ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libev/ev.pod

Comparing libev/ev.pod (file contents):
Revision 1.30 by root, Fri Nov 23 04:36:03 2007 UTC vs.
Revision 1.319 by root, Fri Oct 22 10:09:12 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_loop (0);
		49
		50	// initialise an io watcher, then start it
		51	// this one will watch for stdin to become readable
		52	ev_io_init (&stdin_watcher, stdin_cb, /*STDIN_FILENO*/ 0, EV_READ);
		53	ev_io_start (loop, &stdin_watcher);
		54
		55	// initialise a timer watcher, then start it
		56	// simple non-repeating 5.5 second timeout
		57	ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
		58	ev_timer_start (loop, &timeout_watcher);
		59
		60	// now wait for events to arrive
		61	ev_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 interetsing 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 ABI mismatches :).
		199
		200	assert (("libev version mismatch",
		201	ev_version_major () == EV_VERSION_MAJOR
		202	&& ev_version_minor () >= EV_VERSION_MINOR));
		203
		204	=item unsigned int ev_supported_backends ()
		205
		206	Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
		207	value) compiled into this binary of libev (independent of their
		208	availability on the system you are running on). See C<ev_default_loop> for
		209	a description of the set values.
		210
		211	Example: make sure we have the epoll method, because yeah this is cool and
		212	a must have and can we have a torrent of it please!!!11
		213
		214	assert (("sorry, no epoll, no sex",
		215	ev_supported_backends () & EVBACKEND_EPOLL));
		216
		217	=item unsigned int ev_recommended_backends ()
		218
		219	Return the set of all backends compiled into this binary of libev and
		220	also recommended for this platform, meaning it will work for most file
		221	descriptor types. This set is often smaller than the one returned by
		222	C<ev_supported_backends>, as for example kqueue is broken on most BSDs
		223	and will not be auto-detected unless you explicitly request it (assuming
		224	you know what you are doing). This is the set of backends that libev will
		225	probe for if you specify no backends explicitly.
		226
		227	=item unsigned int ev_embeddable_backends ()
		228
		229	Returns the set of backends that are embeddable in other event loops. This
		230	value is platform-specific but can include backends not available on the
		231	current system. To find which embeddable backends might be supported on
		232	the current system, you would need to look at C<ev_embeddable_backends ()
		233	& ev_supported_backends ()>, likewise for recommended ones.
		234
		235	See the description of C<ev_embed> watchers for more info.
		236
80	=item ev_set_allocator (void *(*cb)(void *ptr, long size))	237	=item ev_set_allocator (void *(*cb)(void *ptr, long size)) [NOT REENTRANT]
81		238
82	Sets the allocation function to use (the prototype is similar to the	239	Sets the allocation function to use (the prototype is similar - the
83	realloc C function, the semantics are identical). It is used to allocate	240	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	241	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	242	when memory needs to be allocated (C<size != 0>), the library might abort
86	destructive action. The default is your system realloc function.	243	or take some potentially destructive action.
		244
		245	Since some systems (at least OpenBSD and Darwin) fail to implement
		246	correct C<realloc> semantics, libev will use a wrapper around the system
		247	C<realloc> and C<free> functions by default.
87		248
88	You could override this function in high-availability programs to, say,	249	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,	250	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.	251	or even to sleep a while and retry until some memory is available.
91		252
		253	Example: Replace the libev allocator with one that waits a bit and then
		254	retries (example requires a standards-compliant C<realloc>).
		255
		256	static void *
		257	persistent_realloc (void *ptr, size_t size)
		258	{
		259	for (;;)
		260	{
		261	void *newptr = realloc (ptr, size);
		262
		263	if (newptr)
		264	return newptr;
		265
		266	sleep (60);
		267	}
		268	}
		269
		270	...
		271	ev_set_allocator (persistent_realloc);
		272
92	=item ev_set_syserr_cb (void (*cb)(const char *msg));	273	=item ev_set_syserr_cb (void (*cb)(const char *msg)); [NOT REENTRANT]
93		274
94	Set the callback function to call on a retryable syscall error (such	275	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	276	as failed select, poll, epoll_wait). The message is a printable string
96	indicating the system call or subsystem causing the problem. If this	277	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	278	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	279	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	280	requested operation, or, if the condition doesn't go away, do bad stuff
100	(such as abort).	281	(such as abort).
101		282
		283	Example: This is basically the same thing that libev does internally, too.
		284
		285	static void
		286	fatal_error (const char *msg)
		287	{
		288	perror (msg);
		289	abort ();
		290	}
		291
		292	...
		293	ev_set_syserr_cb (fatal_error);
		294
102	=back	295	=back
103		296
104	=head1 FUNCTIONS CONTROLLING THE EVENT LOOP	297	=head1 FUNCTIONS CONTROLLING THE EVENT LOOP
105		298
106	An event loop is described by a C<struct ev_loop *>. The library knows two	299	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	300	I<not> optional in this case unless libev 3 compatibility is disabled, as
108	events, and dynamically created loops which do not.	301	libev 3 had an C<ev_loop> function colliding with the struct name).
109		302
110	If you use threads, a common model is to run the default event loop	303	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	304	supports signals and child events, and dynamically created event loops
112	create, you also create another event loop. Libev itself does no locking	305	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		306
117	=over 4	307	=over 4
118		308
119	=item struct ev_loop *ev_default_loop (unsigned int flags)	309	=item struct ev_loop *ev_default_loop (unsigned int flags)
120		310
121	This will initialise the default event loop if it hasn't been initialised	311	This will initialise the default event loop if it hasn't been initialised
122	yet and return it. If the default loop could not be initialised, returns	312	yet and return it. If the default loop could not be initialised, returns
123	false. If it already was initialised it simply returns it (and ignores the	313	false. If it already was initialised it simply returns it (and ignores the
124	flags).	314	flags. If that is troubling you, check C<ev_backend ()> afterwards).
125		315
126	If you don't know what event loop to use, use the one returned from this	316	If you don't know what event loop to use, use the one returned from this
127	function.	317	function.
128		318
		319	Note that this function is I<not> thread-safe, so if you want to use it
		320	from multiple threads, you have to lock (note also that this is unlikely,
		321	as loops cannot be shared easily between threads anyway).
		322
		323	The default loop is the only loop that can handle C<ev_signal> and
		324	C<ev_child> watchers, and to do this, it always registers a handler
		325	for C<SIGCHLD>. If this is a problem for your application you can either
		326	create a dynamic loop with C<ev_loop_new> that doesn't do that, or you
		327	can simply overwrite the C<SIGCHLD> signal handler I<after> calling
		328	C<ev_default_init>.
		329
129	The flags argument can be used to specify special behaviour or specific	330	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).	331	backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
131		332
132	It supports the following flags:	333	The following flags are supported:
133		334
134	=over 4	335	=over 4
135		336
136	=item C<EVFLAG_AUTO>	337	=item C<EVFLAG_AUTO>
137		338
138	The default flags value. Use this if you have no clue (it's the right	339	The default flags value. Use this if you have no clue (it's the right
139	thing, believe me).	340	thing, believe me).
140		341
141	=item C<EVFLAG_NOENV>	342	=item C<EVFLAG_NOENV>
142		343
143	If this flag bit is ored into the flag value (or the program runs setuid	344	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	345	or setgid) then libev will I<not> look at the environment variable
145	C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will	346	C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
146	override the flags completely if it is found in the environment. This is	347	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	348	useful to try out specific backends to test their performance, or to work
148	around bugs.	349	around bugs.
149		350
		351	=item C<EVFLAG_FORKCHECK>
		352
		353	Instead of calling C<ev_loop_fork> manually after a fork, you can also
		354	make libev check for a fork in each iteration by enabling this flag.
		355
		356	This works by calling C<getpid ()> on every iteration of the loop,
		357	and thus this might slow down your event loop if you do a lot of loop
		358	iterations and little real work, but is usually not noticeable (on my
		359	GNU/Linux system for example, C<getpid> is actually a simple 5-insn sequence
		360	without a system call and thus I<very> fast, but my GNU/Linux system also has
		361	C<pthread_atfork> which is even faster).
		362
		363	The big advantage of this flag is that you can forget about fork (and
		364	forget about forgetting to tell libev about forking) when you use this
		365	flag.
		366
		367	This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
		368	environment variable.
		369
		370	=item C<EVFLAG_NOINOTIFY>
		371
		372	When this flag is specified, then libev will not attempt to use the
		373	I<inotify> API for it's C<ev_stat> watchers. Apart from debugging and
		374	testing, this flag can be useful to conserve inotify file descriptors, as
		375	otherwise each loop using C<ev_stat> watchers consumes one inotify handle.
		376
		377	=item C<EVFLAG_SIGNALFD>
		378
		379	When this flag is specified, then libev will attempt to use the
		380	I<signalfd> API for it's C<ev_signal> (and C<ev_child>) watchers. This API
		381	delivers signals synchronously, which makes it both faster and might make
		382	it possible to get the queued signal data. It can also simplify signal
		383	handling with threads, as long as you properly block signals in your
		384	threads that are not interested in handling them.
		385
		386	Signalfd will not be used by default as this changes your signal mask, and
		387	there are a lot of shoddy libraries and programs (glib's threadpool for
		388	example) that can't properly initialise their signal masks.
		389
150	=item C<EVMETHOD_SELECT> (value 1, portable select backend)	390	=item C<EVBACKEND_SELECT> (value 1, portable select backend)
151		391
152	This is your standard select(2) backend. Not I<completely> standard, as	392	This is your standard select(2) backend. Not I<completely> standard, as
153	libev tries to roll its own fd_set with no limits on the number of fds,	393	libev tries to roll its own fd_set with no limits on the number of fds,
154	but if that fails, expect a fairly low limit on the number of fds when	394	but if that fails, expect a fairly low limit on the number of fds when
155	using this backend. It doesn't scale too well (O(highest_fd)), but its usually	395	using this backend. It doesn't scale too well (O(highest_fd)), but its
156	the fastest backend for a low number of fds.	396	usually the fastest backend for a low number of (low-numbered :) fds.
157		397
		398	To get good performance out of this backend you need a high amount of
		399	parallelism (most of the file descriptors should be busy). If you are
		400	writing a server, you should C<accept ()> in a loop to accept as many
		401	connections as possible during one iteration. You might also want to have
		402	a look at C<ev_set_io_collect_interval ()> to increase the amount of
		403	readiness notifications you get per iteration.
		404
		405	This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
		406	C<writefds> set (and to work around Microsoft Windows bugs, also onto the
		407	C<exceptfds> set on that platform).
		408
158	=item C<EVMETHOD_POLL> (value 2, poll backend, available everywhere except on windows)	409	=item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
159		410
160	And this is your standard poll(2) backend. It's more complicated than	411	And this is your standard poll(2) backend. It's more complicated
161	select, but handles sparse fds better and has no artificial limit on the	412	than select, but handles sparse fds better and has no artificial
162	number of fds you can use (except it will slow down considerably with a	413	limit on the number of fds you can use (except it will slow down
163	lot of inactive fds). It scales similarly to select, i.e. O(total_fds).	414	considerably with a lot of inactive fds). It scales similarly to select,
		415	i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
		416	performance tips.
164		417
		418	This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
		419	C<EV_WRITE> to C<POLLOUT | POLLERR | POLLHUP>.
		420
165	=item C<EVMETHOD_EPOLL> (value 4, Linux)	421	=item C<EVBACKEND_EPOLL> (value 4, Linux)
		422
		423	Use the linux-specific epoll(7) interface (for both pre- and post-2.6.9
		424	kernels).
166		425
167	For few fds, this backend is a bit little slower than poll and select,	426	For few fds, this backend is a bit little slower than poll and select,
168	but it scales phenomenally better. While poll and select usually scale like	427	but it scales phenomenally better. While poll and select usually scale
169	O(total_fds) where n is the total number of fds (or the highest fd), epoll scales	428	like O(total_fds) where n is the total number of fds (or the highest fd),
170	either O(1) or O(active_fds).	429	epoll scales either O(1) or O(active_fds).
171		430
		431	The epoll mechanism deserves honorable mention as the most misdesigned
		432	of the more advanced event mechanisms: mere annoyances include silently
		433	dropping file descriptors, requiring a system call per change per file
		434	descriptor (and unnecessary guessing of parameters), problems with dup and
		435	so on. The biggest issue is fork races, however - if a program forks then
		436	I<both> parent and child process have to recreate the epoll set, which can
		437	take considerable time (one syscall per file descriptor) and is of course
		438	hard to detect.
		439
		440	Epoll is also notoriously buggy - embedding epoll fds I<should> work, but
		441	of course I<doesn't>, and epoll just loves to report events for totally
		442	I<different> file descriptors (even already closed ones, so one cannot
		443	even remove them from the set) than registered in the set (especially
		444	on SMP systems). Libev tries to counter these spurious notifications by
		445	employing an additional generation counter and comparing that against the
		446	events to filter out spurious ones, recreating the set when required. Last
		447	not least, it also refuses to work with some file descriptors which work
		448	perfectly fine with C<select> (files, many character devices...).
		449
172	While stopping and starting an I/O watcher in the same iteration will	450	While stopping, setting and starting an I/O watcher in the same iteration
173	result in some caching, there is still a syscall per such incident	451	will result in some caching, there is still a system call per such
174	(because the fd could point to a different file description now), so its	452	incident (because the same I<file descriptor> could point to a different
175	best to avoid that. Also, dup()ed file descriptors might not work very	453	I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
176	well if you register events for both fds.	454	file descriptors might not work very well if you register events for both
		455	file descriptors.
177		456
		457	Best performance from this backend is achieved by not unregistering all
		458	watchers for a file descriptor until it has been closed, if possible,
		459	i.e. keep at least one watcher active per fd at all times. Stopping and
		460	starting a watcher (without re-setting it) also usually doesn't cause
		461	extra overhead. A fork can both result in spurious notifications as well
		462	as in libev having to destroy and recreate the epoll object, which can
		463	take considerable time and thus should be avoided.
		464
		465	All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
		466	faster than epoll for maybe up to a hundred file descriptors, depending on
		467	the usage. So sad.
		468
		469	While nominally embeddable in other event loops, this feature is broken in
		470	all kernel versions tested so far.
		471
		472	This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
		473	C<EVBACKEND_POLL>.
		474
178	=item C<EVMETHOD_KQUEUE> (value 8, most BSD clones)	475	=item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
179		476
180	Kqueue deserves special mention, as at the time of this writing, it	477	Kqueue deserves special mention, as at the time of this writing, it
181	was broken on all BSDs except NetBSD (usually it doesn't work with	478	was broken on all BSDs except NetBSD (usually it doesn't work reliably
182	anything but sockets and pipes, except on Darwin, where of course its	479	with anything but sockets and pipes, except on Darwin, where of course
183	completely useless). For this reason its not being "autodetected" unless	480	it's completely useless). Unlike epoll, however, whose brokenness
184	you explicitly specify the flags (i.e. you don't use EVFLAG_AUTO).	481	is by design, these kqueue bugs can (and eventually will) be fixed
		482	without API changes to existing programs. For this reason it's not being
		483	"auto-detected" unless you explicitly specify it in the flags (i.e. using
		484	C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
		485	system like NetBSD.
		486
		487	You still can embed kqueue into a normal poll or select backend and use it
		488	only for sockets (after having made sure that sockets work with kqueue on
		489	the target platform). See C<ev_embed> watchers for more info.
185		490
186	It scales in the same way as the epoll backend, but the interface to the	491	It scales in the same way as the epoll backend, but the interface to the
187	kernel is more efficient (which says nothing about its actual speed, of	492	kernel is more efficient (which says nothing about its actual speed, of
188	course). While starting and stopping an I/O watcher does not cause an	493	course). While stopping, setting and starting an I/O watcher does never
189	extra syscall as with epoll, it still adds up to four event changes per	494	cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
190	incident, so its best to avoid that.	495	two event changes per incident. Support for C<fork ()> is very bad (but
		496	sane, unlike epoll) and it drops fds silently in similarly hard-to-detect
		497	cases
191		498
		499	This backend usually performs well under most conditions.
		500
		501	While nominally embeddable in other event loops, this doesn't work
		502	everywhere, so you might need to test for this. And since it is broken
		503	almost everywhere, you should only use it when you have a lot of sockets
		504	(for which it usually works), by embedding it into another event loop
		505	(e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
		506	also broken on OS X)) and, did I mention it, using it only for sockets.
		507
		508	This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
		509	C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
		510	C<NOTE_EOF>.
		511
192	=item C<EVMETHOD_DEVPOLL> (value 16, Solaris 8)	512	=item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
193		513
194	This is not implemented yet (and might never be).	514	This is not implemented yet (and might never be, unless you send me an
		515	implementation). According to reports, C</dev/poll> only supports sockets
		516	and is not embeddable, which would limit the usefulness of this backend
		517	immensely.
195		518
196	=item C<EVMETHOD_PORT> (value 32, Solaris 10)	519	=item C<EVBACKEND_PORT> (value 32, Solaris 10)
197		520
198	This uses the Solaris 10 port mechanism. As with everything on Solaris,	521	This uses the Solaris 10 event port mechanism. As with everything on Solaris,
199	it's really slow, but it still scales very well (O(active_fds)).	522	it's really slow, but it still scales very well (O(active_fds)).
200		523
		524	Please note that Solaris event ports can deliver a lot of spurious
		525	notifications, so you need to use non-blocking I/O or other means to avoid
		526	blocking when no data (or space) is available.
		527
		528	While this backend scales well, it requires one system call per active
		529	file descriptor per loop iteration. For small and medium numbers of file
		530	descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
		531	might perform better.
		532
		533	On the positive side, with the exception of the spurious readiness
		534	notifications, this backend actually performed fully to specification
		535	in all tests and is fully embeddable, which is a rare feat among the
		536	OS-specific backends (I vastly prefer correctness over speed hacks).
		537
		538	This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
		539	C<EVBACKEND_POLL>.
		540
201	=item C<EVMETHOD_ALL>	541	=item C<EVBACKEND_ALL>
202		542
203	Try all backends (even potentially broken ones that wouldn't be tried	543	Try all backends (even potentially broken ones that wouldn't be tried
204	with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as	544	with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
205	C<EVMETHOD_ALL & ~EVMETHOD_KQUEUE>.	545	C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.
		546
		547	It is definitely not recommended to use this flag.
206		548
207	=back	549	=back
208		550
209	If one or more of these are ored into the flags value, then only these	551	If one or more of the backend flags are or'ed into the flags value,
210	backends will be tried (in the reverse order as given here). If none are	552	then only these backends will be tried (in the reverse order as listed
211	specified, most compiled-in backend will be tried, usually in reverse	553	here). If none are specified, all backends in C<ev_recommended_backends
212	order of their flag values :)	554	()> will be tried.
		555
		556	Example: This is the most typical usage.
		557
		558	if (!ev_default_loop (0))
		559	fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
		560
		561	Example: Restrict libev to the select and poll backends, and do not allow
		562	environment settings to be taken into account:
		563
		564	ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
		565
		566	Example: Use whatever libev has to offer, but make sure that kqueue is
		567	used if available (warning, breaks stuff, best use only with your own
		568	private event loop and only if you know the OS supports your types of
		569	fds):
		570
		571	ev_default_loop (ev_recommended_backends () | EVBACKEND_KQUEUE);
213		572
214	=item struct ev_loop *ev_loop_new (unsigned int flags)	573	=item struct ev_loop *ev_loop_new (unsigned int flags)
215		574
216	Similar to C<ev_default_loop>, but always creates a new event loop that is	575	Similar to C<ev_default_loop>, but always creates a new event loop that is
217	always distinct from the default loop. Unlike the default loop, it cannot	576	always distinct from the default loop.
218	handle signal and child watchers, and attempts to do so will be greeted by	577
219	undefined behaviour (or a failed assertion if assertions are enabled).	578	Note that this function I<is> thread-safe, and one common way to use
		579	libev with threads is indeed to create one loop per thread, and using the
		580	default loop in the "main" or "initial" thread.
		581
		582	Example: Try to create a event loop that uses epoll and nothing else.
		583
		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");
220		587
221	=item ev_default_destroy ()	588	=item ev_default_destroy ()
222		589
223	Destroys the default loop again (frees all memory and kernel state	590	Destroys the default loop (frees all memory and kernel state etc.). None
224	etc.). This stops all registered event watchers (by not touching them in	591	of the active event watchers will be stopped in the normal sense, so
225	any way whatsoever, although you cannot rely on this :).	592	e.g. C<ev_is_active> might still return true. It is your responsibility to
		593	either stop all watchers cleanly yourself I<before> calling this function,
		594	or cope with the fact afterwards (which is usually the easiest thing, you
		595	can just ignore the watchers and/or C<free ()> them for example).
		596
		597	Note that certain global state, such as signal state (and installed signal
		598	handlers), will not be freed by this function, and related watchers (such
		599	as signal and child watchers) would need to be stopped manually.
		600
		601	In general it is not advisable to call this function except in the
		602	rare occasion where you really need to free e.g. the signal handling
		603	pipe fds. If you need dynamically allocated loops it is better to use
		604	C<ev_loop_new> and C<ev_loop_destroy>.
226		605
227	=item ev_loop_destroy (loop)	606	=item ev_loop_destroy (loop)
228		607
229	Like C<ev_default_destroy>, but destroys an event loop created by an	608	Like C<ev_default_destroy>, but destroys an event loop created by an
230	earlier call to C<ev_loop_new>.	609	earlier call to C<ev_loop_new>.
231		610
232	=item ev_default_fork ()	611	=item ev_default_fork ()
233		612
		613	This function sets a flag that causes subsequent C<ev_run> iterations
234	This function reinitialises the kernel state for backends that have	614	to reinitialise the kernel state for backends that have one. Despite the
235	one. Despite the name, you can call it anytime, but it makes most sense	615	name, you can call it anytime, but it makes most sense after forking, in
236	after forking, in either the parent or child process (or both, but that	616	the child process (or both child and parent, but that again makes little
237	again makes little sense).	617	sense). You I<must> call it in the child before using any of the libev
		618	functions, and it will only take effect at the next C<ev_run> iteration.
238		619
239	You I<must> call this function in the child process after forking if and	620	Again, you I<have> to call it on I<any> loop that you want to re-use after
240	only if you want to use the event library in both processes. If you just	621	a fork, I<even if you do not plan to use the loop in the parent>. This is
241	fork+exec, you don't have to call it.	622	because some kernel interfaces *cough* I<kqueue> *cough* do funny things
		623	during fork.
