ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libev/ev.pod

Comparing libev/ev.pod (file contents):
Revision 1.27 by root, Wed Nov 14 05:02:07 2007 UTC vs.
Revision 1.330 by sf-exg, Sun Oct 24 21:01:58 2010 UTC

…		…
2		2
3	libev - a high performance full-featured event loop written in C	3	libev - a high performance full-featured event loop written in C
4		4
5	=head1 SYNOPSIS	5	=head1 SYNOPSIS
6		6
7	#include <ev.h>	7	#include <ev.h>
8		8
9	=head1 DESCRIPTION	9	=head2 EXAMPLE PROGRAM
		10
		11	// a single header file is required
		12	#include <ev.h>
		13
		14	#include <stdio.h> // for puts
		15
		16	// every watcher type has its own typedef'd struct
		17	// with the name ev_TYPE
		18	ev_io stdin_watcher;
		19	ev_timer timeout_watcher;
		20
		21	// all watcher callbacks have a similar signature
		22	// this callback is called when data is readable on stdin
		23	static void
		24	stdin_cb (EV_P_ ev_io *w, int revents)
		25	{
		26	puts ("stdin ready");
		27	// for one-shot events, one must manually stop the watcher
		28	// with its corresponding stop function.
		29	ev_io_stop (EV_A_ w);
		30
		31	// this causes all nested ev_run's to stop iterating
		32	ev_break (EV_A_ EVBREAK_ALL);
		33	}
		34
		35	// another callback, this time for a time-out
		36	static void
		37	timeout_cb (EV_P_ ev_timer *w, int revents)
		38	{
		39	puts ("timeout");
		40	// this causes the innermost ev_run to stop iterating
		41	ev_break (EV_A_ EVBREAK_ONE);
		42	}
		43
		44	int
		45	main (void)
		46	{
		47	// use the default event loop unless you have special needs
		48	struct ev_loop *loop = EV_DEFAULT;
		49
		50	// initialise an io watcher, then start it
		51	// this one will watch for stdin to become readable
		52	ev_io_init (&stdin_watcher, stdin_cb, /*STDIN_FILENO*/ 0, EV_READ);
		53	ev_io_start (loop, &stdin_watcher);
		54
		55	// initialise a timer watcher, then start it
		56	// simple non-repeating 5.5 second timeout
		57	ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
		58	ev_timer_start (loop, &timeout_watcher);
		59
		60	// now wait for events to arrive
		61	ev_run (loop, 0);
		62
		63	// unloop was called, so exit
		64	return 0;
		65	}
		66
		67	=head1 ABOUT THIS DOCUMENT
		68
		69	This document documents the libev software package.
		70
		71	The newest version of this document is also available as an html-formatted
		72	web page you might find easier to navigate when reading it for the first
		73	time: L<http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod>.
		74
		75	While this document tries to be as complete as possible in documenting
		76	libev, its usage and the rationale behind its design, it is not a tutorial
		77	on event-based programming, nor will it introduce event-based programming
		78	with libev.
		79
		80	Familiarity with event based programming techniques in general is assumed
		81	throughout this document.
		82
		83	=head1 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 practice
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.
…		…
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. Please note that the	168	Returns the current time as libev would use it. Please note that the
62	C<ev_now> function is usually faster and also often returns the timestamp	169	C<ev_now> function is usually faster and also often returns the timestamp
63	you actually want to know.	170	you actually want to know. Also interesting is the combination of
		171	C<ev_update_now> and C<ev_now>.
		172
		173	=item ev_sleep (ev_tstamp interval)
		174
		175	Sleep for the given interval: The current thread will be blocked until
		176	either it is interrupted or the given time interval has passed. Basically
		177	this is a sub-second-resolution C<sleep ()>.
64		178
65	=item int ev_version_major ()	179	=item int ev_version_major ()
66		180
67	=item int ev_version_minor ()	181	=item int ev_version_minor ()
68		182
69	You can find out the major and minor version numbers of the library	183	You can find out the major and minor ABI version numbers of the library
70	you linked against by calling the functions C<ev_version_major> and	184	you linked against by calling the functions C<ev_version_major> and
71	C<ev_version_minor>. If you want, you can compare against the global	185	C<ev_version_minor>. If you want, you can compare against the global
72	symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the	186	symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the
73	version of the library your program was compiled against.	187	version of the library your program was compiled against.
74		188
		189	These version numbers refer to the ABI version of the library, not the
		190	release version.
		191
75	Usually, it's a good idea to terminate if the major versions mismatch,	192	Usually, it's a good idea to terminate if the major versions mismatch,
76	as this indicates an incompatible change. Minor versions are usually	193	as this indicates an incompatible change. Minor versions are usually
77	compatible to older versions, so a larger minor version alone is usually	194	compatible to older versions, so a larger minor version alone is usually
78	not a problem.	195	not a problem.
79		196
		197	Example: Make sure we haven't accidentally been linked against the wrong
		198	version (note, however, that this will not detect other ABI mismatches,
		199	such as LFS or reentrancy).
		200
		201	assert (("libev version mismatch",
		202	ev_version_major () == EV_VERSION_MAJOR
		203	&& ev_version_minor () >= EV_VERSION_MINOR));
		204
		205	=item unsigned int ev_supported_backends ()
		206
		207	Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
		208	value) compiled into this binary of libev (independent of their
		209	availability on the system you are running on). See C<ev_default_loop> for
		210	a description of the set values.
		211
		212	Example: make sure we have the epoll method, because yeah this is cool and
		213	a must have and can we have a torrent of it please!!!11
		214
		215	assert (("sorry, no epoll, no sex",
		216	ev_supported_backends () & EVBACKEND_EPOLL));
		217
		218	=item unsigned int ev_recommended_backends ()
		219
		220	Return the set of all backends compiled into this binary of libev and
		221	also recommended for this platform, meaning it will work for most file
		222	descriptor types. This set is often smaller than the one returned by
		223	C<ev_supported_backends>, as for example kqueue is broken on most BSDs
		224	and will not be auto-detected unless you explicitly request it (assuming
		225	you know what you are doing). This is the set of backends that libev will
		226	probe for if you specify no backends explicitly.
		227
		228	=item unsigned int ev_embeddable_backends ()
		229
		230	Returns the set of backends that are embeddable in other event loops. This
		231	value is platform-specific but can include backends not available on the
		232	current system. To find which embeddable backends might be supported on
		233	the current system, you would need to look at C<ev_embeddable_backends ()
		234	& ev_supported_backends ()>, likewise for recommended ones.
		235
		236	See the description of C<ev_embed> watchers for more info.
		237
80	=item ev_set_allocator (void *(*cb)(void *ptr, long size))	238	=item ev_set_allocator (void *(*cb)(void *ptr, long size)) [NOT REENTRANT]
81		239
82	Sets the allocation function to use (the prototype is similar to the	240	Sets the allocation function to use (the prototype is similar - the
83	realloc C function, the semantics are identical). It is used to allocate	241	semantics are identical to the C<realloc> C89/SuS/POSIX function). It is
84	and free memory (no surprises here). If it returns zero when memory	242	used to allocate and free memory (no surprises here). If it returns zero
85	needs to be allocated, the library might abort or take some potentially	243	when memory needs to be allocated (C<size != 0>), the library might abort
86	destructive action. The default is your system realloc function.	244	or take some potentially destructive action.
		245
		246	Since some systems (at least OpenBSD and Darwin) fail to implement
		247	correct C<realloc> semantics, libev will use a wrapper around the system
		248	C<realloc> and C<free> functions by default.
87		249
88	You could override this function in high-availability programs to, say,	250	You could override this function in high-availability programs to, say,
89	free some memory if it cannot allocate memory, to use a special allocator,	251	free some memory if it cannot allocate memory, to use a special allocator,
90	or even to sleep a while and retry until some memory is available.	252	or even to sleep a while and retry until some memory is available.
91		253
		254	Example: Replace the libev allocator with one that waits a bit and then
		255	retries (example requires a standards-compliant C<realloc>).
		256
		257	static void *
		258	persistent_realloc (void *ptr, size_t size)
		259	{
		260	for (;;)
		261	{
		262	void *newptr = realloc (ptr, size);
		263
		264	if (newptr)
		265	return newptr;
		266
		267	sleep (60);
		268	}
		269	}
		270
		271	...
		272	ev_set_allocator (persistent_realloc);
		273
92	=item ev_set_syserr_cb (void (*cb)(const char *msg));	274	=item ev_set_syserr_cb (void (*cb)(const char *msg)); [NOT REENTRANT]
93		275
94	Set the callback function to call on a retryable syscall error (such	276	Set the callback function to call on a retryable system call error (such
95	as failed select, poll, epoll_wait). The message is a printable string	277	as failed select, poll, epoll_wait). The message is a printable string
96	indicating the system call or subsystem causing the problem. If this	278	indicating the system call or subsystem causing the problem. If this
97	callback is set, then libev will expect it to remedy the sitution, no	279	callback is set, then libev will expect it to remedy the situation, no
98	matter what, when it returns. That is, libev will generally retry the	280	matter what, when it returns. That is, libev will generally retry the
99	requested operation, or, if the condition doesn't go away, do bad stuff	281	requested operation, or, if the condition doesn't go away, do bad stuff
100	(such as abort).	282	(such as abort).
101		283
		284	Example: This is basically the same thing that libev does internally, too.
		285
		286	static void
		287	fatal_error (const char *msg)
		288	{
		289	perror (msg);
		290	abort ();
		291	}
		292
		293	...
		294	ev_set_syserr_cb (fatal_error);
		295
102	=back	296	=back
103		297
104	=head1 FUNCTIONS CONTROLLING THE EVENT LOOP	298	=head1 FUNCTIONS CONTROLLING EVENT LOOPS
105		299
106	An event loop is described by a C<struct ev_loop *>. The library knows two	300	An event loop is described by a C<struct ev_loop *> (the C<struct> is
107	types of such loops, the I<default> loop, which supports signals and child	301	I<not> optional in this case unless libev 3 compatibility is disabled, as
108	events, and dynamically created loops which do not.	302	libev 3 had an C<ev_loop> function colliding with the struct name).
109		303
110	If you use threads, a common model is to run the default event loop	304	The library knows two types of such loops, the I<default> loop, which
111	in your main thread (or in a separate thread) and for each thread you	305	supports signals and child events, and dynamically created event loops
112	create, you also create another event loop. Libev itself does no locking	306	which do not.
113	whatsoever, so if you mix calls to the same event loop in different
114	threads, make sure you lock (this is usually a bad idea, though, even if
115	done correctly, because it's hideous and inefficient).
116		307
117	=over 4	308	=over 4
118		309
119	=item struct ev_loop *ev_default_loop (unsigned int flags)	310	=item struct ev_loop *ev_default_loop (unsigned int flags)
120		311
121	This will initialise the default event loop if it hasn't been initialised	312	This returns the "default" event loop object, which is what you should
122	yet and return it. If the default loop could not be initialised, returns	313	normally use when you just need "the event loop". Event loop objects and
123	false. If it already was initialised it simply returns it (and ignores the	314	the C<flags> parameter are described in more detail in the entry for
124	flags).	315	C<ev_loop_new>.
		316
		317	If the default loop is already initialised then this function simply
		318	returns it (and ignores the flags. If that is troubling you, check
		319	C<ev_backend ()> afterwards). Otherwise it will create it with the given
		320	flags, which should almost always be C<0>, unless the caller is also the
		321	one calling C<ev_run> or otherwise qualifies as "the main program".
125		322
126	If you don't know what event loop to use, use the one returned from this	323	If you don't know what event loop to use, use the one returned from this
127	function.	324	function (or via the C<EV_DEFAULT> macro).
		325
		326	Note that this function is I<not> thread-safe, so if you want to use it
		327	from multiple threads, you have to employ some kind of mutex (note also
		328	that this case is unlikely, as loops cannot be shared easily between
		329	threads anyway).
		330
		331	The default loop is the only loop that can handle C<ev_child> watchers,
		332	and to do this, it always registers a handler for C<SIGCHLD>. If this is
		333	a problem for your application you can either create a dynamic loop with
		334	C<ev_loop_new> which doesn't do that, or you can simply overwrite the
		335	C<SIGCHLD> signal handler I<after> calling C<ev_default_init>.
		336
		337	Example: This is the most typical usage.
		338
		339	if (!ev_default_loop (0))
		340	fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
		341
		342	Example: Restrict libev to the select and poll backends, and do not allow
		343	environment settings to be taken into account:
		344
		345	ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
		346
		347	=item struct ev_loop *ev_loop_new (unsigned int flags)
		348
		349	This will create and initialise a new event loop object. If the loop
		350	could not be initialised, returns false.
		351
		352	Note that this function I<is> thread-safe, and one common way to use
		353	libev with threads is indeed to create one loop per thread, and using the
		354	default loop in the "main" or "initial" thread.
128		355
129	The flags argument can be used to specify special behaviour or specific	356	The flags argument can be used to specify special behaviour or specific
130	backends to use, and is usually specified as 0 (or EVFLAG_AUTO).	357	backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
131		358
132	It supports the following flags:	359	The following flags are supported:
133		360
134	=over 4	361	=over 4
135		362
136	=item C<EVFLAG_AUTO>	363	=item C<EVFLAG_AUTO>
137		364
138	The default flags value. Use this if you have no clue (it's the right	365	The default flags value. Use this if you have no clue (it's the right
139	thing, believe me).	366	thing, believe me).
140		367
141	=item C<EVFLAG_NOENV>	368	=item C<EVFLAG_NOENV>
142		369
143	If this flag bit is ored into the flag value (or the program runs setuid	370	If this flag bit is or'ed into the flag value (or the program runs setuid
144	or setgid) then libev will I<not> look at the environment variable	371	or setgid) then libev will I<not> look at the environment variable
145	C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will	372	C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
146	override the flags completely if it is found in the environment. This is	373	override the flags completely if it is found in the environment. This is
147	useful to try out specific backends to test their performance, or to work	374	useful to try out specific backends to test their performance, or to work
148	around bugs.	375	around bugs.
149		376
		377	=item C<EVFLAG_FORKCHECK>
		378
		379	Instead of calling C<ev_loop_fork> manually after a fork, you can also
		380	make libev check for a fork in each iteration by enabling this flag.
		381
		382	This works by calling C<getpid ()> on every iteration of the loop,
		383	and thus this might slow down your event loop if you do a lot of loop
		384	iterations and little real work, but is usually not noticeable (on my
		385	GNU/Linux system for example, C<getpid> is actually a simple 5-insn sequence
		386	without a system call and thus I<very> fast, but my GNU/Linux system also has
		387	C<pthread_atfork> which is even faster).
		388
		389	The big advantage of this flag is that you can forget about fork (and
		390	forget about forgetting to tell libev about forking) when you use this
		391	flag.
		392
		393	This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
		394	environment variable.
		395
		396	=item C<EVFLAG_NOINOTIFY>
		397
		398	When this flag is specified, then libev will not attempt to use the
		399	I<inotify> API for it's C<ev_stat> watchers. Apart from debugging and
		400	testing, this flag can be useful to conserve inotify file descriptors, as
		401	otherwise each loop using C<ev_stat> watchers consumes one inotify handle.
		402
		403	=item C<EVFLAG_SIGNALFD>
		404
		405	When this flag is specified, then libev will attempt to use the
		406	I<signalfd> API for it's C<ev_signal> (and C<ev_child>) watchers. This API
		407	delivers signals synchronously, which makes it both faster and might make
		408	it possible to get the queued signal data. It can also simplify signal
		409	handling with threads, as long as you properly block signals in your
		410	threads that are not interested in handling them.
		411
		412	Signalfd will not be used by default as this changes your signal mask, and
		413	there are a lot of shoddy libraries and programs (glib's threadpool for
		414	example) that can't properly initialise their signal masks.
		415
150	=item C<EVMETHOD_SELECT> (portable select backend)	416	=item C<EVBACKEND_SELECT> (value 1, portable select backend)
151		417
		418	This is your standard select(2) backend. Not I<completely> standard, as
		419	libev tries to roll its own fd_set with no limits on the number of fds,
		420	but if that fails, expect a fairly low limit on the number of fds when
		421	using this backend. It doesn't scale too well (O(highest_fd)), but its
		422	usually the fastest backend for a low number of (low-numbered :) fds.
		423
		424	To get good performance out of this backend you need a high amount of
		425	parallelism (most of the file descriptors should be busy). If you are
		426	writing a server, you should C<accept ()> in a loop to accept as many
		427	connections as possible during one iteration. You might also want to have
		428	a look at C<ev_set_io_collect_interval ()> to increase the amount of
		429	readiness notifications you get per iteration.
		430
		431	This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
		432	C<writefds> set (and to work around Microsoft Windows bugs, also onto the
		433	C<exceptfds> set on that platform).
		434
152	=item C<EVMETHOD_POLL> (poll backend, available everywhere except on windows)	435	=item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
153		436
154	=item C<EVMETHOD_EPOLL> (linux only)	437	And this is your standard poll(2) backend. It's more complicated
		438	than select, but handles sparse fds better and has no artificial
		439	limit on the number of fds you can use (except it will slow down
		440	considerably with a lot of inactive fds). It scales similarly to select,
		441	i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
		442	performance tips.
155		443
156	=item C<EVMETHOD_KQUEUE> (some bsds only)	444	This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
		445	C<EV_WRITE> to C<POLLOUT | POLLERR | POLLHUP>.
157		446
158	=item C<EVMETHOD_DEVPOLL> (solaris 8 only)	447	=item C<EVBACKEND_EPOLL> (value 4, Linux)
159		448
160	=item C<EVMETHOD_PORT> (solaris 10 only)	449	Use the linux-specific epoll(7) interface (for both pre- and post-2.6.9
		450	kernels).
161		451
162	If one or more of these are ored into the flags value, then only these	452	For few fds, this backend is a bit little slower than poll and select,
163	backends will be tried (in the reverse order as given here). If one are	453	but it scales phenomenally better. While poll and select usually scale
164	specified, any backend will do.	454	like O(total_fds) where n is the total number of fds (or the highest fd),
		455	epoll scales either O(1) or O(active_fds).
		456
		457	The epoll mechanism deserves honorable mention as the most misdesigned
		458	of the more advanced event mechanisms: mere annoyances include silently
		459	dropping file descriptors, requiring a system call per change per file
		460	descriptor (and unnecessary guessing of parameters), problems with dup and
		461	so on. The biggest issue is fork races, however - if a program forks then
		462	I<both> parent and child process have to recreate the epoll set, which can
		463	take considerable time (one syscall per file descriptor) and is of course
		464	hard to detect.
		465
		466	Epoll is also notoriously buggy - embedding epoll fds I<should> work, but
		467	of course I<doesn't>, and epoll just loves to report events for totally
		468	I<different> file descriptors (even already closed ones, so one cannot
		469	even remove them from the set) than registered in the set (especially
		470	on SMP systems). Libev tries to counter these spurious notifications by
		471	employing an additional generation counter and comparing that against the
		472	events to filter out spurious ones, recreating the set when required. Last
		473	not least, it also refuses to work with some file descriptors which work
		474	perfectly fine with C<select> (files, many character devices...).
		475
		476	While stopping, setting and starting an I/O watcher in the same iteration
		477	will result in some caching, there is still a system call per such
		478	incident (because the same I<file descriptor> could point to a different
		479	I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
		480	file descriptors might not work very well if you register events for both
		481	file descriptors.
		482
		483	Best performance from this backend is achieved by not unregistering all
		484	watchers for a file descriptor until it has been closed, if possible,
		485	i.e. keep at least one watcher active per fd at all times. Stopping and
		486	starting a watcher (without re-setting it) also usually doesn't cause
		487	extra overhead. A fork can both result in spurious notifications as well
		488	as in libev having to destroy and recreate the epoll object, which can
		489	take considerable time and thus should be avoided.
		490
		491	All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
		492	faster than epoll for maybe up to a hundred file descriptors, depending on
		493	the usage. So sad.
		494
		495	While nominally embeddable in other event loops, this feature is broken in
		496	all kernel versions tested so far.
		497
		498	This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
		499	C<EVBACKEND_POLL>.
		500
		501	=item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
		502
		503	Kqueue deserves special mention, as at the time of this writing, it
		504	was broken on all BSDs except NetBSD (usually it doesn't work reliably
		505	with anything but sockets and pipes, except on Darwin, where of course
		506	it's completely useless). Unlike epoll, however, whose brokenness
		507	is by design, these kqueue bugs can (and eventually will) be fixed
		508	without API changes to existing programs. For this reason it's not being
		509	"auto-detected" unless you explicitly specify it in the flags (i.e. using
		510	C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
		511	system like NetBSD.
		512
		513	You still can embed kqueue into a normal poll or select backend and use it
		514	only for sockets (after having made sure that sockets work with kqueue on
		515	the target platform). See C<ev_embed> watchers for more info.
		516
		517	It scales in the same way as the epoll backend, but the interface to the
		518	kernel is more efficient (which says nothing about its actual speed, of
		519	course). While stopping, setting and starting an I/O watcher does never
		520	cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
		521	two event changes per incident. Support for C<fork ()> is very bad (but
		522	sane, unlike epoll) and it drops fds silently in similarly hard-to-detect
		523	cases
		524
		525	This backend usually performs well under most conditions.
		526
		527	While nominally embeddable in other event loops, this doesn't work
		528	everywhere, so you might need to test for this. And since it is broken
		529	almost everywhere, you should only use it when you have a lot of sockets
		530	(for which it usually works), by embedding it into another event loop
		531	(e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
		532	also broken on OS X)) and, did I mention it, using it only for sockets.
		533
		534	This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
		535	C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
		536	C<NOTE_EOF>.
		537
		538	=item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
		539
		540	This is not implemented yet (and might never be, unless you send me an
		541	implementation). According to reports, C</dev/poll> only supports sockets
		542	and is not embeddable, which would limit the usefulness of this backend
		543	immensely.
		544
		545	=item C<EVBACKEND_PORT> (value 32, Solaris 10)
		546
		547	This uses the Solaris 10 event port mechanism. As with everything on Solaris,
		548	it's really slow, but it still scales very well (O(active_fds)).
		549
		550	Please note that Solaris event ports can deliver a lot of spurious
		551	notifications, so you need to use non-blocking I/O or other means to avoid
		552	blocking when no data (or space) is available.
		553
		554	While this backend scales well, it requires one system call per active
		555	file descriptor per loop iteration. For small and medium numbers of file
		556	descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
		557	might perform better.
		558
		559	On the positive side, with the exception of the spurious readiness
		560	notifications, this backend actually performed fully to specification
		561	in all tests and is fully embeddable, which is a rare feat among the
		562	OS-specific backends (I vastly prefer correctness over speed hacks).
		563
		564	This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
		565	C<EVBACKEND_POLL>.
		566
		567	=item C<EVBACKEND_ALL>
		568
		569	Try all backends (even potentially broken ones that wouldn't be tried
		570	with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
		571	C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.
		572
		573	It is definitely not recommended to use this flag.
165		574
166	=back	575	=back
167		576
168	=item struct ev_loop *ev_loop_new (unsigned int flags)	577	If one or more of the backend flags are or'ed into the flags value,
		578	then only these backends will be tried (in the reverse order as listed
		579	here). If none are specified, all backends in C<ev_recommended_backends
		580	()> will be tried.
169		581
170	Similar to C<ev_default_loop>, but always creates a new event loop that is	582	Example: Try to create a event loop that uses epoll and nothing else.
171	always distinct from the default loop. Unlike the default loop, it cannot
172	handle signal and child watchers, and attempts to do so will be greeted by
173	undefined behaviour (or a failed assertion if assertions are enabled).
174		583
175	=item ev_default_destroy ()	584	struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
		585	if (!epoller)
		586	fatal ("no epoll found here, maybe it hides under your chair");
176		587
177	Destroys the default loop again (frees all memory and kernel state	588	Example: Use whatever libev has to offer, but make sure that kqueue is
178	etc.). This stops all registered event watchers (by not touching them in	589	used if available.
179	any way whatsoever, although you cannot rely on this :).	590
		591	struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_KQUEUE);
180		592
181	=item ev_loop_destroy (loop)	593	=item ev_loop_destroy (loop)
182		594
183	Like C<ev_default_destroy>, but destroys an event loop created by an	595	Destroys an event loop object (frees all memory and kernel state
184	earlier call to C<ev_loop_new>.	596	etc.). None of the active event watchers will be stopped in the normal
		597	sense, so e.g. C<ev_is_active> might still return true. It is your
		598	responsibility to either stop all watchers cleanly yourself I<before>
		599	calling this function, or cope with the fact afterwards (which is usually
		600	the easiest thing, you can just ignore the watchers and/or C<free ()> them
		601	for example).
185		602
186	=item ev_default_fork ()	603	Note that certain global state, such as signal state (and installed signal
		604	handlers), will not be freed by this function, and related watchers (such
		605	as signal and child watchers) would need to be stopped manually.
187		606
		607	This function is normally used on loop objects allocated by
		608	C<ev_loop_new>, but it can also be used on the default loop returned by
		609	C<ev_default_loop>, in which case it is not thread-safe.
		610
		611	Note that it is not advisable to call this function on the default loop
		612	except in the rare occasion where you really need to free it's resources.
		613	If you need dynamically allocated loops it is better to use C<ev_loop_new>
		614	and C<ev_loop_destroy>.
		615
		616	=item ev_loop_fork (loop)
		617
		618	This function sets a flag that causes subsequent C<ev_run> iterations to
188	This function reinitialises the kernel state for backends that have	619	reinitialise the kernel state for backends that have one. Despite the
189	one. Despite the name, you can call it anytime, but it makes most sense	620	name, you can call it anytime, but it makes most sense after forking, in
190	after forking, in either the parent or child process (or both, but that	621	the child process. You I<must> call it (or use C<EVFLAG_FORKCHECK>) in the
191	again makes little sense).	622	child before resuming or calling C<ev_run>.
192		623
193	You I<must> call this function after forking if and only if you want to	624	Again, you I<have> to call it on I<any> loop that you want to re-use after
194	use the event library in both processes. If you just fork+exec, you don't	625	a fork, I<even if you do not plan to use the loop in the parent>. This is
195	have to call it.	626	because some kernel interfaces *cough* I<kqueue> *cough* do funny things
		627	during fork.
		628
		629	On the other hand, you only need to call this function in the child
		630	process if and only if you want to use the event loop in the child. If
		631	you just fork+exec or create a new loop in the child, you don't have to
		632	call it at all (in fact, C<epoll> is so badly broken that it makes a
		633	difference, but libev will usually detect this case on its own and do a
		634	costly reset of the backend).
196		635
197	The function itself is quite fast and it's usually not a problem to call	636	The function itself is quite fast and it's usually not a problem to call
198	it just in case after a fork. To make this easy, the function will fit in	637	it just in case after a fork.
199	quite nicely into a call to C<pthread_atfork>:
200		638
		639	Example: Automate calling C<ev_loop_fork> on the default loop when
		640	using pthreads.
		641
		642	static void
		643	post_fork_child (void)
		644	{
		645	ev_loop_fork (EV_DEFAULT);
		646	}
		647
		648	...
201	pthread_atfork (0, 0, ev_default_fork);	649	pthread_atfork (0, 0, post_fork_child);
202		650
203	=item ev_loop_fork (loop)	651	=item int ev_is_default_loop (loop)
204		652
205	Like C<ev_default_fork>, but acts on an event loop created by	653	Returns true when the given loop is, in fact, the default loop, and false
206	C<ev_loop_new>. Yes, you have to call this on every allocated event loop	654	otherwise.
207	after fork, and how you do this is entirely your own problem.
208		655
		656	=item unsigned int ev_iteration (loop)
		657
		658	Returns the current iteration count for the event loop, which is identical
		659	to the number of times libev did poll for new events. It starts at C<0>
		660	and happily wraps around with enough iterations.
