (ev_signal *, callback)	2706	=item ev_idle_init (ev_idle *, callback)
665		2707
666	Initialises and configures the idle watcher - it has no parameters of any	2708	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,	2709	kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
668	believe me.	2710	believe me.
669		2711
670	=back	2712	=back
671		2713
		2714	=head3 Examples
		2715
		2716	Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
		2717	callback, free it. Also, use no error checking, as usual.
		2718
		2719	static void
		2720	idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
		2721	{
		2722	free (w);
		2723	// now do something you wanted to do when the program has
		2724	// no longer anything immediate to do.
		2725	}
		2726
		2727	ev_idle *idle_watcher = malloc (sizeof (ev_idle));
		2728	ev_idle_init (idle_watcher, idle_cb);
		2729	ev_idle_start (loop, idle_watcher);
		2730
		2731
672	=head2 C<ev_prepare> and C<ev_check> - customise your event loop	2732	=head2 C<ev_prepare> and C<ev_check> - customise your event loop!
673		2733
674	Prepare and check watchers are usually (but not always) used in tandem:	2734	Prepare and check watchers are usually (but not always) used in pairs:
675	Prepare watchers get invoked before the process blocks and check watchers	2735	prepare watchers get invoked before the process blocks and check watchers
676	afterwards.	2736	afterwards.
677		2737
		2738	You I<must not> call C<ev_run> or similar functions that enter
		2739	the current event loop from either C<ev_prepare> or C<ev_check>
		2740	watchers. Other loops than the current one are fine, however. The
		2741	rationale behind this is that you do not need to check for recursion in
		2742	those watchers, i.e. the sequence will always be C<ev_prepare>, blocking,
		2743	C<ev_check> so if you have one watcher of each kind they will always be
		2744	called in pairs bracketing the blocking call.
		2745
678	Their main purpose is to integrate other event mechanisms into libev. This	2746	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	2747	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.	2748	variable changes, implement your own watchers, integrate net-snmp or a
		2749	coroutine library and lots more. They are also occasionally useful if
		2750	you cache some data and want to flush it before blocking (for example,
		2751	in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
		2752	watcher).
681		2753
682	This is done by examining in each prepare call which file descriptors need	2754	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	2755	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	2756	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	2757	libraries provide exactly this functionality). Then, in the check watcher,
686	any events that occured (by checking the pending status of all watchers	2758	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	2759	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,	2760	I/O and timer callbacks will never actually be called (but must be valid
689	because you never know, you know?).	2761	nevertheless, because you never know, you know?).
690		2762
691	As another example, the Perl Coro module uses these hooks to integrate	2763	As another example, the Perl Coro module uses these hooks to integrate
692	coroutines into libev programs, by yielding to other active coroutines	2764	coroutines into libev programs, by yielding to other active coroutines
693	during each prepare and only letting the process block if no coroutines	2765	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	2766	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,	2767	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	2768	of lower priority, but only once, using idle watchers to keep the event
697	coroutines exist, thus mapping low-priority coroutines to idle/background	2769	loop from blocking if lower-priority coroutines are active, thus mapping
698	tasks).	2770	low-priority coroutines to idle/background tasks).
		2771
		2772	It is recommended to give C<ev_check> watchers highest (C<EV_MAXPRI>)
		2773	priority, to ensure that they are being run before any other watchers
		2774	after the poll (this doesn't matter for C<ev_prepare> watchers).
		2775
		2776	Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
		2777	activate ("feed") events into libev. While libev fully supports this, they
		2778	might get executed before other C<ev_check> watchers did their job. As
		2779	C<ev_check> watchers are often used to embed other (non-libev) event
		2780	loops those other event loops might be in an unusable state until their
		2781	C<ev_check> watcher ran (always remind yourself to coexist peacefully with
		2782	others).
		2783
		2784	=head3 Watcher-Specific Functions and Data Members
699		2785
700	=over 4	2786	=over 4
701		2787
702	=item ev_prepare_init (ev_prepare *, callback)	2788	=item ev_prepare_init (ev_prepare *, callback)
703		2789
704	=item ev_check_init (ev_check *, callback)	2790	=item ev_check_init (ev_check *, callback)
705		2791
706	Initialises and configures the prepare or check watcher - they have no	2792	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>	2793	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.	2794	macros, but using them is utterly, utterly, utterly and completely
		2795	pointless.
709		2796
710	=back	2797	=back
711		2798
		2799	=head3 Examples
		2800
		2801	There are a number of principal ways to embed other event loops or modules
		2802	into libev. Here are some ideas on how to include libadns into libev
		2803	(there is a Perl module named C<EV::ADNS> that does this, which you could
		2804	use as a working example. Another Perl module named C<EV::Glib> embeds a
		2805	Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
		2806	Glib event loop).
		2807
		2808	Method 1: Add IO watchers and a timeout watcher in a prepare handler,
		2809	and in a check watcher, destroy them and call into libadns. What follows
		2810	is pseudo-code only of course. This requires you to either use a low
		2811	priority for the check watcher or use C<ev_clear_pending> explicitly, as
		2812	the callbacks for the IO/timeout watchers might not have been called yet.
		2813
		2814	static ev_io iow [nfd];
		2815	static ev_timer tw;
		2816
		2817	static void
		2818	io_cb (struct ev_loop *loop, ev_io *w, int revents)
		2819	{
		2820	}
		2821
		2822	// create io watchers for each fd and a timer before blocking
		2823	static void
		2824	adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
		2825	{
		2826	int timeout = 3600000;
		2827	struct pollfd fds [nfd];
		2828	// actual code will need to loop here and realloc etc.
		2829	adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
		2830
		2831	/* the callback is illegal, but won't be called as we stop during check */
		2832	ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
		2833	ev_timer_start (loop, &tw);
		2834
		2835	// create one ev_io per pollfd
		2836	for (int i = 0; i < nfd; ++i)
		2837	{
		2838	ev_io_init (iow + i, io_cb, fds [i].fd,
		2839	((fds [i].events & POLLIN ? EV_READ : 0)
		2840	| (fds [i].events & POLLOUT ? EV_WRITE : 0)));
		2841
		2842	fds [i].revents = 0;
		2843	ev_io_start (loop, iow + i);
		2844	}
		2845	}
		2846
		2847	// stop all watchers after blocking
		2848	static void
		2849	adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
		2850	{
		2851	ev_timer_stop (loop, &tw);
		2852
		2853	for (int i = 0; i < nfd; ++i)
		2854	{
		2855	// set the relevant poll flags
		2856	// could also call adns_processreadable etc. here
		2857	struct pollfd *fd = fds + i;
		2858	int revents = ev_clear_pending (iow + i);
		2859	if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
		2860	if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
		2861
		2862	// now stop the watcher
		2863	ev_io_stop (loop, iow + i);
		2864	}
		2865
		2866	adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
		2867	}
		2868
		2869	Method 2: This would be just like method 1, but you run C<adns_afterpoll>
		2870	in the prepare watcher and would dispose of the check watcher.
		2871
		2872	Method 3: If the module to be embedded supports explicit event
		2873	notification (libadns does), you can also make use of the actual watcher
		2874	callbacks, and only destroy/create the watchers in the prepare watcher.
		2875
		2876	static void
		2877	timer_cb (EV_P_ ev_timer *w, int revents)
		2878	{
		2879	adns_state ads = (adns_state)w->data;
		2880	update_now (EV_A);
		2881
		2882	adns_processtimeouts (ads, &tv_now);
		2883	}
		2884
		2885	static void
		2886	io_cb (EV_P_ ev_io *w, int revents)
		2887	{
		2888	adns_state ads = (adns_state)w->data;
		2889	update_now (EV_A);
		2890
		2891	if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
		2892	if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
		2893	}
		2894
		2895	// do not ever call adns_afterpoll
		2896
		2897	Method 4: Do not use a prepare or check watcher because the module you
		2898	want to embed is not flexible enough to support it. Instead, you can
		2899	override their poll function. The drawback with this solution is that the
		2900	main loop is now no longer controllable by EV. The C<Glib::EV> module uses
		2901	this approach, effectively embedding EV as a client into the horrible
		2902	libglib event loop.
		2903
		2904	static gint
		2905	event_poll_func (GPollFD *fds, guint nfds, gint timeout)
		2906	{
		2907	int got_events = 0;
		2908
		2909	for (n = 0; n < nfds; ++n)
		2910	// create/start io watcher that sets the relevant bits in fds[n] and increment got_events
		2911
		2912	if (timeout >= 0)
		2913	// create/start timer
		2914
		2915	// poll
		2916	ev_run (EV_A_ 0);
		2917
		2918	// stop timer again
		2919	if (timeout >= 0)
		2920	ev_timer_stop (EV_A_ &to);
		2921
		2922	// stop io watchers again - their callbacks should have set
		2923	for (n = 0; n < nfds; ++n)
		2924	ev_io_stop (EV_A_ iow [n]);
		2925
		2926	return got_events;
		2927	}
		2928
		2929
		2930	=head2 C<ev_embed> - when one backend isn't enough...
		2931
		2932	This is a rather advanced watcher type that lets you embed one event loop
		2933	into another (currently only C<ev_io> events are supported in the embedded
		2934	loop, other types of watchers might be handled in a delayed or incorrect
		2935	fashion and must not be used).
		2936
		2937	There are primarily two reasons you would want that: work around bugs and
		2938	prioritise I/O.
		2939
		2940	As an example for a bug workaround, the kqueue backend might only support
		2941	sockets on some platform, so it is unusable as generic backend, but you
		2942	still want to make use of it because you have many sockets and it scales
		2943	so nicely. In this case, you would create a kqueue-based loop and embed
		2944	it into your default loop (which might use e.g. poll). Overall operation
		2945	will be a bit slower because first libev has to call C<poll> and then
		2946	C<kevent>, but at least you can use both mechanisms for what they are
		2947	best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
		2948
		2949	As for prioritising I/O: under rare circumstances you have the case where
		2950	some fds have to be watched and handled very quickly (with low latency),
		2951	and even priorities and idle watchers might have too much overhead. In
		2952	this case you would put all the high priority stuff in one loop and all
		2953	the rest in a second one, and embed the second one in the first.
		2954
		2955	As long as the watcher is active, the callback will be invoked every
		2956	time there might be events pending in the embedded loop. The callback
		2957	must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
		2958	sweep and invoke their callbacks (the callback doesn't need to invoke the
		2959	C<ev_embed_sweep> function directly, it could also start an idle watcher
		2960	to give the embedded loop strictly lower priority for example).
		2961
		2962	You can also set the callback to C<0>, in which case the embed watcher
		2963	will automatically execute the embedded loop sweep whenever necessary.
		2964
		2965	Fork detection will be handled transparently while the C<ev_embed> watcher
		2966	is active, i.e., the embedded loop will automatically be forked when the
		2967	embedding loop forks. In other cases, the user is responsible for calling
		2968	C<ev_loop_fork> on the embedded loop.
		2969
		2970	Unfortunately, not all backends are embeddable: only the ones returned by
		2971	C<ev_embeddable_backends> are, which, unfortunately, does not include any
		2972	portable one.
		2973
		2974	So when you want to use this feature you will always have to be prepared
		2975	that you cannot get an embeddable loop. The recommended way to get around
		2976	this is to have a separate variables for your embeddable loop, try to
		2977	create it, and if that fails, use the normal loop for everything.
		2978
		2979	=head3 C<ev_embed> and fork
		2980
		2981	While the C<ev_embed> watcher is running, forks in the embedding loop will
		2982	automatically be applied to the embedded loop as well, so no special
		2983	fork handling is required in that case. When the watcher is not running,
		2984	however, it is still the task of the libev user to call C<ev_loop_fork ()>
		2985	as applicable.