		624
		625	On the other hand, you only need to call this function in the child
		626	process if and only if you want to use the event loop in the child. If
		627	you just fork+exec or create a new loop in the child, you don't have to
		628	call it at all (in fact, C<epoll> is so badly broken that it makes a
		629	difference, but libev will usually detect this case on its own and do a
		630	costly reset of the backend).
242		631
243	The function itself is quite fast and it's usually not a problem to call	632	The function itself is quite fast and it's usually not a problem to call
244	it just in case after a fork. To make this easy, the function will fit in	633	it just in case after a fork. To make this easy, the function will fit in
245	quite nicely into a call to C<pthread_atfork>:	634	quite nicely into a call to C<pthread_atfork>:
246		635
…		…
248		637
249	=item ev_loop_fork (loop)	638	=item ev_loop_fork (loop)
250		639
251	Like C<ev_default_fork>, but acts on an event loop created by	640	Like C<ev_default_fork>, but acts on an event loop created by
252	C<ev_loop_new>. Yes, you have to call this on every allocated event loop	641	C<ev_loop_new>. Yes, you have to call this on every allocated event loop
253	after fork, and how you do this is entirely your own problem.	642	after fork that you want to re-use in the child, and how you keep track of
		643	them is entirely your own problem.
254		644
		645	=item int ev_is_default_loop (loop)
		646
		647	Returns true when the given loop is, in fact, the default loop, and false
		648	otherwise.
		649
		650	=item unsigned int ev_iteration (loop)
		651
		652	Returns the current iteration count for the event loop, which is identical
		653	to the number of times libev did poll for new events. It starts at C<0>
		654	and happily wraps around with enough iterations.
		655
		656	This value can sometimes be useful as a generation counter of sorts (it
		657	"ticks" the number of loop iterations), as it roughly corresponds with
		658	C<ev_prepare> and C<ev_check> calls - and is incremented between the
		659	prepare and check phases.
		660
255	=item unsigned int ev_method (loop)	661	=item unsigned int ev_depth (loop)
256		662
		663	Returns the number of times C<ev_run> was entered minus the number of
		664	times C<ev_run> was exited, in other words, the recursion depth.
		665
		666	Outside C<ev_run>, this number is zero. In a callback, this number is
		667	C<1>, unless C<ev_run> was invoked recursively (or from another thread),
		668	in which case it is higher.
		669
		670	Leaving C<ev_run> abnormally (setjmp/longjmp, cancelling the thread
		671	etc.), doesn't count as "exit" - consider this as a hint to avoid such
		672	ungentleman-like behaviour unless it's really convenient.
		673
		674	=item unsigned int ev_backend (loop)
		675
257	Returns one of the C<EVMETHOD_*> flags indicating the event backend in	676	Returns one of the C<EVBACKEND_*> flags indicating the event backend in
258	use.	677	use.
259		678
260	=item ev_tstamp ev_now (loop)	679	=item ev_tstamp ev_now (loop)
261		680
262	Returns the current "event loop time", which is the time the event loop	681	Returns the current "event loop time", which is the time the event loop
263	got events and started processing them. This timestamp does not change	682	received events and started processing them. This timestamp does not
264	as long as callbacks are being processed, and this is also the base time	683	change as long as callbacks are being processed, and this is also the base
265	used for relative timers. You can treat it as the timestamp of the event	684	time used for relative timers. You can treat it as the timestamp of the
266	occuring (or more correctly, the mainloop finding out about it).	685	event occurring (or more correctly, libev finding out about it).
267		686
		687	=item ev_now_update (loop)
		688
		689	Establishes the current time by querying the kernel, updating the time
		690	returned by C<ev_now ()> in the progress. This is a costly operation and
		691	is usually done automatically within C<ev_run ()>.
		692
		693	This function is rarely useful, but when some event callback runs for a
		694	very long time without entering the event loop, updating libev's idea of
		695	the current time is a good idea.
		696
		697	See also L<The special problem of time updates> in the C<ev_timer> section.
		698
		699	=item ev_suspend (loop)
		700
		701	=item ev_resume (loop)
		702
		703	These two functions suspend and resume an event loop, for use when the
		704	loop is not used for a while and timeouts should not be processed.
		705
		706	A typical use case would be an interactive program such as a game: When
		707	the user presses C<^Z> to suspend the game and resumes it an hour later it
		708	would be best to handle timeouts as if no time had actually passed while
		709	the program was suspended. This can be achieved by calling C<ev_suspend>
		710	in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
		711	C<ev_resume> directly afterwards to resume timer processing.
		712
		713	Effectively, all C<ev_timer> watchers will be delayed by the time spend
		714	between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
		715	will be rescheduled (that is, they will lose any events that would have
		716	occurred while suspended).
		717
		718	After calling C<ev_suspend> you B<must not> call I<any> function on the
		719	given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
		720	without a previous call to C<ev_suspend>.
		721
		722	Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
		723	event loop time (see C<ev_now_update>).
		724
268	=item ev_loop (loop, int flags)	725	=item ev_run (loop, int flags)
269		726
270	Finally, this is it, the event handler. This function usually is called	727	Finally, this is it, the event handler. This function usually is called
271	after you initialised all your watchers and you want to start handling	728	after you have initialised all your watchers and you want to start
272	events.	729	handling events. It will ask the operating system for any new events, call
		730	the watcher callbacks, an then repeat the whole process indefinitely: This
		731	is why event loops are called I<loops>.
273		732
274	If the flags argument is specified as 0, it will not return until either	733	If the flags argument is specified as C<0>, it will keep handling events
275	no event watchers are active anymore or C<ev_unloop> was called.	734	until either no event watchers are active anymore or C<ev_break> was
		735	called.
276		736
		737	Please note that an explicit C<ev_break> is usually better than
		738	relying on all watchers to be stopped when deciding when a program has
		739	finished (especially in interactive programs), but having a program
		740	that automatically loops as long as it has to and no longer by virtue
		741	of relying on its watchers stopping correctly, that is truly a thing of
		742	beauty.
		743
277	A flags value of C<EVLOOP_NONBLOCK> will look for new events, will handle	744	A flags value of C<EVRUN_NOWAIT> will look for new events, will handle
278	those events and any outstanding ones, but will not block your process in	745	those events and any already outstanding ones, but will not wait and
279	case there are no events and will return after one iteration of the loop.	746	block your process in case there are no events and will return after one
		747	iteration of the loop. This is sometimes useful to poll and handle new
		748	events while doing lengthy calculations, to keep the program responsive.
280		749
281	A flags value of C<EVLOOP_ONESHOT> will look for new events (waiting if	750	A flags value of C<EVRUN_ONCE> will look for new events (waiting if
282	neccessary) and will handle those and any outstanding ones. It will block	751	necessary) and will handle those and any already outstanding ones. It
283	your process until at least one new event arrives, and will return after	752	will block your process until at least one new event arrives (which could
		753	be an event internal to libev itself, so there is no guarantee that a
		754	user-registered callback will be called), and will return after one
284	one iteration of the loop.	755	iteration of the loop.
285		756
286	This flags value could be used to implement alternative looping	757	This is useful if you are waiting for some external event in conjunction
287	constructs, but the C<prepare> and C<check> watchers provide a better and	758	with something not expressible using other libev watchers (i.e. "roll your
288	more generic mechanism.	759	own C<ev_run>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
		760	usually a better approach for this kind of thing.
289		761
290	Here are the gory details of what ev_loop does:	762	Here are the gory details of what C<ev_run> does:
291		763
292	1. If there are no active watchers (reference count is zero), return.	764	- Increment loop depth.
		765	- Reset the ev_break status.
		766	- Before the first iteration, call any pending watchers.
		767	LOOP:
		768	- If EVFLAG_FORKCHECK was used, check for a fork.
		769	- If a fork was detected (by any means), queue and call all fork watchers.
293	2. Queue and immediately call all prepare watchers.	770	- Queue and call all prepare watchers.
		771	- If ev_break was called, goto FINISH.
294	3. If we have been forked, recreate the kernel state.	772	- If we have been forked, detach and recreate the kernel state
		773	as to not disturb the other process.
295	4. Update the kernel state with all outstanding changes.	774	- Update the kernel state with all outstanding changes.
296	5. Update the "event loop time".	775	- Update the "event loop time" (ev_now ()).
297	6. Calculate for how long to block.	776	- Calculate for how long to sleep or block, if at all
		777	(active idle watchers, EVRUN_NOWAIT or not having
		778	any active watchers at all will result in not sleeping).
		779	- Sleep if the I/O and timer collect interval say so.
		780	- Increment loop iteration counter.
298	7. Block the process, waiting for events.	781	- Block the process, waiting for any events.
		782	- Queue all outstanding I/O (fd) events.
299	8. Update the "event loop time" and do time jump handling.	783	- Update the "event loop time" (ev_now ()), and do time jump adjustments.
300	9. Queue all outstanding timers.	784	- Queue all expired timers.
301	10. Queue all outstanding periodics.	785	- Queue all expired periodics.
302	11. If no events are pending now, queue all idle watchers.	786	- Queue all idle watchers with priority higher than that of pending events.
303	12. Queue all check watchers.	787	- Queue all check watchers.
304	13. Call all queued watchers in reverse order (i.e. check watchers first).	788	- Call all queued watchers in reverse order (i.e. check watchers first).
305	14. If ev_unloop has been called or EVLOOP_ONESHOT or EVLOOP_NONBLOCK	789	Signals and child watchers are implemented as I/O watchers, and will
306	was used, return, otherwise continue with step #1.	790	be handled here by queueing them when their watcher gets executed.
		791	- If ev_break has been called, or EVRUN_ONCE or EVRUN_NOWAIT
		792	were used, or there are no active watchers, goto FINISH, otherwise
		793	continue with step LOOP.
		794	FINISH:
		795	- Reset the ev_break status iff it was EVBREAK_ONE.
		796	- Decrement the loop depth.
		797	- Return.
307		798
		799	Example: Queue some jobs and then loop until no events are outstanding
		800	anymore.
		801
		802	... queue jobs here, make sure they register event watchers as long
		803	... as they still have work to do (even an idle watcher will do..)
		804	ev_run (my_loop, 0);
		805	... jobs done or somebody called unloop. yeah!
		806
308	=item ev_unloop (loop, how)	807	=item ev_break (loop, how)
309		808
310	Can be used to make a call to C<ev_loop> return early (but only after it	809	Can be used to make a call to C<ev_run> return early (but only after it
311	has processed all outstanding events). The C<how> argument must be either	810	has processed all outstanding events). The C<how> argument must be either
312	C<EVUNLOOP_ONE>, which will make the innermost C<ev_loop> call return, or	811	C<EVBREAK_ONE>, which will make the innermost C<ev_run> call return, or
313	C<EVUNLOOP_ALL>, which will make all nested C<ev_loop> calls return.	812	C<EVBREAK_ALL>, which will make all nested C<ev_run> calls return.
		813
		814	This "unloop state" will be cleared when entering C<ev_run> again.
		815
		816	It is safe to call C<ev_break> from outside any C<ev_run> calls. ##TODO##
314		817
315	=item ev_ref (loop)	818	=item ev_ref (loop)
316		819
317	=item ev_unref (loop)	820	=item ev_unref (loop)
318		821
319	Ref/unref can be used to add or remove a reference count on the event	822	Ref/unref can be used to add or remove a reference count on the event
320	loop: Every watcher keeps one reference, and as long as the reference	823	loop: Every watcher keeps one reference, and as long as the reference
321	count is nonzero, C<ev_loop> will not return on its own. If you have	824	count is nonzero, C<ev_run> will not return on its own.
322	a watcher you never unregister that should not keep C<ev_loop> from	825
323	returning, ev_unref() after starting, and ev_ref() before stopping it. For	826	This is useful when you have a watcher that you never intend to
		827	unregister, but that nevertheless should not keep C<ev_run> from
		828	returning. In such a case, call C<ev_unref> after starting, and C<ev_ref>
		829	before stopping it.
		830
324	example, libev itself uses this for its internal signal pipe: It is not	831	As an example, libev itself uses this for its internal signal pipe: It
325	visible to the libev user and should not keep C<ev_loop> from exiting if	832	is not visible to the libev user and should not keep C<ev_run> from
326	no event watchers registered by it are active. It is also an excellent	833	exiting if no event watchers registered by it are active. It is also an
327	way to do this for generic recurring timers or from within third-party	834	excellent way to do this for generic recurring timers or from within
328	libraries. Just remember to I<unref after start> and I<ref before stop>.	835	third-party libraries. Just remember to I<unref after start> and I<ref
		836	before stop> (but only if the watcher wasn't active before, or was active
		837	before, respectively. Note also that libev might stop watchers itself
		838	(e.g. non-repeating timers) in which case you have to C<ev_ref>
		839	in the callback).
		840
		841	Example: Create a signal watcher, but keep it from keeping C<ev_run>
		842	running when nothing else is active.
		843
		844	ev_signal exitsig;
		845	ev_signal_init (&exitsig, sig_cb, SIGINT);
		846	ev_signal_start (loop, &exitsig);
		847	evf_unref (loop);
		848
		849	Example: For some weird reason, unregister the above signal handler again.
		850
		851	ev_ref (loop);
		852	ev_signal_stop (loop, &exitsig);
		853
		854	=item ev_set_io_collect_interval (loop, ev_tstamp interval)
		855
		856	=item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
		857
		858	These advanced functions influence the time that libev will spend waiting
		859	for events. Both time intervals are by default C<0>, meaning that libev
		860	will try to invoke timer/periodic callbacks and I/O callbacks with minimum
		861	latency.
		862
		863	Setting these to a higher value (the C<interval> I<must> be >= C<0>)
		864	allows libev to delay invocation of I/O and timer/periodic callbacks
		865	to increase efficiency of loop iterations (or to increase power-saving
		866	opportunities).
		867
		868	The idea is that sometimes your program runs just fast enough to handle
		869	one (or very few) event(s) per loop iteration. While this makes the
		870	program responsive, it also wastes a lot of CPU time to poll for new
		871	events, especially with backends like C<select ()> which have a high
		872	overhead for the actual polling but can deliver many events at once.
		873
		874	By setting a higher I<io collect interval> you allow libev to spend more
		875	time collecting I/O events, so you can handle more events per iteration,
		876	at the cost of increasing latency. Timeouts (both C<ev_periodic> and
		877	C<ev_timer>) will be not affected. Setting this to a non-null value will
		878	introduce an additional C<ev_sleep ()> call into most loop iterations. The
		879	sleep time ensures that libev will not poll for I/O events more often then
		880	once per this interval, on average.
		881
		882	Likewise, by setting a higher I<timeout collect interval> you allow libev
		883	to spend more time collecting timeouts, at the expense of increased
		884	latency/jitter/inexactness (the watcher callback will be called
		885	later). C<ev_io> watchers will not be affected. Setting this to a non-null
		886	value will not introduce any overhead in libev.
		887
		888	Many (busy) programs can usually benefit by setting the I/O collect
		889	interval to a value near C<0.1> or so, which is often enough for
		890	interactive servers (of course not for games), likewise for timeouts. It
		891	usually doesn't make much sense to set it to a lower value than C<0.01>,
		892	as this approaches the timing granularity of most systems. Note that if
		893	you do transactions with the outside world and you can't increase the
		894	parallelity, then this setting will limit your transaction rate (if you
		895	need to poll once per transaction and the I/O collect interval is 0.01,
		896	then you can't do more than 100 transactions per second).
		897
		898	Setting the I<timeout collect interval> can improve the opportunity for
		899	saving power, as the program will "bundle" timer callback invocations that
		900	are "near" in time together, by delaying some, thus reducing the number of
		901	times the process sleeps and wakes up again. Another useful technique to
		902	reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
		903	they fire on, say, one-second boundaries only.
		904
		905	Example: we only need 0.1s timeout granularity, and we wish not to poll
		906	more often than 100 times per second:
		907
		908	ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
		909	ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
		910
		911	=item ev_invoke_pending (loop)
		912
		913	This call will simply invoke all pending watchers while resetting their
		914	pending state. Normally, C<ev_run> does this automatically when required,
		915	but when overriding the invoke callback this call comes handy. This
		916	function can be invoked from a watcher - this can be useful for example
		917	when you want to do some lengthy calculation and want to pass further
		918	event handling to another thread (you still have to make sure only one
		919	thread executes within C<ev_invoke_pending> or C<ev_run> of course).
		920
		921	=item int ev_pending_count (loop)
		922
		923	Returns the number of pending watchers - zero indicates that no watchers
		924	are pending.
		925
		926	=item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
		927
		928	This overrides the invoke pending functionality of the loop: Instead of
		929	invoking all pending watchers when there are any, C<ev_run> will call
		930	this callback instead. This is useful, for example, when you want to
		931	invoke the actual watchers inside another context (another thread etc.).
		932
		933	If you want to reset the callback, use C<ev_invoke_pending> as new
		934	callback.
		935
		936	=item ev_set_loop_release_cb (loop, void (*release)(EV_P), void (*acquire)(EV_P))
		937
		938	Sometimes you want to share the same loop between multiple threads. This
		939	can be done relatively simply by putting mutex_lock/unlock calls around
		940	each call to a libev function.
		941
		942	However, C<ev_run> can run an indefinite time, so it is not feasible
		943	to wait for it to return. One way around this is to wake up the event
		944	loop via C<ev_break> and C<av_async_send>, another way is to set these
		945	I<release> and I<acquire> callbacks on the loop.
		946
		947	When set, then C<release> will be called just before the thread is
		948	suspended waiting for new events, and C<acquire> is called just
		949	afterwards.
		950
		951	Ideally, C<release> will just call your mutex_unlock function, and
		952	C<acquire> will just call the mutex_lock function again.
		953
		954	While event loop modifications are allowed between invocations of
		955	C<release> and C<acquire> (that's their only purpose after all), no
		956	modifications done will affect the event loop, i.e. adding watchers will
		957	have no effect on the set of file descriptors being watched, or the time
		958	waited. Use an C<ev_async> watcher to wake up C<ev_run> when you want it
		959	to take note of any changes you made.
		960
		961	In theory, threads executing C<ev_run> will be async-cancel safe between
		962	invocations of C<release> and C<acquire>.
		963
		964	See also the locking example in the C<THREADS> section later in this
		965	document.
		966
		967	=item ev_set_userdata (loop, void *data)
		968
		969	=item ev_userdata (loop)
		970
		971	Set and retrieve a single C<void *> associated with a loop. When
		972	C<ev_set_userdata> has never been called, then C<ev_userdata> returns
		973	C<0.>
		974
		975	These two functions can be used to associate arbitrary data with a loop,
		976	and are intended solely for the C<invoke_pending_cb>, C<release> and
		977	C<acquire> callbacks described above, but of course can be (ab-)used for
		978	any other purpose as well.
		979
		980	=item ev_verify (loop)
		981
		982	This function only does something when C<EV_VERIFY> support has been
		983	compiled in, which is the default for non-minimal builds. It tries to go
		984	through all internal structures and checks them for validity. If anything
		985	is found to be inconsistent, it will print an error message to standard
		986	error and call C<abort ()>.
		987
		988	This can be used to catch bugs inside libev itself: under normal
		989	circumstances, this function will never abort as of course libev keeps its
		990	data structures consistent.
329		991
330	=back	992	=back
331		993
		994
332	=head1 ANATOMY OF A WATCHER	995	=head1 ANATOMY OF A WATCHER
333		996
		997	In the following description, uppercase C<TYPE> in names stands for the
		998	watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
		999	watchers and C<ev_io_start> for I/O watchers.
		1000
334	A watcher is a structure that you create and register to record your	1001	A watcher is an opaque structure that you allocate and register to record
335	interest in some event. For instance, if you want to wait for STDIN to	1002	your interest in some event. To make a concrete example, imagine you want
336	become readable, you would create an C<ev_io> watcher for that:	1003	to wait for STDIN to become readable, you would create an C<ev_io> watcher
		1004	for that:
337		1005
338	static void my_cb (struct ev_loop *loop, struct ev_io *w, int revents)	1006	static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
339	{	1007	{
340	ev_io_stop (w);	1008	ev_io_stop (w);
341	ev_unloop (loop, EVUNLOOP_ALL);	1009	ev_break (loop, EVBREAK_ALL);
342	}	1010	}
343		1011
344	struct ev_loop *loop = ev_default_loop (0);	1012	struct ev_loop *loop = ev_default_loop (0);
		1013
345	struct ev_io stdin_watcher;	1014	ev_io stdin_watcher;
		1015
346	ev_init (&stdin_watcher, my_cb);	1016	ev_init (&stdin_watcher, my_cb);
347	ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);	1017	ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
348	ev_io_start (loop, &stdin_watcher);	1018	ev_io_start (loop, &stdin_watcher);
		1019
349	ev_loop (loop, 0);	1020	ev_run (loop, 0);
350		1021
351	As you can see, you are responsible for allocating the memory for your	1022	As you can see, you are responsible for allocating the memory for your
352	watcher structures (and it is usually a bad idea to do this on the stack,	1023	watcher structures (and it is I<usually> a bad idea to do this on the
353	although this can sometimes be quite valid).	1024	stack).
354		1025
		1026	Each watcher has an associated watcher structure (called C<struct ev_TYPE>
		1027	or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
		1028
355	Each watcher structure must be initialised by a call to C<ev_init	1029	Each watcher structure must be initialised by a call to C<ev_init (watcher
356	(watcher *, callback)>, which expects a callback to be provided. This	1030	*, callback)>, which expects a callback to be provided. This callback is
357	callback gets invoked each time the event occurs (or, in the case of io	1031	invoked each time the event occurs (or, in the case of I/O watchers, each
358	watchers, each time the event loop detects that the file descriptor given	1032	time the event loop detects that the file descriptor given is readable
359	is readable and/or writable).	1033	and/or writable).
360		1034
361	Each watcher type has its own C<< ev_<type>_set (watcher *, ...) >> macro	1035	Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
362	with arguments specific to this watcher type. There is also a macro	1036	macro to configure it, with arguments specific to the watcher type. There
363	to combine initialisation and setting in one call: C<< ev_<type>_init	1037	is also a macro to combine initialisation and setting in one call: C<<
364	(watcher *, callback, ...) >>.	1038	ev_TYPE_init (watcher *, callback, ...) >>.
