ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libev/ev.pod

Comparing libev/ev.pod (file contents):
Revision 1.8 by root, Mon Nov 12 08:20:02 2007 UTC vs.
Revision 1.272 by root, Tue Nov 24 06:39:28 2009 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_loop's to stop iterating
		32	ev_unloop (EV_A_ EVUNLOOP_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_loop to stop iterating
		41	ev_unloop (EV_A_ EVUNLOOP_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_loop (0);
		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_loop (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	Familarity with event based programming techniques in general is assumed
		81	throughout this document.
		82
		83	=head1 ABOUT LIBEV
10		84
11	Libev is an event loop: you register interest in certain events (such as a	85	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	86	file descriptor being readable or a timeout occurring), and it will manage
13	these event sources and provide your program with events.	87	these event sources and provide your program with events.
14		88
15	To do this, it must take more or less complete control over your process	89	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	90	(or thread) by executing the I<event loop> handler, and will then
17	communicate events via a callback mechanism.	91	communicate events via a callback mechanism.
…		…
19	You register interest in certain events by registering so-called I<event	93	You register interest in certain events by registering so-called I<event
20	watchers>, which are relatively small C structures you initialise with the	94	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	95	details of the event, and then hand it over to libev by I<starting> the
22	watcher.	96	watcher.
23		97
24	=head1 FEATURES	98	=head2 FEATURES
25		99
26	Libev supports select, poll, the linux-specific epoll and the bsd-specific	100	Libev supports C<select>, C<poll>, the Linux-specific C<epoll>, the
27	kqueue mechanisms for file descriptor events, relative timers, absolute	101	BSD-specific C<kqueue> and the Solaris-specific event port mechanisms
28	timers with customised rescheduling, signal events, process status change	102	for file descriptor events (C<ev_io>), the Linux C<inotify> interface
		103	(for C<ev_stat>), Linux eventfd/signalfd (for faster and cleaner
		104	inter-thread wakeup (C<ev_async>)/signal handling (C<ev_signal>)) relative
		105	timers (C<ev_timer>), absolute timers with customised rescheduling
		106	(C<ev_periodic>), synchronous signals (C<ev_signal>), process status
29	events (related to SIGCHLD), and event watchers dealing with the event	107	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	108	loop mechanism itself (C<ev_idle>, C<ev_embed>, C<ev_prepare> and
		109	C<ev_check> watchers) as well as file watchers (C<ev_stat>) and even
		110	limited support for fork events (C<ev_fork>).
		111
		112	It also is quite fast (see this
31	fast (see this L<benchmark|http://libev.schmorp.de/bench.html> comparing	113	L<benchmark|http://libev.schmorp.de/bench.html> comparing it to libevent
32	it to libevent for example).	114	for example).
33		115
34	=head1 CONVENTIONS	116	=head2 CONVENTIONS
35		117
36	Libev is very configurable. In this manual the default configuration	118	Libev is very configurable. In this manual the default (and most common)
37	will be described, which supports multiple event loops. For more info	119	configuration will be described, which supports multiple event loops. For
38	about various configuration options please have a look at the file	120	more info about various configuration options please have a look at
39	F<README.embed> in the libev distribution. If libev was configured without	121	B<EMBED> section in this manual. If libev was configured without support
40	support for multiple event loops, then all functions taking an initial	122	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 *>)	123	name C<loop> (which is always of type C<ev_loop *>) will not have
42	will not have this argument.	124	this argument.
43		125
44	=head1 TIME AND OTHER GLOBAL FUNCTIONS	126	=head2 TIME REPRESENTATION
45		127
46	Libev represents time as a single floating point number, representing the	128	Libev represents time as a single floating point number, representing
47	(fractional) number of seconds since the (POSIX) epoch (somewhere near	129	the (fractional) number of seconds since the (POSIX) epoch (somewhere
48	the beginning of 1970, details are complicated, don't ask). This type is	130	near the beginning of 1970, details are complicated, don't ask). This
49	called C<ev_tstamp>, which is what you should use too. It usually aliases	131	type is called C<ev_tstamp>, which is what you should use too. It usually
50	to the double type in C.	132	aliases to the C<double> type in C. When you need to do any calculations
		133	on it, you should treat it as some floating point value. Unlike the name
		134	component C<stamp> might indicate, it is also used for time differences
		135	throughout libev.
		136
		137	=head1 ERROR HANDLING
		138
		139	Libev knows three classes of errors: operating system errors, usage errors
		140	and internal errors (bugs).
		141
		142	When libev catches an operating system error it cannot handle (for example
		143	a system call indicating a condition libev cannot fix), it calls the callback
		144	set via C<ev_set_syserr_cb>, which is supposed to fix the problem or
		145	abort. The default is to print a diagnostic message and to call C<abort
		146	()>.
		147
		148	When libev detects a usage error such as a negative timer interval, then
		149	it will print a diagnostic message and abort (via the C<assert> mechanism,
		150	so C<NDEBUG> will disable this checking): these are programming errors in
		151	the libev caller and need to be fixed there.
		152
		153	Libev also has a few internal error-checking C<assert>ions, and also has
		154	extensive consistency checking code. These do not trigger under normal
		155	circumstances, as they indicate either a bug in libev or worse.
		156
		157
		158	=head1 GLOBAL FUNCTIONS
		159
		160	These functions can be called anytime, even before initialising the
		161	library in any way.
51		162
52	=over 4	163	=over 4
53		164
54	=item ev_tstamp ev_time ()	165	=item ev_tstamp ev_time ()
55		166
56	Returns the current time as libev would use it.	167	Returns the current time as libev would use it. Please note that the
		168	C<ev_now> function is usually faster and also often returns the timestamp
		169	you actually want to know.
		170
		171	=item ev_sleep (ev_tstamp interval)
		172
		173	Sleep for the given interval: The current thread will be blocked until
		174	either it is interrupted or the given time interval has passed. Basically
		175	this is a sub-second-resolution C<sleep ()>.
57		176
58	=item int ev_version_major ()	177	=item int ev_version_major ()
59		178
60	=item int ev_version_minor ()	179	=item int ev_version_minor ()
61		180
62	You can find out the major and minor version numbers of the library	181	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	182	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	183	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	184	symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the
66	version of the library your program was compiled against.	185	version of the library your program was compiled against.
67		186
		187	These version numbers refer to the ABI version of the library, not the
		188	release version.
		189
68	Usually, its a good idea to terminate if the major versions mismatch,	190	Usually, it's a good idea to terminate if the major versions mismatch,
69	as this indicates an incompatible change. Minor versions are usually	191	as this indicates an incompatible change. Minor versions are usually
70	compatible to older versions, so a larger minor version alone is usually	192	compatible to older versions, so a larger minor version alone is usually
71	not a problem.	193	not a problem.
72		194
		195	Example: Make sure we haven't accidentally been linked against the wrong
		196	version.
		197
		198	assert (("libev version mismatch",
		199	ev_version_major () == EV_VERSION_MAJOR
		200	&& ev_version_minor () >= EV_VERSION_MINOR));
		201
		202	=item unsigned int ev_supported_backends ()
		203
		204	Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
		205	value) compiled into this binary of libev (independent of their
		206	availability on the system you are running on). See C<ev_default_loop> for
		207	a description of the set values.
		208
		209	Example: make sure we have the epoll method, because yeah this is cool and
		210	a must have and can we have a torrent of it please!!!11
		211
		212	assert (("sorry, no epoll, no sex",
		213	ev_supported_backends () & EVBACKEND_EPOLL));
		214
		215	=item unsigned int ev_recommended_backends ()
		216
		217	Return the set of all backends compiled into this binary of libev and also
		218	recommended for this platform. This set is often smaller than the one
		219	returned by C<ev_supported_backends>, as for example kqueue is broken on
		220	most BSDs and will not be auto-detected unless you explicitly request it
		221	(assuming you know what you are doing). This is the set of backends that
		222	libev will probe for if you specify no backends explicitly.
		223
		224	=item unsigned int ev_embeddable_backends ()
		225
		226	Returns the set of backends that are embeddable in other event loops. This
		227	is the theoretical, all-platform, value. To find which backends
		228	might be supported on the current system, you would need to look at
		229	C<ev_embeddable_backends () & ev_supported_backends ()>, likewise for
		230	recommended ones.
		231
		232	See the description of C<ev_embed> watchers for more info.
		233
73	=item ev_set_allocator (void *(*cb)(void *ptr, long size))	234	=item ev_set_allocator (void *(*cb)(void *ptr, long size)) [NOT REENTRANT]
74		235
75	Sets the allocation function to use (the prototype is similar to the	236	Sets the allocation function to use (the prototype is similar - the
76	realloc C function, the semantics are identical). It is used to allocate	237	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	238	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	239	when memory needs to be allocated (C<size != 0>), the library might abort
79	destructive action. The default is your system realloc function.	240	or take some potentially destructive action.
		241
		242	Since some systems (at least OpenBSD and Darwin) fail to implement
		243	correct C<realloc> semantics, libev will use a wrapper around the system
		244	C<realloc> and C<free> functions by default.
80		245
81	You could override this function in high-availability programs to, say,	246	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,	247	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.	248	or even to sleep a while and retry until some memory is available.
84		249
		250	Example: Replace the libev allocator with one that waits a bit and then
		251	retries (example requires a standards-compliant C<realloc>).
		252
		253	static void *
		254	persistent_realloc (void *ptr, size_t size)
		255	{
		256	for (;;)
		257	{
		258	void *newptr = realloc (ptr, size);
		259
		260	if (newptr)
		261	return newptr;
		262
		263	sleep (60);
		264	}
		265	}
		266
		267	...
		268	ev_set_allocator (persistent_realloc);
		269
85	=item ev_set_syserr_cb (void (*cb)(const char *msg));	270	=item ev_set_syserr_cb (void (*cb)(const char *msg)); [NOT REENTRANT]
86		271
87	Set the callback function to call on a retryable syscall error (such	272	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	273	as failed select, poll, epoll_wait). The message is a printable string
89	indicating the system call or subsystem causing the problem. If this	274	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	275	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	276	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	277	requested operation, or, if the condition doesn't go away, do bad stuff
93	(such as abort).	278	(such as abort).
94		279
		280	Example: This is basically the same thing that libev does internally, too.
		281
		282	static void
		283	fatal_error (const char *msg)
		284	{
		285	perror (msg);
		286	abort ();
		287	}
		288
		289	...
		290	ev_set_syserr_cb (fatal_error);
		291
95	=back	292	=back
96		293
97	=head1 FUNCTIONS CONTROLLING THE EVENT LOOP	294	=head1 FUNCTIONS CONTROLLING THE EVENT LOOP
98		295
99	An event loop is described by a C<struct ev_loop *>. The library knows two	296	An event loop is described by a C<struct ev_loop *> (the C<struct>
100	types of such loops, the I<default> loop, which supports signals and child	297	is I<not> optional in this case, as there is also an C<ev_loop>
101	events, and dynamically created loops which do not.	298	I<function>).
102		299
103	If you use threads, a common model is to run the default event loop	300	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	301	supports signals and child events, and dynamically created loops which do
105	create, you also create another event loop. Libev itself does no locking	302	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 its hideous and inefficient).
109		303
110	=over 4	304	=over 4
111		305
112	=item struct ev_loop *ev_default_loop (unsigned int flags)	306	=item struct ev_loop *ev_default_loop (unsigned int flags)
113		307
114	This will initialise the default event loop if it hasn't been initialised	308	This will initialise the default event loop if it hasn't been initialised
115	yet and return it. If the default loop could not be initialised, returns	309	yet and return it. If the default loop could not be initialised, returns
116	false. If it already was initialised it simply returns it (and ignores the	310	false. If it already was initialised it simply returns it (and ignores the
117	flags).	311	flags. If that is troubling you, check C<ev_backend ()> afterwards).
118		312
119	If you don't know what event loop to use, use the one returned from this	313	If you don't know what event loop to use, use the one returned from this
120	function.	314	function.
121		315
		316	Note that this function is I<not> thread-safe, so if you want to use it
		317	from multiple threads, you have to lock (note also that this is unlikely,
		318	as loops cannot be shared easily between threads anyway).
		319
		320	The default loop is the only loop that can handle C<ev_signal> and
		321	C<ev_child> watchers, and to do this, it always registers a handler
		322	for C<SIGCHLD>. If this is a problem for your application you can either
		323	create a dynamic loop with C<ev_loop_new> that doesn't do that, or you
		324	can simply overwrite the C<SIGCHLD> signal handler I<after> calling
		325	C<ev_default_init>.
		326
122	The flags argument can be used to specify special behaviour or specific	327	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).	328	backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
124		329
125	It supports the following flags:	330	The following flags are supported:
126		331
127	=over 4	332	=over 4
128		333
129	=item EVFLAG_AUTO	334	=item C<EVFLAG_AUTO>
130		335
131	The default flags value. Use this if you have no clue (its the right	336	The default flags value. Use this if you have no clue (it's the right
132	thing, believe me).	337	thing, believe me).
133		338
134	=item EVFLAG_NOENV	339	=item C<EVFLAG_NOENV>
135		340
136	If this flag bit is ored into the flag value (or the program runs setuid	341	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	342	or setgid) then libev will I<not> look at the environment variable
138	C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will	343	C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
139	override the flags completely if it is found in the environment. This is	344	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	345	useful to try out specific backends to test their performance, or to work
141	around bugs.	346	around bugs.
142		347
		348	=item C<EVFLAG_FORKCHECK>
		349
		350	Instead of calling C<ev_default_fork> or C<ev_loop_fork> manually after
		351	a fork, you can also make libev check for a fork in each iteration by
		352	enabling this flag.
		353
		354	This works by calling C<getpid ()> on every iteration of the loop,
		355	and thus this might slow down your event loop if you do a lot of loop
		356	iterations and little real work, but is usually not noticeable (on my
		357	GNU/Linux system for example, C<getpid> is actually a simple 5-insn sequence
		358	without a system call and thus I<very> fast, but my GNU/Linux system also has
		359	C<pthread_atfork> which is even faster).
		360
		361	The big advantage of this flag is that you can forget about fork (and
		362	forget about forgetting to tell libev about forking) when you use this
		363	flag.
		364
		365	This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
		366	environment variable.
		367
		368	=item C<EVFLAG_NOINOTIFY>
		369
		370	When this flag is specified, then libev will not attempt to use the
		371	I<inotify> API for it's C<ev_stat> watchers. Apart from debugging and
		372	testing, this flag can be useful to conserve inotify file descriptors, as
		373	otherwise each loop using C<ev_stat> watchers consumes one inotify handle.
		374
		375	=item C<EVFLAG_NOSIGFD>
		376
		377	When this flag is specified, then libev will not attempt to use the
		378	I<signalfd> API for it's C<ev_signal> (and C<ev_child>) watchers. This is
		379	probably only useful to work around any bugs in libev. Consequently, this
		380	flag might go away once the signalfd functionality is considered stable,
		381	so it's useful mostly in environment variables and not in program code.
		382
143	=item EVMETHOD_SELECT portable select backend	383	=item C<EVBACKEND_SELECT> (value 1, portable select backend)
144		384
145	=item EVMETHOD_POLL poll backend (everywhere except windows)	385	This is your standard select(2) backend. Not I<completely> standard, as
		386	libev tries to roll its own fd_set with no limits on the number of fds,
		387	but if that fails, expect a fairly low limit on the number of fds when
		388	using this backend. It doesn't scale too well (O(highest_fd)), but its
		389	usually the fastest backend for a low number of (low-numbered :) fds.
146		390
147	=item EVMETHOD_EPOLL linux only	391	To get good performance out of this backend you need a high amount of
		392	parallelism (most of the file descriptors should be busy). If you are
		393	writing a server, you should C<accept ()> in a loop to accept as many
		394	connections as possible during one iteration. You might also want to have
		395	a look at C<ev_set_io_collect_interval ()> to increase the amount of
		396	readiness notifications you get per iteration.
148		397
149	=item EVMETHOD_KQUEUE some bsds only	398	This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
		399	C<writefds> set (and to work around Microsoft Windows bugs, also onto the
		400	C<exceptfds> set on that platform).
150		401
151	=item EVMETHOD_DEVPOLL solaris 8 only	402	=item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
152		403
153	=item EVMETHOD_PORT solaris 10 only	404	And this is your standard poll(2) backend. It's more complicated
		405	than select, but handles sparse fds better and has no artificial
		406	limit on the number of fds you can use (except it will slow down
		407	considerably with a lot of inactive fds). It scales similarly to select,
		408	i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
		409	performance tips.
154		410
155	If one or more of these are ored into the flags value, then only these	411	This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
156	backends will be tried (in the reverse order as given here). If one are	412	C<EV_WRITE> to C<POLLOUT | POLLERR | POLLHUP>.
157	specified, any backend will do.	413
		414	=item C<EVBACKEND_EPOLL> (value 4, Linux)
		415
		416	Use the linux-specific epoll(7) interface (for both pre- and post-2.6.9
		417	kernels).
		418
		419	For few fds, this backend is a bit little slower than poll and select,
		420	but it scales phenomenally better. While poll and select usually scale
		421	like O(total_fds) where n is the total number of fds (or the highest fd),
		422	epoll scales either O(1) or O(active_fds).
		423
		424	The epoll mechanism deserves honorable mention as the most misdesigned
		425	of the more advanced event mechanisms: mere annoyances include silently
		426	dropping file descriptors, requiring a system call per change per file
		427	descriptor (and unnecessary guessing of parameters), problems with dup and
		428	so on. The biggest issue is fork races, however - if a program forks then
		429	I<both> parent and child process have to recreate the epoll set, which can
		430	take considerable time (one syscall per file descriptor) and is of course
		431	hard to detect.
		432
		433	Epoll is also notoriously buggy - embedding epoll fds I<should> work, but
		434	of course I<doesn't>, and epoll just loves to report events for totally
		435	I<different> file descriptors (even already closed ones, so one cannot
		436	even remove them from the set) than registered in the set (especially
		437	on SMP systems). Libev tries to counter these spurious notifications by
		438	employing an additional generation counter and comparing that against the
		439	events to filter out spurious ones, recreating the set when required.
		440
		441	While stopping, setting and starting an I/O watcher in the same iteration
		442	will result in some caching, there is still a system call per such
		443	incident (because the same I<file descriptor> could point to a different
		444	I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
		445	file descriptors might not work very well if you register events for both
		446	file descriptors.
		447
		448	Best performance from this backend is achieved by not unregistering all
		449	watchers for a file descriptor until it has been closed, if possible,
		450	i.e. keep at least one watcher active per fd at all times. Stopping and
		451	starting a watcher (without re-setting it) also usually doesn't cause
		452	extra overhead. A fork can both result in spurious notifications as well
		453	as in libev having to destroy and recreate the epoll object, which can
		454	take considerable time and thus should be avoided.
		455
		456	All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
		457	faster than epoll for maybe up to a hundred file descriptors, depending on
		458	the usage. So sad.
		459
		460	While nominally embeddable in other event loops, this feature is broken in
		461	all kernel versions tested so far.
		462
		463	This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
		464	C<EVBACKEND_POLL>.
		465
		466	=item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
		467
		468	Kqueue deserves special mention, as at the time of this writing, it
		469	was broken on all BSDs except NetBSD (usually it doesn't work reliably
		470	with anything but sockets and pipes, except on Darwin, where of course
		471	it's completely useless). Unlike epoll, however, whose brokenness
		472	is by design, these kqueue bugs can (and eventually will) be fixed
		473	without API changes to existing programs. For this reason it's not being
		474	"auto-detected" unless you explicitly specify it in the flags (i.e. using
		475	C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
		476	system like NetBSD.
		477
		478	You still can embed kqueue into a normal poll or select backend and use it
		479	only for sockets (after having made sure that sockets work with kqueue on
		480	the target platform). See C<ev_embed> watchers for more info.
		481
		482	It scales in the same way as the epoll backend, but the interface to the
		483	kernel is more efficient (which says nothing about its actual speed, of
		484	course). While stopping, setting and starting an I/O watcher does never
		485	cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
		486	two event changes per incident. Support for C<fork ()> is very bad (but
		487	sane, unlike epoll) and it drops fds silently in similarly hard-to-detect
		488	cases
		489
		490	This backend usually performs well under most conditions.
		491
		492	While nominally embeddable in other event loops, this doesn't work
		493	everywhere, so you might need to test for this. And since it is broken
		494	almost everywhere, you should only use it when you have a lot of sockets
		495	(for which it usually works), by embedding it into another event loop
		496	(e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
		497	also broken on OS X)) and, did I mention it, using it only for sockets.
		498
		499	This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
		500	C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
		501	C<NOTE_EOF>.
		502
		503	=item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
		504
		505	This is not implemented yet (and might never be, unless you send me an
		506	implementation). According to reports, C</dev/poll> only supports sockets
		507	and is not embeddable, which would limit the usefulness of this backend
		508	immensely.
		509
		510	=item C<EVBACKEND_PORT> (value 32, Solaris 10)
		511
		512	This uses the Solaris 10 event port mechanism. As with everything on Solaris,
		513	it's really slow, but it still scales very well (O(active_fds)).
		514
		515	Please note that Solaris event ports can deliver a lot of spurious
		516	notifications, so you need to use non-blocking I/O or other means to avoid
		517	blocking when no data (or space) is available.
		518
		519	While this backend scales well, it requires one system call per active
		520	file descriptor per loop iteration. For small and medium numbers of file
		521	descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
		522	might perform better.
		523
		524	On the positive side, with the exception of the spurious readiness
		525	notifications, this backend actually performed fully to specification
		526	in all tests and is fully embeddable, which is a rare feat among the
		527	OS-specific backends (I vastly prefer correctness over speed hacks).
		528
		529	This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
		530	C<EVBACKEND_POLL>.
		531
		532	=item C<EVBACKEND_ALL>
		533
		534	Try all backends (even potentially broken ones that wouldn't be tried
		535	with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
		536	C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.
		537
		538	It is definitely not recommended to use this flag.
158		539
159	=back	540	=back
		541
		542	If one or more of the backend flags are or'ed into the flags value,
		543	then only these backends will be tried (in the reverse order as listed
		544	here). If none are specified, all backends in C<ev_recommended_backends
		545	()> will be tried.
		546
		547	Example: This is the most typical usage.
		548
		549	if (!ev_default_loop (0))
		550	fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
		551
		552	Example: Restrict libev to the select and poll backends, and do not allow
		553	environment settings to be taken into account:
		554
		555	ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
		556
		557	Example: Use whatever libev has to offer, but make sure that kqueue is
		558	used if available (warning, breaks stuff, best use only with your own
		559	private event loop and only if you know the OS supports your types of
		560	fds):
		561
		562	ev_default_loop (ev_recommended_backends () | EVBACKEND_KQUEUE);
160		563
161	=item struct ev_loop *ev_loop_new (unsigned int flags)	564	=item struct ev_loop *ev_loop_new (unsigned int flags)
162		565
163	Similar to C<ev_default_loop>, but always creates a new event loop that is	566	Similar to C<ev_default_loop>, but always creates a new event loop that is
164	always distinct from the default loop. Unlike the default loop, it cannot	567	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	568	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).	569	undefined behaviour (or a failed assertion if assertions are enabled).
167		570
		571	Note that this function I<is> thread-safe, and the recommended way to use
		572	libev with threads is indeed to create one loop per thread, and using the
		573	default loop in the "main" or "initial" thread.
		574
		575	Example: Try to create a event loop that uses epoll and nothing else.
		576
		577	struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
		578	if (!epoller)
		579	fatal ("no epoll found here, maybe it hides under your chair");
		580
168	=item ev_default_destroy ()	581	=item ev_default_destroy ()
169		582
170	Destroys the default loop again (frees all memory and kernel state	583	Destroys the default loop again (frees all memory and kernel state
171	etc.). This stops all registered event watchers (by not touching them in	584	etc.). None of the active event watchers will be stopped in the normal
172	any way whatsoever, although you cnanot rely on this :).	585	sense, so e.g. C<ev_is_active> might still return true. It is your
		586	responsibility to either stop all watchers cleanly yourself I<before>
		587	calling this function, or cope with the fact afterwards (which is usually
		588	the easiest thing, you can just ignore the watchers and/or C<free ()> them
		589	for example).
		590
		591	Note that certain global state, such as signal state (and installed signal
		592	handlers), will not be freed by this function, and related watchers (such
		593	as signal and child watchers) would need to be stopped manually.
		594
		595	In general it is not advisable to call this function except in the
		596	rare occasion where you really need to free e.g. the signal handling
		597	pipe fds. If you need dynamically allocated loops it is better to use
		598	C<ev_loop_new> and C<ev_loop_destroy>.
173		599
174	=item ev_loop_destroy (loop)	600	=item ev_loop_destroy (loop)
175		601
176	Like C<ev_default_destroy>, but destroys an event loop created by an	602	Like C<ev_default_destroy>, but destroys an event loop created by an
177	earlier call to C<ev_loop_new>.	603	earlier call to C<ev_loop_new>.
178		604
179	=item ev_default_fork ()	605	=item ev_default_fork ()
180		606
		607	This function sets a flag that causes subsequent C<ev_loop> iterations
181	This function reinitialises the kernel state for backends that have	608	to 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	609	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	610	the child process (or both child and parent, but that again makes little
184	again makes little sense).	611	sense). You I<must> call it in the child before using any of the libev
		612	functions, and it will only take effect at the next C<ev_loop> iteration.