		661
		662	This value can sometimes be useful as a generation counter of sorts (it
		663	"ticks" the number of loop iterations), as it roughly corresponds with
		664	C<ev_prepare> and C<ev_check> calls - and is incremented between the
		665	prepare and check phases.
		666
209	=item unsigned int ev_method (loop)	667	=item unsigned int ev_depth (loop)
210		668
		669	Returns the number of times C<ev_run> was entered minus the number of
		670	times C<ev_run> was exited, in other words, the recursion depth.
		671
		672	Outside C<ev_run>, this number is zero. In a callback, this number is
		673	C<1>, unless C<ev_run> was invoked recursively (or from another thread),
		674	in which case it is higher.
		675
		676	Leaving C<ev_run> abnormally (setjmp/longjmp, cancelling the thread
		677	etc.), doesn't count as "exit" - consider this as a hint to avoid such
		678	ungentleman-like behaviour unless it's really convenient.
		679
		680	=item unsigned int ev_backend (loop)
		681
211	Returns one of the C<EVMETHOD_*> flags indicating the event backend in	682	Returns one of the C<EVBACKEND_*> flags indicating the event backend in
212	use.	683	use.
213		684
214	=item ev_tstamp ev_now (loop)	685	=item ev_tstamp ev_now (loop)
215		686
216	Returns the current "event loop time", which is the time the event loop	687	Returns the current "event loop time", which is the time the event loop
217	got events and started processing them. This timestamp does not change	688	received events and started processing them. This timestamp does not
218	as long as callbacks are being processed, and this is also the base time	689	change as long as callbacks are being processed, and this is also the base
219	used for relative timers. You can treat it as the timestamp of the event	690	time used for relative timers. You can treat it as the timestamp of the
220	occuring (or more correctly, the mainloop finding out about it).	691	event occurring (or more correctly, libev finding out about it).
221		692
		693	=item ev_now_update (loop)
		694
		695	Establishes the current time by querying the kernel, updating the time
		696	returned by C<ev_now ()> in the progress. This is a costly operation and
		697	is usually done automatically within C<ev_run ()>.
		698
		699	This function is rarely useful, but when some event callback runs for a
		700	very long time without entering the event loop, updating libev's idea of
		701	the current time is a good idea.
		702
		703	See also L<The special problem of time updates> in the C<ev_timer> section.
		704
		705	=item ev_suspend (loop)
		706
		707	=item ev_resume (loop)
		708
		709	These two functions suspend and resume an event loop, for use when the
		710	loop is not used for a while and timeouts should not be processed.
		711
		712	A typical use case would be an interactive program such as a game: When
		713	the user presses C<^Z> to suspend the game and resumes it an hour later it
		714	would be best to handle timeouts as if no time had actually passed while
		715	the program was suspended. This can be achieved by calling C<ev_suspend>
		716	in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
		717	C<ev_resume> directly afterwards to resume timer processing.
		718
		719	Effectively, all C<ev_timer> watchers will be delayed by the time spend
		720	between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
		721	will be rescheduled (that is, they will lose any events that would have
		722	occurred while suspended).
		723
		724	After calling C<ev_suspend> you B<must not> call I<any> function on the
		725	given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
		726	without a previous call to C<ev_suspend>.
		727
		728	Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
		729	event loop time (see C<ev_now_update>).
		730
222	=item ev_loop (loop, int flags)	731	=item ev_run (loop, int flags)
223		732
224	Finally, this is it, the event handler. This function usually is called	733	Finally, this is it, the event handler. This function usually is called
225	after you initialised all your watchers and you want to start handling	734	after you have initialised all your watchers and you want to start
226	events.	735	handling events. It will ask the operating system for any new events, call
		736	the watcher callbacks, an then repeat the whole process indefinitely: This
		737	is why event loops are called I<loops>.
227		738
228	If the flags argument is specified as 0, it will not return until either	739	If the flags argument is specified as C<0>, it will keep handling events
229	no event watchers are active anymore or C<ev_unloop> was called.	740	until either no event watchers are active anymore or C<ev_break> was
		741	called.
230		742
		743	Please note that an explicit C<ev_break> is usually better than
		744	relying on all watchers to be stopped when deciding when a program has
		745	finished (especially in interactive programs), but having a program
		746	that automatically loops as long as it has to and no longer by virtue
		747	of relying on its watchers stopping correctly, that is truly a thing of
		748	beauty.
		749
231	A flags value of C<EVLOOP_NONBLOCK> will look for new events, will handle	750	A flags value of C<EVRUN_NOWAIT> will look for new events, will handle
232	those events and any outstanding ones, but will not block your process in	751	those events and any already outstanding ones, but will not wait and
233	case there are no events and will return after one iteration of the loop.	752	block your process in case there are no events and will return after one
		753	iteration of the loop. This is sometimes useful to poll and handle new
		754	events while doing lengthy calculations, to keep the program responsive.
234		755
235	A flags value of C<EVLOOP_ONESHOT> will look for new events (waiting if	756	A flags value of C<EVRUN_ONCE> will look for new events (waiting if
236	neccessary) and will handle those and any outstanding ones. It will block	757	necessary) and will handle those and any already outstanding ones. It
237	your process until at least one new event arrives, and will return after	758	will block your process until at least one new event arrives (which could
		759	be an event internal to libev itself, so there is no guarantee that a
		760	user-registered callback will be called), and will return after one
238	one iteration of the loop.	761	iteration of the loop.
239		762
240	This flags value could be used to implement alternative looping	763	This is useful if you are waiting for some external event in conjunction
241	constructs, but the C<prepare> and C<check> watchers provide a better and	764	with something not expressible using other libev watchers (i.e. "roll your
242	more generic mechanism.	765	own C<ev_run>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
		766	usually a better approach for this kind of thing.
243		767
244	Here are the gory details of what ev_loop does:	768	Here are the gory details of what C<ev_run> does:
245		769
246	1. If there are no active watchers (reference count is zero), return.	770	- Increment loop depth.
		771	- Reset the ev_break status.
		772	- Before the first iteration, call any pending watchers.
		773	LOOP:
		774	- If EVFLAG_FORKCHECK was used, check for a fork.
		775	- If a fork was detected (by any means), queue and call all fork watchers.
247	2. Queue and immediately call all prepare watchers.	776	- Queue and call all prepare watchers.
		777	- If ev_break was called, goto FINISH.
248	3. If we have been forked, recreate the kernel state.	778	- If we have been forked, detach and recreate the kernel state
		779	as to not disturb the other process.
249	4. Update the kernel state with all outstanding changes.	780	- Update the kernel state with all outstanding changes.
250	5. Update the "event loop time".	781	- Update the "event loop time" (ev_now ()).
251	6. Calculate for how long to block.	782	- Calculate for how long to sleep or block, if at all
		783	(active idle watchers, EVRUN_NOWAIT or not having
		784	any active watchers at all will result in not sleeping).
		785	- Sleep if the I/O and timer collect interval say so.
		786	- Increment loop iteration counter.
252	7. Block the process, waiting for events.	787	- Block the process, waiting for any events.
		788	- Queue all outstanding I/O (fd) events.
253	8. Update the "event loop time" and do time jump handling.	789	- Update the "event loop time" (ev_now ()), and do time jump adjustments.
254	9. Queue all outstanding timers.	790	- Queue all expired timers.
255	10. Queue all outstanding periodics.	791	- Queue all expired periodics.
256	11. If no events are pending now, queue all idle watchers.	792	- Queue all idle watchers with priority higher than that of pending events.
257	12. Queue all check watchers.	793	- Queue all check watchers.
258	13. Call all queued watchers in reverse order (i.e. check watchers first).	794	- Call all queued watchers in reverse order (i.e. check watchers first).
259	14. If ev_unloop has been called or EVLOOP_ONESHOT or EVLOOP_NONBLOCK	795	Signals and child watchers are implemented as I/O watchers, and will
260	was used, return, otherwise continue with step #1.	796	be handled here by queueing them when their watcher gets executed.
		797	- If ev_break has been called, or EVRUN_ONCE or EVRUN_NOWAIT
		798	were used, or there are no active watchers, goto FINISH, otherwise
		799	continue with step LOOP.
		800	FINISH:
		801	- Reset the ev_break status iff it was EVBREAK_ONE.
		802	- Decrement the loop depth.
		803	- Return.
261		804
		805	Example: Queue some jobs and then loop until no events are outstanding
		806	anymore.
		807
		808	... queue jobs here, make sure they register event watchers as long
		809	... as they still have work to do (even an idle watcher will do..)
		810	ev_run (my_loop, 0);
		811	... jobs done or somebody called unloop. yeah!
		812
262	=item ev_unloop (loop, how)	813	=item ev_break (loop, how)
263		814
264	Can be used to make a call to C<ev_loop> return early (but only after it	815	Can be used to make a call to C<ev_run> return early (but only after it
265	has processed all outstanding events). The C<how> argument must be either	816	has processed all outstanding events). The C<how> argument must be either
266	C<EVUNLOOP_ONE>, which will make the innermost C<ev_loop> call return, or	817	C<EVBREAK_ONE>, which will make the innermost C<ev_run> call return, or
267	C<EVUNLOOP_ALL>, which will make all nested C<ev_loop> calls return.	818	C<EVBREAK_ALL>, which will make all nested C<ev_run> calls return.
		819
		820	This "unloop state" will be cleared when entering C<ev_run> again.
		821
		822	It is safe to call C<ev_break> from outside any C<ev_run> calls. ##TODO##
268		823
269	=item ev_ref (loop)	824	=item ev_ref (loop)
270		825
271	=item ev_unref (loop)	826	=item ev_unref (loop)
272		827
273	Ref/unref can be used to add or remove a reference count on the event	828	Ref/unref can be used to add or remove a reference count on the event
274	loop: Every watcher keeps one reference, and as long as the reference	829	loop: Every watcher keeps one reference, and as long as the reference
275	count is nonzero, C<ev_loop> will not return on its own. If you have	830	count is nonzero, C<ev_run> will not return on its own.
276	a watcher you never unregister that should not keep C<ev_loop> from	831
277	returning, ev_unref() after starting, and ev_ref() before stopping it. For	832	This is useful when you have a watcher that you never intend to
		833	unregister, but that nevertheless should not keep C<ev_run> from
		834	returning. In such a case, call C<ev_unref> after starting, and C<ev_ref>
		835	before stopping it.
		836
278	example, libev itself uses this for its internal signal pipe: It is not	837	As an example, libev itself uses this for its internal signal pipe: It
279	visible to the libev user and should not keep C<ev_loop> from exiting if	838	is not visible to the libev user and should not keep C<ev_run> from
280	no event watchers registered by it are active. It is also an excellent	839	exiting if no event watchers registered by it are active. It is also an
281	way to do this for generic recurring timers or from within third-party	840	excellent way to do this for generic recurring timers or from within
282	libraries. Just remember to I<unref after start> and I<ref before stop>.	841	third-party libraries. Just remember to I<unref after start> and I<ref
		842	before stop> (but only if the watcher wasn't active before, or was active
		843	before, respectively. Note also that libev might stop watchers itself
		844	(e.g. non-repeating timers) in which case you have to C<ev_ref>
		845	in the callback).
		846
		847	Example: Create a signal watcher, but keep it from keeping C<ev_run>
		848	running when nothing else is active.
		849
		850	ev_signal exitsig;
		851	ev_signal_init (&exitsig, sig_cb, SIGINT);
		852	ev_signal_start (loop, &exitsig);
		853	evf_unref (loop);
		854
		855	Example: For some weird reason, unregister the above signal handler again.
		856
		857	ev_ref (loop);
		858	ev_signal_stop (loop, &exitsig);
		859
		860	=item ev_set_io_collect_interval (loop, ev_tstamp interval)
		861
		862	=item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
		863
		864	These advanced functions influence the time that libev will spend waiting
		865	for events. Both time intervals are by default C<0>, meaning that libev
		866	will try to invoke timer/periodic callbacks and I/O callbacks with minimum
		867	latency.
		868
		869	Setting these to a higher value (the C<interval> I<must> be >= C<0>)
		870	allows libev to delay invocation of I/O and timer/periodic callbacks
		871	to increase efficiency of loop iterations (or to increase power-saving
		872	opportunities).
		873
		874	The idea is that sometimes your program runs just fast enough to handle
		875	one (or very few) event(s) per loop iteration. While this makes the
		876	program responsive, it also wastes a lot of CPU time to poll for new
		877	events, especially with backends like C<select ()> which have a high
		878	overhead for the actual polling but can deliver many events at once.
		879
		880	By setting a higher I<io collect interval> you allow libev to spend more
		881	time collecting I/O events, so you can handle more events per iteration,
		882	at the cost of increasing latency. Timeouts (both C<ev_periodic> and
		883	C<ev_timer>) will be not affected. Setting this to a non-null value will
		884	introduce an additional C<ev_sleep ()> call into most loop iterations. The
		885	sleep time ensures that libev will not poll for I/O events more often then
		886	once per this interval, on average.
		887
		888	Likewise, by setting a higher I<timeout collect interval> you allow libev
		889	to spend more time collecting timeouts, at the expense of increased
		890	latency/jitter/inexactness (the watcher callback will be called
		891	later). C<ev_io> watchers will not be affected. Setting this to a non-null
		892	value will not introduce any overhead in libev.
		893
		894	Many (busy) programs can usually benefit by setting the I/O collect
		895	interval to a value near C<0.1> or so, which is often enough for
		896	interactive servers (of course not for games), likewise for timeouts. It
		897	usually doesn't make much sense to set it to a lower value than C<0.01>,
		898	as this approaches the timing granularity of most systems. Note that if
		899	you do transactions with the outside world and you can't increase the
		900	parallelity, then this setting will limit your transaction rate (if you
		901	need to poll once per transaction and the I/O collect interval is 0.01,
		902	then you can't do more than 100 transactions per second).
		903
		904	Setting the I<timeout collect interval> can improve the opportunity for
		905	saving power, as the program will "bundle" timer callback invocations that
		906	are "near" in time together, by delaying some, thus reducing the number of
		907	times the process sleeps and wakes up again. Another useful technique to
		908	reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
		909	they fire on, say, one-second boundaries only.
		910
		911	Example: we only need 0.1s timeout granularity, and we wish not to poll
		912	more often than 100 times per second:
		913
		914	ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
		915	ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
		916
		917	=item ev_invoke_pending (loop)
		918
		919	This call will simply invoke all pending watchers while resetting their
		920	pending state. Normally, C<ev_run> does this automatically when required,
		921	but when overriding the invoke callback this call comes handy. This
		922	function can be invoked from a watcher - this can be useful for example
		923	when you want to do some lengthy calculation and want to pass further
		924	event handling to another thread (you still have to make sure only one
		925	thread executes within C<ev_invoke_pending> or C<ev_run> of course).
		926
		927	=item int ev_pending_count (loop)
		928
		929	Returns the number of pending watchers - zero indicates that no watchers
		930	are pending.
		931
		932	=item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
		933
		934	This overrides the invoke pending functionality of the loop: Instead of
		935	invoking all pending watchers when there are any, C<ev_run> will call
		936	this callback instead. This is useful, for example, when you want to
		937	invoke the actual watchers inside another context (another thread etc.).
		938
		939	If you want to reset the callback, use C<ev_invoke_pending> as new
		940	callback.
		941
		942	=item ev_set_loop_release_cb (loop, void (*release)(EV_P), void (*acquire)(EV_P))
		943
		944	Sometimes you want to share the same loop between multiple threads. This
		945	can be done relatively simply by putting mutex_lock/unlock calls around
		946	each call to a libev function.
		947
		948	However, C<ev_run> can run an indefinite time, so it is not feasible
		949	to wait for it to return. One way around this is to wake up the event
		950	loop via C<ev_break> and C<av_async_send>, another way is to set these
		951	I<release> and I<acquire> callbacks on the loop.
		952
		953	When set, then C<release> will be called just before the thread is
		954	suspended waiting for new events, and C<acquire> is called just
		955	afterwards.
		956
		957	Ideally, C<release> will just call your mutex_unlock function, and
		958	C<acquire> will just call the mutex_lock function again.
		959
		960	While event loop modifications are allowed between invocations of
		961	C<release> and C<acquire> (that's their only purpose after all), no
		962	modifications done will affect the event loop, i.e. adding watchers will
		963	have no effect on the set of file descriptors being watched, or the time
		964	waited. Use an C<ev_async> watcher to wake up C<ev_run> when you want it
		965	to take note of any changes you made.
		966
		967	In theory, threads executing C<ev_run> will be async-cancel safe between
		968	invocations of C<release> and C<acquire>.
		969
		970	See also the locking example in the C<THREADS> section later in this
		971	document.
		972
		973	=item ev_set_userdata (loop, void *data)
		974
		975	=item ev_userdata (loop)
		976
		977	Set and retrieve a single C<void *> associated with a loop. When
		978	C<ev_set_userdata> has never been called, then C<ev_userdata> returns
		979	C<0.>
		980
		981	These two functions can be used to associate arbitrary data with a loop,
		982	and are intended solely for the C<invoke_pending_cb>, C<release> and
		983	C<acquire> callbacks described above, but of course can be (ab-)used for
		984	any other purpose as well.
		985
		986	=item ev_verify (loop)
		987
		988	This function only does something when C<EV_VERIFY> support has been
		989	compiled in, which is the default for non-minimal builds. It tries to go
		990	through all internal structures and checks them for validity. If anything
		991	is found to be inconsistent, it will print an error message to standard
		992	error and call C<abort ()>.
		993
		994	This can be used to catch bugs inside libev itself: under normal
		995	circumstances, this function will never abort as of course libev keeps its
		996	data structures consistent.
283		997
284	=back	998	=back
285		999
		1000
286	=head1 ANATOMY OF A WATCHER	1001	=head1 ANATOMY OF A WATCHER
287		1002
		1003	In the following description, uppercase C<TYPE> in names stands for the
		1004	watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
		1005	watchers and C<ev_io_start> for I/O watchers.
		1006
288	A watcher is a structure that you create and register to record your	1007	A watcher is an opaque structure that you allocate and register to record
289	interest in some event. For instance, if you want to wait for STDIN to	1008	your interest in some event. To make a concrete example, imagine you want
290	become readable, you would create an C<ev_io> watcher for that:	1009	to wait for STDIN to become readable, you would create an C<ev_io> watcher
		1010	for that:
291		1011
292	static void my_cb (struct ev_loop *loop, struct ev_io *w, int revents)	1012	static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
293	{	1013	{
294	ev_io_stop (w);	1014	ev_io_stop (w);
295	ev_unloop (loop, EVUNLOOP_ALL);	1015	ev_break (loop, EVBREAK_ALL);
296	}	1016	}
297		1017
298	struct ev_loop *loop = ev_default_loop (0);	1018	struct ev_loop *loop = ev_default_loop (0);
		1019
299	struct ev_io stdin_watcher;	1020	ev_io stdin_watcher;
		1021
300	ev_init (&stdin_watcher, my_cb);	1022	ev_init (&stdin_watcher, my_cb);
301	ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);	1023	ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
302	ev_io_start (loop, &stdin_watcher);	1024	ev_io_start (loop, &stdin_watcher);
		1025
303	ev_loop (loop, 0);	1026	ev_run (loop, 0);
304		1027
305	As you can see, you are responsible for allocating the memory for your	1028	As you can see, you are responsible for allocating the memory for your
306	watcher structures (and it is usually a bad idea to do this on the stack,	1029	watcher structures (and it is I<usually> a bad idea to do this on the
307	although this can sometimes be quite valid).	1030	stack).
308		1031
		1032	Each watcher has an associated watcher structure (called C<struct ev_TYPE>
		1033	or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
		1034
309	Each watcher structure must be initialised by a call to C<ev_init	1035	Each watcher structure must be initialised by a call to C<ev_init (watcher
310	(watcher *, callback)>, which expects a callback to be provided. This	1036	*, callback)>, which expects a callback to be provided. This callback is
311	callback gets invoked each time the event occurs (or, in the case of io	1037	invoked each time the event occurs (or, in the case of I/O watchers, each
312	watchers, each time the event loop detects that the file descriptor given	1038	time the event loop detects that the file descriptor given is readable
313	is readable and/or writable).	1039	and/or writable).
314		1040
315	Each watcher type has its own C<< ev_<type>_set (watcher *, ...) >> macro	1041	Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
316	with arguments specific to this watcher type. There is also a macro	1042	macro to configure it, with arguments specific to the watcher type. There
317	to combine initialisation and setting in one call: C<< ev_<type>_init	1043	is also a macro to combine initialisation and setting in one call: C<<
318	(watcher *, callback, ...) >>.	1044	ev_TYPE_init (watcher *, callback, ...) >>.
319		1045
320	To make the watcher actually watch out for events, you have to start it	1046	To make the watcher actually watch out for events, you have to start it
321	with a watcher-specific start function (C<< ev_<type>_start (loop, watcher	1047	with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
322	*) >>), and you can stop watching for events at any time by calling the	1048	*) >>), and you can stop watching for events at any time by calling the
323	corresponding stop function (C<< ev_<type>_stop (loop, watcher *) >>.	1049	corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
324		1050
325	As long as your watcher is active (has been started but not stopped) you	1051	As long as your watcher is active (has been started but not stopped) you
326	must not touch the values stored in it. Most specifically you must never	1052	must not touch the values stored in it. Most specifically you must never
327	reinitialise it or call its set method.	1053	reinitialise it or call its C<ev_TYPE_set> macro.
328
329	You can check whether an event is active by calling the C<ev_is_active
330	(watcher *)> macro. To see whether an event is outstanding (but the
331	callback for it has not been called yet) you can use the C<ev_is_pending
332	(watcher *)> macro.
333		1054
334	Each and every callback receives the event loop pointer as first, the	1055	Each and every callback receives the event loop pointer as first, the
335	registered watcher structure as second, and a bitset of received events as	1056	registered watcher structure as second, and a bitset of received events as
336	third argument.	1057	third argument.
337		1058
…		…
346	=item C<EV_WRITE>	1067	=item C<EV_WRITE>
347		1068
348	The file descriptor in the C<ev_io> watcher has become readable and/or	1069	The file descriptor in the C<ev_io> watcher has become readable and/or
349	writable.	1070	writable.
350		1071
351	=item C<EV_TIMEOUT>	1072	=item C<EV_TIMER>
352		1073
353	The C<ev_timer> watcher has timed out.	1074	The C<ev_timer> watcher has timed out.
354		1075
355	=item C<EV_PERIODIC>	1076	=item C<EV_PERIODIC>
356		1077
…		…
362		1083
363	=item C<EV_CHILD>	1084	=item C<EV_CHILD>
364		1085
365	The pid specified in the C<ev_child> watcher has received a status change.	1086	The pid specified in the C<ev_child> watcher has received a status change.
366		1087
		1088	=item C<EV_STAT>
		1089
		1090	The path specified in the C<ev_stat> watcher changed its attributes somehow.
		1091
367	=item C<EV_IDLE>	1092	=item C<EV_IDLE>
368		1093
369	The C<ev_idle> watcher has determined that you have nothing better to do.	1094	The C<ev_idle> watcher has determined that you have nothing better to do.
370		1095
371	=item C<EV_PREPARE>	1096	=item C<EV_PREPARE>
372		1097
373	=item C<EV_CHECK>	1098	=item C<EV_CHECK>
374		1099
375	All C<ev_prepare> watchers are invoked just I<before> C<ev_loop> starts	1100	All C<ev_prepare> watchers are invoked just I<before> C<ev_run> starts
376	to gather new events, and all C<ev_check> watchers are invoked just after	1101	to gather new events, and all C<ev_check> watchers are invoked just after
377	C<ev_loop> has gathered them, but before it invokes any callbacks for any	1102	C<ev_run> has gathered them, but before it invokes any callbacks for any
378	received events. Callbacks of both watcher types can start and stop as	1103	received events. Callbacks of both watcher types can start and stop as
379	many watchers as they want, and all of them will be taken into account	1104	many watchers as they want, and all of them will be taken into account
380	(for example, a C<ev_prepare> watcher might start an idle watcher to keep	1105	(for example, a C<ev_prepare> watcher might start an idle watcher to keep
381	C<ev_loop> from blocking).	1106	C<ev_run> from blocking).
		1107
		1108	=item C<EV_EMBED>
		1109
		1110	The embedded event loop specified in the C<ev_embed> watcher needs attention.
		1111
		1112	=item C<EV_FORK>
		1113
		1114	The event loop has been resumed in the child process after fork (see
		1115	C<ev_fork>).
		1116
		1117	=item C<EV_CLEANUP>
		1118
		1119	The event loop is about to be destroyed (see C<ev_cleanup>).
		1120
		1121	=item C<EV_ASYNC>
		1122
		1123	The given async watcher has been asynchronously notified (see C<ev_async>).
		1124
		1125	=item C<EV_CUSTOM>
		1126
		1127	Not ever sent (or otherwise used) by libev itself, but can be freely used
		1128	by libev users to signal watchers (e.g. via C<ev_feed_event>).
382		1129
383	=item C<EV_ERROR>	1130	=item C<EV_ERROR>
384		1131
385	An unspecified error has occured, the watcher has been stopped. This might	1132	An unspecified error has occurred, the watcher has been stopped. This might
386	happen because the watcher could not be properly started because libev	1133	happen because the watcher could not be properly started because libev
387	ran out of memory, a file descriptor was found to be closed or any other	1134	ran out of memory, a file descriptor was found to be closed or any other
		1135	problem. Libev considers these application bugs.
		1136
388	problem. You best act on it by reporting the problem and somehow coping	1137	You best act on it by reporting the problem and somehow coping with the
389	with the watcher being stopped.	1138	watcher being stopped. Note that well-written programs should not receive
		1139	an error ever, so when your watcher receives it, this usually indicates a
		1140	bug in your program.
390		1141
391	Libev will usually signal a few "dummy" events together with an error,	1142	Libev will usually signal a few "dummy" events together with an error, for
392	for example it might indicate that a fd is readable or writable, and if	1143	example it might indicate that a fd is readable or writable, and if your
393	your callbacks is well-written it can just attempt the operation and cope	1144	callbacks is well-written it can just attempt the operation and cope with
394	with the error from read() or write(). This will not work in multithreaded	1145	the error from read() or write(). This will not work in multi-threaded
395	programs, though, so beware.	1146	programs, though, as the fd could already be closed and reused for another
		1147	thing, so beware.
396		1148
397	=back	1149	=back
398		1150
		1151	=head2 WATCHER STATES
		1152
		1153	There are various watcher states mentioned throughout this manual -
		1154	active, pending and so on. In this section these states and the rules to
		1155	transition between them will be described in more detail - and while these
		1156	rules might look complicated, they usually do "the right thing".
		1157
		1158	=over 4
		1159
		1160	=item initialiased
		1161
		1162	Before a watcher can be registered with the event looop it has to be
		1163	initialised. This can be done with a call to C<ev_TYPE_init>, or calls to
		1164	C<ev_init> followed by the watcher-specific C<ev_TYPE_set> function.
		1165
		1166	In this state it is simply some block of memory that is suitable for use
		1167	in an event loop. It can be moved around, freed, reused etc. at will.
		1168
		1169	=item started/running/active
		1170
		1171	Once a watcher has been started with a call to C<ev_TYPE_start> it becomes
		1172	property of the event loop, and is actively waiting for events. While in
		1173	this state it cannot be accessed (except in a few documented ways), moved,
		1174	freed or anything else - the only legal thing is to keep a pointer to it,
		1175	and call libev functions on it that are documented to work on active watchers.
		1176
		1177	=item pending
		1178
		1179	If a watcher is active and libev determines that an event it is interested
		1180	in has occurred (such as a timer expiring), it will become pending. It will
		1181	stay in this pending state until either it is stopped or its callback is
		1182	about to be invoked, so it is not normally pending inside the watcher
		1183	callback.