365		1039
366	To make the watcher actually watch out for events, you have to start it	1040	To make the watcher actually watch out for events, you have to start it
367	with a watcher-specific start function (C<< ev_<type>_start (loop, watcher	1041	with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
368	*) >>), and you can stop watching for events at any time by calling the	1042	*) >>), and you can stop watching for events at any time by calling the
369	corresponding stop function (C<< ev_<type>_stop (loop, watcher *) >>.	1043	corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
370		1044
371	As long as your watcher is active (has been started but not stopped) you	1045	As long as your watcher is active (has been started but not stopped) you
372	must not touch the values stored in it. Most specifically you must never	1046	must not touch the values stored in it. Most specifically you must never
373	reinitialise it or call its set method.	1047	reinitialise it or call its C<ev_TYPE_set> macro.
374
375	You can check whether an event is active by calling the C<ev_is_active
376	(watcher *)> macro. To see whether an event is outstanding (but the
377	callback for it has not been called yet) you can use the C<ev_is_pending
378	(watcher *)> macro.
379		1048
380	Each and every callback receives the event loop pointer as first, the	1049	Each and every callback receives the event loop pointer as first, the
381	registered watcher structure as second, and a bitset of received events as	1050	registered watcher structure as second, and a bitset of received events as
382	third argument.	1051	third argument.
383		1052
…		…
392	=item C<EV_WRITE>	1061	=item C<EV_WRITE>
393		1062
394	The file descriptor in the C<ev_io> watcher has become readable and/or	1063	The file descriptor in the C<ev_io> watcher has become readable and/or
395	writable.	1064	writable.
396		1065
397	=item C<EV_TIMEOUT>	1066	=item C<EV_TIMER>
398		1067
399	The C<ev_timer> watcher has timed out.	1068	The C<ev_timer> watcher has timed out.
400		1069
401	=item C<EV_PERIODIC>	1070	=item C<EV_PERIODIC>
402		1071
…		…
408		1077
409	=item C<EV_CHILD>	1078	=item C<EV_CHILD>
410		1079
411	The pid specified in the C<ev_child> watcher has received a status change.	1080	The pid specified in the C<ev_child> watcher has received a status change.
412		1081
		1082	=item C<EV_STAT>
		1083
		1084	The path specified in the C<ev_stat> watcher changed its attributes somehow.
		1085
413	=item C<EV_IDLE>	1086	=item C<EV_IDLE>
414		1087
415	The C<ev_idle> watcher has determined that you have nothing better to do.	1088	The C<ev_idle> watcher has determined that you have nothing better to do.
416		1089
417	=item C<EV_PREPARE>	1090	=item C<EV_PREPARE>
418		1091
419	=item C<EV_CHECK>	1092	=item C<EV_CHECK>
420		1093
421	All C<ev_prepare> watchers are invoked just I<before> C<ev_loop> starts	1094	All C<ev_prepare> watchers are invoked just I<before> C<ev_run> starts
422	to gather new events, and all C<ev_check> watchers are invoked just after	1095	to gather new events, and all C<ev_check> watchers are invoked just after
423	C<ev_loop> has gathered them, but before it invokes any callbacks for any	1096	C<ev_run> has gathered them, but before it invokes any callbacks for any
424	received events. Callbacks of both watcher types can start and stop as	1097	received events. Callbacks of both watcher types can start and stop as
425	many watchers as they want, and all of them will be taken into account	1098	many watchers as they want, and all of them will be taken into account
426	(for example, a C<ev_prepare> watcher might start an idle watcher to keep	1099	(for example, a C<ev_prepare> watcher might start an idle watcher to keep
427	C<ev_loop> from blocking).	1100	C<ev_run> from blocking).
		1101
		1102	=item C<EV_EMBED>
		1103
		1104	The embedded event loop specified in the C<ev_embed> watcher needs attention.
		1105
		1106	=item C<EV_FORK>
		1107
		1108	The event loop has been resumed in the child process after fork (see
		1109	C<ev_fork>).
		1110
		1111	=item C<EV_ASYNC>
		1112
		1113	The given async watcher has been asynchronously notified (see C<ev_async>).
		1114
		1115	=item C<EV_CUSTOM>
		1116
		1117	Not ever sent (or otherwise used) by libev itself, but can be freely used
		1118	by libev users to signal watchers (e.g. via C<ev_feed_event>).
428		1119
429	=item C<EV_ERROR>	1120	=item C<EV_ERROR>
430		1121
431	An unspecified error has occured, the watcher has been stopped. This might	1122	An unspecified error has occurred, the watcher has been stopped. This might
432	happen because the watcher could not be properly started because libev	1123	happen because the watcher could not be properly started because libev
433	ran out of memory, a file descriptor was found to be closed or any other	1124	ran out of memory, a file descriptor was found to be closed or any other
		1125	problem. Libev considers these application bugs.
		1126
434	problem. You best act on it by reporting the problem and somehow coping	1127	You best act on it by reporting the problem and somehow coping with the
435	with the watcher being stopped.	1128	watcher being stopped. Note that well-written programs should not receive
		1129	an error ever, so when your watcher receives it, this usually indicates a
		1130	bug in your program.
436		1131
437	Libev will usually signal a few "dummy" events together with an error,	1132	Libev will usually signal a few "dummy" events together with an error, for
438	for example it might indicate that a fd is readable or writable, and if	1133	example it might indicate that a fd is readable or writable, and if your
439	your callbacks is well-written it can just attempt the operation and cope	1134	callbacks is well-written it can just attempt the operation and cope with
440	with the error from read() or write(). This will not work in multithreaded	1135	the error from read() or write(). This will not work in multi-threaded
441	programs, though, so beware.	1136	programs, though, as the fd could already be closed and reused for another
		1137	thing, so beware.
442		1138
443	=back	1139	=back
444		1140
		1141	=head2 WATCHER STATES
		1142
		1143	There are various watcher states mentioned throughout this manual -
		1144	active, pending and so on. In this section these states and the rules to
		1145	transition between them will be described in more detail - and while these
		1146	rules might look complicated, they usually do "the right thing".
		1147
		1148	=over 4
		1149
		1150	=item initialiased
		1151
		1152	Before a watcher can be registered with the event looop it has to be
		1153	initialised. This can be done with a call to C<ev_TYPE_init>, or calls to
		1154	C<ev_init> followed by the watcher-specific C<ev_TYPE_set> function.
		1155
		1156	In this state it is simply some block of memory that is suitable for use
		1157	in an event loop. It can be moved around, freed, reused etc. at will.
		1158
		1159	=item started/running/active
		1160
		1161	Once a watcher has been started with a call to C<ev_TYPE_start> it becomes
		1162	property of the event loop, and is actively waiting for events. While in
		1163	this state it cannot be accessed (except in a few documented ways), moved,
		1164	freed or anything else - the only legal thing is to keep a pointer to it,
		1165	and call libev functions on it that are documented to work on active watchers.
		1166
		1167	=item pending
		1168
		1169	If a watcher is active and libev determines that an event it is interested
		1170	in has occurred (such as a timer expiring), it will become pending. It will
		1171	stay in this pending state until either it is stopped or its callback is
		1172	about to be invoked, so it is not normally pending inside the watcher
		1173	callback.
		1174
		1175	The watcher might or might not be active while it is pending (for example,
		1176	an expired non-repeating timer can be pending but no longer active). If it
		1177	is stopped, it can be freely accessed (e.g. by calling C<ev_TYPE_set>),
		1178	but it is still property of the event loop at this time, so cannot be
		1179	moved, freed or reused. And if it is active the rules described in the
		1180	previous item still apply.
		1181
		1182	It is also possible to feed an event on a watcher that is not active (e.g.
		1183	via C<ev_feed_event>), in which case it becomes pending without being
		1184	active.
		1185
		1186	=item stopped
		1187
		1188	A watcher can be stopped implicitly by libev (in which case it might still
		1189	be pending), or explicitly by calling its C<ev_TYPE_stop> function. The
		1190	latter will clear any pending state the watcher might be in, regardless
		1191	of whether it was active or not, so stopping a watcher explicitly before
		1192	freeing it is often a good idea.
		1193
		1194	While stopped (and not pending) the watcher is essentially in the
		1195	initialised state, that is it can be reused, moved, modified in any way
		1196	you wish.
		1197
		1198	=back
		1199
		1200	=head2 GENERIC WATCHER FUNCTIONS
		1201
		1202	=over 4
		1203
		1204	=item C<ev_init> (ev_TYPE *watcher, callback)
		1205
		1206	This macro initialises the generic portion of a watcher. The contents
		1207	of the watcher object can be arbitrary (so C<malloc> will do). Only
		1208	the generic parts of the watcher are initialised, you I<need> to call
		1209	the type-specific C<ev_TYPE_set> macro afterwards to initialise the
		1210	type-specific parts. For each type there is also a C<ev_TYPE_init> macro
		1211	which rolls both calls into one.
		1212
		1213	You can reinitialise a watcher at any time as long as it has been stopped
		1214	(or never started) and there are no pending events outstanding.
		1215
		1216	The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
		1217	int revents)>.
		1218
		1219	Example: Initialise an C<ev_io> watcher in two steps.
		1220
		1221	ev_io w;
		1222	ev_init (&w, my_cb);
		1223	ev_io_set (&w, STDIN_FILENO, EV_READ);
		1224
		1225	=item C<ev_TYPE_set> (ev_TYPE *watcher, [args])
		1226
		1227	This macro initialises the type-specific parts of a watcher. You need to
		1228	call C<ev_init> at least once before you call this macro, but you can
		1229	call C<ev_TYPE_set> any number of times. You must not, however, call this
		1230	macro on a watcher that is active (it can be pending, however, which is a
		1231	difference to the C<ev_init> macro).
		1232
		1233	Although some watcher types do not have type-specific arguments
		1234	(e.g. C<ev_prepare>) you still need to call its C<set> macro.
		1235
		1236	See C<ev_init>, above, for an example.
		1237
		1238	=item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
		1239
		1240	This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
		1241	calls into a single call. This is the most convenient method to initialise
		1242	a watcher. The same limitations apply, of course.
		1243
		1244	Example: Initialise and set an C<ev_io> watcher in one step.
		1245
		1246	ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
		1247
		1248	=item C<ev_TYPE_start> (loop, ev_TYPE *watcher)
		1249
		1250	Starts (activates) the given watcher. Only active watchers will receive
		1251	events. If the watcher is already active nothing will happen.
		1252
		1253	Example: Start the C<ev_io> watcher that is being abused as example in this
		1254	whole section.
		1255
		1256	ev_io_start (EV_DEFAULT_UC, &w);
		1257
		1258	=item C<ev_TYPE_stop> (loop, ev_TYPE *watcher)
		1259
		1260	Stops the given watcher if active, and clears the pending status (whether
		1261	the watcher was active or not).
		1262
		1263	It is possible that stopped watchers are pending - for example,
		1264	non-repeating timers are being stopped when they become pending - but
		1265	calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
		1266	pending. If you want to free or reuse the memory used by the watcher it is
		1267	therefore a good idea to always call its C<ev_TYPE_stop> function.
		1268
		1269	=item bool ev_is_active (ev_TYPE *watcher)
		1270
		1271	Returns a true value iff the watcher is active (i.e. it has been started
		1272	and not yet been stopped). As long as a watcher is active you must not modify
		1273	it.
		1274
		1275	=item bool ev_is_pending (ev_TYPE *watcher)
		1276
		1277	Returns a true value iff the watcher is pending, (i.e. it has outstanding
		1278	events but its callback has not yet been invoked). As long as a watcher
		1279	is pending (but not active) you must not call an init function on it (but
		1280	C<ev_TYPE_set> is safe), you must not change its priority, and you must
		1281	make sure the watcher is available to libev (e.g. you cannot C<free ()>
		1282	it).
		1283
		1284	=item callback ev_cb (ev_TYPE *watcher)
		1285
		1286	Returns the callback currently set on the watcher.
		1287
		1288	=item ev_cb_set (ev_TYPE *watcher, callback)
		1289
		1290	Change the callback. You can change the callback at virtually any time
		1291	(modulo threads).
		1292
		1293	=item ev_set_priority (ev_TYPE *watcher, int priority)
		1294
		1295	=item int ev_priority (ev_TYPE *watcher)
		1296
		1297	Set and query the priority of the watcher. The priority is a small
		1298	integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
		1299	(default: C<-2>). Pending watchers with higher priority will be invoked
		1300	before watchers with lower priority, but priority will not keep watchers
		1301	from being executed (except for C<ev_idle> watchers).
		1302
		1303	If you need to suppress invocation when higher priority events are pending
		1304	you need to look at C<ev_idle> watchers, which provide this functionality.
		1305
		1306	You I<must not> change the priority of a watcher as long as it is active or
		1307	pending.
		1308
		1309	Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
		1310	fine, as long as you do not mind that the priority value you query might
		1311	or might not have been clamped to the valid range.
		1312
		1313	The default priority used by watchers when no priority has been set is
		1314	always C<0>, which is supposed to not be too high and not be too low :).
		1315
		1316	See L<WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
		1317	priorities.
		1318
		1319	=item ev_invoke (loop, ev_TYPE *watcher, int revents)
		1320
		1321	Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
		1322	C<loop> nor C<revents> need to be valid as long as the watcher callback
		1323	can deal with that fact, as both are simply passed through to the
		1324	callback.
		1325
		1326	=item int ev_clear_pending (loop, ev_TYPE *watcher)
		1327
		1328	If the watcher is pending, this function clears its pending status and
		1329	returns its C<revents> bitset (as if its callback was invoked). If the
		1330	watcher isn't pending it does nothing and returns C<0>.
		1331
		1332	Sometimes it can be useful to "poll" a watcher instead of waiting for its
		1333	callback to be invoked, which can be accomplished with this function.
		1334
		1335	=item ev_feed_event (loop, ev_TYPE *watcher, int revents)
		1336
		1337	Feeds the given event set into the event loop, as if the specified event
		1338	had happened for the specified watcher (which must be a pointer to an
		1339	initialised but not necessarily started event watcher). Obviously you must
		1340	not free the watcher as long as it has pending events.
		1341
		1342	Stopping the watcher, letting libev invoke it, or calling
		1343	C<ev_clear_pending> will clear the pending event, even if the watcher was
		1344	not started in the first place.
		1345
		1346	See also C<ev_feed_fd_event> and C<ev_feed_signal_event> for related
		1347	functions that do not need a watcher.
		1348
		1349	=back
		1350
		1351
445	=head2 ASSOCIATING CUSTOM DATA WITH A WATCHER	1352	=head2 ASSOCIATING CUSTOM DATA WITH A WATCHER
446		1353
447	Each watcher has, by default, a member C<void *data> that you can change	1354	Each watcher has, by default, a member C<void *data> that you can change
448	and read at any time, libev will completely ignore it. This can be used	1355	and read at any time: libev will completely ignore it. This can be used
449	to associate arbitrary data with your watcher. If you need more data and	1356	to associate arbitrary data with your watcher. If you need more data and
450	don't want to allocate memory and store a pointer to it in that data	1357	don't want to allocate memory and store a pointer to it in that data
451	member, you can also "subclass" the watcher type and provide your own	1358	member, you can also "subclass" the watcher type and provide your own
452	data:	1359	data:
453		1360
454	struct my_io	1361	struct my_io
455	{	1362	{
456	struct ev_io io;	1363	ev_io io;
457	int otherfd;	1364	int otherfd;
458	void *somedata;	1365	void *somedata;
459	struct whatever *mostinteresting;	1366	struct whatever *mostinteresting;
460	}	1367	};
		1368
		1369	...
		1370	struct my_io w;
		1371	ev_io_init (&w.io, my_cb, fd, EV_READ);
461		1372
462	And since your callback will be called with a pointer to the watcher, you	1373	And since your callback will be called with a pointer to the watcher, you
463	can cast it back to your own type:	1374	can cast it back to your own type:
464		1375
465	static void my_cb (struct ev_loop *loop, struct ev_io *w_, int revents)	1376	static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
466	{	1377	{
467	struct my_io *w = (struct my_io *)w_;	1378	struct my_io *w = (struct my_io *)w_;
468	...	1379	...
469	}	1380	}
470		1381
471	More interesting and less C-conformant ways of catsing your callback type	1382	More interesting and less C-conformant ways of casting your callback type
472	have been omitted....	1383	instead have been omitted.
		1384
		1385	Another common scenario is to use some data structure with multiple
		1386	embedded watchers:
		1387
		1388	struct my_biggy
		1389	{
		1390	int some_data;
		1391	ev_timer t1;
		1392	ev_timer t2;
		1393	}
		1394
		1395	In this case getting the pointer to C<my_biggy> is a bit more
		1396	complicated: Either you store the address of your C<my_biggy> struct
		1397	in the C<data> member of the watcher (for woozies), or you need to use
		1398	some pointer arithmetic using C<offsetof> inside your watchers (for real
		1399	programmers):
		1400
		1401	#include <stddef.h>
		1402
		1403	static void
		1404	t1_cb (EV_P_ ev_timer *w, int revents)
		1405	{
		1406	struct my_biggy big = (struct my_biggy *)
		1407	(((char *)w) - offsetof (struct my_biggy, t1));
		1408	}
		1409
		1410	static void
		1411	t2_cb (EV_P_ ev_timer *w, int revents)
		1412	{
		1413	struct my_biggy big = (struct my_biggy *)
		1414	(((char *)w) - offsetof (struct my_biggy, t2));
		1415	}
		1416
		1417	=head2 WATCHER PRIORITY MODELS
		1418
		1419	Many event loops support I<watcher priorities>, which are usually small
		1420	integers that influence the ordering of event callback invocation
		1421	between watchers in some way, all else being equal.
		1422
		1423	In libev, Watcher priorities can be set using C<ev_set_priority>. See its
		1424	description for the more technical details such as the actual priority
		1425	range.
		1426
		1427	There are two common ways how these these priorities are being interpreted
		1428	by event loops:
		1429
		1430	In the more common lock-out model, higher priorities "lock out" invocation
		1431	of lower priority watchers, which means as long as higher priority
		1432	watchers receive events, lower priority watchers are not being invoked.
		1433
		1434	The less common only-for-ordering model uses priorities solely to order
		1435	callback invocation within a single event loop iteration: Higher priority
		1436	watchers are invoked before lower priority ones, but they all get invoked
		1437	before polling for new events.
		1438
		1439	Libev uses the second (only-for-ordering) model for all its watchers
		1440	except for idle watchers (which use the lock-out model).
		1441
		1442	The rationale behind this is that implementing the lock-out model for
		1443	watchers is not well supported by most kernel interfaces, and most event
		1444	libraries will just poll for the same events again and again as long as
		1445	their callbacks have not been executed, which is very inefficient in the
		1446	common case of one high-priority watcher locking out a mass of lower
		1447	priority ones.
		1448
		1449	Static (ordering) priorities are most useful when you have two or more
		1450	watchers handling the same resource: a typical usage example is having an
		1451	C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
		1452	timeouts. Under load, data might be received while the program handles
		1453	other jobs, but since timers normally get invoked first, the timeout
		1454	handler will be executed before checking for data. In that case, giving
		1455	the timer a lower priority than the I/O watcher ensures that I/O will be
		1456	handled first even under adverse conditions (which is usually, but not
		1457	always, what you want).
		1458
		1459	Since idle watchers use the "lock-out" model, meaning that idle watchers
		1460	will only be executed when no same or higher priority watchers have
		1461	received events, they can be used to implement the "lock-out" model when
		1462	required.
		1463
		1464	For example, to emulate how many other event libraries handle priorities,
		1465	you can associate an C<ev_idle> watcher to each such watcher, and in
		1466	the normal watcher callback, you just start the idle watcher. The real
		1467	processing is done in the idle watcher callback. This causes libev to
		1468	continuously poll and process kernel event data for the watcher, but when
		1469	the lock-out case is known to be rare (which in turn is rare :), this is
		1470	workable.
		1471
		1472	Usually, however, the lock-out model implemented that way will perform
		1473	miserably under the type of load it was designed to handle. In that case,
		1474	it might be preferable to stop the real watcher before starting the
		1475	idle watcher, so the kernel will not have to process the event in case
		1476	the actual processing will be delayed for considerable time.
		1477
		1478	Here is an example of an I/O watcher that should run at a strictly lower
		1479	priority than the default, and which should only process data when no
		1480	other events are pending:
		1481
		1482	ev_idle idle; // actual processing watcher
		1483	ev_io io; // actual event watcher
		1484
		1485	static void
		1486	io_cb (EV_P_ ev_io *w, int revents)
		1487	{
		1488	// stop the I/O watcher, we received the event, but
		1489	// are not yet ready to handle it.
		1490	ev_io_stop (EV_A_ w);
		1491
		1492	// start the idle watcher to handle the actual event.
		1493	// it will not be executed as long as other watchers
		1494	// with the default priority are receiving events.
		1495	ev_idle_start (EV_A_ &idle);
		1496	}
		1497
		1498	static void
		1499	idle_cb (EV_P_ ev_idle *w, int revents)
		1500	{
		1501	// actual processing
		1502	read (STDIN_FILENO, ...);
		1503
		1504	// have to start the I/O watcher again, as
		1505	// we have handled the event
		1506	ev_io_start (EV_P_ &io);
		1507	}
		1508
		1509	// initialisation
		1510	ev_idle_init (&idle, idle_cb);
		1511	ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
		1512	ev_io_start (EV_DEFAULT_ &io);
		1513
		1514	In the "real" world, it might also be beneficial to start a timer, so that
		1515	low-priority connections can not be locked out forever under load. This
		1516	enables your program to keep a lower latency for important connections
		1517	during short periods of high load, while not completely locking out less
		1518	important ones.
473		1519
474		1520
475	=head1 WATCHER TYPES	1521	=head1 WATCHER TYPES
476		1522
477	This section describes each watcher in detail, but will not repeat	1523	This section describes each watcher in detail, but will not repeat
478	information given in the last section.	1524	information given in the last section. Any initialisation/set macros,
		1525	functions and members specific to the watcher type are explained.
479		1526
		1527	Members are additionally marked with either I<[read-only]>, meaning that,
		1528	while the watcher is active, you can look at the member and expect some
		1529	sensible content, but you must not modify it (you can modify it while the
		1530	watcher is stopped to your hearts content), or I<[read-write]>, which
		1531	means you can expect it to have some sensible content while the watcher
		1532	is active, but you can also modify it. Modifying it may not do something
		1533	sensible or take immediate effect (or do anything at all), but libev will
		1534	not crash or malfunction in any way.
		1535
		1536
480	=head2 C<ev_io> - is this file descriptor readable or writable	1537	=head2 C<ev_io> - is this file descriptor readable or writable?