		2986
		2987	=head3 Watcher-Specific Functions and Data Members
		2988
		2989	=over 4
		2990
		2991	=item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
		2992
		2993	=item ev_embed_set (ev_embed *, callback, struct ev_loop *embedded_loop)
		2994
		2995	Configures the watcher to embed the given loop, which must be
		2996	embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
		2997	invoked automatically, otherwise it is the responsibility of the callback
		2998	to invoke it (it will continue to be called until the sweep has been done,
		2999	if you do not want that, you need to temporarily stop the embed watcher).
		3000
		3001	=item ev_embed_sweep (loop, ev_embed *)
		3002
		3003	Make a single, non-blocking sweep over the embedded loop. This works
		3004	similarly to C<ev_run (embedded_loop, EVRUN_NOWAIT)>, but in the most
		3005	appropriate way for embedded loops.
		3006
		3007	=item struct ev_loop *other [read-only]
		3008
		3009	The embedded event loop.
		3010
		3011	=back
		3012
		3013	=head3 Examples
		3014
		3015	Example: Try to get an embeddable event loop and embed it into the default
		3016	event loop. If that is not possible, use the default loop. The default
		3017	loop is stored in C<loop_hi>, while the embeddable loop is stored in
		3018	C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
		3019	used).
		3020
		3021	struct ev_loop *loop_hi = ev_default_init (0);
		3022	struct ev_loop *loop_lo = 0;
		3023	ev_embed embed;
		3024
		3025	// see if there is a chance of getting one that works
		3026	// (remember that a flags value of 0 means autodetection)
		3027	loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
		3028	? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
		3029	: 0;
		3030
		3031	// if we got one, then embed it, otherwise default to loop_hi
		3032	if (loop_lo)
		3033	{
		3034	ev_embed_init (&embed, 0, loop_lo);
		3035	ev_embed_start (loop_hi, &embed);
		3036	}
		3037	else
		3038	loop_lo = loop_hi;
		3039
		3040	Example: Check if kqueue is available but not recommended and create
		3041	a kqueue backend for use with sockets (which usually work with any
		3042	kqueue implementation). Store the kqueue/socket-only event loop in
		3043	C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
		3044
		3045	struct ev_loop *loop = ev_default_init (0);
		3046	struct ev_loop *loop_socket = 0;
		3047	ev_embed embed;
		3048
		3049	if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
		3050	if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
		3051	{
		3052	ev_embed_init (&embed, 0, loop_socket);
		3053	ev_embed_start (loop, &embed);
		3054	}
		3055
		3056	if (!loop_socket)
		3057	loop_socket = loop;
		3058
		3059	// now use loop_socket for all sockets, and loop for everything else
		3060
		3061
		3062	=head2 C<ev_fork> - the audacity to resume the event loop after a fork
		3063
		3064	Fork watchers are called when a C<fork ()> was detected (usually because
		3065	whoever is a good citizen cared to tell libev about it by calling
		3066	C<ev_default_fork> or C<ev_loop_fork>). The invocation is done before the
		3067	event loop blocks next and before C<ev_check> watchers are being called,
		3068	and only in the child after the fork. If whoever good citizen calling
		3069	C<ev_default_fork> cheats and calls it in the wrong process, the fork
		3070	handlers will be invoked, too, of course.
		3071
		3072	=head3 The special problem of life after fork - how is it possible?
		3073
		3074	Most uses of C<fork()> consist of forking, then some simple calls to set
		3075	up/change the process environment, followed by a call to C<exec()>. This
		3076	sequence should be handled by libev without any problems.
		3077
		3078	This changes when the application actually wants to do event handling
		3079	in the child, or both parent in child, in effect "continuing" after the
		3080	fork.
		3081
		3082	The default mode of operation (for libev, with application help to detect
		3083	forks) is to duplicate all the state in the child, as would be expected
		3084	when I<either> the parent I<or> the child process continues.
		3085
		3086	When both processes want to continue using libev, then this is usually the
		3087	wrong result. In that case, usually one process (typically the parent) is
		3088	supposed to continue with all watchers in place as before, while the other
		3089	process typically wants to start fresh, i.e. without any active watchers.
		3090
		3091	The cleanest and most efficient way to achieve that with libev is to
		3092	simply create a new event loop, which of course will be "empty", and
		3093	use that for new watchers. This has the advantage of not touching more
		3094	memory than necessary, and thus avoiding the copy-on-write, and the
		3095	disadvantage of having to use multiple event loops (which do not support
		3096	signal watchers).
		3097
		3098	When this is not possible, or you want to use the default loop for
		3099	other reasons, then in the process that wants to start "fresh", call
		3100	C<ev_loop_destroy (EV_DEFAULT)> followed by C<ev_default_loop (...)>.
		3101	Destroying the default loop will "orphan" (not stop) all registered
		3102	watchers, so you have to be careful not to execute code that modifies
		3103	those watchers. Note also that in that case, you have to re-register any
		3104	signal watchers.
		3105
		3106	=head3 Watcher-Specific Functions and Data Members
		3107
		3108	=over 4
		3109
		3110	=item ev_fork_init (ev_fork *, callback)
		3111
		3112	Initialises and configures the fork watcher - it has no parameters of any
		3113	kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
		3114	really.
		3115
		3116	=back
		3117
		3118
		3119	=head2 C<ev_cleanup> - even the best things end
		3120
		3121	Cleanup watchers are called just before the event loop is being destroyed
		3122	by a call to C<ev_loop_destroy>.
		3123
		3124	While there is no guarantee that the event loop gets destroyed, cleanup
		3125	watchers provide a convenient method to install cleanup hooks for your
		3126	program, worker threads and so on - you just to make sure to destroy the
		3127	loop when you want them to be invoked.
		3128
		3129	Cleanup watchers are invoked in the same way as any other watcher. Unlike
		3130	all other watchers, they do not keep a reference to the event loop (which
		3131	makes a lot of sense if you think about it). Like all other watchers, you
		3132	can call libev functions in the callback, except C<ev_cleanup_start>.
		3133
		3134	=head3 Watcher-Specific Functions and Data Members
		3135
		3136	=over 4
		3137
		3138	=item ev_cleanup_init (ev_cleanup *, callback)
		3139
		3140	Initialises and configures the cleanup watcher - it has no parameters of
		3141	any kind. There is a C<ev_cleanup_set> macro, but using it is utterly
		3142	pointless, I assure you.
		3143
		3144	=back
		3145
		3146	Example: Register an atexit handler to destroy the default loop, so any
		3147	cleanup functions are called.
		3148
		3149	static void
		3150	program_exits (void)
		3151	{
		3152	ev_loop_destroy (EV_DEFAULT_UC);
		3153	}
		3154
		3155	...
		3156	atexit (program_exits);
		3157
		3158
		3159	=head2 C<ev_async> - how to wake up an event loop
		3160
		3161	In general, you cannot use an C<ev_run> from multiple threads or other
		3162	asynchronous sources such as signal handlers (as opposed to multiple event
		3163	loops - those are of course safe to use in different threads).
		3164
		3165	Sometimes, however, you need to wake up an event loop you do not control,
		3166	for example because it belongs to another thread. This is what C<ev_async>
		3167	watchers do: as long as the C<ev_async> watcher is active, you can signal
		3168	it by calling C<ev_async_send>, which is thread- and signal safe.
		3169
		3170	This functionality is very similar to C<ev_signal> watchers, as signals,
		3171	too, are asynchronous in nature, and signals, too, will be compressed
		3172	(i.e. the number of callback invocations may be less than the number of
		3173	C<ev_async_sent> calls).
		3174
		3175	Unlike C<ev_signal> watchers, C<ev_async> works with any event loop, not
		3176	just the default loop.
		3177
		3178	=head3 Queueing
		3179
		3180	C<ev_async> does not support queueing of data in any way. The reason
		3181	is that the author does not know of a simple (or any) algorithm for a
		3182	multiple-writer-single-reader queue that works in all cases and doesn't
		3183	need elaborate support such as pthreads or unportable memory access
		3184	semantics.
		3185
		3186	That means that if you want to queue data, you have to provide your own
		3187	queue. But at least I can tell you how to implement locking around your
		3188	queue:
		3189
		3190	=over 4
		3191
		3192	=item queueing from a signal handler context
		3193
		3194	To implement race-free queueing, you simply add to the queue in the signal
		3195	handler but you block the signal handler in the watcher callback. Here is
		3196	an example that does that for some fictitious SIGUSR1 handler:
		3197
		3198	static ev_async mysig;
		3199
		3200	static void
		3201	sigusr1_handler (void)
		3202	{
		3203	sometype data;
		3204
		3205	// no locking etc.
		3206	queue_put (data);
		3207	ev_async_send (EV_DEFAULT_ &mysig);
		3208	}
		3209
		3210	static void
		3211	mysig_cb (EV_P_ ev_async *w, int revents)
		3212	{
		3213	sometype data;
		3214	sigset_t block, prev;
		3215
		3216	sigemptyset (&block);
		3217	sigaddset (&block, SIGUSR1);
		3218	sigprocmask (SIG_BLOCK, &block, &prev);
		3219
		3220	while (queue_get (&data))
		3221	process (data);
		3222
		3223	if (sigismember (&prev, SIGUSR1)
		3224	sigprocmask (SIG_UNBLOCK, &block, 0);
		3225	}
		3226
		3227	(Note: pthreads in theory requires you to use C<pthread_setmask>
		3228	instead of C<sigprocmask> when you use threads, but libev doesn't do it
		3229	either...).
		3230
		3231	=item queueing from a thread context
		3232
		3233	The strategy for threads is different, as you cannot (easily) block
		3234	threads but you can easily preempt them, so to queue safely you need to
		3235	employ a traditional mutex lock, such as in this pthread example:
		3236
		3237	static ev_async mysig;
		3238	static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
		3239
		3240	static void
		3241	otherthread (void)
		3242	{
		3243	// only need to lock the actual queueing operation
		3244	pthread_mutex_lock (&mymutex);
		3245	queue_put (data);
		3246	pthread_mutex_unlock (&mymutex);
		3247
		3248	ev_async_send (EV_DEFAULT_ &mysig);
		3249	}
		3250
		3251	static void
		3252	mysig_cb (EV_P_ ev_async *w, int revents)
		3253	{
		3254	pthread_mutex_lock (&mymutex);
		3255
		3256	while (queue_get (&data))
		3257	process (data);
		3258
		3259	pthread_mutex_unlock (&mymutex);
		3260	}
		3261
		3262	=back
		3263
		3264
		3265	=head3 Watcher-Specific Functions and Data Members
		3266
		3267	=over 4
		3268
		3269	=item ev_async_init (ev_async *, callback)
		3270
		3271	Initialises and configures the async watcher - it has no parameters of any
		3272	kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
		3273	trust me.
		3274
		3275	=item ev_async_send (loop, ev_async *)
		3276
		3277	Sends/signals/activates the given C<ev_async> watcher, that is, feeds
		3278	an C<EV_ASYNC> event on the watcher into the event loop. Unlike
		3279	C<ev_feed_event>, this call is safe to do from other threads, signal or
		3280	similar contexts (see the discussion of C<EV_ATOMIC_T> in the embedding
		3281	section below on what exactly this means).
		3282
		3283	Note that, as with other watchers in libev, multiple events might get
		3284	compressed into a single callback invocation (another way to look at this
		3285	is that C<ev_async> watchers are level-triggered, set on C<ev_async_send>,
		3286	reset when the event loop detects that).
		3287
		3288	This call incurs the overhead of a system call only once per event loop
		3289	iteration, so while the overhead might be noticeable, it doesn't apply to
		3290	repeated calls to C<ev_async_send> for the same event loop.
		3291
		3292	=item bool = ev_async_pending (ev_async *)
		3293
		3294	Returns a non-zero value when C<ev_async_send> has been called on the
		3295	watcher but the event has not yet been processed (or even noted) by the
		3296	event loop.