		1184
		1185	The watcher might or might not be active while it is pending (for example,
		1186	an expired non-repeating timer can be pending but no longer active). If it
		1187	is stopped, it can be freely accessed (e.g. by calling C<ev_TYPE_set>),
		1188	but it is still property of the event loop at this time, so cannot be
		1189	moved, freed or reused. And if it is active the rules described in the
		1190	previous item still apply.
		1191
		1192	It is also possible to feed an event on a watcher that is not active (e.g.
		1193	via C<ev_feed_event>), in which case it becomes pending without being
		1194	active.
		1195
		1196	=item stopped
		1197
		1198	A watcher can be stopped implicitly by libev (in which case it might still
		1199	be pending), or explicitly by calling its C<ev_TYPE_stop> function. The
		1200	latter will clear any pending state the watcher might be in, regardless
		1201	of whether it was active or not, so stopping a watcher explicitly before
		1202	freeing it is often a good idea.
		1203
		1204	While stopped (and not pending) the watcher is essentially in the
		1205	initialised state, that is it can be reused, moved, modified in any way
		1206	you wish.
		1207
		1208	=back
		1209
		1210	=head2 GENERIC WATCHER FUNCTIONS
		1211
		1212	=over 4
		1213
		1214	=item C<ev_init> (ev_TYPE *watcher, callback)
		1215
		1216	This macro initialises the generic portion of a watcher. The contents
		1217	of the watcher object can be arbitrary (so C<malloc> will do). Only
		1218	the generic parts of the watcher are initialised, you I<need> to call
		1219	the type-specific C<ev_TYPE_set> macro afterwards to initialise the
		1220	type-specific parts. For each type there is also a C<ev_TYPE_init> macro
		1221	which rolls both calls into one.
		1222
		1223	You can reinitialise a watcher at any time as long as it has been stopped
		1224	(or never started) and there are no pending events outstanding.
		1225
		1226	The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
		1227	int revents)>.
		1228
		1229	Example: Initialise an C<ev_io> watcher in two steps.
		1230
		1231	ev_io w;
		1232	ev_init (&w, my_cb);
		1233	ev_io_set (&w, STDIN_FILENO, EV_READ);
		1234
		1235	=item C<ev_TYPE_set> (ev_TYPE *watcher, [args])
		1236
		1237	This macro initialises the type-specific parts of a watcher. You need to
		1238	call C<ev_init> at least once before you call this macro, but you can
		1239	call C<ev_TYPE_set> any number of times. You must not, however, call this
		1240	macro on a watcher that is active (it can be pending, however, which is a
		1241	difference to the C<ev_init> macro).
		1242
		1243	Although some watcher types do not have type-specific arguments
		1244	(e.g. C<ev_prepare>) you still need to call its C<set> macro.
		1245
		1246	See C<ev_init>, above, for an example.
		1247
		1248	=item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
		1249
		1250	This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
		1251	calls into a single call. This is the most convenient method to initialise
		1252	a watcher. The same limitations apply, of course.
		1253
		1254	Example: Initialise and set an C<ev_io> watcher in one step.
		1255
		1256	ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
		1257
		1258	=item C<ev_TYPE_start> (loop, ev_TYPE *watcher)
		1259
		1260	Starts (activates) the given watcher. Only active watchers will receive
		1261	events. If the watcher is already active nothing will happen.
		1262
		1263	Example: Start the C<ev_io> watcher that is being abused as example in this
		1264	whole section.
		1265
		1266	ev_io_start (EV_DEFAULT_UC, &w);
		1267
		1268	=item C<ev_TYPE_stop> (loop, ev_TYPE *watcher)
		1269
		1270	Stops the given watcher if active, and clears the pending status (whether
		1271	the watcher was active or not).
		1272
		1273	It is possible that stopped watchers are pending - for example,
		1274	non-repeating timers are being stopped when they become pending - but
		1275	calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
		1276	pending. If you want to free or reuse the memory used by the watcher it is
		1277	therefore a good idea to always call its C<ev_TYPE_stop> function.
		1278
		1279	=item bool ev_is_active (ev_TYPE *watcher)
		1280
		1281	Returns a true value iff the watcher is active (i.e. it has been started
		1282	and not yet been stopped). As long as a watcher is active you must not modify
		1283	it.
		1284
		1285	=item bool ev_is_pending (ev_TYPE *watcher)
		1286
		1287	Returns a true value iff the watcher is pending, (i.e. it has outstanding
		1288	events but its callback has not yet been invoked). As long as a watcher
		1289	is pending (but not active) you must not call an init function on it (but
		1290	C<ev_TYPE_set> is safe), you must not change its priority, and you must
		1291	make sure the watcher is available to libev (e.g. you cannot C<free ()>
		1292	it).
		1293
		1294	=item callback ev_cb (ev_TYPE *watcher)
		1295
		1296	Returns the callback currently set on the watcher.
		1297
		1298	=item ev_cb_set (ev_TYPE *watcher, callback)
		1299
		1300	Change the callback. You can change the callback at virtually any time
		1301	(modulo threads).
		1302
		1303	=item ev_set_priority (ev_TYPE *watcher, int priority)
		1304
		1305	=item int ev_priority (ev_TYPE *watcher)
		1306
		1307	Set and query the priority of the watcher. The priority is a small
		1308	integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
		1309	(default: C<-2>). Pending watchers with higher priority will be invoked
		1310	before watchers with lower priority, but priority will not keep watchers
		1311	from being executed (except for C<ev_idle> watchers).
		1312
		1313	If you need to suppress invocation when higher priority events are pending
		1314	you need to look at C<ev_idle> watchers, which provide this functionality.
		1315
		1316	You I<must not> change the priority of a watcher as long as it is active or
		1317	pending.
		1318
		1319	Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
		1320	fine, as long as you do not mind that the priority value you query might
		1321	or might not have been clamped to the valid range.
		1322
		1323	The default priority used by watchers when no priority has been set is
		1324	always C<0>, which is supposed to not be too high and not be too low :).
		1325
		1326	See L<WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
		1327	priorities.
		1328
		1329	=item ev_invoke (loop, ev_TYPE *watcher, int revents)
		1330
		1331	Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
		1332	C<loop> nor C<revents> need to be valid as long as the watcher callback
		1333	can deal with that fact, as both are simply passed through to the
		1334	callback.
		1335
		1336	=item int ev_clear_pending (loop, ev_TYPE *watcher)
		1337
		1338	If the watcher is pending, this function clears its pending status and
		1339	returns its C<revents> bitset (as if its callback was invoked). If the
		1340	watcher isn't pending it does nothing and returns C<0>.
		1341
		1342	Sometimes it can be useful to "poll" a watcher instead of waiting for its
		1343	callback to be invoked, which can be accomplished with this function.
		1344
		1345	=item ev_feed_event (loop, ev_TYPE *watcher, int revents)
		1346
		1347	Feeds the given event set into the event loop, as if the specified event
		1348	had happened for the specified watcher (which must be a pointer to an
		1349	initialised but not necessarily started event watcher). Obviously you must
		1350	not free the watcher as long as it has pending events.
		1351
		1352	Stopping the watcher, letting libev invoke it, or calling
		1353	C<ev_clear_pending> will clear the pending event, even if the watcher was
		1354	not started in the first place.
		1355
		1356	See also C<ev_feed_fd_event> and C<ev_feed_signal_event> for related
		1357	functions that do not need a watcher.
		1358
		1359	=back
		1360
		1361
399	=head2 ASSOCIATING CUSTOM DATA WITH A WATCHER	1362	=head2 ASSOCIATING CUSTOM DATA WITH A WATCHER
400		1363
401	Each watcher has, by default, a member C<void *data> that you can change	1364	Each watcher has, by default, a member C<void *data> that you can change
402	and read at any time, libev will completely ignore it. This can be used	1365	and read at any time: libev will completely ignore it. This can be used
403	to associate arbitrary data with your watcher. If you need more data and	1366	to associate arbitrary data with your watcher. If you need more data and
404	don't want to allocate memory and store a pointer to it in that data	1367	don't want to allocate memory and store a pointer to it in that data
405	member, you can also "subclass" the watcher type and provide your own	1368	member, you can also "subclass" the watcher type and provide your own
406	data:	1369	data:
407		1370
408	struct my_io	1371	struct my_io
409	{	1372	{
410	struct ev_io io;	1373	ev_io io;
411	int otherfd;	1374	int otherfd;
412	void *somedata;	1375	void *somedata;
413	struct whatever *mostinteresting;	1376	struct whatever *mostinteresting;
414	}	1377	};
		1378
		1379	...
		1380	struct my_io w;
		1381	ev_io_init (&w.io, my_cb, fd, EV_READ);
415		1382
416	And since your callback will be called with a pointer to the watcher, you	1383	And since your callback will be called with a pointer to the watcher, you
417	can cast it back to your own type:	1384	can cast it back to your own type:
418		1385
419	static void my_cb (struct ev_loop *loop, struct ev_io *w_, int revents)	1386	static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
420	{	1387	{
421	struct my_io *w = (struct my_io *)w_;	1388	struct my_io *w = (struct my_io *)w_;
422	...	1389	...
423	}	1390	}
424		1391
425	More interesting and less C-conformant ways of catsing your callback type	1392	More interesting and less C-conformant ways of casting your callback type
426	have been omitted....	1393	instead have been omitted.
		1394
		1395	Another common scenario is to use some data structure with multiple
		1396	embedded watchers:
		1397
		1398	struct my_biggy
		1399	{
		1400	int some_data;
		1401	ev_timer t1;
		1402	ev_timer t2;
		1403	}
		1404
		1405	In this case getting the pointer to C<my_biggy> is a bit more
		1406	complicated: Either you store the address of your C<my_biggy> struct
		1407	in the C<data> member of the watcher (for woozies), or you need to use
		1408	some pointer arithmetic using C<offsetof> inside your watchers (for real
		1409	programmers):
		1410
		1411	#include <stddef.h>
		1412
		1413	static void
		1414	t1_cb (EV_P_ ev_timer *w, int revents)
		1415	{
		1416	struct my_biggy big = (struct my_biggy *)
		1417	(((char *)w) - offsetof (struct my_biggy, t1));
		1418	}
		1419
		1420	static void
		1421	t2_cb (EV_P_ ev_timer *w, int revents)
		1422	{
		1423	struct my_biggy big = (struct my_biggy *)
		1424	(((char *)w) - offsetof (struct my_biggy, t2));
		1425	}
		1426
		1427	=head2 WATCHER PRIORITY MODELS
		1428
		1429	Many event loops support I<watcher priorities>, which are usually small
		1430	integers that influence the ordering of event callback invocation
		1431	between watchers in some way, all else being equal.
		1432
		1433	In libev, Watcher priorities can be set using C<ev_set_priority>. See its
		1434	description for the more technical details such as the actual priority
		1435	range.
		1436
		1437	There are two common ways how these these priorities are being interpreted
		1438	by event loops:
		1439
		1440	In the more common lock-out model, higher priorities "lock out" invocation
		1441	of lower priority watchers, which means as long as higher priority
		1442	watchers receive events, lower priority watchers are not being invoked.
		1443
		1444	The less common only-for-ordering model uses priorities solely to order
		1445	callback invocation within a single event loop iteration: Higher priority
		1446	watchers are invoked before lower priority ones, but they all get invoked
		1447	before polling for new events.
		1448
		1449	Libev uses the second (only-for-ordering) model for all its watchers
		1450	except for idle watchers (which use the lock-out model).
		1451
		1452	The rationale behind this is that implementing the lock-out model for
		1453	watchers is not well supported by most kernel interfaces, and most event
		1454	libraries will just poll for the same events again and again as long as
		1455	their callbacks have not been executed, which is very inefficient in the
		1456	common case of one high-priority watcher locking out a mass of lower
		1457	priority ones.
		1458
		1459	Static (ordering) priorities are most useful when you have two or more
		1460	watchers handling the same resource: a typical usage example is having an
		1461	C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
		1462	timeouts. Under load, data might be received while the program handles
		1463	other jobs, but since timers normally get invoked first, the timeout
		1464	handler will be executed before checking for data. In that case, giving
		1465	the timer a lower priority than the I/O watcher ensures that I/O will be
		1466	handled first even under adverse conditions (which is usually, but not
		1467	always, what you want).
		1468
		1469	Since idle watchers use the "lock-out" model, meaning that idle watchers
		1470	will only be executed when no same or higher priority watchers have
		1471	received events, they can be used to implement the "lock-out" model when
		1472	required.
		1473
		1474	For example, to emulate how many other event libraries handle priorities,
		1475	you can associate an C<ev_idle> watcher to each such watcher, and in
		1476	the normal watcher callback, you just start the idle watcher. The real
		1477	processing is done in the idle watcher callback. This causes libev to
		1478	continuously poll and process kernel event data for the watcher, but when
		1479	the lock-out case is known to be rare (which in turn is rare :), this is
		1480	workable.
		1481
		1482	Usually, however, the lock-out model implemented that way will perform
		1483	miserably under the type of load it was designed to handle. In that case,
		1484	it might be preferable to stop the real watcher before starting the
		1485	idle watcher, so the kernel will not have to process the event in case
		1486	the actual processing will be delayed for considerable time.
		1487
		1488	Here is an example of an I/O watcher that should run at a strictly lower
		1489	priority than the default, and which should only process data when no
		1490	other events are pending:
		1491
		1492	ev_idle idle; // actual processing watcher
		1493	ev_io io; // actual event watcher
		1494
		1495	static void
		1496	io_cb (EV_P_ ev_io *w, int revents)
		1497	{
		1498	// stop the I/O watcher, we received the event, but
		1499	// are not yet ready to handle it.
		1500	ev_io_stop (EV_A_ w);
		1501
		1502	// start the idle watcher to handle the actual event.
		1503	// it will not be executed as long as other watchers
		1504	// with the default priority are receiving events.
		1505	ev_idle_start (EV_A_ &idle);
		1506	}
		1507
		1508	static void
		1509	idle_cb (EV_P_ ev_idle *w, int revents)
		1510	{
		1511	// actual processing
		1512	read (STDIN_FILENO, ...);
		1513
		1514	// have to start the I/O watcher again, as
		1515	// we have handled the event
		1516	ev_io_start (EV_P_ &io);
		1517	}
		1518
		1519	// initialisation
		1520	ev_idle_init (&idle, idle_cb);
		1521	ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
		1522	ev_io_start (EV_DEFAULT_ &io);
		1523
		1524	In the "real" world, it might also be beneficial to start a timer, so that
		1525	low-priority connections can not be locked out forever under load. This
		1526	enables your program to keep a lower latency for important connections
		1527	during short periods of high load, while not completely locking out less
		1528	important ones.
427		1529
428		1530
429	=head1 WATCHER TYPES	1531	=head1 WATCHER TYPES
430		1532
431	This section describes each watcher in detail, but will not repeat	1533	This section describes each watcher in detail, but will not repeat
432	information given in the last section.	1534	information given in the last section. Any initialisation/set macros,
		1535	functions and members specific to the watcher type are explained.
433		1536
		1537	Members are additionally marked with either I<[read-only]>, meaning that,
		1538	while the watcher is active, you can look at the member and expect some
		1539	sensible content, but you must not modify it (you can modify it while the
		1540	watcher is stopped to your hearts content), or I<[read-write]>, which
		1541	means you can expect it to have some sensible content while the watcher
		1542	is active, but you can also modify it. Modifying it may not do something
		1543	sensible or take immediate effect (or do anything at all), but libev will
		1544	not crash or malfunction in any way.
		1545
		1546
434	=head2 C<ev_io> - is this file descriptor readable or writable	1547	=head2 C<ev_io> - is this file descriptor readable or writable?
435		1548
436	I/O watchers check whether a file descriptor is readable or writable	1549	I/O watchers check whether a file descriptor is readable or writable
437	in each iteration of the event loop (This behaviour is called	1550	in each iteration of the event loop, or, more precisely, when reading
438	level-triggering because you keep receiving events as long as the	1551	would not block the process and writing would at least be able to write
439	condition persists. Remember you can stop the watcher if you don't want to	1552	some data. This behaviour is called level-triggering because you keep
440	act on the event and neither want to receive future events).	1553	receiving events as long as the condition persists. Remember you can stop
		1554	the watcher if you don't want to act on the event and neither want to
		1555	receive future events.
441		1556
442	In general you can register as many read and/or write event watchers per	1557	In general you can register as many read and/or write event watchers per
443	fd as you want (as long as you don't confuse yourself). Setting all file	1558	fd as you want (as long as you don't confuse yourself). Setting all file
444	descriptors to non-blocking mode is also usually a good idea (but not	1559	descriptors to non-blocking mode is also usually a good idea (but not
445	required if you know what you are doing).	1560	required if you know what you are doing).
446		1561
447	You have to be careful with dup'ed file descriptors, though. Some backends	1562	If you cannot use non-blocking mode, then force the use of a
448	(the linux epoll backend is a notable example) cannot handle dup'ed file	1563	known-to-be-good backend (at the time of this writing, this includes only
449	descriptors correctly if you register interest in two or more fds pointing	1564	C<EVBACKEND_SELECT> and C<EVBACKEND_POLL>). The same applies to file
450	to the same underlying file/socket etc. description (that is, they share	1565	descriptors for which non-blocking operation makes no sense (such as
451	the same underlying "file open").	1566	files) - libev doesn't guarantee any specific behaviour in that case.
452		1567
453	If you must do this, then force the use of a known-to-be-good backend	1568	Another thing you have to watch out for is that it is quite easy to
454	(at the time of this writing, this includes only EVMETHOD_SELECT and	1569	receive "spurious" readiness notifications, that is your callback might
455	EVMETHOD_POLL).	1570	be called with C<EV_READ> but a subsequent C<read>(2) will actually block
		1571	because there is no data. Not only are some backends known to create a
		1572	lot of those (for example Solaris ports), it is very easy to get into
		1573	this situation even with a relatively standard program structure. Thus
		1574	it is best to always use non-blocking I/O: An extra C<read>(2) returning
		1575	C<EAGAIN> is far preferable to a program hanging until some data arrives.
		1576
		1577	If you cannot run the fd in non-blocking mode (for example you should
		1578	not play around with an Xlib connection), then you have to separately
		1579	re-test whether a file descriptor is really ready with a known-to-be good
		1580	interface such as poll (fortunately in our Xlib example, Xlib already
		1581	does this on its own, so its quite safe to use). Some people additionally
		1582	use C<SIGALRM> and an interval timer, just to be sure you won't block
		1583	indefinitely.
		1584
		1585	But really, best use non-blocking mode.
		1586
		1587	=head3 The special problem of disappearing file descriptors
		1588
		1589	Some backends (e.g. kqueue, epoll) need to be told about closing a file
		1590	descriptor (either due to calling C<close> explicitly or any other means,
		1591	such as C<dup2>). The reason is that you register interest in some file
		1592	descriptor, but when it goes away, the operating system will silently drop
		1593	this interest. If another file descriptor with the same number then is
		1594	registered with libev, there is no efficient way to see that this is, in
		1595	fact, a different file descriptor.
		1596
		1597	To avoid having to explicitly tell libev about such cases, libev follows
		1598	the following policy: Each time C<ev_io_set> is being called, libev
		1599	will assume that this is potentially a new file descriptor, otherwise
		1600	it is assumed that the file descriptor stays the same. That means that
		1601	you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
		1602	descriptor even if the file descriptor number itself did not change.
		1603
		1604	This is how one would do it normally anyway, the important point is that
		1605	the libev application should not optimise around libev but should leave
		1606	optimisations to libev.
		1607
		1608	=head3 The special problem of dup'ed file descriptors
		1609
		1610	Some backends (e.g. epoll), cannot register events for file descriptors,
		1611	but only events for the underlying file descriptions. That means when you
		1612	have C<dup ()>'ed file descriptors or weirder constellations, and register
		1613	events for them, only one file descriptor might actually receive events.
		1614
		1615	There is no workaround possible except not registering events
		1616	for potentially C<dup ()>'ed file descriptors, or to resort to
		1617	C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
		1618
		1619	=head3 The special problem of fork
		1620
		1621	Some backends (epoll, kqueue) do not support C<fork ()> at all or exhibit
		1622	useless behaviour. Libev fully supports fork, but needs to be told about
		1623	it in the child.
		1624
		1625	To support fork in your programs, you either have to call
		1626	C<ev_default_fork ()> or C<ev_loop_fork ()> after a fork in the child,
		1627	enable C<EVFLAG_FORKCHECK>, or resort to C<EVBACKEND_SELECT> or
		1628	C<EVBACKEND_POLL>.
		1629
		1630	=head3 The special problem of SIGPIPE
		1631
		1632	While not really specific to libev, it is easy to forget about C<SIGPIPE>:
		1633	when writing to a pipe whose other end has been closed, your program gets
		1634	sent a SIGPIPE, which, by default, aborts your program. For most programs
		1635	this is sensible behaviour, for daemons, this is usually undesirable.
		1636
		1637	So when you encounter spurious, unexplained daemon exits, make sure you
		1638	ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
		1639	somewhere, as that would have given you a big clue).
		1640
		1641	=head3 The special problem of accept()ing when you can't
		1642
		1643	Many implementations of the POSIX C<accept> function (for example,
		1644	found in post-2004 Linux) have the peculiar behaviour of not removing a
		1645	connection from the pending queue in all error cases.
		1646
		1647	For example, larger servers often run out of file descriptors (because
		1648	of resource limits), causing C<accept> to fail with C<ENFILE> but not
		1649	rejecting the connection, leading to libev signalling readiness on
		1650	the next iteration again (the connection still exists after all), and
		1651	typically causing the program to loop at 100% CPU usage.
		1652
		1653	Unfortunately, the set of errors that cause this issue differs between
		1654	operating systems, there is usually little the app can do to remedy the
		1655	situation, and no known thread-safe method of removing the connection to
		1656	cope with overload is known (to me).
		1657
		1658	One of the easiest ways to handle this situation is to just ignore it
		1659	- when the program encounters an overload, it will just loop until the
		1660	situation is over. While this is a form of busy waiting, no OS offers an
		1661	event-based way to handle this situation, so it's the best one can do.
		1662
		1663	A better way to handle the situation is to log any errors other than
		1664	C<EAGAIN> and C<EWOULDBLOCK>, making sure not to flood the log with such
		1665	messages, and continue as usual, which at least gives the user an idea of
		1666	what could be wrong ("raise the ulimit!"). For extra points one could stop
		1667	the C<ev_io> watcher on the listening fd "for a while", which reduces CPU
		1668	usage.
		1669
		1670	If your program is single-threaded, then you could also keep a dummy file
		1671	descriptor for overload situations (e.g. by opening F</dev/null>), and
		1672	when you run into C<ENFILE> or C<EMFILE>, close it, run C<accept>,
		1673	close that fd, and create a new dummy fd. This will gracefully refuse
		1674	clients under typical overload conditions.
		1675
		1676	The last way to handle it is to simply log the error and C<exit>, as
		1677	is often done with C<malloc> failures, but this results in an easy
		1678	opportunity for a DoS attack.
		1679
		1680	=head3 Watcher-Specific Functions
456		1681
457	=over 4	1682	=over 4
458		1683
459	=item ev_io_init (ev_io *, callback, int fd, int events)	1684	=item ev_io_init (ev_io *, callback, int fd, int events)
460		1685
461	=item ev_io_set (ev_io *, int fd, int events)	1686	=item ev_io_set (ev_io *, int fd, int events)
462		1687
463	Configures an C<ev_io> watcher. The fd is the file descriptor to rceeive	1688	Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
464	events for and events is either C<EV_READ>, C<EV_WRITE> or C<EV_READ |	1689	receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
465	EV_WRITE> to receive the given events.	1690	C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
		1691
		1692	=item int fd [read-only]
		1693
		1694	The file descriptor being watched.
		1695
		1696	=item int events [read-only]
		1697
		1698	The events being watched.
466		1699
467	=back	1700	=back
468		1701
		1702	=head3 Examples
		1703
		1704	Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
		1705	readable, but only once. Since it is likely line-buffered, you could
		1706	attempt to read a whole line in the callback.
		1707
		1708	static void
		1709	stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
		1710	{
		1711	ev_io_stop (loop, w);
		1712	.. read from stdin here (or from w->fd) and handle any I/O errors
		1713	}
		1714
		1715	...
		1716	struct ev_loop *loop = ev_default_init (0);
		1717	ev_io stdin_readable;
		1718	ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
		1719	ev_io_start (loop, &stdin_readable);
		1720	ev_run (loop, 0);
		1721
		1722
469	=head2 C<ev_timer> - relative and optionally recurring timeouts	1723	=head2 C<ev_timer> - relative and optionally repeating timeouts
470		1724
471	Timer watchers are simple relative timers that generate an event after a	1725	Timer watchers are simple relative timers that generate an event after a
472	given time, and optionally repeating in regular intervals after that.	1726	given time, and optionally repeating in regular intervals after that.
473		1727
474	The timers are based on real time, that is, if you register an event that	1728	The timers are based on real time, that is, if you register an event that
475	times out after an hour and you reset your system clock to last years	1729	times out after an hour and you reset your system clock to January last
476	time, it will still time out after (roughly) and hour. "Roughly" because	1730	year, it will still time out after (roughly) one hour. "Roughly" because
477	detecting time jumps is hard, and soem inaccuracies are unavoidable (the	1731	detecting time jumps is hard, and some inaccuracies are unavoidable (the
478	monotonic clock option helps a lot here).	1732	monotonic clock option helps a lot here).
		1733
		1734	The callback is guaranteed to be invoked only I<after> its timeout has
		1735	passed (not I<at>, so on systems with very low-resolution clocks this
		1736	might introduce a small delay). If multiple timers become ready during the
		1737	same loop iteration then the ones with earlier time-out values are invoked
		1738	before ones of the same priority with later time-out values (but this is
		1739	no longer true when a callback calls C<ev_run> recursively).
		1740
		1741	=head3 Be smart about timeouts
		1742
		1743	Many real-world problems involve some kind of timeout, usually for error
		1744	recovery. A typical example is an HTTP request - if the other side hangs,
		1745	you want to raise some error after a while.
		1746
		1747	What follows are some ways to handle this problem, from obvious and
		1748	inefficient to smart and efficient.
		1749
		1750	In the following, a 60 second activity timeout is assumed - a timeout that
		1751	gets reset to 60 seconds each time there is activity (e.g. each time some
		1752	data or other life sign was received).
		1753
		1754	=over 4
		1755
		1756	=item 1. Use a timer and stop, reinitialise and start it on activity.
		1757
		1758	This is the most obvious, but not the most simple way: In the beginning,
		1759	start the watcher:
		1760
		1761	ev_timer_init (timer, callback, 60., 0.);
		1762	ev_timer_start (loop, timer);
		1763
		1764	Then, each time there is some activity, C<ev_timer_stop> it, initialise it
		1765	and start it again:
		1766
		1767	ev_timer_stop (loop, timer);
		1768	ev_timer_set (timer, 60., 0.);
		1769	ev_timer_start (loop, timer);
		1770
		1771	This is relatively simple to implement, but means that each time there is
		1772	some activity, libev will first have to remove the timer from its internal
		1773	data structure and then add it again. Libev tries to be fast, but it's
		1774	still not a constant-time operation.
		1775
		1776	=item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
		1777
		1778	This is the easiest way, and involves using C<ev_timer_again> instead of
		1779	C<ev_timer_start>.
		1780
		1781	To implement this, configure an C<ev_timer> with a C<repeat> value
		1782	of C<60> and then call C<ev_timer_again> at start and each time you
		1783	successfully read or write some data. If you go into an idle state where
		1784	you do not expect data to travel on the socket, you can C<ev_timer_stop>
		1785	the timer, and C<ev_timer_again> will automatically restart it if need be.
		1786
		1787	That means you can ignore both the C<ev_timer_start> function and the
		1788	C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
		1789	member and C<ev_timer_again>.
