ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libev/ev.pod

Comparing libev/ev.pod (file contents):
Revision 1.22 by root, Mon Nov 12 18:36:42 2007 UTC vs.
Revision 1.306 by root, Mon Oct 18 07:36:05 2010 UTC

…		…
2		2
3	libev - a high performance full-featured event loop written in C	3	libev - a high performance full-featured event loop written in C
4		4
5	=head1 SYNOPSIS	5	=head1 SYNOPSIS
6		6
7	#include <ev.h>	7	#include <ev.h>
8		8
9	=head1 DESCRIPTION	9	=head2 EXAMPLE PROGRAM
		10
		11	// a single header file is required
		12	#include <ev.h>
		13
		14	#include <stdio.h> // for puts
		15
		16	// every watcher type has its own typedef'd struct
		17	// with the name ev_TYPE
		18	ev_io stdin_watcher;
		19	ev_timer timeout_watcher;
		20
		21	// all watcher callbacks have a similar signature
		22	// this callback is called when data is readable on stdin
		23	static void
		24	stdin_cb (EV_P_ ev_io *w, int revents)
		25	{
		26	puts ("stdin ready");
		27	// for one-shot events, one must manually stop the watcher
		28	// with its corresponding stop function.
		29	ev_io_stop (EV_A_ w);
		30
		31	// this causes all nested ev_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	Familiarity 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<struct ev_loop *>) will not have
42	will not have this argument.	124	this argument.
43		125
44	=head1 TIME REPRESENTATION	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 (in practise
48	the beginning of 1970, details are complicated, don't ask). This type is	130	somewhere near the beginning of 1970, details are complicated, don't
49	called C<ev_tstamp>, which is what you should use too. It usually aliases	131	ask). This type is called C<ev_tstamp>, which is what you should use
50	to the double type in C.	132	too. It usually aliases to the C<double> type in C. When you need to do
		133	any calculations on it, you should treat it as some floating point value.
		134
		135	Unlike the name component C<stamp> might indicate, it is also used for
		136	time differences (e.g. delays) throughout libev.
		137
		138	=head1 ERROR HANDLING
		139
		140	Libev knows three classes of errors: operating system errors, usage errors
		141	and internal errors (bugs).
		142
		143	When libev catches an operating system error it cannot handle (for example
		144	a system call indicating a condition libev cannot fix), it calls the callback
		145	set via C<ev_set_syserr_cb>, which is supposed to fix the problem or
		146	abort. The default is to print a diagnostic message and to call C<abort
		147	()>.
		148
		149	When libev detects a usage error such as a negative timer interval, then
		150	it will print a diagnostic message and abort (via the C<assert> mechanism,
		151	so C<NDEBUG> will disable this checking): these are programming errors in
		152	the libev caller and need to be fixed there.
		153
		154	Libev also has a few internal error-checking C<assert>ions, and also has
		155	extensive consistency checking code. These do not trigger under normal
		156	circumstances, as they indicate either a bug in libev or worse.
		157
51		158
52	=head1 GLOBAL FUNCTIONS	159	=head1 GLOBAL FUNCTIONS
53		160
54	These functions can be called anytime, even before initialising the	161	These functions can be called anytime, even before initialising the
55	library in any way.	162	library in any way.
56		163
57	=over 4	164	=over 4
58		165
59	=item ev_tstamp ev_time ()	166	=item ev_tstamp ev_time ()
60		167
61	Returns the current time as libev would use it.	168	Returns the current time as libev would use it. Please note that the
		169	C<ev_now> function is usually faster and also often returns the timestamp
		170	you actually want to know.
		171
		172	=item ev_sleep (ev_tstamp interval)
		173
		174	Sleep for the given interval: The current thread will be blocked until
		175	either it is interrupted or the given time interval has passed. Basically
		176	this is a sub-second-resolution C<sleep ()>.
62		177
63	=item int ev_version_major ()	178	=item int ev_version_major ()
64		179
65	=item int ev_version_minor ()	180	=item int ev_version_minor ()
66		181
67	You can find out the major and minor version numbers of the library	182	You can find out the major and minor ABI version numbers of the library
68	you linked against by calling the functions C<ev_version_major> and	183	you linked against by calling the functions C<ev_version_major> and
69	C<ev_version_minor>. If you want, you can compare against the global	184	C<ev_version_minor>. If you want, you can compare against the global
70	symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the	185	symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the
71	version of the library your program was compiled against.	186	version of the library your program was compiled against.
72		187
		188	These version numbers refer to the ABI version of the library, not the
		189	release version.
		190
73	Usually, it's a good idea to terminate if the major versions mismatch,	191	Usually, it's a good idea to terminate if the major versions mismatch,
74	as this indicates an incompatible change. Minor versions are usually	192	as this indicates an incompatible change. Minor versions are usually
75	compatible to older versions, so a larger minor version alone is usually	193	compatible to older versions, so a larger minor version alone is usually
76	not a problem.	194	not a problem.
77		195
		196	Example: Make sure we haven't accidentally been linked against the wrong
		197	version (note, however, that this will not detect ABI mismatches :).
		198
		199	assert (("libev version mismatch",
		200	ev_version_major () == EV_VERSION_MAJOR
		201	&& ev_version_minor () >= EV_VERSION_MINOR));
		202
		203	=item unsigned int ev_supported_backends ()
		204
		205	Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
		206	value) compiled into this binary of libev (independent of their
		207	availability on the system you are running on). See C<ev_default_loop> for
		208	a description of the set values.
		209
		210	Example: make sure we have the epoll method, because yeah this is cool and
		211	a must have and can we have a torrent of it please!!!11
		212
		213	assert (("sorry, no epoll, no sex",
		214	ev_supported_backends () & EVBACKEND_EPOLL));
		215
		216	=item unsigned int ev_recommended_backends ()
		217
		218	Return the set of all backends compiled into this binary of libev and also
		219	recommended for this platform. This set is often smaller than the one
		220	returned by C<ev_supported_backends>, as for example kqueue is broken on
		221	most BSDs and will not be auto-detected unless you explicitly request it
		222	(assuming you know what you are doing). This is the set of backends that
		223	libev will probe for if you specify no backends explicitly.
		224
		225	=item unsigned int ev_embeddable_backends ()
		226
		227	Returns the set of backends that are embeddable in other event loops. This
		228	is the theoretical, all-platform, value. To find which backends
		229	might be supported on the current system, you would need to look at
		230	C<ev_embeddable_backends () & ev_supported_backends ()>, likewise for
		231	recommended ones.
		232
		233	See the description of C<ev_embed> watchers for more info.
		234
78	=item ev_set_allocator (void *(*cb)(void *ptr, long size))	235	=item ev_set_allocator (void *(*cb)(void *ptr, long size)) [NOT REENTRANT]
79		236
80	Sets the allocation function to use (the prototype is similar to the	237	Sets the allocation function to use (the prototype is similar - the
81	realloc C function, the semantics are identical). It is used to allocate	238	semantics are identical to the C<realloc> C89/SuS/POSIX function). It is
82	and free memory (no surprises here). If it returns zero when memory	239	used to allocate and free memory (no surprises here). If it returns zero
83	needs to be allocated, the library might abort or take some potentially	240	when memory needs to be allocated (C<size != 0>), the library might abort
84	destructive action. The default is your system realloc function.	241	or take some potentially destructive action.
		242
		243	Since some systems (at least OpenBSD and Darwin) fail to implement
		244	correct C<realloc> semantics, libev will use a wrapper around the system
		245	C<realloc> and C<free> functions by default.
85		246
86	You could override this function in high-availability programs to, say,	247	You could override this function in high-availability programs to, say,
87	free some memory if it cannot allocate memory, to use a special allocator,	248	free some memory if it cannot allocate memory, to use a special allocator,
88	or even to sleep a while and retry until some memory is available.	249	or even to sleep a while and retry until some memory is available.
89		250
		251	Example: Replace the libev allocator with one that waits a bit and then
		252	retries (example requires a standards-compliant C<realloc>).
		253
		254	static void *
		255	persistent_realloc (void *ptr, size_t size)
		256	{
		257	for (;;)
		258	{
		259	void *newptr = realloc (ptr, size);
		260
		261	if (newptr)
		262	return newptr;
		263
		264	sleep (60);
		265	}
		266	}
		267
		268	...
		269	ev_set_allocator (persistent_realloc);
		270
90	=item ev_set_syserr_cb (void (*cb)(const char *msg));	271	=item ev_set_syserr_cb (void (*cb)(const char *msg)); [NOT REENTRANT]
91		272
92	Set the callback function to call on a retryable syscall error (such	273	Set the callback function to call on a retryable system call error (such
93	as failed select, poll, epoll_wait). The message is a printable string	274	as failed select, poll, epoll_wait). The message is a printable string
94	indicating the system call or subsystem causing the problem. If this	275	indicating the system call or subsystem causing the problem. If this
95	callback is set, then libev will expect it to remedy the sitution, no	276	callback is set, then libev will expect it to remedy the situation, no
96	matter what, when it returns. That is, libev will generally retry the	277	matter what, when it returns. That is, libev will generally retry the
97	requested operation, or, if the condition doesn't go away, do bad stuff	278	requested operation, or, if the condition doesn't go away, do bad stuff
98	(such as abort).	279	(such as abort).
99		280
		281	Example: This is basically the same thing that libev does internally, too.
		282
		283	static void
		284	fatal_error (const char *msg)
		285	{
		286	perror (msg);
		287	abort ();
		288	}
		289
		290	...
		291	ev_set_syserr_cb (fatal_error);
		292
100	=back	293	=back
101		294
102	=head1 FUNCTIONS CONTROLLING THE EVENT LOOP	295	=head1 FUNCTIONS CONTROLLING THE EVENT LOOP
103		296
104	An event loop is described by a C<struct ev_loop *>. The library knows two	297	An event loop is described by a C<struct ev_loop *> (the C<struct>
105	types of such loops, the I<default> loop, which supports signals and child	298	is I<not> optional in this case, as there is also an C<ev_loop>
106	events, and dynamically created loops which do not.	299	I<function>).
107		300
108	If you use threads, a common model is to run the default event loop	301	The library knows two types of such loops, the I<default> loop, which
109	in your main thread (or in a separate thread) and for each thread you	302	supports signals and child events, and dynamically created loops which do
110	create, you also create another event loop. Libev itself does no locking	303	not.
111	whatsoever, so if you mix calls to the same event loop in different
112	threads, make sure you lock (this is usually a bad idea, though, even if
113	done correctly, because it's hideous and inefficient).
114		304
115	=over 4	305	=over 4
116		306
117	=item struct ev_loop *ev_default_loop (unsigned int flags)	307	=item struct ev_loop *ev_default_loop (unsigned int flags)
118		308
119	This will initialise the default event loop if it hasn't been initialised	309	This will initialise the default event loop if it hasn't been initialised
120	yet and return it. If the default loop could not be initialised, returns	310	yet and return it. If the default loop could not be initialised, returns
121	false. If it already was initialised it simply returns it (and ignores the	311	false. If it already was initialised it simply returns it (and ignores the
122	flags).	312	flags. If that is troubling you, check C<ev_backend ()> afterwards).
123		313
124	If you don't know what event loop to use, use the one returned from this	314	If you don't know what event loop to use, use the one returned from this
125	function.	315	function.
126		316
		317	Note that this function is I<not> thread-safe, so if you want to use it
		318	from multiple threads, you have to lock (note also that this is unlikely,
		319	as loops cannot be shared easily between threads anyway).
		320
		321	The default loop is the only loop that can handle C<ev_signal> and
		322	C<ev_child> watchers, and to do this, it always registers a handler
		323	for C<SIGCHLD>. If this is a problem for your application you can either
		324	create a dynamic loop with C<ev_loop_new> that doesn't do that, or you
		325	can simply overwrite the C<SIGCHLD> signal handler I<after> calling
		326	C<ev_default_init>.
		327
127	The flags argument can be used to specify special behaviour or specific	328	The flags argument can be used to specify special behaviour or specific
128	backends to use, and is usually specified as 0 (or EVFLAG_AUTO).	329	backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
129		330
130	It supports the following flags:	331	The following flags are supported:
131		332
132	=over 4	333	=over 4
133		334
134	=item C<EVFLAG_AUTO>	335	=item C<EVFLAG_AUTO>
135		336
136	The default flags value. Use this if you have no clue (it's the right	337	The default flags value. Use this if you have no clue (it's the right
137	thing, believe me).	338	thing, believe me).
138		339
139	=item C<EVFLAG_NOENV>	340	=item C<EVFLAG_NOENV>
140		341
141	If this flag bit is ored into the flag value (or the program runs setuid	342	If this flag bit is or'ed into the flag value (or the program runs setuid
142	or setgid) then libev will I<not> look at the environment variable	343	or setgid) then libev will I<not> look at the environment variable
143	C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will	344	C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
144	override the flags completely if it is found in the environment. This is	345	override the flags completely if it is found in the environment. This is
145	useful to try out specific backends to test their performance, or to work	346	useful to try out specific backends to test their performance, or to work
146	around bugs.	347	around bugs.
147		348
		349	=item C<EVFLAG_FORKCHECK>
		350
		351	Instead of calling C<ev_loop_fork> manually after a fork, you can also
		352	make libev check for a fork in each iteration by enabling this flag.
		353
		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_SIGNALFD>
		376
		377	When this flag is specified, then libev will attempt to use the
		378	I<signalfd> API for it's C<ev_signal> (and C<ev_child>) watchers. This API
		379	delivers signals synchronously, which makes it both faster and might make
		380	it possible to get the queued signal data. It can also simplify signal
		381	handling with threads, as long as you properly block signals in your
		382	threads that are not interested in handling them.
		383
		384	Signalfd will not be used by default as this changes your signal mask, and
		385	there are a lot of shoddy libraries and programs (glib's threadpool for
		386	example) that can't properly initialise their signal masks.
		387
148	=item C<EVMETHOD_SELECT> (portable select backend)	388	=item C<EVBACKEND_SELECT> (value 1, portable select backend)
149		389
		390	This is your standard select(2) backend. Not I<completely> standard, as
		391	libev tries to roll its own fd_set with no limits on the number of fds,
		392	but if that fails, expect a fairly low limit on the number of fds when
		393	using this backend. It doesn't scale too well (O(highest_fd)), but its
		394	usually the fastest backend for a low number of (low-numbered :) fds.
		395
		396	To get good performance out of this backend you need a high amount of
		397	parallelism (most of the file descriptors should be busy). If you are
		398	writing a server, you should C<accept ()> in a loop to accept as many
		399	connections as possible during one iteration. You might also want to have
		400	a look at C<ev_set_io_collect_interval ()> to increase the amount of
		401	readiness notifications you get per iteration.
		402
		403	This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
		404	C<writefds> set (and to work around Microsoft Windows bugs, also onto the
		405	C<exceptfds> set on that platform).
		406
150	=item C<EVMETHOD_POLL> (poll backend, available everywhere except on windows)	407	=item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
151		408
152	=item C<EVMETHOD_EPOLL> (linux only)	409	And this is your standard poll(2) backend. It's more complicated
		410	than select, but handles sparse fds better and has no artificial
		411	limit on the number of fds you can use (except it will slow down
		412	considerably with a lot of inactive fds). It scales similarly to select,
		413	i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
		414	performance tips.
153		415
154	=item C<EVMETHOD_KQUEUE> (some bsds only)	416	This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
		417	C<EV_WRITE> to C<POLLOUT | POLLERR | POLLHUP>.
155		418
156	=item C<EVMETHOD_DEVPOLL> (solaris 8 only)	419	=item C<EVBACKEND_EPOLL> (value 4, Linux)
157		420
158	=item C<EVMETHOD_PORT> (solaris 10 only)	421	Use the linux-specific epoll(7) interface (for both pre- and post-2.6.9
		422	kernels).
159		423
160	If one or more of these are ored into the flags value, then only these	424	For few fds, this backend is a bit little slower than poll and select,
161	backends will be tried (in the reverse order as given here). If one are	425	but it scales phenomenally better. While poll and select usually scale
162	specified, any backend will do.	426	like O(total_fds) where n is the total number of fds (or the highest fd),
		427	epoll scales either O(1) or O(active_fds).
		428
		429	The epoll mechanism deserves honorable mention as the most misdesigned
		430	of the more advanced event mechanisms: mere annoyances include silently
		431	dropping file descriptors, requiring a system call per change per file
		432	descriptor (and unnecessary guessing of parameters), problems with dup and
		433	so on. The biggest issue is fork races, however - if a program forks then
		434	I<both> parent and child process have to recreate the epoll set, which can
		435	take considerable time (one syscall per file descriptor) and is of course
		436	hard to detect.
		437
		438	Epoll is also notoriously buggy - embedding epoll fds I<should> work, but
		439	of course I<doesn't>, and epoll just loves to report events for totally
		440	I<different> file descriptors (even already closed ones, so one cannot
		441	even remove them from the set) than registered in the set (especially
		442	on SMP systems). Libev tries to counter these spurious notifications by
		443	employing an additional generation counter and comparing that against the
		444	events to filter out spurious ones, recreating the set when required. Last
		445	not least, it also refuses to work with some file descriptors which work
		446	perfectly fine with C<select> (files, many character devices...).
		447
		448	While stopping, setting and starting an I/O watcher in the same iteration
		449	will result in some caching, there is still a system call per such
		450	incident (because the same I<file descriptor> could point to a different
		451	I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
		452	file descriptors might not work very well if you register events for both
		453	file descriptors.
		454
		455	Best performance from this backend is achieved by not unregistering all
		456	watchers for a file descriptor until it has been closed, if possible,
		457	i.e. keep at least one watcher active per fd at all times. Stopping and
		458	starting a watcher (without re-setting it) also usually doesn't cause
		459	extra overhead. A fork can both result in spurious notifications as well
		460	as in libev having to destroy and recreate the epoll object, which can
		461	take considerable time and thus should be avoided.
		462
		463	All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
		464	faster than epoll for maybe up to a hundred file descriptors, depending on
		465	the usage. So sad.
		466
		467	While nominally embeddable in other event loops, this feature is broken in
		468	all kernel versions tested so far.
		469
		470	This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
		471	C<EVBACKEND_POLL>.
		472
		473	=item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
		474
		475	Kqueue deserves special mention, as at the time of this writing, it
		476	was broken on all BSDs except NetBSD (usually it doesn't work reliably
		477	with anything but sockets and pipes, except on Darwin, where of course
		478	it's completely useless). Unlike epoll, however, whose brokenness
		479	is by design, these kqueue bugs can (and eventually will) be fixed
		480	without API changes to existing programs. For this reason it's not being
		481	"auto-detected" unless you explicitly specify it in the flags (i.e. using
		482	C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
		483	system like NetBSD.
		484
		485	You still can embed kqueue into a normal poll or select backend and use it
		486	only for sockets (after having made sure that sockets work with kqueue on
		487	the target platform). See C<ev_embed> watchers for more info.
		488
		489	It scales in the same way as the epoll backend, but the interface to the
		490	kernel is more efficient (which says nothing about its actual speed, of
		491	course). While stopping, setting and starting an I/O watcher does never
		492	cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
		493	two event changes per incident. Support for C<fork ()> is very bad (but
		494	sane, unlike epoll) and it drops fds silently in similarly hard-to-detect
		495	cases
		496
		497	This backend usually performs well under most conditions.
		498
		499	While nominally embeddable in other event loops, this doesn't work
		500	everywhere, so you might need to test for this. And since it is broken
		501	almost everywhere, you should only use it when you have a lot of sockets
		502	(for which it usually works), by embedding it into another event loop
		503	(e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
		504	also broken on OS X)) and, did I mention it, using it only for sockets.
		505
		506	This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
		507	C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
		508	C<NOTE_EOF>.
		509
		510	=item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
		511
		512	This is not implemented yet (and might never be, unless you send me an
		513	implementation). According to reports, C</dev/poll> only supports sockets
		514	and is not embeddable, which would limit the usefulness of this backend
		515	immensely.
		516
		517	=item C<EVBACKEND_PORT> (value 32, Solaris 10)
		518
		519	This uses the Solaris 10 event port mechanism. As with everything on Solaris,
		520	it's really slow, but it still scales very well (O(active_fds)).
		521
		522	Please note that Solaris event ports can deliver a lot of spurious
		523	notifications, so you need to use non-blocking I/O or other means to avoid
		524	blocking when no data (or space) is available.
		525
		526	While this backend scales well, it requires one system call per active
		527	file descriptor per loop iteration. For small and medium numbers of file
		528	descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
		529	might perform better.
		530
		531	On the positive side, with the exception of the spurious readiness
		532	notifications, this backend actually performed fully to specification
		533	in all tests and is fully embeddable, which is a rare feat among the
		534	OS-specific backends (I vastly prefer correctness over speed hacks).
		535
		536	This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
		537	C<EVBACKEND_POLL>.
		538
		539	=item C<EVBACKEND_ALL>
		540
		541	Try all backends (even potentially broken ones that wouldn't be tried
		542	with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
		543	C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.
		544
		545	It is definitely not recommended to use this flag.
163		546
164	=back	547	=back
165		548
		549	If one or more of the backend flags are or'ed into the flags value,
		550	then only these backends will be tried (in the reverse order as listed
		551	here). If none are specified, all backends in C<ev_recommended_backends
		552	()> will be tried.
		553
		554	Example: This is the most typical usage.
		555
		556	if (!ev_default_loop (0))
		557	fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
		558
		559	Example: Restrict libev to the select and poll backends, and do not allow
		560	environment settings to be taken into account:
		561
		562	ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
		563
		564	Example: Use whatever libev has to offer, but make sure that kqueue is
		565	used if available (warning, breaks stuff, best use only with your own
		566	private event loop and only if you know the OS supports your types of
		567	fds):
		568
		569	ev_default_loop (ev_recommended_backends () | EVBACKEND_KQUEUE);
		570
166	=item struct ev_loop *ev_loop_new (unsigned int flags)	571	=item struct ev_loop *ev_loop_new (unsigned int flags)
167		572
168	Similar to C<ev_default_loop>, but always creates a new event loop that is	573	Similar to C<ev_default_loop>, but always creates a new event loop that is
169	always distinct from the default loop. Unlike the default loop, it cannot	574	always distinct from the default loop.
170	handle signal and child watchers, and attempts to do so will be greeted by	575
171	undefined behaviour (or a failed assertion if assertions are enabled).	576	Note that this function I<is> thread-safe, and one common way to use
		577	libev with threads is indeed to create one loop per thread, and using the
		578	default loop in the "main" or "initial" thread.
		579
		580	Example: Try to create a event loop that uses epoll and nothing else.
		581
		582	struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
		583	if (!epoller)
		584	fatal ("no epoll found here, maybe it hides under your chair");
172		585
173	=item ev_default_destroy ()	586	=item ev_default_destroy ()
174		587
175	Destroys the default loop again (frees all memory and kernel state	588	Destroys the default loop (frees all memory and kernel state etc.). None
176	etc.). This stops all registered event watchers (by not touching them in	589	of the active event watchers will be stopped in the normal sense, so
177	any way whatsoever, although you cannot rely on this :).	590	e.g. C<ev_is_active> might still return true. It is your responsibility to
		591	either stop all watchers cleanly yourself I<before> calling this function,
		592	or cope with the fact afterwards (which is usually the easiest thing, you
		593	can just ignore the watchers and/or C<free ()> them for example).
		594
		595	Note that certain global state, such as signal state (and installed signal
		596	handlers), will not be freed by this function, and related watchers (such
		597	as signal and child watchers) would need to be stopped manually.
		598
		599	In general it is not advisable to call this function except in the
		600	rare occasion where you really need to free e.g. the signal handling
		601	pipe fds. If you need dynamically allocated loops it is better to use
		602	C<ev_loop_new> and C<ev_loop_destroy>.
178		603
179	=item ev_loop_destroy (loop)	604	=item ev_loop_destroy (loop)
180		605
181	Like C<ev_default_destroy>, but destroys an event loop created by an	606	Like C<ev_default_destroy>, but destroys an event loop created by an
182	earlier call to C<ev_loop_new>.	607	earlier call to C<ev_loop_new>.
183		608
184	=item ev_default_fork ()	609	=item ev_default_fork ()
185		610
		611	This function sets a flag that causes subsequent C<ev_loop> iterations
186	This function reinitialises the kernel state for backends that have	612	to reinitialise the kernel state for backends that have one. Despite the
187	one. Despite the name, you can call it anytime, but it makes most sense	613	name, you can call it anytime, but it makes most sense after forking, in
188	after forking, in either the parent or child process (or both, but that	614	the child process (or both child and parent, but that again makes little
189	again makes little sense).	615	sense). You I<must> call it in the child before using any of the libev
		616	functions, and it will only take effect at the next C<ev_loop> iteration.
190		617
191	You I<must> call this function after forking if and only if you want to	618	Again, you I<have> to call it on I<any> loop that you want to re-use after
192	use the event library in both processes. If you just fork+exec, you don't	619	a fork, I<even if you do not plan to use the loop in the parent>. This is
193	have to call it.	620	because some kernel interfaces *cough* I<kqueue> *cough* do funny things
		621	during fork.
		622
		623	On the other hand, you only need to call this function in the child
		624	process if and only if you want to use the event loop in the child. If you
		625	just fork+exec or create a new loop in the child, you don't have to call
		626	it at all.
194		627
195	The function itself is quite fast and it's usually not a problem to call	628	The function itself is quite fast and it's usually not a problem to call
196	it just in case after a fork. To make this easy, the function will fit in	629	it just in case after a fork. To make this easy, the function will fit in
197	quite nicely into a call to C<pthread_atfork>:	630	quite nicely into a call to C<pthread_atfork>:
198		631
…		…
200		633
201	=item ev_loop_fork (loop)	634	=item ev_loop_fork (loop)
202		635
203	Like C<ev_default_fork>, but acts on an event loop created by	636	Like C<ev_default_fork>, but acts on an event loop created by
204	C<ev_loop_new>. Yes, you have to call this on every allocated event loop	637	C<ev_loop_new>. Yes, you have to call this on every allocated event loop
205	after fork, and how you do this is entirely your own problem.	638	after fork that you want to re-use in the child, and how you keep track of
		639	them is entirely your own problem.
206		640
		641	=item int ev_is_default_loop (loop)
		642
		643	Returns true when the given loop is, in fact, the default loop, and false
		644	otherwise.
		645
		646	=item unsigned int ev_iteration (loop)
		647
		648	Returns the current iteration count for the loop, which is identical to
		649	the number of times libev did poll for new events. It starts at C<0> and
		650	happily wraps around with enough iterations.