481		1538
482	I/O watchers check whether a file descriptor is readable or writable	1539	I/O watchers check whether a file descriptor is readable or writable
483	in each iteration of the event loop (This behaviour is called	1540	in each iteration of the event loop, or, more precisely, when reading
484	level-triggering because you keep receiving events as long as the	1541	would not block the process and writing would at least be able to write
485	condition persists. Remember you can stop the watcher if you don't want to	1542	some data. This behaviour is called level-triggering because you keep
486	act on the event and neither want to receive future events).	1543	receiving events as long as the condition persists. Remember you can stop
		1544	the watcher if you don't want to act on the event and neither want to
		1545	receive future events.
487		1546
488	In general you can register as many read and/or write event watchers per	1547	In general you can register as many read and/or write event watchers per
489	fd as you want (as long as you don't confuse yourself). Setting all file	1548	fd as you want (as long as you don't confuse yourself). Setting all file
490	descriptors to non-blocking mode is also usually a good idea (but not	1549	descriptors to non-blocking mode is also usually a good idea (but not
491	required if you know what you are doing).	1550	required if you know what you are doing).
492		1551
493	You have to be careful with dup'ed file descriptors, though. Some backends	1552	If you cannot use non-blocking mode, then force the use of a
494	(the linux epoll backend is a notable example) cannot handle dup'ed file	1553	known-to-be-good backend (at the time of this writing, this includes only
495	descriptors correctly if you register interest in two or more fds pointing	1554	C<EVBACKEND_SELECT> and C<EVBACKEND_POLL>). The same applies to file
496	to the same underlying file/socket etc. description (that is, they share	1555	descriptors for which non-blocking operation makes no sense (such as
497	the same underlying "file open").	1556	files) - libev doesn't guarantee any specific behaviour in that case.
498		1557
499	If you must do this, then force the use of a known-to-be-good backend	1558	Another thing you have to watch out for is that it is quite easy to
500	(at the time of this writing, this includes only EVMETHOD_SELECT and	1559	receive "spurious" readiness notifications, that is your callback might
501	EVMETHOD_POLL).	1560	be called with C<EV_READ> but a subsequent C<read>(2) will actually block
		1561	because there is no data. Not only are some backends known to create a
		1562	lot of those (for example Solaris ports), it is very easy to get into
		1563	this situation even with a relatively standard program structure. Thus
		1564	it is best to always use non-blocking I/O: An extra C<read>(2) returning
		1565	C<EAGAIN> is far preferable to a program hanging until some data arrives.
		1566
		1567	If you cannot run the fd in non-blocking mode (for example you should
		1568	not play around with an Xlib connection), then you have to separately
		1569	re-test whether a file descriptor is really ready with a known-to-be good
		1570	interface such as poll (fortunately in our Xlib example, Xlib already
		1571	does this on its own, so its quite safe to use). Some people additionally
		1572	use C<SIGALRM> and an interval timer, just to be sure you won't block
		1573	indefinitely.
		1574
		1575	But really, best use non-blocking mode.
		1576
		1577	=head3 The special problem of disappearing file descriptors
		1578
		1579	Some backends (e.g. kqueue, epoll) need to be told about closing a file
		1580	descriptor (either due to calling C<close> explicitly or any other means,
		1581	such as C<dup2>). The reason is that you register interest in some file
		1582	descriptor, but when it goes away, the operating system will silently drop
		1583	this interest. If another file descriptor with the same number then is
		1584	registered with libev, there is no efficient way to see that this is, in
		1585	fact, a different file descriptor.
		1586
		1587	To avoid having to explicitly tell libev about such cases, libev follows
		1588	the following policy: Each time C<ev_io_set> is being called, libev
		1589	will assume that this is potentially a new file descriptor, otherwise
		1590	it is assumed that the file descriptor stays the same. That means that
		1591	you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
		1592	descriptor even if the file descriptor number itself did not change.
		1593
		1594	This is how one would do it normally anyway, the important point is that
		1595	the libev application should not optimise around libev but should leave
		1596	optimisations to libev.
		1597
		1598	=head3 The special problem of dup'ed file descriptors
		1599
		1600	Some backends (e.g. epoll), cannot register events for file descriptors,
		1601	but only events for the underlying file descriptions. That means when you
		1602	have C<dup ()>'ed file descriptors or weirder constellations, and register
		1603	events for them, only one file descriptor might actually receive events.
		1604
		1605	There is no workaround possible except not registering events
		1606	for potentially C<dup ()>'ed file descriptors, or to resort to
		1607	C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
		1608
		1609	=head3 The special problem of fork
		1610
		1611	Some backends (epoll, kqueue) do not support C<fork ()> at all or exhibit
		1612	useless behaviour. Libev fully supports fork, but needs to be told about
		1613	it in the child.
		1614
		1615	To support fork in your programs, you either have to call
		1616	C<ev_default_fork ()> or C<ev_loop_fork ()> after a fork in the child,
		1617	enable C<EVFLAG_FORKCHECK>, or resort to C<EVBACKEND_SELECT> or
		1618	C<EVBACKEND_POLL>.
		1619
		1620	=head3 The special problem of SIGPIPE
		1621
		1622	While not really specific to libev, it is easy to forget about C<SIGPIPE>:
		1623	when writing to a pipe whose other end has been closed, your program gets
		1624	sent a SIGPIPE, which, by default, aborts your program. For most programs
		1625	this is sensible behaviour, for daemons, this is usually undesirable.
		1626
		1627	So when you encounter spurious, unexplained daemon exits, make sure you
		1628	ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
		1629	somewhere, as that would have given you a big clue).
		1630
		1631	=head3 The special problem of accept()ing when you can't
		1632
		1633	Many implementations of the POSIX C<accept> function (for example,
		1634	found in post-2004 Linux) have the peculiar behaviour of not removing a
		1635	connection from the pending queue in all error cases.
		1636
		1637	For example, larger servers often run out of file descriptors (because
		1638	of resource limits), causing C<accept> to fail with C<ENFILE> but not
		1639	rejecting the connection, leading to libev signalling readiness on
		1640	the next iteration again (the connection still exists after all), and
		1641	typically causing the program to loop at 100% CPU usage.
		1642
		1643	Unfortunately, the set of errors that cause this issue differs between
		1644	operating systems, there is usually little the app can do to remedy the
		1645	situation, and no known thread-safe method of removing the connection to
		1646	cope with overload is known (to me).
		1647
		1648	One of the easiest ways to handle this situation is to just ignore it
		1649	- when the program encounters an overload, it will just loop until the
		1650	situation is over. While this is a form of busy waiting, no OS offers an
		1651	event-based way to handle this situation, so it's the best one can do.
		1652
		1653	A better way to handle the situation is to log any errors other than
		1654	C<EAGAIN> and C<EWOULDBLOCK>, making sure not to flood the log with such
		1655	messages, and continue as usual, which at least gives the user an idea of
		1656	what could be wrong ("raise the ulimit!"). For extra points one could stop
		1657	the C<ev_io> watcher on the listening fd "for a while", which reduces CPU
		1658	usage.
		1659
		1660	If your program is single-threaded, then you could also keep a dummy file
		1661	descriptor for overload situations (e.g. by opening F</dev/null>), and
		1662	when you run into C<ENFILE> or C<EMFILE>, close it, run C<accept>,
		1663	close that fd, and create a new dummy fd. This will gracefully refuse
		1664	clients under typical overload conditions.
		1665
		1666	The last way to handle it is to simply log the error and C<exit>, as
		1667	is often done with C<malloc> failures, but this results in an easy
		1668	opportunity for a DoS attack.
		1669
		1670	=head3 Watcher-Specific Functions
502		1671
503	=over 4	1672	=over 4
504		1673
505	=item ev_io_init (ev_io *, callback, int fd, int events)	1674	=item ev_io_init (ev_io *, callback, int fd, int events)
506		1675
507	=item ev_io_set (ev_io *, int fd, int events)	1676	=item ev_io_set (ev_io *, int fd, int events)
508		1677
509	Configures an C<ev_io> watcher. The fd is the file descriptor to rceeive	1678	Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
510	events for and events is either C<EV_READ>, C<EV_WRITE> or C<EV_READ |	1679	receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
511	EV_WRITE> to receive the given events.	1680	C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
		1681
		1682	=item int fd [read-only]
		1683
		1684	The file descriptor being watched.
		1685
		1686	=item int events [read-only]
		1687
		1688	The events being watched.
512		1689
513	=back	1690	=back
514		1691
		1692	=head3 Examples
		1693
		1694	Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
		1695	readable, but only once. Since it is likely line-buffered, you could
		1696	attempt to read a whole line in the callback.
		1697
		1698	static void
		1699	stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
		1700	{
		1701	ev_io_stop (loop, w);
		1702	.. read from stdin here (or from w->fd) and handle any I/O errors
		1703	}
		1704
		1705	...
		1706	struct ev_loop *loop = ev_default_init (0);
		1707	ev_io stdin_readable;
		1708	ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
		1709	ev_io_start (loop, &stdin_readable);
		1710	ev_run (loop, 0);
		1711
		1712
515	=head2 C<ev_timer> - relative and optionally recurring timeouts	1713	=head2 C<ev_timer> - relative and optionally repeating timeouts
516		1714
517	Timer watchers are simple relative timers that generate an event after a	1715	Timer watchers are simple relative timers that generate an event after a
518	given time, and optionally repeating in regular intervals after that.	1716	given time, and optionally repeating in regular intervals after that.
519		1717
520	The timers are based on real time, that is, if you register an event that	1718	The timers are based on real time, that is, if you register an event that
521	times out after an hour and you reset your system clock to last years	1719	times out after an hour and you reset your system clock to January last
522	time, it will still time out after (roughly) and hour. "Roughly" because	1720	year, it will still time out after (roughly) one hour. "Roughly" because
523	detecting time jumps is hard, and some inaccuracies are unavoidable (the	1721	detecting time jumps is hard, and some inaccuracies are unavoidable (the
524	monotonic clock option helps a lot here).	1722	monotonic clock option helps a lot here).
		1723
		1724	The callback is guaranteed to be invoked only I<after> its timeout has
		1725	passed (not I<at>, so on systems with very low-resolution clocks this
		1726	might introduce a small delay). If multiple timers become ready during the
		1727	same loop iteration then the ones with earlier time-out values are invoked
		1728	before ones of the same priority with later time-out values (but this is
		1729	no longer true when a callback calls C<ev_run> recursively).
		1730
		1731	=head3 Be smart about timeouts
		1732
		1733	Many real-world problems involve some kind of timeout, usually for error
		1734	recovery. A typical example is an HTTP request - if the other side hangs,
		1735	you want to raise some error after a while.
		1736
		1737	What follows are some ways to handle this problem, from obvious and
		1738	inefficient to smart and efficient.
		1739
		1740	In the following, a 60 second activity timeout is assumed - a timeout that
		1741	gets reset to 60 seconds each time there is activity (e.g. each time some
		1742	data or other life sign was received).
		1743
		1744	=over 4
		1745
		1746	=item 1. Use a timer and stop, reinitialise and start it on activity.
		1747
		1748	This is the most obvious, but not the most simple way: In the beginning,
		1749	start the watcher:
		1750
		1751	ev_timer_init (timer, callback, 60., 0.);
		1752	ev_timer_start (loop, timer);
		1753
		1754	Then, each time there is some activity, C<ev_timer_stop> it, initialise it
		1755	and start it again:
		1756
		1757	ev_timer_stop (loop, timer);
		1758	ev_timer_set (timer, 60., 0.);
		1759	ev_timer_start (loop, timer);
		1760
		1761	This is relatively simple to implement, but means that each time there is
		1762	some activity, libev will first have to remove the timer from its internal
		1763	data structure and then add it again. Libev tries to be fast, but it's
		1764	still not a constant-time operation.
		1765
		1766	=item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
		1767
		1768	This is the easiest way, and involves using C<ev_timer_again> instead of
		1769	C<ev_timer_start>.
		1770
		1771	To implement this, configure an C<ev_timer> with a C<repeat> value
		1772	of C<60> and then call C<ev_timer_again> at start and each time you
		1773	successfully read or write some data. If you go into an idle state where
		1774	you do not expect data to travel on the socket, you can C<ev_timer_stop>
		1775	the timer, and C<ev_timer_again> will automatically restart it if need be.
		1776
		1777	That means you can ignore both the C<ev_timer_start> function and the
		1778	C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
		1779	member and C<ev_timer_again>.
		1780
		1781	At start:
		1782
		1783	ev_init (timer, callback);
		1784	timer->repeat = 60.;
		1785	ev_timer_again (loop, timer);
		1786
		1787	Each time there is some activity:
		1788
		1789	ev_timer_again (loop, timer);
		1790
		1791	It is even possible to change the time-out on the fly, regardless of
		1792	whether the watcher is active or not:
		1793
		1794	timer->repeat = 30.;
		1795	ev_timer_again (loop, timer);
		1796
		1797	This is slightly more efficient then stopping/starting the timer each time
		1798	you want to modify its timeout value, as libev does not have to completely
		1799	remove and re-insert the timer from/into its internal data structure.
		1800
		1801	It is, however, even simpler than the "obvious" way to do it.
		1802
		1803	=item 3. Let the timer time out, but then re-arm it as required.
		1804
		1805	This method is more tricky, but usually most efficient: Most timeouts are
		1806	relatively long compared to the intervals between other activity - in
		1807	our example, within 60 seconds, there are usually many I/O events with
		1808	associated activity resets.
		1809
		1810	In this case, it would be more efficient to leave the C<ev_timer> alone,
		1811	but remember the time of last activity, and check for a real timeout only
		1812	within the callback:
		1813
		1814	ev_tstamp last_activity; // time of last activity
		1815
		1816	static void
		1817	callback (EV_P_ ev_timer *w, int revents)
		1818	{
		1819	ev_tstamp now = ev_now (EV_A);
		1820	ev_tstamp timeout = last_activity + 60.;
		1821
		1822	// if last_activity + 60. is older than now, we did time out
		1823	if (timeout < now)
		1824	{
		1825	// timeout occurred, take action
		1826	}
		1827	else
		1828	{
		1829	// callback was invoked, but there was some activity, re-arm
		1830	// the watcher to fire in last_activity + 60, which is
		1831	// guaranteed to be in the future, so "again" is positive:
		1832	w->repeat = timeout - now;
		1833	ev_timer_again (EV_A_ w);
		1834	}
		1835	}
		1836
		1837	To summarise the callback: first calculate the real timeout (defined
		1838	as "60 seconds after the last activity"), then check if that time has
		1839	been reached, which means something I<did>, in fact, time out. Otherwise
		1840	the callback was invoked too early (C<timeout> is in the future), so
		1841	re-schedule the timer to fire at that future time, to see if maybe we have
		1842	a timeout then.
		1843
		1844	Note how C<ev_timer_again> is used, taking advantage of the
		1845	C<ev_timer_again> optimisation when the timer is already running.
		1846
		1847	This scheme causes more callback invocations (about one every 60 seconds
		1848	minus half the average time between activity), but virtually no calls to
		1849	libev to change the timeout.
		1850
		1851	To start the timer, simply initialise the watcher and set C<last_activity>
		1852	to the current time (meaning we just have some activity :), then call the
		1853	callback, which will "do the right thing" and start the timer:
		1854
		1855	ev_init (timer, callback);
		1856	last_activity = ev_now (loop);
		1857	callback (loop, timer, EV_TIMER);
		1858
		1859	And when there is some activity, simply store the current time in
		1860	C<last_activity>, no libev calls at all:
		1861
		1862	last_activity = ev_now (loop);
		1863
		1864	This technique is slightly more complex, but in most cases where the
		1865	time-out is unlikely to be triggered, much more efficient.
		1866
		1867	Changing the timeout is trivial as well (if it isn't hard-coded in the
		1868	callback :) - just change the timeout and invoke the callback, which will
		1869	fix things for you.
		1870
		1871	=item 4. Wee, just use a double-linked list for your timeouts.
		1872
		1873	If there is not one request, but many thousands (millions...), all
		1874	employing some kind of timeout with the same timeout value, then one can
		1875	do even better:
		1876
		1877	When starting the timeout, calculate the timeout value and put the timeout
		1878	at the I<end> of the list.
		1879
		1880	Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
		1881	the list is expected to fire (for example, using the technique #3).
		1882
		1883	When there is some activity, remove the timer from the list, recalculate
		1884	the timeout, append it to the end of the list again, and make sure to
		1885	update the C<ev_timer> if it was taken from the beginning of the list.
		1886
		1887	This way, one can manage an unlimited number of timeouts in O(1) time for
		1888	starting, stopping and updating the timers, at the expense of a major
		1889	complication, and having to use a constant timeout. The constant timeout
		1890	ensures that the list stays sorted.
		1891
		1892	=back
		1893
		1894	So which method the best?
		1895
		1896	Method #2 is a simple no-brain-required solution that is adequate in most
		1897	situations. Method #3 requires a bit more thinking, but handles many cases
		1898	better, and isn't very complicated either. In most case, choosing either
		1899	one is fine, with #3 being better in typical situations.
		1900
		1901	Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
		1902	rather complicated, but extremely efficient, something that really pays
		1903	off after the first million or so of active timers, i.e. it's usually
		1904	overkill :)
		1905
		1906	=head3 The special problem of time updates
		1907
		1908	Establishing the current time is a costly operation (it usually takes at
		1909	least two system calls): EV therefore updates its idea of the current
		1910	time only before and after C<ev_run> collects new events, which causes a
		1911	growing difference between C<ev_now ()> and C<ev_time ()> when handling
		1912	lots of events in one iteration.
525		1913
526	The relative timeouts are calculated relative to the C<ev_now ()>	1914	The relative timeouts are calculated relative to the C<ev_now ()>
527	time. This is usually the right thing as this timestamp refers to the time	1915	time. This is usually the right thing as this timestamp refers to the time
528	of the event triggering whatever timeout you are modifying/starting. If	1916	of the event triggering whatever timeout you are modifying/starting. If
529	you suspect event processing to be delayed and you I<need> to base the timeout	1917	you suspect event processing to be delayed and you I<need> to base the
530	on the current time, use something like this to adjust for this:	1918	timeout on the current time, use something like this to adjust for this:
531		1919
532	ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);	1920	ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);
533		1921
534	The callback is guarenteed to be invoked only when its timeout has passed,	1922	If the event loop is suspended for a long time, you can also force an
535	but if multiple timers become ready during the same loop iteration then	1923	update of the time returned by C<ev_now ()> by calling C<ev_now_update
536	order of execution is undefined.	1924	()>.
		1925
		1926	=head3 The special problems of suspended animation
		1927
		1928	When you leave the server world it is quite customary to hit machines that
		1929	can suspend/hibernate - what happens to the clocks during such a suspend?
		1930
		1931	Some quick tests made with a Linux 2.6.28 indicate that a suspend freezes
		1932	all processes, while the clocks (C<times>, C<CLOCK_MONOTONIC>) continue
		1933	to run until the system is suspended, but they will not advance while the
		1934	system is suspended. That means, on resume, it will be as if the program
		1935	was frozen for a few seconds, but the suspend time will not be counted
		1936	towards C<ev_timer> when a monotonic clock source is used. The real time
		1937	clock advanced as expected, but if it is used as sole clocksource, then a
		1938	long suspend would be detected as a time jump by libev, and timers would
		1939	be adjusted accordingly.
		1940
		1941	I would not be surprised to see different behaviour in different between
		1942	operating systems, OS versions or even different hardware.
		1943
		1944	The other form of suspend (job control, or sending a SIGSTOP) will see a
		1945	time jump in the monotonic clocks and the realtime clock. If the program
		1946	is suspended for a very long time, and monotonic clock sources are in use,
		1947	then you can expect C<ev_timer>s to expire as the full suspension time
		1948	will be counted towards the timers. When no monotonic clock source is in
		1949	use, then libev will again assume a timejump and adjust accordingly.
		1950
		1951	It might be beneficial for this latter case to call C<ev_suspend>
		1952	and C<ev_resume> in code that handles C<SIGTSTP>, to at least get
		1953	deterministic behaviour in this case (you can do nothing against
		1954	C<SIGSTOP>).
		1955
		1956	=head3 Watcher-Specific Functions and Data Members
537		1957
538	=over 4	1958	=over 4
539		1959
540	=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)	1960	=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
541		1961
542	=item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)	1962	=item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
543		1963
544	Configure the timer to trigger after C<after> seconds. If C<repeat> is	1964	Configure the timer to trigger after C<after> seconds. If C<repeat>
545	C<0.>, then it will automatically be stopped. If it is positive, then the	1965	is C<0.>, then it will automatically be stopped once the timeout is
546	timer will automatically be configured to trigger again C<repeat> seconds	1966	reached. If it is positive, then the timer will automatically be
547	later, again, and again, until stopped manually.	1967	configured to trigger again C<repeat> seconds later, again, and again,
		1968	until stopped manually.
548		1969
549	The timer itself will do a best-effort at avoiding drift, that is, if you	1970	The timer itself will do a best-effort at avoiding drift, that is, if
550	configure a timer to trigger every 10 seconds, then it will trigger at	1971	you configure a timer to trigger every 10 seconds, then it will normally
551	exactly 10 second intervals. If, however, your program cannot keep up with	1972	trigger at exactly 10 second intervals. If, however, your program cannot
552	the timer (because it takes longer than those 10 seconds to do stuff) the	1973	keep up with the timer (because it takes longer than those 10 seconds to
553	timer will not fire more than once per event loop iteration.	1974	do stuff) the timer will not fire more than once per event loop iteration.
554		1975
555	=item ev_timer_again (loop)	1976	=item ev_timer_again (loop, ev_timer *)
556		1977
557	This will act as if the timer timed out and restart it again if it is	1978	This will act as if the timer timed out and restart it again if it is
558	repeating. The exact semantics are:	1979	repeating. The exact semantics are:
559		1980
		1981	If the timer is pending, its pending status is cleared.
		1982
560	If the timer is started but nonrepeating, stop it.	1983	If the timer is started but non-repeating, stop it (as if it timed out).
561		1984
562	If the timer is repeating, either start it if necessary (with the repeat	1985	If the timer is repeating, either start it if necessary (with the
563	value), or reset the running timer to the repeat value.	1986	C<repeat> value), or reset the running timer to the C<repeat> value.
564		1987
565	This sounds a bit complicated, but here is a useful and typical	1988	This sounds a bit complicated, see L<Be smart about timeouts>, above, for a
566	example: Imagine you have a tcp connection and you want a so-called idle	1989	usage example.