		1790
		1791	At start:
		1792
		1793	ev_init (timer, callback);
		1794	timer->repeat = 60.;
		1795	ev_timer_again (loop, timer);
		1796
		1797	Each time there is some activity:
		1798
		1799	ev_timer_again (loop, timer);
		1800
		1801	It is even possible to change the time-out on the fly, regardless of
		1802	whether the watcher is active or not:
		1803
		1804	timer->repeat = 30.;
		1805	ev_timer_again (loop, timer);
		1806
		1807	This is slightly more efficient then stopping/starting the timer each time
		1808	you want to modify its timeout value, as libev does not have to completely
		1809	remove and re-insert the timer from/into its internal data structure.
		1810
		1811	It is, however, even simpler than the "obvious" way to do it.
		1812
		1813	=item 3. Let the timer time out, but then re-arm it as required.
		1814
		1815	This method is more tricky, but usually most efficient: Most timeouts are
		1816	relatively long compared to the intervals between other activity - in
		1817	our example, within 60 seconds, there are usually many I/O events with
		1818	associated activity resets.
		1819
		1820	In this case, it would be more efficient to leave the C<ev_timer> alone,
		1821	but remember the time of last activity, and check for a real timeout only
		1822	within the callback:
		1823
		1824	ev_tstamp last_activity; // time of last activity
		1825
		1826	static void
		1827	callback (EV_P_ ev_timer *w, int revents)
		1828	{
		1829	ev_tstamp now = ev_now (EV_A);
		1830	ev_tstamp timeout = last_activity + 60.;
		1831
		1832	// if last_activity + 60. is older than now, we did time out
		1833	if (timeout < now)
		1834	{
		1835	// timeout occurred, take action
		1836	}
		1837	else
		1838	{
		1839	// callback was invoked, but there was some activity, re-arm
		1840	// the watcher to fire in last_activity + 60, which is
		1841	// guaranteed to be in the future, so "again" is positive:
		1842	w->repeat = timeout - now;
		1843	ev_timer_again (EV_A_ w);
		1844	}
		1845	}
		1846
		1847	To summarise the callback: first calculate the real timeout (defined
		1848	as "60 seconds after the last activity"), then check if that time has
		1849	been reached, which means something I<did>, in fact, time out. Otherwise
		1850	the callback was invoked too early (C<timeout> is in the future), so
		1851	re-schedule the timer to fire at that future time, to see if maybe we have
		1852	a timeout then.
		1853
		1854	Note how C<ev_timer_again> is used, taking advantage of the
		1855	C<ev_timer_again> optimisation when the timer is already running.
		1856
		1857	This scheme causes more callback invocations (about one every 60 seconds
		1858	minus half the average time between activity), but virtually no calls to
		1859	libev to change the timeout.
		1860
		1861	To start the timer, simply initialise the watcher and set C<last_activity>
		1862	to the current time (meaning we just have some activity :), then call the
		1863	callback, which will "do the right thing" and start the timer:
		1864
		1865	ev_init (timer, callback);
		1866	last_activity = ev_now (loop);
		1867	callback (loop, timer, EV_TIMER);
		1868
		1869	And when there is some activity, simply store the current time in
		1870	C<last_activity>, no libev calls at all:
		1871
		1872	last_activity = ev_now (loop);
		1873
		1874	This technique is slightly more complex, but in most cases where the
		1875	time-out is unlikely to be triggered, much more efficient.
		1876
		1877	Changing the timeout is trivial as well (if it isn't hard-coded in the
		1878	callback :) - just change the timeout and invoke the callback, which will
		1879	fix things for you.
		1880
		1881	=item 4. Wee, just use a double-linked list for your timeouts.
		1882
		1883	If there is not one request, but many thousands (millions...), all
		1884	employing some kind of timeout with the same timeout value, then one can
		1885	do even better:
		1886
		1887	When starting the timeout, calculate the timeout value and put the timeout
		1888	at the I<end> of the list.
		1889
		1890	Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
		1891	the list is expected to fire (for example, using the technique #3).
		1892
		1893	When there is some activity, remove the timer from the list, recalculate
		1894	the timeout, append it to the end of the list again, and make sure to
		1895	update the C<ev_timer> if it was taken from the beginning of the list.
		1896
		1897	This way, one can manage an unlimited number of timeouts in O(1) time for
		1898	starting, stopping and updating the timers, at the expense of a major
		1899	complication, and having to use a constant timeout. The constant timeout
		1900	ensures that the list stays sorted.
		1901
		1902	=back
		1903
		1904	So which method the best?
		1905
		1906	Method #2 is a simple no-brain-required solution that is adequate in most
		1907	situations. Method #3 requires a bit more thinking, but handles many cases
		1908	better, and isn't very complicated either. In most case, choosing either
		1909	one is fine, with #3 being better in typical situations.
		1910
		1911	Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
		1912	rather complicated, but extremely efficient, something that really pays
		1913	off after the first million or so of active timers, i.e. it's usually
		1914	overkill :)
		1915
		1916	=head3 The special problem of time updates
		1917
		1918	Establishing the current time is a costly operation (it usually takes at
		1919	least two system calls): EV therefore updates its idea of the current
		1920	time only before and after C<ev_run> collects new events, which causes a
		1921	growing difference between C<ev_now ()> and C<ev_time ()> when handling
		1922	lots of events in one iteration.
479		1923
480	The relative timeouts are calculated relative to the C<ev_now ()>	1924	The relative timeouts are calculated relative to the C<ev_now ()>
481	time. This is usually the right thing as this timestamp refers to the time	1925	time. This is usually the right thing as this timestamp refers to the time
482	of the event triggering whatever timeout you are modifying/starting. If	1926	of the event triggering whatever timeout you are modifying/starting. If
483	you suspect event processing to be delayed and you *need* to base the timeout	1927	you suspect event processing to be delayed and you I<need> to base the
484	on the current time, use something like this to adjust for this:	1928	timeout on the current time, use something like this to adjust for this:
485		1929
486	ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);	1930	ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);
487		1931
		1932	If the event loop is suspended for a long time, you can also force an
		1933	update of the time returned by C<ev_now ()> by calling C<ev_now_update
		1934	()>.
		1935
		1936	=head3 The special problems of suspended animation
		1937
		1938	When you leave the server world it is quite customary to hit machines that
		1939	can suspend/hibernate - what happens to the clocks during such a suspend?
		1940
		1941	Some quick tests made with a Linux 2.6.28 indicate that a suspend freezes
		1942	all processes, while the clocks (C<times>, C<CLOCK_MONOTONIC>) continue
		1943	to run until the system is suspended, but they will not advance while the
		1944	system is suspended. That means, on resume, it will be as if the program
		1945	was frozen for a few seconds, but the suspend time will not be counted
		1946	towards C<ev_timer> when a monotonic clock source is used. The real time
		1947	clock advanced as expected, but if it is used as sole clocksource, then a
		1948	long suspend would be detected as a time jump by libev, and timers would
		1949	be adjusted accordingly.
		1950
		1951	I would not be surprised to see different behaviour in different between
		1952	operating systems, OS versions or even different hardware.
		1953
		1954	The other form of suspend (job control, or sending a SIGSTOP) will see a
		1955	time jump in the monotonic clocks and the realtime clock. If the program
		1956	is suspended for a very long time, and monotonic clock sources are in use,
		1957	then you can expect C<ev_timer>s to expire as the full suspension time
		1958	will be counted towards the timers. When no monotonic clock source is in
		1959	use, then libev will again assume a timejump and adjust accordingly.
		1960
		1961	It might be beneficial for this latter case to call C<ev_suspend>
		1962	and C<ev_resume> in code that handles C<SIGTSTP>, to at least get
		1963	deterministic behaviour in this case (you can do nothing against
		1964	C<SIGSTOP>).
		1965
		1966	=head3 Watcher-Specific Functions and Data Members
		1967
488	=over 4	1968	=over 4
489		1969
490	=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)	1970	=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
491		1971
492	=item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)	1972	=item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
493		1973
494	Configure the timer to trigger after C<after> seconds. If C<repeat> is	1974	Configure the timer to trigger after C<after> seconds. If C<repeat>
495	C<0.>, then it will automatically be stopped. If it is positive, then the	1975	is C<0.>, then it will automatically be stopped once the timeout is
496	timer will automatically be configured to trigger again C<repeat> seconds	1976	reached. If it is positive, then the timer will automatically be
497	later, again, and again, until stopped manually.	1977	configured to trigger again C<repeat> seconds later, again, and again,
		1978	until stopped manually.
498		1979
499	The timer itself will do a best-effort at avoiding drift, that is, if you	1980	The timer itself will do a best-effort at avoiding drift, that is, if
500	configure a timer to trigger every 10 seconds, then it will trigger at	1981	you configure a timer to trigger every 10 seconds, then it will normally
501	exactly 10 second intervals. If, however, your program cannot keep up with	1982	trigger at exactly 10 second intervals. If, however, your program cannot
502	the timer (because it takes longer than those 10 seconds to do stuff) the	1983	keep up with the timer (because it takes longer than those 10 seconds to
503	timer will not fire more than once per event loop iteration.	1984	do stuff) the timer will not fire more than once per event loop iteration.
504		1985
505	=item ev_timer_again (loop)	1986	=item ev_timer_again (loop, ev_timer *)
506		1987
507	This will act as if the timer timed out and restart it again if it is	1988	This will act as if the timer timed out and restart it again if it is
508	repeating. The exact semantics are:	1989	repeating. The exact semantics are:
509		1990
		1991	If the timer is pending, its pending status is cleared.
		1992
510	If the timer is started but nonrepeating, stop it.	1993	If the timer is started but non-repeating, stop it (as if it timed out).
511		1994
512	If the timer is repeating, either start it if necessary (with the repeat	1995	If the timer is repeating, either start it if necessary (with the
513	value), or reset the running timer to the repeat value.	1996	C<repeat> value), or reset the running timer to the C<repeat> value.
514		1997
515	This sounds a bit complicated, but here is a useful and typical	1998	This sounds a bit complicated, see L<Be smart about timeouts>, above, for a
516	example: Imagine you have a tcp connection and you want a so-called idle	1999	usage example.
517	timeout, that is, you want to be called when there have been, say, 60	2000
518	seconds of inactivity on the socket. The easiest way to do this is to	2001	=item ev_tstamp ev_timer_remaining (loop, ev_timer *)
519	configure an C<ev_timer> with after=repeat=60 and calling ev_timer_again each	2002
520	time you successfully read or write some data. If you go into an idle	2003	Returns the remaining time until a timer fires. If the timer is active,
521	state where you do not expect data to travel on the socket, you can stop	2004	then this time is relative to the current event loop time, otherwise it's
522	the timer, and again will automatically restart it if need be.	2005	the timeout value currently configured.
		2006
		2007	That is, after an C<ev_timer_set (w, 5, 7)>, C<ev_timer_remaining> returns
		2008	C<5>. When the timer is started and one second passes, C<ev_timer_remaining>
		2009	will return C<4>. When the timer expires and is restarted, it will return
		2010	roughly C<7> (likely slightly less as callback invocation takes some time,
		2011	too), and so on.
		2012
		2013	=item ev_tstamp repeat [read-write]
		2014
		2015	The current C<repeat> value. Will be used each time the watcher times out
		2016	or C<ev_timer_again> is called, and determines the next timeout (if any),
		2017	which is also when any modifications are taken into account.
523		2018
524	=back	2019	=back
525		2020
		2021	=head3 Examples
		2022
		2023	Example: Create a timer that fires after 60 seconds.
		2024
		2025	static void
		2026	one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
		2027	{
		2028	.. one minute over, w is actually stopped right here
		2029	}
		2030
		2031	ev_timer mytimer;
		2032	ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
		2033	ev_timer_start (loop, &mytimer);
		2034
		2035	Example: Create a timeout timer that times out after 10 seconds of
		2036	inactivity.
		2037
		2038	static void
		2039	timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
		2040	{
		2041	.. ten seconds without any activity
		2042	}
		2043
		2044	ev_timer mytimer;
		2045	ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
		2046	ev_timer_again (&mytimer); /* start timer */
		2047	ev_run (loop, 0);
		2048
		2049	// and in some piece of code that gets executed on any "activity":
		2050	// reset the timeout to start ticking again at 10 seconds
		2051	ev_timer_again (&mytimer);
		2052
		2053
526	=head2 C<ev_periodic> - to cron or not to cron	2054	=head2 C<ev_periodic> - to cron or not to cron?
527		2055
528	Periodic watchers are also timers of a kind, but they are very versatile	2056	Periodic watchers are also timers of a kind, but they are very versatile
529	(and unfortunately a bit complex).	2057	(and unfortunately a bit complex).
530		2058
531	Unlike C<ev_timer>'s, they are not based on real time (or relative time)	2059	Unlike C<ev_timer>, periodic watchers are not based on real time (or
532	but on wallclock time (absolute time). You can tell a periodic watcher	2060	relative time, the physical time that passes) but on wall clock time
533	to trigger "at" some specific point in time. For example, if you tell a	2061	(absolute time, the thing you can read on your calender or clock). The
534	periodic watcher to trigger in 10 seconds (by specifiying e.g. c<ev_now ()	2062	difference is that wall clock time can run faster or slower than real
535	+ 10.>) and then reset your system clock to the last year, then it will	2063	time, and time jumps are not uncommon (e.g. when you adjust your
536	take a year to trigger the event (unlike an C<ev_timer>, which would trigger	2064	wrist-watch).
537	roughly 10 seconds later and of course not if you reset your system time
538	again).
539		2065
540	They can also be used to implement vastly more complex timers, such as	2066	You can tell a periodic watcher to trigger after some specific point
		2067	in time: for example, if you tell a periodic watcher to trigger "in 10
		2068	seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
		2069	not a delay) and then reset your system clock to January of the previous
		2070	year, then it will take a year or more to trigger the event (unlike an
		2071	C<ev_timer>, which would still trigger roughly 10 seconds after starting
		2072	it, as it uses a relative timeout).
		2073
		2074	C<ev_periodic> watchers can also be used to implement vastly more complex
541	triggering an event on eahc midnight, local time.	2075	timers, such as triggering an event on each "midnight, local time", or
		2076	other complicated rules. This cannot be done with C<ev_timer> watchers, as
		2077	those cannot react to time jumps.
		2078
		2079	As with timers, the callback is guaranteed to be invoked only when the
		2080	point in time where it is supposed to trigger has passed. If multiple
		2081	timers become ready during the same loop iteration then the ones with
		2082	earlier time-out values are invoked before ones with later time-out values
		2083	(but this is no longer true when a callback calls C<ev_run> recursively).
		2084
		2085	=head3 Watcher-Specific Functions and Data Members
542		2086
543	=over 4	2087	=over 4
544		2088
545	=item ev_periodic_init (ev_periodic *, callback, ev_tstamp at, ev_tstamp interval, reschedule_cb)	2089	=item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
546		2090
547	=item ev_periodic_set (ev_periodic *, ev_tstamp after, ev_tstamp repeat, reschedule_cb)	2091	=item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
548		2092
549	Lots of arguments, lets sort it out... There are basically three modes of	2093	Lots of arguments, let's sort it out... There are basically three modes of
550	operation, and we will explain them from simplest to complex:	2094	operation, and we will explain them from simplest to most complex:
551
552		2095
553	=over 4	2096	=over 4
554		2097
555	=item * absolute timer (interval = reschedule_cb = 0)	2098	=item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
556		2099
557	In this configuration the watcher triggers an event at the wallclock time	2100	In this configuration the watcher triggers an event after the wall clock
558	C<at> and doesn't repeat. It will not adjust when a time jump occurs,	2101	time C<offset> has passed. It will not repeat and will not adjust when a
559	that is, if it is to be run at January 1st 2011 then it will run when the	2102	time jump occurs, that is, if it is to be run at January 1st 2011 then it
560	system time reaches or surpasses this time.	2103	will be stopped and invoked when the system clock reaches or surpasses
		2104	this point in time.
561		2105
562	=item * non-repeating interval timer (interval > 0, reschedule_cb = 0)	2106	=item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
563		2107
564	In this mode the watcher will always be scheduled to time out at the next	2108	In this mode the watcher will always be scheduled to time out at the next
565	C<at + N * interval> time (for some integer N) and then repeat, regardless	2109	C<offset + N * interval> time (for some integer N, which can also be
566	of any time jumps.	2110	negative) and then repeat, regardless of any time jumps. The C<offset>
		2111	argument is merely an offset into the C<interval> periods.
567		2112
568	This can be used to create timers that do not drift with respect to system	2113	This can be used to create timers that do not drift with respect to the
569	time:	2114	system clock, for example, here is an C<ev_periodic> that triggers each
		2115	hour, on the hour (with respect to UTC):
570		2116
571	ev_periodic_set (&periodic, 0., 3600., 0);	2117	ev_periodic_set (&periodic, 0., 3600., 0);
572		2118
573	This doesn't mean there will always be 3600 seconds in between triggers,	2119	This doesn't mean there will always be 3600 seconds in between triggers,
574	but only that the the callback will be called when the system time shows a	2120	but only that the callback will be called when the system time shows a
575	full hour (UTC), or more correctly, when the system time is evenly divisible	2121	full hour (UTC), or more correctly, when the system time is evenly divisible
576	by 3600.	2122	by 3600.
577		2123
578	Another way to think about it (for the mathematically inclined) is that	2124	Another way to think about it (for the mathematically inclined) is that
579	C<ev_periodic> will try to run the callback in this mode at the next possible	2125	C<ev_periodic> will try to run the callback in this mode at the next possible
580	time where C<time = at (mod interval)>, regardless of any time jumps.	2126	time where C<time = offset (mod interval)>, regardless of any time jumps.
581		2127
		2128	For numerical stability it is preferable that the C<offset> value is near
		2129	C<ev_now ()> (the current time), but there is no range requirement for
		2130	this value, and in fact is often specified as zero.
		2131
		2132	Note also that there is an upper limit to how often a timer can fire (CPU
		2133	speed for example), so if C<interval> is very small then timing stability
		2134	will of course deteriorate. Libev itself tries to be exact to be about one
		2135	millisecond (if the OS supports it and the machine is fast enough).
		2136
582	=item * manual reschedule mode (reschedule_cb = callback)	2137	=item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
583		2138
584	In this mode the values for C<interval> and C<at> are both being	2139	In this mode the values for C<interval> and C<offset> are both being
585	ignored. Instead, each time the periodic watcher gets scheduled, the	2140	ignored. Instead, each time the periodic watcher gets scheduled, the
586	reschedule callback will be called with the watcher as first, and the	2141	reschedule callback will be called with the watcher as first, and the
587	current time as second argument.	2142	current time as second argument.
588		2143
589	NOTE: I<This callback MUST NOT stop or destroy any periodic watcher,	2144	NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, ever,
590	ever, or make any event loop modifications>. If you need to stop it,	2145	or make ANY other event loop modifications whatsoever, unless explicitly
591	return C<now + 1e30> (or so, fudge fudge) and stop it afterwards (e.g. by	2146	allowed by documentation here>.
592	starting a prepare watcher).
593		2147
		2148	If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
		2149	it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
		2150	only event loop modification you are allowed to do).
		2151
594	Its prototype is C<ev_tstamp (*reschedule_cb)(struct ev_periodic *w,	2152	The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
595	ev_tstamp now)>, e.g.:	2153	*w, ev_tstamp now)>, e.g.:
596		2154
		2155	static ev_tstamp
597	static ev_tstamp my_rescheduler (struct ev_periodic *w, ev_tstamp now)	2156	my_rescheduler (ev_periodic *w, ev_tstamp now)
598	{	2157	{
599	return now + 60.;	2158	return now + 60.;
600	}	2159	}
601		2160
602	It must return the next time to trigger, based on the passed time value	2161	It must return the next time to trigger, based on the passed time value
603	(that is, the lowest time value larger than to the second argument). It	2162	(that is, the lowest time value larger than to the second argument). It
604	will usually be called just before the callback will be triggered, but	2163	will usually be called just before the callback will be triggered, but
605	might be called at other times, too.	2164	might be called at other times, too.
606		2165
607	NOTE: I<< This callback must always return a time that is later than the	2166	NOTE: I<< This callback must always return a time that is higher than or
608	passed C<now> value >>. Not even C<now> itself will do, it I<must> be larger.	2167	equal to the passed C<now> value >>.
609		2168
610	This can be used to create very complex timers, such as a timer that	2169	This can be used to create very complex timers, such as a timer that
611	triggers on each midnight, local time. To do this, you would calculate the	2170	triggers on "next midnight, local time". To do this, you would calculate the
612	next midnight after C<now> and return the timestamp value for this. How	2171	next midnight after C<now> and return the timestamp value for this. How
613	you do this is, again, up to you (but it is not trivial, which is the main	2172	you do this is, again, up to you (but it is not trivial, which is the main
614	reason I omitted it as an example).	2173	reason I omitted it as an example).
615		2174
616	=back	2175	=back
…		…
620	Simply stops and restarts the periodic watcher again. This is only useful	2179	Simply stops and restarts the periodic watcher again. This is only useful
621	when you changed some parameters or the reschedule callback would return	2180	when you changed some parameters or the reschedule callback would return
622	a different time than the last time it was called (e.g. in a crond like	2181	a different time than the last time it was called (e.g. in a crond like
623	program when the crontabs have changed).	2182	program when the crontabs have changed).
624		2183
		2184	=item ev_tstamp ev_periodic_at (ev_periodic *)
		2185
		2186	When active, returns the absolute time that the watcher is supposed
		2187	to trigger next. This is not the same as the C<offset> argument to
		2188	C<ev_periodic_set>, but indeed works even in interval and manual
		2189	rescheduling modes.
		2190
		2191	=item ev_tstamp offset [read-write]
		2192
		2193	When repeating, this contains the offset value, otherwise this is the
		2194	absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
		2195	although libev might modify this value for better numerical stability).
		2196
		2197	Can be modified any time, but changes only take effect when the periodic
		2198	timer fires or C<ev_periodic_again> is being called.
		2199
		2200	=item ev_tstamp interval [read-write]
		2201
		2202	The current interval value. Can be modified any time, but changes only
		2203	take effect when the periodic timer fires or C<ev_periodic_again> is being
		2204	called.
		2205
		2206	=item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
		2207
		2208	The current reschedule callback, or C<0>, if this functionality is
		2209	switched off. Can be changed any time, but changes only take effect when
		2210	the periodic timer fires or C<ev_periodic_again> is being called.
		2211
625	=back	2212	=back
626		2213
		2214	=head3 Examples
		2215
		2216	Example: Call a callback every hour, or, more precisely, whenever the
		2217	system time is divisible by 3600. The callback invocation times have
		2218	potentially a lot of jitter, but good long-term stability.
		2219
		2220	static void
		2221	clock_cb (struct ev_loop *loop, ev_periodic *w, int revents)
		2222	{
		2223	... its now a full hour (UTC, or TAI or whatever your clock follows)
		2224	}
		2225
		2226	ev_periodic hourly_tick;
		2227	ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
		2228	ev_periodic_start (loop, &hourly_tick);
		2229
		2230	Example: The same as above, but use a reschedule callback to do it:
		2231
		2232	#include <math.h>
		2233
		2234	static ev_tstamp
		2235	my_scheduler_cb (ev_periodic *w, ev_tstamp now)
		2236	{
		2237	return now + (3600. - fmod (now, 3600.));
		2238	}
		2239
		2240	ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
		2241
		2242	Example: Call a callback every hour, starting now:
		2243
		2244	ev_periodic hourly_tick;
		2245	ev_periodic_init (&hourly_tick, clock_cb,
		2246	fmod (ev_now (loop), 3600.), 3600., 0);
		2247	ev_periodic_start (loop, &hourly_tick);
		2248
		2249
627	=head2 C<ev_signal> - signal me when a signal gets signalled	2250	=head2 C<ev_signal> - signal me when a signal gets signalled!
628		2251
629	Signal watchers will trigger an event when the process receives a specific	2252	Signal watchers will trigger an event when the process receives a specific
630	signal one or more times. Even though signals are very asynchronous, libev	2253	signal one or more times. Even though signals are very asynchronous, libev
631	will try it's best to deliver signals synchronously, i.e. as part of the	2254	will try it's best to deliver signals synchronously, i.e. as part of the
632	normal event processing, like any other event.	2255	normal event processing, like any other event.
633		2256
		2257	If you want signals to be delivered truly asynchronously, just use
		2258	C<sigaction> as you would do without libev and forget about sharing
		2259	the signal. You can even use C<ev_async> from a signal handler to
		2260	synchronously wake up an event loop.
		2261
634	You can configure as many watchers as you like per signal. Only when the	2262	You can configure as many watchers as you like for the same signal, but
		2263	only within the same loop, i.e. you can watch for C<SIGINT> in your
		2264	default loop and for C<SIGIO> in another loop, but you cannot watch for
		2265	C<SIGINT> in both the default loop and another loop at the same time. At
		2266	the moment, C<SIGCHLD> is permanently tied to the default loop.
		2267
635	first watcher gets started will libev actually register a signal watcher	2268	When the first watcher gets started will libev actually register something
636	with the kernel (thus it coexists with your own signal handlers as long	2269	with the kernel (thus it coexists with your own signal handlers as long as
637	as you don't register any with libev). Similarly, when the last signal	2270	you don't register any with libev for the same signal).
638	watcher for a signal is stopped libev will reset the signal handler to	2271
639	SIG_DFL (regardless of what it was set to before).	2272	If possible and supported, libev will install its handlers with
		2273	C<SA_RESTART> (or equivalent) behaviour enabled, so system calls should
		2274	not be unduly interrupted. If you have a problem with system calls getting
		2275	interrupted by signals you can block all signals in an C<ev_check> watcher
		2276	and unblock them in an C<ev_prepare> watcher.
		2277
		2278	=head3 The special problem of inheritance over fork/execve/pthread_create
		2279
		2280	Both the signal mask (C<sigprocmask>) and the signal disposition
		2281	(C<sigaction>) are unspecified after starting a signal watcher (and after
		2282	stopping it again), that is, libev might or might not block the signal,
		2283	and might or might not set or restore the installed signal handler.
		2284
		2285	While this does not matter for the signal disposition (libev never
		2286	sets signals to C<SIG_IGN>, so handlers will be reset to C<SIG_DFL> on
		2287	C<execve>), this matters for the signal mask: many programs do not expect
		2288	certain signals to be blocked.
		2289
		2290	This means that before calling C<exec> (from the child) you should reset
		2291	the signal mask to whatever "default" you expect (all clear is a good
		2292	choice usually).
		2293
		2294	The simplest way to ensure that the signal mask is reset in the child is
		2295	to install a fork handler with C<pthread_atfork> that resets it. That will
		2296	catch fork calls done by libraries (such as the libc) as well.
		2297
		2298	In current versions of libev, the signal will not be blocked indefinitely
		2299	unless you use the C<signalfd> API (C<EV_SIGNALFD>). While this reduces
		2300	the window of opportunity for problems, it will not go away, as libev
		2301	I<has> to modify the signal mask, at least temporarily.
		2302
		2303	So I can't stress this enough: I<If you do not reset your signal mask when
		2304	you expect it to be empty, you have a race condition in your code>. This
		2305	is not a libev-specific thing, this is true for most event libraries.
		2306
		2307	=head3 Watcher-Specific Functions and Data Members
640		2308
641	=over 4	2309	=over 4
642		2310
643	=item ev_signal_init (ev_signal *, callback, int signum)	2311	=item ev_signal_init (ev_signal *, callback, int signum)
644		2312
645	=item ev_signal_set (ev_signal *, int signum)	2313	=item ev_signal_set (ev_signal *, int signum)
646		2314
647	Configures the watcher to trigger on the given signal number (usually one	2315	Configures the watcher to trigger on the given signal number (usually one
648	of the C<SIGxxx> constants).	2316	of the C<SIGxxx> constants).