		3297
		3298	C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
		3299	the loop iterates next and checks for the watcher to have become active,
		3300	it will reset the flag again. C<ev_async_pending> can be used to very
		3301	quickly check whether invoking the loop might be a good idea.
		3302
		3303	Not that this does I<not> check whether the watcher itself is pending,
		3304	only whether it has been requested to make this watcher pending: there
		3305	is a time window between the event loop checking and resetting the async
		3306	notification, and the callback being invoked.
		3307
		3308	=back
		3309
		3310
712	=head1 OTHER FUNCTIONS	3311	=head1 OTHER FUNCTIONS
713		3312
714	There are some other functions of possible interest. Described. Here. Now.	3313	There are some other functions of possible interest. Described. Here. Now.
715		3314
716	=over 4	3315	=over 4
717		3316
718	=item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)	3317	=item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)
719		3318
720	This function combines a simple timer and an I/O watcher, calls your	3319	This function combines a simple timer and an I/O watcher, calls your
721	callback on whichever event happens first and automatically stop both	3320	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	3321	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	3322	or timeout without having to allocate/configure/start/stop/free one or
724	more watchers yourself.	3323	more watchers yourself.
725		3324
726	If C<fd> is less than 0, then no I/O watcher will be started and events	3325	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	3326	C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
728	C<events> set will be craeted and started.	3327	the given C<fd> and C<events> set will be created and started.
729		3328
730	If C<timeout> is less than 0, then no timeout watcher will be	3329	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	3330	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	3331	repeat = 0) will be started. C<0> is a valid timeout.
733	dubious value.
734		3332
735	The callback has the type C<void (*cb)(int revents, void *arg)> and gets	3333	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	3334	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>	3335	C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMER>) and the C<arg>
738	value passed to C<ev_once>:	3336	value passed to C<ev_once>. Note that it is possible to receive I<both>
		3337	a timeout and an io event at the same time - you probably should give io
		3338	events precedence.
739		3339
		3340	Example: wait up to ten seconds for data to appear on STDIN_FILENO.
		3341
740	static void stdin_ready (int revents, void *arg)	3342	static void stdin_ready (int revents, void *arg)
741	{	3343	{
742	if (revents & EV_TIMEOUT)
743	/* doh, nothing entered */;
744	else if (revents & EV_READ)	3344	if (revents & EV_READ)
745	/* stdin might have data for us, joy! */;	3345	/* stdin might have data for us, joy! */;
		3346	else if (revents & EV_TIMER)
		3347	/* doh, nothing entered */;
746	}	3348	}
747		3349
748	ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);	3350	ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
749
750	=item ev_feed_event (loop, watcher, int events)
751
752	Feeds the given event set into the event loop, as if the specified event
753	had happened for the specified watcher (which must be a pointer to an
754	initialised but not necessarily started event watcher).
755		3351
756	=item ev_feed_fd_event (loop, int fd, int revents)	3352	=item ev_feed_fd_event (loop, int fd, int revents)
757		3353
758	Feed an event on the given fd, as if a file descriptor backend detected	3354	Feed an event on the given fd, as if a file descriptor backend detected
759	the given events it.	3355	the given events it.
760		3356
761	=item ev_feed_signal_event (loop, int signum)	3357	=item ev_feed_signal_event (loop, int signum)
762		3358
763	Feed an event as if the given signal occured (loop must be the default loop!).	3359	Feed an event as if the given signal occurred (C<loop> must be the default
		3360	loop!).
764		3361
765	=back	3362	=back
766		3363
		3364
		3365	=head1 LIBEVENT EMULATION
		3366
		3367	Libev offers a compatibility emulation layer for libevent. It cannot
		3368	emulate the internals of libevent, so here are some usage hints:
		3369
		3370	=over 4
		3371
		3372	=item * Use it by including <event.h>, as usual.
		3373
		3374	=item * The following members are fully supported: ev_base, ev_callback,
		3375	ev_arg, ev_fd, ev_res, ev_events.
		3376
		3377	=item * Avoid using ev_flags and the EVLIST_*-macros, while it is
		3378	maintained by libev, it does not work exactly the same way as in libevent (consider
		3379	it a private API).
		3380
		3381	=item * Priorities are not currently supported. Initialising priorities
		3382	will fail and all watchers will have the same priority, even though there
		3383	is an ev_pri field.
		3384
		3385	=item * In libevent, the last base created gets the signals, in libev, the
		3386	first base created (== the default loop) gets the signals.
		3387
		3388	=item * Other members are not supported.
		3389
		3390	=item * The libev emulation is I<not> ABI compatible to libevent, you need
		3391	to use the libev header file and library.
		3392
		3393	=back
		3394
		3395	=head1 C++ SUPPORT
		3396
		3397	Libev comes with some simplistic wrapper classes for C++ that mainly allow
		3398	you to use some convenience methods to start/stop watchers and also change
		3399	the callback model to a model using method callbacks on objects.
		3400
		3401	To use it,
		3402
		3403	#include <ev++.h>
		3404
		3405	This automatically includes F<ev.h> and puts all of its definitions (many
		3406	of them macros) into the global namespace. All C++ specific things are
		3407	put into the C<ev> namespace. It should support all the same embedding
		3408	options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
		3409
		3410	Care has been taken to keep the overhead low. The only data member the C++
		3411	classes add (compared to plain C-style watchers) is the event loop pointer
		3412	that the watcher is associated with (or no additional members at all if
		3413	you disable C<EV_MULTIPLICITY> when embedding libev).
		3414
		3415	Currently, functions, and static and non-static member functions can be
		3416	used as callbacks. Other types should be easy to add as long as they only
		3417	need one additional pointer for context. If you need support for other
		3418	types of functors please contact the author (preferably after implementing
		3419	it).
		3420
		3421	Here is a list of things available in the C<ev> namespace:
		3422
		3423	=over 4
		3424
		3425	=item C<ev::READ>, C<ev::WRITE> etc.
		3426
		3427	These are just enum values with the same values as the C<EV_READ> etc.
		3428	macros from F<ev.h>.
		3429
		3430	=item C<ev::tstamp>, C<ev::now>
		3431
		3432	Aliases to the same types/functions as with the C<ev_> prefix.
		3433
		3434	=item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
		3435
		3436	For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
		3437	the same name in the C<ev> namespace, with the exception of C<ev_signal>
		3438	which is called C<ev::sig> to avoid clashes with the C<signal> macro
		3439	defines by many implementations.
		3440
		3441	All of those classes have these methods:
		3442
		3443	=over 4
		3444
		3445	=item ev::TYPE::TYPE ()
		3446
		3447	=item ev::TYPE::TYPE (loop)
		3448
		3449	=item ev::TYPE::~TYPE
		3450
		3451	The constructor (optionally) takes an event loop to associate the watcher
		3452	with. If it is omitted, it will use C<EV_DEFAULT>.
		3453
		3454	The constructor calls C<ev_init> for you, which means you have to call the
		3455	C<set> method before starting it.
		3456
		3457	It will not set a callback, however: You have to call the templated C<set>
		3458	method to set a callback before you can start the watcher.
		3459
		3460	(The reason why you have to use a method is a limitation in C++ which does
		3461	not allow explicit template arguments for constructors).
		3462
		3463	The destructor automatically stops the watcher if it is active.
		3464
		3465	=item w->set<class, &class::method> (object *)
		3466
		3467	This method sets the callback method to call. The method has to have a
		3468	signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
		3469	first argument and the C<revents> as second. The object must be given as
		3470	parameter and is stored in the C<data> member of the watcher.
		3471
		3472	This method synthesizes efficient thunking code to call your method from
		3473	the C callback that libev requires. If your compiler can inline your
		3474	callback (i.e. it is visible to it at the place of the C<set> call and
		3475	your compiler is good :), then the method will be fully inlined into the
		3476	thunking function, making it as fast as a direct C callback.
		3477
		3478	Example: simple class declaration and watcher initialisation
		3479
		3480	struct myclass
		3481	{
		3482	void io_cb (ev::io &w, int revents) { }
		3483	}
		3484
		3485	myclass obj;
		3486	ev::io iow;
		3487	iow.set <myclass, &myclass::io_cb> (&obj);
		3488
		3489	=item w->set (object *)
		3490
		3491	This is a variation of a method callback - leaving out the method to call
		3492	will default the method to C<operator ()>, which makes it possible to use
		3493	functor objects without having to manually specify the C<operator ()> all
		3494	the time. Incidentally, you can then also leave out the template argument
		3495	list.
		3496
		3497	The C<operator ()> method prototype must be C<void operator ()(watcher &w,
		3498	int revents)>.
		3499
		3500	See the method-C<set> above for more details.
		3501
		3502	Example: use a functor object as callback.
		3503
		3504	struct myfunctor
		3505	{
		3506	void operator() (ev::io &w, int revents)
		3507	{
		3508	...
		3509	}
		3510	}
		3511
		3512	myfunctor f;
		3513
		3514	ev::io w;
		3515	w.set (&f);
		3516
		3517	=item w->set<function> (void *data = 0)
		3518
		3519	Also sets a callback, but uses a static method or plain function as
		3520	callback. The optional C<data> argument will be stored in the watcher's
		3521	C<data> member and is free for you to use.
		3522
		3523	The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
		3524
		3525	See the method-C<set> above for more details.
		3526
		3527	Example: Use a plain function as callback.
		3528
		3529	static void io_cb (ev::io &w, int revents) { }
		3530	iow.set <io_cb> ();
		3531
		3532	=item w->set (loop)
		3533
		3534	Associates a different C<struct ev_loop> with this watcher. You can only
		3535	do this when the watcher is inactive (and not pending either).
		3536
		3537	=item w->set ([arguments])
		3538
		3539	Basically the same as C<ev_TYPE_set>, with the same arguments. Either this
		3540	method or a suitable start method must be called at least once. Unlike the
		3541	C counterpart, an active watcher gets automatically stopped and restarted
		3542	when reconfiguring it with this method.
		3543
		3544	=item w->start ()
		3545
		3546	Starts the watcher. Note that there is no C<loop> argument, as the
		3547	constructor already stores the event loop.
		3548
		3549	=item w->start ([arguments])
		3550
		3551	Instead of calling C<set> and C<start> methods separately, it is often
		3552	convenient to wrap them in one call. Uses the same type of arguments as
		3553	the configure C<set> method of the watcher.
		3554
		3555	=item w->stop ()
		3556
		3557	Stops the watcher if it is active. Again, no C<loop> argument.
		3558
		3559	=item w->again () (C<ev::timer>, C<ev::periodic> only)
		3560
		3561	For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
		3562	C<ev_TYPE_again> function.
		3563
		3564	=item w->sweep () (C<ev::embed> only)
		3565
		3566	Invokes C<ev_embed_sweep>.
		3567
		3568	=item w->update () (C<ev::stat> only)
		3569
		3570	Invokes C<ev_stat_stat>.
		3571
		3572	=back
		3573
		3574	=back
		3575
		3576	Example: Define a class with two I/O and idle watchers, start the I/O
		3577	watchers in the constructor.
		3578
		3579	class myclass
		3580	{
		3581	ev::io io ; void io_cb (ev::io &w, int revents);
		3582	ev::io2 io2 ; void io2_cb (ev::io &w, int revents);
		3583	ev::idle idle; void idle_cb (ev::idle &w, int revents);
		3584
		3585	myclass (int fd)
		3586	{
		3587	io .set <myclass, &myclass::io_cb > (this);
		3588	io2 .set <myclass, &myclass::io2_cb > (this);
		3589	idle.set <myclass, &myclass::idle_cb> (this);
		3590
		3591	io.set (fd, ev::WRITE); // configure the watcher
		3592	io.start (); // start it whenever convenient
		3593
		3594	io2.start (fd, ev::READ); // set + start in one call
		3595	}
		3596	};
		3597
		3598
		3599	=head1 OTHER LANGUAGE BINDINGS
		3600
		3601	Libev does not offer other language bindings itself, but bindings for a
		3602	number of languages exist in the form of third-party packages. If you know
		3603	any interesting language binding in addition to the ones listed here, drop
		3604	me a note.