185		613
186	You I<must> call this function after forking if and only if you want to	614	On the other hand, you only need to call this function in the child
187	use the event library in both processes. If you just fork+exec, you don't	615	process if and only if you want to use the event library in the child. If
188	have to call it.	616	you just fork+exec, you don't have to call it at all.
189		617
190	The function itself is quite fast and its usually not a problem to call	618	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	619	it just in case after a fork. To make this easy, the function will fit in
192	quite nicely into a call to C<pthread_atfork>:	620	quite nicely into a call to C<pthread_atfork>:
193		621
194	pthread_atfork (0, 0, ev_default_fork);	622	pthread_atfork (0, 0, ev_default_fork);
195		623
196	=item ev_loop_fork (loop)	624	=item ev_loop_fork (loop)
197		625
198	Like C<ev_default_fork>, but acts on an event loop created by	626	Like C<ev_default_fork>, but acts on an event loop created by
199	C<ev_loop_new>. Yes, you have to call this on every allocated event loop	627	C<ev_loop_new>. Yes, you have to call this on every allocated event loop
200	after fork, and how you do this is entirely your own problem.	628	after fork that you want to re-use in the child, and how you do this is
		629	entirely your own problem.
201		630
		631	=item int ev_is_default_loop (loop)
		632
		633	Returns true when the given loop is, in fact, the default loop, and false
		634	otherwise.
		635
		636	=item unsigned int ev_loop_count (loop)
		637
		638	Returns the count of loop iterations for the loop, which is identical to
		639	the number of times libev did poll for new events. It starts at C<0> and
		640	happily wraps around with enough iterations.
		641
		642	This value can sometimes be useful as a generation counter of sorts (it
		643	"ticks" the number of loop iterations), as it roughly corresponds with
		644	C<ev_prepare> and C<ev_check> calls.
		645
		646	=item unsigned int ev_loop_depth (loop)
		647
		648	Returns the number of times C<ev_loop> was entered minus the number of
		649	times C<ev_loop> was exited, in other words, the recursion depth.
		650
		651	Outside C<ev_loop>, this number is zero. In a callback, this number is
		652	C<1>, unless C<ev_loop> was invoked recursively (or from another thread),
		653	in which case it is higher.
		654
		655	Leaving C<ev_loop> abnormally (setjmp/longjmp, cancelling the thread
		656	etc.), doesn't count as exit.
		657
202	=item unsigned int ev_method (loop)	658	=item unsigned int ev_backend (loop)
203		659
204	Returns one of the C<EVMETHOD_*> flags indicating the event backend in	660	Returns one of the C<EVBACKEND_*> flags indicating the event backend in
205	use.	661	use.
206		662
207	=item ev_tstamp = ev_now (loop)	663	=item ev_tstamp ev_now (loop)
208		664
209	Returns the current "event loop time", which is the time the event loop	665	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	666	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	667	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	668	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).	669	event occurring (or more correctly, libev finding out about it).
		670
		671	=item ev_now_update (loop)
		672
		673	Establishes the current time by querying the kernel, updating the time
		674	returned by C<ev_now ()> in the progress. This is a costly operation and
		675	is usually done automatically within C<ev_loop ()>.
		676
		677	This function is rarely useful, but when some event callback runs for a
		678	very long time without entering the event loop, updating libev's idea of
		679	the current time is a good idea.
		680
		681	See also L<The special problem of time updates> in the C<ev_timer> section.
		682
		683	=item ev_suspend (loop)
		684
		685	=item ev_resume (loop)
		686
		687	These two functions suspend and resume a loop, for use when the loop is
		688	not used for a while and timeouts should not be processed.
		689
		690	A typical use case would be an interactive program such as a game: When
		691	the user presses C<^Z> to suspend the game and resumes it an hour later it
		692	would be best to handle timeouts as if no time had actually passed while
		693	the program was suspended. This can be achieved by calling C<ev_suspend>
		694	in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
		695	C<ev_resume> directly afterwards to resume timer processing.
		696
		697	Effectively, all C<ev_timer> watchers will be delayed by the time spend
		698	between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
		699	will be rescheduled (that is, they will lose any events that would have
		700	occured while suspended).
		701
		702	After calling C<ev_suspend> you B<must not> call I<any> function on the
		703	given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
		704	without a previous call to C<ev_suspend>.
		705
		706	Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
		707	event loop time (see C<ev_now_update>).
214		708
215	=item ev_loop (loop, int flags)	709	=item ev_loop (loop, int flags)
216		710
217	Finally, this is it, the event handler. This function usually is called	711	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	712	after you have initialised all your watchers and you want to start
219	events.	713	handling events.
220		714
221	If the flags argument is specified as 0, it will not return until either	715	If the flags argument is specified as C<0>, it will not return until
222	no event watchers are active anymore or C<ev_unloop> was called.	716	either no event watchers are active anymore or C<ev_unloop> was called.
		717
		718	Please note that an explicit C<ev_unloop> is usually better than
		719	relying on all watchers to be stopped when deciding when a program has
		720	finished (especially in interactive programs), but having a program
		721	that automatically loops as long as it has to and no longer by virtue
		722	of relying on its watchers stopping correctly, that is truly a thing of
		723	beauty.
223		724
224	A flags value of C<EVLOOP_NONBLOCK> will look for new events, will handle	725	A flags value of C<EVLOOP_NONBLOCK> will look for new events, will handle
225	those events and any outstanding ones, but will not block your process in	726	those events and any already outstanding ones, but will not block your
226	case there are no events.	727	process in case there are no events and will return after one iteration of
		728	the loop.
227		729
228	A flags value of C<EVLOOP_ONESHOT> will look for new events (waiting if	730	A flags value of C<EVLOOP_ONESHOT> will look for new events (waiting if
229	neccessary) and will handle those and any outstanding ones. It will block	731	necessary) and will handle those and any already outstanding ones. It
230	your process until at least one new event arrives.	732	will block your process until at least one new event arrives (which could
		733	be an event internal to libev itself, so there is no guarantee that a
		734	user-registered callback will be called), and will return after one
		735	iteration of the loop.
231		736
232	This flags value could be used to implement alternative looping	737	This is useful if you are waiting for some external event in conjunction
233	constructs, but the C<prepare> and C<check> watchers provide a better and	738	with something not expressible using other libev watchers (i.e. "roll your
234	more generic mechanism.	739	own C<ev_loop>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
		740	usually a better approach for this kind of thing.
		741
		742	Here are the gory details of what C<ev_loop> does:
		743
		744	- Before the first iteration, call any pending watchers.
		745	* If EVFLAG_FORKCHECK was used, check for a fork.
		746	- If a fork was detected (by any means), queue and call all fork watchers.
		747	- Queue and call all prepare watchers.
		748	- If we have been forked, detach and recreate the kernel state
		749	as to not disturb the other process.
		750	- Update the kernel state with all outstanding changes.
		751	- Update the "event loop time" (ev_now ()).
		752	- Calculate for how long to sleep or block, if at all
		753	(active idle watchers, EVLOOP_NONBLOCK or not having
		754	any active watchers at all will result in not sleeping).
		755	- Sleep if the I/O and timer collect interval say so.
		756	- Block the process, waiting for any events.
		757	- Queue all outstanding I/O (fd) events.
		758	- Update the "event loop time" (ev_now ()), and do time jump adjustments.
		759	- Queue all expired timers.
		760	- Queue all expired periodics.
		761	- Unless any events are pending now, queue all idle watchers.
		762	- Queue all check watchers.
		763	- Call all queued watchers in reverse order (i.e. check watchers first).
		764	Signals and child watchers are implemented as I/O watchers, and will
		765	be handled here by queueing them when their watcher gets executed.
		766	- If ev_unloop has been called, or EVLOOP_ONESHOT or EVLOOP_NONBLOCK
		767	were used, or there are no active watchers, return, otherwise
		768	continue with step *.
		769
		770	Example: Queue some jobs and then loop until no events are outstanding
		771	anymore.
		772
		773	... queue jobs here, make sure they register event watchers as long
		774	... as they still have work to do (even an idle watcher will do..)
		775	ev_loop (my_loop, 0);
		776	... jobs done or somebody called unloop. yeah!
235		777
236	=item ev_unloop (loop, how)	778	=item ev_unloop (loop, how)
237		779
238	Can be used to make a call to C<ev_loop> return early. The C<how> argument	780	Can be used to make a call to C<ev_loop> return early (but only after it
		781	has processed all outstanding events). The C<how> argument must be either
239	must be either C<EVUNLOOP_ONCE>, which will make the innermost C<ev_loop>	782	C<EVUNLOOP_ONE>, which will make the innermost C<ev_loop> call return, or
240	call return, or C<EVUNLOOP_ALL>, which will make all nested C<ev_loop>	783	C<EVUNLOOP_ALL>, which will make all nested C<ev_loop> calls return.
241	calls return.	784
		785	This "unloop state" will be cleared when entering C<ev_loop> again.
		786
		787	It is safe to call C<ev_unloop> from otuside any C<ev_loop> calls.
242		788
243	=item ev_ref (loop)	789	=item ev_ref (loop)
244		790
245	=item ev_unref (loop)	791	=item ev_unref (loop)
246		792
247	Ref/unref can be used to add or remove a refcount on the event loop: Every	793	Ref/unref can be used to add or remove a reference count on the event
248	watcher keeps one reference. If you have a long-runing watcher you never	794	loop: Every watcher keeps one reference, and as long as the reference
249	unregister that should not keep ev_loop from running, ev_unref() after	795	count is nonzero, C<ev_loop> will not return on its own.
250	starting, and ev_ref() before stopping it. Libev itself uses this for	796
251	example for its internal signal pipe: It is not visible to you as a user	797	If you have a watcher you never unregister that should not keep C<ev_loop>
252	and should not keep C<ev_loop> from exiting if the work is done. It is	798	from returning, call ev_unref() after starting, and ev_ref() before
		799	stopping it.
		800
		801	As an example, libev itself uses this for its internal signal pipe: It
		802	is not visible to the libev user and should not keep C<ev_loop> from
		803	exiting if no event watchers registered by it are active. It is also an
253	also an excellent way to do this for generic recurring timers or from	804	excellent way to do this for generic recurring timers or from within
254	within third-party libraries. Just remember to unref after start and ref	805	third-party libraries. Just remember to I<unref after start> and I<ref
255	before stop.	806	before stop> (but only if the watcher wasn't active before, or was active
		807	before, respectively. Note also that libev might stop watchers itself
		808	(e.g. non-repeating timers) in which case you have to C<ev_ref>
		809	in the callback).
		810
		811	Example: Create a signal watcher, but keep it from keeping C<ev_loop>
		812	running when nothing else is active.
		813
		814	ev_signal exitsig;
		815	ev_signal_init (&exitsig, sig_cb, SIGINT);
		816	ev_signal_start (loop, &exitsig);
		817	evf_unref (loop);
		818
		819	Example: For some weird reason, unregister the above signal handler again.
		820
		821	ev_ref (loop);
		822	ev_signal_stop (loop, &exitsig);
		823
		824	=item ev_set_io_collect_interval (loop, ev_tstamp interval)
		825
		826	=item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
		827
		828	These advanced functions influence the time that libev will spend waiting
		829	for events. Both time intervals are by default C<0>, meaning that libev
		830	will try to invoke timer/periodic callbacks and I/O callbacks with minimum
		831	latency.
		832
		833	Setting these to a higher value (the C<interval> I<must> be >= C<0>)
		834	allows libev to delay invocation of I/O and timer/periodic callbacks
		835	to increase efficiency of loop iterations (or to increase power-saving
		836	opportunities).
		837
		838	The idea is that sometimes your program runs just fast enough to handle
		839	one (or very few) event(s) per loop iteration. While this makes the
		840	program responsive, it also wastes a lot of CPU time to poll for new
		841	events, especially with backends like C<select ()> which have a high
		842	overhead for the actual polling but can deliver many events at once.
		843
		844	By setting a higher I<io collect interval> you allow libev to spend more
		845	time collecting I/O events, so you can handle more events per iteration,
		846	at the cost of increasing latency. Timeouts (both C<ev_periodic> and
		847	C<ev_timer>) will be not affected. Setting this to a non-null value will
		848	introduce an additional C<ev_sleep ()> call into most loop iterations. The
		849	sleep time ensures that libev will not poll for I/O events more often then
		850	once per this interval, on average.
		851
		852	Likewise, by setting a higher I<timeout collect interval> you allow libev
		853	to spend more time collecting timeouts, at the expense of increased
		854	latency/jitter/inexactness (the watcher callback will be called
		855	later). C<ev_io> watchers will not be affected. Setting this to a non-null
		856	value will not introduce any overhead in libev.
		857
		858	Many (busy) programs can usually benefit by setting the I/O collect
		859	interval to a value near C<0.1> or so, which is often enough for
		860	interactive servers (of course not for games), likewise for timeouts. It
		861	usually doesn't make much sense to set it to a lower value than C<0.01>,
		862	as this approaches the timing granularity of most systems. Note that if
		863	you do transactions with the outside world and you can't increase the
		864	parallelity, then this setting will limit your transaction rate (if you
		865	need to poll once per transaction and the I/O collect interval is 0.01,
		866	then you can't do more than 100 transations per second).
		867
		868	Setting the I<timeout collect interval> can improve the opportunity for
		869	saving power, as the program will "bundle" timer callback invocations that
		870	are "near" in time together, by delaying some, thus reducing the number of
		871	times the process sleeps and wakes up again. Another useful technique to
		872	reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
		873	they fire on, say, one-second boundaries only.
		874
		875	Example: we only need 0.1s timeout granularity, and we wish not to poll
		876	more often than 100 times per second:
		877
		878	ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
		879	ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
		880
		881	=item ev_invoke_pending (loop)
		882
		883	This call will simply invoke all pending watchers while resetting their
		884	pending state. Normally, C<ev_loop> does this automatically when required,
		885	but when overriding the invoke callback this call comes handy.
		886
		887	=item int ev_pending_count (loop)
		888
		889	Returns the number of pending watchers - zero indicates that no watchers
		890	are pending.
		891
		892	=item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
		893
		894	This overrides the invoke pending functionality of the loop: Instead of
		895	invoking all pending watchers when there are any, C<ev_loop> will call
		896	this callback instead. This is useful, for example, when you want to
		897	invoke the actual watchers inside another context (another thread etc.).
		898
		899	If you want to reset the callback, use C<ev_invoke_pending> as new
		900	callback.
		901
		902	=item ev_set_loop_release_cb (loop, void (*release)(EV_P), void (*acquire)(EV_P))
		903
		904	Sometimes you want to share the same loop between multiple threads. This
		905	can be done relatively simply by putting mutex_lock/unlock calls around
		906	each call to a libev function.
		907
		908	However, C<ev_loop> can run an indefinite time, so it is not feasible to
		909	wait for it to return. One way around this is to wake up the loop via
		910	C<ev_unloop> and C<av_async_send>, another way is to set these I<release>
		911	and I<acquire> callbacks on the loop.
		912
		913	When set, then C<release> will be called just before the thread is
		914	suspended waiting for new events, and C<acquire> is called just
		915	afterwards.
		916
		917	Ideally, C<release> will just call your mutex_unlock function, and
		918	C<acquire> will just call the mutex_lock function again.
		919
		920	While event loop modifications are allowed between invocations of
		921	C<release> and C<acquire> (that's their only purpose after all), no
		922	modifications done will affect the event loop, i.e. adding watchers will
		923	have no effect on the set of file descriptors being watched, or the time
		924	waited. USe an C<ev_async> watcher to wake up C<ev_loop> when you want it
		925	to take note of any changes you made.
		926
		927	In theory, threads executing C<ev_loop> will be async-cancel safe between
		928	invocations of C<release> and C<acquire>.
		929
		930	See also the locking example in the C<THREADS> section later in this
		931	document.
		932
		933	=item ev_set_userdata (loop, void *data)
		934
		935	=item ev_userdata (loop)
		936
		937	Set and retrieve a single C<void *> associated with a loop. When
		938	C<ev_set_userdata> has never been called, then C<ev_userdata> returns
		939	C<0.>
		940
		941	These two functions can be used to associate arbitrary data with a loop,
		942	and are intended solely for the C<invoke_pending_cb>, C<release> and
		943	C<acquire> callbacks described above, but of course can be (ab-)used for
		944	any other purpose as well.
		945
		946	=item ev_loop_verify (loop)
		947
		948	This function only does something when C<EV_VERIFY> support has been
		949	compiled in, which is the default for non-minimal builds. It tries to go
		950	through all internal structures and checks them for validity. If anything
		951	is found to be inconsistent, it will print an error message to standard
		952	error and call C<abort ()>.
		953
		954	This can be used to catch bugs inside libev itself: under normal
		955	circumstances, this function will never abort as of course libev keeps its
		956	data structures consistent.
256		957
257	=back	958	=back
258		959
		960
259	=head1 ANATOMY OF A WATCHER	961	=head1 ANATOMY OF A WATCHER
		962
		963	In the following description, uppercase C<TYPE> in names stands for the
		964	watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
		965	watchers and C<ev_io_start> for I/O watchers.
260		966
261	A watcher is a structure that you create and register to record your	967	A watcher is a structure that you create and register to record your
262	interest in some event. For instance, if you want to wait for STDIN to	968	interest in some event. For instance, if you want to wait for STDIN to
263	become readable, you would create an ev_io watcher for that:	969	become readable, you would create an C<ev_io> watcher for that:
264		970
265	static void my_cb (struct ev_loop *loop, struct ev_io *w, int revents)	971	static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
266	{	972	{
267	ev_io_stop (w);	973	ev_io_stop (w);
268	ev_unloop (loop, EVUNLOOP_ALL);	974	ev_unloop (loop, EVUNLOOP_ALL);
269	}	975	}
270		976
271	struct ev_loop *loop = ev_default_loop (0);	977	struct ev_loop *loop = ev_default_loop (0);
		978
272	struct ev_io stdin_watcher;	979	ev_io stdin_watcher;
		980
273	ev_init (&stdin_watcher, my_cb);	981	ev_init (&stdin_watcher, my_cb);
274	ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);	982	ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
275	ev_io_start (loop, &stdin_watcher);	983	ev_io_start (loop, &stdin_watcher);
		984
276	ev_loop (loop, 0);	985	ev_loop (loop, 0);
277		986
278	As you can see, you are responsible for allocating the memory for your	987	As you can see, you are responsible for allocating the memory for your
279	watcher structures (and it is usually a bad idea to do this on the stack,	988	watcher structures (and it is I<usually> a bad idea to do this on the
280	although this can sometimes be quite valid).	989	stack).
		990
		991	Each watcher has an associated watcher structure (called C<struct ev_TYPE>
		992	or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
281		993
282	Each watcher structure must be initialised by a call to C<ev_init	994	Each watcher structure must be initialised by a call to C<ev_init
283	(watcher *, callback)>, which expects a callback to be provided. This	995	(watcher *, callback)>, which expects a callback to be provided. This
284	callback gets invoked each time the event occurs (or, in the case of io	996	callback gets invoked each time the event occurs (or, in the case of I/O
285	watchers, each time the event loop detects that the file descriptor given	997	watchers, each time the event loop detects that the file descriptor given
286	is readable and/or writable).	998	is readable and/or writable).
287		999
288	Each watcher type has its own C<< ev_<type>_set (watcher *, ...) >> macro	1000	Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
289	with arguments specific to this watcher type. There is also a macro	1001	macro to configure it, with arguments specific to the watcher type. There
290	to combine initialisation and setting in one call: C<< ev_<type>_init	1002	is also a macro to combine initialisation and setting in one call: C<<
291	(watcher *, callback, ...) >>.	1003	ev_TYPE_init (watcher *, callback, ...) >>.
292		1004
293	To make the watcher actually watch out for events, you have to start it	1005	To make the watcher actually watch out for events, you have to start it
294	with a watcher-specific start function (C<< ev_<type>_start (loop, watcher	1006	with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
295	*) >>), and you can stop watching for events at any time by calling the	1007	*) >>), and you can stop watching for events at any time by calling the
296	corresponding stop function (C<< ev_<type>_stop (loop, watcher *) >>.	1008	corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
297		1009
298	As long as your watcher is active (has been started but not stopped) you	1010	As long as your watcher is active (has been started but not stopped) you
299	must not touch the values stored in it. Most specifically you must never	1011	must not touch the values stored in it. Most specifically you must never
300	reinitialise it or call its set method.	1012	reinitialise it or call its C<ev_TYPE_set> macro.
301
302	You cna check whether an event is active by calling the C<ev_is_active
303	(watcher *)> macro. To see whether an event is outstanding (but the
304	callback for it has not been called yet) you cna use the C<ev_is_pending
305	(watcher *)> macro.
306		1013
307	Each and every callback receives the event loop pointer as first, the	1014	Each and every callback receives the event loop pointer as first, the
308	registered watcher structure as second, and a bitset of received events as	1015	registered watcher structure as second, and a bitset of received events as
309	third argument.	1016	third argument.
310		1017
311	The rceeived events usually include a single bit per event type received	1018	The received events usually include a single bit per event type received
312	(you can receive multiple events at the same time). The possible bit masks	1019	(you can receive multiple events at the same time). The possible bit masks
313	are:	1020	are:
314		1021
315	=over 4	1022	=over 4
316		1023
317	=item EV_READ	1024	=item C<EV_READ>
318		1025
319	=item EV_WRITE	1026	=item C<EV_WRITE>
320		1027
321	The file descriptor in the ev_io watcher has become readable and/or	1028	The file descriptor in the C<ev_io> watcher has become readable and/or
322	writable.	1029	writable.
323		1030
324	=item EV_TIMEOUT	1031	=item C<EV_TIMEOUT>
325		1032
326	The ev_timer watcher has timed out.	1033	The C<ev_timer> watcher has timed out.
327		1034
328	=item EV_PERIODIC	1035	=item C<EV_PERIODIC>
329		1036
330	The ev_periodic watcher has timed out.	1037	The C<ev_periodic> watcher has timed out.
331		1038
332	=item EV_SIGNAL	1039	=item C<EV_SIGNAL>
333		1040
334	The signal specified in the ev_signal watcher has been received by a thread.	1041	The signal specified in the C<ev_signal> watcher has been received by a thread.
335		1042
336	=item EV_CHILD	1043	=item C<EV_CHILD>
337		1044
338	The pid specified in the ev_child watcher has received a status change.	1045	The pid specified in the C<ev_child> watcher has received a status change.
339		1046
		1047	=item C<EV_STAT>
		1048
		1049	The path specified in the C<ev_stat> watcher changed its attributes somehow.
		1050
340	=item EV_IDLE	1051	=item C<EV_IDLE>
341		1052
342	The ev_idle watcher has determined that you have nothing better to do.	1053	The C<ev_idle> watcher has determined that you have nothing better to do.
343		1054
344	=item EV_PREPARE	1055	=item C<EV_PREPARE>
345		1056
346	=item EV_CHECK	1057	=item C<EV_CHECK>
347		1058
348	All ev_prepare watchers are invoked just I<before> C<ev_loop> starts	1059	All C<ev_prepare> watchers are invoked just I<before> C<ev_loop> starts
349	to gather new events, and all ev_check watchers are invoked just after	1060	to gather new events, and all C<ev_check> watchers are invoked just after
350	C<ev_loop> has gathered them, but before it invokes any callbacks for any	1061	C<ev_loop> has gathered them, but before it invokes any callbacks for any
351	received events. Callbacks of both watcher types can start and stop as	1062	received events. Callbacks of both watcher types can start and stop as
352	many watchers as they want, and all of them will be taken into account	1063	many watchers as they want, and all of them will be taken into account
353	(for example, a ev_prepare watcher might start an idle watcher to keep	1064	(for example, a C<ev_prepare> watcher might start an idle watcher to keep
354	C<ev_loop> from blocking).	1065	C<ev_loop> from blocking).
355		1066
		1067	=item C<EV_EMBED>
		1068
		1069	The embedded event loop specified in the C<ev_embed> watcher needs attention.
		1070
		1071	=item C<EV_FORK>
		1072
		1073	The event loop has been resumed in the child process after fork (see
		1074	C<ev_fork>).
		1075
		1076	=item C<EV_ASYNC>
		1077
		1078	The given async watcher has been asynchronously notified (see C<ev_async>).
		1079
		1080	=item C<EV_CUSTOM>
		1081
		1082	Not ever sent (or otherwise used) by libev itself, but can be freely used
		1083	by libev users to signal watchers (e.g. via C<ev_feed_event>).
		1084
356	=item EV_ERROR	1085	=item C<EV_ERROR>
357		1086
358	An unspecified error has occured, the watcher has been stopped. This might	1087	An unspecified error has occurred, the watcher has been stopped. This might
359	happen because the watcher could not be properly started because libev	1088	happen because the watcher could not be properly started because libev
360	ran out of memory, a file descriptor was found to be closed or any other	1089	ran out of memory, a file descriptor was found to be closed or any other
		1090	problem. Libev considers these application bugs.
		1091
361	problem. You best act on it by reporting the problem and somehow coping	1092	You best act on it by reporting the problem and somehow coping with the
362	with the watcher being stopped.	1093	watcher being stopped. Note that well-written programs should not receive
		1094	an error ever, so when your watcher receives it, this usually indicates a
		1095	bug in your program.