649		2317
		2318	=item int signum [read-only]
		2319
		2320	The signal the watcher watches out for.
		2321
650	=back	2322	=back
651		2323
		2324	=head3 Examples
		2325
		2326	Example: Try to exit cleanly on SIGINT.
		2327
		2328	static void
		2329	sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
		2330	{
		2331	ev_break (loop, EVBREAK_ALL);
		2332	}
		2333
		2334	ev_signal signal_watcher;
		2335	ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
		2336	ev_signal_start (loop, &signal_watcher);
		2337
		2338
652	=head2 C<ev_child> - wait for pid status changes	2339	=head2 C<ev_child> - watch out for process status changes
653		2340
654	Child watchers trigger when your process receives a SIGCHLD in response to	2341	Child watchers trigger when your process receives a SIGCHLD in response to
655	some child status changes (most typically when a child of yours dies).	2342	some child status changes (most typically when a child of yours dies or
		2343	exits). It is permissible to install a child watcher I<after> the child
		2344	has been forked (which implies it might have already exited), as long
		2345	as the event loop isn't entered (or is continued from a watcher), i.e.,
		2346	forking and then immediately registering a watcher for the child is fine,
		2347	but forking and registering a watcher a few event loop iterations later or
		2348	in the next callback invocation is not.
		2349
		2350	Only the default event loop is capable of handling signals, and therefore
		2351	you can only register child watchers in the default event loop.
		2352
		2353	Due to some design glitches inside libev, child watchers will always be
		2354	handled at maximum priority (their priority is set to C<EV_MAXPRI> by
		2355	libev)
		2356
		2357	=head3 Process Interaction
		2358
		2359	Libev grabs C<SIGCHLD> as soon as the default event loop is
		2360	initialised. This is necessary to guarantee proper behaviour even if the
		2361	first child watcher is started after the child exits. The occurrence
		2362	of C<SIGCHLD> is recorded asynchronously, but child reaping is done
		2363	synchronously as part of the event loop processing. Libev always reaps all
		2364	children, even ones not watched.
		2365
		2366	=head3 Overriding the Built-In Processing
		2367
		2368	Libev offers no special support for overriding the built-in child
		2369	processing, but if your application collides with libev's default child
		2370	handler, you can override it easily by installing your own handler for
		2371	C<SIGCHLD> after initialising the default loop, and making sure the
		2372	default loop never gets destroyed. You are encouraged, however, to use an
		2373	event-based approach to child reaping and thus use libev's support for
		2374	that, so other libev users can use C<ev_child> watchers freely.
		2375
		2376	=head3 Stopping the Child Watcher
		2377
		2378	Currently, the child watcher never gets stopped, even when the
		2379	child terminates, so normally one needs to stop the watcher in the
		2380	callback. Future versions of libev might stop the watcher automatically
		2381	when a child exit is detected (calling C<ev_child_stop> twice is not a
		2382	problem).
		2383
		2384	=head3 Watcher-Specific Functions and Data Members
656		2385
657	=over 4	2386	=over 4
658		2387
659	=item ev_child_init (ev_child *, callback, int pid)	2388	=item ev_child_init (ev_child *, callback, int pid, int trace)
660		2389
661	=item ev_child_set (ev_child *, int pid)	2390	=item ev_child_set (ev_child *, int pid, int trace)
662		2391
663	Configures the watcher to wait for status changes of process C<pid> (or	2392	Configures the watcher to wait for status changes of process C<pid> (or
664	I<any> process if C<pid> is specified as C<0>). The callback can look	2393	I<any> process if C<pid> is specified as C<0>). The callback can look
665	at the C<rstatus> member of the C<ev_child> watcher structure to see	2394	at the C<rstatus> member of the C<ev_child> watcher structure to see
666	the status word (use the macros from C<sys/wait.h> and see your systems	2395	the status word (use the macros from C<sys/wait.h> and see your systems
667	C<waitpid> documentation). The C<rpid> member contains the pid of the	2396	C<waitpid> documentation). The C<rpid> member contains the pid of the
668	process causing the status change.	2397	process causing the status change. C<trace> must be either C<0> (only
		2398	activate the watcher when the process terminates) or C<1> (additionally
		2399	activate the watcher when the process is stopped or continued).
		2400
		2401	=item int pid [read-only]
		2402
		2403	The process id this watcher watches out for, or C<0>, meaning any process id.
		2404
		2405	=item int rpid [read-write]
		2406
		2407	The process id that detected a status change.
		2408
		2409	=item int rstatus [read-write]
		2410
		2411	The process exit/trace status caused by C<rpid> (see your systems
		2412	C<waitpid> and C<sys/wait.h> documentation for details).
669		2413
670	=back	2414	=back
671		2415
		2416	=head3 Examples
		2417
		2418	Example: C<fork()> a new process and install a child handler to wait for
		2419	its completion.
		2420
		2421	ev_child cw;
		2422
		2423	static void
		2424	child_cb (EV_P_ ev_child *w, int revents)
		2425	{
		2426	ev_child_stop (EV_A_ w);
		2427	printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
		2428	}
		2429
		2430	pid_t pid = fork ();
		2431
		2432	if (pid < 0)
		2433	// error
		2434	else if (pid == 0)
		2435	{
		2436	// the forked child executes here
		2437	exit (1);
		2438	}
		2439	else
		2440	{
		2441	ev_child_init (&cw, child_cb, pid, 0);
		2442	ev_child_start (EV_DEFAULT_ &cw);
		2443	}
		2444
		2445
		2446	=head2 C<ev_stat> - did the file attributes just change?
		2447
		2448	This watches a file system path for attribute changes. That is, it calls
		2449	C<stat> on that path in regular intervals (or when the OS says it changed)
		2450	and sees if it changed compared to the last time, invoking the callback if
		2451	it did.
		2452
		2453	The path does not need to exist: changing from "path exists" to "path does
		2454	not exist" is a status change like any other. The condition "path does not
		2455	exist" (or more correctly "path cannot be stat'ed") is signified by the
		2456	C<st_nlink> field being zero (which is otherwise always forced to be at
		2457	least one) and all the other fields of the stat buffer having unspecified
		2458	contents.
		2459
		2460	The path I<must not> end in a slash or contain special components such as
		2461	C<.> or C<..>. The path I<should> be absolute: If it is relative and
		2462	your working directory changes, then the behaviour is undefined.
		2463
		2464	Since there is no portable change notification interface available, the
		2465	portable implementation simply calls C<stat(2)> regularly on the path
		2466	to see if it changed somehow. You can specify a recommended polling
		2467	interval for this case. If you specify a polling interval of C<0> (highly
		2468	recommended!) then a I<suitable, unspecified default> value will be used
		2469	(which you can expect to be around five seconds, although this might
		2470	change dynamically). Libev will also impose a minimum interval which is
		2471	currently around C<0.1>, but that's usually overkill.
		2472
		2473	This watcher type is not meant for massive numbers of stat watchers,
		2474	as even with OS-supported change notifications, this can be
		2475	resource-intensive.
		2476
		2477	At the time of this writing, the only OS-specific interface implemented
		2478	is the Linux inotify interface (implementing kqueue support is left as an
		2479	exercise for the reader. Note, however, that the author sees no way of
		2480	implementing C<ev_stat> semantics with kqueue, except as a hint).
		2481
		2482	=head3 ABI Issues (Largefile Support)
		2483
		2484	Libev by default (unless the user overrides this) uses the default
		2485	compilation environment, which means that on systems with large file
		2486	support disabled by default, you get the 32 bit version of the stat
		2487	structure. When using the library from programs that change the ABI to
		2488	use 64 bit file offsets the programs will fail. In that case you have to
		2489	compile libev with the same flags to get binary compatibility. This is
		2490	obviously the case with any flags that change the ABI, but the problem is
		2491	most noticeably displayed with ev_stat and large file support.
		2492
		2493	The solution for this is to lobby your distribution maker to make large
		2494	file interfaces available by default (as e.g. FreeBSD does) and not
		2495	optional. Libev cannot simply switch on large file support because it has
		2496	to exchange stat structures with application programs compiled using the
		2497	default compilation environment.
		2498
		2499	=head3 Inotify and Kqueue
		2500
		2501	When C<inotify (7)> support has been compiled into libev and present at
		2502	runtime, it will be used to speed up change detection where possible. The
		2503	inotify descriptor will be created lazily when the first C<ev_stat>
		2504	watcher is being started.
		2505
		2506	Inotify presence does not change the semantics of C<ev_stat> watchers
		2507	except that changes might be detected earlier, and in some cases, to avoid
		2508	making regular C<stat> calls. Even in the presence of inotify support
		2509	there are many cases where libev has to resort to regular C<stat> polling,
		2510	but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
		2511	many bugs), the path exists (i.e. stat succeeds), and the path resides on
		2512	a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
		2513	xfs are fully working) libev usually gets away without polling.
		2514
		2515	There is no support for kqueue, as apparently it cannot be used to
		2516	implement this functionality, due to the requirement of having a file
		2517	descriptor open on the object at all times, and detecting renames, unlinks
		2518	etc. is difficult.
		2519
		2520	=head3 C<stat ()> is a synchronous operation
		2521
		2522	Libev doesn't normally do any kind of I/O itself, and so is not blocking
		2523	the process. The exception are C<ev_stat> watchers - those call C<stat
		2524	()>, which is a synchronous operation.
		2525
		2526	For local paths, this usually doesn't matter: unless the system is very
		2527	busy or the intervals between stat's are large, a stat call will be fast,
		2528	as the path data is usually in memory already (except when starting the
		2529	watcher).
		2530
		2531	For networked file systems, calling C<stat ()> can block an indefinite
		2532	time due to network issues, and even under good conditions, a stat call
		2533	often takes multiple milliseconds.
		2534
		2535	Therefore, it is best to avoid using C<ev_stat> watchers on networked
		2536	paths, although this is fully supported by libev.
		2537
		2538	=head3 The special problem of stat time resolution
		2539
		2540	The C<stat ()> system call only supports full-second resolution portably,
		2541	and even on systems where the resolution is higher, most file systems
		2542	still only support whole seconds.
		2543
		2544	That means that, if the time is the only thing that changes, you can
		2545	easily miss updates: on the first update, C<ev_stat> detects a change and
		2546	calls your callback, which does something. When there is another update
		2547	within the same second, C<ev_stat> will be unable to detect unless the
		2548	stat data does change in other ways (e.g. file size).
		2549
		2550	The solution to this is to delay acting on a change for slightly more
		2551	than a second (or till slightly after the next full second boundary), using
		2552	a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
		2553	ev_timer_again (loop, w)>).
		2554
		2555	The C<.02> offset is added to work around small timing inconsistencies
		2556	of some operating systems (where the second counter of the current time
		2557	might be be delayed. One such system is the Linux kernel, where a call to
		2558	C<gettimeofday> might return a timestamp with a full second later than
		2559	a subsequent C<time> call - if the equivalent of C<time ()> is used to
		2560	update file times then there will be a small window where the kernel uses
		2561	the previous second to update file times but libev might already execute
		2562	the timer callback).
		2563
		2564	=head3 Watcher-Specific Functions and Data Members
		2565
		2566	=over 4
		2567
		2568	=item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
		2569
		2570	=item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
		2571
		2572	Configures the watcher to wait for status changes of the given
		2573	C<path>. The C<interval> is a hint on how quickly a change is expected to
		2574	be detected and should normally be specified as C<0> to let libev choose
		2575	a suitable value. The memory pointed to by C<path> must point to the same
		2576	path for as long as the watcher is active.
		2577
		2578	The callback will receive an C<EV_STAT> event when a change was detected,
		2579	relative to the attributes at the time the watcher was started (or the
		2580	last change was detected).
		2581
		2582	=item ev_stat_stat (loop, ev_stat *)
		2583
		2584	Updates the stat buffer immediately with new values. If you change the
		2585	watched path in your callback, you could call this function to avoid
		2586	detecting this change (while introducing a race condition if you are not
		2587	the only one changing the path). Can also be useful simply to find out the
		2588	new values.
		2589
		2590	=item ev_statdata attr [read-only]
		2591
		2592	The most-recently detected attributes of the file. Although the type is
		2593	C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
		2594	suitable for your system, but you can only rely on the POSIX-standardised
		2595	members to be present. If the C<st_nlink> member is C<0>, then there was
		2596	some error while C<stat>ing the file.
		2597
		2598	=item ev_statdata prev [read-only]
		2599
		2600	The previous attributes of the file. The callback gets invoked whenever
		2601	C<prev> != C<attr>, or, more precisely, one or more of these members
		2602	differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
		2603	C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
		2604
		2605	=item ev_tstamp interval [read-only]
		2606
		2607	The specified interval.
		2608
		2609	=item const char *path [read-only]
		2610
		2611	The file system path that is being watched.
		2612
		2613	=back
		2614
		2615	=head3 Examples
		2616
		2617	Example: Watch C</etc/passwd> for attribute changes.
		2618
		2619	static void
		2620	passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
		2621	{
		2622	/* /etc/passwd changed in some way */
		2623	if (w->attr.st_nlink)
		2624	{
		2625	printf ("passwd current size %ld\n", (long)w->attr.st_size);
		2626	printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
		2627	printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
		2628	}
		2629	else
		2630	/* you shalt not abuse printf for puts */
		2631	puts ("wow, /etc/passwd is not there, expect problems. "
		2632	"if this is windows, they already arrived\n");
		2633	}
		2634
		2635	...
		2636	ev_stat passwd;
		2637
		2638	ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
		2639	ev_stat_start (loop, &passwd);
		2640
		2641	Example: Like above, but additionally use a one-second delay so we do not
		2642	miss updates (however, frequent updates will delay processing, too, so
		2643	one might do the work both on C<ev_stat> callback invocation I<and> on
		2644	C<ev_timer> callback invocation).
		2645
		2646	static ev_stat passwd;
		2647	static ev_timer timer;
		2648
		2649	static void
		2650	timer_cb (EV_P_ ev_timer *w, int revents)
		2651	{
		2652	ev_timer_stop (EV_A_ w);
		2653
		2654	/* now it's one second after the most recent passwd change */
		2655	}
		2656
		2657	static void
		2658	stat_cb (EV_P_ ev_stat *w, int revents)
		2659	{
		2660	/* reset the one-second timer */
		2661	ev_timer_again (EV_A_ &timer);
		2662	}
		2663
		2664	...
		2665	ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
		2666	ev_stat_start (loop, &passwd);
		2667	ev_timer_init (&timer, timer_cb, 0., 1.02);
		2668
		2669
672	=head2 C<ev_idle> - when you've got nothing better to do	2670	=head2 C<ev_idle> - when you've got nothing better to do...
673		2671
674	Idle watchers trigger events when there are no other events are pending	2672	Idle watchers trigger events when no other events of the same or higher
675	(prepare, check and other idle watchers do not count). That is, as long	2673	priority are pending (prepare, check and other idle watchers do not count
676	as your process is busy handling sockets or timeouts (or even signals,	2674	as receiving "events").
677	imagine) it will not be triggered. But when your process is idle all idle	2675
678	watchers are being called again and again, once per event loop iteration -	2676	That is, as long as your process is busy handling sockets or timeouts
		2677	(or even signals, imagine) of the same or higher priority it will not be
		2678	triggered. But when your process is idle (or only lower-priority watchers
		2679	are pending), the idle watchers are being called once per event loop
679	until stopped, that is, or your process receives more events and becomes	2680	iteration - until stopped, that is, or your process receives more events
680	busy.	2681	and becomes busy again with higher priority stuff.
681		2682
682	The most noteworthy effect is that as long as any idle watchers are	2683	The most noteworthy effect is that as long as any idle watchers are
683	active, the process will not block when waiting for new events.	2684	active, the process will not block when waiting for new events.
684		2685
685	Apart from keeping your process non-blocking (which is a useful	2686	Apart from keeping your process non-blocking (which is a useful
686	effect on its own sometimes), idle watchers are a good place to do	2687	effect on its own sometimes), idle watchers are a good place to do
687	"pseudo-background processing", or delay processing stuff to after the	2688	"pseudo-background processing", or delay processing stuff to after the
688	event loop has handled all outstanding events.	2689	event loop has handled all outstanding events.
689		2690
		2691	=head3 Watcher-Specific Functions and Data Members
		2692
690	=over 4	2693	=over 4
691		2694
692	=item ev_idle_init (ev_signal *, callback)	2695	=item ev_idle_init (ev_idle *, callback)
693		2696
694	Initialises and configures the idle watcher - it has no parameters of any	2697	Initialises and configures the idle watcher - it has no parameters of any
695	kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,	2698	kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
696	believe me.	2699	believe me.
697		2700
698	=back	2701	=back
699		2702
		2703	=head3 Examples
		2704
		2705	Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
		2706	callback, free it. Also, use no error checking, as usual.
		2707
		2708	static void
		2709	idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
		2710	{
		2711	free (w);
		2712	// now do something you wanted to do when the program has
		2713	// no longer anything immediate to do.
		2714	}
		2715
		2716	ev_idle *idle_watcher = malloc (sizeof (ev_idle));
		2717	ev_idle_init (idle_watcher, idle_cb);
		2718	ev_idle_start (loop, idle_watcher);
		2719
		2720
700	=head2 C<ev_prepare> and C<ev_check> - customise your event loop	2721	=head2 C<ev_prepare> and C<ev_check> - customise your event loop!
701		2722
702	Prepare and check watchers are usually (but not always) used in tandem:	2723	Prepare and check watchers are usually (but not always) used in pairs:
703	prepare watchers get invoked before the process blocks and check watchers	2724	prepare watchers get invoked before the process blocks and check watchers
704	afterwards.	2725	afterwards.
705		2726
		2727	You I<must not> call C<ev_run> or similar functions that enter
		2728	the current event loop from either C<ev_prepare> or C<ev_check>
		2729	watchers. Other loops than the current one are fine, however. The
		2730	rationale behind this is that you do not need to check for recursion in
		2731	those watchers, i.e. the sequence will always be C<ev_prepare>, blocking,
		2732	C<ev_check> so if you have one watcher of each kind they will always be
		2733	called in pairs bracketing the blocking call.
		2734
706	Their main purpose is to integrate other event mechanisms into libev. This	2735	Their main purpose is to integrate other event mechanisms into libev and
707	could be used, for example, to track variable changes, implement your own	2736	their use is somewhat advanced. They could be used, for example, to track
708	watchers, integrate net-snmp or a coroutine library and lots more.	2737	variable changes, implement your own watchers, integrate net-snmp or a
		2738	coroutine library and lots more. They are also occasionally useful if
		2739	you cache some data and want to flush it before blocking (for example,
		2740	in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
		2741	watcher).
709		2742
710	This is done by examining in each prepare call which file descriptors need	2743	This is done by examining in each prepare call which file descriptors
711	to be watched by the other library, registering C<ev_io> watchers for	2744	need to be watched by the other library, registering C<ev_io> watchers
712	them and starting an C<ev_timer> watcher for any timeouts (many libraries	2745	for them and starting an C<ev_timer> watcher for any timeouts (many
713	provide just this functionality). Then, in the check watcher you check for	2746	libraries provide exactly this functionality). Then, in the check watcher,
714	any events that occured (by checking the pending status of all watchers	2747	you check for any events that occurred (by checking the pending status
715	and stopping them) and call back into the library. The I/O and timer	2748	of all watchers and stopping them) and call back into the library. The
716	callbacks will never actually be called (but must be valid nevertheless,	2749	I/O and timer callbacks will never actually be called (but must be valid
717	because you never know, you know?).	2750	nevertheless, because you never know, you know?).
718		2751
719	As another example, the Perl Coro module uses these hooks to integrate	2752	As another example, the Perl Coro module uses these hooks to integrate
720	coroutines into libev programs, by yielding to other active coroutines	2753	coroutines into libev programs, by yielding to other active coroutines
721	during each prepare and only letting the process block if no coroutines	2754	during each prepare and only letting the process block if no coroutines
722	are ready to run (it's actually more complicated: it only runs coroutines	2755	are ready to run (it's actually more complicated: it only runs coroutines
723	with priority higher than or equal to the event loop and one coroutine	2756	with priority higher than or equal to the event loop and one coroutine
724	of lower priority, but only once, using idle watchers to keep the event	2757	of lower priority, but only once, using idle watchers to keep the event
725	loop from blocking if lower-priority coroutines are active, thus mapping	2758	loop from blocking if lower-priority coroutines are active, thus mapping
726	low-priority coroutines to idle/background tasks).	2759	low-priority coroutines to idle/background tasks).
727		2760
		2761	It is recommended to give C<ev_check> watchers highest (C<EV_MAXPRI>)
		2762	priority, to ensure that they are being run before any other watchers
		2763	after the poll (this doesn't matter for C<ev_prepare> watchers).
		2764
		2765	Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
		2766	activate ("feed") events into libev. While libev fully supports this, they
		2767	might get executed before other C<ev_check> watchers did their job. As
		2768	C<ev_check> watchers are often used to embed other (non-libev) event
		2769	loops those other event loops might be in an unusable state until their
		2770	C<ev_check> watcher ran (always remind yourself to coexist peacefully with
		2771	others).
		2772
		2773	=head3 Watcher-Specific Functions and Data Members
		2774
728	=over 4	2775	=over 4
729		2776
730	=item ev_prepare_init (ev_prepare *, callback)	2777	=item ev_prepare_init (ev_prepare *, callback)
731		2778
732	=item ev_check_init (ev_check *, callback)	2779	=item ev_check_init (ev_check *, callback)
733		2780
734	Initialises and configures the prepare or check watcher - they have no	2781	Initialises and configures the prepare or check watcher - they have no
735	parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>	2782	parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
736	macros, but using them is utterly, utterly and completely pointless.	2783	macros, but using them is utterly, utterly, utterly and completely
		2784	pointless.
737		2785
738	=back	2786	=back
739		2787
		2788	=head3 Examples
		2789
		2790	There are a number of principal ways to embed other event loops or modules
		2791	into libev. Here are some ideas on how to include libadns into libev
		2792	(there is a Perl module named C<EV::ADNS> that does this, which you could
		2793	use as a working example. Another Perl module named C<EV::Glib> embeds a
		2794	Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
		2795	Glib event loop).
		2796
		2797	Method 1: Add IO watchers and a timeout watcher in a prepare handler,
		2798	and in a check watcher, destroy them and call into libadns. What follows
		2799	is pseudo-code only of course. This requires you to either use a low
		2800	priority for the check watcher or use C<ev_clear_pending> explicitly, as
		2801	the callbacks for the IO/timeout watchers might not have been called yet.
		2802
		2803	static ev_io iow [nfd];
		2804	static ev_timer tw;
		2805
		2806	static void
		2807	io_cb (struct ev_loop *loop, ev_io *w, int revents)
		2808	{
		2809	}
		2810
		2811	// create io watchers for each fd and a timer before blocking
		2812	static void
		2813	adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
		2814	{
		2815	int timeout = 3600000;
		2816	struct pollfd fds [nfd];
		2817	// actual code will need to loop here and realloc etc.
		2818	adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
		2819
		2820	/* the callback is illegal, but won't be called as we stop during check */
		2821	ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
		2822	ev_timer_start (loop, &tw);
		2823
		2824	// create one ev_io per pollfd
		2825	for (int i = 0; i < nfd; ++i)
		2826	{
		2827	ev_io_init (iow + i, io_cb, fds [i].fd,
		2828	((fds [i].events & POLLIN ? EV_READ : 0)
		2829	| (fds [i].events & POLLOUT ? EV_WRITE : 0)));
		2830
		2831	fds [i].revents = 0;
		2832	ev_io_start (loop, iow + i);
		2833	}
		2834	}
		2835
		2836	// stop all watchers after blocking
		2837	static void
		2838	adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
		2839	{
		2840	ev_timer_stop (loop, &tw);
		2841
		2842	for (int i = 0; i < nfd; ++i)
		2843	{
		2844	// set the relevant poll flags
		2845	// could also call adns_processreadable etc. here
		2846	struct pollfd *fd = fds + i;
		2847	int revents = ev_clear_pending (iow + i);
		2848	if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
		2849	if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
		2850
		2851	// now stop the watcher
		2852	ev_io_stop (loop, iow + i);
		2853	}
		2854
		2855	adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
		2856	}
		2857
		2858	Method 2: This would be just like method 1, but you run C<adns_afterpoll>
		2859	in the prepare watcher and would dispose of the check watcher.
		2860
		2861	Method 3: If the module to be embedded supports explicit event
		2862	notification (libadns does), you can also make use of the actual watcher
		2863	callbacks, and only destroy/create the watchers in the prepare watcher.
		2864
		2865	static void
		2866	timer_cb (EV_P_ ev_timer *w, int revents)
		2867	{
		2868	adns_state ads = (adns_state)w->data;
		2869	update_now (EV_A);
		2870
		2871	adns_processtimeouts (ads, &tv_now);
		2872	}
		2873
		2874	static void
		2875	io_cb (EV_P_ ev_io *w, int revents)
		2876	{
		2877	adns_state ads = (adns_state)w->data;
		2878	update_now (EV_A);
		2879
		2880	if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
		2881	if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
		2882	}
		2883
		2884	// do not ever call adns_afterpoll
		2885
		2886	Method 4: Do not use a prepare or check watcher because the module you
		2887	want to embed is not flexible enough to support it. Instead, you can
		2888	override their poll function. The drawback with this solution is that the
		2889	main loop is now no longer controllable by EV. The C<Glib::EV> module uses
		2890	this approach, effectively embedding EV as a client into the horrible
		2891	libglib event loop.
		2892
		2893	static gint
		2894	event_poll_func (GPollFD *fds, guint nfds, gint timeout)
		2895	{
		2896	int got_events = 0;
		2897
		2898	for (n = 0; n < nfds; ++n)
		2899	// create/start io watcher that sets the relevant bits in fds[n] and increment got_events
		2900
		2901	if (timeout >= 0)
		2902	// create/start timer
		2903
		2904	// poll
		2905	ev_run (EV_A_ 0);
		2906
		2907	// stop timer again
		2908	if (timeout >= 0)
		2909	ev_timer_stop (EV_A_ &to);
		2910
		2911	// stop io watchers again - their callbacks should have set
		2912	for (n = 0; n < nfds; ++n)
		2913	ev_io_stop (EV_A_ iow [n]);
		2914
		2915	return got_events;
		2916	}
		2917
		2918
		2919	=head2 C<ev_embed> - when one backend isn't enough...
		2920
		2921	This is a rather advanced watcher type that lets you embed one event loop
		2922	into another (currently only C<ev_io> events are supported in the embedded
		2923	loop, other types of watchers might be handled in a delayed or incorrect
		2924	fashion and must not be used).
		2925
		2926	There are primarily two reasons you would want that: work around bugs and
		2927	prioritise I/O.
		2928
		2929	As an example for a bug workaround, the kqueue backend might only support
		2930	sockets on some platform, so it is unusable as generic backend, but you
		2931	still want to make use of it because you have many sockets and it scales
		2932	so nicely. In this case, you would create a kqueue-based loop and embed
		2933	it into your default loop (which might use e.g. poll). Overall operation
		2934	will be a bit slower because first libev has to call C<poll> and then
		2935	C<kevent>, but at least you can use both mechanisms for what they are
		2936	best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
		2937
		2938	As for prioritising I/O: under rare circumstances you have the case where
		2939	some fds have to be watched and handled very quickly (with low latency),
		2940	and even priorities and idle watchers might have too much overhead. In
		2941	this case you would put all the high priority stuff in one loop and all
		2942	the rest in a second one, and embed the second one in the first.
		2943
		2944	As long as the watcher is active, the callback will be invoked every
		2945	time there might be events pending in the embedded loop. The callback
		2946	must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
		2947	sweep and invoke their callbacks (the callback doesn't need to invoke the
		2948	C<ev_embed_sweep> function directly, it could also start an idle watcher
		2949	to give the embedded loop strictly lower priority for example).