		3605
		3606	=over 4
		3607
		3608	=item Perl
		3609
		3610	The EV module implements the full libev API and is actually used to test
		3611	libev. EV is developed together with libev. Apart from the EV core module,
		3612	there are additional modules that implement libev-compatible interfaces
		3613	to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
		3614	C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
		3615	and C<EV::Glib>).
		3616
		3617	It can be found and installed via CPAN, its homepage is at
		3618	L<http://software.schmorp.de/pkg/EV>.
		3619
		3620	=item Python
		3621
		3622	Python bindings can be found at L<http://code.google.com/p/pyev/>. It
		3623	seems to be quite complete and well-documented.
		3624
		3625	=item Ruby
		3626
		3627	Tony Arcieri has written a ruby extension that offers access to a subset
		3628	of the libev API and adds file handle abstractions, asynchronous DNS and
		3629	more on top of it. It can be found via gem servers. Its homepage is at
		3630	L<http://rev.rubyforge.org/>.
		3631
		3632	Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
		3633	makes rev work even on mingw.
		3634
		3635	=item Haskell
		3636
		3637	A haskell binding to libev is available at
		3638	L<http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
		3639
		3640	=item D
		3641
		3642	Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
		3643	be found at L<http://proj.llucax.com.ar/wiki/evd>.
		3644
		3645	=item Ocaml
		3646
		3647	Erkki Seppala has written Ocaml bindings for libev, to be found at
		3648	L<http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
		3649
		3650	=item Lua
		3651
		3652	Brian Maher has written a partial interface to libev for lua (at the
		3653	time of this writing, only C<ev_io> and C<ev_timer>), to be found at
		3654	L<http://github.com/brimworks/lua-ev>.
		3655
		3656	=back
		3657
		3658
		3659	=head1 MACRO MAGIC
		3660
		3661	Libev can be compiled with a variety of options, the most fundamental
		3662	of which is C<EV_MULTIPLICITY>. This option determines whether (most)
		3663	functions and callbacks have an initial C<struct ev_loop *> argument.
		3664
		3665	To make it easier to write programs that cope with either variant, the
		3666	following macros are defined:
		3667
		3668	=over 4
		3669
		3670	=item C<EV_A>, C<EV_A_>
		3671
		3672	This provides the loop I<argument> for functions, if one is required ("ev
		3673	loop argument"). The C<EV_A> form is used when this is the sole argument,
		3674	C<EV_A_> is used when other arguments are following. Example:
		3675
		3676	ev_unref (EV_A);
		3677	ev_timer_add (EV_A_ watcher);
		3678	ev_run (EV_A_ 0);
		3679
		3680	It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
		3681	which is often provided by the following macro.
		3682
		3683	=item C<EV_P>, C<EV_P_>
		3684
		3685	This provides the loop I<parameter> for functions, if one is required ("ev
		3686	loop parameter"). The C<EV_P> form is used when this is the sole parameter,
		3687	C<EV_P_> is used when other parameters are following. Example:
		3688
		3689	// this is how ev_unref is being declared
		3690	static void ev_unref (EV_P);
		3691
		3692	// this is how you can declare your typical callback
		3693	static void cb (EV_P_ ev_timer *w, int revents)
		3694
		3695	It declares a parameter C<loop> of type C<struct ev_loop *>, quite
		3696	suitable for use with C<EV_A>.
		3697
		3698	=item C<EV_DEFAULT>, C<EV_DEFAULT_>
		3699
		3700	Similar to the other two macros, this gives you the value of the default
		3701	loop, if multiple loops are supported ("ev loop default").
		3702
		3703	=item C<EV_DEFAULT_UC>, C<EV_DEFAULT_UC_>
		3704
		3705	Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
		3706	default loop has been initialised (C<UC> == unchecked). Their behaviour
		3707	is undefined when the default loop has not been initialised by a previous
		3708	execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
		3709
		3710	It is often prudent to use C<EV_DEFAULT> when initialising the first
		3711	watcher in a function but use C<EV_DEFAULT_UC> afterwards.
		3712
		3713	=back
		3714
		3715	Example: Declare and initialise a check watcher, utilising the above
		3716	macros so it will work regardless of whether multiple loops are supported
		3717	or not.
		3718
		3719	static void
		3720	check_cb (EV_P_ ev_timer *w, int revents)
		3721	{
		3722	ev_check_stop (EV_A_ w);
		3723	}
		3724
		3725	ev_check check;
		3726	ev_check_init (&check, check_cb);
		3727	ev_check_start (EV_DEFAULT_ &check);
		3728	ev_run (EV_DEFAULT_ 0);
		3729
		3730	=head1 EMBEDDING
		3731
		3732	Libev can (and often is) directly embedded into host
		3733	applications. Examples of applications that embed it include the Deliantra
		3734	Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
		3735	and rxvt-unicode.
		3736
		3737	The goal is to enable you to just copy the necessary files into your
		3738	source directory without having to change even a single line in them, so
		3739	you can easily upgrade by simply copying (or having a checked-out copy of
		3740	libev somewhere in your source tree).
		3741
		3742	=head2 FILESETS
		3743
		3744	Depending on what features you need you need to include one or more sets of files
		3745	in your application.
		3746
		3747	=head3 CORE EVENT LOOP
		3748
		3749	To include only the libev core (all the C<ev_*> functions), with manual
		3750	configuration (no autoconf):
		3751
		3752	#define EV_STANDALONE 1
		3753	#include "ev.c"
		3754
		3755	This will automatically include F<ev.h>, too, and should be done in a
		3756	single C source file only to provide the function implementations. To use
		3757	it, do the same for F<ev.h> in all files wishing to use this API (best
		3758	done by writing a wrapper around F<ev.h> that you can include instead and
		3759	where you can put other configuration options):
		3760
		3761	#define EV_STANDALONE 1
		3762	#include "ev.h"
		3763
		3764	Both header files and implementation files can be compiled with a C++
		3765	compiler (at least, that's a stated goal, and breakage will be treated
		3766	as a bug).
		3767
		3768	You need the following files in your source tree, or in a directory
		3769	in your include path (e.g. in libev/ when using -Ilibev):
		3770
		3771	ev.h
		3772	ev.c
		3773	ev_vars.h
		3774	ev_wrap.h
		3775
		3776	ev_win32.c required on win32 platforms only
		3777
		3778	ev_select.c only when select backend is enabled (which is enabled by default)
		3779	ev_poll.c only when poll backend is enabled (disabled by default)
		3780	ev_epoll.c only when the epoll backend is enabled (disabled by default)
		3781	ev_kqueue.c only when the kqueue backend is enabled (disabled by default)
		3782	ev_port.c only when the solaris port backend is enabled (disabled by default)
		3783
		3784	F<ev.c> includes the backend files directly when enabled, so you only need
		3785	to compile this single file.
		3786
		3787	=head3 LIBEVENT COMPATIBILITY API
		3788
		3789	To include the libevent compatibility API, also include:
		3790
		3791	#include "event.c"
		3792
		3793	in the file including F<ev.c>, and:
		3794
		3795	#include "event.h"
		3796
		3797	in the files that want to use the libevent API. This also includes F<ev.h>.
		3798
		3799	You need the following additional files for this:
		3800
		3801	event.h
		3802	event.c
		3803
		3804	=head3 AUTOCONF SUPPORT
		3805
		3806	Instead of using C<EV_STANDALONE=1> and providing your configuration in
		3807	whatever way you want, you can also C<m4_include([libev.m4])> in your
		3808	F<configure.ac> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
		3809	include F<config.h> and configure itself accordingly.
		3810
		3811	For this of course you need the m4 file:
		3812
		3813	libev.m4
		3814
		3815	=head2 PREPROCESSOR SYMBOLS/MACROS
		3816
		3817	Libev can be configured via a variety of preprocessor symbols you have to
		3818	define before including (or compiling) any of its files. The default in
		3819	the absence of autoconf is documented for every option.
		3820
		3821	Symbols marked with "(h)" do not change the ABI, and can have different
		3822	values when compiling libev vs. including F<ev.h>, so it is permissible
		3823	to redefine them before including F<ev.h> without breaking compatibility
		3824	to a compiled library. All other symbols change the ABI, which means all
		3825	users of libev and the libev code itself must be compiled with compatible
		3826	settings.
		3827
		3828	=over 4
		3829
		3830	=item EV_COMPAT3 (h)
		3831
		3832	Backwards compatibility is a major concern for libev. This is why this
		3833	release of libev comes with wrappers for the functions and symbols that
		3834	have been renamed between libev version 3 and 4.
		3835
		3836	You can disable these wrappers (to test compatibility with future
		3837	versions) by defining C<EV_COMPAT3> to C<0> when compiling your
		3838	sources. This has the additional advantage that you can drop the C<struct>
		3839	from C<struct ev_loop> declarations, as libev will provide an C<ev_loop>
		3840	typedef in that case.
		3841
		3842	In some future version, the default for C<EV_COMPAT3> will become C<0>,
		3843	and in some even more future version the compatibility code will be
		3844	removed completely.
		3845
		3846	=item EV_STANDALONE (h)
		3847
		3848	Must always be C<1> if you do not use autoconf configuration, which
		3849	keeps libev from including F<config.h>, and it also defines dummy
		3850	implementations for some libevent functions (such as logging, which is not
		3851	supported). It will also not define any of the structs usually found in
		3852	F<event.h> that are not directly supported by the libev core alone.
		3853
		3854	In standalone mode, libev will still try to automatically deduce the
		3855	configuration, but has to be more conservative.
		3856
		3857	=item EV_USE_MONOTONIC
		3858
		3859	If defined to be C<1>, libev will try to detect the availability of the
		3860	monotonic clock option at both compile time and runtime. Otherwise no
		3861	use of the monotonic clock option will be attempted. If you enable this,
		3862	you usually have to link against librt or something similar. Enabling it
		3863	when the functionality isn't available is safe, though, although you have
		3864	to make sure you link against any libraries where the C<clock_gettime>
		3865	function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
		3866
		3867	=item EV_USE_REALTIME
		3868
		3869	If defined to be C<1>, libev will try to detect the availability of the
		3870	real-time clock option at compile time (and assume its availability
		3871	at runtime if successful). Otherwise no use of the real-time clock
		3872	option will be attempted. This effectively replaces C<gettimeofday>
		3873	by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
		3874	correctness. See the note about libraries in the description of
		3875	C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
		3876	C<EV_USE_CLOCK_SYSCALL>.
		3877
		3878	=item EV_USE_CLOCK_SYSCALL
		3879
		3880	If defined to be C<1>, libev will try to use a direct syscall instead
		3881	of calling the system-provided C<clock_gettime> function. This option
		3882	exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
		3883	unconditionally pulls in C<libpthread>, slowing down single-threaded
		3884	programs needlessly. Using a direct syscall is slightly slower (in
		3885	theory), because no optimised vdso implementation can be used, but avoids
		3886	the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
		3887	higher, as it simplifies linking (no need for C<-lrt>).
		3888
		3889	=item EV_USE_NANOSLEEP
		3890
		3891	If defined to be C<1>, libev will assume that C<nanosleep ()> is available
		3892	and will use it for delays. Otherwise it will use C<select ()>.
		3893
		3894	=item EV_USE_EVENTFD
		3895
		3896	If defined to be C<1>, then libev will assume that C<eventfd ()> is
		3897	available and will probe for kernel support at runtime. This will improve
		3898	C<ev_signal> and C<ev_async> performance and reduce resource consumption.
		3899	If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
		3900	2.7 or newer, otherwise disabled.