567	timeout, that is, you want to be called when there have been, say, 60	1990
568	seconds of inactivity on the socket. The easiest way to do this is to	1991	=item ev_tstamp ev_timer_remaining (loop, ev_timer *)
569	configure an C<ev_timer> with after=repeat=60 and calling ev_timer_again each	1992
570	time you successfully read or write some data. If you go into an idle	1993	Returns the remaining time until a timer fires. If the timer is active,
571	state where you do not expect data to travel on the socket, you can stop	1994	then this time is relative to the current event loop time, otherwise it's
572	the timer, and again will automatically restart it if need be.	1995	the timeout value currently configured.
		1996
		1997	That is, after an C<ev_timer_set (w, 5, 7)>, C<ev_timer_remaining> returns
		1998	C<5>. When the timer is started and one second passes, C<ev_timer_remaining>
		1999	will return C<4>. When the timer expires and is restarted, it will return
		2000	roughly C<7> (likely slightly less as callback invocation takes some time,
		2001	too), and so on.
		2002
		2003	=item ev_tstamp repeat [read-write]
		2004
		2005	The current C<repeat> value. Will be used each time the watcher times out
		2006	or C<ev_timer_again> is called, and determines the next timeout (if any),
		2007	which is also when any modifications are taken into account.
573		2008
574	=back	2009	=back
575		2010
		2011	=head3 Examples
		2012
		2013	Example: Create a timer that fires after 60 seconds.
		2014
		2015	static void
		2016	one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
		2017	{
		2018	.. one minute over, w is actually stopped right here
		2019	}
		2020
		2021	ev_timer mytimer;
		2022	ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
		2023	ev_timer_start (loop, &mytimer);
		2024
		2025	Example: Create a timeout timer that times out after 10 seconds of
		2026	inactivity.
		2027
		2028	static void
		2029	timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
		2030	{
		2031	.. ten seconds without any activity
		2032	}
		2033
		2034	ev_timer mytimer;
		2035	ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
		2036	ev_timer_again (&mytimer); /* start timer */
		2037	ev_run (loop, 0);
		2038
		2039	// and in some piece of code that gets executed on any "activity":
		2040	// reset the timeout to start ticking again at 10 seconds
		2041	ev_timer_again (&mytimer);
		2042
		2043
576	=head2 C<ev_periodic> - to cron or not to cron	2044	=head2 C<ev_periodic> - to cron or not to cron?
577		2045
578	Periodic watchers are also timers of a kind, but they are very versatile	2046	Periodic watchers are also timers of a kind, but they are very versatile
579	(and unfortunately a bit complex).	2047	(and unfortunately a bit complex).
580		2048
581	Unlike C<ev_timer>'s, they are not based on real time (or relative time)	2049	Unlike C<ev_timer>, periodic watchers are not based on real time (or
582	but on wallclock time (absolute time). You can tell a periodic watcher	2050	relative time, the physical time that passes) but on wall clock time
583	to trigger "at" some specific point in time. For example, if you tell a	2051	(absolute time, the thing you can read on your calender or clock). The
584	periodic watcher to trigger in 10 seconds (by specifiying e.g. c<ev_now ()	2052	difference is that wall clock time can run faster or slower than real
585	+ 10.>) and then reset your system clock to the last year, then it will	2053	time, and time jumps are not uncommon (e.g. when you adjust your
586	take a year to trigger the event (unlike an C<ev_timer>, which would trigger	2054	wrist-watch).
587	roughly 10 seconds later and of course not if you reset your system time
588	again).
589		2055
590	They can also be used to implement vastly more complex timers, such as	2056	You can tell a periodic watcher to trigger after some specific point
		2057	in time: for example, if you tell a periodic watcher to trigger "in 10
		2058	seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
		2059	not a delay) and then reset your system clock to January of the previous
		2060	year, then it will take a year or more to trigger the event (unlike an
		2061	C<ev_timer>, which would still trigger roughly 10 seconds after starting
		2062	it, as it uses a relative timeout).
		2063
		2064	C<ev_periodic> watchers can also be used to implement vastly more complex
591	triggering an event on eahc midnight, local time.	2065	timers, such as triggering an event on each "midnight, local time", or
		2066	other complicated rules. This cannot be done with C<ev_timer> watchers, as
		2067	those cannot react to time jumps.
592		2068
593	As with timers, the callback is guarenteed to be invoked only when the	2069	As with timers, the callback is guaranteed to be invoked only when the
594	time (C<at>) has been passed, but if multiple periodic timers become ready	2070	point in time where it is supposed to trigger has passed. If multiple
595	during the same loop iteration then order of execution is undefined.	2071	timers become ready during the same loop iteration then the ones with
		2072	earlier time-out values are invoked before ones with later time-out values
		2073	(but this is no longer true when a callback calls C<ev_run> recursively).
		2074
		2075	=head3 Watcher-Specific Functions and Data Members
596		2076
597	=over 4	2077	=over 4
598		2078
599	=item ev_periodic_init (ev_periodic *, callback, ev_tstamp at, ev_tstamp interval, reschedule_cb)	2079	=item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
600		2080
601	=item ev_periodic_set (ev_periodic *, ev_tstamp after, ev_tstamp repeat, reschedule_cb)	2081	=item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
602		2082
603	Lots of arguments, lets sort it out... There are basically three modes of	2083	Lots of arguments, let's sort it out... There are basically three modes of
604	operation, and we will explain them from simplest to complex:	2084	operation, and we will explain them from simplest to most complex:
605		2085
606	=over 4	2086	=over 4
607		2087
608	=item * absolute timer (interval = reschedule_cb = 0)	2088	=item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
609		2089
610	In this configuration the watcher triggers an event at the wallclock time	2090	In this configuration the watcher triggers an event after the wall clock
611	C<at> and doesn't repeat. It will not adjust when a time jump occurs,	2091	time C<offset> has passed. It will not repeat and will not adjust when a
612	that is, if it is to be run at January 1st 2011 then it will run when the	2092	time jump occurs, that is, if it is to be run at January 1st 2011 then it
613	system time reaches or surpasses this time.	2093	will be stopped and invoked when the system clock reaches or surpasses
		2094	this point in time.
614		2095
615	=item * non-repeating interval timer (interval > 0, reschedule_cb = 0)	2096	=item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
616		2097
617	In this mode the watcher will always be scheduled to time out at the next	2098	In this mode the watcher will always be scheduled to time out at the next
618	C<at + N * interval> time (for some integer N) and then repeat, regardless	2099	C<offset + N * interval> time (for some integer N, which can also be
619	of any time jumps.	2100	negative) and then repeat, regardless of any time jumps. The C<offset>
		2101	argument is merely an offset into the C<interval> periods.
620		2102
621	This can be used to create timers that do not drift with respect to system	2103	This can be used to create timers that do not drift with respect to the
622	time:	2104	system clock, for example, here is an C<ev_periodic> that triggers each
		2105	hour, on the hour (with respect to UTC):
623		2106
624	ev_periodic_set (&periodic, 0., 3600., 0);	2107	ev_periodic_set (&periodic, 0., 3600., 0);
625		2108
626	This doesn't mean there will always be 3600 seconds in between triggers,	2109	This doesn't mean there will always be 3600 seconds in between triggers,
627	but only that the the callback will be called when the system time shows a	2110	but only that the callback will be called when the system time shows a
628	full hour (UTC), or more correctly, when the system time is evenly divisible	2111	full hour (UTC), or more correctly, when the system time is evenly divisible
629	by 3600.	2112	by 3600.
630		2113
631	Another way to think about it (for the mathematically inclined) is that	2114	Another way to think about it (for the mathematically inclined) is that
632	C<ev_periodic> will try to run the callback in this mode at the next possible	2115	C<ev_periodic> will try to run the callback in this mode at the next possible
633	time where C<time = at (mod interval)>, regardless of any time jumps.	2116	time where C<time = offset (mod interval)>, regardless of any time jumps.
634		2117
		2118	For numerical stability it is preferable that the C<offset> value is near
		2119	C<ev_now ()> (the current time), but there is no range requirement for
		2120	this value, and in fact is often specified as zero.
		2121
		2122	Note also that there is an upper limit to how often a timer can fire (CPU
		2123	speed for example), so if C<interval> is very small then timing stability
		2124	will of course deteriorate. Libev itself tries to be exact to be about one
		2125	millisecond (if the OS supports it and the machine is fast enough).
		2126
635	=item * manual reschedule mode (reschedule_cb = callback)	2127	=item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
636		2128
637	In this mode the values for C<interval> and C<at> are both being	2129	In this mode the values for C<interval> and C<offset> are both being
638	ignored. Instead, each time the periodic watcher gets scheduled, the	2130	ignored. Instead, each time the periodic watcher gets scheduled, the
639	reschedule callback will be called with the watcher as first, and the	2131	reschedule callback will be called with the watcher as first, and the
640	current time as second argument.	2132	current time as second argument.
641		2133
642	NOTE: I<This callback MUST NOT stop or destroy any periodic watcher,	2134	NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, ever,
643	ever, or make any event loop modifications>. If you need to stop it,	2135	or make ANY other event loop modifications whatsoever, unless explicitly
644	return C<now + 1e30> (or so, fudge fudge) and stop it afterwards (e.g. by	2136	allowed by documentation here>.
645	starting a prepare watcher).
646		2137
		2138	If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
		2139	it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
		2140	only event loop modification you are allowed to do).
		2141
647	Its prototype is C<ev_tstamp (*reschedule_cb)(struct ev_periodic *w,	2142	The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
648	ev_tstamp now)>, e.g.:	2143	*w, ev_tstamp now)>, e.g.:
649		2144
		2145	static ev_tstamp
650	static ev_tstamp my_rescheduler (struct ev_periodic *w, ev_tstamp now)	2146	my_rescheduler (ev_periodic *w, ev_tstamp now)
651	{	2147	{
652	return now + 60.;	2148	return now + 60.;
653	}	2149	}
654		2150
655	It must return the next time to trigger, based on the passed time value	2151	It must return the next time to trigger, based on the passed time value
656	(that is, the lowest time value larger than to the second argument). It	2152	(that is, the lowest time value larger than to the second argument). It
657	will usually be called just before the callback will be triggered, but	2153	will usually be called just before the callback will be triggered, but
658	might be called at other times, too.	2154	might be called at other times, too.
659		2155
660	NOTE: I<< This callback must always return a time that is later than the	2156	NOTE: I<< This callback must always return a time that is higher than or
661	passed C<now> value >>. Not even C<now> itself will do, it I<must> be larger.	2157	equal to the passed C<now> value >>.
662		2158
663	This can be used to create very complex timers, such as a timer that	2159	This can be used to create very complex timers, such as a timer that
664	triggers on each midnight, local time. To do this, you would calculate the	2160	triggers on "next midnight, local time". To do this, you would calculate the
665	next midnight after C<now> and return the timestamp value for this. How	2161	next midnight after C<now> and return the timestamp value for this. How
666	you do this is, again, up to you (but it is not trivial, which is the main	2162	you do this is, again, up to you (but it is not trivial, which is the main
667	reason I omitted it as an example).	2163	reason I omitted it as an example).
668		2164
669	=back	2165	=back
…		…
673	Simply stops and restarts the periodic watcher again. This is only useful	2169	Simply stops and restarts the periodic watcher again. This is only useful
674	when you changed some parameters or the reschedule callback would return	2170	when you changed some parameters or the reschedule callback would return
675	a different time than the last time it was called (e.g. in a crond like	2171	a different time than the last time it was called (e.g. in a crond like
676	program when the crontabs have changed).	2172	program when the crontabs have changed).
677		2173
		2174	=item ev_tstamp ev_periodic_at (ev_periodic *)
		2175
		2176	When active, returns the absolute time that the watcher is supposed
		2177	to trigger next. This is not the same as the C<offset> argument to
		2178	C<ev_periodic_set>, but indeed works even in interval and manual
		2179	rescheduling modes.
		2180
		2181	=item ev_tstamp offset [read-write]
		2182
		2183	When repeating, this contains the offset value, otherwise this is the
		2184	absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
		2185	although libev might modify this value for better numerical stability).
		2186
		2187	Can be modified any time, but changes only take effect when the periodic
		2188	timer fires or C<ev_periodic_again> is being called.
		2189
		2190	=item ev_tstamp interval [read-write]
		2191
		2192	The current interval value. Can be modified any time, but changes only
		2193	take effect when the periodic timer fires or C<ev_periodic_again> is being
		2194	called.
		2195
		2196	=item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
		2197
		2198	The current reschedule callback, or C<0>, if this functionality is
		2199	switched off. Can be changed any time, but changes only take effect when
		2200	the periodic timer fires or C<ev_periodic_again> is being called.
		2201
678	=back	2202	=back
679		2203
		2204	=head3 Examples
		2205
		2206	Example: Call a callback every hour, or, more precisely, whenever the
		2207	system time is divisible by 3600. The callback invocation times have
		2208	potentially a lot of jitter, but good long-term stability.
		2209
		2210	static void
		2211	clock_cb (struct ev_loop *loop, ev_periodic *w, int revents)
		2212	{
		2213	... its now a full hour (UTC, or TAI or whatever your clock follows)
		2214	}
		2215
		2216	ev_periodic hourly_tick;
		2217	ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
		2218	ev_periodic_start (loop, &hourly_tick);
		2219
		2220	Example: The same as above, but use a reschedule callback to do it:
		2221
		2222	#include <math.h>
		2223
		2224	static ev_tstamp
		2225	my_scheduler_cb (ev_periodic *w, ev_tstamp now)
		2226	{
		2227	return now + (3600. - fmod (now, 3600.));
		2228	}
		2229
		2230	ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
		2231
		2232	Example: Call a callback every hour, starting now:
		2233
		2234	ev_periodic hourly_tick;
		2235	ev_periodic_init (&hourly_tick, clock_cb,
		2236	fmod (ev_now (loop), 3600.), 3600., 0);
		2237	ev_periodic_start (loop, &hourly_tick);
		2238
		2239
680	=head2 C<ev_signal> - signal me when a signal gets signalled	2240	=head2 C<ev_signal> - signal me when a signal gets signalled!
681		2241
682	Signal watchers will trigger an event when the process receives a specific	2242	Signal watchers will trigger an event when the process receives a specific
683	signal one or more times. Even though signals are very asynchronous, libev	2243	signal one or more times. Even though signals are very asynchronous, libev
684	will try it's best to deliver signals synchronously, i.e. as part of the	2244	will try it's best to deliver signals synchronously, i.e. as part of the
685	normal event processing, like any other event.	2245	normal event processing, like any other event.
686		2246
		2247	If you want signals to be delivered truly asynchronously, just use
		2248	C<sigaction> as you would do without libev and forget about sharing
		2249	the signal. You can even use C<ev_async> from a signal handler to
		2250	synchronously wake up an event loop.
		2251
687	You can configure as many watchers as you like per signal. Only when the	2252	You can configure as many watchers as you like for the same signal, but
		2253	only within the same loop, i.e. you can watch for C<SIGINT> in your
		2254	default loop and for C<SIGIO> in another loop, but you cannot watch for
		2255	C<SIGINT> in both the default loop and another loop at the same time. At
		2256	the moment, C<SIGCHLD> is permanently tied to the default loop.
		2257
688	first watcher gets started will libev actually register a signal watcher	2258	When the first watcher gets started will libev actually register something
689	with the kernel (thus it coexists with your own signal handlers as long	2259	with the kernel (thus it coexists with your own signal handlers as long as
690	as you don't register any with libev). Similarly, when the last signal	2260	you don't register any with libev for the same signal).
691	watcher for a signal is stopped libev will reset the signal handler to	2261
692	SIG_DFL (regardless of what it was set to before).	2262	If possible and supported, libev will install its handlers with
		2263	C<SA_RESTART> (or equivalent) behaviour enabled, so system calls should
		2264	not be unduly interrupted. If you have a problem with system calls getting
		2265	interrupted by signals you can block all signals in an C<ev_check> watcher
		2266	and unblock them in an C<ev_prepare> watcher.
		2267
		2268	=head3 The special problem of inheritance over fork/execve/pthread_create
		2269
		2270	Both the signal mask (C<sigprocmask>) and the signal disposition
		2271	(C<sigaction>) are unspecified after starting a signal watcher (and after
		2272	stopping it again), that is, libev might or might not block the signal,
		2273	and might or might not set or restore the installed signal handler.
		2274
		2275	While this does not matter for the signal disposition (libev never
		2276	sets signals to C<SIG_IGN>, so handlers will be reset to C<SIG_DFL> on
		2277	C<execve>), this matters for the signal mask: many programs do not expect
		2278	certain signals to be blocked.
		2279
		2280	This means that before calling C<exec> (from the child) you should reset
		2281	the signal mask to whatever "default" you expect (all clear is a good
		2282	choice usually).
		2283
		2284	The simplest way to ensure that the signal mask is reset in the child is
		2285	to install a fork handler with C<pthread_atfork> that resets it. That will
		2286	catch fork calls done by libraries (such as the libc) as well.
		2287
		2288	In current versions of libev, the signal will not be blocked indefinitely
		2289	unless you use the C<signalfd> API (C<EV_SIGNALFD>). While this reduces
		2290	the window of opportunity for problems, it will not go away, as libev
		2291	I<has> to modify the signal mask, at least temporarily.
		2292
		2293	So I can't stress this enough: I<If you do not reset your signal mask when
		2294	you expect it to be empty, you have a race condition in your code>. This
		2295	is not a libev-specific thing, this is true for most event libraries.
		2296
		2297	=head3 Watcher-Specific Functions and Data Members
693		2298
694	=over 4	2299	=over 4
695		2300
696	=item ev_signal_init (ev_signal *, callback, int signum)	2301	=item ev_signal_init (ev_signal *, callback, int signum)
697		2302
698	=item ev_signal_set (ev_signal *, int signum)	2303	=item ev_signal_set (ev_signal *, int signum)
699		2304
700	Configures the watcher to trigger on the given signal number (usually one	2305	Configures the watcher to trigger on the given signal number (usually one
701	of the C<SIGxxx> constants).	2306	of the C<SIGxxx> constants).
702		2307
		2308	=item int signum [read-only]
		2309
		2310	The signal the watcher watches out for.
		2311
703	=back	2312	=back
704		2313
		2314	=head3 Examples
		2315
		2316	Example: Try to exit cleanly on SIGINT.
		2317
		2318	static void
		2319	sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
		2320	{
		2321	ev_break (loop, EVBREAK_ALL);
		2322	}
		2323
		2324	ev_signal signal_watcher;
		2325	ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
		2326	ev_signal_start (loop, &signal_watcher);
		2327
		2328
705	=head2 C<ev_child> - wait for pid status changes	2329	=head2 C<ev_child> - watch out for process status changes
706		2330
707	Child watchers trigger when your process receives a SIGCHLD in response to	2331	Child watchers trigger when your process receives a SIGCHLD in response to
708	some child status changes (most typically when a child of yours dies).	2332	some child status changes (most typically when a child of yours dies or
		2333	exits). It is permissible to install a child watcher I<after> the child
		2334	has been forked (which implies it might have already exited), as long
		2335	as the event loop isn't entered (or is continued from a watcher), i.e.,
		2336	forking and then immediately registering a watcher for the child is fine,
		2337	but forking and registering a watcher a few event loop iterations later or
		2338	in the next callback invocation is not.
		2339
		2340	Only the default event loop is capable of handling signals, and therefore
		2341	you can only register child watchers in the default event loop.
		2342
		2343	Due to some design glitches inside libev, child watchers will always be
		2344	handled at maximum priority (their priority is set to C<EV_MAXPRI> by
		2345	libev)
		2346
		2347	=head3 Process Interaction
		2348
		2349	Libev grabs C<SIGCHLD> as soon as the default event loop is
		2350	initialised. This is necessary to guarantee proper behaviour even if the
		2351	first child watcher is started after the child exits. The occurrence
		2352	of C<SIGCHLD> is recorded asynchronously, but child reaping is done
		2353	synchronously as part of the event loop processing. Libev always reaps all
		2354	children, even ones not watched.
		2355
		2356	=head3 Overriding the Built-In Processing
		2357
		2358	Libev offers no special support for overriding the built-in child
		2359	processing, but if your application collides with libev's default child
		2360	handler, you can override it easily by installing your own handler for
		2361	C<SIGCHLD> after initialising the default loop, and making sure the
		2362	default loop never gets destroyed. You are encouraged, however, to use an
		2363	event-based approach to child reaping and thus use libev's support for
		2364	that, so other libev users can use C<ev_child> watchers freely.
		2365
		2366	=head3 Stopping the Child Watcher
		2367
		2368	Currently, the child watcher never gets stopped, even when the
		2369	child terminates, so normally one needs to stop the watcher in the
		2370	callback. Future versions of libev might stop the watcher automatically
		2371	when a child exit is detected (calling C<ev_child_stop> twice is not a
		2372	problem).
		2373
		2374	=head3 Watcher-Specific Functions and Data Members
709		2375
710	=over 4	2376	=over 4
711		2377
712	=item ev_child_init (ev_child *, callback, int pid)	2378	=item ev_child_init (ev_child *, callback, int pid, int trace)
713		2379
714	=item ev_child_set (ev_child *, int pid)	2380	=item ev_child_set (ev_child *, int pid, int trace)
715		2381
716	Configures the watcher to wait for status changes of process C<pid> (or	2382	Configures the watcher to wait for status changes of process C<pid> (or
717	I<any> process if C<pid> is specified as C<0>). The callback can look	2383	I<any> process if C<pid> is specified as C<0>). The callback can look
718	at the C<rstatus> member of the C<ev_child> watcher structure to see	2384	at the C<rstatus> member of the C<ev_child> watcher structure to see
719	the status word (use the macros from C<sys/wait.h> and see your systems	2385	the status word (use the macros from C<sys/wait.h> and see your systems
720	C<waitpid> documentation). The C<rpid> member contains the pid of the	2386	C<waitpid> documentation). The C<rpid> member contains the pid of the
721	process causing the status change.	2387	process causing the status change. C<trace> must be either C<0> (only
		2388	activate the watcher when the process terminates) or C<1> (additionally
		2389	activate the watcher when the process is stopped or continued).
		2390
		2391	=item int pid [read-only]
		2392
		2393	The process id this watcher watches out for, or C<0>, meaning any process id.
		2394
		2395	=item int rpid [read-write]
		2396
		2397	The process id that detected a status change.
		2398
		2399	=item int rstatus [read-write]
		2400
		2401	The process exit/trace status caused by C<rpid> (see your systems
		2402	C<waitpid> and C<sys/wait.h> documentation for details).
722		2403
723	=back	2404	=back
724		2405
		2406	=head3 Examples
		2407
		2408	Example: C<fork()> a new process and install a child handler to wait for
		2409	its completion.