		2950
		2951	You can also set the callback to C<0>, in which case the embed watcher
		2952	will automatically execute the embedded loop sweep whenever necessary.
		2953
		2954	Fork detection will be handled transparently while the C<ev_embed> watcher
		2955	is active, i.e., the embedded loop will automatically be forked when the
		2956	embedding loop forks. In other cases, the user is responsible for calling
		2957	C<ev_loop_fork> on the embedded loop.
		2958
		2959	Unfortunately, not all backends are embeddable: only the ones returned by
		2960	C<ev_embeddable_backends> are, which, unfortunately, does not include any
		2961	portable one.
		2962
		2963	So when you want to use this feature you will always have to be prepared
		2964	that you cannot get an embeddable loop. The recommended way to get around
		2965	this is to have a separate variables for your embeddable loop, try to
		2966	create it, and if that fails, use the normal loop for everything.
		2967
		2968	=head3 C<ev_embed> and fork
		2969
		2970	While the C<ev_embed> watcher is running, forks in the embedding loop will
		2971	automatically be applied to the embedded loop as well, so no special
		2972	fork handling is required in that case. When the watcher is not running,
		2973	however, it is still the task of the libev user to call C<ev_loop_fork ()>
		2974	as applicable.
		2975
		2976	=head3 Watcher-Specific Functions and Data Members
		2977
		2978	=over 4
		2979
		2980	=item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
		2981
		2982	=item ev_embed_set (ev_embed *, callback, struct ev_loop *embedded_loop)
		2983
		2984	Configures the watcher to embed the given loop, which must be
		2985	embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
		2986	invoked automatically, otherwise it is the responsibility of the callback
		2987	to invoke it (it will continue to be called until the sweep has been done,
		2988	if you do not want that, you need to temporarily stop the embed watcher).
		2989
		2990	=item ev_embed_sweep (loop, ev_embed *)
		2991
		2992	Make a single, non-blocking sweep over the embedded loop. This works
		2993	similarly to C<ev_run (embedded_loop, EVRUN_NOWAIT)>, but in the most
		2994	appropriate way for embedded loops.
		2995
		2996	=item struct ev_loop *other [read-only]
		2997
		2998	The embedded event loop.
		2999
		3000	=back
		3001
		3002	=head3 Examples
		3003
		3004	Example: Try to get an embeddable event loop and embed it into the default
		3005	event loop. If that is not possible, use the default loop. The default
		3006	loop is stored in C<loop_hi>, while the embeddable loop is stored in
		3007	C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
		3008	used).
		3009
		3010	struct ev_loop *loop_hi = ev_default_init (0);
		3011	struct ev_loop *loop_lo = 0;
		3012	ev_embed embed;
		3013
		3014	// see if there is a chance of getting one that works
		3015	// (remember that a flags value of 0 means autodetection)
		3016	loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
		3017	? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
		3018	: 0;
		3019
		3020	// if we got one, then embed it, otherwise default to loop_hi
		3021	if (loop_lo)
		3022	{
		3023	ev_embed_init (&embed, 0, loop_lo);
		3024	ev_embed_start (loop_hi, &embed);
		3025	}
		3026	else
		3027	loop_lo = loop_hi;
		3028
		3029	Example: Check if kqueue is available but not recommended and create
		3030	a kqueue backend for use with sockets (which usually work with any
		3031	kqueue implementation). Store the kqueue/socket-only event loop in
		3032	C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
		3033
		3034	struct ev_loop *loop = ev_default_init (0);
		3035	struct ev_loop *loop_socket = 0;
		3036	ev_embed embed;
		3037
		3038	if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
		3039	if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
		3040	{
		3041	ev_embed_init (&embed, 0, loop_socket);
		3042	ev_embed_start (loop, &embed);
		3043	}
		3044
		3045	if (!loop_socket)
		3046	loop_socket = loop;
		3047
		3048	// now use loop_socket for all sockets, and loop for everything else
		3049
		3050
		3051	=head2 C<ev_fork> - the audacity to resume the event loop after a fork
		3052
		3053	Fork watchers are called when a C<fork ()> was detected (usually because
		3054	whoever is a good citizen cared to tell libev about it by calling
		3055	C<ev_default_fork> or C<ev_loop_fork>). The invocation is done before the
		3056	event loop blocks next and before C<ev_check> watchers are being called,
		3057	and only in the child after the fork. If whoever good citizen calling
		3058	C<ev_default_fork> cheats and calls it in the wrong process, the fork
		3059	handlers will be invoked, too, of course.
		3060
		3061	=head3 The special problem of life after fork - how is it possible?
		3062
		3063	Most uses of C<fork()> consist of forking, then some simple calls to set
		3064	up/change the process environment, followed by a call to C<exec()>. This
		3065	sequence should be handled by libev without any problems.
		3066
		3067	This changes when the application actually wants to do event handling
		3068	in the child, or both parent in child, in effect "continuing" after the
		3069	fork.
		3070
		3071	The default mode of operation (for libev, with application help to detect
		3072	forks) is to duplicate all the state in the child, as would be expected
		3073	when I<either> the parent I<or> the child process continues.
		3074
		3075	When both processes want to continue using libev, then this is usually the
		3076	wrong result. In that case, usually one process (typically the parent) is
		3077	supposed to continue with all watchers in place as before, while the other
		3078	process typically wants to start fresh, i.e. without any active watchers.
		3079
		3080	The cleanest and most efficient way to achieve that with libev is to
		3081	simply create a new event loop, which of course will be "empty", and
		3082	use that for new watchers. This has the advantage of not touching more
		3083	memory than necessary, and thus avoiding the copy-on-write, and the
		3084	disadvantage of having to use multiple event loops (which do not support
		3085	signal watchers).
		3086
		3087	When this is not possible, or you want to use the default loop for
		3088	other reasons, then in the process that wants to start "fresh", call
		3089	C<ev_loop_destroy (EV_DEFAULT)> followed by C<ev_default_loop (...)>.
		3090	Destroying the default loop will "orphan" (not stop) all registered
		3091	watchers, so you have to be careful not to execute code that modifies
		3092	those watchers. Note also that in that case, you have to re-register any
		3093	signal watchers.
		3094
		3095	=head3 Watcher-Specific Functions and Data Members
		3096
		3097	=over 4
		3098
		3099	=item ev_fork_init (ev_fork *, callback)
		3100
		3101	Initialises and configures the fork watcher - it has no parameters of any
		3102	kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
		3103	really.
		3104
		3105	=back
		3106
		3107
		3108	=head2 C<ev_cleanup> - even the best things end
		3109
		3110	Cleanup watchers are called just before the event loop is being destroyed
		3111	by a call to C<ev_loop_destroy>.
		3112
		3113	While there is no guarantee that the event loop gets destroyed, cleanup
		3114	watchers provide a convenient method to install cleanup hooks for your
		3115	program, worker threads and so on - you just to make sure to destroy the
		3116	loop when you want them to be invoked.
		3117
		3118	Cleanup watchers are invoked in the same way as any other watcher. Unlike
		3119	all other watchers, they do not keep a reference to the event loop (which
		3120	makes a lot of sense if you think about it). Like all other watchers, you
		3121	can call libev functions in the callback, except C<ev_cleanup_start>.
		3122
		3123	=head3 Watcher-Specific Functions and Data Members
		3124
		3125	=over 4
		3126
		3127	=item ev_cleanup_init (ev_cleanup *, callback)
		3128
		3129	Initialises and configures the cleanup watcher - it has no parameters of
		3130	any kind. There is a C<ev_cleanup_set> macro, but using it is utterly
		3131	pointless, I assure you.
		3132
		3133	=back
		3134
		3135	Example: Register an atexit handler to destroy the default loop, so any
		3136	cleanup functions are called.
		3137
		3138	static void
		3139	program_exits (void)
		3140	{
		3141	ev_loop_destroy (EV_DEFAULT_UC);
		3142	}
		3143
		3144	...
		3145	atexit (program_exits);
		3146
		3147
		3148	=head2 C<ev_async> - how to wake up an event loop
		3149
		3150	In general, you cannot use an C<ev_run> from multiple threads or other
		3151	asynchronous sources such as signal handlers (as opposed to multiple event
		3152	loops - those are of course safe to use in different threads).
		3153
		3154	Sometimes, however, you need to wake up an event loop you do not control,
		3155	for example because it belongs to another thread. This is what C<ev_async>
		3156	watchers do: as long as the C<ev_async> watcher is active, you can signal
		3157	it by calling C<ev_async_send>, which is thread- and signal safe.
		3158
		3159	This functionality is very similar to C<ev_signal> watchers, as signals,
		3160	too, are asynchronous in nature, and signals, too, will be compressed
		3161	(i.e. the number of callback invocations may be less than the number of
		3162	C<ev_async_sent> calls).
		3163
		3164	Unlike C<ev_signal> watchers, C<ev_async> works with any event loop, not
		3165	just the default loop.
		3166
		3167	=head3 Queueing
		3168
		3169	C<ev_async> does not support queueing of data in any way. The reason
		3170	is that the author does not know of a simple (or any) algorithm for a
		3171	multiple-writer-single-reader queue that works in all cases and doesn't
		3172	need elaborate support such as pthreads or unportable memory access
		3173	semantics.
		3174
		3175	That means that if you want to queue data, you have to provide your own
		3176	queue. But at least I can tell you how to implement locking around your
		3177	queue:
		3178
		3179	=over 4
		3180
		3181	=item queueing from a signal handler context
		3182
		3183	To implement race-free queueing, you simply add to the queue in the signal
		3184	handler but you block the signal handler in the watcher callback. Here is
		3185	an example that does that for some fictitious SIGUSR1 handler:
		3186
		3187	static ev_async mysig;
		3188
		3189	static void
		3190	sigusr1_handler (void)
		3191	{
		3192	sometype data;
		3193
		3194	// no locking etc.
		3195	queue_put (data);
		3196	ev_async_send (EV_DEFAULT_ &mysig);
		3197	}
		3198
		3199	static void
		3200	mysig_cb (EV_P_ ev_async *w, int revents)
		3201	{
		3202	sometype data;
		3203	sigset_t block, prev;
		3204
		3205	sigemptyset (&block);
		3206	sigaddset (&block, SIGUSR1);
		3207	sigprocmask (SIG_BLOCK, &block, &prev);
		3208
		3209	while (queue_get (&data))
		3210	process (data);
		3211
		3212	if (sigismember (&prev, SIGUSR1)
		3213	sigprocmask (SIG_UNBLOCK, &block, 0);
		3214	}
		3215
		3216	(Note: pthreads in theory requires you to use C<pthread_setmask>
		3217	instead of C<sigprocmask> when you use threads, but libev doesn't do it
		3218	either...).
		3219
		3220	=item queueing from a thread context
		3221
		3222	The strategy for threads is different, as you cannot (easily) block
		3223	threads but you can easily preempt them, so to queue safely you need to
		3224	employ a traditional mutex lock, such as in this pthread example:
		3225
		3226	static ev_async mysig;
		3227	static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
		3228
		3229	static void
		3230	otherthread (void)
		3231	{
		3232	// only need to lock the actual queueing operation
		3233	pthread_mutex_lock (&mymutex);
		3234	queue_put (data);
		3235	pthread_mutex_unlock (&mymutex);
		3236
		3237	ev_async_send (EV_DEFAULT_ &mysig);
		3238	}
		3239
		3240	static void
		3241	mysig_cb (EV_P_ ev_async *w, int revents)
		3242	{
		3243	pthread_mutex_lock (&mymutex);
		3244
		3245	while (queue_get (&data))
		3246	process (data);
		3247
		3248	pthread_mutex_unlock (&mymutex);
		3249	}
		3250
		3251	=back
		3252
		3253
		3254	=head3 Watcher-Specific Functions and Data Members
		3255
		3256	=over 4
		3257
		3258	=item ev_async_init (ev_async *, callback)
		3259
		3260	Initialises and configures the async watcher - it has no parameters of any
		3261	kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
		3262	trust me.
		3263
		3264	=item ev_async_send (loop, ev_async *)
		3265
		3266	Sends/signals/activates the given C<ev_async> watcher, that is, feeds
		3267	an C<EV_ASYNC> event on the watcher into the event loop. Unlike
		3268	C<ev_feed_event>, this call is safe to do from other threads, signal or
		3269	similar contexts (see the discussion of C<EV_ATOMIC_T> in the embedding
		3270	section below on what exactly this means).
		3271
		3272	Note that, as with other watchers in libev, multiple events might get
		3273	compressed into a single callback invocation (another way to look at this
		3274	is that C<ev_async> watchers are level-triggered, set on C<ev_async_send>,
		3275	reset when the event loop detects that).
		3276
		3277	This call incurs the overhead of a system call only once per event loop
		3278	iteration, so while the overhead might be noticeable, it doesn't apply to
		3279	repeated calls to C<ev_async_send> for the same event loop.
		3280
		3281	=item bool = ev_async_pending (ev_async *)
		3282
		3283	Returns a non-zero value when C<ev_async_send> has been called on the
		3284	watcher but the event has not yet been processed (or even noted) by the
		3285	event loop.
		3286
		3287	C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
		3288	the loop iterates next and checks for the watcher to have become active,
		3289	it will reset the flag again. C<ev_async_pending> can be used to very
		3290	quickly check whether invoking the loop might be a good idea.
		3291
		3292	Not that this does I<not> check whether the watcher itself is pending,
		3293	only whether it has been requested to make this watcher pending: there
		3294	is a time window between the event loop checking and resetting the async
		3295	notification, and the callback being invoked.
		3296
		3297	=back
		3298
		3299
740	=head1 OTHER FUNCTIONS	3300	=head1 OTHER FUNCTIONS
741		3301
742	There are some other functions of possible interest. Described. Here. Now.	3302	There are some other functions of possible interest. Described. Here. Now.
743		3303
744	=over 4	3304	=over 4
745		3305
746	=item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)	3306	=item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)
747		3307
748	This function combines a simple timer and an I/O watcher, calls your	3308	This function combines a simple timer and an I/O watcher, calls your
749	callback on whichever event happens first and automatically stop both	3309	callback on whichever event happens first and automatically stops both
750	watchers. This is useful if you want to wait for a single event on an fd	3310	watchers. This is useful if you want to wait for a single event on an fd
751	or timeout without having to allocate/configure/start/stop/free one or	3311	or timeout without having to allocate/configure/start/stop/free one or
752	more watchers yourself.	3312	more watchers yourself.
753		3313
754	If C<fd> is less than 0, then no I/O watcher will be started and events	3314	If C<fd> is less than 0, then no I/O watcher will be started and the
755	is being ignored. Otherwise, an C<ev_io> watcher for the given C<fd> and	3315	C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
756	C<events> set will be craeted and started.	3316	the given C<fd> and C<events> set will be created and started.
757		3317
758	If C<timeout> is less than 0, then no timeout watcher will be	3318	If C<timeout> is less than 0, then no timeout watcher will be
759	started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and	3319	started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
760	repeat = 0) will be started. While C<0> is a valid timeout, it is of	3320	repeat = 0) will be started. C<0> is a valid timeout.
761	dubious value.
762		3321
763	The callback has the type C<void (*cb)(int revents, void *arg)> and gets	3322	The callback has the type C<void (*cb)(int revents, void *arg)> and is
764	passed an C<revents> set like normal event callbacks (a combination of	3323	passed an C<revents> set like normal event callbacks (a combination of
765	C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMEOUT>) and the C<arg>	3324	C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMER>) and the C<arg>
766	value passed to C<ev_once>:	3325	value passed to C<ev_once>. Note that it is possible to receive I<both>
		3326	a timeout and an io event at the same time - you probably should give io
		3327	events precedence.
767		3328
		3329	Example: wait up to ten seconds for data to appear on STDIN_FILENO.
		3330
768	static void stdin_ready (int revents, void *arg)	3331	static void stdin_ready (int revents, void *arg)
769	{	3332	{
770	if (revents & EV_TIMEOUT)
771	/* doh, nothing entered */;
772	else if (revents & EV_READ)	3333	if (revents & EV_READ)
773	/* stdin might have data for us, joy! */;	3334	/* stdin might have data for us, joy! */;
		3335	else if (revents & EV_TIMER)
		3336	/* doh, nothing entered */;
774	}	3337	}
775		3338
776	ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);	3339	ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
777
778	=item ev_feed_event (loop, watcher, int events)
779
780	Feeds the given event set into the event loop, as if the specified event
781	had happened for the specified watcher (which must be a pointer to an
782	initialised but not necessarily started event watcher).
783		3340
784	=item ev_feed_fd_event (loop, int fd, int revents)	3341	=item ev_feed_fd_event (loop, int fd, int revents)
785		3342
786	Feed an event on the given fd, as if a file descriptor backend detected	3343	Feed an event on the given fd, as if a file descriptor backend detected
787	the given events it.	3344	the given events it.
788		3345
789	=item ev_feed_signal_event (loop, int signum)	3346	=item ev_feed_signal_event (loop, int signum)
790		3347
791	Feed an event as if the given signal occured (loop must be the default loop!).	3348	Feed an event as if the given signal occurred (C<loop> must be the default
		3349	loop!).
792		3350
793	=back	3351	=back
		3352
794		3353
795	=head1 LIBEVENT EMULATION	3354	=head1 LIBEVENT EMULATION
796		3355
797	Libev offers a compatibility emulation layer for libevent. It cannot	3356	Libev offers a compatibility emulation layer for libevent. It cannot
798	emulate the internals of libevent, so here are some usage hints:	3357	emulate the internals of libevent, so here are some usage hints:
…		…
810		3369
811	=item * Priorities are not currently supported. Initialising priorities	3370	=item * Priorities are not currently supported. Initialising priorities
812	will fail and all watchers will have the same priority, even though there	3371	will fail and all watchers will have the same priority, even though there
813	is an ev_pri field.	3372	is an ev_pri field.
814		3373
		3374	=item * In libevent, the last base created gets the signals, in libev, the
		3375	first base created (== the default loop) gets the signals.
		3376
815	=item * Other members are not supported.	3377	=item * Other members are not supported.
816		3378
817	=item * The libev emulation is I<not> ABI compatible to libevent, you need	3379	=item * The libev emulation is I<not> ABI compatible to libevent, you need
818	to use the libev header file and library.	3380	to use the libev header file and library.
819		3381
820	=back	3382	=back
821		3383
822	=head1 C++ SUPPORT	3384	=head1 C++ SUPPORT
823		3385
824	TBD.	3386	Libev comes with some simplistic wrapper classes for C++ that mainly allow
		3387	you to use some convenience methods to start/stop watchers and also change
		3388	the callback model to a model using method callbacks on objects.
		3389
		3390	To use it,
		3391
		3392	#include <ev++.h>
		3393
		3394	This automatically includes F<ev.h> and puts all of its definitions (many
		3395	of them macros) into the global namespace. All C++ specific things are
		3396	put into the C<ev> namespace. It should support all the same embedding
		3397	options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
		3398
		3399	Care has been taken to keep the overhead low. The only data member the C++
		3400	classes add (compared to plain C-style watchers) is the event loop pointer
		3401	that the watcher is associated with (or no additional members at all if
		3402	you disable C<EV_MULTIPLICITY> when embedding libev).
		3403
		3404	Currently, functions, and static and non-static member functions can be
		3405	used as callbacks. Other types should be easy to add as long as they only
		3406	need one additional pointer for context. If you need support for other
		3407	types of functors please contact the author (preferably after implementing
		3408	it).
		3409
		3410	Here is a list of things available in the C<ev> namespace:
		3411
		3412	=over 4
		3413
		3414	=item C<ev::READ>, C<ev::WRITE> etc.
		3415
		3416	These are just enum values with the same values as the C<EV_READ> etc.
		3417	macros from F<ev.h>.
		3418
		3419	=item C<ev::tstamp>, C<ev::now>
		3420
		3421	Aliases to the same types/functions as with the C<ev_> prefix.
		3422
		3423	=item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
		3424
		3425	For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
		3426	the same name in the C<ev> namespace, with the exception of C<ev_signal>
		3427	which is called C<ev::sig> to avoid clashes with the C<signal> macro
		3428	defines by many implementations.
		3429
		3430	All of those classes have these methods:
		3431
		3432	=over 4
		3433
		3434	=item ev::TYPE::TYPE ()
		3435
		3436	=item ev::TYPE::TYPE (loop)
		3437
		3438	=item ev::TYPE::~TYPE
		3439
		3440	The constructor (optionally) takes an event loop to associate the watcher
		3441	with. If it is omitted, it will use C<EV_DEFAULT>.
		3442
		3443	The constructor calls C<ev_init> for you, which means you have to call the
		3444	C<set> method before starting it.
		3445
		3446	It will not set a callback, however: You have to call the templated C<set>
		3447	method to set a callback before you can start the watcher.
		3448
		3449	(The reason why you have to use a method is a limitation in C++ which does
		3450	not allow explicit template arguments for constructors).
		3451
		3452	The destructor automatically stops the watcher if it is active.
		3453
		3454	=item w->set<class, &class::method> (object *)
		3455
		3456	This method sets the callback method to call. The method has to have a
		3457	signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
		3458	first argument and the C<revents> as second. The object must be given as
		3459	parameter and is stored in the C<data> member of the watcher.
		3460
		3461	This method synthesizes efficient thunking code to call your method from
		3462	the C callback that libev requires. If your compiler can inline your
		3463	callback (i.e. it is visible to it at the place of the C<set> call and
		3464	your compiler is good :), then the method will be fully inlined into the
		3465	thunking function, making it as fast as a direct C callback.
		3466
		3467	Example: simple class declaration and watcher initialisation
		3468
		3469	struct myclass
		3470	{
		3471	void io_cb (ev::io &w, int revents) { }
		3472	}
		3473
		3474	myclass obj;
		3475	ev::io iow;
		3476	iow.set <myclass, &myclass::io_cb> (&obj);
		3477
		3478	=item w->set (object *)
		3479
		3480	This is a variation of a method callback - leaving out the method to call
		3481	will default the method to C<operator ()>, which makes it possible to use
		3482	functor objects without having to manually specify the C<operator ()> all
		3483	the time. Incidentally, you can then also leave out the template argument
		3484	list.
		3485
		3486	The C<operator ()> method prototype must be C<void operator ()(watcher &w,
		3487	int revents)>.
		3488
		3489	See the method-C<set> above for more details.
		3490
		3491	Example: use a functor object as callback.
		3492
		3493	struct myfunctor
		3494	{
		3495	void operator() (ev::io &w, int revents)
		3496	{
		3497	...
		3498	}
		3499	}
		3500
		3501	myfunctor f;
		3502
		3503	ev::io w;
		3504	w.set (&f);
		3505
		3506	=item w->set<function> (void *data = 0)
		3507
		3508	Also sets a callback, but uses a static method or plain function as
		3509	callback. The optional C<data> argument will be stored in the watcher's
		3510	C<data> member and is free for you to use.
		3511
		3512	The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
		3513
		3514	See the method-C<set> above for more details.
		3515
		3516	Example: Use a plain function as callback.
		3517
		3518	static void io_cb (ev::io &w, int revents) { }
		3519	iow.set <io_cb> ();
		3520
		3521	=item w->set (loop)
		3522
		3523	Associates a different C<struct ev_loop> with this watcher. You can only
		3524	do this when the watcher is inactive (and not pending either).
		3525
		3526	=item w->set ([arguments])
		3527
		3528	Basically the same as C<ev_TYPE_set>, with the same arguments. Either this
		3529	method or a suitable start method must be called at least once. Unlike the
		3530	C counterpart, an active watcher gets automatically stopped and restarted
		3531	when reconfiguring it with this method.
		3532
		3533	=item w->start ()
		3534
		3535	Starts the watcher. Note that there is no C<loop> argument, as the
		3536	constructor already stores the event loop.
		3537
		3538	=item w->start ([arguments])
		3539
		3540	Instead of calling C<set> and C<start> methods separately, it is often
		3541	convenient to wrap them in one call. Uses the same type of arguments as
		3542	the configure C<set> method of the watcher.
		3543
		3544	=item w->stop ()
		3545
		3546	Stops the watcher if it is active. Again, no C<loop> argument.
		3547
		3548	=item w->again () (C<ev::timer>, C<ev::periodic> only)
		3549
		3550	For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
		3551	C<ev_TYPE_again> function.
		3552
		3553	=item w->sweep () (C<ev::embed> only)
		3554
		3555	Invokes C<ev_embed_sweep>.
		3556
		3557	=item w->update () (C<ev::stat> only)
		3558
		3559	Invokes C<ev_stat_stat>.
		3560
		3561	=back
		3562
		3563	=back
		3564
		3565	Example: Define a class with two I/O and idle watchers, start the I/O
		3566	watchers in the constructor.
		3567
		3568	class myclass
		3569	{
		3570	ev::io io ; void io_cb (ev::io &w, int revents);
		3571	ev::io2 io2 ; void io2_cb (ev::io &w, int revents);
		3572	ev::idle idle; void idle_cb (ev::idle &w, int revents);
		3573
		3574	myclass (int fd)
		3575	{
		3576	io .set <myclass, &myclass::io_cb > (this);
		3577	io2 .set <myclass, &myclass::io2_cb > (this);
		3578	idle.set <myclass, &myclass::idle_cb> (this);
		3579
		3580	io.set (fd, ev::WRITE); // configure the watcher
		3581	io.start (); // start it whenever convenient
		3582
		3583	io2.start (fd, ev::READ); // set + start in one call
		3584	}
		3585	};
		3586
		3587
		3588	=head1 OTHER LANGUAGE BINDINGS
		3589
		3590	Libev does not offer other language bindings itself, but bindings for a
		3591	number of languages exist in the form of third-party packages. If you know
		3592	any interesting language binding in addition to the ones listed here, drop
		3593	me a note.
		3594
		3595	=over 4
		3596
		3597	=item Perl
		3598
		3599	The EV module implements the full libev API and is actually used to test
		3600	libev. EV is developed together with libev. Apart from the EV core module,
		3601	there are additional modules that implement libev-compatible interfaces
		3602	to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
		3603	C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
		3604	and C<EV::Glib>).
		3605
		3606	It can be found and installed via CPAN, its homepage is at
		3607	L<http://software.schmorp.de/pkg/EV>.
		3608
		3609	=item Python
		3610
		3611	Python bindings can be found at L<http://code.google.com/p/pyev/>. It
		3612	seems to be quite complete and well-documented.
		3613
		3614	=item Ruby
		3615
		3616	Tony Arcieri has written a ruby extension that offers access to a subset
		3617	of the libev API and adds file handle abstractions, asynchronous DNS and
		3618	more on top of it. It can be found via gem servers. Its homepage is at
		3619	L<http://rev.rubyforge.org/>.
		3620
		3621	Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
		3622	makes rev work even on mingw.
		3623
		3624	=item Haskell
		3625
		3626	A haskell binding to libev is available at
		3627	L<http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
		3628
		3629	=item D
		3630
		3631	Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
		3632	be found at L<http://proj.llucax.com.ar/wiki/evd>.
		3633
		3634	=item Ocaml
		3635
		3636	Erkki Seppala has written Ocaml bindings for libev, to be found at
		3637	L<http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
		3638
		3639	=item Lua
		3640
		3641	Brian Maher has written a partial interface to libev for lua (at the
		3642	time of this writing, only C<ev_io> and C<ev_timer>), to be found at
		3643	L<http://github.com/brimworks/lua-ev>.
		3644
		3645	=back
		3646
		3647
		3648	=head1 MACRO MAGIC
		3649
		3650	Libev can be compiled with a variety of options, the most fundamental
		3651	of which is C<EV_MULTIPLICITY>. This option determines whether (most)
		3652	functions and callbacks have an initial C<struct ev_loop *> argument.