363		1096
364	Libev will usually signal a few "dummy" events together with an error,	1097	Libev will usually signal a few "dummy" events together with an error, for
365	for example it might indicate that a fd is readable or writable, and if	1098	example it might indicate that a fd is readable or writable, and if your
366	your callbacks is well-written it can just attempt the operation and cope	1099	callbacks is well-written it can just attempt the operation and cope with
367	with the error from read() or write(). This will not work in multithreaded	1100	the error from read() or write(). This will not work in multi-threaded
368	programs, though, so beware.	1101	programs, though, as the fd could already be closed and reused for another
		1102	thing, so beware.
369		1103
370	=back	1104	=back
371		1105
		1106	=head2 GENERIC WATCHER FUNCTIONS
		1107
		1108	=over 4
		1109
		1110	=item C<ev_init> (ev_TYPE *watcher, callback)
		1111
		1112	This macro initialises the generic portion of a watcher. The contents
		1113	of the watcher object can be arbitrary (so C<malloc> will do). Only
		1114	the generic parts of the watcher are initialised, you I<need> to call
		1115	the type-specific C<ev_TYPE_set> macro afterwards to initialise the
		1116	type-specific parts. For each type there is also a C<ev_TYPE_init> macro
		1117	which rolls both calls into one.
		1118
		1119	You can reinitialise a watcher at any time as long as it has been stopped
		1120	(or never started) and there are no pending events outstanding.
		1121
		1122	The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
		1123	int revents)>.
		1124
		1125	Example: Initialise an C<ev_io> watcher in two steps.
		1126
		1127	ev_io w;
		1128	ev_init (&w, my_cb);
		1129	ev_io_set (&w, STDIN_FILENO, EV_READ);
		1130
		1131	=item C<ev_TYPE_set> (ev_TYPE *, [args])
		1132
		1133	This macro initialises the type-specific parts of a watcher. You need to
		1134	call C<ev_init> at least once before you call this macro, but you can
		1135	call C<ev_TYPE_set> any number of times. You must not, however, call this
		1136	macro on a watcher that is active (it can be pending, however, which is a
		1137	difference to the C<ev_init> macro).
		1138
		1139	Although some watcher types do not have type-specific arguments
		1140	(e.g. C<ev_prepare>) you still need to call its C<set> macro.
		1141
		1142	See C<ev_init>, above, for an example.
		1143
		1144	=item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
		1145
		1146	This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
		1147	calls into a single call. This is the most convenient method to initialise
		1148	a watcher. The same limitations apply, of course.
		1149
		1150	Example: Initialise and set an C<ev_io> watcher in one step.
		1151
		1152	ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
		1153
		1154	=item C<ev_TYPE_start> (loop *, ev_TYPE *watcher)
		1155
		1156	Starts (activates) the given watcher. Only active watchers will receive
		1157	events. If the watcher is already active nothing will happen.
		1158
		1159	Example: Start the C<ev_io> watcher that is being abused as example in this
		1160	whole section.
		1161
		1162	ev_io_start (EV_DEFAULT_UC, &w);
		1163
		1164	=item C<ev_TYPE_stop> (loop *, ev_TYPE *watcher)
		1165
		1166	Stops the given watcher if active, and clears the pending status (whether
		1167	the watcher was active or not).
		1168
		1169	It is possible that stopped watchers are pending - for example,
		1170	non-repeating timers are being stopped when they become pending - but
		1171	calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
		1172	pending. If you want to free or reuse the memory used by the watcher it is
		1173	therefore a good idea to always call its C<ev_TYPE_stop> function.
		1174
		1175	=item bool ev_is_active (ev_TYPE *watcher)
		1176
		1177	Returns a true value iff the watcher is active (i.e. it has been started
		1178	and not yet been stopped). As long as a watcher is active you must not modify
		1179	it.
		1180
		1181	=item bool ev_is_pending (ev_TYPE *watcher)
		1182
		1183	Returns a true value iff the watcher is pending, (i.e. it has outstanding
		1184	events but its callback has not yet been invoked). As long as a watcher
		1185	is pending (but not active) you must not call an init function on it (but
		1186	C<ev_TYPE_set> is safe), you must not change its priority, and you must
		1187	make sure the watcher is available to libev (e.g. you cannot C<free ()>
		1188	it).
		1189
		1190	=item callback ev_cb (ev_TYPE *watcher)
		1191
		1192	Returns the callback currently set on the watcher.
		1193
		1194	=item ev_cb_set (ev_TYPE *watcher, callback)
		1195
		1196	Change the callback. You can change the callback at virtually any time
		1197	(modulo threads).
		1198
		1199	=item ev_set_priority (ev_TYPE *watcher, priority)
		1200
		1201	=item int ev_priority (ev_TYPE *watcher)
		1202
		1203	Set and query the priority of the watcher. The priority is a small
		1204	integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
		1205	(default: C<-2>). Pending watchers with higher priority will be invoked
		1206	before watchers with lower priority, but priority will not keep watchers
		1207	from being executed (except for C<ev_idle> watchers).
		1208
		1209	If you need to suppress invocation when higher priority events are pending
		1210	you need to look at C<ev_idle> watchers, which provide this functionality.
		1211
		1212	You I<must not> change the priority of a watcher as long as it is active or
		1213	pending.
		1214
		1215	Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
		1216	fine, as long as you do not mind that the priority value you query might
		1217	or might not have been clamped to the valid range.
		1218
		1219	The default priority used by watchers when no priority has been set is
		1220	always C<0>, which is supposed to not be too high and not be too low :).
		1221
		1222	See L<WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
		1223	priorities.
		1224
		1225	=item ev_invoke (loop, ev_TYPE *watcher, int revents)
		1226
		1227	Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
		1228	C<loop> nor C<revents> need to be valid as long as the watcher callback
		1229	can deal with that fact, as both are simply passed through to the
		1230	callback.
		1231
		1232	=item int ev_clear_pending (loop, ev_TYPE *watcher)
		1233
		1234	If the watcher is pending, this function clears its pending status and
		1235	returns its C<revents> bitset (as if its callback was invoked). If the
		1236	watcher isn't pending it does nothing and returns C<0>.
		1237
		1238	Sometimes it can be useful to "poll" a watcher instead of waiting for its
		1239	callback to be invoked, which can be accomplished with this function.
		1240
		1241	=back
		1242
		1243
372	=head2 ASSOCIATING CUSTOM DATA WITH A WATCHER	1244	=head2 ASSOCIATING CUSTOM DATA WITH A WATCHER
373		1245
374	Each watcher has, by default, a member C<void *data> that you can change	1246	Each watcher has, by default, a member C<void *data> that you can change
375	and read at any time, libev will completely ignore it. This cna be used	1247	and read at any time: libev will completely ignore it. This can be used
376	to associate arbitrary data with your watcher. If you need more data and	1248	to associate arbitrary data with your watcher. If you need more data and
377	don't want to allocate memory and store a pointer to it in that data	1249	don't want to allocate memory and store a pointer to it in that data
378	member, you can also "subclass" the watcher type and provide your own	1250	member, you can also "subclass" the watcher type and provide your own
379	data:	1251	data:
380		1252
381	struct my_io	1253	struct my_io
382	{	1254	{
383	struct ev_io io;	1255	ev_io io;
384	int otherfd;	1256	int otherfd;
385	void *somedata;	1257	void *somedata;
386	struct whatever *mostinteresting;	1258	struct whatever *mostinteresting;
387	}	1259	};
		1260
		1261	...
		1262	struct my_io w;
		1263	ev_io_init (&w.io, my_cb, fd, EV_READ);
388		1264
389	And since your callback will be called with a pointer to the watcher, you	1265	And since your callback will be called with a pointer to the watcher, you
390	can cast it back to your own type:	1266	can cast it back to your own type:
391		1267
392	static void my_cb (struct ev_loop *loop, struct ev_io *w_, int revents)	1268	static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
393	{	1269	{
394	struct my_io *w = (struct my_io *)w_;	1270	struct my_io *w = (struct my_io *)w_;
395	...	1271	...
396	}	1272	}
397		1273
398	More interesting and less C-conformant ways of catsing your callback type	1274	More interesting and less C-conformant ways of casting your callback type
399	have been omitted....	1275	instead have been omitted.
		1276
		1277	Another common scenario is to use some data structure with multiple
		1278	embedded watchers:
		1279
		1280	struct my_biggy
		1281	{
		1282	int some_data;
		1283	ev_timer t1;
		1284	ev_timer t2;
		1285	}
		1286
		1287	In this case getting the pointer to C<my_biggy> is a bit more
		1288	complicated: Either you store the address of your C<my_biggy> struct
		1289	in the C<data> member of the watcher (for woozies), or you need to use
		1290	some pointer arithmetic using C<offsetof> inside your watchers (for real
		1291	programmers):
		1292
		1293	#include <stddef.h>
		1294
		1295	static void
		1296	t1_cb (EV_P_ ev_timer *w, int revents)
		1297	{
		1298	struct my_biggy big = (struct my_biggy *)
		1299	(((char *)w) - offsetof (struct my_biggy, t1));
		1300	}
		1301
		1302	static void
		1303	t2_cb (EV_P_ ev_timer *w, int revents)
		1304	{
		1305	struct my_biggy big = (struct my_biggy *)
		1306	(((char *)w) - offsetof (struct my_biggy, t2));
		1307	}
		1308
		1309	=head2 WATCHER PRIORITY MODELS
		1310
		1311	Many event loops support I<watcher priorities>, which are usually small
		1312	integers that influence the ordering of event callback invocation
		1313	between watchers in some way, all else being equal.
		1314
		1315	In libev, Watcher priorities can be set using C<ev_set_priority>. See its
		1316	description for the more technical details such as the actual priority
		1317	range.
		1318
		1319	There are two common ways how these these priorities are being interpreted
		1320	by event loops:
		1321
		1322	In the more common lock-out model, higher priorities "lock out" invocation
		1323	of lower priority watchers, which means as long as higher priority
		1324	watchers receive events, lower priority watchers are not being invoked.
		1325
		1326	The less common only-for-ordering model uses priorities solely to order
		1327	callback invocation within a single event loop iteration: Higher priority
		1328	watchers are invoked before lower priority ones, but they all get invoked
		1329	before polling for new events.
		1330
		1331	Libev uses the second (only-for-ordering) model for all its watchers
		1332	except for idle watchers (which use the lock-out model).
		1333
		1334	The rationale behind this is that implementing the lock-out model for
		1335	watchers is not well supported by most kernel interfaces, and most event
		1336	libraries will just poll for the same events again and again as long as
		1337	their callbacks have not been executed, which is very inefficient in the
		1338	common case of one high-priority watcher locking out a mass of lower
		1339	priority ones.
		1340
		1341	Static (ordering) priorities are most useful when you have two or more
		1342	watchers handling the same resource: a typical usage example is having an
		1343	C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
		1344	timeouts. Under load, data might be received while the program handles
		1345	other jobs, but since timers normally get invoked first, the timeout
		1346	handler will be executed before checking for data. In that case, giving
		1347	the timer a lower priority than the I/O watcher ensures that I/O will be
		1348	handled first even under adverse conditions (which is usually, but not
		1349	always, what you want).
		1350
		1351	Since idle watchers use the "lock-out" model, meaning that idle watchers
		1352	will only be executed when no same or higher priority watchers have
		1353	received events, they can be used to implement the "lock-out" model when
		1354	required.
		1355
		1356	For example, to emulate how many other event libraries handle priorities,
		1357	you can associate an C<ev_idle> watcher to each such watcher, and in
		1358	the normal watcher callback, you just start the idle watcher. The real
		1359	processing is done in the idle watcher callback. This causes libev to
		1360	continously poll and process kernel event data for the watcher, but when
		1361	the lock-out case is known to be rare (which in turn is rare :), this is
		1362	workable.
		1363
		1364	Usually, however, the lock-out model implemented that way will perform
		1365	miserably under the type of load it was designed to handle. In that case,
		1366	it might be preferable to stop the real watcher before starting the
		1367	idle watcher, so the kernel will not have to process the event in case
		1368	the actual processing will be delayed for considerable time.
		1369
		1370	Here is an example of an I/O watcher that should run at a strictly lower
		1371	priority than the default, and which should only process data when no
		1372	other events are pending:
		1373
		1374	ev_idle idle; // actual processing watcher
		1375	ev_io io; // actual event watcher
		1376
		1377	static void
		1378	io_cb (EV_P_ ev_io *w, int revents)
		1379	{
		1380	// stop the I/O watcher, we received the event, but
		1381	// are not yet ready to handle it.
		1382	ev_io_stop (EV_A_ w);
		1383
		1384	// start the idle watcher to ahndle the actual event.
		1385	// it will not be executed as long as other watchers
		1386	// with the default priority are receiving events.
		1387	ev_idle_start (EV_A_ &idle);
		1388	}
		1389
		1390	static void
		1391	idle_cb (EV_P_ ev_idle *w, int revents)
		1392	{
		1393	// actual processing
		1394	read (STDIN_FILENO, ...);
		1395
		1396	// have to start the I/O watcher again, as
		1397	// we have handled the event
		1398	ev_io_start (EV_P_ &io);
		1399	}
		1400
		1401	// initialisation
		1402	ev_idle_init (&idle, idle_cb);
		1403	ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
		1404	ev_io_start (EV_DEFAULT_ &io);
		1405
		1406	In the "real" world, it might also be beneficial to start a timer, so that
		1407	low-priority connections can not be locked out forever under load. This
		1408	enables your program to keep a lower latency for important connections
		1409	during short periods of high load, while not completely locking out less
		1410	important ones.
400		1411
401		1412
402	=head1 WATCHER TYPES	1413	=head1 WATCHER TYPES
403		1414
404	This section describes each watcher in detail, but will not repeat	1415	This section describes each watcher in detail, but will not repeat
405	information given in the last section.	1416	information given in the last section. Any initialisation/set macros,
		1417	functions and members specific to the watcher type are explained.
406		1418
		1419	Members are additionally marked with either I<[read-only]>, meaning that,
		1420	while the watcher is active, you can look at the member and expect some
		1421	sensible content, but you must not modify it (you can modify it while the
		1422	watcher is stopped to your hearts content), or I<[read-write]>, which
		1423	means you can expect it to have some sensible content while the watcher
		1424	is active, but you can also modify it. Modifying it may not do something
		1425	sensible or take immediate effect (or do anything at all), but libev will
		1426	not crash or malfunction in any way.
		1427
		1428
407	=head2 struct ev_io - is my file descriptor readable or writable	1429	=head2 C<ev_io> - is this file descriptor readable or writable?
408		1430
409	I/O watchers check whether a file descriptor is readable or writable	1431	I/O watchers check whether a file descriptor is readable or writable
410	in each iteration of the event loop (This behaviour is called	1432	in each iteration of the event loop, or, more precisely, when reading
411	level-triggering because you keep receiving events as long as the	1433	would not block the process and writing would at least be able to write
412	condition persists. Remember you cna stop the watcher if you don't want to	1434	some data. This behaviour is called level-triggering because you keep
413	act on the event and neither want to receive future events).	1435	receiving events as long as the condition persists. Remember you can stop
		1436	the watcher if you don't want to act on the event and neither want to
		1437	receive future events.
414		1438
415	In general you can register as many read and/or write event watchers oer	1439	In general you can register as many read and/or write event watchers per
416	fd as you want (as long as you don't confuse yourself). Setting all file	1440	fd as you want (as long as you don't confuse yourself). Setting all file
417	descriptors to non-blocking mode is also usually a good idea (but not	1441	descriptors to non-blocking mode is also usually a good idea (but not
418	required if you know what you are doing).	1442	required if you know what you are doing).
419		1443
420	You have to be careful with dup'ed file descriptors, though. Some backends	1444	If you cannot use non-blocking mode, then force the use of a
421	(the linux epoll backend is a notable example) cannot handle dup'ed file	1445	known-to-be-good backend (at the time of this writing, this includes only
422	descriptors correctly if you register interest in two or more fds pointing	1446	C<EVBACKEND_SELECT> and C<EVBACKEND_POLL>). The same applies to file
423	to the same file/socket etc. description.	1447	descriptors for which non-blocking operation makes no sense (such as
		1448	files) - libev doesn't guarentee any specific behaviour in that case.
424		1449
425	If you must do this, then force the use of a known-to-be-good backend	1450	Another thing you have to watch out for is that it is quite easy to
426	(at the time of this writing, this includes only EVMETHOD_SELECT and	1451	receive "spurious" readiness notifications, that is your callback might
427	EVMETHOD_POLL).	1452	be called with C<EV_READ> but a subsequent C<read>(2) will actually block
		1453	because there is no data. Not only are some backends known to create a
		1454	lot of those (for example Solaris ports), it is very easy to get into
		1455	this situation even with a relatively standard program structure. Thus
		1456	it is best to always use non-blocking I/O: An extra C<read>(2) returning
		1457	C<EAGAIN> is far preferable to a program hanging until some data arrives.
		1458
		1459	If you cannot run the fd in non-blocking mode (for example you should
		1460	not play around with an Xlib connection), then you have to separately
		1461	re-test whether a file descriptor is really ready with a known-to-be good
		1462	interface such as poll (fortunately in our Xlib example, Xlib already
		1463	does this on its own, so its quite safe to use). Some people additionally
		1464	use C<SIGALRM> and an interval timer, just to be sure you won't block
		1465	indefinitely.
		1466
		1467	But really, best use non-blocking mode.
		1468
		1469	=head3 The special problem of disappearing file descriptors
		1470
		1471	Some backends (e.g. kqueue, epoll) need to be told about closing a file
		1472	descriptor (either due to calling C<close> explicitly or any other means,
		1473	such as C<dup2>). The reason is that you register interest in some file
		1474	descriptor, but when it goes away, the operating system will silently drop
		1475	this interest. If another file descriptor with the same number then is
		1476	registered with libev, there is no efficient way to see that this is, in
		1477	fact, a different file descriptor.
		1478
		1479	To avoid having to explicitly tell libev about such cases, libev follows
		1480	the following policy: Each time C<ev_io_set> is being called, libev
		1481	will assume that this is potentially a new file descriptor, otherwise
		1482	it is assumed that the file descriptor stays the same. That means that
		1483	you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
		1484	descriptor even if the file descriptor number itself did not change.
		1485
		1486	This is how one would do it normally anyway, the important point is that
		1487	the libev application should not optimise around libev but should leave
		1488	optimisations to libev.
		1489
		1490	=head3 The special problem of dup'ed file descriptors
		1491
		1492	Some backends (e.g. epoll), cannot register events for file descriptors,
		1493	but only events for the underlying file descriptions. That means when you
		1494	have C<dup ()>'ed file descriptors or weirder constellations, and register
		1495	events for them, only one file descriptor might actually receive events.
		1496
		1497	There is no workaround possible except not registering events
		1498	for potentially C<dup ()>'ed file descriptors, or to resort to
		1499	C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
		1500
		1501	=head3 The special problem of fork
		1502
		1503	Some backends (epoll, kqueue) do not support C<fork ()> at all or exhibit
		1504	useless behaviour. Libev fully supports fork, but needs to be told about
		1505	it in the child.
		1506
		1507	To support fork in your programs, you either have to call
		1508	C<ev_default_fork ()> or C<ev_loop_fork ()> after a fork in the child,
		1509	enable C<EVFLAG_FORKCHECK>, or resort to C<EVBACKEND_SELECT> or
		1510	C<EVBACKEND_POLL>.
		1511
		1512	=head3 The special problem of SIGPIPE
		1513
		1514	While not really specific to libev, it is easy to forget about C<SIGPIPE>:
		1515	when writing to a pipe whose other end has been closed, your program gets
		1516	sent a SIGPIPE, which, by default, aborts your program. For most programs
		1517	this is sensible behaviour, for daemons, this is usually undesirable.
		1518
		1519	So when you encounter spurious, unexplained daemon exits, make sure you
		1520	ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
		1521	somewhere, as that would have given you a big clue).
		1522
		1523
		1524	=head3 Watcher-Specific Functions
428		1525
429	=over 4	1526	=over 4
430		1527
431	=item ev_io_init (ev_io *, callback, int fd, int events)	1528	=item ev_io_init (ev_io *, callback, int fd, int events)
432		1529
433	=item ev_io_set (ev_io *, int fd, int events)	1530	=item ev_io_set (ev_io *, int fd, int events)
434		1531
435	Configures an ev_io watcher. The fd is the file descriptor to rceeive	1532	Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
436	events for and events is either C<EV_READ>, C<EV_WRITE> or C<EV_READ |	1533	receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
437	EV_WRITE> to receive the given events.	1534	C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
		1535
		1536	=item int fd [read-only]
		1537
		1538	The file descriptor being watched.
		1539
		1540	=item int events [read-only]
		1541
		1542	The events being watched.
438		1543
439	=back	1544	=back
440		1545
		1546	=head3 Examples
		1547
		1548	Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
		1549	readable, but only once. Since it is likely line-buffered, you could
		1550	attempt to read a whole line in the callback.
		1551
		1552	static void
		1553	stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
		1554	{
		1555	ev_io_stop (loop, w);
		1556	.. read from stdin here (or from w->fd) and handle any I/O errors
		1557	}
		1558
		1559	...
		1560	struct ev_loop *loop = ev_default_init (0);
		1561	ev_io stdin_readable;
		1562	ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
		1563	ev_io_start (loop, &stdin_readable);
		1564	ev_loop (loop, 0);
		1565
		1566
441	=head2 struct ev_timer - relative and optionally recurring timeouts	1567	=head2 C<ev_timer> - relative and optionally repeating timeouts
442		1568
443	Timer watchers are simple relative timers that generate an event after a	1569	Timer watchers are simple relative timers that generate an event after a
444	given time, and optionally repeating in regular intervals after that.	1570	given time, and optionally repeating in regular intervals after that.
445		1571
446	The timers are based on real time, that is, if you register an event that	1572	The timers are based on real time, that is, if you register an event that
447	times out after an hour and youreset your system clock to last years	1573	times out after an hour and you reset your system clock to January last
448	time, it will still time out after (roughly) and hour. "Roughly" because	1574	year, it will still time out after (roughly) one hour. "Roughly" because
449	detecting time jumps is hard, and soem inaccuracies are unavoidable (the	1575	detecting time jumps is hard, and some inaccuracies are unavoidable (the
450	monotonic clock option helps a lot here).	1576	monotonic clock option helps a lot here).
451		1577
		1578	The callback is guaranteed to be invoked only I<after> its timeout has
		1579	passed (not I<at>, so on systems with very low-resolution clocks this
		1580	might introduce a small delay). If multiple timers become ready during the
		1581	same loop iteration then the ones with earlier time-out values are invoked
		1582	before ones of the same priority with later time-out values (but this is
		1583	no longer true when a callback calls C<ev_loop> recursively).
		1584
		1585	=head3 Be smart about timeouts
		1586
		1587	Many real-world problems involve some kind of timeout, usually for error
		1588	recovery. A typical example is an HTTP request - if the other side hangs,
		1589	you want to raise some error after a while.
		1590
		1591	What follows are some ways to handle this problem, from obvious and
		1592	inefficient to smart and efficient.
		1593
		1594	In the following, a 60 second activity timeout is assumed - a timeout that
		1595	gets reset to 60 seconds each time there is activity (e.g. each time some
		1596	data or other life sign was received).
		1597
452	=over 4	1598	=over 4
453		1599
		1600	=item 1. Use a timer and stop, reinitialise and start it on activity.
		1601
		1602	This is the most obvious, but not the most simple way: In the beginning,
		1603	start the watcher:
		1604
		1605	ev_timer_init (timer, callback, 60., 0.);
		1606	ev_timer_start (loop, timer);
		1607
		1608	Then, each time there is some activity, C<ev_timer_stop> it, initialise it
		1609	and start it again:
		1610
		1611	ev_timer_stop (loop, timer);
		1612	ev_timer_set (timer, 60., 0.);
		1613	ev_timer_start (loop, timer);
		1614
		1615	This is relatively simple to implement, but means that each time there is
		1616	some activity, libev will first have to remove the timer from its internal
		1617	data structure and then add it again. Libev tries to be fast, but it's
		1618	still not a constant-time operation.
		1619
		1620	=item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
		1621
		1622	This is the easiest way, and involves using C<ev_timer_again> instead of
		1623	C<ev_timer_start>.
		1624
		1625	To implement this, configure an C<ev_timer> with a C<repeat> value
		1626	of C<60> and then call C<ev_timer_again> at start and each time you
		1627	successfully read or write some data. If you go into an idle state where
		1628	you do not expect data to travel on the socket, you can C<ev_timer_stop>
		1629	the timer, and C<ev_timer_again> will automatically restart it if need be.
		1630
		1631	That means you can ignore both the C<ev_timer_start> function and the
		1632	C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
		1633	member and C<ev_timer_again>.
		1634
		1635	At start:
		1636
		1637	ev_init (timer, callback);
		1638	timer->repeat = 60.;
		1639	ev_timer_again (loop, timer);
		1640
		1641	Each time there is some activity:
		1642
		1643	ev_timer_again (loop, timer);
		1644
		1645	It is even possible to change the time-out on the fly, regardless of
		1646	whether the watcher is active or not:
		1647
		1648	timer->repeat = 30.;
		1649	ev_timer_again (loop, timer);
		1650
		1651	This is slightly more efficient then stopping/starting the timer each time
		1652	you want to modify its timeout value, as libev does not have to completely
		1653	remove and re-insert the timer from/into its internal data structure.