		651
		652	This value can sometimes be useful as a generation counter of sorts (it
		653	"ticks" the number of loop iterations), as it roughly corresponds with
		654	C<ev_prepare> and C<ev_check> calls - and is incremented between the
		655	prepare and check phases.
		656
207	=item unsigned int ev_method (loop)	657	=item unsigned int ev_depth (loop)
208		658
		659	Returns the number of times C<ev_loop> was entered minus the number of
		660	times C<ev_loop> was exited, in other words, the recursion depth.
		661
		662	Outside C<ev_loop>, this number is zero. In a callback, this number is
		663	C<1>, unless C<ev_loop> was invoked recursively (or from another thread),
		664	in which case it is higher.
		665
		666	Leaving C<ev_loop> abnormally (setjmp/longjmp, cancelling the thread
		667	etc.), doesn't count as "exit" - consider this as a hint to avoid such
		668	ungentleman behaviour unless it's really convenient.
		669
		670	=item unsigned int ev_backend (loop)
		671
209	Returns one of the C<EVMETHOD_*> flags indicating the event backend in	672	Returns one of the C<EVBACKEND_*> flags indicating the event backend in
210	use.	673	use.
211		674
212	=item ev_tstamp ev_now (loop)	675	=item ev_tstamp ev_now (loop)
213		676
214	Returns the current "event loop time", which is the time the event loop	677	Returns the current "event loop time", which is the time the event loop
215	got events and started processing them. This timestamp does not change	678	received events and started processing them. This timestamp does not
216	as long as callbacks are being processed, and this is also the base time	679	change as long as callbacks are being processed, and this is also the base
217	used for relative timers. You can treat it as the timestamp of the event	680	time used for relative timers. You can treat it as the timestamp of the
218	occuring (or more correctly, the mainloop finding out about it).	681	event occurring (or more correctly, libev finding out about it).
		682
		683	=item ev_now_update (loop)
		684
		685	Establishes the current time by querying the kernel, updating the time
		686	returned by C<ev_now ()> in the progress. This is a costly operation and
		687	is usually done automatically within C<ev_loop ()>.
		688
		689	This function is rarely useful, but when some event callback runs for a
		690	very long time without entering the event loop, updating libev's idea of
		691	the current time is a good idea.
		692
		693	See also L<The special problem of time updates> in the C<ev_timer> section.
		694
		695	=item ev_suspend (loop)
		696
		697	=item ev_resume (loop)
		698
		699	These two functions suspend and resume a loop, for use when the loop is
		700	not used for a while and timeouts should not be processed.
		701
		702	A typical use case would be an interactive program such as a game: When
		703	the user presses C<^Z> to suspend the game and resumes it an hour later it
		704	would be best to handle timeouts as if no time had actually passed while
		705	the program was suspended. This can be achieved by calling C<ev_suspend>
		706	in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
		707	C<ev_resume> directly afterwards to resume timer processing.
		708
		709	Effectively, all C<ev_timer> watchers will be delayed by the time spend
		710	between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
		711	will be rescheduled (that is, they will lose any events that would have
		712	occurred while suspended).
		713
		714	After calling C<ev_suspend> you B<must not> call I<any> function on the
		715	given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
		716	without a previous call to C<ev_suspend>.
		717
		718	Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
		719	event loop time (see C<ev_now_update>).
219		720
220	=item ev_loop (loop, int flags)	721	=item ev_loop (loop, int flags)
221		722
222	Finally, this is it, the event handler. This function usually is called	723	Finally, this is it, the event handler. This function usually is called
223	after you initialised all your watchers and you want to start handling	724	after you have initialised all your watchers and you want to start
224	events.	725	handling events.
225		726
226	If the flags argument is specified as 0, it will not return until either	727	If the flags argument is specified as C<0>, it will not return until
227	no event watchers are active anymore or C<ev_unloop> was called.	728	either no event watchers are active anymore or C<ev_unloop> was called.
		729
		730	Please note that an explicit C<ev_unloop> is usually better than
		731	relying on all watchers to be stopped when deciding when a program has
		732	finished (especially in interactive programs), but having a program
		733	that automatically loops as long as it has to and no longer by virtue
		734	of relying on its watchers stopping correctly, that is truly a thing of
		735	beauty.
228		736
229	A flags value of C<EVLOOP_NONBLOCK> will look for new events, will handle	737	A flags value of C<EVLOOP_NONBLOCK> will look for new events, will handle
230	those events and any outstanding ones, but will not block your process in	738	those events and any already outstanding ones, but will not block your
231	case there are no events and will return after one iteration of the loop.	739	process in case there are no events and will return after one iteration of
		740	the loop.
232		741
233	A flags value of C<EVLOOP_ONESHOT> will look for new events (waiting if	742	A flags value of C<EVLOOP_ONESHOT> will look for new events (waiting if
234	neccessary) and will handle those and any outstanding ones. It will block	743	necessary) and will handle those and any already outstanding ones. It
235	your process until at least one new event arrives, and will return after	744	will block your process until at least one new event arrives (which could
		745	be an event internal to libev itself, so there is no guarantee that a
		746	user-registered callback will be called), and will return after one
236	one iteration of the loop.	747	iteration of the loop.
237		748
238	This flags value could be used to implement alternative looping	749	This is useful if you are waiting for some external event in conjunction
239	constructs, but the C<prepare> and C<check> watchers provide a better and	750	with something not expressible using other libev watchers (i.e. "roll your
240	more generic mechanism.	751	own C<ev_loop>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
		752	usually a better approach for this kind of thing.
		753
		754	Here are the gory details of what C<ev_loop> does:
		755
		756	- Before the first iteration, call any pending watchers.
		757	* If EVFLAG_FORKCHECK was used, check for a fork.
		758	- If a fork was detected (by any means), queue and call all fork watchers.
		759	- Queue and call all prepare watchers.
		760	- If we have been forked, detach and recreate the kernel state
		761	as to not disturb the other process.
		762	- Update the kernel state with all outstanding changes.
		763	- Update the "event loop time" (ev_now ()).
		764	- Calculate for how long to sleep or block, if at all
		765	(active idle watchers, EVLOOP_NONBLOCK or not having
		766	any active watchers at all will result in not sleeping).
		767	- Sleep if the I/O and timer collect interval say so.
		768	- Block the process, waiting for any events.
		769	- Queue all outstanding I/O (fd) events.
		770	- Update the "event loop time" (ev_now ()), and do time jump adjustments.
		771	- Queue all expired timers.
		772	- Queue all expired periodics.
		773	- Unless any events are pending now, queue all idle watchers.
		774	- Queue all check watchers.
		775	- Call all queued watchers in reverse order (i.e. check watchers first).
		776	Signals and child watchers are implemented as I/O watchers, and will
		777	be handled here by queueing them when their watcher gets executed.
		778	- If ev_unloop has been called, or EVLOOP_ONESHOT or EVLOOP_NONBLOCK
		779	were used, or there are no active watchers, return, otherwise
		780	continue with step *.
		781
		782	Example: Queue some jobs and then loop until no events are outstanding
		783	anymore.
		784
		785	... queue jobs here, make sure they register event watchers as long
		786	... as they still have work to do (even an idle watcher will do..)
		787	ev_loop (my_loop, 0);
		788	... jobs done or somebody called unloop. yeah!
241		789
242	=item ev_unloop (loop, how)	790	=item ev_unloop (loop, how)
243		791
244	Can be used to make a call to C<ev_loop> return early (but only after it	792	Can be used to make a call to C<ev_loop> return early (but only after it
245	has processed all outstanding events). The C<how> argument must be either	793	has processed all outstanding events). The C<how> argument must be either
246	C<EVUNLOOP_ONCE>, which will make the innermost C<ev_loop> call return, or	794	C<EVUNLOOP_ONE>, which will make the innermost C<ev_loop> call return, or
247	C<EVUNLOOP_ALL>, which will make all nested C<ev_loop> calls return.	795	C<EVUNLOOP_ALL>, which will make all nested C<ev_loop> calls return.
		796
		797	This "unloop state" will be cleared when entering C<ev_loop> again.
		798
		799	It is safe to call C<ev_unloop> from outside any C<ev_loop> calls.
248		800
249	=item ev_ref (loop)	801	=item ev_ref (loop)
250		802
251	=item ev_unref (loop)	803	=item ev_unref (loop)
252		804
253	Ref/unref can be used to add or remove a reference count on the event	805	Ref/unref can be used to add or remove a reference count on the event
254	loop: Every watcher keeps one reference, and as long as the reference	806	loop: Every watcher keeps one reference, and as long as the reference
255	count is nonzero, C<ev_loop> will not return on its own. If you have	807	count is nonzero, C<ev_loop> will not return on its own.
256	a watcher you never unregister that should not keep C<ev_loop> from	808
257	returning, ev_unref() after starting, and ev_ref() before stopping it. For	809	This is useful when you have a watcher that you never intend to
		810	unregister, but that nevertheless should not keep C<ev_loop> from
		811	returning. In such a case, call C<ev_unref> after starting, and C<ev_ref>
		812	before stopping it.
		813
258	example, libev itself uses this for its internal signal pipe: It is not	814	As an example, libev itself uses this for its internal signal pipe: It
259	visible to the libev user and should not keep C<ev_loop> from exiting if	815	is not visible to the libev user and should not keep C<ev_loop> from
260	no event watchers registered by it are active. It is also an excellent	816	exiting if no event watchers registered by it are active. It is also an
261	way to do this for generic recurring timers or from within third-party	817	excellent way to do this for generic recurring timers or from within
262	libraries. Just remember to I<unref after start> and I<ref before stop>.	818	third-party libraries. Just remember to I<unref after start> and I<ref
		819	before stop> (but only if the watcher wasn't active before, or was active
		820	before, respectively. Note also that libev might stop watchers itself
		821	(e.g. non-repeating timers) in which case you have to C<ev_ref>
		822	in the callback).
		823
		824	Example: Create a signal watcher, but keep it from keeping C<ev_loop>
		825	running when nothing else is active.
		826
		827	ev_signal exitsig;
		828	ev_signal_init (&exitsig, sig_cb, SIGINT);
		829	ev_signal_start (loop, &exitsig);
		830	evf_unref (loop);
		831
		832	Example: For some weird reason, unregister the above signal handler again.
		833
		834	ev_ref (loop);
		835	ev_signal_stop (loop, &exitsig);
		836
		837	=item ev_set_io_collect_interval (loop, ev_tstamp interval)
		838
		839	=item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
		840
		841	These advanced functions influence the time that libev will spend waiting
		842	for events. Both time intervals are by default C<0>, meaning that libev
		843	will try to invoke timer/periodic callbacks and I/O callbacks with minimum
		844	latency.
		845
		846	Setting these to a higher value (the C<interval> I<must> be >= C<0>)
		847	allows libev to delay invocation of I/O and timer/periodic callbacks
		848	to increase efficiency of loop iterations (or to increase power-saving
		849	opportunities).
		850
		851	The idea is that sometimes your program runs just fast enough to handle
		852	one (or very few) event(s) per loop iteration. While this makes the
		853	program responsive, it also wastes a lot of CPU time to poll for new
		854	events, especially with backends like C<select ()> which have a high
		855	overhead for the actual polling but can deliver many events at once.
		856
		857	By setting a higher I<io collect interval> you allow libev to spend more
		858	time collecting I/O events, so you can handle more events per iteration,
		859	at the cost of increasing latency. Timeouts (both C<ev_periodic> and
		860	C<ev_timer>) will be not affected. Setting this to a non-null value will
		861	introduce an additional C<ev_sleep ()> call into most loop iterations. The
		862	sleep time ensures that libev will not poll for I/O events more often then
		863	once per this interval, on average.
		864
		865	Likewise, by setting a higher I<timeout collect interval> you allow libev
		866	to spend more time collecting timeouts, at the expense of increased
		867	latency/jitter/inexactness (the watcher callback will be called
		868	later). C<ev_io> watchers will not be affected. Setting this to a non-null
		869	value will not introduce any overhead in libev.
		870
		871	Many (busy) programs can usually benefit by setting the I/O collect
		872	interval to a value near C<0.1> or so, which is often enough for
		873	interactive servers (of course not for games), likewise for timeouts. It
		874	usually doesn't make much sense to set it to a lower value than C<0.01>,
		875	as this approaches the timing granularity of most systems. Note that if
		876	you do transactions with the outside world and you can't increase the
		877	parallelity, then this setting will limit your transaction rate (if you
		878	need to poll once per transaction and the I/O collect interval is 0.01,
		879	then you can't do more than 100 transactions per second).
		880
		881	Setting the I<timeout collect interval> can improve the opportunity for
		882	saving power, as the program will "bundle" timer callback invocations that
		883	are "near" in time together, by delaying some, thus reducing the number of
		884	times the process sleeps and wakes up again. Another useful technique to
		885	reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
		886	they fire on, say, one-second boundaries only.
		887
		888	Example: we only need 0.1s timeout granularity, and we wish not to poll
		889	more often than 100 times per second:
		890
		891	ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
		892	ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
		893
		894	=item ev_invoke_pending (loop)
		895
		896	This call will simply invoke all pending watchers while resetting their
		897	pending state. Normally, C<ev_loop> does this automatically when required,
		898	but when overriding the invoke callback this call comes handy.
		899
		900	=item int ev_pending_count (loop)
		901
		902	Returns the number of pending watchers - zero indicates that no watchers
		903	are pending.
		904
		905	=item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
		906
		907	This overrides the invoke pending functionality of the loop: Instead of
		908	invoking all pending watchers when there are any, C<ev_loop> will call
		909	this callback instead. This is useful, for example, when you want to
		910	invoke the actual watchers inside another context (another thread etc.).
		911
		912	If you want to reset the callback, use C<ev_invoke_pending> as new
		913	callback.
		914
		915	=item ev_set_loop_release_cb (loop, void (*release)(EV_P), void (*acquire)(EV_P))
		916
		917	Sometimes you want to share the same loop between multiple threads. This
		918	can be done relatively simply by putting mutex_lock/unlock calls around
		919	each call to a libev function.
		920
		921	However, C<ev_loop> can run an indefinite time, so it is not feasible to
		922	wait for it to return. One way around this is to wake up the loop via
		923	C<ev_unloop> and C<av_async_send>, another way is to set these I<release>
		924	and I<acquire> callbacks on the loop.
		925
		926	When set, then C<release> will be called just before the thread is
		927	suspended waiting for new events, and C<acquire> is called just
		928	afterwards.
		929
		930	Ideally, C<release> will just call your mutex_unlock function, and
		931	C<acquire> will just call the mutex_lock function again.
		932
		933	While event loop modifications are allowed between invocations of
		934	C<release> and C<acquire> (that's their only purpose after all), no
		935	modifications done will affect the event loop, i.e. adding watchers will
		936	have no effect on the set of file descriptors being watched, or the time
		937	waited. Use an C<ev_async> watcher to wake up C<ev_loop> when you want it
		938	to take note of any changes you made.
		939
		940	In theory, threads executing C<ev_loop> will be async-cancel safe between
		941	invocations of C<release> and C<acquire>.
		942
		943	See also the locking example in the C<THREADS> section later in this
		944	document.
		945
		946	=item ev_set_userdata (loop, void *data)
		947
		948	=item ev_userdata (loop)
		949
		950	Set and retrieve a single C<void *> associated with a loop. When
		951	C<ev_set_userdata> has never been called, then C<ev_userdata> returns
		952	C<0.>
		953
		954	These two functions can be used to associate arbitrary data with a loop,
		955	and are intended solely for the C<invoke_pending_cb>, C<release> and
		956	C<acquire> callbacks described above, but of course can be (ab-)used for
		957	any other purpose as well.
		958
		959	=item ev_loop_verify (loop)
		960
		961	This function only does something when C<EV_VERIFY> support has been
		962	compiled in, which is the default for non-minimal builds. It tries to go
		963	through all internal structures and checks them for validity. If anything
		964	is found to be inconsistent, it will print an error message to standard
		965	error and call C<abort ()>.
		966
		967	This can be used to catch bugs inside libev itself: under normal
		968	circumstances, this function will never abort as of course libev keeps its
		969	data structures consistent.
263		970
264	=back	971	=back
265		972
		973
266	=head1 ANATOMY OF A WATCHER	974	=head1 ANATOMY OF A WATCHER
		975
		976	In the following description, uppercase C<TYPE> in names stands for the
		977	watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
		978	watchers and C<ev_io_start> for I/O watchers.
267		979
268	A watcher is a structure that you create and register to record your	980	A watcher is a structure that you create and register to record your
269	interest in some event. For instance, if you want to wait for STDIN to	981	interest in some event. For instance, if you want to wait for STDIN to
270	become readable, you would create an C<ev_io> watcher for that:	982	become readable, you would create an C<ev_io> watcher for that:
271		983
272	static void my_cb (struct ev_loop *loop, struct ev_io *w, int revents)	984	static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
273	{	985	{
274	ev_io_stop (w);	986	ev_io_stop (w);
275	ev_unloop (loop, EVUNLOOP_ALL);	987	ev_unloop (loop, EVUNLOOP_ALL);
276	}	988	}
277		989
278	struct ev_loop *loop = ev_default_loop (0);	990	struct ev_loop *loop = ev_default_loop (0);
		991
279	struct ev_io stdin_watcher;	992	ev_io stdin_watcher;
		993
280	ev_init (&stdin_watcher, my_cb);	994	ev_init (&stdin_watcher, my_cb);
281	ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);	995	ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
282	ev_io_start (loop, &stdin_watcher);	996	ev_io_start (loop, &stdin_watcher);
		997
283	ev_loop (loop, 0);	998	ev_loop (loop, 0);
284		999
285	As you can see, you are responsible for allocating the memory for your	1000	As you can see, you are responsible for allocating the memory for your
286	watcher structures (and it is usually a bad idea to do this on the stack,	1001	watcher structures (and it is I<usually> a bad idea to do this on the
287	although this can sometimes be quite valid).	1002	stack).
		1003
		1004	Each watcher has an associated watcher structure (called C<struct ev_TYPE>
		1005	or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
288		1006
289	Each watcher structure must be initialised by a call to C<ev_init	1007	Each watcher structure must be initialised by a call to C<ev_init
290	(watcher *, callback)>, which expects a callback to be provided. This	1008	(watcher *, callback)>, which expects a callback to be provided. This
291	callback gets invoked each time the event occurs (or, in the case of io	1009	callback gets invoked each time the event occurs (or, in the case of I/O
292	watchers, each time the event loop detects that the file descriptor given	1010	watchers, each time the event loop detects that the file descriptor given
293	is readable and/or writable).	1011	is readable and/or writable).
294		1012
295	Each watcher type has its own C<< ev_<type>_set (watcher *, ...) >> macro	1013	Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
296	with arguments specific to this watcher type. There is also a macro	1014	macro to configure it, with arguments specific to the watcher type. There
297	to combine initialisation and setting in one call: C<< ev_<type>_init	1015	is also a macro to combine initialisation and setting in one call: C<<
298	(watcher *, callback, ...) >>.	1016	ev_TYPE_init (watcher *, callback, ...) >>.
299		1017
300	To make the watcher actually watch out for events, you have to start it	1018	To make the watcher actually watch out for events, you have to start it
301	with a watcher-specific start function (C<< ev_<type>_start (loop, watcher	1019	with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
302	*) >>), and you can stop watching for events at any time by calling the	1020	*) >>), and you can stop watching for events at any time by calling the
303	corresponding stop function (C<< ev_<type>_stop (loop, watcher *) >>.	1021	corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
304		1022
305	As long as your watcher is active (has been started but not stopped) you	1023	As long as your watcher is active (has been started but not stopped) you
306	must not touch the values stored in it. Most specifically you must never	1024	must not touch the values stored in it. Most specifically you must never
307	reinitialise it or call its set method.	1025	reinitialise it or call its C<ev_TYPE_set> macro.
308
309	You can check whether an event is active by calling the C<ev_is_active
310	(watcher *)> macro. To see whether an event is outstanding (but the
311	callback for it has not been called yet) you can use the C<ev_is_pending
312	(watcher *)> macro.
313		1026
314	Each and every callback receives the event loop pointer as first, the	1027	Each and every callback receives the event loop pointer as first, the
315	registered watcher structure as second, and a bitset of received events as	1028	registered watcher structure as second, and a bitset of received events as
316	third argument.	1029	third argument.
317		1030
…		…
326	=item C<EV_WRITE>	1039	=item C<EV_WRITE>
327		1040
328	The file descriptor in the C<ev_io> watcher has become readable and/or	1041	The file descriptor in the C<ev_io> watcher has become readable and/or
329	writable.	1042	writable.
330		1043
331	=item C<EV_TIMEOUT>	1044	=item C<EV_TIMER>
332		1045
333	The C<ev_timer> watcher has timed out.	1046	The C<ev_timer> watcher has timed out.
334		1047
335	=item C<EV_PERIODIC>	1048	=item C<EV_PERIODIC>
336		1049
…		…
341	The signal specified in the C<ev_signal> watcher has been received by a thread.	1054	The signal specified in the C<ev_signal> watcher has been received by a thread.
342		1055
343	=item C<EV_CHILD>	1056	=item C<EV_CHILD>
344		1057
345	The pid specified in the C<ev_child> watcher has received a status change.	1058	The pid specified in the C<ev_child> watcher has received a status change.
		1059
		1060	=item C<EV_STAT>
		1061
		1062	The path specified in the C<ev_stat> watcher changed its attributes somehow.
346		1063
347	=item C<EV_IDLE>	1064	=item C<EV_IDLE>
348		1065
349	The C<ev_idle> watcher has determined that you have nothing better to do.	1066	The C<ev_idle> watcher has determined that you have nothing better to do.
350		1067
…		…
358	received events. Callbacks of both watcher types can start and stop as	1075	received events. Callbacks of both watcher types can start and stop as
359	many watchers as they want, and all of them will be taken into account	1076	many watchers as they want, and all of them will be taken into account
360	(for example, a C<ev_prepare> watcher might start an idle watcher to keep	1077	(for example, a C<ev_prepare> watcher might start an idle watcher to keep
361	C<ev_loop> from blocking).	1078	C<ev_loop> from blocking).
362		1079
		1080	=item C<EV_EMBED>
		1081
		1082	The embedded event loop specified in the C<ev_embed> watcher needs attention.
		1083
		1084	=item C<EV_FORK>
		1085
		1086	The event loop has been resumed in the child process after fork (see
		1087	C<ev_fork>).
		1088
		1089	=item C<EV_ASYNC>
		1090
		1091	The given async watcher has been asynchronously notified (see C<ev_async>).
		1092
		1093	=item C<EV_CUSTOM>
		1094
		1095	Not ever sent (or otherwise used) by libev itself, but can be freely used
		1096	by libev users to signal watchers (e.g. via C<ev_feed_event>).
		1097
363	=item C<EV_ERROR>	1098	=item C<EV_ERROR>
364		1099
365	An unspecified error has occured, the watcher has been stopped. This might	1100	An unspecified error has occurred, the watcher has been stopped. This might
366	happen because the watcher could not be properly started because libev	1101	happen because the watcher could not be properly started because libev
367	ran out of memory, a file descriptor was found to be closed or any other	1102	ran out of memory, a file descriptor was found to be closed or any other
		1103	problem. Libev considers these application bugs.
		1104
368	problem. You best act on it by reporting the problem and somehow coping	1105	You best act on it by reporting the problem and somehow coping with the
369	with the watcher being stopped.	1106	watcher being stopped. Note that well-written programs should not receive
		1107	an error ever, so when your watcher receives it, this usually indicates a
		1108	bug in your program.
370		1109
371	Libev will usually signal a few "dummy" events together with an error,	1110	Libev will usually signal a few "dummy" events together with an error, for
372	for example it might indicate that a fd is readable or writable, and if	1111	example it might indicate that a fd is readable or writable, and if your
373	your callbacks is well-written it can just attempt the operation and cope	1112	callbacks is well-written it can just attempt the operation and cope with
374	with the error from read() or write(). This will not work in multithreaded	1113	the error from read() or write(). This will not work in multi-threaded
375	programs, though, so beware.	1114	programs, though, as the fd could already be closed and reused for another
		1115	thing, so beware.
376		1116
377	=back	1117	=back
378		1118
		1119	=head2 GENERIC WATCHER FUNCTIONS
		1120
		1121	=over 4
		1122
		1123	=item C<ev_init> (ev_TYPE *watcher, callback)
		1124
		1125	This macro initialises the generic portion of a watcher. The contents
		1126	of the watcher object can be arbitrary (so C<malloc> will do). Only
		1127	the generic parts of the watcher are initialised, you I<need> to call
		1128	the type-specific C<ev_TYPE_set> macro afterwards to initialise the
		1129	type-specific parts. For each type there is also a C<ev_TYPE_init> macro
		1130	which rolls both calls into one.
		1131
		1132	You can reinitialise a watcher at any time as long as it has been stopped
		1133	(or never started) and there are no pending events outstanding.
		1134
		1135	The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
		1136	int revents)>.
		1137
		1138	Example: Initialise an C<ev_io> watcher in two steps.
		1139
		1140	ev_io w;
		1141	ev_init (&w, my_cb);
		1142	ev_io_set (&w, STDIN_FILENO, EV_READ);
		1143
		1144	=item C<ev_TYPE_set> (ev_TYPE *watcher, [args])
		1145
		1146	This macro initialises the type-specific parts of a watcher. You need to
		1147	call C<ev_init> at least once before you call this macro, but you can
		1148	call C<ev_TYPE_set> any number of times. You must not, however, call this
		1149	macro on a watcher that is active (it can be pending, however, which is a
		1150	difference to the C<ev_init> macro).
		1151
		1152	Although some watcher types do not have type-specific arguments
		1153	(e.g. C<ev_prepare>) you still need to call its C<set> macro.
		1154
		1155	See C<ev_init>, above, for an example.
		1156
		1157	=item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
		1158
		1159	This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
		1160	calls into a single call. This is the most convenient method to initialise
		1161	a watcher. The same limitations apply, of course.
		1162
		1163	Example: Initialise and set an C<ev_io> watcher in one step.
		1164
		1165	ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
		1166
		1167	=item C<ev_TYPE_start> (loop, ev_TYPE *watcher)
		1168
		1169	Starts (activates) the given watcher. Only active watchers will receive
		1170	events. If the watcher is already active nothing will happen.
		1171
		1172	Example: Start the C<ev_io> watcher that is being abused as example in this
		1173	whole section.
		1174
		1175	ev_io_start (EV_DEFAULT_UC, &w);
		1176
		1177	=item C<ev_TYPE_stop> (loop, ev_TYPE *watcher)
		1178
		1179	Stops the given watcher if active, and clears the pending status (whether
		1180	the watcher was active or not).