		2410
		2411	ev_child cw;
		2412
		2413	static void
		2414	child_cb (EV_P_ ev_child *w, int revents)
		2415	{
		2416	ev_child_stop (EV_A_ w);
		2417	printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
		2418	}
		2419
		2420	pid_t pid = fork ();
		2421
		2422	if (pid < 0)
		2423	// error
		2424	else if (pid == 0)
		2425	{
		2426	// the forked child executes here
		2427	exit (1);
		2428	}
		2429	else
		2430	{
		2431	ev_child_init (&cw, child_cb, pid, 0);
		2432	ev_child_start (EV_DEFAULT_ &cw);
		2433	}
		2434
		2435
		2436	=head2 C<ev_stat> - did the file attributes just change?
		2437
		2438	This watches a file system path for attribute changes. That is, it calls
		2439	C<stat> on that path in regular intervals (or when the OS says it changed)
		2440	and sees if it changed compared to the last time, invoking the callback if
		2441	it did.
		2442
		2443	The path does not need to exist: changing from "path exists" to "path does
		2444	not exist" is a status change like any other. The condition "path does not
		2445	exist" (or more correctly "path cannot be stat'ed") is signified by the
		2446	C<st_nlink> field being zero (which is otherwise always forced to be at
		2447	least one) and all the other fields of the stat buffer having unspecified
		2448	contents.
		2449
		2450	The path I<must not> end in a slash or contain special components such as
		2451	C<.> or C<..>. The path I<should> be absolute: If it is relative and
		2452	your working directory changes, then the behaviour is undefined.
		2453
		2454	Since there is no portable change notification interface available, the
		2455	portable implementation simply calls C<stat(2)> regularly on the path
		2456	to see if it changed somehow. You can specify a recommended polling
		2457	interval for this case. If you specify a polling interval of C<0> (highly
		2458	recommended!) then a I<suitable, unspecified default> value will be used
		2459	(which you can expect to be around five seconds, although this might
		2460	change dynamically). Libev will also impose a minimum interval which is
		2461	currently around C<0.1>, but that's usually overkill.
		2462
		2463	This watcher type is not meant for massive numbers of stat watchers,
		2464	as even with OS-supported change notifications, this can be
		2465	resource-intensive.
		2466
		2467	At the time of this writing, the only OS-specific interface implemented
		2468	is the Linux inotify interface (implementing kqueue support is left as an
		2469	exercise for the reader. Note, however, that the author sees no way of
		2470	implementing C<ev_stat> semantics with kqueue, except as a hint).
		2471
		2472	=head3 ABI Issues (Largefile Support)
		2473
		2474	Libev by default (unless the user overrides this) uses the default
		2475	compilation environment, which means that on systems with large file
		2476	support disabled by default, you get the 32 bit version of the stat
		2477	structure. When using the library from programs that change the ABI to
		2478	use 64 bit file offsets the programs will fail. In that case you have to
		2479	compile libev with the same flags to get binary compatibility. This is
		2480	obviously the case with any flags that change the ABI, but the problem is
		2481	most noticeably displayed with ev_stat and large file support.
		2482
		2483	The solution for this is to lobby your distribution maker to make large
		2484	file interfaces available by default (as e.g. FreeBSD does) and not
		2485	optional. Libev cannot simply switch on large file support because it has
		2486	to exchange stat structures with application programs compiled using the
		2487	default compilation environment.
		2488
		2489	=head3 Inotify and Kqueue
		2490
		2491	When C<inotify (7)> support has been compiled into libev and present at
		2492	runtime, it will be used to speed up change detection where possible. The
		2493	inotify descriptor will be created lazily when the first C<ev_stat>
		2494	watcher is being started.
		2495
		2496	Inotify presence does not change the semantics of C<ev_stat> watchers
		2497	except that changes might be detected earlier, and in some cases, to avoid
		2498	making regular C<stat> calls. Even in the presence of inotify support
		2499	there are many cases where libev has to resort to regular C<stat> polling,
		2500	but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
		2501	many bugs), the path exists (i.e. stat succeeds), and the path resides on
		2502	a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
		2503	xfs are fully working) libev usually gets away without polling.
		2504
		2505	There is no support for kqueue, as apparently it cannot be used to
		2506	implement this functionality, due to the requirement of having a file
		2507	descriptor open on the object at all times, and detecting renames, unlinks
		2508	etc. is difficult.
		2509
		2510	=head3 C<stat ()> is a synchronous operation
		2511
		2512	Libev doesn't normally do any kind of I/O itself, and so is not blocking
		2513	the process. The exception are C<ev_stat> watchers - those call C<stat
		2514	()>, which is a synchronous operation.
		2515
		2516	For local paths, this usually doesn't matter: unless the system is very
		2517	busy or the intervals between stat's are large, a stat call will be fast,
		2518	as the path data is usually in memory already (except when starting the
		2519	watcher).
		2520
		2521	For networked file systems, calling C<stat ()> can block an indefinite
		2522	time due to network issues, and even under good conditions, a stat call
		2523	often takes multiple milliseconds.
		2524
		2525	Therefore, it is best to avoid using C<ev_stat> watchers on networked
		2526	paths, although this is fully supported by libev.
		2527
		2528	=head3 The special problem of stat time resolution
		2529
		2530	The C<stat ()> system call only supports full-second resolution portably,
		2531	and even on systems where the resolution is higher, most file systems
		2532	still only support whole seconds.
		2533
		2534	That means that, if the time is the only thing that changes, you can
		2535	easily miss updates: on the first update, C<ev_stat> detects a change and
		2536	calls your callback, which does something. When there is another update
		2537	within the same second, C<ev_stat> will be unable to detect unless the
		2538	stat data does change in other ways (e.g. file size).
		2539
		2540	The solution to this is to delay acting on a change for slightly more
		2541	than a second (or till slightly after the next full second boundary), using
		2542	a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
		2543	ev_timer_again (loop, w)>).
		2544
		2545	The C<.02> offset is added to work around small timing inconsistencies
		2546	of some operating systems (where the second counter of the current time
		2547	might be be delayed. One such system is the Linux kernel, where a call to
		2548	C<gettimeofday> might return a timestamp with a full second later than
		2549	a subsequent C<time> call - if the equivalent of C<time ()> is used to
		2550	update file times then there will be a small window where the kernel uses
		2551	the previous second to update file times but libev might already execute
		2552	the timer callback).
		2553
		2554	=head3 Watcher-Specific Functions and Data Members
		2555
		2556	=over 4
		2557
		2558	=item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
		2559
		2560	=item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
		2561
		2562	Configures the watcher to wait for status changes of the given
		2563	C<path>. The C<interval> is a hint on how quickly a change is expected to
		2564	be detected and should normally be specified as C<0> to let libev choose
		2565	a suitable value. The memory pointed to by C<path> must point to the same
		2566	path for as long as the watcher is active.
		2567
		2568	The callback will receive an C<EV_STAT> event when a change was detected,
		2569	relative to the attributes at the time the watcher was started (or the
		2570	last change was detected).
		2571
		2572	=item ev_stat_stat (loop, ev_stat *)
		2573
		2574	Updates the stat buffer immediately with new values. If you change the
		2575	watched path in your callback, you could call this function to avoid
		2576	detecting this change (while introducing a race condition if you are not
		2577	the only one changing the path). Can also be useful simply to find out the
		2578	new values.
		2579
		2580	=item ev_statdata attr [read-only]
		2581
		2582	The most-recently detected attributes of the file. Although the type is
		2583	C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
		2584	suitable for your system, but you can only rely on the POSIX-standardised
		2585	members to be present. If the C<st_nlink> member is C<0>, then there was
		2586	some error while C<stat>ing the file.
		2587
		2588	=item ev_statdata prev [read-only]
		2589
		2590	The previous attributes of the file. The callback gets invoked whenever
		2591	C<prev> != C<attr>, or, more precisely, one or more of these members
		2592	differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
		2593	C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
		2594
		2595	=item ev_tstamp interval [read-only]
		2596
		2597	The specified interval.
		2598
		2599	=item const char *path [read-only]
		2600
		2601	The file system path that is being watched.
		2602
		2603	=back
		2604
		2605	=head3 Examples
		2606
		2607	Example: Watch C</etc/passwd> for attribute changes.
		2608
		2609	static void
		2610	passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
		2611	{
		2612	/* /etc/passwd changed in some way */
		2613	if (w->attr.st_nlink)
		2614	{
		2615	printf ("passwd current size %ld\n", (long)w->attr.st_size);
		2616	printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
		2617	printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
		2618	}
		2619	else
		2620	/* you shalt not abuse printf for puts */
		2621	puts ("wow, /etc/passwd is not there, expect problems. "
		2622	"if this is windows, they already arrived\n");
		2623	}
		2624
		2625	...
		2626	ev_stat passwd;
		2627
		2628	ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
		2629	ev_stat_start (loop, &passwd);
		2630
		2631	Example: Like above, but additionally use a one-second delay so we do not
		2632	miss updates (however, frequent updates will delay processing, too, so
		2633	one might do the work both on C<ev_stat> callback invocation I<and> on
		2634	C<ev_timer> callback invocation).
		2635
		2636	static ev_stat passwd;
		2637	static ev_timer timer;
		2638
		2639	static void
		2640	timer_cb (EV_P_ ev_timer *w, int revents)
		2641	{
		2642	ev_timer_stop (EV_A_ w);
		2643
		2644	/* now it's one second after the most recent passwd change */
		2645	}
		2646
		2647	static void
		2648	stat_cb (EV_P_ ev_stat *w, int revents)
		2649	{
		2650	/* reset the one-second timer */
		2651	ev_timer_again (EV_A_ &timer);
		2652	}
		2653
		2654	...
		2655	ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
		2656	ev_stat_start (loop, &passwd);
		2657	ev_timer_init (&timer, timer_cb, 0., 1.02);
		2658
		2659
725	=head2 C<ev_idle> - when you've got nothing better to do	2660	=head2 C<ev_idle> - when you've got nothing better to do...
726		2661
727	Idle watchers trigger events when there are no other events are pending	2662	Idle watchers trigger events when no other events of the same or higher
728	(prepare, check and other idle watchers do not count). That is, as long	2663	priority are pending (prepare, check and other idle watchers do not count
729	as your process is busy handling sockets or timeouts (or even signals,	2664	as receiving "events").
730	imagine) it will not be triggered. But when your process is idle all idle	2665
731	watchers are being called again and again, once per event loop iteration -	2666	That is, as long as your process is busy handling sockets or timeouts
		2667	(or even signals, imagine) of the same or higher priority it will not be
		2668	triggered. But when your process is idle (or only lower-priority watchers
		2669	are pending), the idle watchers are being called once per event loop
732	until stopped, that is, or your process receives more events and becomes	2670	iteration - until stopped, that is, or your process receives more events
733	busy.	2671	and becomes busy again with higher priority stuff.
734		2672
735	The most noteworthy effect is that as long as any idle watchers are	2673	The most noteworthy effect is that as long as any idle watchers are
736	active, the process will not block when waiting for new events.	2674	active, the process will not block when waiting for new events.
737		2675
738	Apart from keeping your process non-blocking (which is a useful	2676	Apart from keeping your process non-blocking (which is a useful
739	effect on its own sometimes), idle watchers are a good place to do	2677	effect on its own sometimes), idle watchers are a good place to do
740	"pseudo-background processing", or delay processing stuff to after the	2678	"pseudo-background processing", or delay processing stuff to after the
741	event loop has handled all outstanding events.	2679	event loop has handled all outstanding events.
742		2680
		2681	=head3 Watcher-Specific Functions and Data Members
		2682
743	=over 4	2683	=over 4
744		2684
745	=item ev_idle_init (ev_signal *, callback)	2685	=item ev_idle_init (ev_idle *, callback)
746		2686
747	Initialises and configures the idle watcher - it has no parameters of any	2687	Initialises and configures the idle watcher - it has no parameters of any
748	kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,	2688	kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
749	believe me.	2689	believe me.
750		2690
751	=back	2691	=back
752		2692
		2693	=head3 Examples
		2694
		2695	Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
		2696	callback, free it. Also, use no error checking, as usual.
		2697
		2698	static void
		2699	idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
		2700	{
		2701	free (w);
		2702	// now do something you wanted to do when the program has
		2703	// no longer anything immediate to do.
		2704	}
		2705
		2706	ev_idle *idle_watcher = malloc (sizeof (ev_idle));
		2707	ev_idle_init (idle_watcher, idle_cb);
		2708	ev_idle_start (loop, idle_watcher);
		2709
		2710
753	=head2 C<ev_prepare> and C<ev_check> - customise your event loop	2711	=head2 C<ev_prepare> and C<ev_check> - customise your event loop!
754		2712
755	Prepare and check watchers are usually (but not always) used in tandem:	2713	Prepare and check watchers are usually (but not always) used in pairs:
756	prepare watchers get invoked before the process blocks and check watchers	2714	prepare watchers get invoked before the process blocks and check watchers
757	afterwards.	2715	afterwards.
758		2716
		2717	You I<must not> call C<ev_run> or similar functions that enter
		2718	the current event loop from either C<ev_prepare> or C<ev_check>
		2719	watchers. Other loops than the current one are fine, however. The
		2720	rationale behind this is that you do not need to check for recursion in
		2721	those watchers, i.e. the sequence will always be C<ev_prepare>, blocking,
		2722	C<ev_check> so if you have one watcher of each kind they will always be
		2723	called in pairs bracketing the blocking call.
		2724
759	Their main purpose is to integrate other event mechanisms into libev. This	2725	Their main purpose is to integrate other event mechanisms into libev and
760	could be used, for example, to track variable changes, implement your own	2726	their use is somewhat advanced. They could be used, for example, to track
761	watchers, integrate net-snmp or a coroutine library and lots more.	2727	variable changes, implement your own watchers, integrate net-snmp or a
		2728	coroutine library and lots more. They are also occasionally useful if
		2729	you cache some data and want to flush it before blocking (for example,
		2730	in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
		2731	watcher).
762		2732
763	This is done by examining in each prepare call which file descriptors need	2733	This is done by examining in each prepare call which file descriptors
764	to be watched by the other library, registering C<ev_io> watchers for	2734	need to be watched by the other library, registering C<ev_io> watchers
765	them and starting an C<ev_timer> watcher for any timeouts (many libraries	2735	for them and starting an C<ev_timer> watcher for any timeouts (many
766	provide just this functionality). Then, in the check watcher you check for	2736	libraries provide exactly this functionality). Then, in the check watcher,
767	any events that occured (by checking the pending status of all watchers	2737	you check for any events that occurred (by checking the pending status
768	and stopping them) and call back into the library. The I/O and timer	2738	of all watchers and stopping them) and call back into the library. The
769	callbacks will never actually be called (but must be valid nevertheless,	2739	I/O and timer callbacks will never actually be called (but must be valid
770	because you never know, you know?).	2740	nevertheless, because you never know, you know?).
771		2741
772	As another example, the Perl Coro module uses these hooks to integrate	2742	As another example, the Perl Coro module uses these hooks to integrate
773	coroutines into libev programs, by yielding to other active coroutines	2743	coroutines into libev programs, by yielding to other active coroutines
774	during each prepare and only letting the process block if no coroutines	2744	during each prepare and only letting the process block if no coroutines
775	are ready to run (it's actually more complicated: it only runs coroutines	2745	are ready to run (it's actually more complicated: it only runs coroutines
776	with priority higher than or equal to the event loop and one coroutine	2746	with priority higher than or equal to the event loop and one coroutine
777	of lower priority, but only once, using idle watchers to keep the event	2747	of lower priority, but only once, using idle watchers to keep the event
778	loop from blocking if lower-priority coroutines are active, thus mapping	2748	loop from blocking if lower-priority coroutines are active, thus mapping
779	low-priority coroutines to idle/background tasks).	2749	low-priority coroutines to idle/background tasks).
780		2750
		2751	It is recommended to give C<ev_check> watchers highest (C<EV_MAXPRI>)
		2752	priority, to ensure that they are being run before any other watchers
		2753	after the poll (this doesn't matter for C<ev_prepare> watchers).
		2754
		2755	Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
		2756	activate ("feed") events into libev. While libev fully supports this, they
		2757	might get executed before other C<ev_check> watchers did their job. As
		2758	C<ev_check> watchers are often used to embed other (non-libev) event
		2759	loops those other event loops might be in an unusable state until their
		2760	C<ev_check> watcher ran (always remind yourself to coexist peacefully with
		2761	others).
		2762
		2763	=head3 Watcher-Specific Functions and Data Members
		2764
781	=over 4	2765	=over 4
782		2766
783	=item ev_prepare_init (ev_prepare *, callback)	2767	=item ev_prepare_init (ev_prepare *, callback)
784		2768
785	=item ev_check_init (ev_check *, callback)	2769	=item ev_check_init (ev_check *, callback)
786		2770
787	Initialises and configures the prepare or check watcher - they have no	2771	Initialises and configures the prepare or check watcher - they have no
788	parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>	2772	parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
789	macros, but using them is utterly, utterly and completely pointless.	2773	macros, but using them is utterly, utterly, utterly and completely
		2774	pointless.
790		2775
791	=back	2776	=back
792		2777
		2778	=head3 Examples
		2779
		2780	There are a number of principal ways to embed other event loops or modules
		2781	into libev. Here are some ideas on how to include libadns into libev
		2782	(there is a Perl module named C<EV::ADNS> that does this, which you could
		2783	use as a working example. Another Perl module named C<EV::Glib> embeds a
		2784	Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
		2785	Glib event loop).
		2786
		2787	Method 1: Add IO watchers and a timeout watcher in a prepare handler,
		2788	and in a check watcher, destroy them and call into libadns. What follows
		2789	is pseudo-code only of course. This requires you to either use a low
		2790	priority for the check watcher or use C<ev_clear_pending> explicitly, as
		2791	the callbacks for the IO/timeout watchers might not have been called yet.
		2792
		2793	static ev_io iow [nfd];
		2794	static ev_timer tw;
		2795
		2796	static void
		2797	io_cb (struct ev_loop *loop, ev_io *w, int revents)
		2798	{
		2799	}
		2800
		2801	// create io watchers for each fd and a timer before blocking
		2802	static void
		2803	adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
		2804	{
		2805	int timeout = 3600000;
		2806	struct pollfd fds [nfd];
		2807	// actual code will need to loop here and realloc etc.
		2808	adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
		2809
		2810	/* the callback is illegal, but won't be called as we stop during check */
		2811	ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
		2812	ev_timer_start (loop, &tw);
		2813
		2814	// create one ev_io per pollfd
		2815	for (int i = 0; i < nfd; ++i)
		2816	{
		2817	ev_io_init (iow + i, io_cb, fds [i].fd,
		2818	((fds [i].events & POLLIN ? EV_READ : 0)
		2819	| (fds [i].events & POLLOUT ? EV_WRITE : 0)));
		2820
		2821	fds [i].revents = 0;
		2822	ev_io_start (loop, iow + i);
		2823	}
		2824	}
		2825
		2826	// stop all watchers after blocking
		2827	static void
		2828	adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
		2829	{
		2830	ev_timer_stop (loop, &tw);
		2831
		2832	for (int i = 0; i < nfd; ++i)
		2833	{
		2834	// set the relevant poll flags
		2835	// could also call adns_processreadable etc. here
		2836	struct pollfd *fd = fds + i;
		2837	int revents = ev_clear_pending (iow + i);
		2838	if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
		2839	if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
		2840
		2841	// now stop the watcher
		2842	ev_io_stop (loop, iow + i);
		2843	}
		2844
		2845	adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
		2846	}
		2847
		2848	Method 2: This would be just like method 1, but you run C<adns_afterpoll>
		2849	in the prepare watcher and would dispose of the check watcher.
		2850
		2851	Method 3: If the module to be embedded supports explicit event
		2852	notification (libadns does), you can also make use of the actual watcher
		2853	callbacks, and only destroy/create the watchers in the prepare watcher.
		2854
		2855	static void
		2856	timer_cb (EV_P_ ev_timer *w, int revents)
		2857	{
		2858	adns_state ads = (adns_state)w->data;
		2859	update_now (EV_A);
		2860
		2861	adns_processtimeouts (ads, &tv_now);
		2862	}
		2863
		2864	static void
		2865	io_cb (EV_P_ ev_io *w, int revents)
		2866	{
		2867	adns_state ads = (adns_state)w->data;
		2868	update_now (EV_A);
		2869
		2870	if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
		2871	if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
		2872	}
		2873
		2874	// do not ever call adns_afterpoll
		2875
		2876	Method 4: Do not use a prepare or check watcher because the module you
		2877	want to embed is not flexible enough to support it. Instead, you can
		2878	override their poll function. The drawback with this solution is that the
		2879	main loop is now no longer controllable by EV. The C<Glib::EV> module uses
		2880	this approach, effectively embedding EV as a client into the horrible
		2881	libglib event loop.
		2882
		2883	static gint
		2884	event_poll_func (GPollFD *fds, guint nfds, gint timeout)
		2885	{
		2886	int got_events = 0;
		2887
		2888	for (n = 0; n < nfds; ++n)
		2889	// create/start io watcher that sets the relevant bits in fds[n] and increment got_events
		2890
		2891	if (timeout >= 0)
		2892	// create/start timer
		2893
		2894	// poll
		2895	ev_run (EV_A_ 0);
		2896
		2897	// stop timer again
		2898	if (timeout >= 0)
		2899	ev_timer_stop (EV_A_ &to);
		2900
		2901	// stop io watchers again - their callbacks should have set
		2902	for (n = 0; n < nfds; ++n)
		2903	ev_io_stop (EV_A_ iow [n]);
		2904
		2905	return got_events;
		2906	}
		2907
		2908
		2909	=head2 C<ev_embed> - when one backend isn't enough...
		2910
		2911	This is a rather advanced watcher type that lets you embed one event loop
		2912	into another (currently only C<ev_io> events are supported in the embedded
		2913	loop, other types of watchers might be handled in a delayed or incorrect
		2914	fashion and must not be used).
		2915
		2916	There are primarily two reasons you would want that: work around bugs and
		2917	prioritise I/O.