		1654
		1655	It is, however, even simpler than the "obvious" way to do it.
		1656
		1657	=item 3. Let the timer time out, but then re-arm it as required.
		1658
		1659	This method is more tricky, but usually most efficient: Most timeouts are
		1660	relatively long compared to the intervals between other activity - in
		1661	our example, within 60 seconds, there are usually many I/O events with
		1662	associated activity resets.
		1663
		1664	In this case, it would be more efficient to leave the C<ev_timer> alone,
		1665	but remember the time of last activity, and check for a real timeout only
		1666	within the callback:
		1667
		1668	ev_tstamp last_activity; // time of last activity
		1669
		1670	static void
		1671	callback (EV_P_ ev_timer *w, int revents)
		1672	{
		1673	ev_tstamp now = ev_now (EV_A);
		1674	ev_tstamp timeout = last_activity + 60.;
		1675
		1676	// if last_activity + 60. is older than now, we did time out
		1677	if (timeout < now)
		1678	{
		1679	// timeout occured, take action
		1680	}
		1681	else
		1682	{
		1683	// callback was invoked, but there was some activity, re-arm
		1684	// the watcher to fire in last_activity + 60, which is
		1685	// guaranteed to be in the future, so "again" is positive:
		1686	w->repeat = timeout - now;
		1687	ev_timer_again (EV_A_ w);
		1688	}
		1689	}
		1690
		1691	To summarise the callback: first calculate the real timeout (defined
		1692	as "60 seconds after the last activity"), then check if that time has
		1693	been reached, which means something I<did>, in fact, time out. Otherwise
		1694	the callback was invoked too early (C<timeout> is in the future), so
		1695	re-schedule the timer to fire at that future time, to see if maybe we have
		1696	a timeout then.
		1697
		1698	Note how C<ev_timer_again> is used, taking advantage of the
		1699	C<ev_timer_again> optimisation when the timer is already running.
		1700
		1701	This scheme causes more callback invocations (about one every 60 seconds
		1702	minus half the average time between activity), but virtually no calls to
		1703	libev to change the timeout.
		1704
		1705	To start the timer, simply initialise the watcher and set C<last_activity>
		1706	to the current time (meaning we just have some activity :), then call the
		1707	callback, which will "do the right thing" and start the timer:
		1708
		1709	ev_init (timer, callback);
		1710	last_activity = ev_now (loop);
		1711	callback (loop, timer, EV_TIMEOUT);
		1712
		1713	And when there is some activity, simply store the current time in
		1714	C<last_activity>, no libev calls at all:
		1715
		1716	last_actiivty = ev_now (loop);
		1717
		1718	This technique is slightly more complex, but in most cases where the
		1719	time-out is unlikely to be triggered, much more efficient.
		1720
		1721	Changing the timeout is trivial as well (if it isn't hard-coded in the
		1722	callback :) - just change the timeout and invoke the callback, which will
		1723	fix things for you.
		1724
		1725	=item 4. Wee, just use a double-linked list for your timeouts.
		1726
		1727	If there is not one request, but many thousands (millions...), all
		1728	employing some kind of timeout with the same timeout value, then one can
		1729	do even better:
		1730
		1731	When starting the timeout, calculate the timeout value and put the timeout
		1732	at the I<end> of the list.
		1733
		1734	Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
		1735	the list is expected to fire (for example, using the technique #3).
		1736
		1737	When there is some activity, remove the timer from the list, recalculate
		1738	the timeout, append it to the end of the list again, and make sure to
		1739	update the C<ev_timer> if it was taken from the beginning of the list.
		1740
		1741	This way, one can manage an unlimited number of timeouts in O(1) time for
		1742	starting, stopping and updating the timers, at the expense of a major
		1743	complication, and having to use a constant timeout. The constant timeout
		1744	ensures that the list stays sorted.
		1745
		1746	=back
		1747
		1748	So which method the best?
		1749
		1750	Method #2 is a simple no-brain-required solution that is adequate in most
		1751	situations. Method #3 requires a bit more thinking, but handles many cases
		1752	better, and isn't very complicated either. In most case, choosing either
		1753	one is fine, with #3 being better in typical situations.
		1754
		1755	Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
		1756	rather complicated, but extremely efficient, something that really pays
		1757	off after the first million or so of active timers, i.e. it's usually
		1758	overkill :)
		1759
		1760	=head3 The special problem of time updates
		1761
		1762	Establishing the current time is a costly operation (it usually takes at
		1763	least two system calls): EV therefore updates its idea of the current
		1764	time only before and after C<ev_loop> collects new events, which causes a
		1765	growing difference between C<ev_now ()> and C<ev_time ()> when handling
		1766	lots of events in one iteration.
		1767
		1768	The relative timeouts are calculated relative to the C<ev_now ()>
		1769	time. This is usually the right thing as this timestamp refers to the time
		1770	of the event triggering whatever timeout you are modifying/starting. If
		1771	you suspect event processing to be delayed and you I<need> to base the
		1772	timeout on the current time, use something like this to adjust for this:
		1773
		1774	ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);
		1775
		1776	If the event loop is suspended for a long time, you can also force an
		1777	update of the time returned by C<ev_now ()> by calling C<ev_now_update
		1778	()>.
		1779
		1780	=head3 The special problems of suspended animation
		1781
		1782	When you leave the server world it is quite customary to hit machines that
		1783	can suspend/hibernate - what happens to the clocks during such a suspend?
		1784
		1785	Some quick tests made with a Linux 2.6.28 indicate that a suspend freezes
		1786	all processes, while the clocks (C<times>, C<CLOCK_MONOTONIC>) continue
		1787	to run until the system is suspended, but they will not advance while the
		1788	system is suspended. That means, on resume, it will be as if the program
		1789	was frozen for a few seconds, but the suspend time will not be counted
		1790	towards C<ev_timer> when a monotonic clock source is used. The real time
		1791	clock advanced as expected, but if it is used as sole clocksource, then a
		1792	long suspend would be detected as a time jump by libev, and timers would
		1793	be adjusted accordingly.
		1794
		1795	I would not be surprised to see different behaviour in different between
		1796	operating systems, OS versions or even different hardware.
		1797
		1798	The other form of suspend (job control, or sending a SIGSTOP) will see a
		1799	time jump in the monotonic clocks and the realtime clock. If the program
		1800	is suspended for a very long time, and monotonic clock sources are in use,
		1801	then you can expect C<ev_timer>s to expire as the full suspension time
		1802	will be counted towards the timers. When no monotonic clock source is in
		1803	use, then libev will again assume a timejump and adjust accordingly.
		1804
		1805	It might be beneficial for this latter case to call C<ev_suspend>
		1806	and C<ev_resume> in code that handles C<SIGTSTP>, to at least get
		1807	deterministic behaviour in this case (you can do nothing against
		1808	C<SIGSTOP>).
		1809
		1810	=head3 Watcher-Specific Functions and Data Members
		1811
		1812	=over 4
		1813
454	=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)	1814	=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
455		1815
456	=item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)	1816	=item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
457		1817
458	Configure the timer to trigger after C<after> seconds. If C<repeat> is	1818	Configure the timer to trigger after C<after> seconds. If C<repeat>
459	C<0.>, then it will automatically be stopped. If it is positive, then the	1819	is C<0.>, then it will automatically be stopped once the timeout is
460	timer will automatically be configured to trigger again C<repeat> seconds	1820	reached. If it is positive, then the timer will automatically be
461	later, again, and again, until stopped manually.	1821	configured to trigger again C<repeat> seconds later, again, and again,
		1822	until stopped manually.
462		1823
463	The timer itself will do a best-effort at avoiding drift, that is, if you	1824	The timer itself will do a best-effort at avoiding drift, that is, if
464	configure a timer to trigger every 10 seconds, then it will trigger at	1825	you configure a timer to trigger every 10 seconds, then it will normally
465	exactly 10 second intervals. If, however, your program cannot keep up with	1826	trigger at exactly 10 second intervals. If, however, your program cannot
466	the timer (ecause it takes longer than those 10 seconds to do stuff) the	1827	keep up with the timer (because it takes longer than those 10 seconds to
467	timer will not fire more than once per event loop iteration.	1828	do stuff) the timer will not fire more than once per event loop iteration.
468		1829
469	=item ev_timer_again (loop)	1830	=item ev_timer_again (loop, ev_timer *)
470		1831
471	This will act as if the timer timed out and restart it again if it is	1832	This will act as if the timer timed out and restart it again if it is
472	repeating. The exact semantics are:	1833	repeating. The exact semantics are:
473		1834
		1835	If the timer is pending, its pending status is cleared.
		1836
474	If the timer is started but nonrepeating, stop it.	1837	If the timer is started but non-repeating, stop it (as if it timed out).
475		1838
476	If the timer is repeating, either start it if necessary (with the repeat	1839	If the timer is repeating, either start it if necessary (with the
477	value), or reset the running timer to the repeat value.	1840	C<repeat> value), or reset the running timer to the C<repeat> value.
478		1841
479	This sounds a bit complicated, but here is a useful and typical	1842	This sounds a bit complicated, see L<Be smart about timeouts>, above, for a
480	example: Imagine you have a tcp connection and you want a so-called idle	1843	usage example.
481	timeout, that is, you want to be called when there have been, say, 60	1844
482	seconds of inactivity on the socket. The easiest way to do this is to	1845	=item ev_timer_remaining (loop, ev_timer *)
483	configure an ev_timer with after=repeat=60 and calling ev_timer_again each	1846
484	time you successfully read or write some data. If you go into an idle	1847	Returns the remaining time until a timer fires. If the timer is active,
485	state where you do not expect data to travel on the socket, you can stop	1848	then this time is relative to the current event loop time, otherwise it's
486	the timer, and again will automatically restart it if need be.	1849	the timeout value currently configured.
		1850
		1851	That is, after an C<ev_timer_set (w, 5, 7)>, C<ev_timer_remaining> returns
		1852	C<5>. When the timer is started and one second passes, C<ev_timer_remain>
		1853	will return C<4>. When the timer expires and is restarted, it will return
		1854	roughly C<7> (likely slightly less as callback invocation takes some time,
		1855	too), and so on.
		1856
		1857	=item ev_tstamp repeat [read-write]
		1858
		1859	The current C<repeat> value. Will be used each time the watcher times out
		1860	or C<ev_timer_again> is called, and determines the next timeout (if any),
		1861	which is also when any modifications are taken into account.
487		1862
488	=back	1863	=back
489		1864
		1865	=head3 Examples
		1866
		1867	Example: Create a timer that fires after 60 seconds.
		1868
		1869	static void
		1870	one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
		1871	{
		1872	.. one minute over, w is actually stopped right here
		1873	}
		1874
		1875	ev_timer mytimer;
		1876	ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
		1877	ev_timer_start (loop, &mytimer);
		1878
		1879	Example: Create a timeout timer that times out after 10 seconds of
		1880	inactivity.
		1881
		1882	static void
		1883	timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
		1884	{
		1885	.. ten seconds without any activity
		1886	}
		1887
		1888	ev_timer mytimer;
		1889	ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
		1890	ev_timer_again (&mytimer); /* start timer */
		1891	ev_loop (loop, 0);
		1892
		1893	// and in some piece of code that gets executed on any "activity":
		1894	// reset the timeout to start ticking again at 10 seconds
		1895	ev_timer_again (&mytimer);
		1896
		1897
490	=head2 ev_periodic - to cron or not to cron it	1898	=head2 C<ev_periodic> - to cron or not to cron?
491		1899
492	Periodic watchers are also timers of a kind, but they are very versatile	1900	Periodic watchers are also timers of a kind, but they are very versatile
493	(and unfortunately a bit complex).	1901	(and unfortunately a bit complex).
494		1902
495	Unlike ev_timer's, they are not based on real time (or relative time)	1903	Unlike C<ev_timer>, periodic watchers are not based on real time (or
496	but on wallclock time (absolute time). You can tell a periodic watcher	1904	relative time, the physical time that passes) but on wall clock time
497	to trigger "at" some specific point in time. For example, if you tell a	1905	(absolute time, the thing you can read on your calender or clock). The
498	periodic watcher to trigger in 10 seconds (by specifiying e.g. c<ev_now ()	1906	difference is that wall clock time can run faster or slower than real
499	+ 10.>) and then reset your system clock to the last year, then it will	1907	time, and time jumps are not uncommon (e.g. when you adjust your
500	take a year to trigger the event (unlike an ev_timer, which would trigger	1908	wrist-watch).
501	roughly 10 seconds later and of course not if you reset your system time
502	again).
503		1909
504	They can also be used to implement vastly more complex timers, such as	1910	You can tell a periodic watcher to trigger after some specific point
		1911	in time: for example, if you tell a periodic watcher to trigger "in 10
		1912	seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
		1913	not a delay) and then reset your system clock to January of the previous
		1914	year, then it will take a year or more to trigger the event (unlike an
		1915	C<ev_timer>, which would still trigger roughly 10 seconds after starting
		1916	it, as it uses a relative timeout).
		1917
		1918	C<ev_periodic> watchers can also be used to implement vastly more complex
505	triggering an event on eahc midnight, local time.	1919	timers, such as triggering an event on each "midnight, local time", or
		1920	other complicated rules. This cannot be done with C<ev_timer> watchers, as
		1921	those cannot react to time jumps.
		1922
		1923	As with timers, the callback is guaranteed to be invoked only when the
		1924	point in time where it is supposed to trigger has passed. If multiple
		1925	timers become ready during the same loop iteration then the ones with
		1926	earlier time-out values are invoked before ones with later time-out values
		1927	(but this is no longer true when a callback calls C<ev_loop> recursively).
		1928
		1929	=head3 Watcher-Specific Functions and Data Members
506		1930
507	=over 4	1931	=over 4
508		1932
509	=item ev_periodic_init (ev_periodic *, callback, ev_tstamp at, ev_tstamp interval, reschedule_cb)	1933	=item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
510		1934
511	=item ev_periodic_set (ev_periodic *, ev_tstamp after, ev_tstamp repeat, reschedule_cb)	1935	=item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
512		1936
513	Lots of arguments, lets sort it out... There are basically three modes of	1937	Lots of arguments, let's sort it out... There are basically three modes of
514	operation, and we will explain them from simplest to complex:	1938	operation, and we will explain them from simplest to most complex:
515
516		1939
517	=over 4	1940	=over 4
518		1941
519	=item * absolute timer (interval = reschedule_cb = 0)	1942	=item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
520		1943
521	In this configuration the watcher triggers an event at the wallclock time	1944	In this configuration the watcher triggers an event after the wall clock
522	C<at> and doesn't repeat. It will not adjust when a time jump occurs,	1945	time C<offset> has passed. It will not repeat and will not adjust when a
523	that is, if it is to be run at January 1st 2011 then it will run when the	1946	time jump occurs, that is, if it is to be run at January 1st 2011 then it
524	system time reaches or surpasses this time.	1947	will be stopped and invoked when the system clock reaches or surpasses
		1948	this point in time.
525		1949
526	=item * non-repeating interval timer (interval > 0, reschedule_cb = 0)	1950	=item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
527		1951
528	In this mode the watcher will always be scheduled to time out at the next	1952	In this mode the watcher will always be scheduled to time out at the next
529	C<at + N * interval> time (for some integer N) and then repeat, regardless	1953	C<offset + N * interval> time (for some integer N, which can also be
530	of any time jumps.	1954	negative) and then repeat, regardless of any time jumps. The C<offset>
		1955	argument is merely an offset into the C<interval> periods.
531		1956
532	This can be used to create timers that do not drift with respect to system	1957	This can be used to create timers that do not drift with respect to the
533	time:	1958	system clock, for example, here is an C<ev_periodic> that triggers each
		1959	hour, on the hour (with respect to UTC):
534		1960
535	ev_periodic_set (&periodic, 0., 3600., 0);	1961	ev_periodic_set (&periodic, 0., 3600., 0);
536		1962
537	This doesn't mean there will always be 3600 seconds in between triggers,	1963	This doesn't mean there will always be 3600 seconds in between triggers,
538	but only that the the callback will be called when the system time shows a	1964	but only that the callback will be called when the system time shows a
539	full hour (UTC), or more correct, when the system time is evenly divisible	1965	full hour (UTC), or more correctly, when the system time is evenly divisible
540	by 3600.	1966	by 3600.
541		1967
542	Another way to think about it (for the mathematically inclined) is that	1968	Another way to think about it (for the mathematically inclined) is that
543	ev_periodic will try to run the callback in this mode at the next possible	1969	C<ev_periodic> will try to run the callback in this mode at the next possible
544	time where C<time = at (mod interval)>, regardless of any time jumps.	1970	time where C<time = offset (mod interval)>, regardless of any time jumps.
545		1971
		1972	For numerical stability it is preferable that the C<offset> value is near
		1973	C<ev_now ()> (the current time), but there is no range requirement for
		1974	this value, and in fact is often specified as zero.
		1975
		1976	Note also that there is an upper limit to how often a timer can fire (CPU
		1977	speed for example), so if C<interval> is very small then timing stability
		1978	will of course deteriorate. Libev itself tries to be exact to be about one
		1979	millisecond (if the OS supports it and the machine is fast enough).
		1980
546	=item * manual reschedule mode (reschedule_cb = callback)	1981	=item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
547		1982
548	In this mode the values for C<interval> and C<at> are both being	1983	In this mode the values for C<interval> and C<offset> are both being
549	ignored. Instead, each time the periodic watcher gets scheduled, the	1984	ignored. Instead, each time the periodic watcher gets scheduled, the
550	reschedule callback will be called with the watcher as first, and the	1985	reschedule callback will be called with the watcher as first, and the
551	current time as second argument.	1986	current time as second argument.
552		1987
553	NOTE: I<This callback MUST NOT stop or destroy the periodic or any other	1988	NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, ever,
554	periodic watcher, ever, or make any event loop modificstions>. If you need	1989	or make ANY other event loop modifications whatsoever, unless explicitly
555	to stop it, return 1e30 (or so, fudge fudge) and stop it afterwards.	1990	allowed by documentation here>.
556		1991
		1992	If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
		1993	it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
		1994	only event loop modification you are allowed to do).
		1995
557	Its prototype is c<ev_tstamp (*reschedule_cb)(struct ev_periodic *w,	1996	The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
558	ev_tstamp now)>, e.g.:	1997	*w, ev_tstamp now)>, e.g.:
559		1998
		1999	static ev_tstamp
560	static ev_tstamp my_rescheduler (struct ev_periodic *w, ev_tstamp now)	2000	my_rescheduler (ev_periodic *w, ev_tstamp now)
561	{	2001	{
562	return now + 60.;	2002	return now + 60.;
563	}	2003	}
564		2004
565	It must return the next time to trigger, based on the passed time value	2005	It must return the next time to trigger, based on the passed time value
566	(that is, the lowest time value larger than to the second argument). It	2006	(that is, the lowest time value larger than to the second argument). It
567	will usually be called just before the callback will be triggered, but	2007	will usually be called just before the callback will be triggered, but
568	might be called at other times, too.	2008	might be called at other times, too.
569		2009
		2010	NOTE: I<< This callback must always return a time that is higher than or
		2011	equal to the passed C<now> value >>.
		2012
570	This can be used to create very complex timers, such as a timer that	2013	This can be used to create very complex timers, such as a timer that
571	triggers on each midnight, local time. To do this, you would calculate the	2014	triggers on "next midnight, local time". To do this, you would calculate the
572	next midnight after C<now> and return the timestamp value for this. How you do this	2015	next midnight after C<now> and return the timestamp value for this. How
573	is, again, up to you (but it is not trivial).	2016	you do this is, again, up to you (but it is not trivial, which is the main
		2017	reason I omitted it as an example).
574		2018
575	=back	2019	=back
576		2020
577	=item ev_periodic_again (loop, ev_periodic *)	2021	=item ev_periodic_again (loop, ev_periodic *)
578		2022
579	Simply stops and restarts the periodic watcher again. This is only useful	2023	Simply stops and restarts the periodic watcher again. This is only useful
580	when you changed some parameters or the reschedule callback would return	2024	when you changed some parameters or the reschedule callback would return
581	a different time than the last time it was called (e.g. in a crond like	2025	a different time than the last time it was called (e.g. in a crond like
582	program when the crontabs have changed).	2026	program when the crontabs have changed).
583		2027
		2028	=item ev_tstamp ev_periodic_at (ev_periodic *)
		2029
		2030	When active, returns the absolute time that the watcher is supposed
		2031	to trigger next. This is not the same as the C<offset> argument to
		2032	C<ev_periodic_set>, but indeed works even in interval and manual
		2033	rescheduling modes.
		2034
		2035	=item ev_tstamp offset [read-write]
		2036
		2037	When repeating, this contains the offset value, otherwise this is the
		2038	absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
		2039	although libev might modify this value for better numerical stability).
		2040
		2041	Can be modified any time, but changes only take effect when the periodic
		2042	timer fires or C<ev_periodic_again> is being called.
		2043
		2044	=item ev_tstamp interval [read-write]
		2045
		2046	The current interval value. Can be modified any time, but changes only
		2047	take effect when the periodic timer fires or C<ev_periodic_again> is being
		2048	called.
		2049
		2050	=item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
		2051
		2052	The current reschedule callback, or C<0>, if this functionality is
		2053	switched off. Can be changed any time, but changes only take effect when
		2054	the periodic timer fires or C<ev_periodic_again> is being called.
		2055
584	=back	2056	=back
585		2057
		2058	=head3 Examples
		2059
		2060	Example: Call a callback every hour, or, more precisely, whenever the
		2061	system time is divisible by 3600. The callback invocation times have
		2062	potentially a lot of jitter, but good long-term stability.
		2063
		2064	static void
		2065	clock_cb (struct ev_loop *loop, ev_io *w, int revents)
		2066	{
		2067	... its now a full hour (UTC, or TAI or whatever your clock follows)
		2068	}
		2069
		2070	ev_periodic hourly_tick;
		2071	ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
		2072	ev_periodic_start (loop, &hourly_tick);
		2073
		2074	Example: The same as above, but use a reschedule callback to do it:
		2075
		2076	#include <math.h>
		2077
		2078	static ev_tstamp
		2079	my_scheduler_cb (ev_periodic *w, ev_tstamp now)
		2080	{
		2081	return now + (3600. - fmod (now, 3600.));
		2082	}
		2083
		2084	ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
		2085
		2086	Example: Call a callback every hour, starting now:
		2087
		2088	ev_periodic hourly_tick;
		2089	ev_periodic_init (&hourly_tick, clock_cb,
		2090	fmod (ev_now (loop), 3600.), 3600., 0);
		2091	ev_periodic_start (loop, &hourly_tick);
		2092
		2093
586	=head2 ev_signal - signal me when a signal gets signalled	2094	=head2 C<ev_signal> - signal me when a signal gets signalled!
587		2095
588	Signal watchers will trigger an event when the process receives a specific	2096	Signal watchers will trigger an event when the process receives a specific
589	signal one or more times. Even though signals are very asynchronous, libev	2097	signal one or more times. Even though signals are very asynchronous, libev
590	will try its best to deliver signals synchronously, i.e. as part of the	2098	will try it's best to deliver signals synchronously, i.e. as part of the
591	normal event processing, like any other event.	2099	normal event processing, like any other event.
592		2100
		2101	If you want signals to be delivered truly asynchronously, just use
		2102	C<sigaction> as you would do without libev and forget about sharing
		2103	the signal. You can even use C<ev_async> from a signal handler to
		2104	synchronously wake up an event loop.
		2105
593	You cna configure as many watchers as you like per signal. Only when the	2106	You can configure as many watchers as you like for the same signal, but
		2107	only within the same loop, i.e. you can watch for C<SIGINT> in your
		2108	default loop and for C<SIGIO> in another loop, but you cannot watch for
		2109	C<SIGINT> in both the default loop and another loop at the same time. At
		2110	the moment, C<SIGCHLD> is permanently tied to the default loop.
		2111
594	first watcher gets started will libev actually register a signal watcher	2112	When the first watcher gets started will libev actually register something
595	with the kernel (thus it coexists with your own signal handlers as long	2113	with the kernel (thus it coexists with your own signal handlers as long as
596	as you don't register any with libev). Similarly, when the last signal	2114	you don't register any with libev for the same signal).
597	watcher for a signal is stopped libev will reset the signal handler to	2115
598	SIG_DFL (regardless of what it was set to before).	2116	If possible and supported, libev will install its handlers with
		2117	C<SA_RESTART> (or equivalent) behaviour enabled, so system calls should
		2118	not be unduly interrupted. If you have a problem with system calls getting
		2119	interrupted by signals you can block all signals in an C<ev_check> watcher
		2120	and unblock them in an C<ev_prepare> watcher.
		2121
		2122	=head3 The special problem of inheritance over execve
		2123
		2124	Both the signal mask (C<sigprocmask>) and the signal disposition
		2125	(C<sigaction>) are unspecified after starting a signal watcher (and after
		2126	stopping it again), that is, libev might or might not block the signal,
		2127	and might or might not set or restore the installed signal handler.