		3901
		3902	=item EV_USE_SELECT
		3903
		3904	If undefined or defined to be C<1>, libev will compile in support for the
		3905	C<select>(2) backend. No attempt at auto-detection will be done: if no
		3906	other method takes over, select will be it. Otherwise the select backend
		3907	will not be compiled in.
		3908
		3909	=item EV_SELECT_USE_FD_SET
		3910
		3911	If defined to C<1>, then the select backend will use the system C<fd_set>
		3912	structure. This is useful if libev doesn't compile due to a missing
		3913	C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
		3914	on exotic systems. This usually limits the range of file descriptors to
		3915	some low limit such as 1024 or might have other limitations (winsocket
		3916	only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
		3917	configures the maximum size of the C<fd_set>.
		3918
		3919	=item EV_SELECT_IS_WINSOCKET
		3920
		3921	When defined to C<1>, the select backend will assume that
		3922	select/socket/connect etc. don't understand file descriptors but
		3923	wants osf handles on win32 (this is the case when the select to
		3924	be used is the winsock select). This means that it will call
		3925	C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
		3926	it is assumed that all these functions actually work on fds, even
		3927	on win32. Should not be defined on non-win32 platforms.
		3928
		3929	=item EV_FD_TO_WIN32_HANDLE(fd)
		3930
		3931	If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
		3932	file descriptors to socket handles. When not defining this symbol (the
		3933	default), then libev will call C<_get_osfhandle>, which is usually
		3934	correct. In some cases, programs use their own file descriptor management,
		3935	in which case they can provide this function to map fds to socket handles.
		3936
		3937	=item EV_WIN32_HANDLE_TO_FD(handle)
		3938
		3939	If C<EV_SELECT_IS_WINSOCKET> then libev maps handles to file descriptors
		3940	using the standard C<_open_osfhandle> function. For programs implementing
		3941	their own fd to handle mapping, overwriting this function makes it easier
		3942	to do so. This can be done by defining this macro to an appropriate value.
		3943
		3944	=item EV_WIN32_CLOSE_FD(fd)
		3945
		3946	If programs implement their own fd to handle mapping on win32, then this
		3947	macro can be used to override the C<close> function, useful to unregister
		3948	file descriptors again. Note that the replacement function has to close
		3949	the underlying OS handle.
		3950
		3951	=item EV_USE_POLL
		3952
		3953	If defined to be C<1>, libev will compile in support for the C<poll>(2)
		3954	backend. Otherwise it will be enabled on non-win32 platforms. It
		3955	takes precedence over select.
		3956
		3957	=item EV_USE_EPOLL
		3958
		3959	If defined to be C<1>, libev will compile in support for the Linux
		3960	C<epoll>(7) backend. Its availability will be detected at runtime,
		3961	otherwise another method will be used as fallback. This is the preferred
		3962	backend for GNU/Linux systems. If undefined, it will be enabled if the
		3963	headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
		3964
		3965	=item EV_USE_KQUEUE
		3966
		3967	If defined to be C<1>, libev will compile in support for the BSD style
		3968	C<kqueue>(2) backend. Its actual availability will be detected at runtime,
		3969	otherwise another method will be used as fallback. This is the preferred
		3970	backend for BSD and BSD-like systems, although on most BSDs kqueue only
		3971	supports some types of fds correctly (the only platform we found that
		3972	supports ptys for example was NetBSD), so kqueue might be compiled in, but
		3973	not be used unless explicitly requested. The best way to use it is to find
		3974	out whether kqueue supports your type of fd properly and use an embedded
		3975	kqueue loop.
		3976
		3977	=item EV_USE_PORT
		3978
		3979	If defined to be C<1>, libev will compile in support for the Solaris
		3980	10 port style backend. Its availability will be detected at runtime,
		3981	otherwise another method will be used as fallback. This is the preferred
		3982	backend for Solaris 10 systems.
		3983
		3984	=item EV_USE_DEVPOLL
		3985
		3986	Reserved for future expansion, works like the USE symbols above.
		3987
		3988	=item EV_USE_INOTIFY
		3989
		3990	If defined to be C<1>, libev will compile in support for the Linux inotify
		3991	interface to speed up C<ev_stat> watchers. Its actual availability will
		3992	be detected at runtime. If undefined, it will be enabled if the headers
		3993	indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
		3994
		3995	=item EV_ATOMIC_T
		3996
		3997	Libev requires an integer type (suitable for storing C<0> or C<1>) whose
		3998	access is atomic with respect to other threads or signal contexts. No such
		3999	type is easily found in the C language, so you can provide your own type
		4000	that you know is safe for your purposes. It is used both for signal handler "locking"
		4001	as well as for signal and thread safety in C<ev_async> watchers.
		4002
		4003	In the absence of this define, libev will use C<sig_atomic_t volatile>
		4004	(from F<signal.h>), which is usually good enough on most platforms.
		4005
		4006	=item EV_H (h)
		4007
		4008	The name of the F<ev.h> header file used to include it. The default if
		4009	undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
		4010	used to virtually rename the F<ev.h> header file in case of conflicts.
		4011
		4012	=item EV_CONFIG_H (h)
		4013
		4014	If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
		4015	F<ev.c>'s idea of where to find the F<config.h> file, similarly to
		4016	C<EV_H>, above.
		4017
		4018	=item EV_EVENT_H (h)
		4019
		4020	Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
		4021	of how the F<event.h> header can be found, the default is C<"event.h">.
		4022
		4023	=item EV_PROTOTYPES (h)
		4024
		4025	If defined to be C<0>, then F<ev.h> will not define any function
		4026	prototypes, but still define all the structs and other symbols. This is
		4027	occasionally useful if you want to provide your own wrapper functions
		4028	around libev functions.
		4029
		4030	=item EV_MULTIPLICITY
		4031
		4032	If undefined or defined to C<1>, then all event-loop-specific functions
		4033	will have the C<struct ev_loop *> as first argument, and you can create
		4034	additional independent event loops. Otherwise there will be no support
		4035	for multiple event loops and there is no first event loop pointer
		4036	argument. Instead, all functions act on the single default loop.
		4037
		4038	=item EV_MINPRI
		4039
		4040	=item EV_MAXPRI
		4041
		4042	The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
		4043	C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
		4044	provide for more priorities by overriding those symbols (usually defined
		4045	to be C<-2> and C<2>, respectively).
		4046
		4047	When doing priority-based operations, libev usually has to linearly search
		4048	all the priorities, so having many of them (hundreds) uses a lot of space
		4049	and time, so using the defaults of five priorities (-2 .. +2) is usually
		4050	fine.
		4051
		4052	If your embedding application does not need any priorities, defining these
		4053	both to C<0> will save some memory and CPU.
		4054
		4055	=item EV_PERIODIC_ENABLE, EV_IDLE_ENABLE, EV_EMBED_ENABLE, EV_STAT_ENABLE,
		4056	EV_PREPARE_ENABLE, EV_CHECK_ENABLE, EV_FORK_ENABLE, EV_SIGNAL_ENABLE,
		4057	EV_ASYNC_ENABLE, EV_CHILD_ENABLE.
		4058
		4059	If undefined or defined to be C<1> (and the platform supports it), then
		4060	the respective watcher type is supported. If defined to be C<0>, then it
		4061	is not. Disabling watcher types mainly saves code size.
		4062
		4063	=item EV_FEATURES
		4064
		4065	If you need to shave off some kilobytes of code at the expense of some
		4066	speed (but with the full API), you can define this symbol to request
		4067	certain subsets of functionality. The default is to enable all features
		4068	that can be enabled on the platform.
		4069
		4070	A typical way to use this symbol is to define it to C<0> (or to a bitset
		4071	with some broad features you want) and then selectively re-enable
		4072	additional parts you want, for example if you want everything minimal,
		4073	but multiple event loop support, async and child watchers and the poll
		4074	backend, use this:
		4075
		4076	#define EV_FEATURES 0
		4077	#define EV_MULTIPLICITY 1
		4078	#define EV_USE_POLL 1
		4079	#define EV_CHILD_ENABLE 1
		4080	#define EV_ASYNC_ENABLE 1
		4081
		4082	The actual value is a bitset, it can be a combination of the following
		4083	values:
		4084
		4085	=over 4
		4086
		4087	=item C<1> - faster/larger code
		4088
		4089	Use larger code to speed up some operations.
		4090
		4091	Currently this is used to override some inlining decisions (enlarging the
		4092	code size by roughly 30% on amd64).
		4093
		4094	When optimising for size, use of compiler flags such as C<-Os> with
		4095	gcc is recommended, as well as C<-DNDEBUG>, as libev contains a number of
		4096	assertions.
		4097
		4098	=item C<2> - faster/larger data structures
		4099
		4100	Replaces the small 2-heap for timer management by a faster 4-heap, larger
		4101	hash table sizes and so on. This will usually further increase code size
		4102	and can additionally have an effect on the size of data structures at
		4103	runtime.
		4104
		4105	=item C<4> - full API configuration
		4106
		4107	This enables priorities (sets C<EV_MAXPRI>=2 and C<EV_MINPRI>=-2), and
		4108	enables multiplicity (C<EV_MULTIPLICITY>=1).
		4109
		4110	=item C<8> - full API
		4111
		4112	This enables a lot of the "lesser used" API functions. See C<ev.h> for
		4113	details on which parts of the API are still available without this
		4114	feature, and do not complain if this subset changes over time.
		4115
		4116	=item C<16> - enable all optional watcher types
		4117
		4118	Enables all optional watcher types. If you want to selectively enable
		4119	only some watcher types other than I/O and timers (e.g. prepare,
		4120	embed, async, child...) you can enable them manually by defining
		4121	C<EV_watchertype_ENABLE> to C<1> instead.
		4122
		4123	=item C<32> - enable all backends
		4124
		4125	This enables all backends - without this feature, you need to enable at
		4126	least one backend manually (C<EV_USE_SELECT> is a good choice).
		4127
		4128	=item C<64> - enable OS-specific "helper" APIs
		4129
		4130	Enable inotify, eventfd, signalfd and similar OS-specific helper APIs by
		4131	default.
		4132
		4133	=back
		4134
		4135	Compiling with C<gcc -Os -DEV_STANDALONE -DEV_USE_EPOLL=1 -DEV_FEATURES=0>
		4136	reduces the compiled size of libev from 24.7Kb code/2.8Kb data to 6.5Kb
		4137	code/0.3Kb data on my GNU/Linux amd64 system, while still giving you I/O
		4138	watchers, timers and monotonic clock support.
		4139
		4140	With an intelligent-enough linker (gcc+binutils are intelligent enough
		4141	when you use C<-Wl,--gc-sections -ffunction-sections>) functions unused by
		4142	your program might be left out as well - a binary starting a timer and an
		4143	I/O watcher then might come out at only 5Kb.
		4144
		4145	=item EV_AVOID_STDIO
		4146
		4147	If this is set to C<1> at compiletime, then libev will avoid using stdio
		4148	functions (printf, scanf, perror etc.). This will increase the code size
		4149	somewhat, but if your program doesn't otherwise depend on stdio and your
		4150	libc allows it, this avoids linking in the stdio library which is quite
		4151	big.
		4152
		4153	Note that error messages might become less precise when this option is
		4154	enabled.
		4155
		4156	=item EV_NSIG
		4157
		4158	The highest supported signal number, +1 (or, the number of
		4159	signals): Normally, libev tries to deduce the maximum number of signals
		4160	automatically, but sometimes this fails, in which case it can be
		4161	specified. Also, using a lower number than detected (C<32> should be
		4162	good for about any system in existence) can save some memory, as libev
		4163	statically allocates some 12-24 bytes per signal number.
		4164
		4165	=item EV_PID_HASHSIZE
		4166
		4167	C<ev_child> watchers use a small hash table to distribute workload by
		4168	pid. The default size is C<16> (or C<1> with C<EV_FEATURES> disabled),
		4169	usually more than enough. If you need to manage thousands of children you
		4170	might want to increase this value (I<must> be a power of two).