		2918
		2919	As an example for a bug workaround, the kqueue backend might only support
		2920	sockets on some platform, so it is unusable as generic backend, but you
		2921	still want to make use of it because you have many sockets and it scales
		2922	so nicely. In this case, you would create a kqueue-based loop and embed
		2923	it into your default loop (which might use e.g. poll). Overall operation
		2924	will be a bit slower because first libev has to call C<poll> and then
		2925	C<kevent>, but at least you can use both mechanisms for what they are
		2926	best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
		2927
		2928	As for prioritising I/O: under rare circumstances you have the case where
		2929	some fds have to be watched and handled very quickly (with low latency),
		2930	and even priorities and idle watchers might have too much overhead. In
		2931	this case you would put all the high priority stuff in one loop and all
		2932	the rest in a second one, and embed the second one in the first.
		2933
		2934	As long as the watcher is active, the callback will be invoked every
		2935	time there might be events pending in the embedded loop. The callback
		2936	must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
		2937	sweep and invoke their callbacks (the callback doesn't need to invoke the
		2938	C<ev_embed_sweep> function directly, it could also start an idle watcher
		2939	to give the embedded loop strictly lower priority for example).
		2940
		2941	You can also set the callback to C<0>, in which case the embed watcher
		2942	will automatically execute the embedded loop sweep whenever necessary.
		2943
		2944	Fork detection will be handled transparently while the C<ev_embed> watcher
		2945	is active, i.e., the embedded loop will automatically be forked when the
		2946	embedding loop forks. In other cases, the user is responsible for calling
		2947	C<ev_loop_fork> on the embedded loop.
		2948
		2949	Unfortunately, not all backends are embeddable: only the ones returned by
		2950	C<ev_embeddable_backends> are, which, unfortunately, does not include any
		2951	portable one.
		2952
		2953	So when you want to use this feature you will always have to be prepared
		2954	that you cannot get an embeddable loop. The recommended way to get around
		2955	this is to have a separate variables for your embeddable loop, try to
		2956	create it, and if that fails, use the normal loop for everything.
		2957
		2958	=head3 C<ev_embed> and fork
		2959
		2960	While the C<ev_embed> watcher is running, forks in the embedding loop will
		2961	automatically be applied to the embedded loop as well, so no special
		2962	fork handling is required in that case. When the watcher is not running,
		2963	however, it is still the task of the libev user to call C<ev_loop_fork ()>
		2964	as applicable.
		2965
		2966	=head3 Watcher-Specific Functions and Data Members
		2967
		2968	=over 4
		2969
		2970	=item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
		2971
		2972	=item ev_embed_set (ev_embed *, callback, struct ev_loop *embedded_loop)
		2973
		2974	Configures the watcher to embed the given loop, which must be
		2975	embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
		2976	invoked automatically, otherwise it is the responsibility of the callback
		2977	to invoke it (it will continue to be called until the sweep has been done,
		2978	if you do not want that, you need to temporarily stop the embed watcher).
		2979
		2980	=item ev_embed_sweep (loop, ev_embed *)
		2981
		2982	Make a single, non-blocking sweep over the embedded loop. This works
		2983	similarly to C<ev_run (embedded_loop, EVRUN_NOWAIT)>, but in the most
		2984	appropriate way for embedded loops.
		2985
		2986	=item struct ev_loop *other [read-only]
		2987
		2988	The embedded event loop.
		2989
		2990	=back
		2991
		2992	=head3 Examples
		2993
		2994	Example: Try to get an embeddable event loop and embed it into the default
		2995	event loop. If that is not possible, use the default loop. The default
		2996	loop is stored in C<loop_hi>, while the embeddable loop is stored in
		2997	C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
		2998	used).
		2999
		3000	struct ev_loop *loop_hi = ev_default_init (0);
		3001	struct ev_loop *loop_lo = 0;
		3002	ev_embed embed;
		3003
		3004	// see if there is a chance of getting one that works
		3005	// (remember that a flags value of 0 means autodetection)
		3006	loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
		3007	? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
		3008	: 0;
		3009
		3010	// if we got one, then embed it, otherwise default to loop_hi
		3011	if (loop_lo)
		3012	{
		3013	ev_embed_init (&embed, 0, loop_lo);
		3014	ev_embed_start (loop_hi, &embed);
		3015	}
		3016	else
		3017	loop_lo = loop_hi;
		3018
		3019	Example: Check if kqueue is available but not recommended and create
		3020	a kqueue backend for use with sockets (which usually work with any
		3021	kqueue implementation). Store the kqueue/socket-only event loop in
		3022	C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
		3023
		3024	struct ev_loop *loop = ev_default_init (0);
		3025	struct ev_loop *loop_socket = 0;
		3026	ev_embed embed;
		3027
		3028	if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
		3029	if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
		3030	{
		3031	ev_embed_init (&embed, 0, loop_socket);
		3032	ev_embed_start (loop, &embed);
		3033	}
		3034
		3035	if (!loop_socket)
		3036	loop_socket = loop;
		3037
		3038	// now use loop_socket for all sockets, and loop for everything else
		3039
		3040
		3041	=head2 C<ev_fork> - the audacity to resume the event loop after a fork
		3042
		3043	Fork watchers are called when a C<fork ()> was detected (usually because
		3044	whoever is a good citizen cared to tell libev about it by calling
		3045	C<ev_default_fork> or C<ev_loop_fork>). The invocation is done before the
		3046	event loop blocks next and before C<ev_check> watchers are being called,
		3047	and only in the child after the fork. If whoever good citizen calling
		3048	C<ev_default_fork> cheats and calls it in the wrong process, the fork
		3049	handlers will be invoked, too, of course.
		3050
		3051	=head3 The special problem of life after fork - how is it possible?
		3052
		3053	Most uses of C<fork()> consist of forking, then some simple calls to set
		3054	up/change the process environment, followed by a call to C<exec()>. This
		3055	sequence should be handled by libev without any problems.
		3056
		3057	This changes when the application actually wants to do event handling
		3058	in the child, or both parent in child, in effect "continuing" after the
		3059	fork.
		3060
		3061	The default mode of operation (for libev, with application help to detect
		3062	forks) is to duplicate all the state in the child, as would be expected
		3063	when I<either> the parent I<or> the child process continues.
		3064
		3065	When both processes want to continue using libev, then this is usually the
		3066	wrong result. In that case, usually one process (typically the parent) is
		3067	supposed to continue with all watchers in place as before, while the other
		3068	process typically wants to start fresh, i.e. without any active watchers.
		3069
		3070	The cleanest and most efficient way to achieve that with libev is to
		3071	simply create a new event loop, which of course will be "empty", and
		3072	use that for new watchers. This has the advantage of not touching more
		3073	memory than necessary, and thus avoiding the copy-on-write, and the
		3074	disadvantage of having to use multiple event loops (which do not support
		3075	signal watchers).
		3076
		3077	When this is not possible, or you want to use the default loop for
		3078	other reasons, then in the process that wants to start "fresh", call
		3079	C<ev_default_destroy ()> followed by C<ev_default_loop (...)>. Destroying
		3080	the default loop will "orphan" (not stop) all registered watchers, so you
		3081	have to be careful not to execute code that modifies those watchers. Note
		3082	also that in that case, you have to re-register any signal watchers.
		3083
		3084	=head3 Watcher-Specific Functions and Data Members
		3085
		3086	=over 4
		3087
		3088	=item ev_fork_init (ev_signal *, callback)
		3089
		3090	Initialises and configures the fork watcher - it has no parameters of any
		3091	kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
		3092	believe me.
		3093
		3094	=back
		3095
		3096
		3097	=head2 C<ev_async> - how to wake up an event loop
		3098
		3099	In general, you cannot use an C<ev_run> from multiple threads or other
		3100	asynchronous sources such as signal handlers (as opposed to multiple event
		3101	loops - those are of course safe to use in different threads).
		3102
		3103	Sometimes, however, you need to wake up an event loop you do not control,
		3104	for example because it belongs to another thread. This is what C<ev_async>
		3105	watchers do: as long as the C<ev_async> watcher is active, you can signal
		3106	it by calling C<ev_async_send>, which is thread- and signal safe.
		3107
		3108	This functionality is very similar to C<ev_signal> watchers, as signals,
		3109	too, are asynchronous in nature, and signals, too, will be compressed
		3110	(i.e. the number of callback invocations may be less than the number of
		3111	C<ev_async_sent> calls).
		3112
		3113	Unlike C<ev_signal> watchers, C<ev_async> works with any event loop, not
		3114	just the default loop.
		3115
		3116	=head3 Queueing
		3117
		3118	C<ev_async> does not support queueing of data in any way. The reason
		3119	is that the author does not know of a simple (or any) algorithm for a
		3120	multiple-writer-single-reader queue that works in all cases and doesn't
		3121	need elaborate support such as pthreads or unportable memory access
		3122	semantics.
		3123
		3124	That means that if you want to queue data, you have to provide your own
		3125	queue. But at least I can tell you how to implement locking around your
		3126	queue:
		3127
		3128	=over 4
		3129
		3130	=item queueing from a signal handler context
		3131
		3132	To implement race-free queueing, you simply add to the queue in the signal
		3133	handler but you block the signal handler in the watcher callback. Here is
		3134	an example that does that for some fictitious SIGUSR1 handler:
		3135
		3136	static ev_async mysig;
		3137
		3138	static void
		3139	sigusr1_handler (void)
		3140	{
		3141	sometype data;
		3142
		3143	// no locking etc.
		3144	queue_put (data);
		3145	ev_async_send (EV_DEFAULT_ &mysig);
		3146	}
		3147
		3148	static void
		3149	mysig_cb (EV_P_ ev_async *w, int revents)
		3150	{
		3151	sometype data;
		3152	sigset_t block, prev;
		3153
		3154	sigemptyset (&block);
		3155	sigaddset (&block, SIGUSR1);
		3156	sigprocmask (SIG_BLOCK, &block, &prev);
		3157
		3158	while (queue_get (&data))
		3159	process (data);
		3160
		3161	if (sigismember (&prev, SIGUSR1)
		3162	sigprocmask (SIG_UNBLOCK, &block, 0);
		3163	}
		3164
		3165	(Note: pthreads in theory requires you to use C<pthread_setmask>
		3166	instead of C<sigprocmask> when you use threads, but libev doesn't do it
		3167	either...).
		3168
		3169	=item queueing from a thread context
		3170
		3171	The strategy for threads is different, as you cannot (easily) block
		3172	threads but you can easily preempt them, so to queue safely you need to
		3173	employ a traditional mutex lock, such as in this pthread example:
		3174
		3175	static ev_async mysig;
		3176	static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
		3177
		3178	static void
		3179	otherthread (void)
		3180	{
		3181	// only need to lock the actual queueing operation
		3182	pthread_mutex_lock (&mymutex);
		3183	queue_put (data);
		3184	pthread_mutex_unlock (&mymutex);
		3185
		3186	ev_async_send (EV_DEFAULT_ &mysig);
		3187	}
		3188
		3189	static void
		3190	mysig_cb (EV_P_ ev_async *w, int revents)
		3191	{
		3192	pthread_mutex_lock (&mymutex);
		3193
		3194	while (queue_get (&data))
		3195	process (data);
		3196
		3197	pthread_mutex_unlock (&mymutex);
		3198	}
		3199
		3200	=back
		3201
		3202
		3203	=head3 Watcher-Specific Functions and Data Members
		3204
		3205	=over 4
		3206
		3207	=item ev_async_init (ev_async *, callback)
		3208
		3209	Initialises and configures the async watcher - it has no parameters of any
		3210	kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
		3211	trust me.
		3212
		3213	=item ev_async_send (loop, ev_async *)
		3214
		3215	Sends/signals/activates the given C<ev_async> watcher, that is, feeds
		3216	an C<EV_ASYNC> event on the watcher into the event loop. Unlike
		3217	C<ev_feed_event>, this call is safe to do from other threads, signal or
		3218	similar contexts (see the discussion of C<EV_ATOMIC_T> in the embedding
		3219	section below on what exactly this means).
		3220
		3221	Note that, as with other watchers in libev, multiple events might get
		3222	compressed into a single callback invocation (another way to look at this
		3223	is that C<ev_async> watchers are level-triggered, set on C<ev_async_send>,
		3224	reset when the event loop detects that).
		3225
		3226	This call incurs the overhead of a system call only once per event loop
		3227	iteration, so while the overhead might be noticeable, it doesn't apply to
		3228	repeated calls to C<ev_async_send> for the same event loop.
		3229
		3230	=item bool = ev_async_pending (ev_async *)
		3231
		3232	Returns a non-zero value when C<ev_async_send> has been called on the
		3233	watcher but the event has not yet been processed (or even noted) by the
		3234	event loop.
		3235
		3236	C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
		3237	the loop iterates next and checks for the watcher to have become active,
		3238	it will reset the flag again. C<ev_async_pending> can be used to very
		3239	quickly check whether invoking the loop might be a good idea.
		3240
		3241	Not that this does I<not> check whether the watcher itself is pending,
		3242	only whether it has been requested to make this watcher pending: there
		3243	is a time window between the event loop checking and resetting the async
		3244	notification, and the callback being invoked.
		3245
		3246	=back
		3247
		3248
793	=head1 OTHER FUNCTIONS	3249	=head1 OTHER FUNCTIONS
794		3250
795	There are some other functions of possible interest. Described. Here. Now.	3251	There are some other functions of possible interest. Described. Here. Now.
796		3252
797	=over 4	3253	=over 4
798		3254
799	=item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)	3255	=item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)
800		3256
801	This function combines a simple timer and an I/O watcher, calls your	3257	This function combines a simple timer and an I/O watcher, calls your
802	callback on whichever event happens first and automatically stop both	3258	callback on whichever event happens first and automatically stops both
803	watchers. This is useful if you want to wait for a single event on an fd	3259	watchers. This is useful if you want to wait for a single event on an fd
804	or timeout without having to allocate/configure/start/stop/free one or	3260	or timeout without having to allocate/configure/start/stop/free one or
805	more watchers yourself.	3261	more watchers yourself.
806		3262
807	If C<fd> is less than 0, then no I/O watcher will be started and events	3263	If C<fd> is less than 0, then no I/O watcher will be started and the
808	is being ignored. Otherwise, an C<ev_io> watcher for the given C<fd> and	3264	C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
809	C<events> set will be craeted and started.	3265	the given C<fd> and C<events> set will be created and started.
810		3266
811	If C<timeout> is less than 0, then no timeout watcher will be	3267	If C<timeout> is less than 0, then no timeout watcher will be
812	started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and	3268	started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
813	repeat = 0) will be started. While C<0> is a valid timeout, it is of	3269	repeat = 0) will be started. C<0> is a valid timeout.
814	dubious value.
815		3270
816	The callback has the type C<void (*cb)(int revents, void *arg)> and gets	3271	The callback has the type C<void (*cb)(int revents, void *arg)> and is
817	passed an C<revents> set like normal event callbacks (a combination of	3272	passed an C<revents> set like normal event callbacks (a combination of
818	C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMEOUT>) and the C<arg>	3273	C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMER>) and the C<arg>
819	value passed to C<ev_once>:	3274	value passed to C<ev_once>. Note that it is possible to receive I<both>
		3275	a timeout and an io event at the same time - you probably should give io
		3276	events precedence.
820		3277
		3278	Example: wait up to ten seconds for data to appear on STDIN_FILENO.
		3279
821	static void stdin_ready (int revents, void *arg)	3280	static void stdin_ready (int revents, void *arg)
822	{	3281	{
823	if (revents & EV_TIMEOUT)
824	/* doh, nothing entered */;
825	else if (revents & EV_READ)	3282	if (revents & EV_READ)
826	/* stdin might have data for us, joy! */;	3283	/* stdin might have data for us, joy! */;
		3284	else if (revents & EV_TIMER)
		3285	/* doh, nothing entered */;
827	}	3286	}
828		3287
829	ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);	3288	ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
830
831	=item ev_feed_event (loop, watcher, int events)
832
833	Feeds the given event set into the event loop, as if the specified event
834	had happened for the specified watcher (which must be a pointer to an
835	initialised but not necessarily started event watcher).
836		3289
837	=item ev_feed_fd_event (loop, int fd, int revents)	3290	=item ev_feed_fd_event (loop, int fd, int revents)
838		3291
839	Feed an event on the given fd, as if a file descriptor backend detected	3292	Feed an event on the given fd, as if a file descriptor backend detected
840	the given events it.	3293	the given events it.
841		3294
842	=item ev_feed_signal_event (loop, int signum)	3295	=item ev_feed_signal_event (loop, int signum)
843		3296
844	Feed an event as if the given signal occured (loop must be the default loop!).	3297	Feed an event as if the given signal occurred (C<loop> must be the default
		3298	loop!).
845		3299
846	=back	3300	=back
		3301
847		3302
848	=head1 LIBEVENT EMULATION	3303	=head1 LIBEVENT EMULATION
849		3304
850	Libev offers a compatibility emulation layer for libevent. It cannot	3305	Libev offers a compatibility emulation layer for libevent. It cannot
851	emulate the internals of libevent, so here are some usage hints:	3306	emulate the internals of libevent, so here are some usage hints:
…		…
863		3318
864	=item * Priorities are not currently supported. Initialising priorities	3319	=item * Priorities are not currently supported. Initialising priorities
865	will fail and all watchers will have the same priority, even though there	3320	will fail and all watchers will have the same priority, even though there
866	is an ev_pri field.	3321	is an ev_pri field.
867		3322
		3323	=item * In libevent, the last base created gets the signals, in libev, the
		3324	first base created (== the default loop) gets the signals.
		3325
868	=item * Other members are not supported.	3326	=item * Other members are not supported.
869		3327
870	=item * The libev emulation is I<not> ABI compatible to libevent, you need	3328	=item * The libev emulation is I<not> ABI compatible to libevent, you need
871	to use the libev header file and library.	3329	to use the libev header file and library.
872		3330
873	=back	3331	=back
874		3332
875	=head1 C++ SUPPORT	3333	=head1 C++ SUPPORT
876		3334
877	TBD.	3335	Libev comes with some simplistic wrapper classes for C++ that mainly allow
		3336	you to use some convenience methods to start/stop watchers and also change
		3337	the callback model to a model using method callbacks on objects.
		3338
		3339	To use it,
		3340
		3341	#include <ev++.h>
		3342
		3343	This automatically includes F<ev.h> and puts all of its definitions (many
		3344	of them macros) into the global namespace. All C++ specific things are
		3345	put into the C<ev> namespace. It should support all the same embedding
		3346	options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
		3347
		3348	Care has been taken to keep the overhead low. The only data member the C++
		3349	classes add (compared to plain C-style watchers) is the event loop pointer
		3350	that the watcher is associated with (or no additional members at all if
		3351	you disable C<EV_MULTIPLICITY> when embedding libev).
		3352
		3353	Currently, functions, and static and non-static member functions can be
		3354	used as callbacks. Other types should be easy to add as long as they only
		3355	need one additional pointer for context. If you need support for other
		3356	types of functors please contact the author (preferably after implementing
		3357	it).
		3358
		3359	Here is a list of things available in the C<ev> namespace:
		3360
		3361	=over 4
		3362
		3363	=item C<ev::READ>, C<ev::WRITE> etc.
		3364
		3365	These are just enum values with the same values as the C<EV_READ> etc.
		3366	macros from F<ev.h>.
		3367
		3368	=item C<ev::tstamp>, C<ev::now>
		3369
		3370	Aliases to the same types/functions as with the C<ev_> prefix.
		3371
		3372	=item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
		3373
		3374	For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
		3375	the same name in the C<ev> namespace, with the exception of C<ev_signal>
		3376	which is called C<ev::sig> to avoid clashes with the C<signal> macro
		3377	defines by many implementations.
		3378
		3379	All of those classes have these methods:
		3380
		3381	=over 4
		3382
		3383	=item ev::TYPE::TYPE ()
		3384
		3385	=item ev::TYPE::TYPE (loop)
		3386
		3387	=item ev::TYPE::~TYPE
		3388
		3389	The constructor (optionally) takes an event loop to associate the watcher
		3390	with. If it is omitted, it will use C<EV_DEFAULT>.
		3391
		3392	The constructor calls C<ev_init> for you, which means you have to call the
		3393	C<set> method before starting it.
		3394
		3395	It will not set a callback, however: You have to call the templated C<set>
		3396	method to set a callback before you can start the watcher.
		3397
		3398	(The reason why you have to use a method is a limitation in C++ which does
		3399	not allow explicit template arguments for constructors).
		3400
		3401	The destructor automatically stops the watcher if it is active.
		3402
		3403	=item w->set<class, &class::method> (object *)
		3404
		3405	This method sets the callback method to call. The method has to have a
		3406	signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
		3407	first argument and the C<revents> as second. The object must be given as
		3408	parameter and is stored in the C<data> member of the watcher.
		3409
		3410	This method synthesizes efficient thunking code to call your method from
		3411	the C callback that libev requires. If your compiler can inline your
		3412	callback (i.e. it is visible to it at the place of the C<set> call and
		3413	your compiler is good :), then the method will be fully inlined into the
		3414	thunking function, making it as fast as a direct C callback.
		3415
		3416	Example: simple class declaration and watcher initialisation
		3417
		3418	struct myclass
		3419	{
		3420	void io_cb (ev::io &w, int revents) { }
		3421	}
		3422
		3423	myclass obj;
		3424	ev::io iow;
		3425	iow.set <myclass, &myclass::io_cb> (&obj);
		3426
		3427	=item w->set (object *)
		3428
		3429	This is a variation of a method callback - leaving out the method to call
		3430	will default the method to C<operator ()>, which makes it possible to use
		3431	functor objects without having to manually specify the C<operator ()> all
		3432	the time. Incidentally, you can then also leave out the template argument
		3433	list.
		3434
		3435	The C<operator ()> method prototype must be C<void operator ()(watcher &w,
		3436	int revents)>.
		3437
		3438	See the method-C<set> above for more details.
		3439
		3440	Example: use a functor object as callback.
		3441
		3442	struct myfunctor
		3443	{
		3444	void operator() (ev::io &w, int revents)
		3445	{
		3446	...
		3447	}
		3448	}
		3449
		3450	myfunctor f;
		3451
		3452	ev::io w;
		3453	w.set (&f);
		3454
		3455	=item w->set<function> (void *data = 0)
		3456
		3457	Also sets a callback, but uses a static method or plain function as
		3458	callback. The optional C<data> argument will be stored in the watcher's
		3459	C<data> member and is free for you to use.
		3460
		3461	The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
		3462
		3463	See the method-C<set> above for more details.
		3464
		3465	Example: Use a plain function as callback.
		3466
		3467	static void io_cb (ev::io &w, int revents) { }
		3468	iow.set <io_cb> ();
		3469
		3470	=item w->set (loop)
		3471
		3472	Associates a different C<struct ev_loop> with this watcher. You can only
		3473	do this when the watcher is inactive (and not pending either).