		1181
		1182	It is possible that stopped watchers are pending - for example,
		1183	non-repeating timers are being stopped when they become pending - but
		1184	calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
		1185	pending. If you want to free or reuse the memory used by the watcher it is
		1186	therefore a good idea to always call its C<ev_TYPE_stop> function.
		1187
		1188	=item bool ev_is_active (ev_TYPE *watcher)
		1189
		1190	Returns a true value iff the watcher is active (i.e. it has been started
		1191	and not yet been stopped). As long as a watcher is active you must not modify
		1192	it.
		1193
		1194	=item bool ev_is_pending (ev_TYPE *watcher)
		1195
		1196	Returns a true value iff the watcher is pending, (i.e. it has outstanding
		1197	events but its callback has not yet been invoked). As long as a watcher
		1198	is pending (but not active) you must not call an init function on it (but
		1199	C<ev_TYPE_set> is safe), you must not change its priority, and you must
		1200	make sure the watcher is available to libev (e.g. you cannot C<free ()>
		1201	it).
		1202
		1203	=item callback ev_cb (ev_TYPE *watcher)
		1204
		1205	Returns the callback currently set on the watcher.
		1206
		1207	=item ev_cb_set (ev_TYPE *watcher, callback)
		1208
		1209	Change the callback. You can change the callback at virtually any time
		1210	(modulo threads).
		1211
		1212	=item ev_set_priority (ev_TYPE *watcher, int priority)
		1213
		1214	=item int ev_priority (ev_TYPE *watcher)
		1215
		1216	Set and query the priority of the watcher. The priority is a small
		1217	integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
		1218	(default: C<-2>). Pending watchers with higher priority will be invoked
		1219	before watchers with lower priority, but priority will not keep watchers
		1220	from being executed (except for C<ev_idle> watchers).
		1221
		1222	If you need to suppress invocation when higher priority events are pending
		1223	you need to look at C<ev_idle> watchers, which provide this functionality.
		1224
		1225	You I<must not> change the priority of a watcher as long as it is active or
		1226	pending.
		1227
		1228	Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
		1229	fine, as long as you do not mind that the priority value you query might
		1230	or might not have been clamped to the valid range.
		1231
		1232	The default priority used by watchers when no priority has been set is
		1233	always C<0>, which is supposed to not be too high and not be too low :).
		1234
		1235	See L<WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
		1236	priorities.
		1237
		1238	=item ev_invoke (loop, ev_TYPE *watcher, int revents)
		1239
		1240	Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
		1241	C<loop> nor C<revents> need to be valid as long as the watcher callback
		1242	can deal with that fact, as both are simply passed through to the
		1243	callback.
		1244
		1245	=item int ev_clear_pending (loop, ev_TYPE *watcher)
		1246
		1247	If the watcher is pending, this function clears its pending status and
		1248	returns its C<revents> bitset (as if its callback was invoked). If the
		1249	watcher isn't pending it does nothing and returns C<0>.
		1250
		1251	Sometimes it can be useful to "poll" a watcher instead of waiting for its
		1252	callback to be invoked, which can be accomplished with this function.
		1253
		1254	=item ev_feed_event (loop, ev_TYPE *watcher, int revents)
		1255
		1256	Feeds the given event set into the event loop, as if the specified event
		1257	had happened for the specified watcher (which must be a pointer to an
		1258	initialised but not necessarily started event watcher). Obviously you must
		1259	not free the watcher as long as it has pending events.
		1260
		1261	Stopping the watcher, letting libev invoke it, or calling
		1262	C<ev_clear_pending> will clear the pending event, even if the watcher was
		1263	not started in the first place.
		1264
		1265	See also C<ev_feed_fd_event> and C<ev_feed_signal_event> for related
		1266	functions that do not need a watcher.
		1267
		1268	=back
		1269
		1270
379	=head2 ASSOCIATING CUSTOM DATA WITH A WATCHER	1271	=head2 ASSOCIATING CUSTOM DATA WITH A WATCHER
380		1272
381	Each watcher has, by default, a member C<void *data> that you can change	1273	Each watcher has, by default, a member C<void *data> that you can change
382	and read at any time, libev will completely ignore it. This can be used	1274	and read at any time: libev will completely ignore it. This can be used
383	to associate arbitrary data with your watcher. If you need more data and	1275	to associate arbitrary data with your watcher. If you need more data and
384	don't want to allocate memory and store a pointer to it in that data	1276	don't want to allocate memory and store a pointer to it in that data
385	member, you can also "subclass" the watcher type and provide your own	1277	member, you can also "subclass" the watcher type and provide your own
386	data:	1278	data:
387		1279
388	struct my_io	1280	struct my_io
389	{	1281	{
390	struct ev_io io;	1282	ev_io io;
391	int otherfd;	1283	int otherfd;
392	void *somedata;	1284	void *somedata;
393	struct whatever *mostinteresting;	1285	struct whatever *mostinteresting;
394	}	1286	};
		1287
		1288	...
		1289	struct my_io w;
		1290	ev_io_init (&w.io, my_cb, fd, EV_READ);
395		1291
396	And since your callback will be called with a pointer to the watcher, you	1292	And since your callback will be called with a pointer to the watcher, you
397	can cast it back to your own type:	1293	can cast it back to your own type:
398		1294
399	static void my_cb (struct ev_loop *loop, struct ev_io *w_, int revents)	1295	static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
400	{	1296	{
401	struct my_io *w = (struct my_io *)w_;	1297	struct my_io *w = (struct my_io *)w_;
402	...	1298	...
403	}	1299	}
404		1300
405	More interesting and less C-conformant ways of catsing your callback type	1301	More interesting and less C-conformant ways of casting your callback type
406	have been omitted....	1302	instead have been omitted.
		1303
		1304	Another common scenario is to use some data structure with multiple
		1305	embedded watchers:
		1306
		1307	struct my_biggy
		1308	{
		1309	int some_data;
		1310	ev_timer t1;
		1311	ev_timer t2;
		1312	}
		1313
		1314	In this case getting the pointer to C<my_biggy> is a bit more
		1315	complicated: Either you store the address of your C<my_biggy> struct
		1316	in the C<data> member of the watcher (for woozies), or you need to use
		1317	some pointer arithmetic using C<offsetof> inside your watchers (for real
		1318	programmers):
		1319
		1320	#include <stddef.h>
		1321
		1322	static void
		1323	t1_cb (EV_P_ ev_timer *w, int revents)
		1324	{
		1325	struct my_biggy big = (struct my_biggy *)
		1326	(((char *)w) - offsetof (struct my_biggy, t1));
		1327	}
		1328
		1329	static void
		1330	t2_cb (EV_P_ ev_timer *w, int revents)
		1331	{
		1332	struct my_biggy big = (struct my_biggy *)
		1333	(((char *)w) - offsetof (struct my_biggy, t2));
		1334	}
		1335
		1336	=head2 WATCHER PRIORITY MODELS
		1337
		1338	Many event loops support I<watcher priorities>, which are usually small
		1339	integers that influence the ordering of event callback invocation
		1340	between watchers in some way, all else being equal.
		1341
		1342	In libev, Watcher priorities can be set using C<ev_set_priority>. See its
		1343	description for the more technical details such as the actual priority
		1344	range.
		1345
		1346	There are two common ways how these these priorities are being interpreted
		1347	by event loops:
		1348
		1349	In the more common lock-out model, higher priorities "lock out" invocation
		1350	of lower priority watchers, which means as long as higher priority
		1351	watchers receive events, lower priority watchers are not being invoked.
		1352
		1353	The less common only-for-ordering model uses priorities solely to order
		1354	callback invocation within a single event loop iteration: Higher priority
		1355	watchers are invoked before lower priority ones, but they all get invoked
		1356	before polling for new events.
		1357
		1358	Libev uses the second (only-for-ordering) model for all its watchers
		1359	except for idle watchers (which use the lock-out model).
		1360
		1361	The rationale behind this is that implementing the lock-out model for
		1362	watchers is not well supported by most kernel interfaces, and most event
		1363	libraries will just poll for the same events again and again as long as
		1364	their callbacks have not been executed, which is very inefficient in the
		1365	common case of one high-priority watcher locking out a mass of lower
		1366	priority ones.
		1367
		1368	Static (ordering) priorities are most useful when you have two or more
		1369	watchers handling the same resource: a typical usage example is having an
		1370	C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
		1371	timeouts. Under load, data might be received while the program handles
		1372	other jobs, but since timers normally get invoked first, the timeout
		1373	handler will be executed before checking for data. In that case, giving
		1374	the timer a lower priority than the I/O watcher ensures that I/O will be
		1375	handled first even under adverse conditions (which is usually, but not
		1376	always, what you want).
		1377
		1378	Since idle watchers use the "lock-out" model, meaning that idle watchers
		1379	will only be executed when no same or higher priority watchers have
		1380	received events, they can be used to implement the "lock-out" model when
		1381	required.
		1382
		1383	For example, to emulate how many other event libraries handle priorities,
		1384	you can associate an C<ev_idle> watcher to each such watcher, and in
		1385	the normal watcher callback, you just start the idle watcher. The real
		1386	processing is done in the idle watcher callback. This causes libev to
		1387	continuously poll and process kernel event data for the watcher, but when
		1388	the lock-out case is known to be rare (which in turn is rare :), this is
		1389	workable.
		1390
		1391	Usually, however, the lock-out model implemented that way will perform
		1392	miserably under the type of load it was designed to handle. In that case,
		1393	it might be preferable to stop the real watcher before starting the
		1394	idle watcher, so the kernel will not have to process the event in case
		1395	the actual processing will be delayed for considerable time.
		1396
		1397	Here is an example of an I/O watcher that should run at a strictly lower
		1398	priority than the default, and which should only process data when no
		1399	other events are pending:
		1400
		1401	ev_idle idle; // actual processing watcher
		1402	ev_io io; // actual event watcher
		1403
		1404	static void
		1405	io_cb (EV_P_ ev_io *w, int revents)
		1406	{
		1407	// stop the I/O watcher, we received the event, but
		1408	// are not yet ready to handle it.
		1409	ev_io_stop (EV_A_ w);
		1410
		1411	// start the idle watcher to handle the actual event.
		1412	// it will not be executed as long as other watchers
		1413	// with the default priority are receiving events.
		1414	ev_idle_start (EV_A_ &idle);
		1415	}
		1416
		1417	static void
		1418	idle_cb (EV_P_ ev_idle *w, int revents)
		1419	{
		1420	// actual processing
		1421	read (STDIN_FILENO, ...);
		1422
		1423	// have to start the I/O watcher again, as
		1424	// we have handled the event
		1425	ev_io_start (EV_P_ &io);
		1426	}
		1427
		1428	// initialisation
		1429	ev_idle_init (&idle, idle_cb);
		1430	ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
		1431	ev_io_start (EV_DEFAULT_ &io);
		1432
		1433	In the "real" world, it might also be beneficial to start a timer, so that
		1434	low-priority connections can not be locked out forever under load. This
		1435	enables your program to keep a lower latency for important connections
		1436	during short periods of high load, while not completely locking out less
		1437	important ones.
407		1438
408		1439
409	=head1 WATCHER TYPES	1440	=head1 WATCHER TYPES
410		1441
411	This section describes each watcher in detail, but will not repeat	1442	This section describes each watcher in detail, but will not repeat
412	information given in the last section.	1443	information given in the last section. Any initialisation/set macros,
		1444	functions and members specific to the watcher type are explained.
413		1445
		1446	Members are additionally marked with either I<[read-only]>, meaning that,
		1447	while the watcher is active, you can look at the member and expect some
		1448	sensible content, but you must not modify it (you can modify it while the
		1449	watcher is stopped to your hearts content), or I<[read-write]>, which
		1450	means you can expect it to have some sensible content while the watcher
		1451	is active, but you can also modify it. Modifying it may not do something
		1452	sensible or take immediate effect (or do anything at all), but libev will
		1453	not crash or malfunction in any way.
		1454
		1455
414	=head2 C<ev_io> - is this file descriptor readable or writable	1456	=head2 C<ev_io> - is this file descriptor readable or writable?
415		1457
416	I/O watchers check whether a file descriptor is readable or writable	1458	I/O watchers check whether a file descriptor is readable or writable
417	in each iteration of the event loop (This behaviour is called	1459	in each iteration of the event loop, or, more precisely, when reading
418	level-triggering because you keep receiving events as long as the	1460	would not block the process and writing would at least be able to write
419	condition persists. Remember you can stop the watcher if you don't want to	1461	some data. This behaviour is called level-triggering because you keep
420	act on the event and neither want to receive future events).	1462	receiving events as long as the condition persists. Remember you can stop
		1463	the watcher if you don't want to act on the event and neither want to
		1464	receive future events.
421		1465
422	In general you can register as many read and/or write event watchers oer	1466	In general you can register as many read and/or write event watchers per
423	fd as you want (as long as you don't confuse yourself). Setting all file	1467	fd as you want (as long as you don't confuse yourself). Setting all file
424	descriptors to non-blocking mode is also usually a good idea (but not	1468	descriptors to non-blocking mode is also usually a good idea (but not
425	required if you know what you are doing).	1469	required if you know what you are doing).
426		1470
427	You have to be careful with dup'ed file descriptors, though. Some backends	1471	If you cannot use non-blocking mode, then force the use of a
428	(the linux epoll backend is a notable example) cannot handle dup'ed file	1472	known-to-be-good backend (at the time of this writing, this includes only
429	descriptors correctly if you register interest in two or more fds pointing	1473	C<EVBACKEND_SELECT> and C<EVBACKEND_POLL>). The same applies to file
430	to the same file/socket etc. description (that is, they share the same	1474	descriptors for which non-blocking operation makes no sense (such as
431	underlying "file open").	1475	files) - libev doesn't guarantee any specific behaviour in that case.
432		1476
433	If you must do this, then force the use of a known-to-be-good backend	1477	Another thing you have to watch out for is that it is quite easy to
434	(at the time of this writing, this includes only EVMETHOD_SELECT and	1478	receive "spurious" readiness notifications, that is your callback might
435	EVMETHOD_POLL).	1479	be called with C<EV_READ> but a subsequent C<read>(2) will actually block
		1480	because there is no data. Not only are some backends known to create a
		1481	lot of those (for example Solaris ports), it is very easy to get into
		1482	this situation even with a relatively standard program structure. Thus
		1483	it is best to always use non-blocking I/O: An extra C<read>(2) returning
		1484	C<EAGAIN> is far preferable to a program hanging until some data arrives.
		1485
		1486	If you cannot run the fd in non-blocking mode (for example you should
		1487	not play around with an Xlib connection), then you have to separately
		1488	re-test whether a file descriptor is really ready with a known-to-be good
		1489	interface such as poll (fortunately in our Xlib example, Xlib already
		1490	does this on its own, so its quite safe to use). Some people additionally
		1491	use C<SIGALRM> and an interval timer, just to be sure you won't block
		1492	indefinitely.
		1493
		1494	But really, best use non-blocking mode.
		1495
		1496	=head3 The special problem of disappearing file descriptors
		1497
		1498	Some backends (e.g. kqueue, epoll) need to be told about closing a file
		1499	descriptor (either due to calling C<close> explicitly or any other means,
		1500	such as C<dup2>). The reason is that you register interest in some file
		1501	descriptor, but when it goes away, the operating system will silently drop
		1502	this interest. If another file descriptor with the same number then is
		1503	registered with libev, there is no efficient way to see that this is, in
		1504	fact, a different file descriptor.
		1505
		1506	To avoid having to explicitly tell libev about such cases, libev follows
		1507	the following policy: Each time C<ev_io_set> is being called, libev
		1508	will assume that this is potentially a new file descriptor, otherwise
		1509	it is assumed that the file descriptor stays the same. That means that
		1510	you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
		1511	descriptor even if the file descriptor number itself did not change.
		1512
		1513	This is how one would do it normally anyway, the important point is that
		1514	the libev application should not optimise around libev but should leave
		1515	optimisations to libev.
		1516
		1517	=head3 The special problem of dup'ed file descriptors
		1518
		1519	Some backends (e.g. epoll), cannot register events for file descriptors,
		1520	but only events for the underlying file descriptions. That means when you
		1521	have C<dup ()>'ed file descriptors or weirder constellations, and register
		1522	events for them, only one file descriptor might actually receive events.
		1523
		1524	There is no workaround possible except not registering events
		1525	for potentially C<dup ()>'ed file descriptors, or to resort to
		1526	C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
		1527
		1528	=head3 The special problem of fork
		1529
		1530	Some backends (epoll, kqueue) do not support C<fork ()> at all or exhibit
		1531	useless behaviour. Libev fully supports fork, but needs to be told about
		1532	it in the child.
		1533
		1534	To support fork in your programs, you either have to call
		1535	C<ev_default_fork ()> or C<ev_loop_fork ()> after a fork in the child,
		1536	enable C<EVFLAG_FORKCHECK>, or resort to C<EVBACKEND_SELECT> or
		1537	C<EVBACKEND_POLL>.
		1538
		1539	=head3 The special problem of SIGPIPE
		1540
		1541	While not really specific to libev, it is easy to forget about C<SIGPIPE>:
		1542	when writing to a pipe whose other end has been closed, your program gets
		1543	sent a SIGPIPE, which, by default, aborts your program. For most programs
		1544	this is sensible behaviour, for daemons, this is usually undesirable.
		1545
		1546	So when you encounter spurious, unexplained daemon exits, make sure you
		1547	ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
		1548	somewhere, as that would have given you a big clue).
		1549
		1550	=head3 The special problem of accept()ing when you can't
		1551
		1552	Many implementations of the POSIX C<accept> function (for example,
		1553	found in post-2004 Linux) have the peculiar behaviour of not removing a
		1554	connection from the pending queue in all error cases.
		1555
		1556	For example, larger servers often run out of file descriptors (because
		1557	of resource limits), causing C<accept> to fail with C<ENFILE> but not
		1558	rejecting the connection, leading to libev signalling readiness on
		1559	the next iteration again (the connection still exists after all), and
		1560	typically causing the program to loop at 100% CPU usage.
		1561
		1562	Unfortunately, the set of errors that cause this issue differs between
		1563	operating systems, there is usually little the app can do to remedy the
		1564	situation, and no known thread-safe method of removing the connection to
		1565	cope with overload is known (to me).
		1566
		1567	One of the easiest ways to handle this situation is to just ignore it
		1568	- when the program encounters an overload, it will just loop until the
		1569	situation is over. While this is a form of busy waiting, no OS offers an
		1570	event-based way to handle this situation, so it's the best one can do.
		1571
		1572	A better way to handle the situation is to log any errors other than
		1573	C<EAGAIN> and C<EWOULDBLOCK>, making sure not to flood the log with such
		1574	messages, and continue as usual, which at least gives the user an idea of
		1575	what could be wrong ("raise the ulimit!"). For extra points one could stop
		1576	the C<ev_io> watcher on the listening fd "for a while", which reduces CPU
		1577	usage.
		1578
		1579	If your program is single-threaded, then you could also keep a dummy file
		1580	descriptor for overload situations (e.g. by opening F</dev/null>), and
		1581	when you run into C<ENFILE> or C<EMFILE>, close it, run C<accept>,
		1582	close that fd, and create a new dummy fd. This will gracefully refuse
		1583	clients under typical overload conditions.
		1584
		1585	The last way to handle it is to simply log the error and C<exit>, as
		1586	is often done with C<malloc> failures, but this results in an easy
		1587	opportunity for a DoS attack.
		1588
		1589	=head3 Watcher-Specific Functions
436		1590
437	=over 4	1591	=over 4
438		1592
439	=item ev_io_init (ev_io *, callback, int fd, int events)	1593	=item ev_io_init (ev_io *, callback, int fd, int events)
440		1594
441	=item ev_io_set (ev_io *, int fd, int events)	1595	=item ev_io_set (ev_io *, int fd, int events)
442		1596
443	Configures an C<ev_io> watcher. The fd is the file descriptor to rceeive	1597	Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
444	events for and events is either C<EV_READ>, C<EV_WRITE> or C<EV_READ |	1598	receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
445	EV_WRITE> to receive the given events.	1599	C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
		1600
		1601	=item int fd [read-only]
		1602
		1603	The file descriptor being watched.
		1604
		1605	=item int events [read-only]
		1606
		1607	The events being watched.
446		1608
447	=back	1609	=back
448		1610
		1611	=head3 Examples
		1612
		1613	Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
		1614	readable, but only once. Since it is likely line-buffered, you could
		1615	attempt to read a whole line in the callback.
		1616
		1617	static void
		1618	stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
		1619	{
		1620	ev_io_stop (loop, w);
		1621	.. read from stdin here (or from w->fd) and handle any I/O errors
		1622	}
		1623
		1624	...
		1625	struct ev_loop *loop = ev_default_init (0);
		1626	ev_io stdin_readable;
		1627	ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
		1628	ev_io_start (loop, &stdin_readable);
		1629	ev_loop (loop, 0);
		1630
		1631
449	=head2 C<ev_timer> - relative and optionally recurring timeouts	1632	=head2 C<ev_timer> - relative and optionally repeating timeouts
450		1633
451	Timer watchers are simple relative timers that generate an event after a	1634	Timer watchers are simple relative timers that generate an event after a
452	given time, and optionally repeating in regular intervals after that.	1635	given time, and optionally repeating in regular intervals after that.
453		1636
454	The timers are based on real time, that is, if you register an event that	1637	The timers are based on real time, that is, if you register an event that
455	times out after an hour and you reset your system clock to last years	1638	times out after an hour and you reset your system clock to January last
456	time, it will still time out after (roughly) and hour. "Roughly" because	1639	year, it will still time out after (roughly) one hour. "Roughly" because
457	detecting time jumps is hard, and soem inaccuracies are unavoidable (the	1640	detecting time jumps is hard, and some inaccuracies are unavoidable (the
458	monotonic clock option helps a lot here).	1641	monotonic clock option helps a lot here).
		1642
		1643	The callback is guaranteed to be invoked only I<after> its timeout has
		1644	passed (not I<at>, so on systems with very low-resolution clocks this
		1645	might introduce a small delay). If multiple timers become ready during the
		1646	same loop iteration then the ones with earlier time-out values are invoked
		1647	before ones of the same priority with later time-out values (but this is
		1648	no longer true when a callback calls C<ev_loop> recursively).
		1649
		1650	=head3 Be smart about timeouts
		1651
		1652	Many real-world problems involve some kind of timeout, usually for error
		1653	recovery. A typical example is an HTTP request - if the other side hangs,
		1654	you want to raise some error after a while.
		1655
		1656	What follows are some ways to handle this problem, from obvious and
		1657	inefficient to smart and efficient.
		1658
		1659	In the following, a 60 second activity timeout is assumed - a timeout that
		1660	gets reset to 60 seconds each time there is activity (e.g. each time some
		1661	data or other life sign was received).
		1662
		1663	=over 4
		1664
		1665	=item 1. Use a timer and stop, reinitialise and start it on activity.
		1666
		1667	This is the most obvious, but not the most simple way: In the beginning,
		1668	start the watcher:
		1669
		1670	ev_timer_init (timer, callback, 60., 0.);
		1671	ev_timer_start (loop, timer);
		1672
		1673	Then, each time there is some activity, C<ev_timer_stop> it, initialise it
		1674	and start it again:
		1675
		1676	ev_timer_stop (loop, timer);
		1677	ev_timer_set (timer, 60., 0.);
		1678	ev_timer_start (loop, timer);
		1679
		1680	This is relatively simple to implement, but means that each time there is
		1681	some activity, libev will first have to remove the timer from its internal
		1682	data structure and then add it again. Libev tries to be fast, but it's
		1683	still not a constant-time operation.
		1684
		1685	=item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
		1686
		1687	This is the easiest way, and involves using C<ev_timer_again> instead of
		1688	C<ev_timer_start>.
		1689
		1690	To implement this, configure an C<ev_timer> with a C<repeat> value
		1691	of C<60> and then call C<ev_timer_again> at start and each time you
		1692	successfully read or write some data. If you go into an idle state where
		1693	you do not expect data to travel on the socket, you can C<ev_timer_stop>
		1694	the timer, and C<ev_timer_again> will automatically restart it if need be.
		1695
		1696	That means you can ignore both the C<ev_timer_start> function and the
		1697	C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
		1698	member and C<ev_timer_again>.
		1699
		1700	At start:
		1701
		1702	ev_init (timer, callback);
		1703	timer->repeat = 60.;
		1704	ev_timer_again (loop, timer);
		1705
		1706	Each time there is some activity:
		1707
		1708	ev_timer_again (loop, timer);
		1709
		1710	It is even possible to change the time-out on the fly, regardless of
		1711	whether the watcher is active or not:
		1712
		1713	timer->repeat = 30.;
		1714	ev_timer_again (loop, timer);
		1715
		1716	This is slightly more efficient then stopping/starting the timer each time
		1717	you want to modify its timeout value, as libev does not have to completely
		1718	remove and re-insert the timer from/into its internal data structure.
		1719
		1720	It is, however, even simpler than the "obvious" way to do it.
		1721
		1722	=item 3. Let the timer time out, but then re-arm it as required.
		1723
		1724	This method is more tricky, but usually most efficient: Most timeouts are
		1725	relatively long compared to the intervals between other activity - in
		1726	our example, within 60 seconds, there are usually many I/O events with
		1727	associated activity resets.
		1728
		1729	In this case, it would be more efficient to leave the C<ev_timer> alone,
		1730	but remember the time of last activity, and check for a real timeout only
		1731	within the callback:
		1732
		1733	ev_tstamp last_activity; // time of last activity
		1734
		1735	static void
		1736	callback (EV_P_ ev_timer *w, int revents)
		1737	{
		1738	ev_tstamp now = ev_now (EV_A);
		1739	ev_tstamp timeout = last_activity + 60.;
		1740
		1741	// if last_activity + 60. is older than now, we did time out
		1742	if (timeout < now)
		1743	{
		1744	// timeout occurred, take action
		1745	}
		1746	else
		1747	{
		1748	// callback was invoked, but there was some activity, re-arm
		1749	// the watcher to fire in last_activity + 60, which is
		1750	// guaranteed to be in the future, so "again" is positive:
		1751	w->repeat = timeout - now;
		1752	ev_timer_again (EV_A_ w);
		1753	}
		1754	}
		1755
		1756	To summarise the callback: first calculate the real timeout (defined
		1757	as "60 seconds after the last activity"), then check if that time has
		1758	been reached, which means something I<did>, in fact, time out. Otherwise
		1759	the callback was invoked too early (C<timeout> is in the future), so
		1760	re-schedule the timer to fire at that future time, to see if maybe we have
		1761	a timeout then.