		2128
		2129	While this does not matter for the signal disposition (libev never
		2130	sets signals to C<SIG_IGN>, so handlers will be reset to C<SIG_DFL> on
		2131	C<execve>), this matters for the signal mask: many programs do not expect
		2132	certain signals to be blocked.
		2133
		2134	This means that before calling C<exec> (from the child) you should reset
		2135	the signal mask to whatever "default" you expect (all clear is a good
		2136	choice usually).
		2137
		2138	The simplest way to ensure that the signal mask is reset in the child is
		2139	to install a fork handler with C<pthread_atfork> that resets it. That will
		2140	catch fork calls done by libraries (such as the libc) as well.
		2141
		2142	In current versions of libev, you can also ensure that the signal mask is
		2143	not blocking any signals (except temporarily, so thread users watch out)
		2144	by specifying the C<EVFLAG_NOSIGFD> when creating the event loop. This
		2145	is not guaranteed for future versions, however.
		2146
		2147	=head3 Watcher-Specific Functions and Data Members
599		2148
600	=over 4	2149	=over 4
601		2150
602	=item ev_signal_init (ev_signal *, callback, int signum)	2151	=item ev_signal_init (ev_signal *, callback, int signum)
603		2152
604	=item ev_signal_set (ev_signal *, int signum)	2153	=item ev_signal_set (ev_signal *, int signum)
605		2154
606	Configures the watcher to trigger on the given signal number (usually one	2155	Configures the watcher to trigger on the given signal number (usually one
607	of the C<SIGxxx> constants).	2156	of the C<SIGxxx> constants).
608		2157
		2158	=item int signum [read-only]
		2159
		2160	The signal the watcher watches out for.
		2161
609	=back	2162	=back
610		2163
		2164	=head3 Examples
		2165
		2166	Example: Try to exit cleanly on SIGINT.
		2167
		2168	static void
		2169	sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
		2170	{
		2171	ev_unloop (loop, EVUNLOOP_ALL);
		2172	}
		2173
		2174	ev_signal signal_watcher;
		2175	ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
		2176	ev_signal_start (loop, &signal_watcher);
		2177
		2178
611	=head2 ev_child - wait for pid status changes	2179	=head2 C<ev_child> - watch out for process status changes
612		2180
613	Child watchers trigger when your process receives a SIGCHLD in response to	2181	Child watchers trigger when your process receives a SIGCHLD in response to
614	some child status changes (most typically when a child of yours dies).	2182	some child status changes (most typically when a child of yours dies or
		2183	exits). It is permissible to install a child watcher I<after> the child
		2184	has been forked (which implies it might have already exited), as long
		2185	as the event loop isn't entered (or is continued from a watcher), i.e.,
		2186	forking and then immediately registering a watcher for the child is fine,
		2187	but forking and registering a watcher a few event loop iterations later or
		2188	in the next callback invocation is not.
		2189
		2190	Only the default event loop is capable of handling signals, and therefore
		2191	you can only register child watchers in the default event loop.
		2192
		2193	Due to some design glitches inside libev, child watchers will always be
		2194	handled at maximum priority (their priority is set to C<EV_MAXPRI> by
		2195	libev)
		2196
		2197	=head3 Process Interaction
		2198
		2199	Libev grabs C<SIGCHLD> as soon as the default event loop is
		2200	initialised. This is necessary to guarantee proper behaviour even if the
		2201	first child watcher is started after the child exits. The occurrence
		2202	of C<SIGCHLD> is recorded asynchronously, but child reaping is done
		2203	synchronously as part of the event loop processing. Libev always reaps all
		2204	children, even ones not watched.
		2205
		2206	=head3 Overriding the Built-In Processing
		2207
		2208	Libev offers no special support for overriding the built-in child
		2209	processing, but if your application collides with libev's default child
		2210	handler, you can override it easily by installing your own handler for
		2211	C<SIGCHLD> after initialising the default loop, and making sure the
		2212	default loop never gets destroyed. You are encouraged, however, to use an
		2213	event-based approach to child reaping and thus use libev's support for
		2214	that, so other libev users can use C<ev_child> watchers freely.
		2215
		2216	=head3 Stopping the Child Watcher
		2217
		2218	Currently, the child watcher never gets stopped, even when the
		2219	child terminates, so normally one needs to stop the watcher in the
		2220	callback. Future versions of libev might stop the watcher automatically
		2221	when a child exit is detected (calling C<ev_child_stop> twice is not a
		2222	problem).
		2223
		2224	=head3 Watcher-Specific Functions and Data Members
615		2225
616	=over 4	2226	=over 4
617		2227
618	=item ev_child_init (ev_child *, callback, int pid)	2228	=item ev_child_init (ev_child *, callback, int pid, int trace)
619		2229
620	=item ev_child_set (ev_child *, int pid)	2230	=item ev_child_set (ev_child *, int pid, int trace)
621		2231
622	Configures the watcher to wait for status changes of process C<pid> (or	2232	Configures the watcher to wait for status changes of process C<pid> (or
623	I<any> process if C<pid> is specified as C<0>). The callback can look	2233	I<any> process if C<pid> is specified as C<0>). The callback can look
624	at the C<rstatus> member of the C<ev_child> watcher structure to see	2234	at the C<rstatus> member of the C<ev_child> watcher structure to see
625	the status word (use the macros from C<sys/wait.h>). The C<rpid> member	2235	the status word (use the macros from C<sys/wait.h> and see your systems
626	contains the pid of the process causing the status change.	2236	C<waitpid> documentation). The C<rpid> member contains the pid of the
		2237	process causing the status change. C<trace> must be either C<0> (only
		2238	activate the watcher when the process terminates) or C<1> (additionally
		2239	activate the watcher when the process is stopped or continued).
		2240
		2241	=item int pid [read-only]
		2242
		2243	The process id this watcher watches out for, or C<0>, meaning any process id.
		2244
		2245	=item int rpid [read-write]
		2246
		2247	The process id that detected a status change.
		2248
		2249	=item int rstatus [read-write]
		2250
		2251	The process exit/trace status caused by C<rpid> (see your systems
		2252	C<waitpid> and C<sys/wait.h> documentation for details).
627		2253
628	=back	2254	=back
629		2255
		2256	=head3 Examples
		2257
		2258	Example: C<fork()> a new process and install a child handler to wait for
		2259	its completion.
		2260
		2261	ev_child cw;
		2262
		2263	static void
		2264	child_cb (EV_P_ ev_child *w, int revents)
		2265	{
		2266	ev_child_stop (EV_A_ w);
		2267	printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
		2268	}
		2269
		2270	pid_t pid = fork ();
		2271
		2272	if (pid < 0)
		2273	// error
		2274	else if (pid == 0)
		2275	{
		2276	// the forked child executes here
		2277	exit (1);
		2278	}
		2279	else
		2280	{
		2281	ev_child_init (&cw, child_cb, pid, 0);
		2282	ev_child_start (EV_DEFAULT_ &cw);
		2283	}
		2284
		2285
		2286	=head2 C<ev_stat> - did the file attributes just change?
		2287
		2288	This watches a file system path for attribute changes. That is, it calls
		2289	C<stat> on that path in regular intervals (or when the OS says it changed)
		2290	and sees if it changed compared to the last time, invoking the callback if
		2291	it did.
		2292
		2293	The path does not need to exist: changing from "path exists" to "path does
		2294	not exist" is a status change like any other. The condition "path does not
		2295	exist" (or more correctly "path cannot be stat'ed") is signified by the
		2296	C<st_nlink> field being zero (which is otherwise always forced to be at
		2297	least one) and all the other fields of the stat buffer having unspecified
		2298	contents.
		2299
		2300	The path I<must not> end in a slash or contain special components such as
		2301	C<.> or C<..>. The path I<should> be absolute: If it is relative and
		2302	your working directory changes, then the behaviour is undefined.
		2303
		2304	Since there is no portable change notification interface available, the
		2305	portable implementation simply calls C<stat(2)> regularly on the path
		2306	to see if it changed somehow. You can specify a recommended polling
		2307	interval for this case. If you specify a polling interval of C<0> (highly
		2308	recommended!) then a I<suitable, unspecified default> value will be used
		2309	(which you can expect to be around five seconds, although this might
		2310	change dynamically). Libev will also impose a minimum interval which is
		2311	currently around C<0.1>, but that's usually overkill.
		2312
		2313	This watcher type is not meant for massive numbers of stat watchers,
		2314	as even with OS-supported change notifications, this can be
		2315	resource-intensive.
		2316
		2317	At the time of this writing, the only OS-specific interface implemented
		2318	is the Linux inotify interface (implementing kqueue support is left as an
		2319	exercise for the reader. Note, however, that the author sees no way of
		2320	implementing C<ev_stat> semantics with kqueue, except as a hint).
		2321
		2322	=head3 ABI Issues (Largefile Support)
		2323
		2324	Libev by default (unless the user overrides this) uses the default
		2325	compilation environment, which means that on systems with large file
		2326	support disabled by default, you get the 32 bit version of the stat
		2327	structure. When using the library from programs that change the ABI to
		2328	use 64 bit file offsets the programs will fail. In that case you have to
		2329	compile libev with the same flags to get binary compatibility. This is
		2330	obviously the case with any flags that change the ABI, but the problem is
		2331	most noticeably displayed with ev_stat and large file support.
		2332
		2333	The solution for this is to lobby your distribution maker to make large
		2334	file interfaces available by default (as e.g. FreeBSD does) and not
		2335	optional. Libev cannot simply switch on large file support because it has
		2336	to exchange stat structures with application programs compiled using the
		2337	default compilation environment.
		2338
		2339	=head3 Inotify and Kqueue
		2340
		2341	When C<inotify (7)> support has been compiled into libev and present at
		2342	runtime, it will be used to speed up change detection where possible. The
		2343	inotify descriptor will be created lazily when the first C<ev_stat>
		2344	watcher is being started.
		2345
		2346	Inotify presence does not change the semantics of C<ev_stat> watchers
		2347	except that changes might be detected earlier, and in some cases, to avoid
		2348	making regular C<stat> calls. Even in the presence of inotify support
		2349	there are many cases where libev has to resort to regular C<stat> polling,
		2350	but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
		2351	many bugs), the path exists (i.e. stat succeeds), and the path resides on
		2352	a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
		2353	xfs are fully working) libev usually gets away without polling.
		2354
		2355	There is no support for kqueue, as apparently it cannot be used to
		2356	implement this functionality, due to the requirement of having a file
		2357	descriptor open on the object at all times, and detecting renames, unlinks
		2358	etc. is difficult.
		2359
		2360	=head3 C<stat ()> is a synchronous operation
		2361
		2362	Libev doesn't normally do any kind of I/O itself, and so is not blocking
		2363	the process. The exception are C<ev_stat> watchers - those call C<stat
		2364	()>, which is a synchronous operation.
		2365
		2366	For local paths, this usually doesn't matter: unless the system is very
		2367	busy or the intervals between stat's are large, a stat call will be fast,
		2368	as the path data is usually in memory already (except when starting the
		2369	watcher).
		2370
		2371	For networked file systems, calling C<stat ()> can block an indefinite
		2372	time due to network issues, and even under good conditions, a stat call
		2373	often takes multiple milliseconds.
		2374
		2375	Therefore, it is best to avoid using C<ev_stat> watchers on networked
		2376	paths, although this is fully supported by libev.
		2377
		2378	=head3 The special problem of stat time resolution
		2379
		2380	The C<stat ()> system call only supports full-second resolution portably,
		2381	and even on systems where the resolution is higher, most file systems
		2382	still only support whole seconds.
		2383
		2384	That means that, if the time is the only thing that changes, you can
		2385	easily miss updates: on the first update, C<ev_stat> detects a change and
		2386	calls your callback, which does something. When there is another update
		2387	within the same second, C<ev_stat> will be unable to detect unless the
		2388	stat data does change in other ways (e.g. file size).
		2389
		2390	The solution to this is to delay acting on a change for slightly more
		2391	than a second (or till slightly after the next full second boundary), using
		2392	a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
		2393	ev_timer_again (loop, w)>).
		2394
		2395	The C<.02> offset is added to work around small timing inconsistencies
		2396	of some operating systems (where the second counter of the current time
		2397	might be be delayed. One such system is the Linux kernel, where a call to
		2398	C<gettimeofday> might return a timestamp with a full second later than
		2399	a subsequent C<time> call - if the equivalent of C<time ()> is used to
		2400	update file times then there will be a small window where the kernel uses
		2401	the previous second to update file times but libev might already execute
		2402	the timer callback).
		2403
		2404	=head3 Watcher-Specific Functions and Data Members
		2405
		2406	=over 4
		2407
		2408	=item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
		2409
		2410	=item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
		2411
		2412	Configures the watcher to wait for status changes of the given
		2413	C<path>. The C<interval> is a hint on how quickly a change is expected to
		2414	be detected and should normally be specified as C<0> to let libev choose
		2415	a suitable value. The memory pointed to by C<path> must point to the same
		2416	path for as long as the watcher is active.
		2417
		2418	The callback will receive an C<EV_STAT> event when a change was detected,
		2419	relative to the attributes at the time the watcher was started (or the
		2420	last change was detected).
		2421
		2422	=item ev_stat_stat (loop, ev_stat *)
		2423
		2424	Updates the stat buffer immediately with new values. If you change the
		2425	watched path in your callback, you could call this function to avoid
		2426	detecting this change (while introducing a race condition if you are not
		2427	the only one changing the path). Can also be useful simply to find out the
		2428	new values.
		2429
		2430	=item ev_statdata attr [read-only]
		2431
		2432	The most-recently detected attributes of the file. Although the type is
		2433	C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
		2434	suitable for your system, but you can only rely on the POSIX-standardised
		2435	members to be present. If the C<st_nlink> member is C<0>, then there was
		2436	some error while C<stat>ing the file.
		2437
		2438	=item ev_statdata prev [read-only]
		2439
		2440	The previous attributes of the file. The callback gets invoked whenever
		2441	C<prev> != C<attr>, or, more precisely, one or more of these members
		2442	differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
		2443	C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
		2444
		2445	=item ev_tstamp interval [read-only]
		2446
		2447	The specified interval.
		2448
		2449	=item const char *path [read-only]
		2450
		2451	The file system path that is being watched.
		2452
		2453	=back
		2454
		2455	=head3 Examples
		2456
		2457	Example: Watch C</etc/passwd> for attribute changes.
		2458
		2459	static void
		2460	passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
		2461	{
		2462	/* /etc/passwd changed in some way */
		2463	if (w->attr.st_nlink)
		2464	{
		2465	printf ("passwd current size %ld\n", (long)w->attr.st_size);
		2466	printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
		2467	printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
		2468	}
		2469	else
		2470	/* you shalt not abuse printf for puts */
		2471	puts ("wow, /etc/passwd is not there, expect problems. "
		2472	"if this is windows, they already arrived\n");
		2473	}
		2474
		2475	...
		2476	ev_stat passwd;
		2477
		2478	ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
		2479	ev_stat_start (loop, &passwd);
		2480
		2481	Example: Like above, but additionally use a one-second delay so we do not
		2482	miss updates (however, frequent updates will delay processing, too, so
		2483	one might do the work both on C<ev_stat> callback invocation I<and> on
		2484	C<ev_timer> callback invocation).
		2485
		2486	static ev_stat passwd;
		2487	static ev_timer timer;
		2488
		2489	static void
		2490	timer_cb (EV_P_ ev_timer *w, int revents)
		2491	{
		2492	ev_timer_stop (EV_A_ w);
		2493
		2494	/* now it's one second after the most recent passwd change */
		2495	}
		2496
		2497	static void
		2498	stat_cb (EV_P_ ev_stat *w, int revents)
		2499	{
		2500	/* reset the one-second timer */
		2501	ev_timer_again (EV_A_ &timer);
		2502	}
		2503
		2504	...
		2505	ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
		2506	ev_stat_start (loop, &passwd);
		2507	ev_timer_init (&timer, timer_cb, 0., 1.02);
		2508
		2509
630	=head2 ev_idle - when you've got nothing better to do	2510	=head2 C<ev_idle> - when you've got nothing better to do...
631		2511
632	Idle watchers trigger events when there are no other I/O or timer (or	2512	Idle watchers trigger events when no other events of the same or higher
633	periodic) events pending. That is, as long as your process is busy	2513	priority are pending (prepare, check and other idle watchers do not count
634	handling sockets or timeouts it will not be called. But when your process	2514	as receiving "events").
635	is idle all idle watchers are being called again and again - until	2515
		2516	That is, as long as your process is busy handling sockets or timeouts
		2517	(or even signals, imagine) of the same or higher priority it will not be
		2518	triggered. But when your process is idle (or only lower-priority watchers
		2519	are pending), the idle watchers are being called once per event loop
636	stopped, that is, or your process receives more events.	2520	iteration - until stopped, that is, or your process receives more events
		2521	and becomes busy again with higher priority stuff.
637		2522
638	The most noteworthy effect is that as long as any idle watchers are	2523	The most noteworthy effect is that as long as any idle watchers are
639	active, the process will not block when waiting for new events.	2524	active, the process will not block when waiting for new events.
640		2525
641	Apart from keeping your process non-blocking (which is a useful	2526	Apart from keeping your process non-blocking (which is a useful
642	effect on its own sometimes), idle watchers are a good place to do	2527	effect on its own sometimes), idle watchers are a good place to do
643	"pseudo-background processing", or delay processing stuff to after the	2528	"pseudo-background processing", or delay processing stuff to after the
644	event loop has handled all outstanding events.	2529	event loop has handled all outstanding events.
645		2530
		2531	=head3 Watcher-Specific Functions and Data Members
		2532
646	=over 4	2533	=over 4
647		2534
648	=item ev_idle_init (ev_signal *, callback)	2535	=item ev_idle_init (ev_idle *, callback)
649		2536
650	Initialises and configures the idle watcher - it has no parameters of any	2537	Initialises and configures the idle watcher - it has no parameters of any
651	kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,	2538	kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
652	believe me.	2539	believe me.
653		2540
654	=back	2541	=back
655		2542
656	=head2 prepare and check - your hooks into the event loop	2543	=head3 Examples
657		2544
		2545	Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
		2546	callback, free it. Also, use no error checking, as usual.
		2547
		2548	static void
		2549	idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
		2550	{
		2551	free (w);
		2552	// now do something you wanted to do when the program has
		2553	// no longer anything immediate to do.
		2554	}
		2555
		2556	ev_idle *idle_watcher = malloc (sizeof (ev_idle));
		2557	ev_idle_init (idle_watcher, idle_cb);
		2558	ev_idle_start (loop, idle_watcher);
		2559
		2560
		2561	=head2 C<ev_prepare> and C<ev_check> - customise your event loop!
		2562
658	Prepare and check watchers usually (but not always) are used in	2563	Prepare and check watchers are usually (but not always) used in pairs:
659	tandom. Prepare watchers get invoked before the process blocks and check	2564	prepare watchers get invoked before the process blocks and check watchers
660	watchers afterwards.	2565	afterwards.
661		2566
		2567	You I<must not> call C<ev_loop> or similar functions that enter
		2568	the current event loop from either C<ev_prepare> or C<ev_check>
		2569	watchers. Other loops than the current one are fine, however. The
		2570	rationale behind this is that you do not need to check for recursion in
		2571	those watchers, i.e. the sequence will always be C<ev_prepare>, blocking,
		2572	C<ev_check> so if you have one watcher of each kind they will always be
		2573	called in pairs bracketing the blocking call.
		2574
662	Their main purpose is to integrate other event mechanisms into libev. This	2575	Their main purpose is to integrate other event mechanisms into libev and
663	could be used, for example, to track variable changes, implement your own	2576	their use is somewhat advanced. They could be used, for example, to track
664	watchers, integrate net-snmp or a coroutine library and lots more.	2577	variable changes, implement your own watchers, integrate net-snmp or a
		2578	coroutine library and lots more. They are also occasionally useful if
		2579	you cache some data and want to flush it before blocking (for example,
		2580	in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
		2581	watcher).
665		2582
666	This is done by examining in each prepare call which file descriptors need	2583	This is done by examining in each prepare call which file descriptors
667	to be watched by the other library, registering ev_io watchers for them	2584	need to be watched by the other library, registering C<ev_io> watchers
668	and starting an ev_timer watcher for any timeouts (many libraries provide	2585	for them and starting an C<ev_timer> watcher for any timeouts (many
669	just this functionality). Then, in the check watcher you check for any	2586	libraries provide exactly this functionality). Then, in the check watcher,
670	events that occured (by making your callbacks set soem flags for example)	2587	you check for any events that occurred (by checking the pending status
671	and call back into the library.	2588	of all watchers and stopping them) and call back into the library. The
		2589	I/O and timer callbacks will never actually be called (but must be valid
		2590	nevertheless, because you never know, you know?).
672		2591
673	As another example, the perl Coro module uses these hooks to integrate	2592	As another example, the Perl Coro module uses these hooks to integrate
674	coroutines into libev programs, by yielding to other active coroutines	2593	coroutines into libev programs, by yielding to other active coroutines
675	during each prepare and only letting the process block if no coroutines	2594	during each prepare and only letting the process block if no coroutines
676	are ready to run.	2595	are ready to run (it's actually more complicated: it only runs coroutines
		2596	with priority higher than or equal to the event loop and one coroutine
		2597	of lower priority, but only once, using idle watchers to keep the event
		2598	loop from blocking if lower-priority coroutines are active, thus mapping
		2599	low-priority coroutines to idle/background tasks).
		2600
		2601	It is recommended to give C<ev_check> watchers highest (C<EV_MAXPRI>)
		2602	priority, to ensure that they are being run before any other watchers
		2603	after the poll (this doesn't matter for C<ev_prepare> watchers).
		2604
		2605	Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
		2606	activate ("feed") events into libev. While libev fully supports this, they
		2607	might get executed before other C<ev_check> watchers did their job. As
		2608	C<ev_check> watchers are often used to embed other (non-libev) event
		2609	loops those other event loops might be in an unusable state until their
		2610	C<ev_check> watcher ran (always remind yourself to coexist peacefully with
		2611	others).
		2612
		2613	=head3 Watcher-Specific Functions and Data Members
677		2614
678	=over 4	2615	=over 4
679		2616
680	=item ev_prepare_init (ev_prepare *, callback)	2617	=item ev_prepare_init (ev_prepare *, callback)
681		2618
682	=item ev_check_init (ev_check *, callback)	2619	=item ev_check_init (ev_check *, callback)
683		2620
684	Initialises and configures the prepare or check watcher - they have no	2621	Initialises and configures the prepare or check watcher - they have no
685	parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>	2622	parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
686	macros, but using them is utterly, utterly pointless.	2623	macros, but using them is utterly, utterly, utterly and completely
		2624	pointless.
687		2625
688	=back	2626	=back
689		2627
		2628	=head3 Examples
		2629
		2630	There are a number of principal ways to embed other event loops or modules
		2631	into libev. Here are some ideas on how to include libadns into libev
		2632	(there is a Perl module named C<EV::ADNS> that does this, which you could
		2633	use as a working example. Another Perl module named C<EV::Glib> embeds a
		2634	Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
		2635	Glib event loop).
		2636
		2637	Method 1: Add IO watchers and a timeout watcher in a prepare handler,
		2638	and in a check watcher, destroy them and call into libadns. What follows
		2639	is pseudo-code only of course. This requires you to either use a low
		2640	priority for the check watcher or use C<ev_clear_pending> explicitly, as
		2641	the callbacks for the IO/timeout watchers might not have been called yet.
		2642
		2643	static ev_io iow [nfd];
		2644	static ev_timer tw;
		2645
		2646	static void
		2647	io_cb (struct ev_loop *loop, ev_io *w, int revents)
		2648	{
		2649	}
		2650
		2651	// create io watchers for each fd and a timer before blocking
		2652	static void
		2653	adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
		2654	{
		2655	int timeout = 3600000;
		2656	struct pollfd fds [nfd];
		2657	// actual code will need to loop here and realloc etc.
		2658	adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
		2659
		2660	/* the callback is illegal, but won't be called as we stop during check */
		2661	ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
		2662	ev_timer_start (loop, &tw);
		2663
		2664	// create one ev_io per pollfd
		2665	for (int i = 0; i < nfd; ++i)
		2666	{
		2667	ev_io_init (iow + i, io_cb, fds [i].fd,
		2668	((fds [i].events & POLLIN ? EV_READ : 0)
		2669	| (fds [i].events & POLLOUT ? EV_WRITE : 0)));
		2670
		2671	fds [i].revents = 0;
		2672	ev_io_start (loop, iow + i);
		2673	}
		2674	}
		2675
		2676	// stop all watchers after blocking
		2677	static void
		2678	adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
		2679	{
		2680	ev_timer_stop (loop, &tw);
		2681
		2682	for (int i = 0; i < nfd; ++i)
		2683	{
		2684	// set the relevant poll flags
		2685	// could also call adns_processreadable etc. here
		2686	struct pollfd *fd = fds + i;
		2687	int revents = ev_clear_pending (iow + i);
		2688	if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
		2689	if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
		2690
		2691	// now stop the watcher
		2692	ev_io_stop (loop, iow + i);
		2693	}
		2694
		2695	adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
		2696	}
		2697
		2698	Method 2: This would be just like method 1, but you run C<adns_afterpoll>
		2699	in the prepare watcher and would dispose of the check watcher.