		4171
		4172	=item EV_INOTIFY_HASHSIZE
		4173
		4174	C<ev_stat> watchers use a small hash table to distribute workload by
		4175	inotify watch id. The default size is C<16> (or C<1> with C<EV_FEATURES>
		4176	disabled), usually more than enough. If you need to manage thousands of
		4177	C<ev_stat> watchers you might want to increase this value (I<must> be a
		4178	power of two).
		4179
		4180	=item EV_USE_4HEAP
		4181
		4182	Heaps are not very cache-efficient. To improve the cache-efficiency of the
		4183	timer and periodics heaps, libev uses a 4-heap when this symbol is defined
		4184	to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
		4185	faster performance with many (thousands) of watchers.
		4186
		4187	The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
		4188	will be C<0>.
		4189
		4190	=item EV_HEAP_CACHE_AT
		4191
		4192	Heaps are not very cache-efficient. To improve the cache-efficiency of the
		4193	timer and periodics heaps, libev can cache the timestamp (I<at>) within
		4194	the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
		4195	which uses 8-12 bytes more per watcher and a few hundred bytes more code,
		4196	but avoids random read accesses on heap changes. This improves performance
		4197	noticeably with many (hundreds) of watchers.
		4198
		4199	The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
		4200	will be C<0>.
		4201
		4202	=item EV_VERIFY
		4203
		4204	Controls how much internal verification (see C<ev_verify ()>) will
		4205	be done: If set to C<0>, no internal verification code will be compiled
		4206	in. If set to C<1>, then verification code will be compiled in, but not
		4207	called. If set to C<2>, then the internal verification code will be
		4208	called once per loop, which can slow down libev. If set to C<3>, then the
		4209	verification code will be called very frequently, which will slow down
		4210	libev considerably.
		4211
		4212	The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
		4213	will be C<0>.
		4214
		4215	=item EV_COMMON
		4216
		4217	By default, all watchers have a C<void *data> member. By redefining
		4218	this macro to something else you can include more and other types of
		4219	members. You have to define it each time you include one of the files,
		4220	though, and it must be identical each time.
		4221
		4222	For example, the perl EV module uses something like this:
		4223
		4224	#define EV_COMMON \
		4225	SV *self; /* contains this struct */ \
		4226	SV *cb_sv, *fh /* note no trailing ";" */
		4227
		4228	=item EV_CB_DECLARE (type)
		4229
		4230	=item EV_CB_INVOKE (watcher, revents)
		4231
		4232	=item ev_set_cb (ev, cb)
		4233
		4234	Can be used to change the callback member declaration in each watcher,
		4235	and the way callbacks are invoked and set. Must expand to a struct member
		4236	definition and a statement, respectively. See the F<ev.h> header file for
		4237	their default definitions. One possible use for overriding these is to
		4238	avoid the C<struct ev_loop *> as first argument in all cases, or to use
		4239	method calls instead of plain function calls in C++.
		4240
		4241	=back
		4242
		4243	=head2 EXPORTED API SYMBOLS
		4244
		4245	If you need to re-export the API (e.g. via a DLL) and you need a list of
		4246	exported symbols, you can use the provided F<Symbol.*> files which list
		4247	all public symbols, one per line:
		4248
		4249	Symbols.ev for libev proper
		4250	Symbols.event for the libevent emulation
		4251
		4252	This can also be used to rename all public symbols to avoid clashes with
		4253	multiple versions of libev linked together (which is obviously bad in
		4254	itself, but sometimes it is inconvenient to avoid this).
		4255
		4256	A sed command like this will create wrapper C<#define>'s that you need to
		4257	include before including F<ev.h>:
		4258
		4259	<Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
		4260
		4261	This would create a file F<wrap.h> which essentially looks like this:
		4262
		4263	#define ev_backend myprefix_ev_backend
		4264	#define ev_check_start myprefix_ev_check_start
		4265	#define ev_check_stop myprefix_ev_check_stop
		4266	...
		4267
		4268	=head2 EXAMPLES
		4269
		4270	For a real-world example of a program the includes libev
		4271	verbatim, you can have a look at the EV perl module
		4272	(L<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
		4273	the F<libev/> subdirectory and includes them in the F<EV/EVAPI.h> (public
		4274	interface) and F<EV.xs> (implementation) files. Only the F<EV.xs> file
		4275	will be compiled. It is pretty complex because it provides its own header
		4276	file.
		4277
		4278	The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
		4279	that everybody includes and which overrides some configure choices:
		4280
		4281	#define EV_FEATURES 8
		4282	#define EV_USE_SELECT 1
		4283	#define EV_PREPARE_ENABLE 1
		4284	#define EV_IDLE_ENABLE 1
		4285	#define EV_SIGNAL_ENABLE 1
		4286	#define EV_CHILD_ENABLE 1
		4287	#define EV_USE_STDEXCEPT 0
		4288	#define EV_CONFIG_H <config.h>
		4289
		4290	#include "ev++.h"
		4291
		4292	And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
		4293
		4294	#include "ev_cpp.h"
		4295	#include "ev.c"
		4296
		4297	=head1 INTERACTION WITH OTHER PROGRAMS OR LIBRARIES
		4298
		4299	=head2 THREADS AND COROUTINES
		4300
		4301	=head3 THREADS
		4302
		4303	All libev functions are reentrant and thread-safe unless explicitly
		4304	documented otherwise, but libev implements no locking itself. This means
		4305	that you can use as many loops as you want in parallel, as long as there
		4306	are no concurrent calls into any libev function with the same loop
		4307	parameter (C<ev_default_*> calls have an implicit default loop parameter,
		4308	of course): libev guarantees that different event loops share no data
		4309	structures that need any locking.
		4310
		4311	Or to put it differently: calls with different loop parameters can be done
		4312	concurrently from multiple threads, calls with the same loop parameter
		4313	must be done serially (but can be done from different threads, as long as
		4314	only one thread ever is inside a call at any point in time, e.g. by using
		4315	a mutex per loop).
		4316
		4317	Specifically to support threads (and signal handlers), libev implements
		4318	so-called C<ev_async> watchers, which allow some limited form of
		4319	concurrency on the same event loop, namely waking it up "from the
		4320	outside".
		4321
		4322	If you want to know which design (one loop, locking, or multiple loops
		4323	without or something else still) is best for your problem, then I cannot
		4324	help you, but here is some generic advice:
		4325
		4326	=over 4
		4327
		4328	=item * most applications have a main thread: use the default libev loop
		4329	in that thread, or create a separate thread running only the default loop.
		4330
		4331	This helps integrating other libraries or software modules that use libev
		4332	themselves and don't care/know about threading.
		4333
		4334	=item * one loop per thread is usually a good model.
		4335
		4336	Doing this is almost never wrong, sometimes a better-performance model
		4337	exists, but it is always a good start.
		4338
		4339	=item * other models exist, such as the leader/follower pattern, where one
		4340	loop is handed through multiple threads in a kind of round-robin fashion.
		4341
		4342	Choosing a model is hard - look around, learn, know that usually you can do
		4343	better than you currently do :-)
		4344
		4345	=item * often you need to talk to some other thread which blocks in the
		4346	event loop.
		4347
		4348	C<ev_async> watchers can be used to wake them up from other threads safely
		4349	(or from signal contexts...).
		4350
		4351	An example use would be to communicate signals or other events that only
		4352	work in the default loop by registering the signal watcher with the
		4353	default loop and triggering an C<ev_async> watcher from the default loop
		4354	watcher callback into the event loop interested in the signal.
		4355
		4356	=back
		4357
		4358	=head4 THREAD LOCKING EXAMPLE
		4359
		4360	Here is a fictitious example of how to run an event loop in a different
		4361	thread than where callbacks are being invoked and watchers are
		4362	created/added/removed.
		4363
		4364	For a real-world example, see the C<EV::Loop::Async> perl module,
		4365	which uses exactly this technique (which is suited for many high-level
		4366	languages).
		4367
		4368	The example uses a pthread mutex to protect the loop data, a condition
		4369	variable to wait for callback invocations, an async watcher to notify the
		4370	event loop thread and an unspecified mechanism to wake up the main thread.
		4371
		4372	First, you need to associate some data with the event loop:
		4373
		4374	typedef struct {
		4375	mutex_t lock; /* global loop lock */
		4376	ev_async async_w;
		4377	thread_t tid;
		4378	cond_t invoke_cv;
		4379	} userdata;
		4380
		4381	void prepare_loop (EV_P)
		4382	{
		4383	// for simplicity, we use a static userdata struct.
		4384	static userdata u;
		4385
		4386	ev_async_init (&u->async_w, async_cb);
		4387	ev_async_start (EV_A_ &u->async_w);
		4388
		4389	pthread_mutex_init (&u->lock, 0);
		4390	pthread_cond_init (&u->invoke_cv, 0);
		4391
		4392	// now associate this with the loop
		4393	ev_set_userdata (EV_A_ u);
		4394	ev_set_invoke_pending_cb (EV_A_ l_invoke);
		4395	ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
		4396
		4397	// then create the thread running ev_loop
		4398	pthread_create (&u->tid, 0, l_run, EV_A);
		4399	}
		4400
		4401	The callback for the C<ev_async> watcher does nothing: the watcher is used
		4402	solely to wake up the event loop so it takes notice of any new watchers
		4403	that might have been added:
		4404
		4405	static void
		4406	async_cb (EV_P_ ev_async *w, int revents)
		4407	{
		4408	// just used for the side effects
		4409	}
		4410
		4411	The C<l_release> and C<l_acquire> callbacks simply unlock/lock the mutex
		4412	protecting the loop data, respectively.
		4413
		4414	static void
		4415	l_release (EV_P)
		4416	{
		4417	userdata *u = ev_userdata (EV_A);
		4418	pthread_mutex_unlock (&u->lock);
		4419	}
		4420
		4421	static void
		4422	l_acquire (EV_P)
		4423	{
		4424	userdata *u = ev_userdata (EV_A);
		4425	pthread_mutex_lock (&u->lock);
		4426	}
		4427
		4428	The event loop thread first acquires the mutex, and then jumps straight
		4429	into C<ev_run>:
		4430
		4431	void *
		4432	l_run (void *thr_arg)
		4433	{
		4434	struct ev_loop *loop = (struct ev_loop *)thr_arg;
		4435
		4436	l_acquire (EV_A);
		4437	pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
		4438	ev_run (EV_A_ 0);
		4439	l_release (EV_A);
		4440
		4441	return 0;
		4442	}
		4443
		4444	Instead of invoking all pending watchers, the C<l_invoke> callback will
		4445	signal the main thread via some unspecified mechanism (signals? pipe
		4446	writes? C<Async::Interrupt>?) and then waits until all pending watchers
		4447	have been called (in a while loop because a) spurious wakeups are possible
		4448	and b) skipping inter-thread-communication when there are no pending
		4449	watchers is very beneficial):
		4450
		4451	static void
		4452	l_invoke (EV_P)
		4453	{
		4454	userdata *u = ev_userdata (EV_A);
		4455
		4456	while (ev_pending_count (EV_A))
		4457	{
		4458	wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
		4459	pthread_cond_wait (&u->invoke_cv, &u->lock);
		4460	}
		4461	}
		4462
		4463	Now, whenever the main thread gets told to invoke pending watchers, it
		4464	will grab the lock, call C<ev_invoke_pending> and then signal the loop
		4465	thread to continue:
		4466
		4467	static void
		4468	real_invoke_pending (EV_P)
		4469	{
		4470	userdata *u = ev_userdata (EV_A);
		4471
		4472	pthread_mutex_lock (&u->lock);
		4473	ev_invoke_pending (EV_A);
		4474	pthread_cond_signal (&u->invoke_cv);
		4475	pthread_mutex_unlock (&u->lock);
		4476	}
		4477
		4478	Whenever you want to start/stop a watcher or do other modifications to an
		4479	event loop, you will now have to lock:
		4480
		4481	ev_timer timeout_watcher;
		4482	userdata *u = ev_userdata (EV_A);
		4483
		4484	ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
		4485
		4486	pthread_mutex_lock (&u->lock);
		4487	ev_timer_start (EV_A_ &timeout_watcher);
		4488	ev_async_send (EV_A_ &u->async_w);
		4489	pthread_mutex_unlock (&u->lock);
		4490
		4491	Note that sending the C<ev_async> watcher is required because otherwise
		4492	an event loop currently blocking in the kernel will have no knowledge
		4493	about the newly added timer. By waking up the loop it will pick up any new
		4494	watchers in the next event loop iteration.