		1762
		1763	Note how C<ev_timer_again> is used, taking advantage of the
		1764	C<ev_timer_again> optimisation when the timer is already running.
		1765
		1766	This scheme causes more callback invocations (about one every 60 seconds
		1767	minus half the average time between activity), but virtually no calls to
		1768	libev to change the timeout.
		1769
		1770	To start the timer, simply initialise the watcher and set C<last_activity>
		1771	to the current time (meaning we just have some activity :), then call the
		1772	callback, which will "do the right thing" and start the timer:
		1773
		1774	ev_init (timer, callback);
		1775	last_activity = ev_now (loop);
		1776	callback (loop, timer, EV_TIMER);
		1777
		1778	And when there is some activity, simply store the current time in
		1779	C<last_activity>, no libev calls at all:
		1780
		1781	last_activity = ev_now (loop);
		1782
		1783	This technique is slightly more complex, but in most cases where the
		1784	time-out is unlikely to be triggered, much more efficient.
		1785
		1786	Changing the timeout is trivial as well (if it isn't hard-coded in the
		1787	callback :) - just change the timeout and invoke the callback, which will
		1788	fix things for you.
		1789
		1790	=item 4. Wee, just use a double-linked list for your timeouts.
		1791
		1792	If there is not one request, but many thousands (millions...), all
		1793	employing some kind of timeout with the same timeout value, then one can
		1794	do even better:
		1795
		1796	When starting the timeout, calculate the timeout value and put the timeout
		1797	at the I<end> of the list.
		1798
		1799	Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
		1800	the list is expected to fire (for example, using the technique #3).
		1801
		1802	When there is some activity, remove the timer from the list, recalculate
		1803	the timeout, append it to the end of the list again, and make sure to
		1804	update the C<ev_timer> if it was taken from the beginning of the list.
		1805
		1806	This way, one can manage an unlimited number of timeouts in O(1) time for
		1807	starting, stopping and updating the timers, at the expense of a major
		1808	complication, and having to use a constant timeout. The constant timeout
		1809	ensures that the list stays sorted.
		1810
		1811	=back
		1812
		1813	So which method the best?
		1814
		1815	Method #2 is a simple no-brain-required solution that is adequate in most
		1816	situations. Method #3 requires a bit more thinking, but handles many cases
		1817	better, and isn't very complicated either. In most case, choosing either
		1818	one is fine, with #3 being better in typical situations.
		1819
		1820	Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
		1821	rather complicated, but extremely efficient, something that really pays
		1822	off after the first million or so of active timers, i.e. it's usually
		1823	overkill :)
		1824
		1825	=head3 The special problem of time updates
		1826
		1827	Establishing the current time is a costly operation (it usually takes at
		1828	least two system calls): EV therefore updates its idea of the current
		1829	time only before and after C<ev_loop> collects new events, which causes a
		1830	growing difference between C<ev_now ()> and C<ev_time ()> when handling
		1831	lots of events in one iteration.
459		1832
460	The relative timeouts are calculated relative to the C<ev_now ()>	1833	The relative timeouts are calculated relative to the C<ev_now ()>
461	time. This is usually the right thing as this timestamp refers to the time	1834	time. This is usually the right thing as this timestamp refers to the time
462	of the event triggering whatever timeout you are modifying/starting. If	1835	of the event triggering whatever timeout you are modifying/starting. If
463	you suspect event processing to be delayed and you *need* to base the timeout	1836	you suspect event processing to be delayed and you I<need> to base the
464	on the current time, use something like this to adjust for this:	1837	timeout on the current time, use something like this to adjust for this:
465		1838
466	ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);	1839	ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);
467		1840
		1841	If the event loop is suspended for a long time, you can also force an
		1842	update of the time returned by C<ev_now ()> by calling C<ev_now_update
		1843	()>.
		1844
		1845	=head3 The special problems of suspended animation
		1846
		1847	When you leave the server world it is quite customary to hit machines that
		1848	can suspend/hibernate - what happens to the clocks during such a suspend?
		1849
		1850	Some quick tests made with a Linux 2.6.28 indicate that a suspend freezes
		1851	all processes, while the clocks (C<times>, C<CLOCK_MONOTONIC>) continue
		1852	to run until the system is suspended, but they will not advance while the
		1853	system is suspended. That means, on resume, it will be as if the program
		1854	was frozen for a few seconds, but the suspend time will not be counted
		1855	towards C<ev_timer> when a monotonic clock source is used. The real time
		1856	clock advanced as expected, but if it is used as sole clocksource, then a
		1857	long suspend would be detected as a time jump by libev, and timers would
		1858	be adjusted accordingly.
		1859
		1860	I would not be surprised to see different behaviour in different between
		1861	operating systems, OS versions or even different hardware.
		1862
		1863	The other form of suspend (job control, or sending a SIGSTOP) will see a
		1864	time jump in the monotonic clocks and the realtime clock. If the program
		1865	is suspended for a very long time, and monotonic clock sources are in use,
		1866	then you can expect C<ev_timer>s to expire as the full suspension time
		1867	will be counted towards the timers. When no monotonic clock source is in
		1868	use, then libev will again assume a timejump and adjust accordingly.
		1869
		1870	It might be beneficial for this latter case to call C<ev_suspend>
		1871	and C<ev_resume> in code that handles C<SIGTSTP>, to at least get
		1872	deterministic behaviour in this case (you can do nothing against
		1873	C<SIGSTOP>).
		1874
		1875	=head3 Watcher-Specific Functions and Data Members
		1876
468	=over 4	1877	=over 4
469		1878
470	=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)	1879	=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
471		1880
472	=item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)	1881	=item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
473		1882
474	Configure the timer to trigger after C<after> seconds. If C<repeat> is	1883	Configure the timer to trigger after C<after> seconds. If C<repeat>
475	C<0.>, then it will automatically be stopped. If it is positive, then the	1884	is C<0.>, then it will automatically be stopped once the timeout is
476	timer will automatically be configured to trigger again C<repeat> seconds	1885	reached. If it is positive, then the timer will automatically be
477	later, again, and again, until stopped manually.	1886	configured to trigger again C<repeat> seconds later, again, and again,
		1887	until stopped manually.
478		1888
479	The timer itself will do a best-effort at avoiding drift, that is, if you	1889	The timer itself will do a best-effort at avoiding drift, that is, if
480	configure a timer to trigger every 10 seconds, then it will trigger at	1890	you configure a timer to trigger every 10 seconds, then it will normally
481	exactly 10 second intervals. If, however, your program cannot keep up with	1891	trigger at exactly 10 second intervals. If, however, your program cannot
482	the timer (because it takes longer than those 10 seconds to do stuff) the	1892	keep up with the timer (because it takes longer than those 10 seconds to
483	timer will not fire more than once per event loop iteration.	1893	do stuff) the timer will not fire more than once per event loop iteration.
484		1894
485	=item ev_timer_again (loop)	1895	=item ev_timer_again (loop, ev_timer *)
486		1896
487	This will act as if the timer timed out and restart it again if it is	1897	This will act as if the timer timed out and restart it again if it is
488	repeating. The exact semantics are:	1898	repeating. The exact semantics are:
489		1899
		1900	If the timer is pending, its pending status is cleared.
		1901
490	If the timer is started but nonrepeating, stop it.	1902	If the timer is started but non-repeating, stop it (as if it timed out).
491		1903
492	If the timer is repeating, either start it if necessary (with the repeat	1904	If the timer is repeating, either start it if necessary (with the
493	value), or reset the running timer to the repeat value.	1905	C<repeat> value), or reset the running timer to the C<repeat> value.
494		1906
495	This sounds a bit complicated, but here is a useful and typical	1907	This sounds a bit complicated, see L<Be smart about timeouts>, above, for a
496	example: Imagine you have a tcp connection and you want a so-called idle	1908	usage example.
497	timeout, that is, you want to be called when there have been, say, 60	1909
498	seconds of inactivity on the socket. The easiest way to do this is to	1910	=item ev_tstamp ev_timer_remaining (loop, ev_timer *)
499	configure an C<ev_timer> with after=repeat=60 and calling ev_timer_again each	1911
500	time you successfully read or write some data. If you go into an idle	1912	Returns the remaining time until a timer fires. If the timer is active,
501	state where you do not expect data to travel on the socket, you can stop	1913	then this time is relative to the current event loop time, otherwise it's
502	the timer, and again will automatically restart it if need be.	1914	the timeout value currently configured.
		1915
		1916	That is, after an C<ev_timer_set (w, 5, 7)>, C<ev_timer_remaining> returns
		1917	C<5>. When the timer is started and one second passes, C<ev_timer_remaining>
		1918	will return C<4>. When the timer expires and is restarted, it will return
		1919	roughly C<7> (likely slightly less as callback invocation takes some time,
		1920	too), and so on.
		1921
		1922	=item ev_tstamp repeat [read-write]
		1923
		1924	The current C<repeat> value. Will be used each time the watcher times out
		1925	or C<ev_timer_again> is called, and determines the next timeout (if any),
		1926	which is also when any modifications are taken into account.
503		1927
504	=back	1928	=back
505		1929
		1930	=head3 Examples
		1931
		1932	Example: Create a timer that fires after 60 seconds.
		1933
		1934	static void
		1935	one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
		1936	{
		1937	.. one minute over, w is actually stopped right here
		1938	}
		1939
		1940	ev_timer mytimer;
		1941	ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
		1942	ev_timer_start (loop, &mytimer);
		1943
		1944	Example: Create a timeout timer that times out after 10 seconds of
		1945	inactivity.
		1946
		1947	static void
		1948	timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
		1949	{
		1950	.. ten seconds without any activity
		1951	}
		1952
		1953	ev_timer mytimer;
		1954	ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
		1955	ev_timer_again (&mytimer); /* start timer */
		1956	ev_loop (loop, 0);
		1957
		1958	// and in some piece of code that gets executed on any "activity":
		1959	// reset the timeout to start ticking again at 10 seconds
		1960	ev_timer_again (&mytimer);
		1961
		1962
506	=head2 C<ev_periodic> - to cron or not to cron	1963	=head2 C<ev_periodic> - to cron or not to cron?
507		1964
508	Periodic watchers are also timers of a kind, but they are very versatile	1965	Periodic watchers are also timers of a kind, but they are very versatile
509	(and unfortunately a bit complex).	1966	(and unfortunately a bit complex).
510		1967
511	Unlike C<ev_timer>'s, they are not based on real time (or relative time)	1968	Unlike C<ev_timer>, periodic watchers are not based on real time (or
512	but on wallclock time (absolute time). You can tell a periodic watcher	1969	relative time, the physical time that passes) but on wall clock time
513	to trigger "at" some specific point in time. For example, if you tell a	1970	(absolute time, the thing you can read on your calender or clock). The
514	periodic watcher to trigger in 10 seconds (by specifiying e.g. c<ev_now ()	1971	difference is that wall clock time can run faster or slower than real
515	+ 10.>) and then reset your system clock to the last year, then it will	1972	time, and time jumps are not uncommon (e.g. when you adjust your
516	take a year to trigger the event (unlike an C<ev_timer>, which would trigger	1973	wrist-watch).
517	roughly 10 seconds later and of course not if you reset your system time
518	again).
519		1974
520	They can also be used to implement vastly more complex timers, such as	1975	You can tell a periodic watcher to trigger after some specific point
		1976	in time: for example, if you tell a periodic watcher to trigger "in 10
		1977	seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
		1978	not a delay) and then reset your system clock to January of the previous
		1979	year, then it will take a year or more to trigger the event (unlike an
		1980	C<ev_timer>, which would still trigger roughly 10 seconds after starting
		1981	it, as it uses a relative timeout).
		1982
		1983	C<ev_periodic> watchers can also be used to implement vastly more complex
521	triggering an event on eahc midnight, local time.	1984	timers, such as triggering an event on each "midnight, local time", or
		1985	other complicated rules. This cannot be done with C<ev_timer> watchers, as
		1986	those cannot react to time jumps.
		1987
		1988	As with timers, the callback is guaranteed to be invoked only when the
		1989	point in time where it is supposed to trigger has passed. If multiple
		1990	timers become ready during the same loop iteration then the ones with
		1991	earlier time-out values are invoked before ones with later time-out values
		1992	(but this is no longer true when a callback calls C<ev_loop> recursively).
		1993
		1994	=head3 Watcher-Specific Functions and Data Members
522		1995
523	=over 4	1996	=over 4
524		1997
525	=item ev_periodic_init (ev_periodic *, callback, ev_tstamp at, ev_tstamp interval, reschedule_cb)	1998	=item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
526		1999
527	=item ev_periodic_set (ev_periodic *, ev_tstamp after, ev_tstamp repeat, reschedule_cb)	2000	=item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
528		2001
529	Lots of arguments, lets sort it out... There are basically three modes of	2002	Lots of arguments, let's sort it out... There are basically three modes of
530	operation, and we will explain them from simplest to complex:	2003	operation, and we will explain them from simplest to most complex:
531
532		2004
533	=over 4	2005	=over 4
534		2006
535	=item * absolute timer (interval = reschedule_cb = 0)	2007	=item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
536		2008
537	In this configuration the watcher triggers an event at the wallclock time	2009	In this configuration the watcher triggers an event after the wall clock
538	C<at> and doesn't repeat. It will not adjust when a time jump occurs,	2010	time C<offset> has passed. It will not repeat and will not adjust when a
539	that is, if it is to be run at January 1st 2011 then it will run when the	2011	time jump occurs, that is, if it is to be run at January 1st 2011 then it
540	system time reaches or surpasses this time.	2012	will be stopped and invoked when the system clock reaches or surpasses
		2013	this point in time.
541		2014
542	=item * non-repeating interval timer (interval > 0, reschedule_cb = 0)	2015	=item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
543		2016
544	In this mode the watcher will always be scheduled to time out at the next	2017	In this mode the watcher will always be scheduled to time out at the next
545	C<at + N * interval> time (for some integer N) and then repeat, regardless	2018	C<offset + N * interval> time (for some integer N, which can also be
546	of any time jumps.	2019	negative) and then repeat, regardless of any time jumps. The C<offset>
		2020	argument is merely an offset into the C<interval> periods.
547		2021
548	This can be used to create timers that do not drift with respect to system	2022	This can be used to create timers that do not drift with respect to the
549	time:	2023	system clock, for example, here is an C<ev_periodic> that triggers each
		2024	hour, on the hour (with respect to UTC):
550		2025
551	ev_periodic_set (&periodic, 0., 3600., 0);	2026	ev_periodic_set (&periodic, 0., 3600., 0);
552		2027
553	This doesn't mean there will always be 3600 seconds in between triggers,	2028	This doesn't mean there will always be 3600 seconds in between triggers,
554	but only that the the callback will be called when the system time shows a	2029	but only that the callback will be called when the system time shows a
555	full hour (UTC), or more correctly, when the system time is evenly divisible	2030	full hour (UTC), or more correctly, when the system time is evenly divisible
556	by 3600.	2031	by 3600.
557		2032
558	Another way to think about it (for the mathematically inclined) is that	2033	Another way to think about it (for the mathematically inclined) is that
559	C<ev_periodic> will try to run the callback in this mode at the next possible	2034	C<ev_periodic> will try to run the callback in this mode at the next possible
560	time where C<time = at (mod interval)>, regardless of any time jumps.	2035	time where C<time = offset (mod interval)>, regardless of any time jumps.
561		2036
		2037	For numerical stability it is preferable that the C<offset> value is near
		2038	C<ev_now ()> (the current time), but there is no range requirement for
		2039	this value, and in fact is often specified as zero.
		2040
		2041	Note also that there is an upper limit to how often a timer can fire (CPU
		2042	speed for example), so if C<interval> is very small then timing stability
		2043	will of course deteriorate. Libev itself tries to be exact to be about one
		2044	millisecond (if the OS supports it and the machine is fast enough).
		2045
562	=item * manual reschedule mode (reschedule_cb = callback)	2046	=item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
563		2047
564	In this mode the values for C<interval> and C<at> are both being	2048	In this mode the values for C<interval> and C<offset> are both being
565	ignored. Instead, each time the periodic watcher gets scheduled, the	2049	ignored. Instead, each time the periodic watcher gets scheduled, the
566	reschedule callback will be called with the watcher as first, and the	2050	reschedule callback will be called with the watcher as first, and the
567	current time as second argument.	2051	current time as second argument.
568		2052
569	NOTE: I<This callback MUST NOT stop or destroy any periodic watcher,	2053	NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, ever,
570	ever, or make any event loop modifications>. If you need to stop it,	2054	or make ANY other event loop modifications whatsoever, unless explicitly
571	return C<now + 1e30> (or so, fudge fudge) and stop it afterwards (e.g. by	2055	allowed by documentation here>.
572	starting a prepare watcher).
573		2056
		2057	If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
		2058	it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
		2059	only event loop modification you are allowed to do).
		2060
574	Its prototype is C<ev_tstamp (*reschedule_cb)(struct ev_periodic *w,	2061	The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
575	ev_tstamp now)>, e.g.:	2062	*w, ev_tstamp now)>, e.g.:
576		2063
		2064	static ev_tstamp
577	static ev_tstamp my_rescheduler (struct ev_periodic *w, ev_tstamp now)	2065	my_rescheduler (ev_periodic *w, ev_tstamp now)
578	{	2066	{
579	return now + 60.;	2067	return now + 60.;
580	}	2068	}
581		2069
582	It must return the next time to trigger, based on the passed time value	2070	It must return the next time to trigger, based on the passed time value
583	(that is, the lowest time value larger than to the second argument). It	2071	(that is, the lowest time value larger than to the second argument). It
584	will usually be called just before the callback will be triggered, but	2072	will usually be called just before the callback will be triggered, but
585	might be called at other times, too.	2073	might be called at other times, too.
586		2074
587	NOTE: I<< This callback must always return a time that is later than the	2075	NOTE: I<< This callback must always return a time that is higher than or
588	passed C<now> value >>. Not even C<now> itself will do, it I<must> be larger.	2076	equal to the passed C<now> value >>.
589		2077
590	This can be used to create very complex timers, such as a timer that	2078	This can be used to create very complex timers, such as a timer that
591	triggers on each midnight, local time. To do this, you would calculate the	2079	triggers on "next midnight, local time". To do this, you would calculate the
592	next midnight after C<now> and return the timestamp value for this. How	2080	next midnight after C<now> and return the timestamp value for this. How
593	you do this is, again, up to you (but it is not trivial, which is the main	2081	you do this is, again, up to you (but it is not trivial, which is the main
594	reason I omitted it as an example).	2082	reason I omitted it as an example).
595		2083
596	=back	2084	=back
…		…
600	Simply stops and restarts the periodic watcher again. This is only useful	2088	Simply stops and restarts the periodic watcher again. This is only useful
601	when you changed some parameters or the reschedule callback would return	2089	when you changed some parameters or the reschedule callback would return
602	a different time than the last time it was called (e.g. in a crond like	2090	a different time than the last time it was called (e.g. in a crond like
603	program when the crontabs have changed).	2091	program when the crontabs have changed).
604		2092
		2093	=item ev_tstamp ev_periodic_at (ev_periodic *)
		2094
		2095	When active, returns the absolute time that the watcher is supposed
		2096	to trigger next. This is not the same as the C<offset> argument to
		2097	C<ev_periodic_set>, but indeed works even in interval and manual
		2098	rescheduling modes.
		2099
		2100	=item ev_tstamp offset [read-write]
		2101
		2102	When repeating, this contains the offset value, otherwise this is the
		2103	absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
		2104	although libev might modify this value for better numerical stability).
		2105
		2106	Can be modified any time, but changes only take effect when the periodic
		2107	timer fires or C<ev_periodic_again> is being called.
		2108
		2109	=item ev_tstamp interval [read-write]
		2110
		2111	The current interval value. Can be modified any time, but changes only
		2112	take effect when the periodic timer fires or C<ev_periodic_again> is being
		2113	called.
		2114
		2115	=item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
		2116
		2117	The current reschedule callback, or C<0>, if this functionality is
		2118	switched off. Can be changed any time, but changes only take effect when
		2119	the periodic timer fires or C<ev_periodic_again> is being called.
		2120
605	=back	2121	=back
606		2122
		2123	=head3 Examples
		2124
		2125	Example: Call a callback every hour, or, more precisely, whenever the
		2126	system time is divisible by 3600. The callback invocation times have
		2127	potentially a lot of jitter, but good long-term stability.
		2128
		2129	static void
		2130	clock_cb (struct ev_loop *loop, ev_periodic *w, int revents)
		2131	{
		2132	... its now a full hour (UTC, or TAI or whatever your clock follows)
		2133	}
		2134
		2135	ev_periodic hourly_tick;
		2136	ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
		2137	ev_periodic_start (loop, &hourly_tick);
		2138
		2139	Example: The same as above, but use a reschedule callback to do it:
		2140
		2141	#include <math.h>
		2142
		2143	static ev_tstamp
		2144	my_scheduler_cb (ev_periodic *w, ev_tstamp now)
		2145	{
		2146	return now + (3600. - fmod (now, 3600.));
		2147	}
		2148
		2149	ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
		2150
		2151	Example: Call a callback every hour, starting now:
		2152
		2153	ev_periodic hourly_tick;
		2154	ev_periodic_init (&hourly_tick, clock_cb,
		2155	fmod (ev_now (loop), 3600.), 3600., 0);
		2156	ev_periodic_start (loop, &hourly_tick);
		2157
		2158
607	=head2 C<ev_signal> - signal me when a signal gets signalled	2159	=head2 C<ev_signal> - signal me when a signal gets signalled!
608		2160
609	Signal watchers will trigger an event when the process receives a specific	2161	Signal watchers will trigger an event when the process receives a specific
610	signal one or more times. Even though signals are very asynchronous, libev	2162	signal one or more times. Even though signals are very asynchronous, libev
611	will try it's best to deliver signals synchronously, i.e. as part of the	2163	will try it's best to deliver signals synchronously, i.e. as part of the
612	normal event processing, like any other event.	2164	normal event processing, like any other event.
613		2165
		2166	If you want signals to be delivered truly asynchronously, just use
		2167	C<sigaction> as you would do without libev and forget about sharing
		2168	the signal. You can even use C<ev_async> from a signal handler to
		2169	synchronously wake up an event loop.
		2170
614	You can configure as many watchers as you like per signal. Only when the	2171	You can configure as many watchers as you like for the same signal, but
		2172	only within the same loop, i.e. you can watch for C<SIGINT> in your
		2173	default loop and for C<SIGIO> in another loop, but you cannot watch for
		2174	C<SIGINT> in both the default loop and another loop at the same time. At
		2175	the moment, C<SIGCHLD> is permanently tied to the default loop.
		2176
615	first watcher gets started will libev actually register a signal watcher	2177	When the first watcher gets started will libev actually register something
616	with the kernel (thus it coexists with your own signal handlers as long	2178	with the kernel (thus it coexists with your own signal handlers as long as
617	as you don't register any with libev). Similarly, when the last signal	2179	you don't register any with libev for the same signal).
618	watcher for a signal is stopped libev will reset the signal handler to	2180
619	SIG_DFL (regardless of what it was set to before).	2181	If possible and supported, libev will install its handlers with
		2182	C<SA_RESTART> (or equivalent) behaviour enabled, so system calls should
		2183	not be unduly interrupted. If you have a problem with system calls getting
		2184	interrupted by signals you can block all signals in an C<ev_check> watcher
		2185	and unblock them in an C<ev_prepare> watcher.
		2186
		2187	=head3 The special problem of inheritance over fork/execve/pthread_create
		2188
		2189	Both the signal mask (C<sigprocmask>) and the signal disposition
		2190	(C<sigaction>) are unspecified after starting a signal watcher (and after
		2191	stopping it again), that is, libev might or might not block the signal,
		2192	and might or might not set or restore the installed signal handler.
		2193
		2194	While this does not matter for the signal disposition (libev never
		2195	sets signals to C<SIG_IGN>, so handlers will be reset to C<SIG_DFL> on
		2196	C<execve>), this matters for the signal mask: many programs do not expect
		2197	certain signals to be blocked.
		2198
		2199	This means that before calling C<exec> (from the child) you should reset
		2200	the signal mask to whatever "default" you expect (all clear is a good
		2201	choice usually).
		2202
		2203	The simplest way to ensure that the signal mask is reset in the child is
		2204	to install a fork handler with C<pthread_atfork> that resets it. That will
		2205	catch fork calls done by libraries (such as the libc) as well.
		2206
		2207	In current versions of libev, the signal will not be blocked indefinitely
		2208	unless you use the C<signalfd> API (C<EV_SIGNALFD>). While this reduces
		2209	the window of opportunity for problems, it will not go away, as libev
		2210	I<has> to modify the signal mask, at least temporarily.
		2211
		2212	So I can't stress this enough: I<If you do not reset your signal mask when
		2213	you expect it to be empty, you have a race condition in your code>. This
		2214	is not a libev-specific thing, this is true for most event libraries.
		2215
		2216	=head3 Watcher-Specific Functions and Data Members
620		2217
621	=over 4	2218	=over 4
622		2219
623	=item ev_signal_init (ev_signal *, callback, int signum)	2220	=item ev_signal_init (ev_signal *, callback, int signum)
624		2221
625	=item ev_signal_set (ev_signal *, int signum)	2222	=item ev_signal_set (ev_signal *, int signum)
626		2223
627	Configures the watcher to trigger on the given signal number (usually one	2224	Configures the watcher to trigger on the given signal number (usually one
628	of the C<SIGxxx> constants).	2225	of the C<SIGxxx> constants).
629		2226
		2227	=item int signum [read-only]
		2228
		2229	The signal the watcher watches out for.
		2230
630	=back	2231	=back
631		2232
		2233	=head3 Examples
		2234
		2235	Example: Try to exit cleanly on SIGINT.
		2236
		2237	static void
		2238	sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
		2239	{
		2240	ev_unloop (loop, EVUNLOOP_ALL);
		2241	}
		2242
		2243	ev_signal signal_watcher;
		2244	ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
		2245	ev_signal_start (loop, &signal_watcher);
		2246
		2247
632	=head2 C<ev_child> - wait for pid status changes	2248	=head2 C<ev_child> - watch out for process status changes
633		2249
634	Child watchers trigger when your process receives a SIGCHLD in response to	2250	Child watchers trigger when your process receives a SIGCHLD in response to
635	some child status changes (most typically when a child of yours dies).	2251	some child status changes (most typically when a child of yours dies or
		2252	exits). It is permissible to install a child watcher I<after> the child
		2253	has been forked (which implies it might have already exited), as long
		2254	as the event loop isn't entered (or is continued from a watcher), i.e.,
		2255	forking and then immediately registering a watcher for the child is fine,
		2256	but forking and registering a watcher a few event loop iterations later or
		2257	in the next callback invocation is not.