		2700
		2701	Method 3: If the module to be embedded supports explicit event
		2702	notification (libadns does), you can also make use of the actual watcher
		2703	callbacks, and only destroy/create the watchers in the prepare watcher.
		2704
		2705	static void
		2706	timer_cb (EV_P_ ev_timer *w, int revents)
		2707	{
		2708	adns_state ads = (adns_state)w->data;
		2709	update_now (EV_A);
		2710
		2711	adns_processtimeouts (ads, &tv_now);
		2712	}
		2713
		2714	static void
		2715	io_cb (EV_P_ ev_io *w, int revents)
		2716	{
		2717	adns_state ads = (adns_state)w->data;
		2718	update_now (EV_A);
		2719
		2720	if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
		2721	if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
		2722	}
		2723
		2724	// do not ever call adns_afterpoll
		2725
		2726	Method 4: Do not use a prepare or check watcher because the module you
		2727	want to embed is not flexible enough to support it. Instead, you can
		2728	override their poll function. The drawback with this solution is that the
		2729	main loop is now no longer controllable by EV. The C<Glib::EV> module uses
		2730	this approach, effectively embedding EV as a client into the horrible
		2731	libglib event loop.
		2732
		2733	static gint
		2734	event_poll_func (GPollFD *fds, guint nfds, gint timeout)
		2735	{
		2736	int got_events = 0;
		2737
		2738	for (n = 0; n < nfds; ++n)
		2739	// create/start io watcher that sets the relevant bits in fds[n] and increment got_events
		2740
		2741	if (timeout >= 0)
		2742	// create/start timer
		2743
		2744	// poll
		2745	ev_loop (EV_A_ 0);
		2746
		2747	// stop timer again
		2748	if (timeout >= 0)
		2749	ev_timer_stop (EV_A_ &to);
		2750
		2751	// stop io watchers again - their callbacks should have set
		2752	for (n = 0; n < nfds; ++n)
		2753	ev_io_stop (EV_A_ iow [n]);
		2754
		2755	return got_events;
		2756	}
		2757
		2758
		2759	=head2 C<ev_embed> - when one backend isn't enough...
		2760
		2761	This is a rather advanced watcher type that lets you embed one event loop
		2762	into another (currently only C<ev_io> events are supported in the embedded
		2763	loop, other types of watchers might be handled in a delayed or incorrect
		2764	fashion and must not be used).
		2765
		2766	There are primarily two reasons you would want that: work around bugs and
		2767	prioritise I/O.
		2768
		2769	As an example for a bug workaround, the kqueue backend might only support
		2770	sockets on some platform, so it is unusable as generic backend, but you
		2771	still want to make use of it because you have many sockets and it scales
		2772	so nicely. In this case, you would create a kqueue-based loop and embed
		2773	it into your default loop (which might use e.g. poll). Overall operation
		2774	will be a bit slower because first libev has to call C<poll> and then
		2775	C<kevent>, but at least you can use both mechanisms for what they are
		2776	best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
		2777
		2778	As for prioritising I/O: under rare circumstances you have the case where
		2779	some fds have to be watched and handled very quickly (with low latency),
		2780	and even priorities and idle watchers might have too much overhead. In
		2781	this case you would put all the high priority stuff in one loop and all
		2782	the rest in a second one, and embed the second one in the first.
		2783
		2784	As long as the watcher is active, the callback will be invoked every
		2785	time there might be events pending in the embedded loop. The callback
		2786	must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
		2787	sweep and invoke their callbacks (the callback doesn't need to invoke the
		2788	C<ev_embed_sweep> function directly, it could also start an idle watcher
		2789	to give the embedded loop strictly lower priority for example).
		2790
		2791	You can also set the callback to C<0>, in which case the embed watcher
		2792	will automatically execute the embedded loop sweep whenever necessary.
		2793
		2794	Fork detection will be handled transparently while the C<ev_embed> watcher
		2795	is active, i.e., the embedded loop will automatically be forked when the
		2796	embedding loop forks. In other cases, the user is responsible for calling
		2797	C<ev_loop_fork> on the embedded loop.
		2798
		2799	Unfortunately, not all backends are embeddable: only the ones returned by
		2800	C<ev_embeddable_backends> are, which, unfortunately, does not include any
		2801	portable one.
		2802
		2803	So when you want to use this feature you will always have to be prepared
		2804	that you cannot get an embeddable loop. The recommended way to get around
		2805	this is to have a separate variables for your embeddable loop, try to
		2806	create it, and if that fails, use the normal loop for everything.
		2807
		2808	=head3 C<ev_embed> and fork
		2809
		2810	While the C<ev_embed> watcher is running, forks in the embedding loop will
		2811	automatically be applied to the embedded loop as well, so no special
		2812	fork handling is required in that case. When the watcher is not running,
		2813	however, it is still the task of the libev user to call C<ev_loop_fork ()>
		2814	as applicable.
		2815
		2816	=head3 Watcher-Specific Functions and Data Members
		2817
		2818	=over 4
		2819
		2820	=item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
		2821
		2822	=item ev_embed_set (ev_embed *, callback, struct ev_loop *embedded_loop)
		2823
		2824	Configures the watcher to embed the given loop, which must be
		2825	embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
		2826	invoked automatically, otherwise it is the responsibility of the callback
		2827	to invoke it (it will continue to be called until the sweep has been done,
		2828	if you do not want that, you need to temporarily stop the embed watcher).
		2829
		2830	=item ev_embed_sweep (loop, ev_embed *)
		2831
		2832	Make a single, non-blocking sweep over the embedded loop. This works
		2833	similarly to C<ev_loop (embedded_loop, EVLOOP_NONBLOCK)>, but in the most
		2834	appropriate way for embedded loops.
		2835
		2836	=item struct ev_loop *other [read-only]
		2837
		2838	The embedded event loop.
		2839
		2840	=back
		2841
		2842	=head3 Examples
		2843
		2844	Example: Try to get an embeddable event loop and embed it into the default
		2845	event loop. If that is not possible, use the default loop. The default
		2846	loop is stored in C<loop_hi>, while the embeddable loop is stored in
		2847	C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
		2848	used).
		2849
		2850	struct ev_loop *loop_hi = ev_default_init (0);
		2851	struct ev_loop *loop_lo = 0;
		2852	ev_embed embed;
		2853
		2854	// see if there is a chance of getting one that works
		2855	// (remember that a flags value of 0 means autodetection)
		2856	loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
		2857	? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
		2858	: 0;
		2859
		2860	// if we got one, then embed it, otherwise default to loop_hi
		2861	if (loop_lo)
		2862	{
		2863	ev_embed_init (&embed, 0, loop_lo);
		2864	ev_embed_start (loop_hi, &embed);
		2865	}
		2866	else
		2867	loop_lo = loop_hi;
		2868
		2869	Example: Check if kqueue is available but not recommended and create
		2870	a kqueue backend for use with sockets (which usually work with any
		2871	kqueue implementation). Store the kqueue/socket-only event loop in
		2872	C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
		2873
		2874	struct ev_loop *loop = ev_default_init (0);
		2875	struct ev_loop *loop_socket = 0;
		2876	ev_embed embed;
		2877
		2878	if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
		2879	if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
		2880	{
		2881	ev_embed_init (&embed, 0, loop_socket);
		2882	ev_embed_start (loop, &embed);
		2883	}
		2884
		2885	if (!loop_socket)
		2886	loop_socket = loop;
		2887
		2888	// now use loop_socket for all sockets, and loop for everything else
		2889
		2890
		2891	=head2 C<ev_fork> - the audacity to resume the event loop after a fork
		2892
		2893	Fork watchers are called when a C<fork ()> was detected (usually because
		2894	whoever is a good citizen cared to tell libev about it by calling
		2895	C<ev_default_fork> or C<ev_loop_fork>). The invocation is done before the
		2896	event loop blocks next and before C<ev_check> watchers are being called,
		2897	and only in the child after the fork. If whoever good citizen calling
		2898	C<ev_default_fork> cheats and calls it in the wrong process, the fork
		2899	handlers will be invoked, too, of course.
		2900
		2901	=head3 The special problem of life after fork - how is it possible?
		2902
		2903	Most uses of C<fork()> consist of forking, then some simple calls to ste
		2904	up/change the process environment, followed by a call to C<exec()>. This
		2905	sequence should be handled by libev without any problems.
		2906
		2907	This changes when the application actually wants to do event handling
		2908	in the child, or both parent in child, in effect "continuing" after the
		2909	fork.
		2910
		2911	The default mode of operation (for libev, with application help to detect
		2912	forks) is to duplicate all the state in the child, as would be expected
		2913	when I<either> the parent I<or> the child process continues.
		2914
		2915	When both processes want to continue using libev, then this is usually the
		2916	wrong result. In that case, usually one process (typically the parent) is
		2917	supposed to continue with all watchers in place as before, while the other
		2918	process typically wants to start fresh, i.e. without any active watchers.
		2919
		2920	The cleanest and most efficient way to achieve that with libev is to
		2921	simply create a new event loop, which of course will be "empty", and
		2922	use that for new watchers. This has the advantage of not touching more
		2923	memory than necessary, and thus avoiding the copy-on-write, and the
		2924	disadvantage of having to use multiple event loops (which do not support
		2925	signal watchers).
		2926
		2927	When this is not possible, or you want to use the default loop for
		2928	other reasons, then in the process that wants to start "fresh", call
		2929	C<ev_default_destroy ()> followed by C<ev_default_loop (...)>. Destroying
		2930	the default loop will "orphan" (not stop) all registered watchers, so you
		2931	have to be careful not to execute code that modifies those watchers. Note
		2932	also that in that case, you have to re-register any signal watchers.
		2933
		2934	=head3 Watcher-Specific Functions and Data Members
		2935
		2936	=over 4
		2937
		2938	=item ev_fork_init (ev_signal *, callback)
		2939
		2940	Initialises and configures the fork watcher - it has no parameters of any
		2941	kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
		2942	believe me.
		2943
		2944	=back
		2945
		2946
		2947	=head2 C<ev_async> - how to wake up another event loop
		2948
		2949	In general, you cannot use an C<ev_loop> from multiple threads or other
		2950	asynchronous sources such as signal handlers (as opposed to multiple event
		2951	loops - those are of course safe to use in different threads).
		2952
		2953	Sometimes, however, you need to wake up another event loop you do not
		2954	control, for example because it belongs to another thread. This is what
		2955	C<ev_async> watchers do: as long as the C<ev_async> watcher is active, you
		2956	can signal it by calling C<ev_async_send>, which is thread- and signal
		2957	safe.
		2958
		2959	This functionality is very similar to C<ev_signal> watchers, as signals,
		2960	too, are asynchronous in nature, and signals, too, will be compressed
		2961	(i.e. the number of callback invocations may be less than the number of
		2962	C<ev_async_sent> calls).
		2963
		2964	Unlike C<ev_signal> watchers, C<ev_async> works with any event loop, not
		2965	just the default loop.
		2966
		2967	=head3 Queueing
		2968
		2969	C<ev_async> does not support queueing of data in any way. The reason
		2970	is that the author does not know of a simple (or any) algorithm for a
		2971	multiple-writer-single-reader queue that works in all cases and doesn't
		2972	need elaborate support such as pthreads.
		2973
		2974	That means that if you want to queue data, you have to provide your own
		2975	queue. But at least I can tell you how to implement locking around your
		2976	queue:
		2977
		2978	=over 4
		2979
		2980	=item queueing from a signal handler context
		2981
		2982	To implement race-free queueing, you simply add to the queue in the signal
		2983	handler but you block the signal handler in the watcher callback. Here is
		2984	an example that does that for some fictitious SIGUSR1 handler:
		2985
		2986	static ev_async mysig;
		2987
		2988	static void
		2989	sigusr1_handler (void)
		2990	{
		2991	sometype data;
		2992
		2993	// no locking etc.
		2994	queue_put (data);
		2995	ev_async_send (EV_DEFAULT_ &mysig);
		2996	}
		2997
		2998	static void
		2999	mysig_cb (EV_P_ ev_async *w, int revents)
		3000	{
		3001	sometype data;
		3002	sigset_t block, prev;
		3003
		3004	sigemptyset (&block);
		3005	sigaddset (&block, SIGUSR1);
		3006	sigprocmask (SIG_BLOCK, &block, &prev);
		3007
		3008	while (queue_get (&data))
		3009	process (data);
		3010
		3011	if (sigismember (&prev, SIGUSR1)
		3012	sigprocmask (SIG_UNBLOCK, &block, 0);
		3013	}
		3014
		3015	(Note: pthreads in theory requires you to use C<pthread_setmask>
		3016	instead of C<sigprocmask> when you use threads, but libev doesn't do it
		3017	either...).
		3018
		3019	=item queueing from a thread context
		3020
		3021	The strategy for threads is different, as you cannot (easily) block
		3022	threads but you can easily preempt them, so to queue safely you need to
		3023	employ a traditional mutex lock, such as in this pthread example:
		3024
		3025	static ev_async mysig;
		3026	static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
		3027
		3028	static void
		3029	otherthread (void)
		3030	{
		3031	// only need to lock the actual queueing operation
		3032	pthread_mutex_lock (&mymutex);
		3033	queue_put (data);
		3034	pthread_mutex_unlock (&mymutex);
		3035
		3036	ev_async_send (EV_DEFAULT_ &mysig);
		3037	}
		3038
		3039	static void
		3040	mysig_cb (EV_P_ ev_async *w, int revents)
		3041	{
		3042	pthread_mutex_lock (&mymutex);
		3043
		3044	while (queue_get (&data))
		3045	process (data);
		3046
		3047	pthread_mutex_unlock (&mymutex);
		3048	}
		3049
		3050	=back
		3051
		3052
		3053	=head3 Watcher-Specific Functions and Data Members
		3054
		3055	=over 4
		3056
		3057	=item ev_async_init (ev_async *, callback)
		3058
		3059	Initialises and configures the async watcher - it has no parameters of any
		3060	kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
		3061	trust me.
		3062
		3063	=item ev_async_send (loop, ev_async *)
		3064
		3065	Sends/signals/activates the given C<ev_async> watcher, that is, feeds
		3066	an C<EV_ASYNC> event on the watcher into the event loop. Unlike
		3067	C<ev_feed_event>, this call is safe to do from other threads, signal or
		3068	similar contexts (see the discussion of C<EV_ATOMIC_T> in the embedding
		3069	section below on what exactly this means).
		3070
		3071	Note that, as with other watchers in libev, multiple events might get
		3072	compressed into a single callback invocation (another way to look at this
		3073	is that C<ev_async> watchers are level-triggered, set on C<ev_async_send>,
		3074	reset when the event loop detects that).
		3075
		3076	This call incurs the overhead of a system call only once per event loop
		3077	iteration, so while the overhead might be noticeable, it doesn't apply to
		3078	repeated calls to C<ev_async_send> for the same event loop.
		3079
		3080	=item bool = ev_async_pending (ev_async *)
		3081
		3082	Returns a non-zero value when C<ev_async_send> has been called on the
		3083	watcher but the event has not yet been processed (or even noted) by the
		3084	event loop.
		3085
		3086	C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
		3087	the loop iterates next and checks for the watcher to have become active,
		3088	it will reset the flag again. C<ev_async_pending> can be used to very
		3089	quickly check whether invoking the loop might be a good idea.
		3090
		3091	Not that this does I<not> check whether the watcher itself is pending,
		3092	only whether it has been requested to make this watcher pending: there
		3093	is a time window between the event loop checking and resetting the async
		3094	notification, and the callback being invoked.
		3095
		3096	=back
		3097
		3098
690	=head1 OTHER FUNCTIONS	3099	=head1 OTHER FUNCTIONS
691		3100
692	There are some other fucntions of possible interest. Described. Here. Now.	3101	There are some other functions of possible interest. Described. Here. Now.
693		3102
694	=over 4	3103	=over 4
695		3104
696	=item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)	3105	=item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)
697		3106
698	This function combines a simple timer and an I/O watcher, calls your	3107	This function combines a simple timer and an I/O watcher, calls your
699	callback on whichever event happens first and automatically stop both	3108	callback on whichever event happens first and automatically stops both
700	watchers. This is useful if you want to wait for a single event on an fd	3109	watchers. This is useful if you want to wait for a single event on an fd
701	or timeout without havign to allocate/configure/start/stop/free one or	3110	or timeout without having to allocate/configure/start/stop/free one or
702	more watchers yourself.	3111	more watchers yourself.
703		3112
704	If C<fd> is less than 0, then no I/O watcher will be started and events is	3113	If C<fd> is less than 0, then no I/O watcher will be started and the
705	ignored. Otherwise, an ev_io watcher for the given C<fd> and C<events> set	3114	C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
706	will be craeted and started.	3115	the given C<fd> and C<events> set will be created and started.
707		3116
708	If C<timeout> is less than 0, then no timeout watcher will be	3117	If C<timeout> is less than 0, then no timeout watcher will be
709	started. Otherwise an ev_timer watcher with after = C<timeout> (and repeat	3118	started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
710	= 0) will be started.	3119	repeat = 0) will be started. C<0> is a valid timeout.
711		3120
712	The callback has the type C<void (*cb)(int revents, void *arg)> and	3121	The callback has the type C<void (*cb)(int revents, void *arg)> and gets
713	gets passed an events set (normally a combination of EV_ERROR, EV_READ,	3122	passed an C<revents> set like normal event callbacks (a combination of
714	EV_WRITE or EV_TIMEOUT) and the C<arg> value passed to C<ev_once>:	3123	C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMEOUT>) and the C<arg>
		3124	value passed to C<ev_once>. Note that it is possible to receive I<both>
		3125	a timeout and an io event at the same time - you probably should give io
		3126	events precedence.
715		3127
		3128	Example: wait up to ten seconds for data to appear on STDIN_FILENO.
		3129
716	static void stdin_ready (int revents, void *arg)	3130	static void stdin_ready (int revents, void *arg)
717	{	3131	{
718	if (revents & EV_TIMEOUT)
719	/* doh, nothing entered */
720	else if (revents & EV_READ)	3132	if (revents & EV_READ)
721	/* stdin might have data for us, joy! */	3133	/* stdin might have data for us, joy! */;
		3134	else if (revents & EV_TIMEOUT)
		3135	/* doh, nothing entered */;
722	}	3136	}
723		3137
724	ev_once (STDIN_FILENO, EV_READm 10., stdin_ready, 0);	3138	ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
725		3139
726	=item ev_feed_event (loop, watcher, int events)	3140	=item ev_feed_event (struct ev_loop *, watcher *, int revents)
727		3141
728	Feeds the given event set into the event loop, as if the specified event	3142	Feeds the given event set into the event loop, as if the specified event
729	has happened for the specified watcher (which must be a pointer to an	3143	had happened for the specified watcher (which must be a pointer to an
730	initialised but not necessarily active event watcher).	3144	initialised but not necessarily started event watcher).
731		3145
732	=item ev_feed_fd_event (loop, int fd, int revents)	3146	=item ev_feed_fd_event (struct ev_loop *, int fd, int revents)
733		3147
734	Feed an event on the given fd, as if a file descriptor backend detected it.	3148	Feed an event on the given fd, as if a file descriptor backend detected
		3149	the given events it.
735		3150
736	=item ev_feed_signal_event (loop, int signum)	3151	=item ev_feed_signal_event (struct ev_loop *loop, int signum)
737		3152
738	Feed an event as if the given signal occured (loop must be the default loop!).	3153	Feed an event as if the given signal occurred (C<loop> must be the default
		3154	loop!).
739		3155
740	=back	3156	=back
741		3157
		3158
		3159	=head1 LIBEVENT EMULATION
		3160
		3161	Libev offers a compatibility emulation layer for libevent. It cannot
		3162	emulate the internals of libevent, so here are some usage hints:
		3163
		3164	=over 4
		3165
		3166	=item * Use it by including <event.h>, as usual.
		3167
		3168	=item * The following members are fully supported: ev_base, ev_callback,
		3169	ev_arg, ev_fd, ev_res, ev_events.
		3170
		3171	=item * Avoid using ev_flags and the EVLIST_*-macros, while it is
		3172	maintained by libev, it does not work exactly the same way as in libevent (consider
		3173	it a private API).
		3174
		3175	=item * Priorities are not currently supported. Initialising priorities
		3176	will fail and all watchers will have the same priority, even though there
		3177	is an ev_pri field.
		3178
		3179	=item * In libevent, the last base created gets the signals, in libev, the
		3180	first base created (== the default loop) gets the signals.
		3181
		3182	=item * Other members are not supported.
		3183
		3184	=item * The libev emulation is I<not> ABI compatible to libevent, you need
		3185	to use the libev header file and library.
		3186
		3187	=back
		3188
		3189	=head1 C++ SUPPORT
		3190
		3191	Libev comes with some simplistic wrapper classes for C++ that mainly allow
		3192	you to use some convenience methods to start/stop watchers and also change
		3193	the callback model to a model using method callbacks on objects.
		3194
		3195	To use it,
		3196
		3197	#include <ev++.h>
		3198
		3199	This automatically includes F<ev.h> and puts all of its definitions (many
		3200	of them macros) into the global namespace. All C++ specific things are
		3201	put into the C<ev> namespace. It should support all the same embedding
		3202	options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
		3203
		3204	Care has been taken to keep the overhead low. The only data member the C++
		3205	classes add (compared to plain C-style watchers) is the event loop pointer
		3206	that the watcher is associated with (or no additional members at all if
		3207	you disable C<EV_MULTIPLICITY> when embedding libev).
		3208
		3209	Currently, functions, and static and non-static member functions can be
		3210	used as callbacks. Other types should be easy to add as long as they only
		3211	need one additional pointer for context. If you need support for other
		3212	types of functors please contact the author (preferably after implementing
		3213	it).
		3214
		3215	Here is a list of things available in the C<ev> namespace:
		3216
		3217	=over 4
		3218
		3219	=item C<ev::READ>, C<ev::WRITE> etc.
		3220
		3221	These are just enum values with the same values as the C<EV_READ> etc.
		3222	macros from F<ev.h>.
		3223
		3224	=item C<ev::tstamp>, C<ev::now>
		3225
		3226	Aliases to the same types/functions as with the C<ev_> prefix.
		3227
		3228	=item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
		3229
		3230	For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
		3231	the same name in the C<ev> namespace, with the exception of C<ev_signal>
		3232	which is called C<ev::sig> to avoid clashes with the C<signal> macro
		3233	defines by many implementations.
		3234
		3235	All of those classes have these methods:
		3236
		3237	=over 4
		3238
		3239	=item ev::TYPE::TYPE ()
		3240
		3241	=item ev::TYPE::TYPE (struct ev_loop *)
		3242
		3243	=item ev::TYPE::~TYPE
		3244
		3245	The constructor (optionally) takes an event loop to associate the watcher
		3246	with. If it is omitted, it will use C<EV_DEFAULT>.
		3247
		3248	The constructor calls C<ev_init> for you, which means you have to call the
		3249	C<set> method before starting it.
		3250
		3251	It will not set a callback, however: You have to call the templated C<set>
		3252	method to set a callback before you can start the watcher.
		3253
		3254	(The reason why you have to use a method is a limitation in C++ which does
		3255	not allow explicit template arguments for constructors).
		3256
		3257	The destructor automatically stops the watcher if it is active.
		3258
		3259	=item w->set<class, &class::method> (object *)
		3260
		3261	This method sets the callback method to call. The method has to have a
		3262	signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
		3263	first argument and the C<revents> as second. The object must be given as
		3264	parameter and is stored in the C<data> member of the watcher.
		3265
		3266	This method synthesizes efficient thunking code to call your method from
		3267	the C callback that libev requires. If your compiler can inline your
		3268	callback (i.e. it is visible to it at the place of the C<set> call and
		3269	your compiler is good :), then the method will be fully inlined into the
		3270	thunking function, making it as fast as a direct C callback.
		3271
		3272	Example: simple class declaration and watcher initialisation
		3273
		3274	struct myclass
		3275	{
		3276	void io_cb (ev::io &w, int revents) { }
		3277	}
		3278
		3279	myclass obj;
		3280	ev::io iow;
		3281	iow.set <myclass, &myclass::io_cb> (&obj);
		3282
		3283	=item w->set (object *)
		3284
		3285	This is an B<experimental> feature that might go away in a future version.
		3286
		3287	This is a variation of a method callback - leaving out the method to call
		3288	will default the method to C<operator ()>, which makes it possible to use
		3289	functor objects without having to manually specify the C<operator ()> all
		3290	the time. Incidentally, you can then also leave out the template argument
		3291	list.
		3292
		3293	The C<operator ()> method prototype must be C<void operator ()(watcher &w,
		3294	int revents)>.
		3295
		3296	See the method-C<set> above for more details.
		3297
		3298	Example: use a functor object as callback.
		3299
		3300	struct myfunctor
		3301	{
		3302	void operator() (ev::io &w, int revents)
		3303	{
		3304	...