		3653
		3654	To make it easier to write programs that cope with either variant, the
		3655	following macros are defined:
		3656
		3657	=over 4
		3658
		3659	=item C<EV_A>, C<EV_A_>
		3660
		3661	This provides the loop I<argument> for functions, if one is required ("ev
		3662	loop argument"). The C<EV_A> form is used when this is the sole argument,
		3663	C<EV_A_> is used when other arguments are following. Example:
		3664
		3665	ev_unref (EV_A);
		3666	ev_timer_add (EV_A_ watcher);
		3667	ev_run (EV_A_ 0);
		3668
		3669	It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
		3670	which is often provided by the following macro.
		3671
		3672	=item C<EV_P>, C<EV_P_>
		3673
		3674	This provides the loop I<parameter> for functions, if one is required ("ev
		3675	loop parameter"). The C<EV_P> form is used when this is the sole parameter,
		3676	C<EV_P_> is used when other parameters are following. Example:
		3677
		3678	// this is how ev_unref is being declared
		3679	static void ev_unref (EV_P);
		3680
		3681	// this is how you can declare your typical callback
		3682	static void cb (EV_P_ ev_timer *w, int revents)
		3683
		3684	It declares a parameter C<loop> of type C<struct ev_loop *>, quite
		3685	suitable for use with C<EV_A>.
		3686
		3687	=item C<EV_DEFAULT>, C<EV_DEFAULT_>
		3688
		3689	Similar to the other two macros, this gives you the value of the default
		3690	loop, if multiple loops are supported ("ev loop default").
		3691
		3692	=item C<EV_DEFAULT_UC>, C<EV_DEFAULT_UC_>
		3693
		3694	Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
		3695	default loop has been initialised (C<UC> == unchecked). Their behaviour
		3696	is undefined when the default loop has not been initialised by a previous
		3697	execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
		3698
		3699	It is often prudent to use C<EV_DEFAULT> when initialising the first
		3700	watcher in a function but use C<EV_DEFAULT_UC> afterwards.
		3701
		3702	=back
		3703
		3704	Example: Declare and initialise a check watcher, utilising the above
		3705	macros so it will work regardless of whether multiple loops are supported
		3706	or not.
		3707
		3708	static void
		3709	check_cb (EV_P_ ev_timer *w, int revents)
		3710	{
		3711	ev_check_stop (EV_A_ w);
		3712	}
		3713
		3714	ev_check check;
		3715	ev_check_init (&check, check_cb);
		3716	ev_check_start (EV_DEFAULT_ &check);
		3717	ev_run (EV_DEFAULT_ 0);
		3718
		3719	=head1 EMBEDDING
		3720
		3721	Libev can (and often is) directly embedded into host
		3722	applications. Examples of applications that embed it include the Deliantra
		3723	Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
		3724	and rxvt-unicode.
		3725
		3726	The goal is to enable you to just copy the necessary files into your
		3727	source directory without having to change even a single line in them, so
		3728	you can easily upgrade by simply copying (or having a checked-out copy of
		3729	libev somewhere in your source tree).
		3730
		3731	=head2 FILESETS
		3732
		3733	Depending on what features you need you need to include one or more sets of files
		3734	in your application.
		3735
		3736	=head3 CORE EVENT LOOP
		3737
		3738	To include only the libev core (all the C<ev_*> functions), with manual
		3739	configuration (no autoconf):
		3740
		3741	#define EV_STANDALONE 1
		3742	#include "ev.c"
		3743
		3744	This will automatically include F<ev.h>, too, and should be done in a
		3745	single C source file only to provide the function implementations. To use
		3746	it, do the same for F<ev.h> in all files wishing to use this API (best
		3747	done by writing a wrapper around F<ev.h> that you can include instead and
		3748	where you can put other configuration options):
		3749
		3750	#define EV_STANDALONE 1
		3751	#include "ev.h"
		3752
		3753	Both header files and implementation files can be compiled with a C++
		3754	compiler (at least, that's a stated goal, and breakage will be treated
		3755	as a bug).
		3756
		3757	You need the following files in your source tree, or in a directory
		3758	in your include path (e.g. in libev/ when using -Ilibev):
		3759
		3760	ev.h
		3761	ev.c
		3762	ev_vars.h
		3763	ev_wrap.h
		3764
		3765	ev_win32.c required on win32 platforms only
		3766
		3767	ev_select.c only when select backend is enabled (which is enabled by default)
		3768	ev_poll.c only when poll backend is enabled (disabled by default)
		3769	ev_epoll.c only when the epoll backend is enabled (disabled by default)
		3770	ev_kqueue.c only when the kqueue backend is enabled (disabled by default)
		3771	ev_port.c only when the solaris port backend is enabled (disabled by default)
		3772
		3773	F<ev.c> includes the backend files directly when enabled, so you only need
		3774	to compile this single file.
		3775
		3776	=head3 LIBEVENT COMPATIBILITY API
		3777
		3778	To include the libevent compatibility API, also include:
		3779
		3780	#include "event.c"
		3781
		3782	in the file including F<ev.c>, and:
		3783
		3784	#include "event.h"
		3785
		3786	in the files that want to use the libevent API. This also includes F<ev.h>.
		3787
		3788	You need the following additional files for this:
		3789
		3790	event.h
		3791	event.c
		3792
		3793	=head3 AUTOCONF SUPPORT
		3794
		3795	Instead of using C<EV_STANDALONE=1> and providing your configuration in
		3796	whatever way you want, you can also C<m4_include([libev.m4])> in your
		3797	F<configure.ac> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
		3798	include F<config.h> and configure itself accordingly.
		3799
		3800	For this of course you need the m4 file:
		3801
		3802	libev.m4
		3803
		3804	=head2 PREPROCESSOR SYMBOLS/MACROS
		3805
		3806	Libev can be configured via a variety of preprocessor symbols you have to
		3807	define before including (or compiling) any of its files. The default in
		3808	the absence of autoconf is documented for every option.
		3809
		3810	Symbols marked with "(h)" do not change the ABI, and can have different
		3811	values when compiling libev vs. including F<ev.h>, so it is permissible
		3812	to redefine them before including F<ev.h> without breaking compatibility
		3813	to a compiled library. All other symbols change the ABI, which means all
		3814	users of libev and the libev code itself must be compiled with compatible
		3815	settings.
		3816
		3817	=over 4
		3818
		3819	=item EV_COMPAT3 (h)
		3820
		3821	Backwards compatibility is a major concern for libev. This is why this
		3822	release of libev comes with wrappers for the functions and symbols that
		3823	have been renamed between libev version 3 and 4.
		3824
		3825	You can disable these wrappers (to test compatibility with future
		3826	versions) by defining C<EV_COMPAT3> to C<0> when compiling your
		3827	sources. This has the additional advantage that you can drop the C<struct>
		3828	from C<struct ev_loop> declarations, as libev will provide an C<ev_loop>
		3829	typedef in that case.
		3830
		3831	In some future version, the default for C<EV_COMPAT3> will become C<0>,
		3832	and in some even more future version the compatibility code will be
		3833	removed completely.
		3834
		3835	=item EV_STANDALONE (h)
		3836
		3837	Must always be C<1> if you do not use autoconf configuration, which
		3838	keeps libev from including F<config.h>, and it also defines dummy
		3839	implementations for some libevent functions (such as logging, which is not
		3840	supported). It will also not define any of the structs usually found in
		3841	F<event.h> that are not directly supported by the libev core alone.
		3842
		3843	In standalone mode, libev will still try to automatically deduce the
		3844	configuration, but has to be more conservative.
		3845
		3846	=item EV_USE_MONOTONIC
		3847
		3848	If defined to be C<1>, libev will try to detect the availability of the
		3849	monotonic clock option at both compile time and runtime. Otherwise no
		3850	use of the monotonic clock option will be attempted. If you enable this,
		3851	you usually have to link against librt or something similar. Enabling it
		3852	when the functionality isn't available is safe, though, although you have
		3853	to make sure you link against any libraries where the C<clock_gettime>
		3854	function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
		3855
		3856	=item EV_USE_REALTIME
		3857
		3858	If defined to be C<1>, libev will try to detect the availability of the
		3859	real-time clock option at compile time (and assume its availability
		3860	at runtime if successful). Otherwise no use of the real-time clock
		3861	option will be attempted. This effectively replaces C<gettimeofday>
		3862	by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
		3863	correctness. See the note about libraries in the description of
		3864	C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
		3865	C<EV_USE_CLOCK_SYSCALL>.
		3866
		3867	=item EV_USE_CLOCK_SYSCALL
		3868
		3869	If defined to be C<1>, libev will try to use a direct syscall instead
		3870	of calling the system-provided C<clock_gettime> function. This option
		3871	exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
		3872	unconditionally pulls in C<libpthread>, slowing down single-threaded
		3873	programs needlessly. Using a direct syscall is slightly slower (in
		3874	theory), because no optimised vdso implementation can be used, but avoids
		3875	the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
		3876	higher, as it simplifies linking (no need for C<-lrt>).
		3877
		3878	=item EV_USE_NANOSLEEP
		3879
		3880	If defined to be C<1>, libev will assume that C<nanosleep ()> is available
		3881	and will use it for delays. Otherwise it will use C<select ()>.
		3882
		3883	=item EV_USE_EVENTFD
		3884
		3885	If defined to be C<1>, then libev will assume that C<eventfd ()> is
		3886	available and will probe for kernel support at runtime. This will improve
		3887	C<ev_signal> and C<ev_async> performance and reduce resource consumption.
		3888	If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
		3889	2.7 or newer, otherwise disabled.
		3890
		3891	=item EV_USE_SELECT
		3892
		3893	If undefined or defined to be C<1>, libev will compile in support for the
		3894	C<select>(2) backend. No attempt at auto-detection will be done: if no
		3895	other method takes over, select will be it. Otherwise the select backend
		3896	will not be compiled in.
		3897
		3898	=item EV_SELECT_USE_FD_SET
		3899
		3900	If defined to C<1>, then the select backend will use the system C<fd_set>
		3901	structure. This is useful if libev doesn't compile due to a missing
		3902	C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
		3903	on exotic systems. This usually limits the range of file descriptors to
		3904	some low limit such as 1024 or might have other limitations (winsocket
		3905	only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
		3906	configures the maximum size of the C<fd_set>.
		3907
		3908	=item EV_SELECT_IS_WINSOCKET
		3909
		3910	When defined to C<1>, the select backend will assume that
		3911	select/socket/connect etc. don't understand file descriptors but
		3912	wants osf handles on win32 (this is the case when the select to
		3913	be used is the winsock select). This means that it will call
		3914	C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
		3915	it is assumed that all these functions actually work on fds, even
		3916	on win32. Should not be defined on non-win32 platforms.
		3917
		3918	=item EV_FD_TO_WIN32_HANDLE(fd)
		3919
		3920	If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
		3921	file descriptors to socket handles. When not defining this symbol (the
		3922	default), then libev will call C<_get_osfhandle>, which is usually
		3923	correct. In some cases, programs use their own file descriptor management,
		3924	in which case they can provide this function to map fds to socket handles.
		3925
		3926	=item EV_WIN32_HANDLE_TO_FD(handle)
		3927
		3928	If C<EV_SELECT_IS_WINSOCKET> then libev maps handles to file descriptors
		3929	using the standard C<_open_osfhandle> function. For programs implementing
		3930	their own fd to handle mapping, overwriting this function makes it easier
		3931	to do so. This can be done by defining this macro to an appropriate value.
		3932
		3933	=item EV_WIN32_CLOSE_FD(fd)
		3934
		3935	If programs implement their own fd to handle mapping on win32, then this
		3936	macro can be used to override the C<close> function, useful to unregister
		3937	file descriptors again. Note that the replacement function has to close
		3938	the underlying OS handle.
		3939
		3940	=item EV_USE_POLL
		3941
		3942	If defined to be C<1>, libev will compile in support for the C<poll>(2)
		3943	backend. Otherwise it will be enabled on non-win32 platforms. It
		3944	takes precedence over select.
		3945
		3946	=item EV_USE_EPOLL
		3947
		3948	If defined to be C<1>, libev will compile in support for the Linux
		3949	C<epoll>(7) backend. Its availability will be detected at runtime,
		3950	otherwise another method will be used as fallback. This is the preferred
		3951	backend for GNU/Linux systems. If undefined, it will be enabled if the
		3952	headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
		3953
		3954	=item EV_USE_KQUEUE
		3955
		3956	If defined to be C<1>, libev will compile in support for the BSD style
		3957	C<kqueue>(2) backend. Its actual availability will be detected at runtime,
		3958	otherwise another method will be used as fallback. This is the preferred
		3959	backend for BSD and BSD-like systems, although on most BSDs kqueue only
		3960	supports some types of fds correctly (the only platform we found that
		3961	supports ptys for example was NetBSD), so kqueue might be compiled in, but
		3962	not be used unless explicitly requested. The best way to use it is to find
		3963	out whether kqueue supports your type of fd properly and use an embedded
		3964	kqueue loop.
		3965
		3966	=item EV_USE_PORT
		3967
		3968	If defined to be C<1>, libev will compile in support for the Solaris
		3969	10 port style backend. Its availability will be detected at runtime,
		3970	otherwise another method will be used as fallback. This is the preferred
		3971	backend for Solaris 10 systems.
		3972
		3973	=item EV_USE_DEVPOLL
		3974
		3975	Reserved for future expansion, works like the USE symbols above.
		3976
		3977	=item EV_USE_INOTIFY
		3978
		3979	If defined to be C<1>, libev will compile in support for the Linux inotify
		3980	interface to speed up C<ev_stat> watchers. Its actual availability will
		3981	be detected at runtime. If undefined, it will be enabled if the headers
		3982	indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
		3983
		3984	=item EV_ATOMIC_T
		3985
		3986	Libev requires an integer type (suitable for storing C<0> or C<1>) whose
		3987	access is atomic with respect to other threads or signal contexts. No such
		3988	type is easily found in the C language, so you can provide your own type
		3989	that you know is safe for your purposes. It is used both for signal handler "locking"
		3990	as well as for signal and thread safety in C<ev_async> watchers.
		3991
		3992	In the absence of this define, libev will use C<sig_atomic_t volatile>
		3993	(from F<signal.h>), which is usually good enough on most platforms.
		3994
		3995	=item EV_H (h)
		3996
		3997	The name of the F<ev.h> header file used to include it. The default if
		3998	undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
		3999	used to virtually rename the F<ev.h> header file in case of conflicts.
		4000
		4001	=item EV_CONFIG_H (h)
		4002
		4003	If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
		4004	F<ev.c>'s idea of where to find the F<config.h> file, similarly to
		4005	C<EV_H>, above.
		4006
		4007	=item EV_EVENT_H (h)
		4008
		4009	Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
		4010	of how the F<event.h> header can be found, the default is C<"event.h">.
		4011
		4012	=item EV_PROTOTYPES (h)
		4013
		4014	If defined to be C<0>, then F<ev.h> will not define any function
		4015	prototypes, but still define all the structs and other symbols. This is
		4016	occasionally useful if you want to provide your own wrapper functions
		4017	around libev functions.
		4018
		4019	=item EV_MULTIPLICITY
		4020
		4021	If undefined or defined to C<1>, then all event-loop-specific functions
		4022	will have the C<struct ev_loop *> as first argument, and you can create
		4023	additional independent event loops. Otherwise there will be no support
		4024	for multiple event loops and there is no first event loop pointer
		4025	argument. Instead, all functions act on the single default loop.
		4026
		4027	=item EV_MINPRI
		4028
		4029	=item EV_MAXPRI
		4030
		4031	The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
		4032	C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
		4033	provide for more priorities by overriding those symbols (usually defined
		4034	to be C<-2> and C<2>, respectively).
		4035
		4036	When doing priority-based operations, libev usually has to linearly search
		4037	all the priorities, so having many of them (hundreds) uses a lot of space
		4038	and time, so using the defaults of five priorities (-2 .. +2) is usually
		4039	fine.
		4040
		4041	If your embedding application does not need any priorities, defining these
		4042	both to C<0> will save some memory and CPU.
		4043
		4044	=item EV_PERIODIC_ENABLE, EV_IDLE_ENABLE, EV_EMBED_ENABLE, EV_STAT_ENABLE,
		4045	EV_PREPARE_ENABLE, EV_CHECK_ENABLE, EV_FORK_ENABLE, EV_SIGNAL_ENABLE,
		4046	EV_ASYNC_ENABLE, EV_CHILD_ENABLE.
		4047
		4048	If undefined or defined to be C<1> (and the platform supports it), then
		4049	the respective watcher type is supported. If defined to be C<0>, then it
		4050	is not. Disabling watcher types mainly saves code size.
		4051
		4052	=item EV_FEATURES
		4053
		4054	If you need to shave off some kilobytes of code at the expense of some
		4055	speed (but with the full API), you can define this symbol to request
		4056	certain subsets of functionality. The default is to enable all features
		4057	that can be enabled on the platform.
		4058
		4059	A typical way to use this symbol is to define it to C<0> (or to a bitset
		4060	with some broad features you want) and then selectively re-enable
		4061	additional parts you want, for example if you want everything minimal,
		4062	but multiple event loop support, async and child watchers and the poll
		4063	backend, use this:
		4064
		4065	#define EV_FEATURES 0
		4066	#define EV_MULTIPLICITY 1
		4067	#define EV_USE_POLL 1
		4068	#define EV_CHILD_ENABLE 1
		4069	#define EV_ASYNC_ENABLE 1
		4070
		4071	The actual value is a bitset, it can be a combination of the following
		4072	values:
		4073
		4074	=over 4
		4075
		4076	=item C<1> - faster/larger code
		4077
		4078	Use larger code to speed up some operations.
		4079
		4080	Currently this is used to override some inlining decisions (enlarging the
		4081	code size by roughly 30% on amd64).
		4082
		4083	When optimising for size, use of compiler flags such as C<-Os> with
		4084	gcc is recommended, as well as C<-DNDEBUG>, as libev contains a number of
		4085	assertions.
		4086
		4087	=item C<2> - faster/larger data structures
		4088
		4089	Replaces the small 2-heap for timer management by a faster 4-heap, larger
		4090	hash table sizes and so on. This will usually further increase code size
		4091	and can additionally have an effect on the size of data structures at
		4092	runtime.
		4093
		4094	=item C<4> - full API configuration
		4095
		4096	This enables priorities (sets C<EV_MAXPRI>=2 and C<EV_MINPRI>=-2), and
		4097	enables multiplicity (C<EV_MULTIPLICITY>=1).
		4098
		4099	=item C<8> - full API
		4100
		4101	This enables a lot of the "lesser used" API functions. See C<ev.h> for
		4102	details on which parts of the API are still available without this
		4103	feature, and do not complain if this subset changes over time.
		4104
		4105	=item C<16> - enable all optional watcher types
		4106
		4107	Enables all optional watcher types. If you want to selectively enable
		4108	only some watcher types other than I/O and timers (e.g. prepare,
		4109	embed, async, child...) you can enable them manually by defining
		4110	C<EV_watchertype_ENABLE> to C<1> instead.
		4111
		4112	=item C<32> - enable all backends
		4113
		4114	This enables all backends - without this feature, you need to enable at
		4115	least one backend manually (C<EV_USE_SELECT> is a good choice).
		4116
		4117	=item C<64> - enable OS-specific "helper" APIs
		4118
		4119	Enable inotify, eventfd, signalfd and similar OS-specific helper APIs by
		4120	default.
		4121
		4122	=back
		4123
		4124	Compiling with C<gcc -Os -DEV_STANDALONE -DEV_USE_EPOLL=1 -DEV_FEATURES=0>
		4125	reduces the compiled size of libev from 24.7Kb code/2.8Kb data to 6.5Kb
		4126	code/0.3Kb data on my GNU/Linux amd64 system, while still giving you I/O
		4127	watchers, timers and monotonic clock support.
		4128
		4129	With an intelligent-enough linker (gcc+binutils are intelligent enough
		4130	when you use C<-Wl,--gc-sections -ffunction-sections>) functions unused by
		4131	your program might be left out as well - a binary starting a timer and an
		4132	I/O watcher then might come out at only 5Kb.
		4133
		4134	=item EV_AVOID_STDIO
		4135
		4136	If this is set to C<1> at compiletime, then libev will avoid using stdio
		4137	functions (printf, scanf, perror etc.). This will increase the code size
		4138	somewhat, but if your program doesn't otherwise depend on stdio and your
		4139	libc allows it, this avoids linking in the stdio library which is quite
		4140	big.
		4141
		4142	Note that error messages might become less precise when this option is
		4143	enabled.
		4144
		4145	=item EV_NSIG
		4146
		4147	The highest supported signal number, +1 (or, the number of
		4148	signals): Normally, libev tries to deduce the maximum number of signals
		4149	automatically, but sometimes this fails, in which case it can be
		4150	specified. Also, using a lower number than detected (C<32> should be
		4151	good for about any system in existence) can save some memory, as libev
		4152	statically allocates some 12-24 bytes per signal number.
		4153
		4154	=item EV_PID_HASHSIZE
		4155
		4156	C<ev_child> watchers use a small hash table to distribute workload by
		4157	pid. The default size is C<16> (or C<1> with C<EV_FEATURES> disabled),
		4158	usually more than enough. If you need to manage thousands of children you
		4159	might want to increase this value (I<must> be a power of two).
		4160
		4161	=item EV_INOTIFY_HASHSIZE
		4162
		4163	C<ev_stat> watchers use a small hash table to distribute workload by
		4164	inotify watch id. The default size is C<16> (or C<1> with C<EV_FEATURES>
		4165	disabled), usually more than enough. If you need to manage thousands of
		4166	C<ev_stat> watchers you might want to increase this value (I<must> be a
		4167	power of two).
		4168
		4169	=item EV_USE_4HEAP
		4170
		4171	Heaps are not very cache-efficient. To improve the cache-efficiency of the
		4172	timer and periodics heaps, libev uses a 4-heap when this symbol is defined
		4173	to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
		4174	faster performance with many (thousands) of watchers.
		4175
		4176	The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
		4177	will be C<0>.
		4178
		4179	=item EV_HEAP_CACHE_AT
		4180
		4181	Heaps are not very cache-efficient. To improve the cache-efficiency of the
		4182	timer and periodics heaps, libev can cache the timestamp (I<at>) within
		4183	the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
		4184	which uses 8-12 bytes more per watcher and a few hundred bytes more code,
		4185	but avoids random read accesses on heap changes. This improves performance
		4186	noticeably with many (hundreds) of watchers.
		4187
		4188	The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
		4189	will be C<0>.
		4190
		4191	=item EV_VERIFY
		4192
		4193	Controls how much internal verification (see C<ev_verify ()>) will
		4194	be done: If set to C<0>, no internal verification code will be compiled
		4195	in. If set to C<1>, then verification code will be compiled in, but not
		4196	called. If set to C<2>, then the internal verification code will be
		4197	called once per loop, which can slow down libev. If set to C<3>, then the
		4198	verification code will be called very frequently, which will slow down
		4199	libev considerably.
		4200
		4201	The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
		4202	will be C<0>.
		4203
		4204	=item EV_COMMON
		4205
		4206	By default, all watchers have a C<void *data> member. By redefining
		4207	this macro to something else you can include more and other types of
		4208	members. You have to define it each time you include one of the files,
		4209	though, and it must be identical each time.
		4210
		4211	For example, the perl EV module uses something like this:
		4212
		4213	#define EV_COMMON \
		4214	SV *self; /* contains this struct */ \
		4215	SV *cb_sv, *fh /* note no trailing ";" */
		4216
		4217	=item EV_CB_DECLARE (type)
		4218
		4219	=item EV_CB_INVOKE (watcher, revents)
		4220
		4221	=item ev_set_cb (ev, cb)
		4222
		4223	Can be used to change the callback member declaration in each watcher,
		4224	and the way callbacks are invoked and set. Must expand to a struct member
		4225	definition and a statement, respectively. See the F<ev.h> header file for
		4226	their default definitions. One possible use for overriding these is to
		4227	avoid the C<struct ev_loop *> as first argument in all cases, or to use
		4228	method calls instead of plain function calls in C++.
		4229
		4230	=back
		4231
		4232	=head2 EXPORTED API SYMBOLS
		4233
		4234	If you need to re-export the API (e.g. via a DLL) and you need a list of
		4235	exported symbols, you can use the provided F<Symbol.*> files which list
		4236	all public symbols, one per line:
		4237
		4238	Symbols.ev for libev proper
		4239	Symbols.event for the libevent emulation
		4240
		4241	This can also be used to rename all public symbols to avoid clashes with
		4242	multiple versions of libev linked together (which is obviously bad in
		4243	itself, but sometimes it is inconvenient to avoid this).
		4244
		4245	A sed command like this will create wrapper C<#define>'s that you need to
		4246	include before including F<ev.h>:
		4247
		4248	<Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
		4249
		4250	This would create a file F<wrap.h> which essentially looks like this:
		4251
		4252	#define ev_backend myprefix_ev_backend
		4253	#define ev_check_start myprefix_ev_check_start
		4254	#define ev_check_stop myprefix_ev_check_stop
		4255	...
		4256
		4257	=head2 EXAMPLES
		4258
		4259	For a real-world example of a program the includes libev
		4260	verbatim, you can have a look at the EV perl module
		4261	(L<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
		4262	the F<libev/> subdirectory and includes them in the F<EV/EVAPI.h> (public
		4263	interface) and F<EV.xs> (implementation) files. Only the F<EV.xs> file
		4264	will be compiled. It is pretty complex because it provides its own header
		4265	file.
		4266
		4267	The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
		4268	that everybody includes and which overrides some configure choices:
		4269
		4270	#define EV_FEATURES 8
		4271	#define EV_USE_SELECT 1
		4272	#define EV_PREPARE_ENABLE 1
		4273	#define EV_IDLE_ENABLE 1
		4274	#define EV_SIGNAL_ENABLE 1
		4275	#define EV_CHILD_ENABLE 1
		4276	#define EV_USE_STDEXCEPT 0
		4277	#define EV_CONFIG_H <config.h>
		4278
		4279	#include "ev++.h"
		4280
		4281	And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
		4282
		4283	#include "ev_cpp.h"
		4284	#include "ev.c"
		4285
		4286	=head1 INTERACTION WITH OTHER PROGRAMS OR LIBRARIES
		4287
		4288	=head2 THREADS AND COROUTINES
		4289
		4290	=head3 THREADS
		4291
		4292	All libev functions are reentrant and thread-safe unless explicitly
		4293	documented otherwise, but libev implements no locking itself. This means
		4294	that you can use as many loops as you want in parallel, as long as there
		4295	are no concurrent calls into any libev function with the same loop
		4296	parameter (C<ev_default_*> calls have an implicit default loop parameter,
		4297	of course): libev guarantees that different event loops share no data
		4298	structures that need any locking.
		4299
		4300	Or to put it differently: calls with different loop parameters can be done
		4301	concurrently from multiple threads, calls with the same loop parameter
		4302	must be done serially (but can be done from different threads, as long as
		4303	only one thread ever is inside a call at any point in time, e.g. by using
		4304	a mutex per loop).
		4305
		4306	Specifically to support threads (and signal handlers), libev implements
		4307	so-called C<ev_async> watchers, which allow some limited form of
		4308	concurrency on the same event loop, namely waking it up "from the
		4309	outside".
		4310
		4311	If you want to know which design (one loop, locking, or multiple loops
		4312	without or something else still) is best for your problem, then I cannot
		4313	help you, but here is some generic advice:
		4314
		4315	=over 4
		4316
		4317	=item * most applications have a main thread: use the default libev loop
		4318	in that thread, or create a separate thread running only the default loop.
		4319
		4320	This helps integrating other libraries or software modules that use libev
		4321	themselves and don't care/know about threading.
		4322
		4323	=item * one loop per thread is usually a good model.
		4324
		4325	Doing this is almost never wrong, sometimes a better-performance model
		4326	exists, but it is always a good start.
		4327
		4328	=item * other models exist, such as the leader/follower pattern, where one
		4329	loop is handed through multiple threads in a kind of round-robin fashion.