		2258
		2259	Only the default event loop is capable of handling signals, and therefore
		2260	you can only register child watchers in the default event loop.
		2261
		2262	Due to some design glitches inside libev, child watchers will always be
		2263	handled at maximum priority (their priority is set to C<EV_MAXPRI> by
		2264	libev)
		2265
		2266	=head3 Process Interaction
		2267
		2268	Libev grabs C<SIGCHLD> as soon as the default event loop is
		2269	initialised. This is necessary to guarantee proper behaviour even if the
		2270	first child watcher is started after the child exits. The occurrence
		2271	of C<SIGCHLD> is recorded asynchronously, but child reaping is done
		2272	synchronously as part of the event loop processing. Libev always reaps all
		2273	children, even ones not watched.
		2274
		2275	=head3 Overriding the Built-In Processing
		2276
		2277	Libev offers no special support for overriding the built-in child
		2278	processing, but if your application collides with libev's default child
		2279	handler, you can override it easily by installing your own handler for
		2280	C<SIGCHLD> after initialising the default loop, and making sure the
		2281	default loop never gets destroyed. You are encouraged, however, to use an
		2282	event-based approach to child reaping and thus use libev's support for
		2283	that, so other libev users can use C<ev_child> watchers freely.
		2284
		2285	=head3 Stopping the Child Watcher
		2286
		2287	Currently, the child watcher never gets stopped, even when the
		2288	child terminates, so normally one needs to stop the watcher in the
		2289	callback. Future versions of libev might stop the watcher automatically
		2290	when a child exit is detected (calling C<ev_child_stop> twice is not a
		2291	problem).
		2292
		2293	=head3 Watcher-Specific Functions and Data Members
636		2294
637	=over 4	2295	=over 4
638		2296
639	=item ev_child_init (ev_child *, callback, int pid)	2297	=item ev_child_init (ev_child *, callback, int pid, int trace)
640		2298
641	=item ev_child_set (ev_child *, int pid)	2299	=item ev_child_set (ev_child *, int pid, int trace)
642		2300
643	Configures the watcher to wait for status changes of process C<pid> (or	2301	Configures the watcher to wait for status changes of process C<pid> (or
644	I<any> process if C<pid> is specified as C<0>). The callback can look	2302	I<any> process if C<pid> is specified as C<0>). The callback can look
645	at the C<rstatus> member of the C<ev_child> watcher structure to see	2303	at the C<rstatus> member of the C<ev_child> watcher structure to see
646	the status word (use the macros from C<sys/wait.h> and see your systems	2304	the status word (use the macros from C<sys/wait.h> and see your systems
647	C<waitpid> documentation). The C<rpid> member contains the pid of the	2305	C<waitpid> documentation). The C<rpid> member contains the pid of the
648	process causing the status change.	2306	process causing the status change. C<trace> must be either C<0> (only
		2307	activate the watcher when the process terminates) or C<1> (additionally
		2308	activate the watcher when the process is stopped or continued).
		2309
		2310	=item int pid [read-only]
		2311
		2312	The process id this watcher watches out for, or C<0>, meaning any process id.
		2313
		2314	=item int rpid [read-write]
		2315
		2316	The process id that detected a status change.
		2317
		2318	=item int rstatus [read-write]
		2319
		2320	The process exit/trace status caused by C<rpid> (see your systems
		2321	C<waitpid> and C<sys/wait.h> documentation for details).
649		2322
650	=back	2323	=back
651		2324
		2325	=head3 Examples
		2326
		2327	Example: C<fork()> a new process and install a child handler to wait for
		2328	its completion.
		2329
		2330	ev_child cw;
		2331
		2332	static void
		2333	child_cb (EV_P_ ev_child *w, int revents)
		2334	{
		2335	ev_child_stop (EV_A_ w);
		2336	printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
		2337	}
		2338
		2339	pid_t pid = fork ();
		2340
		2341	if (pid < 0)
		2342	// error
		2343	else if (pid == 0)
		2344	{
		2345	// the forked child executes here
		2346	exit (1);
		2347	}
		2348	else
		2349	{
		2350	ev_child_init (&cw, child_cb, pid, 0);
		2351	ev_child_start (EV_DEFAULT_ &cw);
		2352	}
		2353
		2354
		2355	=head2 C<ev_stat> - did the file attributes just change?
		2356
		2357	This watches a file system path for attribute changes. That is, it calls
		2358	C<stat> on that path in regular intervals (or when the OS says it changed)
		2359	and sees if it changed compared to the last time, invoking the callback if
		2360	it did.
		2361
		2362	The path does not need to exist: changing from "path exists" to "path does
		2363	not exist" is a status change like any other. The condition "path does not
		2364	exist" (or more correctly "path cannot be stat'ed") is signified by the
		2365	C<st_nlink> field being zero (which is otherwise always forced to be at
		2366	least one) and all the other fields of the stat buffer having unspecified
		2367	contents.
		2368
		2369	The path I<must not> end in a slash or contain special components such as
		2370	C<.> or C<..>. The path I<should> be absolute: If it is relative and
		2371	your working directory changes, then the behaviour is undefined.
		2372
		2373	Since there is no portable change notification interface available, the
		2374	portable implementation simply calls C<stat(2)> regularly on the path
		2375	to see if it changed somehow. You can specify a recommended polling
		2376	interval for this case. If you specify a polling interval of C<0> (highly
		2377	recommended!) then a I<suitable, unspecified default> value will be used
		2378	(which you can expect to be around five seconds, although this might
		2379	change dynamically). Libev will also impose a minimum interval which is
		2380	currently around C<0.1>, but that's usually overkill.
		2381
		2382	This watcher type is not meant for massive numbers of stat watchers,
		2383	as even with OS-supported change notifications, this can be
		2384	resource-intensive.
		2385
		2386	At the time of this writing, the only OS-specific interface implemented
		2387	is the Linux inotify interface (implementing kqueue support is left as an
		2388	exercise for the reader. Note, however, that the author sees no way of
		2389	implementing C<ev_stat> semantics with kqueue, except as a hint).
		2390
		2391	=head3 ABI Issues (Largefile Support)
		2392
		2393	Libev by default (unless the user overrides this) uses the default
		2394	compilation environment, which means that on systems with large file
		2395	support disabled by default, you get the 32 bit version of the stat
		2396	structure. When using the library from programs that change the ABI to
		2397	use 64 bit file offsets the programs will fail. In that case you have to
		2398	compile libev with the same flags to get binary compatibility. This is
		2399	obviously the case with any flags that change the ABI, but the problem is
		2400	most noticeably displayed with ev_stat and large file support.
		2401
		2402	The solution for this is to lobby your distribution maker to make large
		2403	file interfaces available by default (as e.g. FreeBSD does) and not
		2404	optional. Libev cannot simply switch on large file support because it has
		2405	to exchange stat structures with application programs compiled using the
		2406	default compilation environment.
		2407
		2408	=head3 Inotify and Kqueue
		2409
		2410	When C<inotify (7)> support has been compiled into libev and present at
		2411	runtime, it will be used to speed up change detection where possible. The
		2412	inotify descriptor will be created lazily when the first C<ev_stat>
		2413	watcher is being started.
		2414
		2415	Inotify presence does not change the semantics of C<ev_stat> watchers
		2416	except that changes might be detected earlier, and in some cases, to avoid
		2417	making regular C<stat> calls. Even in the presence of inotify support
		2418	there are many cases where libev has to resort to regular C<stat> polling,
		2419	but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
		2420	many bugs), the path exists (i.e. stat succeeds), and the path resides on
		2421	a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
		2422	xfs are fully working) libev usually gets away without polling.
		2423
		2424	There is no support for kqueue, as apparently it cannot be used to
		2425	implement this functionality, due to the requirement of having a file
		2426	descriptor open on the object at all times, and detecting renames, unlinks
		2427	etc. is difficult.
		2428
		2429	=head3 C<stat ()> is a synchronous operation
		2430
		2431	Libev doesn't normally do any kind of I/O itself, and so is not blocking
		2432	the process. The exception are C<ev_stat> watchers - those call C<stat
		2433	()>, which is a synchronous operation.
		2434
		2435	For local paths, this usually doesn't matter: unless the system is very
		2436	busy or the intervals between stat's are large, a stat call will be fast,
		2437	as the path data is usually in memory already (except when starting the
		2438	watcher).
		2439
		2440	For networked file systems, calling C<stat ()> can block an indefinite
		2441	time due to network issues, and even under good conditions, a stat call
		2442	often takes multiple milliseconds.
		2443
		2444	Therefore, it is best to avoid using C<ev_stat> watchers on networked
		2445	paths, although this is fully supported by libev.
		2446
		2447	=head3 The special problem of stat time resolution
		2448
		2449	The C<stat ()> system call only supports full-second resolution portably,
		2450	and even on systems where the resolution is higher, most file systems
		2451	still only support whole seconds.
		2452
		2453	That means that, if the time is the only thing that changes, you can
		2454	easily miss updates: on the first update, C<ev_stat> detects a change and
		2455	calls your callback, which does something. When there is another update
		2456	within the same second, C<ev_stat> will be unable to detect unless the
		2457	stat data does change in other ways (e.g. file size).
		2458
		2459	The solution to this is to delay acting on a change for slightly more
		2460	than a second (or till slightly after the next full second boundary), using
		2461	a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
		2462	ev_timer_again (loop, w)>).
		2463
		2464	The C<.02> offset is added to work around small timing inconsistencies
		2465	of some operating systems (where the second counter of the current time
		2466	might be be delayed. One such system is the Linux kernel, where a call to
		2467	C<gettimeofday> might return a timestamp with a full second later than
		2468	a subsequent C<time> call - if the equivalent of C<time ()> is used to
		2469	update file times then there will be a small window where the kernel uses
		2470	the previous second to update file times but libev might already execute
		2471	the timer callback).
		2472
		2473	=head3 Watcher-Specific Functions and Data Members
		2474
		2475	=over 4
		2476
		2477	=item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
		2478
		2479	=item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
		2480
		2481	Configures the watcher to wait for status changes of the given
		2482	C<path>. The C<interval> is a hint on how quickly a change is expected to
		2483	be detected and should normally be specified as C<0> to let libev choose
		2484	a suitable value. The memory pointed to by C<path> must point to the same
		2485	path for as long as the watcher is active.
		2486
		2487	The callback will receive an C<EV_STAT> event when a change was detected,
		2488	relative to the attributes at the time the watcher was started (or the
		2489	last change was detected).
		2490
		2491	=item ev_stat_stat (loop, ev_stat *)
		2492
		2493	Updates the stat buffer immediately with new values. If you change the
		2494	watched path in your callback, you could call this function to avoid
		2495	detecting this change (while introducing a race condition if you are not
		2496	the only one changing the path). Can also be useful simply to find out the
		2497	new values.
		2498
		2499	=item ev_statdata attr [read-only]
		2500
		2501	The most-recently detected attributes of the file. Although the type is
		2502	C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
		2503	suitable for your system, but you can only rely on the POSIX-standardised
		2504	members to be present. If the C<st_nlink> member is C<0>, then there was
		2505	some error while C<stat>ing the file.
		2506
		2507	=item ev_statdata prev [read-only]
		2508
		2509	The previous attributes of the file. The callback gets invoked whenever
		2510	C<prev> != C<attr>, or, more precisely, one or more of these members
		2511	differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
		2512	C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
		2513
		2514	=item ev_tstamp interval [read-only]
		2515
		2516	The specified interval.
		2517
		2518	=item const char *path [read-only]
		2519
		2520	The file system path that is being watched.
		2521
		2522	=back
		2523
		2524	=head3 Examples
		2525
		2526	Example: Watch C</etc/passwd> for attribute changes.
		2527
		2528	static void
		2529	passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
		2530	{
		2531	/* /etc/passwd changed in some way */
		2532	if (w->attr.st_nlink)
		2533	{
		2534	printf ("passwd current size %ld\n", (long)w->attr.st_size);
		2535	printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
		2536	printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
		2537	}
		2538	else
		2539	/* you shalt not abuse printf for puts */
		2540	puts ("wow, /etc/passwd is not there, expect problems. "
		2541	"if this is windows, they already arrived\n");
		2542	}
		2543
		2544	...
		2545	ev_stat passwd;
		2546
		2547	ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
		2548	ev_stat_start (loop, &passwd);
		2549
		2550	Example: Like above, but additionally use a one-second delay so we do not
		2551	miss updates (however, frequent updates will delay processing, too, so
		2552	one might do the work both on C<ev_stat> callback invocation I<and> on
		2553	C<ev_timer> callback invocation).
		2554
		2555	static ev_stat passwd;
		2556	static ev_timer timer;
		2557
		2558	static void
		2559	timer_cb (EV_P_ ev_timer *w, int revents)
		2560	{
		2561	ev_timer_stop (EV_A_ w);
		2562
		2563	/* now it's one second after the most recent passwd change */
		2564	}
		2565
		2566	static void
		2567	stat_cb (EV_P_ ev_stat *w, int revents)
		2568	{
		2569	/* reset the one-second timer */
		2570	ev_timer_again (EV_A_ &timer);
		2571	}
		2572
		2573	...
		2574	ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
		2575	ev_stat_start (loop, &passwd);
		2576	ev_timer_init (&timer, timer_cb, 0., 1.02);
		2577
		2578
652	=head2 C<ev_idle> - when you've got nothing better to do	2579	=head2 C<ev_idle> - when you've got nothing better to do...
653		2580
654	Idle watchers trigger events when there are no other events are pending	2581	Idle watchers trigger events when no other events of the same or higher
655	(prepare, check and other idle watchers do not count). That is, as long	2582	priority are pending (prepare, check and other idle watchers do not count
656	as your process is busy handling sockets or timeouts (or even signals,	2583	as receiving "events").
657	imagine) it will not be triggered. But when your process is idle all idle	2584
658	watchers are being called again and again, once per event loop iteration -	2585	That is, as long as your process is busy handling sockets or timeouts
		2586	(or even signals, imagine) of the same or higher priority it will not be
		2587	triggered. But when your process is idle (or only lower-priority watchers
		2588	are pending), the idle watchers are being called once per event loop
659	until stopped, that is, or your process receives more events and becomes	2589	iteration - until stopped, that is, or your process receives more events
660	busy.	2590	and becomes busy again with higher priority stuff.
661		2591
662	The most noteworthy effect is that as long as any idle watchers are	2592	The most noteworthy effect is that as long as any idle watchers are
663	active, the process will not block when waiting for new events.	2593	active, the process will not block when waiting for new events.
664		2594
665	Apart from keeping your process non-blocking (which is a useful	2595	Apart from keeping your process non-blocking (which is a useful
666	effect on its own sometimes), idle watchers are a good place to do	2596	effect on its own sometimes), idle watchers are a good place to do
667	"pseudo-background processing", or delay processing stuff to after the	2597	"pseudo-background processing", or delay processing stuff to after the
668	event loop has handled all outstanding events.	2598	event loop has handled all outstanding events.
669		2599
		2600	=head3 Watcher-Specific Functions and Data Members
		2601
670	=over 4	2602	=over 4
671		2603
672	=item ev_idle_init (ev_signal *, callback)	2604	=item ev_idle_init (ev_idle *, callback)
673		2605
674	Initialises and configures the idle watcher - it has no parameters of any	2606	Initialises and configures the idle watcher - it has no parameters of any
675	kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,	2607	kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
676	believe me.	2608	believe me.
677		2609
678	=back	2610	=back
679		2611
		2612	=head3 Examples
		2613
		2614	Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
		2615	callback, free it. Also, use no error checking, as usual.
		2616
		2617	static void
		2618	idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
		2619	{
		2620	free (w);
		2621	// now do something you wanted to do when the program has
		2622	// no longer anything immediate to do.
		2623	}
		2624
		2625	ev_idle *idle_watcher = malloc (sizeof (ev_idle));
		2626	ev_idle_init (idle_watcher, idle_cb);
		2627	ev_idle_start (loop, idle_watcher);
		2628
		2629
680	=head2 C<ev_prepare> and C<ev_check> - customise your event loop	2630	=head2 C<ev_prepare> and C<ev_check> - customise your event loop!
681		2631
682	Prepare and check watchers are usually (but not always) used in tandem:	2632	Prepare and check watchers are usually (but not always) used in pairs:
683	prepare watchers get invoked before the process blocks and check watchers	2633	prepare watchers get invoked before the process blocks and check watchers
684	afterwards.	2634	afterwards.
685		2635
		2636	You I<must not> call C<ev_loop> or similar functions that enter
		2637	the current event loop from either C<ev_prepare> or C<ev_check>
		2638	watchers. Other loops than the current one are fine, however. The
		2639	rationale behind this is that you do not need to check for recursion in
		2640	those watchers, i.e. the sequence will always be C<ev_prepare>, blocking,
		2641	C<ev_check> so if you have one watcher of each kind they will always be
		2642	called in pairs bracketing the blocking call.
		2643
686	Their main purpose is to integrate other event mechanisms into libev. This	2644	Their main purpose is to integrate other event mechanisms into libev and
687	could be used, for example, to track variable changes, implement your own	2645	their use is somewhat advanced. They could be used, for example, to track
688	watchers, integrate net-snmp or a coroutine library and lots more.	2646	variable changes, implement your own watchers, integrate net-snmp or a
		2647	coroutine library and lots more. They are also occasionally useful if
		2648	you cache some data and want to flush it before blocking (for example,
		2649	in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
		2650	watcher).
689		2651
690	This is done by examining in each prepare call which file descriptors need	2652	This is done by examining in each prepare call which file descriptors
691	to be watched by the other library, registering C<ev_io> watchers for	2653	need to be watched by the other library, registering C<ev_io> watchers
692	them and starting an C<ev_timer> watcher for any timeouts (many libraries	2654	for them and starting an C<ev_timer> watcher for any timeouts (many
693	provide just this functionality). Then, in the check watcher you check for	2655	libraries provide exactly this functionality). Then, in the check watcher,
694	any events that occured (by checking the pending status of all watchers	2656	you check for any events that occurred (by checking the pending status
695	and stopping them) and call back into the library. The I/O and timer	2657	of all watchers and stopping them) and call back into the library. The
696	callbacks will never actually be called (but must be valid nevertheless,	2658	I/O and timer callbacks will never actually be called (but must be valid
697	because you never know, you know?).	2659	nevertheless, because you never know, you know?).
698		2660
699	As another example, the Perl Coro module uses these hooks to integrate	2661	As another example, the Perl Coro module uses these hooks to integrate
700	coroutines into libev programs, by yielding to other active coroutines	2662	coroutines into libev programs, by yielding to other active coroutines
701	during each prepare and only letting the process block if no coroutines	2663	during each prepare and only letting the process block if no coroutines
702	are ready to run (it's actually more complicated: it only runs coroutines	2664	are ready to run (it's actually more complicated: it only runs coroutines
703	with priority higher than or equal to the event loop and one coroutine	2665	with priority higher than or equal to the event loop and one coroutine
704	of lower priority, but only once, using idle watchers to keep the event	2666	of lower priority, but only once, using idle watchers to keep the event
705	loop from blocking if lower-priority coroutines are active, thus mapping	2667	loop from blocking if lower-priority coroutines are active, thus mapping
706	low-priority coroutines to idle/background tasks).	2668	low-priority coroutines to idle/background tasks).
707		2669
		2670	It is recommended to give C<ev_check> watchers highest (C<EV_MAXPRI>)
		2671	priority, to ensure that they are being run before any other watchers
		2672	after the poll (this doesn't matter for C<ev_prepare> watchers).
		2673
		2674	Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
		2675	activate ("feed") events into libev. While libev fully supports this, they
		2676	might get executed before other C<ev_check> watchers did their job. As
		2677	C<ev_check> watchers are often used to embed other (non-libev) event
		2678	loops those other event loops might be in an unusable state until their
		2679	C<ev_check> watcher ran (always remind yourself to coexist peacefully with
		2680	others).
		2681
		2682	=head3 Watcher-Specific Functions and Data Members
		2683
708	=over 4	2684	=over 4
709		2685
710	=item ev_prepare_init (ev_prepare *, callback)	2686	=item ev_prepare_init (ev_prepare *, callback)
711		2687
712	=item ev_check_init (ev_check *, callback)	2688	=item ev_check_init (ev_check *, callback)
713		2689
714	Initialises and configures the prepare or check watcher - they have no	2690	Initialises and configures the prepare or check watcher - they have no
715	parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>	2691	parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
716	macros, but using them is utterly, utterly and completely pointless.	2692	macros, but using them is utterly, utterly, utterly and completely
		2693	pointless.
717		2694
718	=back	2695	=back
719		2696
		2697	=head3 Examples
		2698
		2699	There are a number of principal ways to embed other event loops or modules
		2700	into libev. Here are some ideas on how to include libadns into libev
		2701	(there is a Perl module named C<EV::ADNS> that does this, which you could
		2702	use as a working example. Another Perl module named C<EV::Glib> embeds a
		2703	Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
		2704	Glib event loop).
		2705
		2706	Method 1: Add IO watchers and a timeout watcher in a prepare handler,
		2707	and in a check watcher, destroy them and call into libadns. What follows
		2708	is pseudo-code only of course. This requires you to either use a low
		2709	priority for the check watcher or use C<ev_clear_pending> explicitly, as
		2710	the callbacks for the IO/timeout watchers might not have been called yet.
		2711
		2712	static ev_io iow [nfd];
		2713	static ev_timer tw;
		2714
		2715	static void
		2716	io_cb (struct ev_loop *loop, ev_io *w, int revents)
		2717	{
		2718	}
		2719
		2720	// create io watchers for each fd and a timer before blocking
		2721	static void
		2722	adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
		2723	{
		2724	int timeout = 3600000;
		2725	struct pollfd fds [nfd];
		2726	// actual code will need to loop here and realloc etc.
		2727	adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
		2728
		2729	/* the callback is illegal, but won't be called as we stop during check */
		2730	ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
		2731	ev_timer_start (loop, &tw);
		2732
		2733	// create one ev_io per pollfd
		2734	for (int i = 0; i < nfd; ++i)
		2735	{
		2736	ev_io_init (iow + i, io_cb, fds [i].fd,
		2737	((fds [i].events & POLLIN ? EV_READ : 0)
		2738	| (fds [i].events & POLLOUT ? EV_WRITE : 0)));
		2739
		2740	fds [i].revents = 0;
		2741	ev_io_start (loop, iow + i);
		2742	}
		2743	}
		2744
		2745	// stop all watchers after blocking
		2746	static void
		2747	adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
		2748	{
		2749	ev_timer_stop (loop, &tw);
		2750
		2751	for (int i = 0; i < nfd; ++i)
		2752	{
		2753	// set the relevant poll flags
		2754	// could also call adns_processreadable etc. here
		2755	struct pollfd *fd = fds + i;
		2756	int revents = ev_clear_pending (iow + i);
		2757	if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
		2758	if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
		2759
		2760	// now stop the watcher
		2761	ev_io_stop (loop, iow + i);
		2762	}
		2763
		2764	adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
		2765	}
		2766
		2767	Method 2: This would be just like method 1, but you run C<adns_afterpoll>
		2768	in the prepare watcher and would dispose of the check watcher.
		2769
		2770	Method 3: If the module to be embedded supports explicit event
		2771	notification (libadns does), you can also make use of the actual watcher
		2772	callbacks, and only destroy/create the watchers in the prepare watcher.
		2773
		2774	static void
		2775	timer_cb (EV_P_ ev_timer *w, int revents)
		2776	{
		2777	adns_state ads = (adns_state)w->data;
		2778	update_now (EV_A);
		2779
		2780	adns_processtimeouts (ads, &tv_now);
		2781	}
		2782
		2783	static void
		2784	io_cb (EV_P_ ev_io *w, int revents)
		2785	{
		2786	adns_state ads = (adns_state)w->data;
		2787	update_now (EV_A);
		2788
		2789	if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
		2790	if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
		2791	}
		2792
		2793	// do not ever call adns_afterpoll
		2794
		2795	Method 4: Do not use a prepare or check watcher because the module you
		2796	want to embed is not flexible enough to support it. Instead, you can
		2797	override their poll function. The drawback with this solution is that the
		2798	main loop is now no longer controllable by EV. The C<Glib::EV> module uses
		2799	this approach, effectively embedding EV as a client into the horrible
		2800	libglib event loop.
		2801
		2802	static gint
		2803	event_poll_func (GPollFD *fds, guint nfds, gint timeout)
		2804	{
		2805	int got_events = 0;
		2806
		2807	for (n = 0; n < nfds; ++n)
		2808	// create/start io watcher that sets the relevant bits in fds[n] and increment got_events
		2809
		2810	if (timeout >= 0)
		2811	// create/start timer
		2812
		2813	// poll
		2814	ev_loop (EV_A_ 0);
		2815
		2816	// stop timer again
		2817	if (timeout >= 0)
		2818	ev_timer_stop (EV_A_ &to);
		2819
		2820	// stop io watchers again - their callbacks should have set
		2821	for (n = 0; n < nfds; ++n)
		2822	ev_io_stop (EV_A_ iow [n]);
		2823
		2824	return got_events;
		2825	}
		2826
		2827
		2828	=head2 C<ev_embed> - when one backend isn't enough...
		2829
		2830	This is a rather advanced watcher type that lets you embed one event loop
		2831	into another (currently only C<ev_io> events are supported in the embedded
		2832	loop, other types of watchers might be handled in a delayed or incorrect
		2833	fashion and must not be used).
		2834
		2835	There are primarily two reasons you would want that: work around bugs and
		2836	prioritise I/O.
		2837
		2838	As an example for a bug workaround, the kqueue backend might only support
		2839	sockets on some platform, so it is unusable as generic backend, but you
		2840	still want to make use of it because you have many sockets and it scales
		2841	so nicely. In this case, you would create a kqueue-based loop and embed
		2842	it into your default loop (which might use e.g. poll). Overall operation
		2843	will be a bit slower because first libev has to call C<poll> and then
		2844	C<kevent>, but at least you can use both mechanisms for what they are
		2845	best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
		2846
		2847	As for prioritising I/O: under rare circumstances you have the case where
		2848	some fds have to be watched and handled very quickly (with low latency),
		2849	and even priorities and idle watchers might have too much overhead. In
		2850	this case you would put all the high priority stuff in one loop and all
		2851	the rest in a second one, and embed the second one in the first.
		2852
		2853	As long as the watcher is active, the callback will be invoked every
		2854	time there might be events pending in the embedded loop. The callback
		2855	must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
		2856	sweep and invoke their callbacks (the callback doesn't need to invoke the
		2857	C<ev_embed_sweep> function directly, it could also start an idle watcher
		2858	to give the embedded loop strictly lower priority for example).