		4495
		4496	=head3 COROUTINES
		4497
		4498	Libev is very accommodating to coroutines ("cooperative threads"):
		4499	libev fully supports nesting calls to its functions from different
		4500	coroutines (e.g. you can call C<ev_run> on the same loop from two
		4501	different coroutines, and switch freely between both coroutines running
		4502	the loop, as long as you don't confuse yourself). The only exception is
		4503	that you must not do this from C<ev_periodic> reschedule callbacks.
		4504
		4505	Care has been taken to ensure that libev does not keep local state inside
		4506	C<ev_run>, and other calls do not usually allow for coroutine switches as
		4507	they do not call any callbacks.
		4508
		4509	=head2 COMPILER WARNINGS
		4510
		4511	Depending on your compiler and compiler settings, you might get no or a
		4512	lot of warnings when compiling libev code. Some people are apparently
		4513	scared by this.
		4514
		4515	However, these are unavoidable for many reasons. For one, each compiler
		4516	has different warnings, and each user has different tastes regarding
		4517	warning options. "Warn-free" code therefore cannot be a goal except when
		4518	targeting a specific compiler and compiler-version.
		4519
		4520	Another reason is that some compiler warnings require elaborate
		4521	workarounds, or other changes to the code that make it less clear and less
		4522	maintainable.
		4523
		4524	And of course, some compiler warnings are just plain stupid, or simply
		4525	wrong (because they don't actually warn about the condition their message
		4526	seems to warn about). For example, certain older gcc versions had some
		4527	warnings that resulted in an extreme number of false positives. These have
		4528	been fixed, but some people still insist on making code warn-free with
		4529	such buggy versions.
		4530
		4531	While libev is written to generate as few warnings as possible,
		4532	"warn-free" code is not a goal, and it is recommended not to build libev
		4533	with any compiler warnings enabled unless you are prepared to cope with
		4534	them (e.g. by ignoring them). Remember that warnings are just that:
		4535	warnings, not errors, or proof of bugs.
		4536
		4537
		4538	=head2 VALGRIND
		4539
		4540	Valgrind has a special section here because it is a popular tool that is
		4541	highly useful. Unfortunately, valgrind reports are very hard to interpret.
		4542
		4543	If you think you found a bug (memory leak, uninitialised data access etc.)
		4544	in libev, then check twice: If valgrind reports something like:
		4545
		4546	==2274== definitely lost: 0 bytes in 0 blocks.
		4547	==2274== possibly lost: 0 bytes in 0 blocks.
		4548	==2274== still reachable: 256 bytes in 1 blocks.
		4549
		4550	Then there is no memory leak, just as memory accounted to global variables
		4551	is not a memleak - the memory is still being referenced, and didn't leak.
		4552
		4553	Similarly, under some circumstances, valgrind might report kernel bugs
		4554	as if it were a bug in libev (e.g. in realloc or in the poll backend,
		4555	although an acceptable workaround has been found here), or it might be
		4556	confused.
		4557
		4558	Keep in mind that valgrind is a very good tool, but only a tool. Don't
		4559	make it into some kind of religion.
		4560
		4561	If you are unsure about something, feel free to contact the mailing list
		4562	with the full valgrind report and an explanation on why you think this
		4563	is a bug in libev (best check the archives, too :). However, don't be
		4564	annoyed when you get a brisk "this is no bug" answer and take the chance
		4565	of learning how to interpret valgrind properly.
		4566
		4567	If you need, for some reason, empty reports from valgrind for your project
		4568	I suggest using suppression lists.
		4569
		4570
		4571	=head1 PORTABILITY NOTES
		4572
		4573	=head2 GNU/LINUX 32 BIT LIMITATIONS
		4574
		4575	GNU/Linux is the only common platform that supports 64 bit file/large file
		4576	interfaces but I<disables> them by default.
		4577
		4578	That means that libev compiled in the default environment doesn't support
		4579	files larger than 2GiB or so, which mainly affects C<ev_stat> watchers.
		4580
		4581	Unfortunately, many programs try to work around this GNU/Linux issue
		4582	by enabling the large file API, which makes them incompatible with the
		4583	standard libev compiled for their system.
		4584
		4585	Likewise, libev cannot enable the large file API itself as this would
		4586	suddenly make it incompatible to the default compile time environment,
		4587	i.e. all programs not using special compile switches.
		4588
		4589	=head2 OS/X AND DARWIN BUGS
		4590
		4591	The whole thing is a bug if you ask me - basically any system interface
		4592	you touch is broken, whether it is locales, poll, kqueue or even the
		4593	OpenGL drivers.
		4594
		4595	=head3 C<kqueue> is buggy
		4596
		4597	The kqueue syscall is broken in all known versions - most versions support
		4598	only sockets, many support pipes.
		4599
		4600	Libev tries to work around this by not using C<kqueue> by default on this
		4601	rotten platform, but of course you can still ask for it when creating a
		4602	loop - embedding a socket-only kqueue loop into a select-based one is
		4603	probably going to work well.
		4604
		4605	=head3 C<poll> is buggy
		4606
		4607	Instead of fixing C<kqueue>, Apple replaced their (working) C<poll>
		4608	implementation by something calling C<kqueue> internally around the 10.5.6
		4609	release, so now C<kqueue> I<and> C<poll> are broken.
		4610
		4611	Libev tries to work around this by not using C<poll> by default on
		4612	this rotten platform, but of course you can still ask for it when creating
		4613	a loop.
		4614
		4615	=head3 C<select> is buggy
		4616
		4617	All that's left is C<select>, and of course Apple found a way to fuck this
		4618	one up as well: On OS/X, C<select> actively limits the number of file
		4619	descriptors you can pass in to 1024 - your program suddenly crashes when
		4620	you use more.
		4621
		4622	There is an undocumented "workaround" for this - defining
		4623	C<_DARWIN_UNLIMITED_SELECT>, which libev tries to use, so select I<should>
		4624	work on OS/X.
		4625
		4626	=head2 SOLARIS PROBLEMS AND WORKAROUNDS
		4627
		4628	=head3 C<errno> reentrancy
		4629
		4630	The default compile environment on Solaris is unfortunately so
		4631	thread-unsafe that you can't even use components/libraries compiled
		4632	without C<-D_REENTRANT> in a threaded program, which, of course, isn't
		4633	defined by default. A valid, if stupid, implementation choice.
		4634
		4635	If you want to use libev in threaded environments you have to make sure
		4636	it's compiled with C<_REENTRANT> defined.
		4637
		4638	=head3 Event port backend
		4639
		4640	The scalable event interface for Solaris is called "event
		4641	ports". Unfortunately, this mechanism is very buggy in all major
		4642	releases. If you run into high CPU usage, your program freezes or you get
		4643	a large number of spurious wakeups, make sure you have all the relevant
		4644	and latest kernel patches applied. No, I don't know which ones, but there
		4645	are multiple ones to apply, and afterwards, event ports actually work
		4646	great.
		4647
		4648	If you can't get it to work, you can try running the program by setting
		4649	the environment variable C<LIBEV_FLAGS=3> to only allow C<poll> and
		4650	C<select> backends.
		4651
		4652	=head2 AIX POLL BUG
		4653
		4654	AIX unfortunately has a broken C<poll.h> header. Libev works around
		4655	this by trying to avoid the poll backend altogether (i.e. it's not even
		4656	compiled in), which normally isn't a big problem as C<select> works fine
		4657	with large bitsets on AIX, and AIX is dead anyway.
		4658
		4659	=head2 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
		4660
		4661	=head3 General issues
		4662
		4663	Win32 doesn't support any of the standards (e.g. POSIX) that libev
		4664	requires, and its I/O model is fundamentally incompatible with the POSIX
		4665	model. Libev still offers limited functionality on this platform in
		4666	the form of the C<EVBACKEND_SELECT> backend, and only supports socket
		4667	descriptors. This only applies when using Win32 natively, not when using
		4668	e.g. cygwin. Actually, it only applies to the microsofts own compilers,
		4669	as every compielr comes with a slightly differently broken/incompatible
		4670	environment.
		4671
		4672	Lifting these limitations would basically require the full
		4673	re-implementation of the I/O system. If you are into this kind of thing,
		4674	then note that glib does exactly that for you in a very portable way (note
		4675	also that glib is the slowest event library known to man).
		4676
		4677	There is no supported compilation method available on windows except
		4678	embedding it into other applications.
		4679
		4680	Sensible signal handling is officially unsupported by Microsoft - libev
		4681	tries its best, but under most conditions, signals will simply not work.
		4682
		4683	Not a libev limitation but worth mentioning: windows apparently doesn't
		4684	accept large writes: instead of resulting in a partial write, windows will
		4685	either accept everything or return C<ENOBUFS> if the buffer is too large,
		4686	so make sure you only write small amounts into your sockets (less than a
		4687	megabyte seems safe, but this apparently depends on the amount of memory
		4688	available).
		4689
		4690	Due to the many, low, and arbitrary limits on the win32 platform and
		4691	the abysmal performance of winsockets, using a large number of sockets
		4692	is not recommended (and not reasonable). If your program needs to use
		4693	more than a hundred or so sockets, then likely it needs to use a totally
		4694	different implementation for windows, as libev offers the POSIX readiness
		4695	notification model, which cannot be implemented efficiently on windows
		4696	(due to Microsoft monopoly games).
		4697
		4698	A typical way to use libev under windows is to embed it (see the embedding
		4699	section for details) and use the following F<evwrap.h> header file instead
		4700	of F<ev.h>:
		4701
		4702	#define EV_STANDALONE /* keeps ev from requiring config.h */
		4703	#define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
		4704
		4705	#include "ev.h"
		4706
		4707	And compile the following F<evwrap.c> file into your project (make sure
		4708	you do I<not> compile the F<ev.c> or any other embedded source files!):
		4709
		4710	#include "evwrap.h"
		4711	#include "ev.c"
		4712
		4713	=head3 The winsocket C<select> function
		4714
		4715	The winsocket C<select> function doesn't follow POSIX in that it
		4716	requires socket I<handles> and not socket I<file descriptors> (it is
		4717	also extremely buggy). This makes select very inefficient, and also
		4718	requires a mapping from file descriptors to socket handles (the Microsoft
		4719	C runtime provides the function C<_open_osfhandle> for this). See the
		4720	discussion of the C<EV_SELECT_USE_FD_SET>, C<EV_SELECT_IS_WINSOCKET> and
		4721	C<EV_FD_TO_WIN32_HANDLE> preprocessor symbols for more info.
		4722
		4723	The configuration for a "naked" win32 using the Microsoft runtime
		4724	libraries and raw winsocket select is:
		4725
		4726	#define EV_USE_SELECT 1
		4727	#define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
		4728
		4729	Note that winsockets handling of fd sets is O(n), so you can easily get a
		4730	complexity in the O(n²) range when using win32.