		4330
		4331	Choosing a model is hard - look around, learn, know that usually you can do
		4332	better than you currently do :-)
		4333
		4334	=item * often you need to talk to some other thread which blocks in the
		4335	event loop.
		4336
		4337	C<ev_async> watchers can be used to wake them up from other threads safely
		4338	(or from signal contexts...).
		4339
		4340	An example use would be to communicate signals or other events that only
		4341	work in the default loop by registering the signal watcher with the
		4342	default loop and triggering an C<ev_async> watcher from the default loop
		4343	watcher callback into the event loop interested in the signal.
		4344
		4345	=back
		4346
		4347	=head4 THREAD LOCKING EXAMPLE
		4348
		4349	Here is a fictitious example of how to run an event loop in a different
		4350	thread than where callbacks are being invoked and watchers are
		4351	created/added/removed.
		4352
		4353	For a real-world example, see the C<EV::Loop::Async> perl module,
		4354	which uses exactly this technique (which is suited for many high-level
		4355	languages).
		4356
		4357	The example uses a pthread mutex to protect the loop data, a condition
		4358	variable to wait for callback invocations, an async watcher to notify the
		4359	event loop thread and an unspecified mechanism to wake up the main thread.
		4360
		4361	First, you need to associate some data with the event loop:
		4362
		4363	typedef struct {
		4364	mutex_t lock; /* global loop lock */
		4365	ev_async async_w;
		4366	thread_t tid;
		4367	cond_t invoke_cv;
		4368	} userdata;
		4369
		4370	void prepare_loop (EV_P)
		4371	{
		4372	// for simplicity, we use a static userdata struct.
		4373	static userdata u;
		4374
		4375	ev_async_init (&u->async_w, async_cb);
		4376	ev_async_start (EV_A_ &u->async_w);
		4377
		4378	pthread_mutex_init (&u->lock, 0);
		4379	pthread_cond_init (&u->invoke_cv, 0);
		4380
		4381	// now associate this with the loop
		4382	ev_set_userdata (EV_A_ u);
		4383	ev_set_invoke_pending_cb (EV_A_ l_invoke);
		4384	ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
		4385
		4386	// then create the thread running ev_loop
		4387	pthread_create (&u->tid, 0, l_run, EV_A);
		4388	}
		4389
		4390	The callback for the C<ev_async> watcher does nothing: the watcher is used
		4391	solely to wake up the event loop so it takes notice of any new watchers
		4392	that might have been added:
		4393
		4394	static void
		4395	async_cb (EV_P_ ev_async *w, int revents)
		4396	{
		4397	// just used for the side effects
		4398	}
		4399
		4400	The C<l_release> and C<l_acquire> callbacks simply unlock/lock the mutex
		4401	protecting the loop data, respectively.
		4402
		4403	static void
		4404	l_release (EV_P)
		4405	{
		4406	userdata *u = ev_userdata (EV_A);
		4407	pthread_mutex_unlock (&u->lock);
		4408	}
		4409
		4410	static void
		4411	l_acquire (EV_P)
		4412	{
		4413	userdata *u = ev_userdata (EV_A);
		4414	pthread_mutex_lock (&u->lock);
		4415	}
		4416
		4417	The event loop thread first acquires the mutex, and then jumps straight
		4418	into C<ev_run>:
		4419
		4420	void *
		4421	l_run (void *thr_arg)
		4422	{
		4423	struct ev_loop *loop = (struct ev_loop *)thr_arg;
		4424
		4425	l_acquire (EV_A);
		4426	pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
		4427	ev_run (EV_A_ 0);
		4428	l_release (EV_A);
		4429
		4430	return 0;
		4431	}
		4432
		4433	Instead of invoking all pending watchers, the C<l_invoke> callback will
		4434	signal the main thread via some unspecified mechanism (signals? pipe
		4435	writes? C<Async::Interrupt>?) and then waits until all pending watchers
		4436	have been called (in a while loop because a) spurious wakeups are possible
		4437	and b) skipping inter-thread-communication when there are no pending
		4438	watchers is very beneficial):
		4439
		4440	static void
		4441	l_invoke (EV_P)
		4442	{
		4443	userdata *u = ev_userdata (EV_A);
		4444
		4445	while (ev_pending_count (EV_A))
		4446	{
		4447	wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
		4448	pthread_cond_wait (&u->invoke_cv, &u->lock);
		4449	}
		4450	}
		4451
		4452	Now, whenever the main thread gets told to invoke pending watchers, it
		4453	will grab the lock, call C<ev_invoke_pending> and then signal the loop
		4454	thread to continue:
		4455
		4456	static void
		4457	real_invoke_pending (EV_P)
		4458	{
		4459	userdata *u = ev_userdata (EV_A);
		4460
		4461	pthread_mutex_lock (&u->lock);
		4462	ev_invoke_pending (EV_A);
		4463	pthread_cond_signal (&u->invoke_cv);
		4464	pthread_mutex_unlock (&u->lock);
		4465	}
		4466
		4467	Whenever you want to start/stop a watcher or do other modifications to an
		4468	event loop, you will now have to lock:
		4469
		4470	ev_timer timeout_watcher;
		4471	userdata *u = ev_userdata (EV_A);
		4472
		4473	ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
		4474
		4475	pthread_mutex_lock (&u->lock);
		4476	ev_timer_start (EV_A_ &timeout_watcher);
		4477	ev_async_send (EV_A_ &u->async_w);
		4478	pthread_mutex_unlock (&u->lock);
		4479
		4480	Note that sending the C<ev_async> watcher is required because otherwise
		4481	an event loop currently blocking in the kernel will have no knowledge
		4482	about the newly added timer. By waking up the loop it will pick up any new
		4483	watchers in the next event loop iteration.
		4484
		4485	=head3 COROUTINES
		4486
		4487	Libev is very accommodating to coroutines ("cooperative threads"):
		4488	libev fully supports nesting calls to its functions from different
		4489	coroutines (e.g. you can call C<ev_run> on the same loop from two
		4490	different coroutines, and switch freely between both coroutines running
		4491	the loop, as long as you don't confuse yourself). The only exception is
		4492	that you must not do this from C<ev_periodic> reschedule callbacks.
		4493
		4494	Care has been taken to ensure that libev does not keep local state inside
		4495	C<ev_run>, and other calls do not usually allow for coroutine switches as
		4496	they do not call any callbacks.
		4497
		4498	=head2 COMPILER WARNINGS
		4499
		4500	Depending on your compiler and compiler settings, you might get no or a
		4501	lot of warnings when compiling libev code. Some people are apparently
		4502	scared by this.
		4503
		4504	However, these are unavoidable for many reasons. For one, each compiler
		4505	has different warnings, and each user has different tastes regarding
		4506	warning options. "Warn-free" code therefore cannot be a goal except when
		4507	targeting a specific compiler and compiler-version.
		4508
		4509	Another reason is that some compiler warnings require elaborate
		4510	workarounds, or other changes to the code that make it less clear and less
		4511	maintainable.
		4512
		4513	And of course, some compiler warnings are just plain stupid, or simply
		4514	wrong (because they don't actually warn about the condition their message
		4515	seems to warn about). For example, certain older gcc versions had some
		4516	warnings that resulted in an extreme number of false positives. These have
		4517	been fixed, but some people still insist on making code warn-free with
		4518	such buggy versions.
		4519
		4520	While libev is written to generate as few warnings as possible,
		4521	"warn-free" code is not a goal, and it is recommended not to build libev
		4522	with any compiler warnings enabled unless you are prepared to cope with
		4523	them (e.g. by ignoring them). Remember that warnings are just that:
		4524	warnings, not errors, or proof of bugs.
		4525
		4526
		4527	=head2 VALGRIND
		4528
		4529	Valgrind has a special section here because it is a popular tool that is
		4530	highly useful. Unfortunately, valgrind reports are very hard to interpret.
		4531
		4532	If you think you found a bug (memory leak, uninitialised data access etc.)
		4533	in libev, then check twice: If valgrind reports something like:
		4534
		4535	==2274== definitely lost: 0 bytes in 0 blocks.
		4536	==2274== possibly lost: 0 bytes in 0 blocks.
		4537	==2274== still reachable: 256 bytes in 1 blocks.
		4538
		4539	Then there is no memory leak, just as memory accounted to global variables
		4540	is not a memleak - the memory is still being referenced, and didn't leak.
		4541
		4542	Similarly, under some circumstances, valgrind might report kernel bugs
		4543	as if it were a bug in libev (e.g. in realloc or in the poll backend,
		4544	although an acceptable workaround has been found here), or it might be
		4545	confused.
		4546
		4547	Keep in mind that valgrind is a very good tool, but only a tool. Don't
		4548	make it into some kind of religion.
		4549
		4550	If you are unsure about something, feel free to contact the mailing list
		4551	with the full valgrind report and an explanation on why you think this
		4552	is a bug in libev (best check the archives, too :). However, don't be
		4553	annoyed when you get a brisk "this is no bug" answer and take the chance
		4554	of learning how to interpret valgrind properly.
		4555
		4556	If you need, for some reason, empty reports from valgrind for your project
		4557	I suggest using suppression lists.
		4558
		4559
		4560	=head1 PORTABILITY NOTES
		4561
		4562	=head2 GNU/LINUX 32 BIT LIMITATIONS
		4563
		4564	GNU/Linux is the only common platform that supports 64 bit file/large file
		4565	interfaces but I<disables> them by default.
		4566
		4567	That means that libev compiled in the default environment doesn't support
		4568	files larger than 2GiB or so, which mainly affects C<ev_stat> watchers.
		4569
		4570	Unfortunately, many programs try to work around this GNU/Linux issue
		4571	by enabling the large file API, which makes them incompatible with the
		4572	standard libev compiled for their system.
		4573
		4574	Likewise, libev cannot enable the large file API itself as this would
		4575	suddenly make it incompatible to the default compile time environment,
		4576	i.e. all programs not using special compile switches.
		4577
		4578	=head2 OS/X AND DARWIN BUGS
		4579
		4580	The whole thing is a bug if you ask me - basically any system interface
		4581	you touch is broken, whether it is locales, poll, kqueue or even the
		4582	OpenGL drivers.
		4583
		4584	=head3 C<kqueue> is buggy
		4585
		4586	The kqueue syscall is broken in all known versions - most versions support
		4587	only sockets, many support pipes.
		4588
		4589	Libev tries to work around this by not using C<kqueue> by default on this
		4590	rotten platform, but of course you can still ask for it when creating a
		4591	loop - embedding a socket-only kqueue loop into a select-based one is
		4592	probably going to work well.
		4593
		4594	=head3 C<poll> is buggy
		4595
		4596	Instead of fixing C<kqueue>, Apple replaced their (working) C<poll>
		4597	implementation by something calling C<kqueue> internally around the 10.5.6
		4598	release, so now C<kqueue> I<and> C<poll> are broken.
		4599
		4600	Libev tries to work around this by not using C<poll> by default on
		4601	this rotten platform, but of course you can still ask for it when creating
		4602	a loop.
		4603
		4604	=head3 C<select> is buggy
		4605
		4606	All that's left is C<select>, and of course Apple found a way to fuck this
		4607	one up as well: On OS/X, C<select> actively limits the number of file
		4608	descriptors you can pass in to 1024 - your program suddenly crashes when
		4609	you use more.
		4610
		4611	There is an undocumented "workaround" for this - defining
		4612	C<_DARWIN_UNLIMITED_SELECT>, which libev tries to use, so select I<should>
		4613	work on OS/X.
		4614
		4615	=head2 SOLARIS PROBLEMS AND WORKAROUNDS
		4616
		4617	=head3 C<errno> reentrancy
		4618
		4619	The default compile environment on Solaris is unfortunately so
		4620	thread-unsafe that you can't even use components/libraries compiled
		4621	without C<-D_REENTRANT> in a threaded program, which, of course, isn't
		4622	defined by default. A valid, if stupid, implementation choice.
		4623
		4624	If you want to use libev in threaded environments you have to make sure
		4625	it's compiled with C<_REENTRANT> defined.
		4626
		4627	=head3 Event port backend
		4628
		4629	The scalable event interface for Solaris is called "event
		4630	ports". Unfortunately, this mechanism is very buggy in all major
		4631	releases. If you run into high CPU usage, your program freezes or you get
		4632	a large number of spurious wakeups, make sure you have all the relevant
		4633	and latest kernel patches applied. No, I don't know which ones, but there
		4634	are multiple ones to apply, and afterwards, event ports actually work
		4635	great.
		4636
		4637	If you can't get it to work, you can try running the program by setting
		4638	the environment variable C<LIBEV_FLAGS=3> to only allow C<poll> and
		4639	C<select> backends.
		4640
		4641	=head2 AIX POLL BUG
		4642
		4643	AIX unfortunately has a broken C<poll.h> header. Libev works around
		4644	this by trying to avoid the poll backend altogether (i.e. it's not even
		4645	compiled in), which normally isn't a big problem as C<select> works fine
		4646	with large bitsets on AIX, and AIX is dead anyway.
		4647
		4648	=head2 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
		4649
		4650	=head3 General issues
		4651
		4652	Win32 doesn't support any of the standards (e.g. POSIX) that libev
		4653	requires, and its I/O model is fundamentally incompatible with the POSIX
		4654	model. Libev still offers limited functionality on this platform in
		4655	the form of the C<EVBACKEND_SELECT> backend, and only supports socket
		4656	descriptors. This only applies when using Win32 natively, not when using
		4657	e.g. cygwin. Actually, it only applies to the microsofts own compilers,
		4658	as every compielr comes with a slightly differently broken/incompatible
		4659	environment.
		4660
		4661	Lifting these limitations would basically require the full
		4662	re-implementation of the I/O system. If you are into this kind of thing,
		4663	then note that glib does exactly that for you in a very portable way (note
		4664	also that glib is the slowest event library known to man).
		4665
		4666	There is no supported compilation method available on windows except
		4667	embedding it into other applications.
		4668
		4669	Sensible signal handling is officially unsupported by Microsoft - libev
		4670	tries its best, but under most conditions, signals will simply not work.
		4671
		4672	Not a libev limitation but worth mentioning: windows apparently doesn't
		4673	accept large writes: instead of resulting in a partial write, windows will
		4674	either accept everything or return C<ENOBUFS> if the buffer is too large,
		4675	so make sure you only write small amounts into your sockets (less than a
		4676	megabyte seems safe, but this apparently depends on the amount of memory
		4677	available).
		4678
		4679	Due to the many, low, and arbitrary limits on the win32 platform and
		4680	the abysmal performance of winsockets, using a large number of sockets
		4681	is not recommended (and not reasonable). If your program needs to use
		4682	more than a hundred or so sockets, then likely it needs to use a totally
		4683	different implementation for windows, as libev offers the POSIX readiness
		4684	notification model, which cannot be implemented efficiently on windows
		4685	(due to Microsoft monopoly games).
		4686
		4687	A typical way to use libev under windows is to embed it (see the embedding
		4688	section for details) and use the following F<evwrap.h> header file instead
		4689	of F<ev.h>:
		4690
		4691	#define EV_STANDALONE /* keeps ev from requiring config.h */
		4692	#define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
		4693
		4694	#include "ev.h"
		4695
		4696	And compile the following F<evwrap.c> file into your project (make sure
		4697	you do I<not> compile the F<ev.c> or any other embedded source files!):
		4698
		4699	#include "evwrap.h"
		4700	#include "ev.c"
		4701
		4702	=head3 The winsocket C<select> function
		4703
		4704	The winsocket C<select> function doesn't follow POSIX in that it
		4705	requires socket I<handles> and not socket I<file descriptors> (it is
		4706	also extremely buggy). This makes select very inefficient, and also
		4707	requires a mapping from file descriptors to socket handles (the Microsoft
		4708	C runtime provides the function C<_open_osfhandle> for this). See the
		4709	discussion of the C<EV_SELECT_USE_FD_SET>, C<EV_SELECT_IS_WINSOCKET> and
		4710	C<EV_FD_TO_WIN32_HANDLE> preprocessor symbols for more info.
		4711
		4712	The configuration for a "naked" win32 using the Microsoft runtime
		4713	libraries and raw winsocket select is:
		4714
		4715	#define EV_USE_SELECT 1
		4716	#define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
		4717
		4718	Note that winsockets handling of fd sets is O(n), so you can easily get a
		4719	complexity in the O(n²) range when using win32.
		4720
		4721	=head3 Limited number of file descriptors
		4722
		4723	Windows has numerous arbitrary (and low) limits on things.
		4724
		4725	Early versions of winsocket's select only supported waiting for a maximum
		4726	of C<64> handles (probably owning to the fact that all windows kernels
		4727	can only wait for C<64> things at the same time internally; Microsoft
		4728	recommends spawning a chain of threads and wait for 63 handles and the
		4729	previous thread in each. Sounds great!).
		4730
		4731	Newer versions support more handles, but you need to define C<FD_SETSIZE>
		4732	to some high number (e.g. C<2048>) before compiling the winsocket select
		4733	call (which might be in libev or elsewhere, for example, perl and many
		4734	other interpreters do their own select emulation on windows).
		4735
		4736	Another limit is the number of file descriptors in the Microsoft runtime
		4737	libraries, which by default is C<64> (there must be a hidden I<64>
		4738	fetish or something like this inside Microsoft). You can increase this
		4739	by calling C<_setmaxstdio>, which can increase this limit to C<2048>
		4740	(another arbitrary limit), but is broken in many versions of the Microsoft
		4741	runtime libraries. This might get you to about C<512> or C<2048> sockets
		4742	(depending on windows version and/or the phase of the moon). To get more,
		4743	you need to wrap all I/O functions and provide your own fd management, but
		4744	the cost of calling select (O(n²)) will likely make this unworkable.
		4745
		4746	=head2 PORTABILITY REQUIREMENTS
		4747
		4748	In addition to a working ISO-C implementation and of course the
		4749	backend-specific APIs, libev relies on a few additional extensions:
		4750
		4751	=over 4
		4752
		4753	=item C<void (*)(ev_watcher_type *, int revents)> must have compatible
		4754	calling conventions regardless of C<ev_watcher_type *>.
		4755
		4756	Libev assumes not only that all watcher pointers have the same internal
		4757	structure (guaranteed by POSIX but not by ISO C for example), but it also
		4758	assumes that the same (machine) code can be used to call any watcher
		4759	callback: The watcher callbacks have different type signatures, but libev
		4760	calls them using an C<ev_watcher *> internally.
		4761
		4762	=item C<sig_atomic_t volatile> must be thread-atomic as well
		4763
		4764	The type C<sig_atomic_t volatile> (or whatever is defined as
		4765	C<EV_ATOMIC_T>) must be atomic with respect to accesses from different
		4766	threads. This is not part of the specification for C<sig_atomic_t>, but is
		4767	believed to be sufficiently portable.
		4768
		4769	=item C<sigprocmask> must work in a threaded environment
		4770
		4771	Libev uses C<sigprocmask> to temporarily block signals. This is not
		4772	allowed in a threaded program (C<pthread_sigmask> has to be used). Typical
		4773	pthread implementations will either allow C<sigprocmask> in the "main
		4774	thread" or will block signals process-wide, both behaviours would
		4775	be compatible with libev. Interaction between C<sigprocmask> and
		4776	C<pthread_sigmask> could complicate things, however.
		4777
		4778	The most portable way to handle signals is to block signals in all threads
		4779	except the initial one, and run the default loop in the initial thread as
		4780	well.
		4781
		4782	=item C<long> must be large enough for common memory allocation sizes
		4783
		4784	To improve portability and simplify its API, libev uses C<long> internally
		4785	instead of C<size_t> when allocating its data structures. On non-POSIX
		4786	systems (Microsoft...) this might be unexpectedly low, but is still at
		4787	least 31 bits everywhere, which is enough for hundreds of millions of
		4788	watchers.
		4789
		4790	=item C<double> must hold a time value in seconds with enough accuracy
		4791
		4792	The type C<double> is used to represent timestamps. It is required to
		4793	have at least 51 bits of mantissa (and 9 bits of exponent), which is
		4794	good enough for at least into the year 4000 with millisecond accuracy
		4795	(the design goal for libev). This requirement is overfulfilled by
		4796	implementations using IEEE 754, which is basically all existing ones. With
		4797	IEEE 754 doubles, you get microsecond accuracy until at least 2200.
		4798
		4799	=back
		4800
		4801	If you know of other additional requirements drop me a note.
		4802
		4803
		4804	=head1 ALGORITHMIC COMPLEXITIES
		4805
		4806	In this section the complexities of (many of) the algorithms used inside
		4807	libev will be documented. For complexity discussions about backends see
		4808	the documentation for C<ev_default_init>.
		4809
		4810	All of the following are about amortised time: If an array needs to be
		4811	extended, libev needs to realloc and move the whole array, but this
		4812	happens asymptotically rarer with higher number of elements, so O(1) might
		4813	mean that libev does a lengthy realloc operation in rare cases, but on
		4814	average it is much faster and asymptotically approaches constant time.
		4815
		4816	=over 4
		4817
		4818	=item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
		4819
		4820	This means that, when you have a watcher that triggers in one hour and
		4821	there are 100 watchers that would trigger before that, then inserting will
		4822	have to skip roughly seven (C<ld 100>) of these watchers.
		4823
		4824	=item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
		4825
		4826	That means that changing a timer costs less than removing/adding them,
		4827	as only the relative motion in the event queue has to be paid for.
		4828
		4829	=item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
		4830
		4831	These just add the watcher into an array or at the head of a list.
		4832
		4833	=item Stopping check/prepare/idle/fork/async watchers: O(1)
		4834
		4835	=item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
		4836
		4837	These watchers are stored in lists, so they need to be walked to find the
		4838	correct watcher to remove. The lists are usually short (you don't usually
		4839	have many watchers waiting for the same fd or signal: one is typical, two
		4840	is rare).
		4841
		4842	=item Finding the next timer in each loop iteration: O(1)
		4843
		4844	By virtue of using a binary or 4-heap, the next timer is always found at a
		4845	fixed position in the storage array.
		4846
		4847	=item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
		4848
		4849	A change means an I/O watcher gets started or stopped, which requires
		4850	libev to recalculate its status (and possibly tell the kernel, depending
		4851	on backend and whether C<ev_io_set> was used).
		4852
		4853	=item Activating one watcher (putting it into the pending state): O(1)
		4854
		4855	=item Priority handling: O(number_of_priorities)
		4856
		4857	Priorities are implemented by allocating some space for each
		4858	priority. When doing priority-based operations, libev usually has to
		4859	linearly search all the priorities, but starting/stopping and activating
		4860	watchers becomes O(1) with respect to priority handling.
		4861
		4862	=item Sending an ev_async: O(1)
		4863
		4864	=item Processing ev_async_send: O(number_of_async_watchers)
		4865
		4866	=item Processing signals: O(max_signal_number)
		4867
		4868	Sending involves a system call I<iff> there were no other C<ev_async_send>
		4869	calls in the current loop iteration. Checking for async and signal events
		4870	involves iterating over all running async watchers or all signal numbers.
		4871
		4872	=back
		4873
		4874
		4875	=head1 PORTING FROM LIBEV 3.X TO 4.X
		4876
		4877	The major version 4 introduced some minor incompatible changes to the API.
		4878
		4879	At the moment, the C<ev.h> header file tries to implement superficial
		4880	compatibility, so most programs should still compile. Those might be
		4881	removed in later versions of libev, so better update early than late.
		4882
		4883	=over 4
		4884
		4885	=item C<ev_default_destroy> and C<ev_default_fork> have been removed
		4886
		4887	These calls can be replaced easily by their C<ev_loop_xxx> counterparts:
		4888
		4889	ev_loop_destroy (EV_DEFAULT_UC);
		4890	ev_loop_fork (EV_DEFAULT);
		4891
		4892	=item function/symbol renames
		4893
		4894	A number of functions and symbols have been renamed:
		4895
		4896	ev_loop => ev_run
		4897	EVLOOP_NONBLOCK => EVRUN_NOWAIT
		4898	EVLOOP_ONESHOT => EVRUN_ONCE
		4899
		4900	ev_unloop => ev_break
		4901	EVUNLOOP_CANCEL => EVBREAK_CANCEL
		4902	EVUNLOOP_ONE => EVBREAK_ONE
		4903	EVUNLOOP_ALL => EVBREAK_ALL
		4904
		4905	EV_TIMEOUT => EV_TIMER
		4906
		4907	ev_loop_count => ev_iteration
		4908	ev_loop_depth => ev_depth
		4909	ev_loop_verify => ev_verify
		4910
		4911	Most functions working on C<struct ev_loop> objects don't have an
		4912	C<ev_loop_> prefix, so it was removed; C<ev_loop>, C<ev_unloop> and
		4913	associated constants have been renamed to not collide with the C<struct
		4914	ev_loop> anymore and C<EV_TIMER> now follows the same naming scheme
		4915	as all other watcher types. Note that C<ev_loop_fork> is still called
		4916	C<ev_loop_fork> because it would otherwise clash with the C<ev_fork>
		4917	typedef.
		4918
		4919	=item C<EV_COMPAT3> backwards compatibility mechanism
		4920
		4921	The backward compatibility mechanism can be controlled by
		4922	C<EV_COMPAT3>. See L<PREPROCESSOR SYMBOLS/MACROS> in the L<EMBEDDING>
		4923	section.
		4924
		4925	=item C<EV_MINIMAL> mechanism replaced by C<EV_FEATURES>
		4926
		4927	The preprocessor symbol C<EV_MINIMAL> has been replaced by a different
		4928	mechanism, C<EV_FEATURES>. Programs using C<EV_MINIMAL> usually compile
		4929	and work, but the library code will of course be larger.
		4930
		4931	=back
		4932
		4933
		4934	=head1 GLOSSARY
		4935
		4936	=over 4
		4937
		4938	=item active
		4939
		4940	A watcher is active as long as it has been started and not yet stopped.
		4941	See L<WATCHER STATES> for details.
		4942
		4943	=item application
		4944
		4945	In this document, an application is whatever is using libev.
		4946
		4947	=item backend
		4948
		4949	The part of the code dealing with the operating system interfaces.
		4950
		4951	=item callback
		4952
		4953	The address of a function that is called when some event has been
		4954	detected. Callbacks are being passed the event loop, the watcher that
		4955	received the event, and the actual event bitset.
		4956
		4957	=item callback/watcher invocation
		4958
		4959	The act of calling the callback associated with a watcher.
		4960
		4961	=item event
		4962
		4963	A change of state of some external event, such as data now being available
		4964	for reading on a file descriptor, time having passed or simply not having
		4965	any other events happening anymore.
		4966
		4967	In libev, events are represented as single bits (such as C<EV_READ> or
		4968	C<EV_TIMER>).
		4969
		4970	=item event library
		4971
		4972	A software package implementing an event model and loop.
		4973
		4974	=item event loop
		4975
		4976	An entity that handles and processes external events and converts them
		4977	into callback invocations.
		4978
		4979	=item event model
		4980
		4981	The model used to describe how an event loop handles and processes
		4982	watchers and events.
		4983
		4984	=item pending
		4985
		4986	A watcher is pending as soon as the corresponding event has been
		4987	detected. See L<WATCHER STATES> for details.
		4988
		4989	=item real time
		4990
		4991	The physical time that is observed. It is apparently strictly monotonic :)
		4992
		4993	=item wall-clock time
		4994
		4995	The time and date as shown on clocks. Unlike real time, it can actually
		4996	be wrong and jump forwards and backwards, e.g. when the you adjust your
		4997	clock.
		4998
		4999	=item watcher
		5000
		5001	A data structure that describes interest in certain events. Watchers need
		5002	to be started (attached to an event loop) before they can receive events.
		5003
		5004	=back
825		5005
826	=head1 AUTHOR	5006	=head1 AUTHOR
827		5007
828	Marc Lehmann <libev@schmorp.de>.	5008	Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael Magnusson.
829		5009

Diff Legend

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