		2859
		2860	You can also set the callback to C<0>, in which case the embed watcher
		2861	will automatically execute the embedded loop sweep whenever necessary.
		2862
		2863	Fork detection will be handled transparently while the C<ev_embed> watcher
		2864	is active, i.e., the embedded loop will automatically be forked when the
		2865	embedding loop forks. In other cases, the user is responsible for calling
		2866	C<ev_loop_fork> on the embedded loop.
		2867
		2868	Unfortunately, not all backends are embeddable: only the ones returned by
		2869	C<ev_embeddable_backends> are, which, unfortunately, does not include any
		2870	portable one.
		2871
		2872	So when you want to use this feature you will always have to be prepared
		2873	that you cannot get an embeddable loop. The recommended way to get around
		2874	this is to have a separate variables for your embeddable loop, try to
		2875	create it, and if that fails, use the normal loop for everything.
		2876
		2877	=head3 C<ev_embed> and fork
		2878
		2879	While the C<ev_embed> watcher is running, forks in the embedding loop will
		2880	automatically be applied to the embedded loop as well, so no special
		2881	fork handling is required in that case. When the watcher is not running,
		2882	however, it is still the task of the libev user to call C<ev_loop_fork ()>
		2883	as applicable.
		2884
		2885	=head3 Watcher-Specific Functions and Data Members
		2886
		2887	=over 4
		2888
		2889	=item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
		2890
		2891	=item ev_embed_set (ev_embed *, callback, struct ev_loop *embedded_loop)
		2892
		2893	Configures the watcher to embed the given loop, which must be
		2894	embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
		2895	invoked automatically, otherwise it is the responsibility of the callback
		2896	to invoke it (it will continue to be called until the sweep has been done,
		2897	if you do not want that, you need to temporarily stop the embed watcher).
		2898
		2899	=item ev_embed_sweep (loop, ev_embed *)
		2900
		2901	Make a single, non-blocking sweep over the embedded loop. This works
		2902	similarly to C<ev_loop (embedded_loop, EVLOOP_NONBLOCK)>, but in the most
		2903	appropriate way for embedded loops.
		2904
		2905	=item struct ev_loop *other [read-only]
		2906
		2907	The embedded event loop.
		2908
		2909	=back
		2910
		2911	=head3 Examples
		2912
		2913	Example: Try to get an embeddable event loop and embed it into the default
		2914	event loop. If that is not possible, use the default loop. The default
		2915	loop is stored in C<loop_hi>, while the embeddable loop is stored in
		2916	C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
		2917	used).
		2918
		2919	struct ev_loop *loop_hi = ev_default_init (0);
		2920	struct ev_loop *loop_lo = 0;
		2921	ev_embed embed;
		2922
		2923	// see if there is a chance of getting one that works
		2924	// (remember that a flags value of 0 means autodetection)
		2925	loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
		2926	? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
		2927	: 0;
		2928
		2929	// if we got one, then embed it, otherwise default to loop_hi
		2930	if (loop_lo)
		2931	{
		2932	ev_embed_init (&embed, 0, loop_lo);
		2933	ev_embed_start (loop_hi, &embed);
		2934	}
		2935	else
		2936	loop_lo = loop_hi;
		2937
		2938	Example: Check if kqueue is available but not recommended and create
		2939	a kqueue backend for use with sockets (which usually work with any
		2940	kqueue implementation). Store the kqueue/socket-only event loop in
		2941	C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
		2942
		2943	struct ev_loop *loop = ev_default_init (0);
		2944	struct ev_loop *loop_socket = 0;
		2945	ev_embed embed;
		2946
		2947	if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
		2948	if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
		2949	{
		2950	ev_embed_init (&embed, 0, loop_socket);
		2951	ev_embed_start (loop, &embed);
		2952	}
		2953
		2954	if (!loop_socket)
		2955	loop_socket = loop;
		2956
		2957	// now use loop_socket for all sockets, and loop for everything else
		2958
		2959
		2960	=head2 C<ev_fork> - the audacity to resume the event loop after a fork
		2961
		2962	Fork watchers are called when a C<fork ()> was detected (usually because
		2963	whoever is a good citizen cared to tell libev about it by calling
		2964	C<ev_default_fork> or C<ev_loop_fork>). The invocation is done before the
		2965	event loop blocks next and before C<ev_check> watchers are being called,
		2966	and only in the child after the fork. If whoever good citizen calling
		2967	C<ev_default_fork> cheats and calls it in the wrong process, the fork
		2968	handlers will be invoked, too, of course.
		2969
		2970	=head3 The special problem of life after fork - how is it possible?
		2971
		2972	Most uses of C<fork()> consist of forking, then some simple calls to set
		2973	up/change the process environment, followed by a call to C<exec()>. This
		2974	sequence should be handled by libev without any problems.
		2975
		2976	This changes when the application actually wants to do event handling
		2977	in the child, or both parent in child, in effect "continuing" after the
		2978	fork.
		2979
		2980	The default mode of operation (for libev, with application help to detect
		2981	forks) is to duplicate all the state in the child, as would be expected
		2982	when I<either> the parent I<or> the child process continues.
		2983
		2984	When both processes want to continue using libev, then this is usually the
		2985	wrong result. In that case, usually one process (typically the parent) is
		2986	supposed to continue with all watchers in place as before, while the other
		2987	process typically wants to start fresh, i.e. without any active watchers.
		2988
		2989	The cleanest and most efficient way to achieve that with libev is to
		2990	simply create a new event loop, which of course will be "empty", and
		2991	use that for new watchers. This has the advantage of not touching more
		2992	memory than necessary, and thus avoiding the copy-on-write, and the
		2993	disadvantage of having to use multiple event loops (which do not support
		2994	signal watchers).
		2995
		2996	When this is not possible, or you want to use the default loop for
		2997	other reasons, then in the process that wants to start "fresh", call
		2998	C<ev_default_destroy ()> followed by C<ev_default_loop (...)>. Destroying
		2999	the default loop will "orphan" (not stop) all registered watchers, so you
		3000	have to be careful not to execute code that modifies those watchers. Note
		3001	also that in that case, you have to re-register any signal watchers.
		3002
		3003	=head3 Watcher-Specific Functions and Data Members
		3004
		3005	=over 4
		3006
		3007	=item ev_fork_init (ev_signal *, callback)
		3008
		3009	Initialises and configures the fork watcher - it has no parameters of any
		3010	kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
		3011	believe me.
		3012
		3013	=back
		3014
		3015
		3016	=head2 C<ev_async> - how to wake up an event loop
		3017
		3018	In general, you cannot use an C<ev_loop> from multiple threads or other
		3019	asynchronous sources such as signal handlers (as opposed to multiple event
		3020	loops - those are of course safe to use in different threads).
		3021
		3022	Sometimes, however, you need to wake up an event loop you do not control,
		3023	for example because it belongs to another thread. This is what C<ev_async>
		3024	watchers do: as long as the C<ev_async> watcher is active, you can signal
		3025	it by calling C<ev_async_send>, which is thread- and signal safe.
		3026
		3027	This functionality is very similar to C<ev_signal> watchers, as signals,
		3028	too, are asynchronous in nature, and signals, too, will be compressed
		3029	(i.e. the number of callback invocations may be less than the number of
		3030	C<ev_async_sent> calls).
		3031
		3032	Unlike C<ev_signal> watchers, C<ev_async> works with any event loop, not
		3033	just the default loop.
		3034
		3035	=head3 Queueing
		3036
		3037	C<ev_async> does not support queueing of data in any way. The reason
		3038	is that the author does not know of a simple (or any) algorithm for a
		3039	multiple-writer-single-reader queue that works in all cases and doesn't
		3040	need elaborate support such as pthreads or unportable memory access
		3041	semantics.
		3042
		3043	That means that if you want to queue data, you have to provide your own
		3044	queue. But at least I can tell you how to implement locking around your
		3045	queue:
		3046
		3047	=over 4
		3048
		3049	=item queueing from a signal handler context
		3050
		3051	To implement race-free queueing, you simply add to the queue in the signal
		3052	handler but you block the signal handler in the watcher callback. Here is
		3053	an example that does that for some fictitious SIGUSR1 handler:
		3054
		3055	static ev_async mysig;
		3056
		3057	static void
		3058	sigusr1_handler (void)
		3059	{
		3060	sometype data;
		3061
		3062	// no locking etc.
		3063	queue_put (data);
		3064	ev_async_send (EV_DEFAULT_ &mysig);
		3065	}
		3066
		3067	static void
		3068	mysig_cb (EV_P_ ev_async *w, int revents)
		3069	{
		3070	sometype data;
		3071	sigset_t block, prev;
		3072
		3073	sigemptyset (&block);
		3074	sigaddset (&block, SIGUSR1);
		3075	sigprocmask (SIG_BLOCK, &block, &prev);
		3076
		3077	while (queue_get (&data))
		3078	process (data);
		3079
		3080	if (sigismember (&prev, SIGUSR1)
		3081	sigprocmask (SIG_UNBLOCK, &block, 0);
		3082	}
		3083
		3084	(Note: pthreads in theory requires you to use C<pthread_setmask>
		3085	instead of C<sigprocmask> when you use threads, but libev doesn't do it
		3086	either...).
		3087
		3088	=item queueing from a thread context
		3089
		3090	The strategy for threads is different, as you cannot (easily) block
		3091	threads but you can easily preempt them, so to queue safely you need to
		3092	employ a traditional mutex lock, such as in this pthread example:
		3093
		3094	static ev_async mysig;
		3095	static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
		3096
		3097	static void
		3098	otherthread (void)
		3099	{
		3100	// only need to lock the actual queueing operation
		3101	pthread_mutex_lock (&mymutex);
		3102	queue_put (data);
		3103	pthread_mutex_unlock (&mymutex);
		3104
		3105	ev_async_send (EV_DEFAULT_ &mysig);
		3106	}
		3107
		3108	static void
		3109	mysig_cb (EV_P_ ev_async *w, int revents)
		3110	{
		3111	pthread_mutex_lock (&mymutex);
		3112
		3113	while (queue_get (&data))
		3114	process (data);
		3115
		3116	pthread_mutex_unlock (&mymutex);
		3117	}
		3118
		3119	=back
		3120
		3121
		3122	=head3 Watcher-Specific Functions and Data Members
		3123
		3124	=over 4
		3125
		3126	=item ev_async_init (ev_async *, callback)
		3127
		3128	Initialises and configures the async watcher - it has no parameters of any
		3129	kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
		3130	trust me.
		3131
		3132	=item ev_async_send (loop, ev_async *)
		3133
		3134	Sends/signals/activates the given C<ev_async> watcher, that is, feeds
		3135	an C<EV_ASYNC> event on the watcher into the event loop. Unlike
		3136	C<ev_feed_event>, this call is safe to do from other threads, signal or
		3137	similar contexts (see the discussion of C<EV_ATOMIC_T> in the embedding
		3138	section below on what exactly this means).
		3139
		3140	Note that, as with other watchers in libev, multiple events might get
		3141	compressed into a single callback invocation (another way to look at this
		3142	is that C<ev_async> watchers are level-triggered, set on C<ev_async_send>,
		3143	reset when the event loop detects that).
		3144
		3145	This call incurs the overhead of a system call only once per event loop
		3146	iteration, so while the overhead might be noticeable, it doesn't apply to
		3147	repeated calls to C<ev_async_send> for the same event loop.
		3148
		3149	=item bool = ev_async_pending (ev_async *)
		3150
		3151	Returns a non-zero value when C<ev_async_send> has been called on the
		3152	watcher but the event has not yet been processed (or even noted) by the
		3153	event loop.
		3154
		3155	C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
		3156	the loop iterates next and checks for the watcher to have become active,
		3157	it will reset the flag again. C<ev_async_pending> can be used to very
		3158	quickly check whether invoking the loop might be a good idea.
		3159
		3160	Not that this does I<not> check whether the watcher itself is pending,
		3161	only whether it has been requested to make this watcher pending: there
		3162	is a time window between the event loop checking and resetting the async
		3163	notification, and the callback being invoked.
		3164
		3165	=back
		3166
		3167
720	=head1 OTHER FUNCTIONS	3168	=head1 OTHER FUNCTIONS
721		3169
722	There are some other functions of possible interest. Described. Here. Now.	3170	There are some other functions of possible interest. Described. Here. Now.
723		3171
724	=over 4	3172	=over 4
725		3173
726	=item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)	3174	=item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)
727		3175
728	This function combines a simple timer and an I/O watcher, calls your	3176	This function combines a simple timer and an I/O watcher, calls your
729	callback on whichever event happens first and automatically stop both	3177	callback on whichever event happens first and automatically stops both
730	watchers. This is useful if you want to wait for a single event on an fd	3178	watchers. This is useful if you want to wait for a single event on an fd
731	or timeout without having to allocate/configure/start/stop/free one or	3179	or timeout without having to allocate/configure/start/stop/free one or
732	more watchers yourself.	3180	more watchers yourself.
733		3181
734	If C<fd> is less than 0, then no I/O watcher will be started and events	3182	If C<fd> is less than 0, then no I/O watcher will be started and the
735	is being ignored. Otherwise, an C<ev_io> watcher for the given C<fd> and	3183	C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
736	C<events> set will be craeted and started.	3184	the given C<fd> and C<events> set will be created and started.
737		3185
738	If C<timeout> is less than 0, then no timeout watcher will be	3186	If C<timeout> is less than 0, then no timeout watcher will be
739	started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and	3187	started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
740	repeat = 0) will be started. While C<0> is a valid timeout, it is of	3188	repeat = 0) will be started. C<0> is a valid timeout.
741	dubious value.
742		3189
743	The callback has the type C<void (*cb)(int revents, void *arg)> and gets	3190	The callback has the type C<void (*cb)(int revents, void *arg)> and is
744	passed an C<revents> set like normal event callbacks (a combination of	3191	passed an C<revents> set like normal event callbacks (a combination of
745	C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMEOUT>) and the C<arg>	3192	C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMER>) and the C<arg>
746	value passed to C<ev_once>:	3193	value passed to C<ev_once>. Note that it is possible to receive I<both>
		3194	a timeout and an io event at the same time - you probably should give io
		3195	events precedence.
747		3196
		3197	Example: wait up to ten seconds for data to appear on STDIN_FILENO.
		3198
748	static void stdin_ready (int revents, void *arg)	3199	static void stdin_ready (int revents, void *arg)
749	{	3200	{
750	if (revents & EV_TIMEOUT)
751	/* doh, nothing entered */;
752	else if (revents & EV_READ)	3201	if (revents & EV_READ)
753	/* stdin might have data for us, joy! */;	3202	/* stdin might have data for us, joy! */;
		3203	else if (revents & EV_TIMER)
		3204	/* doh, nothing entered */;
754	}	3205	}
755		3206
756	ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);	3207	ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
757
758	=item ev_feed_event (loop, watcher, int events)
759
760	Feeds the given event set into the event loop, as if the specified event
761	had happened for the specified watcher (which must be a pointer to an
762	initialised but not necessarily started event watcher).
763		3208
764	=item ev_feed_fd_event (loop, int fd, int revents)	3209	=item ev_feed_fd_event (loop, int fd, int revents)
765		3210
766	Feed an event on the given fd, as if a file descriptor backend detected	3211	Feed an event on the given fd, as if a file descriptor backend detected
767	the given events it.	3212	the given events it.
768		3213
769	=item ev_feed_signal_event (loop, int signum)	3214	=item ev_feed_signal_event (loop, int signum)
770		3215
771	Feed an event as if the given signal occured (loop must be the default loop!).	3216	Feed an event as if the given signal occurred (C<loop> must be the default
		3217	loop!).
772		3218
773	=back	3219	=back
774		3220
		3221
775	=head1 LIBEVENT EMULATION	3222	=head1 LIBEVENT EMULATION
776		3223
777	TBD.	3224	Libev offers a compatibility emulation layer for libevent. It cannot
		3225	emulate the internals of libevent, so here are some usage hints:
		3226
		3227	=over 4
		3228
		3229	=item * Use it by including <event.h>, as usual.
		3230
		3231	=item * The following members are fully supported: ev_base, ev_callback,
		3232	ev_arg, ev_fd, ev_res, ev_events.
		3233
		3234	=item * Avoid using ev_flags and the EVLIST_*-macros, while it is
		3235	maintained by libev, it does not work exactly the same way as in libevent (consider
		3236	it a private API).
		3237
		3238	=item * Priorities are not currently supported. Initialising priorities
		3239	will fail and all watchers will have the same priority, even though there
		3240	is an ev_pri field.
		3241
		3242	=item * In libevent, the last base created gets the signals, in libev, the
		3243	first base created (== the default loop) gets the signals.
		3244
		3245	=item * Other members are not supported.
		3246
		3247	=item * The libev emulation is I<not> ABI compatible to libevent, you need
		3248	to use the libev header file and library.
		3249
		3250	=back
778		3251
779	=head1 C++ SUPPORT	3252	=head1 C++ SUPPORT
780		3253
781	TBD.	3254	Libev comes with some simplistic wrapper classes for C++ that mainly allow
		3255	you to use some convenience methods to start/stop watchers and also change
		3256	the callback model to a model using method callbacks on objects.
		3257
		3258	To use it,
		3259
		3260	#include <ev++.h>
		3261
		3262	This automatically includes F<ev.h> and puts all of its definitions (many
		3263	of them macros) into the global namespace. All C++ specific things are
		3264	put into the C<ev> namespace. It should support all the same embedding
		3265	options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
		3266
		3267	Care has been taken to keep the overhead low. The only data member the C++
		3268	classes add (compared to plain C-style watchers) is the event loop pointer
		3269	that the watcher is associated with (or no additional members at all if
		3270	you disable C<EV_MULTIPLICITY> when embedding libev).
		3271
		3272	Currently, functions, and static and non-static member functions can be
		3273	used as callbacks. Other types should be easy to add as long as they only
		3274	need one additional pointer for context. If you need support for other
		3275	types of functors please contact the author (preferably after implementing
		3276	it).
		3277
		3278	Here is a list of things available in the C<ev> namespace:
		3279
		3280	=over 4
		3281
		3282	=item C<ev::READ>, C<ev::WRITE> etc.
		3283
		3284	These are just enum values with the same values as the C<EV_READ> etc.
		3285	macros from F<ev.h>.
		3286
		3287	=item C<ev::tstamp>, C<ev::now>
		3288
		3289	Aliases to the same types/functions as with the C<ev_> prefix.
		3290
		3291	=item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
		3292
		3293	For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
		3294	the same name in the C<ev> namespace, with the exception of C<ev_signal>
		3295	which is called C<ev::sig> to avoid clashes with the C<signal> macro
		3296	defines by many implementations.
		3297
		3298	All of those classes have these methods:
		3299
		3300	=over 4
		3301
		3302	=item ev::TYPE::TYPE ()
		3303
		3304	=item ev::TYPE::TYPE (loop)
		3305
		3306	=item ev::TYPE::~TYPE
		3307
		3308	The constructor (optionally) takes an event loop to associate the watcher
		3309	with. If it is omitted, it will use C<EV_DEFAULT>.
		3310
		3311	The constructor calls C<ev_init> for you, which means you have to call the
		3312	C<set> method before starting it.
		3313
		3314	It will not set a callback, however: You have to call the templated C<set>
		3315	method to set a callback before you can start the watcher.
		3316
		3317	(The reason why you have to use a method is a limitation in C++ which does
		3318	not allow explicit template arguments for constructors).
		3319
		3320	The destructor automatically stops the watcher if it is active.
		3321
		3322	=item w->set<class, &class::method> (object *)
		3323
		3324	This method sets the callback method to call. The method has to have a
		3325	signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
		3326	first argument and the C<revents> as second. The object must be given as
		3327	parameter and is stored in the C<data> member of the watcher.
		3328
		3329	This method synthesizes efficient thunking code to call your method from
		3330	the C callback that libev requires. If your compiler can inline your
		3331	callback (i.e. it is visible to it at the place of the C<set> call and
		3332	your compiler is good :), then the method will be fully inlined into the
		3333	thunking function, making it as fast as a direct C callback.
		3334
		3335	Example: simple class declaration and watcher initialisation
		3336
		3337	struct myclass
		3338	{
		3339	void io_cb (ev::io &w, int revents) { }
		3340	}
		3341
		3342	myclass obj;
		3343	ev::io iow;
		3344	iow.set <myclass, &myclass::io_cb> (&obj);
		3345
		3346	=item w->set (object *)
		3347
		3348	This is a variation of a method callback - leaving out the method to call
		3349	will default the method to C<operator ()>, which makes it possible to use
		3350	functor objects without having to manually specify the C<operator ()> all
		3351	the time. Incidentally, you can then also leave out the template argument
		3352	list.
		3353
		3354	The C<operator ()> method prototype must be C<void operator ()(watcher &w,
		3355	int revents)>.
		3356
		3357	See the method-C<set> above for more details.
		3358
		3359	Example: use a functor object as callback.
		3360
		3361	struct myfunctor
		3362	{
		3363	void operator() (ev::io &w, int revents)
		3364	{
		3365	...
		3366	}
		3367	}
		3368
		3369	myfunctor f;
		3370
		3371	ev::io w;
		3372	w.set (&f);
		3373
		3374	=item w->set<function> (void *data = 0)
		3375
		3376	Also sets a callback, but uses a static method or plain function as
		3377	callback. The optional C<data> argument will be stored in the watcher's
		3378	C<data> member and is free for you to use.
		3379
		3380	The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
		3381
		3382	See the method-C<set> above for more details.
		3383
		3384	Example: Use a plain function as callback.
		3385
		3386	static void io_cb (ev::io &w, int revents) { }
		3387	iow.set <io_cb> ();
		3388
		3389	=item w->set (loop)
		3390
		3391	Associates a different C<struct ev_loop> with this watcher. You can only
		3392	do this when the watcher is inactive (and not pending either).
		3393
		3394	=item w->set ([arguments])
		3395
		3396	Basically the same as C<ev_TYPE_set>, with the same arguments. Must be
		3397	called at least once. Unlike the C counterpart, an active watcher gets
		3398	automatically stopped and restarted when reconfiguring it with this
		3399	method.
		3400
		3401	=item w->start ()
		3402
		3403	Starts the watcher. Note that there is no C<loop> argument, as the
		3404	constructor already stores the event loop.
		3405
		3406	=item w->stop ()
		3407
		3408	Stops the watcher if it is active. Again, no C<loop> argument.
		3409
		3410	=item w->again () (C<ev::timer>, C<ev::periodic> only)
		3411
		3412	For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
		3413	C<ev_TYPE_again> function.
		3414
		3415	=item w->sweep () (C<ev::embed> only)
		3416
		3417	Invokes C<ev_embed_sweep>.
		3418
		3419	=item w->update () (C<ev::stat> only)
		3420
		3421	Invokes C<ev_stat_stat>.
		3422
		3423	=back
		3424
		3425	=back
		3426
		3427	Example: Define a class with an IO and idle watcher, start one of them in
		3428	the constructor.
		3429
		3430	class myclass
		3431	{
		3432	ev::io io ; void io_cb (ev::io &w, int revents);
		3433	ev::idle idle; void idle_cb (ev::idle &w, int revents);
		3434
		3435	myclass (int fd)
		3436	{
		3437	io .set <myclass, &myclass::io_cb > (this);
		3438	idle.set <myclass, &myclass::idle_cb> (this);
		3439
		3440	io.start (fd, ev::READ);
		3441	}
		3442	};
		3443
		3444
		3445	=head1 OTHER LANGUAGE BINDINGS
		3446
		3447	Libev does not offer other language bindings itself, but bindings for a
		3448	number of languages exist in the form of third-party packages. If you know
		3449	any interesting language binding in addition to the ones listed here, drop
		3450	me a note.
		3451
		3452	=over 4
		3453
		3454	=item Perl
		3455
		3456	The EV module implements the full libev API and is actually used to test
		3457	libev. EV is developed together with libev. Apart from the EV core module,
		3458	there are additional modules that implement libev-compatible interfaces
		3459	to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
		3460	C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
		3461	and C<EV::Glib>).
		3462
		3463	It can be found and installed via CPAN, its homepage is at
		3464	L<http://software.schmorp.de/pkg/EV>.
		3465
		3466	=item Python
		3467
		3468	Python bindings can be found at L<http://code.google.com/p/pyev/>. It
		3469	seems to be quite complete and well-documented.
		3470
		3471	=item Ruby
		3472
		3473	Tony Arcieri has written a ruby extension that offers access to a subset
		3474	of the libev API and adds file handle abstractions, asynchronous DNS and
		3475	more on top of it. It can be found via gem servers. Its homepage is at
		3476	L<http://rev.rubyforge.org/>.
		3477
		3478	Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
		3479	makes rev work even on mingw.
		3480
		3481	=item Haskell
		3482
		3483	A haskell binding to libev is available at
		3484	L<http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
		3485
		3486	=item D
		3487
		3488	Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
		3489	be found at L<http://proj.llucax.com.ar/wiki/evd>.
		3490
		3491	=item Ocaml
		3492
		3493	Erkki Seppala has written Ocaml bindings for libev, to be found at
		3494	L<http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
		3495
		3496	=item Lua
		3497
		3498	Brian Maher has written a partial interface to libev for lua (at the
		3499	time of this writing, only C<ev_io> and C<ev_timer>), to be found at
		3500	L<http://github.com/brimworks/lua-ev>.
		3501
		3502	=back
		3503
		3504
		3505	=head1 MACRO MAGIC
		3506
		3507	Libev can be compiled with a variety of options, the most fundamental
		3508	of which is C<EV_MULTIPLICITY>. This option determines whether (most)
		3509	functions and callbacks have an initial C<struct ev_loop *> argument.
		3510
		3511	To make it easier to write programs that cope with either variant, the
		3512	following macros are defined:
		3513
		3514	=over 4
		3515
		3516	=item C<EV_A>, C<EV_A_>
		3517
		3518	This provides the loop I<argument> for functions, if one is required ("ev
		3519	loop argument"). The C<EV_A> form is used when this is the sole argument,
		3520	C<EV_A_> is used when other arguments are following. Example:
		3521
		3522	ev_unref (EV_A);
		3523	ev_timer_add (EV_A_ watcher);
		3524	ev_loop (EV_A_ 0);
		3525
		3526	It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
		3527	which is often provided by the following macro.
		3528
		3529	=item C<EV_P>, C<EV_P_>
		3530
		3531	This provides the loop I<parameter> for functions, if one is required ("ev
		3532	loop parameter"). The C<EV_P> form is used when this is the sole parameter,
		3533	C<EV_P_> is used when other parameters are following. Example:
		3534
		3535	// this is how ev_unref is being declared
		3536	static void ev_unref (EV_P);
		3537
		3538	// this is how you can declare your typical callback
		3539	static void cb (EV_P_ ev_timer *w, int revents)
		3540
		3541	It declares a parameter C<loop> of type C<struct ev_loop *>, quite
		3542	suitable for use with C<EV_A>.