		3305	}
		3306	}
		3307
		3308	myfunctor f;
		3309
		3310	ev::io w;
		3311	w.set (&f);
		3312
		3313	=item w->set<function> (void *data = 0)
		3314
		3315	Also sets a callback, but uses a static method or plain function as
		3316	callback. The optional C<data> argument will be stored in the watcher's
		3317	C<data> member and is free for you to use.
		3318
		3319	The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
		3320
		3321	See the method-C<set> above for more details.
		3322
		3323	Example: Use a plain function as callback.
		3324
		3325	static void io_cb (ev::io &w, int revents) { }
		3326	iow.set <io_cb> ();
		3327
		3328	=item w->set (struct ev_loop *)
		3329
		3330	Associates a different C<struct ev_loop> with this watcher. You can only
		3331	do this when the watcher is inactive (and not pending either).
		3332
		3333	=item w->set ([arguments])
		3334
		3335	Basically the same as C<ev_TYPE_set>, with the same arguments. Must be
		3336	called at least once. Unlike the C counterpart, an active watcher gets
		3337	automatically stopped and restarted when reconfiguring it with this
		3338	method.
		3339
		3340	=item w->start ()
		3341
		3342	Starts the watcher. Note that there is no C<loop> argument, as the
		3343	constructor already stores the event loop.
		3344
		3345	=item w->stop ()
		3346
		3347	Stops the watcher if it is active. Again, no C<loop> argument.
		3348
		3349	=item w->again () (C<ev::timer>, C<ev::periodic> only)
		3350
		3351	For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
		3352	C<ev_TYPE_again> function.
		3353
		3354	=item w->sweep () (C<ev::embed> only)
		3355
		3356	Invokes C<ev_embed_sweep>.
		3357
		3358	=item w->update () (C<ev::stat> only)
		3359
		3360	Invokes C<ev_stat_stat>.
		3361
		3362	=back
		3363
		3364	=back
		3365
		3366	Example: Define a class with an IO and idle watcher, start one of them in
		3367	the constructor.
		3368
		3369	class myclass
		3370	{
		3371	ev::io io ; void io_cb (ev::io &w, int revents);
		3372	ev::idle idle; void idle_cb (ev::idle &w, int revents);
		3373
		3374	myclass (int fd)
		3375	{
		3376	io .set <myclass, &myclass::io_cb > (this);
		3377	idle.set <myclass, &myclass::idle_cb> (this);
		3378
		3379	io.start (fd, ev::READ);
		3380	}
		3381	};
		3382
		3383
		3384	=head1 OTHER LANGUAGE BINDINGS
		3385
		3386	Libev does not offer other language bindings itself, but bindings for a
		3387	number of languages exist in the form of third-party packages. If you know
		3388	any interesting language binding in addition to the ones listed here, drop
		3389	me a note.
		3390
		3391	=over 4
		3392
		3393	=item Perl
		3394
		3395	The EV module implements the full libev API and is actually used to test
		3396	libev. EV is developed together with libev. Apart from the EV core module,
		3397	there are additional modules that implement libev-compatible interfaces
		3398	to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
		3399	C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
		3400	and C<EV::Glib>).
		3401
		3402	It can be found and installed via CPAN, its homepage is at
		3403	L<http://software.schmorp.de/pkg/EV>.
		3404
		3405	=item Python
		3406
		3407	Python bindings can be found at L<http://code.google.com/p/pyev/>. It
		3408	seems to be quite complete and well-documented.
		3409
		3410	=item Ruby
		3411
		3412	Tony Arcieri has written a ruby extension that offers access to a subset
		3413	of the libev API and adds file handle abstractions, asynchronous DNS and
		3414	more on top of it. It can be found via gem servers. Its homepage is at
		3415	L<http://rev.rubyforge.org/>.
		3416
		3417	Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
		3418	makes rev work even on mingw.
		3419
		3420	=item Haskell
		3421
		3422	A haskell binding to libev is available at
		3423	L<http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
		3424
		3425	=item D
		3426
		3427	Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
		3428	be found at L<http://proj.llucax.com.ar/wiki/evd>.
		3429
		3430	=item Ocaml
		3431
		3432	Erkki Seppala has written Ocaml bindings for libev, to be found at
		3433	L<http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
		3434
		3435	=item Lua
		3436
		3437	Brian Maher has written a partial interface to libev
		3438	for lua (only C<ev_io> and C<ev_timer>), to be found at
		3439	L<http://github.com/brimworks/lua-ev>.
		3440
		3441	=back
		3442
		3443
		3444	=head1 MACRO MAGIC
		3445
		3446	Libev can be compiled with a variety of options, the most fundamental
		3447	of which is C<EV_MULTIPLICITY>. This option determines whether (most)
		3448	functions and callbacks have an initial C<struct ev_loop *> argument.
		3449
		3450	To make it easier to write programs that cope with either variant, the
		3451	following macros are defined:
		3452
		3453	=over 4
		3454
		3455	=item C<EV_A>, C<EV_A_>
		3456
		3457	This provides the loop I<argument> for functions, if one is required ("ev
		3458	loop argument"). The C<EV_A> form is used when this is the sole argument,
		3459	C<EV_A_> is used when other arguments are following. Example:
		3460
		3461	ev_unref (EV_A);
		3462	ev_timer_add (EV_A_ watcher);
		3463	ev_loop (EV_A_ 0);
		3464
		3465	It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
		3466	which is often provided by the following macro.
		3467
		3468	=item C<EV_P>, C<EV_P_>
		3469
		3470	This provides the loop I<parameter> for functions, if one is required ("ev
		3471	loop parameter"). The C<EV_P> form is used when this is the sole parameter,
		3472	C<EV_P_> is used when other parameters are following. Example:
		3473
		3474	// this is how ev_unref is being declared
		3475	static void ev_unref (EV_P);
		3476
		3477	// this is how you can declare your typical callback
		3478	static void cb (EV_P_ ev_timer *w, int revents)
		3479
		3480	It declares a parameter C<loop> of type C<struct ev_loop *>, quite
		3481	suitable for use with C<EV_A>.
		3482
		3483	=item C<EV_DEFAULT>, C<EV_DEFAULT_>
		3484
		3485	Similar to the other two macros, this gives you the value of the default
		3486	loop, if multiple loops are supported ("ev loop default").
		3487
		3488	=item C<EV_DEFAULT_UC>, C<EV_DEFAULT_UC_>
		3489
		3490	Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
		3491	default loop has been initialised (C<UC> == unchecked). Their behaviour
		3492	is undefined when the default loop has not been initialised by a previous
		3493	execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
		3494
		3495	It is often prudent to use C<EV_DEFAULT> when initialising the first
		3496	watcher in a function but use C<EV_DEFAULT_UC> afterwards.
		3497
		3498	=back
		3499
		3500	Example: Declare and initialise a check watcher, utilising the above
		3501	macros so it will work regardless of whether multiple loops are supported
		3502	or not.
		3503
		3504	static void
		3505	check_cb (EV_P_ ev_timer *w, int revents)
		3506	{
		3507	ev_check_stop (EV_A_ w);
		3508	}
		3509
		3510	ev_check check;
		3511	ev_check_init (&check, check_cb);
		3512	ev_check_start (EV_DEFAULT_ &check);
		3513	ev_loop (EV_DEFAULT_ 0);
		3514
		3515	=head1 EMBEDDING
		3516
		3517	Libev can (and often is) directly embedded into host
		3518	applications. Examples of applications that embed it include the Deliantra
		3519	Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
		3520	and rxvt-unicode.
		3521
		3522	The goal is to enable you to just copy the necessary files into your
		3523	source directory without having to change even a single line in them, so
		3524	you can easily upgrade by simply copying (or having a checked-out copy of
		3525	libev somewhere in your source tree).
		3526
		3527	=head2 FILESETS
		3528
		3529	Depending on what features you need you need to include one or more sets of files
		3530	in your application.
		3531
		3532	=head3 CORE EVENT LOOP
		3533
		3534	To include only the libev core (all the C<ev_*> functions), with manual
		3535	configuration (no autoconf):
		3536
		3537	#define EV_STANDALONE 1
		3538	#include "ev.c"
		3539
		3540	This will automatically include F<ev.h>, too, and should be done in a
		3541	single C source file only to provide the function implementations. To use
		3542	it, do the same for F<ev.h> in all files wishing to use this API (best
		3543	done by writing a wrapper around F<ev.h> that you can include instead and
		3544	where you can put other configuration options):
		3545
		3546	#define EV_STANDALONE 1
		3547	#include "ev.h"
		3548
		3549	Both header files and implementation files can be compiled with a C++
		3550	compiler (at least, that's a stated goal, and breakage will be treated
		3551	as a bug).
		3552
		3553	You need the following files in your source tree, or in a directory
		3554	in your include path (e.g. in libev/ when using -Ilibev):
		3555
		3556	ev.h
		3557	ev.c
		3558	ev_vars.h
		3559	ev_wrap.h
		3560
		3561	ev_win32.c required on win32 platforms only
		3562
		3563	ev_select.c only when select backend is enabled (which is enabled by default)
		3564	ev_poll.c only when poll backend is enabled (disabled by default)
		3565	ev_epoll.c only when the epoll backend is enabled (disabled by default)
		3566	ev_kqueue.c only when the kqueue backend is enabled (disabled by default)
		3567	ev_port.c only when the solaris port backend is enabled (disabled by default)
		3568
		3569	F<ev.c> includes the backend files directly when enabled, so you only need
		3570	to compile this single file.
		3571
		3572	=head3 LIBEVENT COMPATIBILITY API
		3573
		3574	To include the libevent compatibility API, also include:
		3575
		3576	#include "event.c"
		3577
		3578	in the file including F<ev.c>, and:
		3579
		3580	#include "event.h"
		3581
		3582	in the files that want to use the libevent API. This also includes F<ev.h>.
		3583
		3584	You need the following additional files for this:
		3585
		3586	event.h
		3587	event.c
		3588
		3589	=head3 AUTOCONF SUPPORT
		3590
		3591	Instead of using C<EV_STANDALONE=1> and providing your configuration in
		3592	whatever way you want, you can also C<m4_include([libev.m4])> in your
		3593	F<configure.ac> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
		3594	include F<config.h> and configure itself accordingly.
		3595
		3596	For this of course you need the m4 file:
		3597
		3598	libev.m4
		3599
		3600	=head2 PREPROCESSOR SYMBOLS/MACROS
		3601
		3602	Libev can be configured via a variety of preprocessor symbols you have to
		3603	define before including any of its files. The default in the absence of
		3604	autoconf is documented for every option.
		3605
		3606	=over 4
		3607
		3608	=item EV_STANDALONE
		3609
		3610	Must always be C<1> if you do not use autoconf configuration, which
		3611	keeps libev from including F<config.h>, and it also defines dummy
		3612	implementations for some libevent functions (such as logging, which is not
		3613	supported). It will also not define any of the structs usually found in
		3614	F<event.h> that are not directly supported by the libev core alone.
		3615
		3616	In standalone mode, libev will still try to automatically deduce the
		3617	configuration, but has to be more conservative.
		3618
		3619	=item EV_USE_MONOTONIC
		3620
		3621	If defined to be C<1>, libev will try to detect the availability of the
		3622	monotonic clock option at both compile time and runtime. Otherwise no
		3623	use of the monotonic clock option will be attempted. If you enable this,
		3624	you usually have to link against librt or something similar. Enabling it
		3625	when the functionality isn't available is safe, though, although you have
		3626	to make sure you link against any libraries where the C<clock_gettime>
		3627	function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
		3628
		3629	=item EV_USE_REALTIME
		3630
		3631	If defined to be C<1>, libev will try to detect the availability of the
		3632	real-time clock option at compile time (and assume its availability
		3633	at runtime if successful). Otherwise no use of the real-time clock
		3634	option will be attempted. This effectively replaces C<gettimeofday>
		3635	by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
		3636	correctness. See the note about libraries in the description of
		3637	C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
		3638	C<EV_USE_CLOCK_SYSCALL>.
		3639
		3640	=item EV_USE_CLOCK_SYSCALL
		3641
		3642	If defined to be C<1>, libev will try to use a direct syscall instead
		3643	of calling the system-provided C<clock_gettime> function. This option
		3644	exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
		3645	unconditionally pulls in C<libpthread>, slowing down single-threaded
		3646	programs needlessly. Using a direct syscall is slightly slower (in
		3647	theory), because no optimised vdso implementation can be used, but avoids
		3648	the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
		3649	higher, as it simplifies linking (no need for C<-lrt>).
		3650
		3651	=item EV_USE_NANOSLEEP
		3652
		3653	If defined to be C<1>, libev will assume that C<nanosleep ()> is available
		3654	and will use it for delays. Otherwise it will use C<select ()>.
		3655
		3656	=item EV_USE_EVENTFD
		3657
		3658	If defined to be C<1>, then libev will assume that C<eventfd ()> is
		3659	available and will probe for kernel support at runtime. This will improve
		3660	C<ev_signal> and C<ev_async> performance and reduce resource consumption.
		3661	If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
		3662	2.7 or newer, otherwise disabled.
		3663
		3664	=item EV_USE_SELECT
		3665
		3666	If undefined or defined to be C<1>, libev will compile in support for the
		3667	C<select>(2) backend. No attempt at auto-detection will be done: if no
		3668	other method takes over, select will be it. Otherwise the select backend
		3669	will not be compiled in.
		3670
		3671	=item EV_SELECT_USE_FD_SET
		3672
		3673	If defined to C<1>, then the select backend will use the system C<fd_set>
		3674	structure. This is useful if libev doesn't compile due to a missing
		3675	C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
		3676	on exotic systems. This usually limits the range of file descriptors to
		3677	some low limit such as 1024 or might have other limitations (winsocket
		3678	only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
		3679	configures the maximum size of the C<fd_set>.
		3680
		3681	=item EV_SELECT_IS_WINSOCKET
		3682
		3683	When defined to C<1>, the select backend will assume that
		3684	select/socket/connect etc. don't understand file descriptors but
		3685	wants osf handles on win32 (this is the case when the select to
		3686	be used is the winsock select). This means that it will call
		3687	C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
		3688	it is assumed that all these functions actually work on fds, even
		3689	on win32. Should not be defined on non-win32 platforms.
		3690
		3691	=item EV_FD_TO_WIN32_HANDLE(fd)
		3692
		3693	If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
		3694	file descriptors to socket handles. When not defining this symbol (the
		3695	default), then libev will call C<_get_osfhandle>, which is usually
		3696	correct. In some cases, programs use their own file descriptor management,
		3697	in which case they can provide this function to map fds to socket handles.
		3698
		3699	=item EV_WIN32_HANDLE_TO_FD(handle)
		3700
		3701	If C<EV_SELECT_IS_WINSOCKET> then libev maps handles to file descriptors
		3702	using the standard C<_open_osfhandle> function. For programs implementing
		3703	their own fd to handle mapping, overwriting this function makes it easier
		3704	to do so. This can be done by defining this macro to an appropriate value.
		3705
		3706	=item EV_WIN32_CLOSE_FD(fd)
		3707
		3708	If programs implement their own fd to handle mapping on win32, then this
		3709	macro can be used to override the C<close> function, useful to unregister
		3710	file descriptors again. Note that the replacement function has to close
		3711	the underlying OS handle.
		3712
		3713	=item EV_USE_POLL
		3714
		3715	If defined to be C<1>, libev will compile in support for the C<poll>(2)
		3716	backend. Otherwise it will be enabled on non-win32 platforms. It
		3717	takes precedence over select.
		3718
		3719	=item EV_USE_EPOLL
		3720
		3721	If defined to be C<1>, libev will compile in support for the Linux
		3722	C<epoll>(7) backend. Its availability will be detected at runtime,
		3723	otherwise another method will be used as fallback. This is the preferred
		3724	backend for GNU/Linux systems. If undefined, it will be enabled if the
		3725	headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
		3726
		3727	=item EV_USE_KQUEUE
		3728
		3729	If defined to be C<1>, libev will compile in support for the BSD style
		3730	C<kqueue>(2) backend. Its actual availability will be detected at runtime,
		3731	otherwise another method will be used as fallback. This is the preferred
		3732	backend for BSD and BSD-like systems, although on most BSDs kqueue only
		3733	supports some types of fds correctly (the only platform we found that
		3734	supports ptys for example was NetBSD), so kqueue might be compiled in, but
		3735	not be used unless explicitly requested. The best way to use it is to find
		3736	out whether kqueue supports your type of fd properly and use an embedded
		3737	kqueue loop.
		3738
		3739	=item EV_USE_PORT
		3740
		3741	If defined to be C<1>, libev will compile in support for the Solaris
		3742	10 port style backend. Its availability will be detected at runtime,
		3743	otherwise another method will be used as fallback. This is the preferred
		3744	backend for Solaris 10 systems.
		3745
		3746	=item EV_USE_DEVPOLL
		3747
		3748	Reserved for future expansion, works like the USE symbols above.
		3749
		3750	=item EV_USE_INOTIFY
		3751
		3752	If defined to be C<1>, libev will compile in support for the Linux inotify
		3753	interface to speed up C<ev_stat> watchers. Its actual availability will
		3754	be detected at runtime. If undefined, it will be enabled if the headers
		3755	indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
		3756
		3757	=item EV_ATOMIC_T
		3758
		3759	Libev requires an integer type (suitable for storing C<0> or C<1>) whose
		3760	access is atomic with respect to other threads or signal contexts. No such
		3761	type is easily found in the C language, so you can provide your own type
		3762	that you know is safe for your purposes. It is used both for signal handler "locking"
		3763	as well as for signal and thread safety in C<ev_async> watchers.
		3764
		3765	In the absence of this define, libev will use C<sig_atomic_t volatile>
		3766	(from F<signal.h>), which is usually good enough on most platforms.
		3767
		3768	=item EV_H
		3769
		3770	The name of the F<ev.h> header file used to include it. The default if
		3771	undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
		3772	used to virtually rename the F<ev.h> header file in case of conflicts.
		3773
		3774	=item EV_CONFIG_H
		3775
		3776	If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
		3777	F<ev.c>'s idea of where to find the F<config.h> file, similarly to
		3778	C<EV_H>, above.
		3779
		3780	=item EV_EVENT_H
		3781
		3782	Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
		3783	of how the F<event.h> header can be found, the default is C<"event.h">.
		3784
		3785	=item EV_PROTOTYPES
		3786
		3787	If defined to be C<0>, then F<ev.h> will not define any function
		3788	prototypes, but still define all the structs and other symbols. This is
		3789	occasionally useful if you want to provide your own wrapper functions
		3790	around libev functions.
		3791
		3792	=item EV_MULTIPLICITY
		3793
		3794	If undefined or defined to C<1>, then all event-loop-specific functions
		3795	will have the C<struct ev_loop *> as first argument, and you can create
		3796	additional independent event loops. Otherwise there will be no support
		3797	for multiple event loops and there is no first event loop pointer
		3798	argument. Instead, all functions act on the single default loop.
		3799
		3800	=item EV_MINPRI
		3801
		3802	=item EV_MAXPRI
		3803
		3804	The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
		3805	C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
		3806	provide for more priorities by overriding those symbols (usually defined
		3807	to be C<-2> and C<2>, respectively).
		3808
		3809	When doing priority-based operations, libev usually has to linearly search
		3810	all the priorities, so having many of them (hundreds) uses a lot of space
		3811	and time, so using the defaults of five priorities (-2 .. +2) is usually
		3812	fine.
		3813
		3814	If your embedding application does not need any priorities, defining these
		3815	both to C<0> will save some memory and CPU.
		3816
		3817	=item EV_PERIODIC_ENABLE
		3818
		3819	If undefined or defined to be C<1>, then periodic timers are supported. If
		3820	defined to be C<0>, then they are not. Disabling them saves a few kB of
		3821	code.
		3822
		3823	=item EV_IDLE_ENABLE
		3824
		3825	If undefined or defined to be C<1>, then idle watchers are supported. If
		3826	defined to be C<0>, then they are not. Disabling them saves a few kB of
		3827	code.
		3828
		3829	=item EV_EMBED_ENABLE
		3830
		3831	If undefined or defined to be C<1>, then embed watchers are supported. If
		3832	defined to be C<0>, then they are not. Embed watchers rely on most other
		3833	watcher types, which therefore must not be disabled.
		3834
		3835	=item EV_STAT_ENABLE
		3836
		3837	If undefined or defined to be C<1>, then stat watchers are supported. If
		3838	defined to be C<0>, then they are not.
		3839
		3840	=item EV_FORK_ENABLE
		3841
		3842	If undefined or defined to be C<1>, then fork watchers are supported. If
		3843	defined to be C<0>, then they are not.
		3844
		3845	=item EV_ASYNC_ENABLE
		3846
		3847	If undefined or defined to be C<1>, then async watchers are supported. If
		3848	defined to be C<0>, then they are not.
		3849
		3850	=item EV_MINIMAL
		3851
		3852	If you need to shave off some kilobytes of code at the expense of some
		3853	speed (but with the full API), define this symbol to C<1>. Currently this
		3854	is used to override some inlining decisions, saves roughly 30% code size
		3855	on amd64. It also selects a much smaller 2-heap for timer management over
		3856	the default 4-heap.
		3857
		3858	You can save even more by disabling watcher types you do not need
		3859	and setting C<EV_MAXPRI> == C<EV_MINPRI>. Also, disabling C<assert>
		3860	(C<-DNDEBUG>) will usually reduce code size a lot.
		3861
		3862	Defining C<EV_MINIMAL> to C<2> will additionally reduce the core API to
		3863	provide a bare-bones event library. See C<ev.h> for details on what parts
		3864	of the API are still available, and do not complain if this subset changes
		3865	over time.
		3866
		3867	=item EV_NSIG
		3868
		3869	The highest supported signal number, +1 (or, the number of
		3870	signals): Normally, libev tries to deduce the maximum number of signals
		3871	automatically, but sometimes this fails, in which case it can be
		3872	specified. Also, using a lower number than detected (C<32> should be
		3873	good for about any system in existance) can save some memory, as libev
		3874	statically allocates some 12-24 bytes per signal number.
		3875
		3876	=item EV_PID_HASHSIZE
		3877
		3878	C<ev_child> watchers use a small hash table to distribute workload by
		3879	pid. The default size is C<16> (or C<1> with C<EV_MINIMAL>), usually more
		3880	than enough. If you need to manage thousands of children you might want to
		3881	increase this value (I<must> be a power of two).
		3882
		3883	=item EV_INOTIFY_HASHSIZE
		3884
		3885	C<ev_stat> watchers use a small hash table to distribute workload by
		3886	inotify watch id. The default size is C<16> (or C<1> with C<EV_MINIMAL>),
		3887	usually more than enough. If you need to manage thousands of C<ev_stat>
		3888	watchers you might want to increase this value (I<must> be a power of
		3889	two).
		3890
		3891	=item EV_USE_4HEAP
		3892
		3893	Heaps are not very cache-efficient. To improve the cache-efficiency of the
		3894	timer and periodics heaps, libev uses a 4-heap when this symbol is defined
		3895	to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
		3896	faster performance with many (thousands) of watchers.
		3897
		3898	The default is C<1> unless C<EV_MINIMAL> is set in which case it is C<0>
		3899	(disabled).
		3900
		3901	=item EV_HEAP_CACHE_AT
		3902
		3903	Heaps are not very cache-efficient. To improve the cache-efficiency of the
		3904	timer and periodics heaps, libev can cache the timestamp (I<at>) within
		3905	the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
		3906	which uses 8-12 bytes more per watcher and a few hundred bytes more code,
		3907	but avoids random read accesses on heap changes. This improves performance
		3908	noticeably with many (hundreds) of watchers.
		3909
		3910	The default is C<1> unless C<EV_MINIMAL> is set in which case it is C<0>
		3911	(disabled).
		3912
		3913	=item EV_VERIFY
		3914
		3915	Controls how much internal verification (see C<ev_loop_verify ()>) will
		3916	be done: If set to C<0>, no internal verification code will be compiled
		3917	in. If set to C<1>, then verification code will be compiled in, but not
		3918	called. If set to C<2>, then the internal verification code will be
		3919	called once per loop, which can slow down libev. If set to C<3>, then the
		3920	verification code will be called very frequently, which will slow down
		3921	libev considerably.
		3922
		3923	The default is C<1>, unless C<EV_MINIMAL> is set, in which case it will be
		3924	C<0>.
		3925
		3926	=item EV_COMMON
		3927
		3928	By default, all watchers have a C<void *data> member. By redefining
		3929	this macro to a something else you can include more and other types of
		3930	members. You have to define it each time you include one of the files,
		3931	though, and it must be identical each time.
		3932
		3933	For example, the perl EV module uses something like this:
		3934
		3935	#define EV_COMMON \
		3936	SV *self; /* contains this struct */ \
		3937	SV *cb_sv, *fh /* note no trailing ";" */
		3938
		3939	=item EV_CB_DECLARE (type)
		3940
		3941	=item EV_CB_INVOKE (watcher, revents)
		3942
		3943	=item ev_set_cb (ev, cb)
		3944
		3945	Can be used to change the callback member declaration in each watcher,
		3946	and the way callbacks are invoked and set. Must expand to a struct member
		3947	definition and a statement, respectively. See the F<ev.h> header file for
		3948	their default definitions. One possible use for overriding these is to
		3949	avoid the C<struct ev_loop *> as first argument in all cases, or to use
		3950	method calls instead of plain function calls in C++.