		4731
		4732	=head3 Limited number of file descriptors
		4733
		4734	Windows has numerous arbitrary (and low) limits on things.
		4735
		4736	Early versions of winsocket's select only supported waiting for a maximum
		4737	of C<64> handles (probably owning to the fact that all windows kernels
		4738	can only wait for C<64> things at the same time internally; Microsoft
		4739	recommends spawning a chain of threads and wait for 63 handles and the
		4740	previous thread in each. Sounds great!).
		4741
		4742	Newer versions support more handles, but you need to define C<FD_SETSIZE>
		4743	to some high number (e.g. C<2048>) before compiling the winsocket select
		4744	call (which might be in libev or elsewhere, for example, perl and many
		4745	other interpreters do their own select emulation on windows).
		4746
		4747	Another limit is the number of file descriptors in the Microsoft runtime
		4748	libraries, which by default is C<64> (there must be a hidden I<64>
		4749	fetish or something like this inside Microsoft). You can increase this
		4750	by calling C<_setmaxstdio>, which can increase this limit to C<2048>
		4751	(another arbitrary limit), but is broken in many versions of the Microsoft
		4752	runtime libraries. This might get you to about C<512> or C<2048> sockets
		4753	(depending on windows version and/or the phase of the moon). To get more,
		4754	you need to wrap all I/O functions and provide your own fd management, but
		4755	the cost of calling select (O(n²)) will likely make this unworkable.
		4756
		4757	=head2 PORTABILITY REQUIREMENTS
		4758
		4759	In addition to a working ISO-C implementation and of course the
		4760	backend-specific APIs, libev relies on a few additional extensions:
		4761
		4762	=over 4
		4763
		4764	=item C<void (*)(ev_watcher_type *, int revents)> must have compatible
		4765	calling conventions regardless of C<ev_watcher_type *>.
		4766
		4767	Libev assumes not only that all watcher pointers have the same internal
		4768	structure (guaranteed by POSIX but not by ISO C for example), but it also
		4769	assumes that the same (machine) code can be used to call any watcher
		4770	callback: The watcher callbacks have different type signatures, but libev
		4771	calls them using an C<ev_watcher *> internally.
		4772
		4773	=item pointer accesses must be thread-atomic
		4774
		4775	Accessing a pointer value must be atomic, it must both be readable and
		4776	writable in one piece - this is the case on all current architectures.
		4777
		4778	=item C<sig_atomic_t volatile> must be thread-atomic as well
		4779
		4780	The type C<sig_atomic_t volatile> (or whatever is defined as
		4781	C<EV_ATOMIC_T>) must be atomic with respect to accesses from different
		4782	threads. This is not part of the specification for C<sig_atomic_t>, but is
		4783	believed to be sufficiently portable.
		4784
		4785	=item C<sigprocmask> must work in a threaded environment
		4786
		4787	Libev uses C<sigprocmask> to temporarily block signals. This is not
		4788	allowed in a threaded program (C<pthread_sigmask> has to be used). Typical
		4789	pthread implementations will either allow C<sigprocmask> in the "main
		4790	thread" or will block signals process-wide, both behaviours would
		4791	be compatible with libev. Interaction between C<sigprocmask> and
		4792	C<pthread_sigmask> could complicate things, however.
		4793
		4794	The most portable way to handle signals is to block signals in all threads
		4795	except the initial one, and run the default loop in the initial thread as
		4796	well.
		4797
		4798	=item C<long> must be large enough for common memory allocation sizes
		4799
		4800	To improve portability and simplify its API, libev uses C<long> internally
		4801	instead of C<size_t> when allocating its data structures. On non-POSIX
		4802	systems (Microsoft...) this might be unexpectedly low, but is still at
		4803	least 31 bits everywhere, which is enough for hundreds of millions of
		4804	watchers.
		4805
		4806	=item C<double> must hold a time value in seconds with enough accuracy
		4807
		4808	The type C<double> is used to represent timestamps. It is required to
		4809	have at least 51 bits of mantissa (and 9 bits of exponent), which is
		4810	good enough for at least into the year 4000 with millisecond accuracy
		4811	(the design goal for libev). This requirement is overfulfilled by
		4812	implementations using IEEE 754, which is basically all existing ones. With
		4813	IEEE 754 doubles, you get microsecond accuracy until at least 2200.
		4814
		4815	=back
		4816
		4817	If you know of other additional requirements drop me a note.
		4818
		4819
		4820	=head1 ALGORITHMIC COMPLEXITIES
		4821
		4822	In this section the complexities of (many of) the algorithms used inside
		4823	libev will be documented. For complexity discussions about backends see
		4824	the documentation for C<ev_default_init>.
		4825
		4826	All of the following are about amortised time: If an array needs to be
		4827	extended, libev needs to realloc and move the whole array, but this
		4828	happens asymptotically rarer with higher number of elements, so O(1) might
		4829	mean that libev does a lengthy realloc operation in rare cases, but on
		4830	average it is much faster and asymptotically approaches constant time.
		4831
		4832	=over 4
		4833
		4834	=item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
		4835
		4836	This means that, when you have a watcher that triggers in one hour and
		4837	there are 100 watchers that would trigger before that, then inserting will
		4838	have to skip roughly seven (C<ld 100>) of these watchers.
		4839
		4840	=item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
		4841
		4842	That means that changing a timer costs less than removing/adding them,
		4843	as only the relative motion in the event queue has to be paid for.
		4844
		4845	=item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
		4846
		4847	These just add the watcher into an array or at the head of a list.
		4848
		4849	=item Stopping check/prepare/idle/fork/async watchers: O(1)
		4850
		4851	=item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
		4852
		4853	These watchers are stored in lists, so they need to be walked to find the
		4854	correct watcher to remove. The lists are usually short (you don't usually
		4855	have many watchers waiting for the same fd or signal: one is typical, two
		4856	is rare).
		4857
		4858	=item Finding the next timer in each loop iteration: O(1)
		4859
		4860	By virtue of using a binary or 4-heap, the next timer is always found at a
		4861	fixed position in the storage array.
		4862
		4863	=item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
		4864
		4865	A change means an I/O watcher gets started or stopped, which requires
		4866	libev to recalculate its status (and possibly tell the kernel, depending
		4867	on backend and whether C<ev_io_set> was used).
		4868
		4869	=item Activating one watcher (putting it into the pending state): O(1)
		4870
		4871	=item Priority handling: O(number_of_priorities)
		4872
		4873	Priorities are implemented by allocating some space for each
		4874	priority. When doing priority-based operations, libev usually has to
		4875	linearly search all the priorities, but starting/stopping and activating
		4876	watchers becomes O(1) with respect to priority handling.
		4877
		4878	=item Sending an ev_async: O(1)
		4879
		4880	=item Processing ev_async_send: O(number_of_async_watchers)
		4881
		4882	=item Processing signals: O(max_signal_number)
		4883
		4884	Sending involves a system call I<iff> there were no other C<ev_async_send>
		4885	calls in the current loop iteration. Checking for async and signal events
		4886	involves iterating over all running async watchers or all signal numbers.
		4887
		4888	=back
		4889
		4890
		4891	=head1 PORTING FROM LIBEV 3.X TO 4.X
		4892
		4893	The major version 4 introduced some incompatible changes to the API.
		4894
		4895	At the moment, the C<ev.h> header file provides compatibility definitions
		4896	for all changes, so most programs should still compile. The compatibility
		4897	layer might be removed in later versions of libev, so better update to the
		4898	new API early than late.
		4899
		4900	=over 4
		4901
		4902	=item C<EV_COMPAT3> backwards compatibility mechanism
		4903
		4904	The backward compatibility mechanism can be controlled by
		4905	C<EV_COMPAT3>. See L<PREPROCESSOR SYMBOLS/MACROS> in the L<EMBEDDING>
		4906	section.
		4907
		4908	=item C<ev_default_destroy> and C<ev_default_fork> have been removed
		4909
		4910	These calls can be replaced easily by their C<ev_loop_xxx> counterparts:
		4911
		4912	ev_loop_destroy (EV_DEFAULT_UC);
		4913	ev_loop_fork (EV_DEFAULT);
		4914
		4915	=item function/symbol renames
		4916
		4917	A number of functions and symbols have been renamed:
		4918
		4919	ev_loop => ev_run
		4920	EVLOOP_NONBLOCK => EVRUN_NOWAIT
		4921	EVLOOP_ONESHOT => EVRUN_ONCE
		4922
		4923	ev_unloop => ev_break
		4924	EVUNLOOP_CANCEL => EVBREAK_CANCEL
		4925	EVUNLOOP_ONE => EVBREAK_ONE
		4926	EVUNLOOP_ALL => EVBREAK_ALL
		4927
		4928	EV_TIMEOUT => EV_TIMER
		4929
		4930	ev_loop_count => ev_iteration
		4931	ev_loop_depth => ev_depth
		4932	ev_loop_verify => ev_verify
		4933
		4934	Most functions working on C<struct ev_loop> objects don't have an
		4935	C<ev_loop_> prefix, so it was removed; C<ev_loop>, C<ev_unloop> and
		4936	associated constants have been renamed to not collide with the C<struct
		4937	ev_loop> anymore and C<EV_TIMER> now follows the same naming scheme
		4938	as all other watcher types. Note that C<ev_loop_fork> is still called
		4939	C<ev_loop_fork> because it would otherwise clash with the C<ev_fork>
		4940	typedef.
		4941
		4942	=item C<EV_MINIMAL> mechanism replaced by C<EV_FEATURES>
		4943
		4944	The preprocessor symbol C<EV_MINIMAL> has been replaced by a different
		4945	mechanism, C<EV_FEATURES>. Programs using C<EV_MINIMAL> usually compile
		4946	and work, but the library code will of course be larger.
		4947
		4948	=back
		4949
		4950
		4951	=head1 GLOSSARY
		4952
		4953	=over 4
		4954
		4955	=item active
		4956
		4957	A watcher is active as long as it has been started and not yet stopped.
		4958	See L<WATCHER STATES> for details.
		4959
		4960	=item application
		4961
		4962	In this document, an application is whatever is using libev.
		4963
		4964	=item backend
		4965
		4966	The part of the code dealing with the operating system interfaces.
		4967
		4968	=item callback
		4969
		4970	The address of a function that is called when some event has been
		4971	detected. Callbacks are being passed the event loop, the watcher that
		4972	received the event, and the actual event bitset.
		4973
		4974	=item callback/watcher invocation
		4975
		4976	The act of calling the callback associated with a watcher.
		4977
		4978	=item event
		4979
		4980	A change of state of some external event, such as data now being available
		4981	for reading on a file descriptor, time having passed or simply not having
		4982	any other events happening anymore.
		4983
		4984	In libev, events are represented as single bits (such as C<EV_READ> or
		4985	C<EV_TIMER>).
		4986
		4987	=item event library
		4988
		4989	A software package implementing an event model and loop.
		4990
		4991	=item event loop
		4992
		4993	An entity that handles and processes external events and converts them
		4994	into callback invocations.
		4995
		4996	=item event model
		4997
		4998	The model used to describe how an event loop handles and processes
		4999	watchers and events.
		5000
		5001	=item pending
		5002
		5003	A watcher is pending as soon as the corresponding event has been
		5004	detected. See L<WATCHER STATES> for details.
		5005
		5006	=item real time
		5007
		5008	The physical time that is observed. It is apparently strictly monotonic :)
		5009
		5010	=item wall-clock time
		5011
		5012	The time and date as shown on clocks. Unlike real time, it can actually
		5013	be wrong and jump forwards and backwards, e.g. when the you adjust your
		5014	clock.
		5015
		5016	=item watcher
		5017
		5018	A data structure that describes interest in certain events. Watchers need
		5019	to be started (attached to an event loop) before they can receive events.
		5020
		5021	=back
		5022
767	=head1 AUTHOR	5023	=head1 AUTHOR
768		5024
769	Marc Lehmann <libev@schmorp.de>.	5025	Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael
		5026	Magnusson and Emanuele Giaquinta.
770		5027

Diff Legend

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