		3543
		3544	=item C<EV_DEFAULT>, C<EV_DEFAULT_>
		3545
		3546	Similar to the other two macros, this gives you the value of the default
		3547	loop, if multiple loops are supported ("ev loop default").
		3548
		3549	=item C<EV_DEFAULT_UC>, C<EV_DEFAULT_UC_>
		3550
		3551	Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
		3552	default loop has been initialised (C<UC> == unchecked). Their behaviour
		3553	is undefined when the default loop has not been initialised by a previous
		3554	execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
		3555
		3556	It is often prudent to use C<EV_DEFAULT> when initialising the first
		3557	watcher in a function but use C<EV_DEFAULT_UC> afterwards.
		3558
		3559	=back
		3560
		3561	Example: Declare and initialise a check watcher, utilising the above
		3562	macros so it will work regardless of whether multiple loops are supported
		3563	or not.
		3564
		3565	static void
		3566	check_cb (EV_P_ ev_timer *w, int revents)
		3567	{
		3568	ev_check_stop (EV_A_ w);
		3569	}
		3570
		3571	ev_check check;
		3572	ev_check_init (&check, check_cb);
		3573	ev_check_start (EV_DEFAULT_ &check);
		3574	ev_loop (EV_DEFAULT_ 0);
		3575
		3576	=head1 EMBEDDING
		3577
		3578	Libev can (and often is) directly embedded into host
		3579	applications. Examples of applications that embed it include the Deliantra
		3580	Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
		3581	and rxvt-unicode.
		3582
		3583	The goal is to enable you to just copy the necessary files into your
		3584	source directory without having to change even a single line in them, so
		3585	you can easily upgrade by simply copying (or having a checked-out copy of
		3586	libev somewhere in your source tree).
		3587
		3588	=head2 FILESETS
		3589
		3590	Depending on what features you need you need to include one or more sets of files
		3591	in your application.
		3592
		3593	=head3 CORE EVENT LOOP
		3594
		3595	To include only the libev core (all the C<ev_*> functions), with manual
		3596	configuration (no autoconf):
		3597
		3598	#define EV_STANDALONE 1
		3599	#include "ev.c"
		3600
		3601	This will automatically include F<ev.h>, too, and should be done in a
		3602	single C source file only to provide the function implementations. To use
		3603	it, do the same for F<ev.h> in all files wishing to use this API (best
		3604	done by writing a wrapper around F<ev.h> that you can include instead and
		3605	where you can put other configuration options):
		3606
		3607	#define EV_STANDALONE 1
		3608	#include "ev.h"
		3609
		3610	Both header files and implementation files can be compiled with a C++
		3611	compiler (at least, that's a stated goal, and breakage will be treated
		3612	as a bug).
		3613
		3614	You need the following files in your source tree, or in a directory
		3615	in your include path (e.g. in libev/ when using -Ilibev):
		3616
		3617	ev.h
		3618	ev.c
		3619	ev_vars.h
		3620	ev_wrap.h
		3621
		3622	ev_win32.c required on win32 platforms only
		3623
		3624	ev_select.c only when select backend is enabled (which is enabled by default)
		3625	ev_poll.c only when poll backend is enabled (disabled by default)
		3626	ev_epoll.c only when the epoll backend is enabled (disabled by default)
		3627	ev_kqueue.c only when the kqueue backend is enabled (disabled by default)
		3628	ev_port.c only when the solaris port backend is enabled (disabled by default)
		3629
		3630	F<ev.c> includes the backend files directly when enabled, so you only need
		3631	to compile this single file.
		3632
		3633	=head3 LIBEVENT COMPATIBILITY API
		3634
		3635	To include the libevent compatibility API, also include:
		3636
		3637	#include "event.c"
		3638
		3639	in the file including F<ev.c>, and:
		3640
		3641	#include "event.h"
		3642
		3643	in the files that want to use the libevent API. This also includes F<ev.h>.
		3644
		3645	You need the following additional files for this:
		3646
		3647	event.h
		3648	event.c
		3649
		3650	=head3 AUTOCONF SUPPORT
		3651
		3652	Instead of using C<EV_STANDALONE=1> and providing your configuration in
		3653	whatever way you want, you can also C<m4_include([libev.m4])> in your
		3654	F<configure.ac> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
		3655	include F<config.h> and configure itself accordingly.
		3656
		3657	For this of course you need the m4 file:
		3658
		3659	libev.m4
		3660
		3661	=head2 PREPROCESSOR SYMBOLS/MACROS
		3662
		3663	Libev can be configured via a variety of preprocessor symbols you have to
		3664	define before including (or compiling) any of its files. The default in
		3665	the absence of autoconf is documented for every option.
		3666
		3667	Symbols marked with "(h)" do not change the ABI, and can have different
		3668	values when compiling libev vs. including F<ev.h>, so it is permissible
		3669	to redefine them before including F<ev.h> without breaking compatibility
		3670	to a compiled library. All other symbols change the ABI, which means all
		3671	users of libev and the libev code itself must be compiled with compatible
		3672	settings.
		3673
		3674	=over 4
		3675
		3676	=item EV_STANDALONE (h)
		3677
		3678	Must always be C<1> if you do not use autoconf configuration, which
		3679	keeps libev from including F<config.h>, and it also defines dummy
		3680	implementations for some libevent functions (such as logging, which is not
		3681	supported). It will also not define any of the structs usually found in
		3682	F<event.h> that are not directly supported by the libev core alone.
		3683
		3684	In standalone mode, libev will still try to automatically deduce the
		3685	configuration, but has to be more conservative.
		3686
		3687	=item EV_USE_MONOTONIC
		3688
		3689	If defined to be C<1>, libev will try to detect the availability of the
		3690	monotonic clock option at both compile time and runtime. Otherwise no
		3691	use of the monotonic clock option will be attempted. If you enable this,
		3692	you usually have to link against librt or something similar. Enabling it
		3693	when the functionality isn't available is safe, though, although you have
		3694	to make sure you link against any libraries where the C<clock_gettime>
		3695	function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
		3696
		3697	=item EV_USE_REALTIME
		3698
		3699	If defined to be C<1>, libev will try to detect the availability of the
		3700	real-time clock option at compile time (and assume its availability
		3701	at runtime if successful). Otherwise no use of the real-time clock
		3702	option will be attempted. This effectively replaces C<gettimeofday>
		3703	by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
		3704	correctness. See the note about libraries in the description of
		3705	C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
		3706	C<EV_USE_CLOCK_SYSCALL>.
		3707
		3708	=item EV_USE_CLOCK_SYSCALL
		3709
		3710	If defined to be C<1>, libev will try to use a direct syscall instead
		3711	of calling the system-provided C<clock_gettime> function. This option
		3712	exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
		3713	unconditionally pulls in C<libpthread>, slowing down single-threaded
		3714	programs needlessly. Using a direct syscall is slightly slower (in
		3715	theory), because no optimised vdso implementation can be used, but avoids
		3716	the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
		3717	higher, as it simplifies linking (no need for C<-lrt>).
		3718
		3719	=item EV_USE_NANOSLEEP
		3720
		3721	If defined to be C<1>, libev will assume that C<nanosleep ()> is available
		3722	and will use it for delays. Otherwise it will use C<select ()>.
		3723
		3724	=item EV_USE_EVENTFD
		3725
		3726	If defined to be C<1>, then libev will assume that C<eventfd ()> is
		3727	available and will probe for kernel support at runtime. This will improve
		3728	C<ev_signal> and C<ev_async> performance and reduce resource consumption.
		3729	If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
		3730	2.7 or newer, otherwise disabled.
		3731
		3732	=item EV_USE_SELECT
		3733
		3734	If undefined or defined to be C<1>, libev will compile in support for the
		3735	C<select>(2) backend. No attempt at auto-detection will be done: if no
		3736	other method takes over, select will be it. Otherwise the select backend
		3737	will not be compiled in.
		3738
		3739	=item EV_SELECT_USE_FD_SET
		3740
		3741	If defined to C<1>, then the select backend will use the system C<fd_set>
		3742	structure. This is useful if libev doesn't compile due to a missing
		3743	C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
		3744	on exotic systems. This usually limits the range of file descriptors to
		3745	some low limit such as 1024 or might have other limitations (winsocket
		3746	only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
		3747	configures the maximum size of the C<fd_set>.
		3748
		3749	=item EV_SELECT_IS_WINSOCKET
		3750
		3751	When defined to C<1>, the select backend will assume that
		3752	select/socket/connect etc. don't understand file descriptors but
		3753	wants osf handles on win32 (this is the case when the select to
		3754	be used is the winsock select). This means that it will call
		3755	C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
		3756	it is assumed that all these functions actually work on fds, even
		3757	on win32. Should not be defined on non-win32 platforms.
		3758
		3759	=item EV_FD_TO_WIN32_HANDLE(fd)
		3760
		3761	If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
		3762	file descriptors to socket handles. When not defining this symbol (the
		3763	default), then libev will call C<_get_osfhandle>, which is usually
		3764	correct. In some cases, programs use their own file descriptor management,
		3765	in which case they can provide this function to map fds to socket handles.
		3766
		3767	=item EV_WIN32_HANDLE_TO_FD(handle)
		3768
		3769	If C<EV_SELECT_IS_WINSOCKET> then libev maps handles to file descriptors
		3770	using the standard C<_open_osfhandle> function. For programs implementing
		3771	their own fd to handle mapping, overwriting this function makes it easier
		3772	to do so. This can be done by defining this macro to an appropriate value.
		3773
		3774	=item EV_WIN32_CLOSE_FD(fd)
		3775
		3776	If programs implement their own fd to handle mapping on win32, then this
		3777	macro can be used to override the C<close> function, useful to unregister
		3778	file descriptors again. Note that the replacement function has to close
		3779	the underlying OS handle.
		3780
		3781	=item EV_USE_POLL
		3782
		3783	If defined to be C<1>, libev will compile in support for the C<poll>(2)
		3784	backend. Otherwise it will be enabled on non-win32 platforms. It
		3785	takes precedence over select.
		3786
		3787	=item EV_USE_EPOLL
		3788
		3789	If defined to be C<1>, libev will compile in support for the Linux
		3790	C<epoll>(7) backend. Its availability will be detected at runtime,
		3791	otherwise another method will be used as fallback. This is the preferred
		3792	backend for GNU/Linux systems. If undefined, it will be enabled if the
		3793	headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
		3794
		3795	=item EV_USE_KQUEUE
		3796
		3797	If defined to be C<1>, libev will compile in support for the BSD style
		3798	C<kqueue>(2) backend. Its actual availability will be detected at runtime,
		3799	otherwise another method will be used as fallback. This is the preferred
		3800	backend for BSD and BSD-like systems, although on most BSDs kqueue only
		3801	supports some types of fds correctly (the only platform we found that
		3802	supports ptys for example was NetBSD), so kqueue might be compiled in, but
		3803	not be used unless explicitly requested. The best way to use it is to find
		3804	out whether kqueue supports your type of fd properly and use an embedded
		3805	kqueue loop.
		3806
		3807	=item EV_USE_PORT
		3808
		3809	If defined to be C<1>, libev will compile in support for the Solaris
		3810	10 port style backend. Its availability will be detected at runtime,
		3811	otherwise another method will be used as fallback. This is the preferred
		3812	backend for Solaris 10 systems.
		3813
		3814	=item EV_USE_DEVPOLL
		3815
		3816	Reserved for future expansion, works like the USE symbols above.
		3817
		3818	=item EV_USE_INOTIFY
		3819
		3820	If defined to be C<1>, libev will compile in support for the Linux inotify
		3821	interface to speed up C<ev_stat> watchers. Its actual availability will
		3822	be detected at runtime. If undefined, it will be enabled if the headers
		3823	indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
		3824
		3825	=item EV_ATOMIC_T
		3826
		3827	Libev requires an integer type (suitable for storing C<0> or C<1>) whose
		3828	access is atomic with respect to other threads or signal contexts. No such
		3829	type is easily found in the C language, so you can provide your own type
		3830	that you know is safe for your purposes. It is used both for signal handler "locking"
		3831	as well as for signal and thread safety in C<ev_async> watchers.
		3832
		3833	In the absence of this define, libev will use C<sig_atomic_t volatile>
		3834	(from F<signal.h>), which is usually good enough on most platforms.
		3835
		3836	=item EV_H (h)
		3837
		3838	The name of the F<ev.h> header file used to include it. The default if
		3839	undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
		3840	used to virtually rename the F<ev.h> header file in case of conflicts.
		3841
		3842	=item EV_CONFIG_H (h)
		3843
		3844	If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
		3845	F<ev.c>'s idea of where to find the F<config.h> file, similarly to
		3846	C<EV_H>, above.
		3847
		3848	=item EV_EVENT_H (h)
		3849
		3850	Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
		3851	of how the F<event.h> header can be found, the default is C<"event.h">.
		3852
		3853	=item EV_PROTOTYPES (h)
		3854
		3855	If defined to be C<0>, then F<ev.h> will not define any function
		3856	prototypes, but still define all the structs and other symbols. This is
		3857	occasionally useful if you want to provide your own wrapper functions
		3858	around libev functions.
		3859
		3860	=item EV_MULTIPLICITY
		3861
		3862	If undefined or defined to C<1>, then all event-loop-specific functions
		3863	will have the C<struct ev_loop *> as first argument, and you can create
		3864	additional independent event loops. Otherwise there will be no support
		3865	for multiple event loops and there is no first event loop pointer
		3866	argument. Instead, all functions act on the single default loop.
		3867
		3868	=item EV_MINPRI
		3869
		3870	=item EV_MAXPRI
		3871
		3872	The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
		3873	C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
		3874	provide for more priorities by overriding those symbols (usually defined
		3875	to be C<-2> and C<2>, respectively).
		3876
		3877	When doing priority-based operations, libev usually has to linearly search
		3878	all the priorities, so having many of them (hundreds) uses a lot of space
		3879	and time, so using the defaults of five priorities (-2 .. +2) is usually
		3880	fine.
		3881
		3882	If your embedding application does not need any priorities, defining these
		3883	both to C<0> will save some memory and CPU.
		3884
		3885	=item EV_PERIODIC_ENABLE, EV_IDLE_ENABLE, EV_EMBED_ENABLE, EV_STAT_ENABLE,
		3886	EV_PREPARE_ENABLE, EV_CHECK_ENABLE, EV_FORK_ENABLE, EV_SIGNAL_ENABLE,
		3887	EV_ASYNC_ENABLE, EV_CHILD_ENABLE.
		3888
		3889	If undefined or defined to be C<1> (and the platform supports it), then
		3890	the respective watcher type is supported. If defined to be C<0>, then it
		3891	is not. Disabling watcher types mainly saves code size.
		3892
		3893	=item EV_FEATURES
		3894
		3895	If you need to shave off some kilobytes of code at the expense of some
		3896	speed (but with the full API), you can define this symbol to request
		3897	certain subsets of functionality. The default is to enable all features
		3898	that can be enabled on the platform.
		3899
		3900	A typical way to use this symbol is to define it to C<0> (or to a bitset
		3901	with some broad features you want) and then selectively re-enable
		3902	additional parts you want, for example if you want everything minimal,
		3903	but multiple event loop support, async and child watchers and the poll
		3904	backend, use this:
		3905
		3906	#define EV_FEATURES 0
		3907	#define EV_MULTIPLICITY 1
		3908	#define EV_USE_POLL 1
		3909	#define EV_CHILD_ENABLE 1
		3910	#define EV_ASYNC_ENABLE 1
		3911
		3912	The actual value is a bitset, it can be a combination of the following
		3913	values:
		3914
		3915	=over 4
		3916
		3917	=item C<1> - faster/larger code
		3918
		3919	Use larger code to speed up some operations.
		3920
		3921	Currently this is used to override some inlining decisions (enlarging the
		3922	code size by roughly 30% on amd64).
		3923
		3924	When optimising for size, use of compiler flags such as C<-Os> with
		3925	gcc is recommended, as well as C<-DNDEBUG>, as libev contains a number of
		3926	assertions.
		3927
		3928	=item C<2> - faster/larger data structures
		3929
		3930	Replaces the small 2-heap for timer management by a faster 4-heap, larger
		3931	hash table sizes and so on. This will usually further increase code size
		3932	and can additionally have an effect on the size of data structures at
		3933	runtime.
		3934
		3935	=item C<4> - full API configuration
		3936
		3937	This enables priorities (sets C<EV_MAXPRI>=2 and C<EV_MINPRI>=-2), and
		3938	enables multiplicity (C<EV_MULTIPLICITY>=1).
		3939
		3940	=item C<8> - full API
		3941
		3942	This enables a lot of the "lesser used" API functions. See C<ev.h> for
		3943	details on which parts of the API are still available without this
		3944	feature, and do not complain if this subset changes over time.
		3945
		3946	=item C<16> - enable all optional watcher types
		3947
		3948	Enables all optional watcher types. If you want to selectively enable
		3949	only some watcher types other than I/O and timers (e.g. prepare,
		3950	embed, async, child...) you can enable them manually by defining
		3951	C<EV_watchertype_ENABLE> to C<1> instead.
		3952
		3953	=item C<32> - enable all backends
		3954
		3955	This enables all backends - without this feature, you need to enable at
		3956	least one backend manually (C<EV_USE_SELECT> is a good choice).
		3957
		3958	=item C<64> - enable OS-specific "helper" APIs
		3959
		3960	Enable inotify, eventfd, signalfd and similar OS-specific helper APIs by
		3961	default.
		3962
		3963	=back
		3964
		3965	Compiling with C<gcc -Os -DEV_STANDALONE -DEV_USE_EPOLL=1 -DEV_FEATURES=0>
		3966	reduces the compiled size of libev from 24.7Kb code/2.8Kb data to 6.5Kb
		3967	code/0.3Kb data on my GNU/Linux amd64 system, while still giving you I/O
		3968	watchers, timers and monotonic clock support.
		3969
		3970	With an intelligent-enough linker (gcc+binutils are intelligent enough
		3971	when you use C<-Wl,--gc-sections -ffunction-sections>) functions unused by
		3972	your program might be left out as well - a binary starting a timer and an
		3973	I/O watcher then might come out at only 5Kb.
		3974
		3975	=item EV_AVOID_STDIO
		3976
		3977	If this is set to C<1> at compiletime, then libev will avoid using stdio
		3978	functions (printf, scanf, perror etc.). This will increase the code size
		3979	somewhat, but if your program doesn't otherwise depend on stdio and your
		3980	libc allows it, this avoids linking in the stdio library which is quite
		3981	big.
		3982
		3983	Note that error messages might become less precise when this option is
		3984	enabled.
		3985
		3986	=item EV_NSIG
		3987
		3988	The highest supported signal number, +1 (or, the number of
		3989	signals): Normally, libev tries to deduce the maximum number of signals
		3990	automatically, but sometimes this fails, in which case it can be
		3991	specified. Also, using a lower number than detected (C<32> should be
		3992	good for about any system in existence) can save some memory, as libev
		3993	statically allocates some 12-24 bytes per signal number.
		3994
		3995	=item EV_PID_HASHSIZE
		3996
		3997	C<ev_child> watchers use a small hash table to distribute workload by
		3998	pid. The default size is C<16> (or C<1> with C<EV_FEATURES> disabled),
		3999	usually more than enough. If you need to manage thousands of children you
		4000	might want to increase this value (I<must> be a power of two).
		4001
		4002	=item EV_INOTIFY_HASHSIZE
		4003
		4004	C<ev_stat> watchers use a small hash table to distribute workload by
		4005	inotify watch id. The default size is C<16> (or C<1> with C<EV_FEATURES>
		4006	disabled), usually more than enough. If you need to manage thousands of
		4007	C<ev_stat> watchers you might want to increase this value (I<must> be a
		4008	power of two).
		4009
		4010	=item EV_USE_4HEAP
		4011
		4012	Heaps are not very cache-efficient. To improve the cache-efficiency of the
		4013	timer and periodics heaps, libev uses a 4-heap when this symbol is defined
		4014	to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
		4015	faster performance with many (thousands) of watchers.
		4016
		4017	The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
		4018	will be C<0>.
		4019
		4020	=item EV_HEAP_CACHE_AT
		4021
		4022	Heaps are not very cache-efficient. To improve the cache-efficiency of the
		4023	timer and periodics heaps, libev can cache the timestamp (I<at>) within
		4024	the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
		4025	which uses 8-12 bytes more per watcher and a few hundred bytes more code,
		4026	but avoids random read accesses on heap changes. This improves performance
		4027	noticeably with many (hundreds) of watchers.
		4028
		4029	The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
		4030	will be C<0>.
		4031
		4032	=item EV_VERIFY
		4033
		4034	Controls how much internal verification (see C<ev_loop_verify ()>) will
		4035	be done: If set to C<0>, no internal verification code will be compiled
		4036	in. If set to C<1>, then verification code will be compiled in, but not
		4037	called. If set to C<2>, then the internal verification code will be
		4038	called once per loop, which can slow down libev. If set to C<3>, then the
		4039	verification code will be called very frequently, which will slow down
		4040	libev considerably.
		4041
		4042	The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
		4043	will be C<0>.
		4044
		4045	=item EV_COMMON
		4046
		4047	By default, all watchers have a C<void *data> member. By redefining
		4048	this macro to something else you can include more and other types of
		4049	members. You have to define it each time you include one of the files,
		4050	though, and it must be identical each time.
		4051
		4052	For example, the perl EV module uses something like this:
		4053
		4054	#define EV_COMMON \
		4055	SV *self; /* contains this struct */ \
		4056	SV *cb_sv, *fh /* note no trailing ";" */
		4057
		4058	=item EV_CB_DECLARE (type)
		4059
		4060	=item EV_CB_INVOKE (watcher, revents)
		4061
		4062	=item ev_set_cb (ev, cb)
		4063
		4064	Can be used to change the callback member declaration in each watcher,
		4065	and the way callbacks are invoked and set. Must expand to a struct member
		4066	definition and a statement, respectively. See the F<ev.h> header file for
		4067	their default definitions. One possible use for overriding these is to
		4068	avoid the C<struct ev_loop *> as first argument in all cases, or to use
		4069	method calls instead of plain function calls in C++.
		4070
		4071	=back
		4072
		4073	=head2 EXPORTED API SYMBOLS
		4074
		4075	If you need to re-export the API (e.g. via a DLL) and you need a list of
		4076	exported symbols, you can use the provided F<Symbol.*> files which list
		4077	all public symbols, one per line:
		4078
		4079	Symbols.ev for libev proper
		4080	Symbols.event for the libevent emulation
		4081
		4082	This can also be used to rename all public symbols to avoid clashes with
		4083	multiple versions of libev linked together (which is obviously bad in
		4084	itself, but sometimes it is inconvenient to avoid this).
		4085
		4086	A sed command like this will create wrapper C<#define>'s that you need to
		4087	include before including F<ev.h>:
		4088
		4089	<Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
		4090
		4091	This would create a file F<wrap.h> which essentially looks like this:
		4092
		4093	#define ev_backend myprefix_ev_backend
		4094	#define ev_check_start myprefix_ev_check_start
		4095	#define ev_check_stop myprefix_ev_check_stop
		4096	...
		4097
		4098	=head2 EXAMPLES
		4099
		4100	For a real-world example of a program the includes libev
		4101	verbatim, you can have a look at the EV perl module
		4102	(L<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
		4103	the F<libev/> subdirectory and includes them in the F<EV/EVAPI.h> (public
		4104	interface) and F<EV.xs> (implementation) files. Only the F<EV.xs> file
		4105	will be compiled. It is pretty complex because it provides its own header
		4106	file.
		4107
		4108	The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
		4109	that everybody includes and which overrides some configure choices:
		4110
		4111	#define EV_FEATURES 8
		4112	#define EV_USE_SELECT 1
		4113	#define EV_PREPARE_ENABLE 1
		4114	#define EV_IDLE_ENABLE 1
		4115	#define EV_SIGNAL_ENABLE 1
		4116	#define EV_CHILD_ENABLE 1
		4117	#define EV_USE_STDEXCEPT 0
		4118	#define EV_CONFIG_H <config.h>
		4119
		4120	#include "ev++.h"
		4121
		4122	And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
		4123
		4124	#include "ev_cpp.h"
		4125	#include "ev.c"
		4126
		4127	=head1 INTERACTION WITH OTHER PROGRAMS OR LIBRARIES
		4128
		4129	=head2 THREADS AND COROUTINES
		4130
		4131	=head3 THREADS
		4132
		4133	All libev functions are reentrant and thread-safe unless explicitly
		4134	documented otherwise, but libev implements no locking itself. This means
		4135	that you can use as many loops as you want in parallel, as long as there
		4136	are no concurrent calls into any libev function with the same loop
		4137	parameter (C<ev_default_*> calls have an implicit default loop parameter,
		4138	of course): libev guarantees that different event loops share no data
		4139	structures that need any locking.
		4140
		4141	Or to put it differently: calls with different loop parameters can be done
		4142	concurrently from multiple threads, calls with the same loop parameter
		4143	must be done serially (but can be done from different threads, as long as
		4144	only one thread ever is inside a call at any point in time, e.g. by using
		4145	a mutex per loop).
		4146
		4147	Specifically to support threads (and signal handlers), libev implements
		4148	so-called C<ev_async> watchers, which allow some limited form of
		4149	concurrency on the same event loop, namely waking it up "from the
		4150	outside".
		4151
		4152	If you want to know which design (one loop, locking, or multiple loops
		4153	without or something else still) is best for your problem, then I cannot
		4154	help you, but here is some generic advice:
		4155
		4156	=over 4
		4157
		4158	=item * most applications have a main thread: use the default libev loop
		4159	in that thread, or create a separate thread running only the default loop.
		4160
		4161	This helps integrating other libraries or software modules that use libev
		4162	themselves and don't care/know about threading.
		4163
		4164	=item * one loop per thread is usually a good model.
		4165
		4166	Doing this is almost never wrong, sometimes a better-performance model
		4167	exists, but it is always a good start.
		4168
		4169	=item * other models exist, such as the leader/follower pattern, where one
		4170	loop is handed through multiple threads in a kind of round-robin fashion.
		4171
		4172	Choosing a model is hard - look around, learn, know that usually you can do
		4173	better than you currently do :-)
		4174
		4175	=item * often you need to talk to some other thread which blocks in the
		4176	event loop.
		4177
		4178	C<ev_async> watchers can be used to wake them up from other threads safely
		4179	(or from signal contexts...).