		3951
		3952	=back
		3953
		3954	=head2 EXPORTED API SYMBOLS
		3955
		3956	If you need to re-export the API (e.g. via a DLL) and you need a list of
		3957	exported symbols, you can use the provided F<Symbol.*> files which list
		3958	all public symbols, one per line:
		3959
		3960	Symbols.ev for libev proper
		3961	Symbols.event for the libevent emulation
		3962
		3963	This can also be used to rename all public symbols to avoid clashes with
		3964	multiple versions of libev linked together (which is obviously bad in
		3965	itself, but sometimes it is inconvenient to avoid this).
		3966
		3967	A sed command like this will create wrapper C<#define>'s that you need to
		3968	include before including F<ev.h>:
		3969
		3970	<Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
		3971
		3972	This would create a file F<wrap.h> which essentially looks like this:
		3973
		3974	#define ev_backend myprefix_ev_backend
		3975	#define ev_check_start myprefix_ev_check_start
		3976	#define ev_check_stop myprefix_ev_check_stop
		3977	...
		3978
		3979	=head2 EXAMPLES
		3980
		3981	For a real-world example of a program the includes libev
		3982	verbatim, you can have a look at the EV perl module
		3983	(L<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
		3984	the F<libev/> subdirectory and includes them in the F<EV/EVAPI.h> (public
		3985	interface) and F<EV.xs> (implementation) files. Only the F<EV.xs> file
		3986	will be compiled. It is pretty complex because it provides its own header
		3987	file.
		3988
		3989	The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
		3990	that everybody includes and which overrides some configure choices:
		3991
		3992	#define EV_MINIMAL 1
		3993	#define EV_USE_POLL 0
		3994	#define EV_MULTIPLICITY 0
		3995	#define EV_PERIODIC_ENABLE 0
		3996	#define EV_STAT_ENABLE 0
		3997	#define EV_FORK_ENABLE 0
		3998	#define EV_CONFIG_H <config.h>
		3999	#define EV_MINPRI 0
		4000	#define EV_MAXPRI 0
		4001
		4002	#include "ev++.h"
		4003
		4004	And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
		4005
		4006	#include "ev_cpp.h"
		4007	#include "ev.c"
		4008
		4009	=head1 INTERACTION WITH OTHER PROGRAMS OR LIBRARIES
		4010
		4011	=head2 THREADS AND COROUTINES
		4012
		4013	=head3 THREADS
		4014
		4015	All libev functions are reentrant and thread-safe unless explicitly
		4016	documented otherwise, but libev implements no locking itself. This means
		4017	that you can use as many loops as you want in parallel, as long as there
		4018	are no concurrent calls into any libev function with the same loop
		4019	parameter (C<ev_default_*> calls have an implicit default loop parameter,
		4020	of course): libev guarantees that different event loops share no data
		4021	structures that need any locking.
		4022
		4023	Or to put it differently: calls with different loop parameters can be done
		4024	concurrently from multiple threads, calls with the same loop parameter
		4025	must be done serially (but can be done from different threads, as long as
		4026	only one thread ever is inside a call at any point in time, e.g. by using
		4027	a mutex per loop).
		4028
		4029	Specifically to support threads (and signal handlers), libev implements
		4030	so-called C<ev_async> watchers, which allow some limited form of
		4031	concurrency on the same event loop, namely waking it up "from the
		4032	outside".
		4033
		4034	If you want to know which design (one loop, locking, or multiple loops
		4035	without or something else still) is best for your problem, then I cannot
		4036	help you, but here is some generic advice:
		4037
		4038	=over 4
		4039
		4040	=item * most applications have a main thread: use the default libev loop
		4041	in that thread, or create a separate thread running only the default loop.
		4042
		4043	This helps integrating other libraries or software modules that use libev
		4044	themselves and don't care/know about threading.
		4045
		4046	=item * one loop per thread is usually a good model.
		4047
		4048	Doing this is almost never wrong, sometimes a better-performance model
		4049	exists, but it is always a good start.
		4050
		4051	=item * other models exist, such as the leader/follower pattern, where one
		4052	loop is handed through multiple threads in a kind of round-robin fashion.
		4053
		4054	Choosing a model is hard - look around, learn, know that usually you can do
		4055	better than you currently do :-)
		4056
		4057	=item * often you need to talk to some other thread which blocks in the
		4058	event loop.
		4059
		4060	C<ev_async> watchers can be used to wake them up from other threads safely
		4061	(or from signal contexts...).
		4062
		4063	An example use would be to communicate signals or other events that only
		4064	work in the default loop by registering the signal watcher with the
		4065	default loop and triggering an C<ev_async> watcher from the default loop
		4066	watcher callback into the event loop interested in the signal.
		4067
		4068	=back
		4069
		4070	=head4 THREAD LOCKING EXAMPLE
		4071
		4072	Here is a fictitious example of how to run an event loop in a different
		4073	thread than where callbacks are being invoked and watchers are
		4074	created/added/removed.
		4075
		4076	For a real-world example, see the C<EV::Loop::Async> perl module,
		4077	which uses exactly this technique (which is suited for many high-level
		4078	languages).
		4079
		4080	The example uses a pthread mutex to protect the loop data, a condition
		4081	variable to wait for callback invocations, an async watcher to notify the
		4082	event loop thread and an unspecified mechanism to wake up the main thread.
		4083
		4084	First, you need to associate some data with the event loop:
		4085
		4086	typedef struct {
		4087	mutex_t lock; /* global loop lock */
		4088	ev_async async_w;
		4089	thread_t tid;
		4090	cond_t invoke_cv;
		4091	} userdata;
		4092
		4093	void prepare_loop (EV_P)
		4094	{
		4095	// for simplicity, we use a static userdata struct.
		4096	static userdata u;
		4097
		4098	ev_async_init (&u->async_w, async_cb);
		4099	ev_async_start (EV_A_ &u->async_w);
		4100
		4101	pthread_mutex_init (&u->lock, 0);
		4102	pthread_cond_init (&u->invoke_cv, 0);
		4103
		4104	// now associate this with the loop
		4105	ev_set_userdata (EV_A_ u);
		4106	ev_set_invoke_pending_cb (EV_A_ l_invoke);
		4107	ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
		4108
		4109	// then create the thread running ev_loop
		4110	pthread_create (&u->tid, 0, l_run, EV_A);
		4111	}
		4112
		4113	The callback for the C<ev_async> watcher does nothing: the watcher is used
		4114	solely to wake up the event loop so it takes notice of any new watchers
		4115	that might have been added:
		4116
		4117	static void
		4118	async_cb (EV_P_ ev_async *w, int revents)
		4119	{
		4120	// just used for the side effects
		4121	}
		4122
		4123	The C<l_release> and C<l_acquire> callbacks simply unlock/lock the mutex
		4124	protecting the loop data, respectively.
		4125
		4126	static void
		4127	l_release (EV_P)
		4128	{
		4129	userdata *u = ev_userdata (EV_A);
		4130	pthread_mutex_unlock (&u->lock);
		4131	}
		4132
		4133	static void
		4134	l_acquire (EV_P)
		4135	{
		4136	userdata *u = ev_userdata (EV_A);
		4137	pthread_mutex_lock (&u->lock);
		4138	}
		4139
		4140	The event loop thread first acquires the mutex, and then jumps straight
		4141	into C<ev_loop>:
		4142
		4143	void *
		4144	l_run (void *thr_arg)
		4145	{
		4146	struct ev_loop *loop = (struct ev_loop *)thr_arg;
		4147
		4148	l_acquire (EV_A);
		4149	pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
		4150	ev_loop (EV_A_ 0);
		4151	l_release (EV_A);
		4152
		4153	return 0;
		4154	}
		4155
		4156	Instead of invoking all pending watchers, the C<l_invoke> callback will
		4157	signal the main thread via some unspecified mechanism (signals? pipe
		4158	writes? C<Async::Interrupt>?) and then waits until all pending watchers
		4159	have been called (in a while loop because a) spurious wakeups are possible
		4160	and b) skipping inter-thread-communication when there are no pending
		4161	watchers is very beneficial):
		4162
		4163	static void
		4164	l_invoke (EV_P)
		4165	{
		4166	userdata *u = ev_userdata (EV_A);
		4167
		4168	while (ev_pending_count (EV_A))
		4169	{
		4170	wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
		4171	pthread_cond_wait (&u->invoke_cv, &u->lock);
		4172	}
		4173	}
		4174
		4175	Now, whenever the main thread gets told to invoke pending watchers, it
		4176	will grab the lock, call C<ev_invoke_pending> and then signal the loop
		4177	thread to continue:
		4178
		4179	static void
		4180	real_invoke_pending (EV_P)
		4181	{
		4182	userdata *u = ev_userdata (EV_A);
		4183
		4184	pthread_mutex_lock (&u->lock);
		4185	ev_invoke_pending (EV_A);
		4186	pthread_cond_signal (&u->invoke_cv);
		4187	pthread_mutex_unlock (&u->lock);
		4188	}
		4189
		4190	Whenever you want to start/stop a watcher or do other modifications to an
		4191	event loop, you will now have to lock:
		4192
		4193	ev_timer timeout_watcher;
		4194	userdata *u = ev_userdata (EV_A);
		4195
		4196	ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
		4197
		4198	pthread_mutex_lock (&u->lock);
		4199	ev_timer_start (EV_A_ &timeout_watcher);
		4200	ev_async_send (EV_A_ &u->async_w);
		4201	pthread_mutex_unlock (&u->lock);
		4202
		4203	Note that sending the C<ev_async> watcher is required because otherwise
		4204	an event loop currently blocking in the kernel will have no knowledge
		4205	about the newly added timer. By waking up the loop it will pick up any new
		4206	watchers in the next event loop iteration.
		4207
		4208	=head3 COROUTINES
		4209
		4210	Libev is very accommodating to coroutines ("cooperative threads"):
		4211	libev fully supports nesting calls to its functions from different
		4212	coroutines (e.g. you can call C<ev_loop> on the same loop from two
		4213	different coroutines, and switch freely between both coroutines running
		4214	the loop, as long as you don't confuse yourself). The only exception is
		4215	that you must not do this from C<ev_periodic> reschedule callbacks.
		4216
		4217	Care has been taken to ensure that libev does not keep local state inside
		4218	C<ev_loop>, and other calls do not usually allow for coroutine switches as
		4219	they do not call any callbacks.
		4220
		4221	=head2 COMPILER WARNINGS
		4222
		4223	Depending on your compiler and compiler settings, you might get no or a
		4224	lot of warnings when compiling libev code. Some people are apparently
		4225	scared by this.
		4226
		4227	However, these are unavoidable for many reasons. For one, each compiler
		4228	has different warnings, and each user has different tastes regarding
		4229	warning options. "Warn-free" code therefore cannot be a goal except when
		4230	targeting a specific compiler and compiler-version.
		4231
		4232	Another reason is that some compiler warnings require elaborate
		4233	workarounds, or other changes to the code that make it less clear and less
		4234	maintainable.
		4235
		4236	And of course, some compiler warnings are just plain stupid, or simply
		4237	wrong (because they don't actually warn about the condition their message
		4238	seems to warn about). For example, certain older gcc versions had some
		4239	warnings that resulted an extreme number of false positives. These have
		4240	been fixed, but some people still insist on making code warn-free with
		4241	such buggy versions.
		4242
		4243	While libev is written to generate as few warnings as possible,
		4244	"warn-free" code is not a goal, and it is recommended not to build libev
		4245	with any compiler warnings enabled unless you are prepared to cope with
		4246	them (e.g. by ignoring them). Remember that warnings are just that:
		4247	warnings, not errors, or proof of bugs.
		4248
		4249
		4250	=head2 VALGRIND
		4251
		4252	Valgrind has a special section here because it is a popular tool that is
		4253	highly useful. Unfortunately, valgrind reports are very hard to interpret.
		4254
		4255	If you think you found a bug (memory leak, uninitialised data access etc.)
		4256	in libev, then check twice: If valgrind reports something like:
		4257
		4258	==2274== definitely lost: 0 bytes in 0 blocks.
		4259	==2274== possibly lost: 0 bytes in 0 blocks.
		4260	==2274== still reachable: 256 bytes in 1 blocks.
		4261
		4262	Then there is no memory leak, just as memory accounted to global variables
		4263	is not a memleak - the memory is still being referenced, and didn't leak.
		4264
		4265	Similarly, under some circumstances, valgrind might report kernel bugs
		4266	as if it were a bug in libev (e.g. in realloc or in the poll backend,
		4267	although an acceptable workaround has been found here), or it might be
		4268	confused.
		4269
		4270	Keep in mind that valgrind is a very good tool, but only a tool. Don't
		4271	make it into some kind of religion.
		4272
		4273	If you are unsure about something, feel free to contact the mailing list
		4274	with the full valgrind report and an explanation on why you think this
		4275	is a bug in libev (best check the archives, too :). However, don't be
		4276	annoyed when you get a brisk "this is no bug" answer and take the chance
		4277	of learning how to interpret valgrind properly.
		4278
		4279	If you need, for some reason, empty reports from valgrind for your project
		4280	I suggest using suppression lists.
		4281
		4282
		4283	=head1 PORTABILITY NOTES
		4284
		4285	=head2 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
		4286
		4287	Win32 doesn't support any of the standards (e.g. POSIX) that libev
		4288	requires, and its I/O model is fundamentally incompatible with the POSIX
		4289	model. Libev still offers limited functionality on this platform in
		4290	the form of the C<EVBACKEND_SELECT> backend, and only supports socket
		4291	descriptors. This only applies when using Win32 natively, not when using
		4292	e.g. cygwin.
		4293
		4294	Lifting these limitations would basically require the full
		4295	re-implementation of the I/O system. If you are into these kinds of
		4296	things, then note that glib does exactly that for you in a very portable
		4297	way (note also that glib is the slowest event library known to man).
		4298
		4299	There is no supported compilation method available on windows except
		4300	embedding it into other applications.
		4301
		4302	Sensible signal handling is officially unsupported by Microsoft - libev
		4303	tries its best, but under most conditions, signals will simply not work.
		4304
		4305	Not a libev limitation but worth mentioning: windows apparently doesn't
		4306	accept large writes: instead of resulting in a partial write, windows will
		4307	either accept everything or return C<ENOBUFS> if the buffer is too large,
		4308	so make sure you only write small amounts into your sockets (less than a
		4309	megabyte seems safe, but this apparently depends on the amount of memory
		4310	available).
		4311
		4312	Due to the many, low, and arbitrary limits on the win32 platform and
		4313	the abysmal performance of winsockets, using a large number of sockets
		4314	is not recommended (and not reasonable). If your program needs to use
		4315	more than a hundred or so sockets, then likely it needs to use a totally
		4316	different implementation for windows, as libev offers the POSIX readiness
		4317	notification model, which cannot be implemented efficiently on windows
		4318	(due to Microsoft monopoly games).
		4319
		4320	A typical way to use libev under windows is to embed it (see the embedding
		4321	section for details) and use the following F<evwrap.h> header file instead
		4322	of F<ev.h>:
		4323
		4324	#define EV_STANDALONE /* keeps ev from requiring config.h */
		4325	#define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
		4326
		4327	#include "ev.h"
		4328
		4329	And compile the following F<evwrap.c> file into your project (make sure
		4330	you do I<not> compile the F<ev.c> or any other embedded source files!):
		4331
		4332	#include "evwrap.h"
		4333	#include "ev.c"
		4334
		4335	=over 4
		4336
		4337	=item The winsocket select function
		4338
		4339	The winsocket C<select> function doesn't follow POSIX in that it
		4340	requires socket I<handles> and not socket I<file descriptors> (it is
		4341	also extremely buggy). This makes select very inefficient, and also
		4342	requires a mapping from file descriptors to socket handles (the Microsoft
		4343	C runtime provides the function C<_open_osfhandle> for this). See the
		4344	discussion of the C<EV_SELECT_USE_FD_SET>, C<EV_SELECT_IS_WINSOCKET> and
		4345	C<EV_FD_TO_WIN32_HANDLE> preprocessor symbols for more info.
		4346
		4347	The configuration for a "naked" win32 using the Microsoft runtime
		4348	libraries and raw winsocket select is:
		4349
		4350	#define EV_USE_SELECT 1
		4351	#define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
		4352
		4353	Note that winsockets handling of fd sets is O(n), so you can easily get a
		4354	complexity in the O(n²) range when using win32.
		4355
		4356	=item Limited number of file descriptors
		4357
		4358	Windows has numerous arbitrary (and low) limits on things.
		4359
		4360	Early versions of winsocket's select only supported waiting for a maximum
		4361	of C<64> handles (probably owning to the fact that all windows kernels
		4362	can only wait for C<64> things at the same time internally; Microsoft
		4363	recommends spawning a chain of threads and wait for 63 handles and the
		4364	previous thread in each. Sounds great!).
		4365
		4366	Newer versions support more handles, but you need to define C<FD_SETSIZE>
		4367	to some high number (e.g. C<2048>) before compiling the winsocket select
		4368	call (which might be in libev or elsewhere, for example, perl and many
		4369	other interpreters do their own select emulation on windows).
		4370
		4371	Another limit is the number of file descriptors in the Microsoft runtime
		4372	libraries, which by default is C<64> (there must be a hidden I<64>
		4373	fetish or something like this inside Microsoft). You can increase this
		4374	by calling C<_setmaxstdio>, which can increase this limit to C<2048>
		4375	(another arbitrary limit), but is broken in many versions of the Microsoft
		4376	runtime libraries. This might get you to about C<512> or C<2048> sockets
		4377	(depending on windows version and/or the phase of the moon). To get more,
		4378	you need to wrap all I/O functions and provide your own fd management, but
		4379	the cost of calling select (O(n²)) will likely make this unworkable.
		4380
		4381	=back
		4382
		4383	=head2 PORTABILITY REQUIREMENTS
		4384
		4385	In addition to a working ISO-C implementation and of course the
		4386	backend-specific APIs, libev relies on a few additional extensions:
		4387
		4388	=over 4
		4389
		4390	=item C<void (*)(ev_watcher_type *, int revents)> must have compatible
		4391	calling conventions regardless of C<ev_watcher_type *>.
		4392
		4393	Libev assumes not only that all watcher pointers have the same internal
		4394	structure (guaranteed by POSIX but not by ISO C for example), but it also
		4395	assumes that the same (machine) code can be used to call any watcher
		4396	callback: The watcher callbacks have different type signatures, but libev
		4397	calls them using an C<ev_watcher *> internally.
		4398
		4399	=item C<sig_atomic_t volatile> must be thread-atomic as well
		4400
		4401	The type C<sig_atomic_t volatile> (or whatever is defined as
		4402	C<EV_ATOMIC_T>) must be atomic with respect to accesses from different
		4403	threads. This is not part of the specification for C<sig_atomic_t>, but is
		4404	believed to be sufficiently portable.
		4405
		4406	=item C<sigprocmask> must work in a threaded environment
		4407
		4408	Libev uses C<sigprocmask> to temporarily block signals. This is not
		4409	allowed in a threaded program (C<pthread_sigmask> has to be used). Typical
		4410	pthread implementations will either allow C<sigprocmask> in the "main
		4411	thread" or will block signals process-wide, both behaviours would
		4412	be compatible with libev. Interaction between C<sigprocmask> and
		4413	C<pthread_sigmask> could complicate things, however.
		4414
		4415	The most portable way to handle signals is to block signals in all threads
		4416	except the initial one, and run the default loop in the initial thread as
		4417	well.
		4418
		4419	=item C<long> must be large enough for common memory allocation sizes
		4420
		4421	To improve portability and simplify its API, libev uses C<long> internally
		4422	instead of C<size_t> when allocating its data structures. On non-POSIX
		4423	systems (Microsoft...) this might be unexpectedly low, but is still at
		4424	least 31 bits everywhere, which is enough for hundreds of millions of
		4425	watchers.
		4426
		4427	=item C<double> must hold a time value in seconds with enough accuracy
		4428
		4429	The type C<double> is used to represent timestamps. It is required to
		4430	have at least 51 bits of mantissa (and 9 bits of exponent), which is good
		4431	enough for at least into the year 4000. This requirement is fulfilled by
		4432	implementations implementing IEEE 754, which is basically all existing
		4433	ones. With IEEE 754 doubles, you get microsecond accuracy until at least
		4434	2200.
		4435
		4436	=back
		4437
		4438	If you know of other additional requirements drop me a note.
		4439
		4440
		4441	=head1 ALGORITHMIC COMPLEXITIES
		4442
		4443	In this section the complexities of (many of) the algorithms used inside
		4444	libev will be documented. For complexity discussions about backends see
		4445	the documentation for C<ev_default_init>.
		4446
		4447	All of the following are about amortised time: If an array needs to be
		4448	extended, libev needs to realloc and move the whole array, but this
		4449	happens asymptotically rarer with higher number of elements, so O(1) might
		4450	mean that libev does a lengthy realloc operation in rare cases, but on
		4451	average it is much faster and asymptotically approaches constant time.
		4452
		4453	=over 4
		4454
		4455	=item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
		4456
		4457	This means that, when you have a watcher that triggers in one hour and
		4458	there are 100 watchers that would trigger before that, then inserting will
		4459	have to skip roughly seven (C<ld 100>) of these watchers.
		4460
		4461	=item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
		4462
		4463	That means that changing a timer costs less than removing/adding them,
		4464	as only the relative motion in the event queue has to be paid for.
		4465
		4466	=item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
		4467
		4468	These just add the watcher into an array or at the head of a list.
		4469
		4470	=item Stopping check/prepare/idle/fork/async watchers: O(1)
		4471
		4472	=item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
		4473
		4474	These watchers are stored in lists, so they need to be walked to find the
		4475	correct watcher to remove. The lists are usually short (you don't usually
		4476	have many watchers waiting for the same fd or signal: one is typical, two
		4477	is rare).
		4478
		4479	=item Finding the next timer in each loop iteration: O(1)
		4480
		4481	By virtue of using a binary or 4-heap, the next timer is always found at a
		4482	fixed position in the storage array.
		4483
		4484	=item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
		4485
		4486	A change means an I/O watcher gets started or stopped, which requires
		4487	libev to recalculate its status (and possibly tell the kernel, depending
		4488	on backend and whether C<ev_io_set> was used).
		4489
		4490	=item Activating one watcher (putting it into the pending state): O(1)
		4491
		4492	=item Priority handling: O(number_of_priorities)
		4493
		4494	Priorities are implemented by allocating some space for each
		4495	priority. When doing priority-based operations, libev usually has to
		4496	linearly search all the priorities, but starting/stopping and activating
		4497	watchers becomes O(1) with respect to priority handling.
		4498
		4499	=item Sending an ev_async: O(1)
		4500
		4501	=item Processing ev_async_send: O(number_of_async_watchers)
		4502
		4503	=item Processing signals: O(max_signal_number)
		4504
		4505	Sending involves a system call I<iff> there were no other C<ev_async_send>
		4506	calls in the current loop iteration. Checking for async and signal events
		4507	involves iterating over all running async watchers or all signal numbers.
		4508
		4509	=back
		4510
		4511
		4512	=head1 GLOSSARY
		4513
		4514	=over 4
		4515
		4516	=item active
		4517
		4518	A watcher is active as long as it has been started (has been attached to
		4519	an event loop) but not yet stopped (disassociated from the event loop).
		4520
		4521	=item application
		4522
		4523	In this document, an application is whatever is using libev.
		4524
		4525	=item callback
		4526
		4527	The address of a function that is called when some event has been
		4528	detected. Callbacks are being passed the event loop, the watcher that
		4529	received the event, and the actual event bitset.
		4530
		4531	=item callback invocation
		4532
		4533	The act of calling the callback associated with a watcher.
		4534
		4535	=item event
		4536
		4537	A change of state of some external event, such as data now being available
		4538	for reading on a file descriptor, time having passed or simply not having
		4539	any other events happening anymore.
		4540
		4541	In libev, events are represented as single bits (such as C<EV_READ> or
		4542	C<EV_TIMEOUT>).
		4543
		4544	=item event library
		4545
		4546	A software package implementing an event model and loop.
		4547
		4548	=item event loop
		4549
		4550	An entity that handles and processes external events and converts them
		4551	into callback invocations.
		4552
		4553	=item event model
		4554
		4555	The model used to describe how an event loop handles and processes
		4556	watchers and events.
		4557
		4558	=item pending
		4559
		4560	A watcher is pending as soon as the corresponding event has been detected,
		4561	and stops being pending as soon as the watcher will be invoked or its
		4562	pending status is explicitly cleared by the application.
		4563
		4564	A watcher can be pending, but not active. Stopping a watcher also clears
		4565	its pending status.
		4566
		4567	=item real time
		4568
		4569	The physical time that is observed. It is apparently strictly monotonic :)
		4570
		4571	=item wall-clock time
		4572
		4573	The time and date as shown on clocks. Unlike real time, it can actually
		4574	be wrong and jump forwards and backwards, e.g. when the you adjust your
		4575	clock.
		4576
		4577	=item watcher
		4578
		4579	A data structure that describes interest in certain events. Watchers need
		4580	to be started (attached to an event loop) before they can receive events.
		4581
		4582	=item watcher invocation
		4583
		4584	The act of calling the callback associated with a watcher.
		4585
		4586	=back
		4587
742	=head1 AUTHOR	4588	=head1 AUTHOR
743		4589
744	Marc Lehmann <libev@schmorp.de>.	4590	Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael Magnusson.
745		4591

Diff Legend

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