		4180
		4181	An example use would be to communicate signals or other events that only
		4182	work in the default loop by registering the signal watcher with the
		4183	default loop and triggering an C<ev_async> watcher from the default loop
		4184	watcher callback into the event loop interested in the signal.
		4185
		4186	=back
		4187
		4188	=head4 THREAD LOCKING EXAMPLE
		4189
		4190	Here is a fictitious example of how to run an event loop in a different
		4191	thread than where callbacks are being invoked and watchers are
		4192	created/added/removed.
		4193
		4194	For a real-world example, see the C<EV::Loop::Async> perl module,
		4195	which uses exactly this technique (which is suited for many high-level
		4196	languages).
		4197
		4198	The example uses a pthread mutex to protect the loop data, a condition
		4199	variable to wait for callback invocations, an async watcher to notify the
		4200	event loop thread and an unspecified mechanism to wake up the main thread.
		4201
		4202	First, you need to associate some data with the event loop:
		4203
		4204	typedef struct {
		4205	mutex_t lock; /* global loop lock */
		4206	ev_async async_w;
		4207	thread_t tid;
		4208	cond_t invoke_cv;
		4209	} userdata;
		4210
		4211	void prepare_loop (EV_P)
		4212	{
		4213	// for simplicity, we use a static userdata struct.
		4214	static userdata u;
		4215
		4216	ev_async_init (&u->async_w, async_cb);
		4217	ev_async_start (EV_A_ &u->async_w);
		4218
		4219	pthread_mutex_init (&u->lock, 0);
		4220	pthread_cond_init (&u->invoke_cv, 0);
		4221
		4222	// now associate this with the loop
		4223	ev_set_userdata (EV_A_ u);
		4224	ev_set_invoke_pending_cb (EV_A_ l_invoke);
		4225	ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
		4226
		4227	// then create the thread running ev_loop
		4228	pthread_create (&u->tid, 0, l_run, EV_A);
		4229	}
		4230
		4231	The callback for the C<ev_async> watcher does nothing: the watcher is used
		4232	solely to wake up the event loop so it takes notice of any new watchers
		4233	that might have been added:
		4234
		4235	static void
		4236	async_cb (EV_P_ ev_async *w, int revents)
		4237	{
		4238	// just used for the side effects
		4239	}
		4240
		4241	The C<l_release> and C<l_acquire> callbacks simply unlock/lock the mutex
		4242	protecting the loop data, respectively.
		4243
		4244	static void
		4245	l_release (EV_P)
		4246	{
		4247	userdata *u = ev_userdata (EV_A);
		4248	pthread_mutex_unlock (&u->lock);
		4249	}
		4250
		4251	static void
		4252	l_acquire (EV_P)
		4253	{
		4254	userdata *u = ev_userdata (EV_A);
		4255	pthread_mutex_lock (&u->lock);
		4256	}
		4257
		4258	The event loop thread first acquires the mutex, and then jumps straight
		4259	into C<ev_loop>:
		4260
		4261	void *
		4262	l_run (void *thr_arg)
		4263	{
		4264	struct ev_loop *loop = (struct ev_loop *)thr_arg;
		4265
		4266	l_acquire (EV_A);
		4267	pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
		4268	ev_loop (EV_A_ 0);
		4269	l_release (EV_A);
		4270
		4271	return 0;
		4272	}
		4273
		4274	Instead of invoking all pending watchers, the C<l_invoke> callback will
		4275	signal the main thread via some unspecified mechanism (signals? pipe
		4276	writes? C<Async::Interrupt>?) and then waits until all pending watchers
		4277	have been called (in a while loop because a) spurious wakeups are possible
		4278	and b) skipping inter-thread-communication when there are no pending
		4279	watchers is very beneficial):
		4280
		4281	static void
		4282	l_invoke (EV_P)
		4283	{
		4284	userdata *u = ev_userdata (EV_A);
		4285
		4286	while (ev_pending_count (EV_A))
		4287	{
		4288	wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
		4289	pthread_cond_wait (&u->invoke_cv, &u->lock);
		4290	}
		4291	}
		4292
		4293	Now, whenever the main thread gets told to invoke pending watchers, it
		4294	will grab the lock, call C<ev_invoke_pending> and then signal the loop
		4295	thread to continue:
		4296
		4297	static void
		4298	real_invoke_pending (EV_P)
		4299	{
		4300	userdata *u = ev_userdata (EV_A);
		4301
		4302	pthread_mutex_lock (&u->lock);
		4303	ev_invoke_pending (EV_A);
		4304	pthread_cond_signal (&u->invoke_cv);
		4305	pthread_mutex_unlock (&u->lock);
		4306	}
		4307
		4308	Whenever you want to start/stop a watcher or do other modifications to an
		4309	event loop, you will now have to lock:
		4310
		4311	ev_timer timeout_watcher;
		4312	userdata *u = ev_userdata (EV_A);
		4313
		4314	ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
		4315
		4316	pthread_mutex_lock (&u->lock);
		4317	ev_timer_start (EV_A_ &timeout_watcher);
		4318	ev_async_send (EV_A_ &u->async_w);
		4319	pthread_mutex_unlock (&u->lock);
		4320
		4321	Note that sending the C<ev_async> watcher is required because otherwise
		4322	an event loop currently blocking in the kernel will have no knowledge
		4323	about the newly added timer. By waking up the loop it will pick up any new
		4324	watchers in the next event loop iteration.
		4325
		4326	=head3 COROUTINES
		4327
		4328	Libev is very accommodating to coroutines ("cooperative threads"):
		4329	libev fully supports nesting calls to its functions from different
		4330	coroutines (e.g. you can call C<ev_loop> on the same loop from two
		4331	different coroutines, and switch freely between both coroutines running
		4332	the loop, as long as you don't confuse yourself). The only exception is
		4333	that you must not do this from C<ev_periodic> reschedule callbacks.
		4334
		4335	Care has been taken to ensure that libev does not keep local state inside
		4336	C<ev_loop>, and other calls do not usually allow for coroutine switches as
		4337	they do not call any callbacks.
		4338
		4339	=head2 COMPILER WARNINGS
		4340
		4341	Depending on your compiler and compiler settings, you might get no or a
		4342	lot of warnings when compiling libev code. Some people are apparently
		4343	scared by this.
		4344
		4345	However, these are unavoidable for many reasons. For one, each compiler
		4346	has different warnings, and each user has different tastes regarding
		4347	warning options. "Warn-free" code therefore cannot be a goal except when
		4348	targeting a specific compiler and compiler-version.
		4349
		4350	Another reason is that some compiler warnings require elaborate
		4351	workarounds, or other changes to the code that make it less clear and less
		4352	maintainable.
		4353
		4354	And of course, some compiler warnings are just plain stupid, or simply
		4355	wrong (because they don't actually warn about the condition their message
		4356	seems to warn about). For example, certain older gcc versions had some
		4357	warnings that resulted in an extreme number of false positives. These have
		4358	been fixed, but some people still insist on making code warn-free with
		4359	such buggy versions.
		4360
		4361	While libev is written to generate as few warnings as possible,
		4362	"warn-free" code is not a goal, and it is recommended not to build libev
		4363	with any compiler warnings enabled unless you are prepared to cope with
		4364	them (e.g. by ignoring them). Remember that warnings are just that:
		4365	warnings, not errors, or proof of bugs.
		4366
		4367
		4368	=head2 VALGRIND
		4369
		4370	Valgrind has a special section here because it is a popular tool that is
		4371	highly useful. Unfortunately, valgrind reports are very hard to interpret.
		4372
		4373	If you think you found a bug (memory leak, uninitialised data access etc.)
		4374	in libev, then check twice: If valgrind reports something like:
		4375
		4376	==2274== definitely lost: 0 bytes in 0 blocks.
		4377	==2274== possibly lost: 0 bytes in 0 blocks.
		4378	==2274== still reachable: 256 bytes in 1 blocks.
		4379
		4380	Then there is no memory leak, just as memory accounted to global variables
		4381	is not a memleak - the memory is still being referenced, and didn't leak.
		4382
		4383	Similarly, under some circumstances, valgrind might report kernel bugs
		4384	as if it were a bug in libev (e.g. in realloc or in the poll backend,
		4385	although an acceptable workaround has been found here), or it might be
		4386	confused.
		4387
		4388	Keep in mind that valgrind is a very good tool, but only a tool. Don't
		4389	make it into some kind of religion.
		4390
		4391	If you are unsure about something, feel free to contact the mailing list
		4392	with the full valgrind report and an explanation on why you think this
		4393	is a bug in libev (best check the archives, too :). However, don't be
		4394	annoyed when you get a brisk "this is no bug" answer and take the chance
		4395	of learning how to interpret valgrind properly.
		4396
		4397	If you need, for some reason, empty reports from valgrind for your project
		4398	I suggest using suppression lists.
		4399
		4400
		4401	=head1 PORTABILITY NOTES
		4402
		4403	=head2 GNU/LINUX 32 BIT LIMITATIONS
		4404
		4405	GNU/Linux is the only common platform that supports 64 bit file/large file
		4406	interfaces but I<disables> them by default.
		4407
		4408	That means that libev compiled in the default environment doesn't support
		4409	files larger than 2GiB or so, which mainly affects C<ev_stat> watchers.
		4410
		4411	Unfortunately, many programs try to work around this GNU/Linux issue
		4412	by enabling the large file API, which makes them incompatible with the
		4413	standard libev compiled for their system.
		4414
		4415	Likewise, libev cannot enable the large file API itself as this would
		4416	suddenly make it incompatible to the default compile time environment,
		4417	i.e. all programs not using special compile switches.
		4418
		4419	=head2 OS/X AND DARWIN BUGS
		4420
		4421	The whole thing is a bug if you ask me - basically any system interface
		4422	you touch is broken, whether it is locales, poll, kqueue or even the
		4423	OpenGL drivers.
		4424
		4425	=head3 C<kqueue> is buggy
		4426
		4427	The kqueue syscall is broken in all known versions - most versions support
		4428	only sockets, many support pipes.
		4429
		4430	Libev tries to work around this by not using C<kqueue> by default on
		4431	this rotten platform, but of course you can still ask for it when creating
		4432	a loop.
		4433
		4434	=head3 C<poll> is buggy
		4435
		4436	Instead of fixing C<kqueue>, Apple replaced their (working) C<poll>
		4437	implementation by something calling C<kqueue> internally around the 10.5.6
		4438	release, so now C<kqueue> I<and> C<poll> are broken.
		4439
		4440	Libev tries to work around this by not using C<poll> by default on
		4441	this rotten platform, but of course you can still ask for it when creating
		4442	a loop.
		4443
		4444	=head3 C<select> is buggy
		4445
		4446	All that's left is C<select>, and of course Apple found a way to fuck this
		4447	one up as well: On OS/X, C<select> actively limits the number of file
		4448	descriptors you can pass in to 1024 - your program suddenly crashes when
		4449	you use more.
		4450
		4451	There is an undocumented "workaround" for this - defining
		4452	C<_DARWIN_UNLIMITED_SELECT>, which libev tries to use, so select I<should>
		4453	work on OS/X.
		4454
		4455	=head2 SOLARIS PROBLEMS AND WORKAROUNDS
		4456
		4457	=head3 C<errno> reentrancy
		4458
		4459	The default compile environment on Solaris is unfortunately so
		4460	thread-unsafe that you can't even use components/libraries compiled
		4461	without C<-D_REENTRANT> (as long as they use C<errno>), which, of course,
		4462	isn't defined by default.
		4463
		4464	If you want to use libev in threaded environments you have to make sure
		4465	it's compiled with C<_REENTRANT> defined.
		4466
		4467	=head3 Event port backend
		4468
		4469	The scalable event interface for Solaris is called "event ports". Unfortunately,
		4470	this mechanism is very buggy. If you run into high CPU usage, your program
		4471	freezes or you get a large number of spurious wakeups, make sure you have
		4472	all the relevant and latest kernel patches applied. No, I don't know which
		4473	ones, but there are multiple ones.
		4474
		4475	If you can't get it to work, you can try running the program by setting
		4476	the environment variable C<LIBEV_FLAGS=3> to only allow C<poll> and
		4477	C<select> backends.
		4478
		4479	=head2 AIX POLL BUG
		4480
		4481	AIX unfortunately has a broken C<poll.h> header. Libev works around
		4482	this by trying to avoid the poll backend altogether (i.e. it's not even
		4483	compiled in), which normally isn't a big problem as C<select> works fine
		4484	with large bitsets, and AIX is dead anyway.
		4485
		4486	=head2 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
		4487
		4488	=head3 General issues
		4489
		4490	Win32 doesn't support any of the standards (e.g. POSIX) that libev
		4491	requires, and its I/O model is fundamentally incompatible with the POSIX
		4492	model. Libev still offers limited functionality on this platform in
		4493	the form of the C<EVBACKEND_SELECT> backend, and only supports socket
		4494	descriptors. This only applies when using Win32 natively, not when using
		4495	e.g. cygwin. Actually, it only applies to the microsofts own compilers,
		4496	as every compielr comes with a slightly differently broken/incompatible
		4497	environment.
		4498
		4499	Lifting these limitations would basically require the full
		4500	re-implementation of the I/O system. If you are into this kind of thing,
		4501	then note that glib does exactly that for you in a very portable way (note
		4502	also that glib is the slowest event library known to man).
		4503
		4504	There is no supported compilation method available on windows except
		4505	embedding it into other applications.
		4506
		4507	Sensible signal handling is officially unsupported by Microsoft - libev
		4508	tries its best, but under most conditions, signals will simply not work.
		4509
		4510	Not a libev limitation but worth mentioning: windows apparently doesn't
		4511	accept large writes: instead of resulting in a partial write, windows will
		4512	either accept everything or return C<ENOBUFS> if the buffer is too large,
		4513	so make sure you only write small amounts into your sockets (less than a
		4514	megabyte seems safe, but this apparently depends on the amount of memory
		4515	available).
		4516
		4517	Due to the many, low, and arbitrary limits on the win32 platform and
		4518	the abysmal performance of winsockets, using a large number of sockets
		4519	is not recommended (and not reasonable). If your program needs to use
		4520	more than a hundred or so sockets, then likely it needs to use a totally
		4521	different implementation for windows, as libev offers the POSIX readiness
		4522	notification model, which cannot be implemented efficiently on windows
		4523	(due to Microsoft monopoly games).
		4524
		4525	A typical way to use libev under windows is to embed it (see the embedding
		4526	section for details) and use the following F<evwrap.h> header file instead
		4527	of F<ev.h>:
		4528
		4529	#define EV_STANDALONE /* keeps ev from requiring config.h */
		4530	#define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
		4531
		4532	#include "ev.h"
		4533
		4534	And compile the following F<evwrap.c> file into your project (make sure
		4535	you do I<not> compile the F<ev.c> or any other embedded source files!):
		4536
		4537	#include "evwrap.h"
		4538	#include "ev.c"
		4539
		4540	=head3 The winsocket C<select> function
		4541
		4542	The winsocket C<select> function doesn't follow POSIX in that it
		4543	requires socket I<handles> and not socket I<file descriptors> (it is
		4544	also extremely buggy). This makes select very inefficient, and also
		4545	requires a mapping from file descriptors to socket handles (the Microsoft
		4546	C runtime provides the function C<_open_osfhandle> for this). See the
		4547	discussion of the C<EV_SELECT_USE_FD_SET>, C<EV_SELECT_IS_WINSOCKET> and
		4548	C<EV_FD_TO_WIN32_HANDLE> preprocessor symbols for more info.
		4549
		4550	The configuration for a "naked" win32 using the Microsoft runtime
		4551	libraries and raw winsocket select is:
		4552
		4553	#define EV_USE_SELECT 1
		4554	#define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
		4555
		4556	Note that winsockets handling of fd sets is O(n), so you can easily get a
		4557	complexity in the O(n²) range when using win32.
		4558
		4559	=head3 Limited number of file descriptors
		4560
		4561	Windows has numerous arbitrary (and low) limits on things.
		4562
		4563	Early versions of winsocket's select only supported waiting for a maximum
		4564	of C<64> handles (probably owning to the fact that all windows kernels
		4565	can only wait for C<64> things at the same time internally; Microsoft
		4566	recommends spawning a chain of threads and wait for 63 handles and the
		4567	previous thread in each. Sounds great!).
		4568
		4569	Newer versions support more handles, but you need to define C<FD_SETSIZE>
		4570	to some high number (e.g. C<2048>) before compiling the winsocket select
		4571	call (which might be in libev or elsewhere, for example, perl and many
		4572	other interpreters do their own select emulation on windows).
		4573
		4574	Another limit is the number of file descriptors in the Microsoft runtime
		4575	libraries, which by default is C<64> (there must be a hidden I<64>
		4576	fetish or something like this inside Microsoft). You can increase this
		4577	by calling C<_setmaxstdio>, which can increase this limit to C<2048>
		4578	(another arbitrary limit), but is broken in many versions of the Microsoft
		4579	runtime libraries. This might get you to about C<512> or C<2048> sockets
		4580	(depending on windows version and/or the phase of the moon). To get more,
		4581	you need to wrap all I/O functions and provide your own fd management, but
		4582	the cost of calling select (O(n²)) will likely make this unworkable.
		4583
		4584	=head2 PORTABILITY REQUIREMENTS
		4585
		4586	In addition to a working ISO-C implementation and of course the
		4587	backend-specific APIs, libev relies on a few additional extensions:
		4588
		4589	=over 4
		4590
		4591	=item C<void (*)(ev_watcher_type *, int revents)> must have compatible
		4592	calling conventions regardless of C<ev_watcher_type *>.
		4593
		4594	Libev assumes not only that all watcher pointers have the same internal
		4595	structure (guaranteed by POSIX but not by ISO C for example), but it also
		4596	assumes that the same (machine) code can be used to call any watcher
		4597	callback: The watcher callbacks have different type signatures, but libev
		4598	calls them using an C<ev_watcher *> internally.
		4599
		4600	=item C<sig_atomic_t volatile> must be thread-atomic as well
		4601
		4602	The type C<sig_atomic_t volatile> (or whatever is defined as
		4603	C<EV_ATOMIC_T>) must be atomic with respect to accesses from different
		4604	threads. This is not part of the specification for C<sig_atomic_t>, but is
		4605	believed to be sufficiently portable.
		4606
		4607	=item C<sigprocmask> must work in a threaded environment
		4608
		4609	Libev uses C<sigprocmask> to temporarily block signals. This is not
		4610	allowed in a threaded program (C<pthread_sigmask> has to be used). Typical
		4611	pthread implementations will either allow C<sigprocmask> in the "main
		4612	thread" or will block signals process-wide, both behaviours would
		4613	be compatible with libev. Interaction between C<sigprocmask> and
		4614	C<pthread_sigmask> could complicate things, however.
		4615
		4616	The most portable way to handle signals is to block signals in all threads
		4617	except the initial one, and run the default loop in the initial thread as
		4618	well.
		4619
		4620	=item C<long> must be large enough for common memory allocation sizes
		4621
		4622	To improve portability and simplify its API, libev uses C<long> internally
		4623	instead of C<size_t> when allocating its data structures. On non-POSIX
		4624	systems (Microsoft...) this might be unexpectedly low, but is still at
		4625	least 31 bits everywhere, which is enough for hundreds of millions of
		4626	watchers.
		4627
		4628	=item C<double> must hold a time value in seconds with enough accuracy
		4629
		4630	The type C<double> is used to represent timestamps. It is required to
		4631	have at least 51 bits of mantissa (and 9 bits of exponent), which is good
		4632	enough for at least into the year 4000. This requirement is fulfilled by
		4633	implementations implementing IEEE 754, which is basically all existing
		4634	ones. With IEEE 754 doubles, you get microsecond accuracy until at least
		4635	2200.
		4636
		4637	=back
		4638
		4639	If you know of other additional requirements drop me a note.
		4640
		4641
		4642	=head1 ALGORITHMIC COMPLEXITIES
		4643
		4644	In this section the complexities of (many of) the algorithms used inside
		4645	libev will be documented. For complexity discussions about backends see
		4646	the documentation for C<ev_default_init>.
		4647
		4648	All of the following are about amortised time: If an array needs to be
		4649	extended, libev needs to realloc and move the whole array, but this
		4650	happens asymptotically rarer with higher number of elements, so O(1) might
		4651	mean that libev does a lengthy realloc operation in rare cases, but on
		4652	average it is much faster and asymptotically approaches constant time.
		4653
		4654	=over 4
		4655
		4656	=item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
		4657
		4658	This means that, when you have a watcher that triggers in one hour and
		4659	there are 100 watchers that would trigger before that, then inserting will
		4660	have to skip roughly seven (C<ld 100>) of these watchers.
		4661
		4662	=item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
		4663
		4664	That means that changing a timer costs less than removing/adding them,
		4665	as only the relative motion in the event queue has to be paid for.
		4666
		4667	=item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
		4668
		4669	These just add the watcher into an array or at the head of a list.
		4670
		4671	=item Stopping check/prepare/idle/fork/async watchers: O(1)
		4672
		4673	=item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
		4674
		4675	These watchers are stored in lists, so they need to be walked to find the
		4676	correct watcher to remove. The lists are usually short (you don't usually
		4677	have many watchers waiting for the same fd or signal: one is typical, two
		4678	is rare).
		4679
		4680	=item Finding the next timer in each loop iteration: O(1)
		4681
		4682	By virtue of using a binary or 4-heap, the next timer is always found at a
		4683	fixed position in the storage array.
		4684
		4685	=item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
		4686
		4687	A change means an I/O watcher gets started or stopped, which requires
		4688	libev to recalculate its status (and possibly tell the kernel, depending
		4689	on backend and whether C<ev_io_set> was used).
		4690
		4691	=item Activating one watcher (putting it into the pending state): O(1)
		4692
		4693	=item Priority handling: O(number_of_priorities)
		4694
		4695	Priorities are implemented by allocating some space for each
		4696	priority. When doing priority-based operations, libev usually has to
		4697	linearly search all the priorities, but starting/stopping and activating
		4698	watchers becomes O(1) with respect to priority handling.
		4699
		4700	=item Sending an ev_async: O(1)
		4701
		4702	=item Processing ev_async_send: O(number_of_async_watchers)
		4703
		4704	=item Processing signals: O(max_signal_number)
		4705
		4706	Sending involves a system call I<iff> there were no other C<ev_async_send>
		4707	calls in the current loop iteration. Checking for async and signal events
		4708	involves iterating over all running async watchers or all signal numbers.
		4709
		4710	=back
		4711
		4712
		4713	=head1 PORTING FROM LIBEV 3.X TO 4.X
		4714
		4715	The major version 4 introduced some minor incompatible changes to the API.
		4716
		4717	At the moment, the C<ev.h> header file tries to implement superficial
		4718	compatibility, so most programs should still compile. Those might be
		4719	removed in later versions of libev, so better update early than late.
		4720
		4721	=over 4
		4722
		4723	=item C<ev_loop_count> renamed to C<ev_iteration>
		4724
		4725	=item C<ev_loop_depth> renamed to C<ev_depth>
		4726
		4727	=item C<ev_loop_verify> renamed to C<ev_verify>
		4728
		4729	Most functions working on C<struct ev_loop> objects don't have an
		4730	C<ev_loop_> prefix, so it was removed. Note that C<ev_loop_fork> is
		4731	still called C<ev_loop_fork> because it would otherwise clash with the
		4732	C<ev_fork> typedef.
		4733
		4734	=item C<EV_TIMEOUT> renamed to C<EV_TIMER> in C<revents>
		4735
		4736	This is a simple rename - all other watcher types use their name
		4737	as revents flag, and now C<ev_timer> does, too.
		4738
		4739	Both C<EV_TIMER> and C<EV_TIMEOUT> symbols were present in 3.x versions
		4740	and continue to be present for the foreseeable future, so this is mostly a
		4741	documentation change.
		4742
		4743	=item C<EV_MINIMAL> mechanism replaced by C<EV_FEATURES>
		4744
		4745	The preprocessor symbol C<EV_MINIMAL> has been replaced by a different
		4746	mechanism, C<EV_FEATURES>. Programs using C<EV_MINIMAL> usually compile
		4747	and work, but the library code will of course be larger.
		4748
		4749	=back
		4750
		4751
		4752	=head1 GLOSSARY
		4753
		4754	=over 4
		4755
		4756	=item active
		4757
		4758	A watcher is active as long as it has been started (has been attached to
		4759	an event loop) but not yet stopped (disassociated from the event loop).
		4760
		4761	=item application
		4762
		4763	In this document, an application is whatever is using libev.
		4764
		4765	=item callback
		4766
		4767	The address of a function that is called when some event has been
		4768	detected. Callbacks are being passed the event loop, the watcher that
		4769	received the event, and the actual event bitset.
		4770
		4771	=item callback invocation
		4772
		4773	The act of calling the callback associated with a watcher.
		4774
		4775	=item event
		4776
		4777	A change of state of some external event, such as data now being available
		4778	for reading on a file descriptor, time having passed or simply not having
		4779	any other events happening anymore.
		4780
		4781	In libev, events are represented as single bits (such as C<EV_READ> or
		4782	C<EV_TIMER>).
		4783
		4784	=item event library
		4785
		4786	A software package implementing an event model and loop.
		4787
		4788	=item event loop
		4789
		4790	An entity that handles and processes external events and converts them
		4791	into callback invocations.
		4792
		4793	=item event model
		4794
		4795	The model used to describe how an event loop handles and processes
		4796	watchers and events.
		4797
		4798	=item pending
		4799
		4800	A watcher is pending as soon as the corresponding event has been detected,
		4801	and stops being pending as soon as the watcher will be invoked or its
		4802	pending status is explicitly cleared by the application.
		4803
		4804	A watcher can be pending, but not active. Stopping a watcher also clears
		4805	its pending status.
		4806
		4807	=item real time
		4808
		4809	The physical time that is observed. It is apparently strictly monotonic :)
		4810
		4811	=item wall-clock time
		4812
		4813	The time and date as shown on clocks. Unlike real time, it can actually
		4814	be wrong and jump forwards and backwards, e.g. when the you adjust your
		4815	clock.
		4816
		4817	=item watcher
		4818
		4819	A data structure that describes interest in certain events. Watchers need
		4820	to be started (attached to an event loop) before they can receive events.
		4821
		4822	=item watcher invocation
		4823
		4824	The act of calling the callback associated with a watcher.
		4825
		4826	=back
782		4827
783	=head1 AUTHOR	4828	=head1 AUTHOR
784		4829
785	Marc Lehmann <libev@schmorp.de>.	4830	Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael Magnusson.
786		4831

Diff Legend

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