		3474
		3475	=item w->set ([arguments])
		3476
		3477	Basically the same as C<ev_TYPE_set>, with the same arguments. Either this
		3478	method or a suitable start method must be called at least once. Unlike the
		3479	C counterpart, an active watcher gets automatically stopped and restarted
		3480	when reconfiguring it with this method.
		3481
		3482	=item w->start ()
		3483
		3484	Starts the watcher. Note that there is no C<loop> argument, as the
		3485	constructor already stores the event loop.
		3486
		3487	=item w->start ([arguments])
		3488
		3489	Instead of calling C<set> and C<start> methods separately, it is often
		3490	convenient to wrap them in one call. Uses the same type of arguments as
		3491	the configure C<set> method of the watcher.
		3492
		3493	=item w->stop ()
		3494
		3495	Stops the watcher if it is active. Again, no C<loop> argument.
		3496
		3497	=item w->again () (C<ev::timer>, C<ev::periodic> only)
		3498
		3499	For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
		3500	C<ev_TYPE_again> function.
		3501
		3502	=item w->sweep () (C<ev::embed> only)
		3503
		3504	Invokes C<ev_embed_sweep>.
		3505
		3506	=item w->update () (C<ev::stat> only)
		3507
		3508	Invokes C<ev_stat_stat>.
		3509
		3510	=back
		3511
		3512	=back
		3513
		3514	Example: Define a class with two I/O and idle watchers, start the I/O
		3515	watchers in the constructor.
		3516
		3517	class myclass
		3518	{
		3519	ev::io io ; void io_cb (ev::io &w, int revents);
		3520	ev::io2 io2 ; void io2_cb (ev::io &w, int revents);
		3521	ev::idle idle; void idle_cb (ev::idle &w, int revents);
		3522
		3523	myclass (int fd)
		3524	{
		3525	io .set <myclass, &myclass::io_cb > (this);
		3526	io2 .set <myclass, &myclass::io2_cb > (this);
		3527	idle.set <myclass, &myclass::idle_cb> (this);
		3528
		3529	io.set (fd, ev::WRITE); // configure the watcher
		3530	io.start (); // start it whenever convenient
		3531
		3532	io2.start (fd, ev::READ); // set + start in one call
		3533	}
		3534	};
		3535
		3536
		3537	=head1 OTHER LANGUAGE BINDINGS
		3538
		3539	Libev does not offer other language bindings itself, but bindings for a
		3540	number of languages exist in the form of third-party packages. If you know
		3541	any interesting language binding in addition to the ones listed here, drop
		3542	me a note.
		3543
		3544	=over 4
		3545
		3546	=item Perl
		3547
		3548	The EV module implements the full libev API and is actually used to test
		3549	libev. EV is developed together with libev. Apart from the EV core module,
		3550	there are additional modules that implement libev-compatible interfaces
		3551	to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
		3552	C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
		3553	and C<EV::Glib>).
		3554
		3555	It can be found and installed via CPAN, its homepage is at
		3556	L<http://software.schmorp.de/pkg/EV>.
		3557
		3558	=item Python
		3559
		3560	Python bindings can be found at L<http://code.google.com/p/pyev/>. It
		3561	seems to be quite complete and well-documented.
		3562
		3563	=item Ruby
		3564
		3565	Tony Arcieri has written a ruby extension that offers access to a subset
		3566	of the libev API and adds file handle abstractions, asynchronous DNS and
		3567	more on top of it. It can be found via gem servers. Its homepage is at
		3568	L<http://rev.rubyforge.org/>.
		3569
		3570	Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
		3571	makes rev work even on mingw.
		3572
		3573	=item Haskell
		3574
		3575	A haskell binding to libev is available at
		3576	L<http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
		3577
		3578	=item D
		3579
		3580	Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
		3581	be found at L<http://proj.llucax.com.ar/wiki/evd>.
		3582
		3583	=item Ocaml
		3584
		3585	Erkki Seppala has written Ocaml bindings for libev, to be found at
		3586	L<http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
		3587
		3588	=item Lua
		3589
		3590	Brian Maher has written a partial interface to libev for lua (at the
		3591	time of this writing, only C<ev_io> and C<ev_timer>), to be found at
		3592	L<http://github.com/brimworks/lua-ev>.
		3593
		3594	=back
		3595
		3596
		3597	=head1 MACRO MAGIC
		3598
		3599	Libev can be compiled with a variety of options, the most fundamental
		3600	of which is C<EV_MULTIPLICITY>. This option determines whether (most)
		3601	functions and callbacks have an initial C<struct ev_loop *> argument.
		3602
		3603	To make it easier to write programs that cope with either variant, the
		3604	following macros are defined:
		3605
		3606	=over 4
		3607
		3608	=item C<EV_A>, C<EV_A_>
		3609
		3610	This provides the loop I<argument> for functions, if one is required ("ev
		3611	loop argument"). The C<EV_A> form is used when this is the sole argument,
		3612	C<EV_A_> is used when other arguments are following. Example:
		3613
		3614	ev_unref (EV_A);
		3615	ev_timer_add (EV_A_ watcher);
		3616	ev_run (EV_A_ 0);
		3617
		3618	It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
		3619	which is often provided by the following macro.
		3620
		3621	=item C<EV_P>, C<EV_P_>
		3622
		3623	This provides the loop I<parameter> for functions, if one is required ("ev
		3624	loop parameter"). The C<EV_P> form is used when this is the sole parameter,
		3625	C<EV_P_> is used when other parameters are following. Example:
		3626
		3627	// this is how ev_unref is being declared
		3628	static void ev_unref (EV_P);
		3629
		3630	// this is how you can declare your typical callback
		3631	static void cb (EV_P_ ev_timer *w, int revents)
		3632
		3633	It declares a parameter C<loop> of type C<struct ev_loop *>, quite
		3634	suitable for use with C<EV_A>.
		3635
		3636	=item C<EV_DEFAULT>, C<EV_DEFAULT_>
		3637
		3638	Similar to the other two macros, this gives you the value of the default
		3639	loop, if multiple loops are supported ("ev loop default").
		3640
		3641	=item C<EV_DEFAULT_UC>, C<EV_DEFAULT_UC_>
		3642
		3643	Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
		3644	default loop has been initialised (C<UC> == unchecked). Their behaviour
		3645	is undefined when the default loop has not been initialised by a previous
		3646	execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
		3647
		3648	It is often prudent to use C<EV_DEFAULT> when initialising the first
		3649	watcher in a function but use C<EV_DEFAULT_UC> afterwards.
		3650
		3651	=back
		3652
		3653	Example: Declare and initialise a check watcher, utilising the above
		3654	macros so it will work regardless of whether multiple loops are supported
		3655	or not.
		3656
		3657	static void
		3658	check_cb (EV_P_ ev_timer *w, int revents)
		3659	{
		3660	ev_check_stop (EV_A_ w);
		3661	}
		3662
		3663	ev_check check;
		3664	ev_check_init (&check, check_cb);
		3665	ev_check_start (EV_DEFAULT_ &check);
		3666	ev_run (EV_DEFAULT_ 0);
		3667
		3668	=head1 EMBEDDING
		3669
		3670	Libev can (and often is) directly embedded into host
		3671	applications. Examples of applications that embed it include the Deliantra
		3672	Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
		3673	and rxvt-unicode.
		3674
		3675	The goal is to enable you to just copy the necessary files into your
		3676	source directory without having to change even a single line in them, so
		3677	you can easily upgrade by simply copying (or having a checked-out copy of
		3678	libev somewhere in your source tree).
		3679
		3680	=head2 FILESETS
		3681
		3682	Depending on what features you need you need to include one or more sets of files
		3683	in your application.
		3684
		3685	=head3 CORE EVENT LOOP
		3686
		3687	To include only the libev core (all the C<ev_*> functions), with manual
		3688	configuration (no autoconf):
		3689
		3690	#define EV_STANDALONE 1
		3691	#include "ev.c"
		3692
		3693	This will automatically include F<ev.h>, too, and should be done in a
		3694	single C source file only to provide the function implementations. To use
		3695	it, do the same for F<ev.h> in all files wishing to use this API (best
		3696	done by writing a wrapper around F<ev.h> that you can include instead and
		3697	where you can put other configuration options):
		3698
		3699	#define EV_STANDALONE 1
		3700	#include "ev.h"
		3701
		3702	Both header files and implementation files can be compiled with a C++
		3703	compiler (at least, that's a stated goal, and breakage will be treated
		3704	as a bug).
		3705
		3706	You need the following files in your source tree, or in a directory
		3707	in your include path (e.g. in libev/ when using -Ilibev):
		3708
		3709	ev.h
		3710	ev.c
		3711	ev_vars.h
		3712	ev_wrap.h
		3713
		3714	ev_win32.c required on win32 platforms only
		3715
		3716	ev_select.c only when select backend is enabled (which is enabled by default)
		3717	ev_poll.c only when poll backend is enabled (disabled by default)
		3718	ev_epoll.c only when the epoll backend is enabled (disabled by default)
		3719	ev_kqueue.c only when the kqueue backend is enabled (disabled by default)
		3720	ev_port.c only when the solaris port backend is enabled (disabled by default)
		3721
		3722	F<ev.c> includes the backend files directly when enabled, so you only need
		3723	to compile this single file.
		3724
		3725	=head3 LIBEVENT COMPATIBILITY API
		3726
		3727	To include the libevent compatibility API, also include:
		3728
		3729	#include "event.c"
		3730
		3731	in the file including F<ev.c>, and:
		3732
		3733	#include "event.h"
		3734
		3735	in the files that want to use the libevent API. This also includes F<ev.h>.
		3736
		3737	You need the following additional files for this:
		3738
		3739	event.h
		3740	event.c
		3741
		3742	=head3 AUTOCONF SUPPORT
		3743
		3744	Instead of using C<EV_STANDALONE=1> and providing your configuration in
		3745	whatever way you want, you can also C<m4_include([libev.m4])> in your
		3746	F<configure.ac> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
		3747	include F<config.h> and configure itself accordingly.
		3748
		3749	For this of course you need the m4 file:
		3750
		3751	libev.m4
		3752
		3753	=head2 PREPROCESSOR SYMBOLS/MACROS
		3754
		3755	Libev can be configured via a variety of preprocessor symbols you have to
		3756	define before including (or compiling) any of its files. The default in
		3757	the absence of autoconf is documented for every option.
		3758
		3759	Symbols marked with "(h)" do not change the ABI, and can have different
		3760	values when compiling libev vs. including F<ev.h>, so it is permissible
		3761	to redefine them before including F<ev.h> without breaking compatibility
		3762	to a compiled library. All other symbols change the ABI, which means all
		3763	users of libev and the libev code itself must be compiled with compatible
		3764	settings.
		3765
		3766	=over 4
		3767
		3768	=item EV_COMPAT3 (h)
		3769
		3770	Backwards compatibility is a major concern for libev. This is why this
		3771	release of libev comes with wrappers for the functions and symbols that
		3772	have been renamed between libev version 3 and 4.
		3773
		3774	You can disable these wrappers (to test compatibility with future
		3775	versions) by defining C<EV_COMPAT3> to C<0> when compiling your
		3776	sources. This has the additional advantage that you can drop the C<struct>
		3777	from C<struct ev_loop> declarations, as libev will provide an C<ev_loop>
		3778	typedef in that case.
		3779
		3780	In some future version, the default for C<EV_COMPAT3> will become C<0>,
		3781	and in some even more future version the compatibility code will be
		3782	removed completely.
		3783
		3784	=item EV_STANDALONE (h)
		3785
		3786	Must always be C<1> if you do not use autoconf configuration, which
		3787	keeps libev from including F<config.h>, and it also defines dummy
		3788	implementations for some libevent functions (such as logging, which is not
		3789	supported). It will also not define any of the structs usually found in
		3790	F<event.h> that are not directly supported by the libev core alone.
		3791
		3792	In standalone mode, libev will still try to automatically deduce the
		3793	configuration, but has to be more conservative.
		3794
		3795	=item EV_USE_MONOTONIC
		3796
		3797	If defined to be C<1>, libev will try to detect the availability of the
		3798	monotonic clock option at both compile time and runtime. Otherwise no
		3799	use of the monotonic clock option will be attempted. If you enable this,
		3800	you usually have to link against librt or something similar. Enabling it
		3801	when the functionality isn't available is safe, though, although you have
		3802	to make sure you link against any libraries where the C<clock_gettime>
		3803	function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
		3804
		3805	=item EV_USE_REALTIME
		3806
		3807	If defined to be C<1>, libev will try to detect the availability of the
		3808	real-time clock option at compile time (and assume its availability
		3809	at runtime if successful). Otherwise no use of the real-time clock
		3810	option will be attempted. This effectively replaces C<gettimeofday>
		3811	by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
		3812	correctness. See the note about libraries in the description of
		3813	C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
		3814	C<EV_USE_CLOCK_SYSCALL>.
		3815
		3816	=item EV_USE_CLOCK_SYSCALL
		3817
		3818	If defined to be C<1>, libev will try to use a direct syscall instead
		3819	of calling the system-provided C<clock_gettime> function. This option
		3820	exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
		3821	unconditionally pulls in C<libpthread>, slowing down single-threaded
		3822	programs needlessly. Using a direct syscall is slightly slower (in
		3823	theory), because no optimised vdso implementation can be used, but avoids
		3824	the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
		3825	higher, as it simplifies linking (no need for C<-lrt>).
		3826
		3827	=item EV_USE_NANOSLEEP
		3828
		3829	If defined to be C<1>, libev will assume that C<nanosleep ()> is available
		3830	and will use it for delays. Otherwise it will use C<select ()>.
		3831
		3832	=item EV_USE_EVENTFD
		3833
		3834	If defined to be C<1>, then libev will assume that C<eventfd ()> is
		3835	available and will probe for kernel support at runtime. This will improve
		3836	C<ev_signal> and C<ev_async> performance and reduce resource consumption.
		3837	If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
		3838	2.7 or newer, otherwise disabled.
		3839
		3840	=item EV_USE_SELECT
		3841
		3842	If undefined or defined to be C<1>, libev will compile in support for the
		3843	C<select>(2) backend. No attempt at auto-detection will be done: if no
		3844	other method takes over, select will be it. Otherwise the select backend
		3845	will not be compiled in.
		3846
		3847	=item EV_SELECT_USE_FD_SET
		3848
		3849	If defined to C<1>, then the select backend will use the system C<fd_set>
		3850	structure. This is useful if libev doesn't compile due to a missing
		3851	C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
		3852	on exotic systems. This usually limits the range of file descriptors to
		3853	some low limit such as 1024 or might have other limitations (winsocket
		3854	only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
		3855	configures the maximum size of the C<fd_set>.
		3856
		3857	=item EV_SELECT_IS_WINSOCKET
		3858
		3859	When defined to C<1>, the select backend will assume that
		3860	select/socket/connect etc. don't understand file descriptors but
		3861	wants osf handles on win32 (this is the case when the select to
		3862	be used is the winsock select). This means that it will call
		3863	C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
		3864	it is assumed that all these functions actually work on fds, even
		3865	on win32. Should not be defined on non-win32 platforms.
		3866
		3867	=item EV_FD_TO_WIN32_HANDLE(fd)
		3868
		3869	If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
		3870	file descriptors to socket handles. When not defining this symbol (the
		3871	default), then libev will call C<_get_osfhandle>, which is usually
		3872	correct. In some cases, programs use their own file descriptor management,
		3873	in which case they can provide this function to map fds to socket handles.
		3874
		3875	=item EV_WIN32_HANDLE_TO_FD(handle)
		3876
		3877	If C<EV_SELECT_IS_WINSOCKET> then libev maps handles to file descriptors
		3878	using the standard C<_open_osfhandle> function. For programs implementing
		3879	their own fd to handle mapping, overwriting this function makes it easier
		3880	to do so. This can be done by defining this macro to an appropriate value.
		3881
		3882	=item EV_WIN32_CLOSE_FD(fd)
		3883
		3884	If programs implement their own fd to handle mapping on win32, then this
		3885	macro can be used to override the C<close> function, useful to unregister
		3886	file descriptors again. Note that the replacement function has to close
		3887	the underlying OS handle.
		3888
		3889	=item EV_USE_POLL
		3890
		3891	If defined to be C<1>, libev will compile in support for the C<poll>(2)
		3892	backend. Otherwise it will be enabled on non-win32 platforms. It
		3893	takes precedence over select.
		3894
		3895	=item EV_USE_EPOLL
		3896
		3897	If defined to be C<1>, libev will compile in support for the Linux
		3898	C<epoll>(7) backend. Its availability will be detected at runtime,
		3899	otherwise another method will be used as fallback. This is the preferred
		3900	backend for GNU/Linux systems. If undefined, it will be enabled if the
		3901	headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
		3902
		3903	=item EV_USE_KQUEUE
		3904
		3905	If defined to be C<1>, libev will compile in support for the BSD style
		3906	C<kqueue>(2) backend. Its actual availability will be detected at runtime,
		3907	otherwise another method will be used as fallback. This is the preferred
		3908	backend for BSD and BSD-like systems, although on most BSDs kqueue only
		3909	supports some types of fds correctly (the only platform we found that
		3910	supports ptys for example was NetBSD), so kqueue might be compiled in, but
		3911	not be used unless explicitly requested. The best way to use it is to find
		3912	out whether kqueue supports your type of fd properly and use an embedded
		3913	kqueue loop.
		3914
		3915	=item EV_USE_PORT
		3916
		3917	If defined to be C<1>, libev will compile in support for the Solaris
		3918	10 port style backend. Its availability will be detected at runtime,
		3919	otherwise another method will be used as fallback. This is the preferred
		3920	backend for Solaris 10 systems.
		3921
		3922	=item EV_USE_DEVPOLL
		3923
		3924	Reserved for future expansion, works like the USE symbols above.
		3925
		3926	=item EV_USE_INOTIFY
		3927
		3928	If defined to be C<1>, libev will compile in support for the Linux inotify
		3929	interface to speed up C<ev_stat> watchers. Its actual availability will
		3930	be detected at runtime. If undefined, it will be enabled if the headers
		3931	indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
		3932
		3933	=item EV_ATOMIC_T
		3934
		3935	Libev requires an integer type (suitable for storing C<0> or C<1>) whose
		3936	access is atomic with respect to other threads or signal contexts. No such
		3937	type is easily found in the C language, so you can provide your own type
		3938	that you know is safe for your purposes. It is used both for signal handler "locking"
		3939	as well as for signal and thread safety in C<ev_async> watchers.
		3940
		3941	In the absence of this define, libev will use C<sig_atomic_t volatile>
		3942	(from F<signal.h>), which is usually good enough on most platforms.
		3943
		3944	=item EV_H (h)
		3945
		3946	The name of the F<ev.h> header file used to include it. The default if
		3947	undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
		3948	used to virtually rename the F<ev.h> header file in case of conflicts.
		3949
		3950	=item EV_CONFIG_H (h)
		3951
		3952	If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
		3953	F<ev.c>'s idea of where to find the F<config.h> file, similarly to
		3954	C<EV_H>, above.
		3955
		3956	=item EV_EVENT_H (h)
		3957
		3958	Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
		3959	of how the F<event.h> header can be found, the default is C<"event.h">.
		3960
		3961	=item EV_PROTOTYPES (h)
		3962
		3963	If defined to be C<0>, then F<ev.h> will not define any function
		3964	prototypes, but still define all the structs and other symbols. This is
		3965	occasionally useful if you want to provide your own wrapper functions
		3966	around libev functions.
		3967
		3968	=item EV_MULTIPLICITY
		3969
		3970	If undefined or defined to C<1>, then all event-loop-specific functions
		3971	will have the C<struct ev_loop *> as first argument, and you can create
		3972	additional independent event loops. Otherwise there will be no support
		3973	for multiple event loops and there is no first event loop pointer
		3974	argument. Instead, all functions act on the single default loop.
		3975
		3976	=item EV_MINPRI
		3977
		3978	=item EV_MAXPRI
		3979
		3980	The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
		3981	C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
		3982	provide for more priorities by overriding those symbols (usually defined
		3983	to be C<-2> and C<2>, respectively).
		3984
		3985	When doing priority-based operations, libev usually has to linearly search
		3986	all the priorities, so having many of them (hundreds) uses a lot of space
		3987	and time, so using the defaults of five priorities (-2 .. +2) is usually
		3988	fine.
		3989
		3990	If your embedding application does not need any priorities, defining these
		3991	both to C<0> will save some memory and CPU.
		3992
		3993	=item EV_PERIODIC_ENABLE, EV_IDLE_ENABLE, EV_EMBED_ENABLE, EV_STAT_ENABLE,
		3994	EV_PREPARE_ENABLE, EV_CHECK_ENABLE, EV_FORK_ENABLE, EV_SIGNAL_ENABLE,
		3995	EV_ASYNC_ENABLE, EV_CHILD_ENABLE.
		3996
		3997	If undefined or defined to be C<1> (and the platform supports it), then
		3998	the respective watcher type is supported. If defined to be C<0>, then it
		3999	is not. Disabling watcher types mainly saves code size.
		4000
		4001	=item EV_FEATURES
		4002
		4003	If you need to shave off some kilobytes of code at the expense of some
		4004	speed (but with the full API), you can define this symbol to request
		4005	certain subsets of functionality. The default is to enable all features
		4006	that can be enabled on the platform.
		4007
		4008	A typical way to use this symbol is to define it to C<0> (or to a bitset
		4009	with some broad features you want) and then selectively re-enable
		4010	additional parts you want, for example if you want everything minimal,
		4011	but multiple event loop support, async and child watchers and the poll
		4012	backend, use this:
		4013
		4014	#define EV_FEATURES 0
		4015	#define EV_MULTIPLICITY 1
		4016	#define EV_USE_POLL 1
		4017	#define EV_CHILD_ENABLE 1
		4018	#define EV_ASYNC_ENABLE 1
		4019
		4020	The actual value is a bitset, it can be a combination of the following
		4021	values:
		4022
		4023	=over 4
		4024
		4025	=item C<1> - faster/larger code
		4026
		4027	Use larger code to speed up some operations.
		4028
		4029	Currently this is used to override some inlining decisions (enlarging the
		4030	code size by roughly 30% on amd64).
		4031
		4032	When optimising for size, use of compiler flags such as C<-Os> with
		4033	gcc is recommended, as well as C<-DNDEBUG>, as libev contains a number of
		4034	assertions.
		4035
		4036	=item C<2> - faster/larger data structures
		4037
		4038	Replaces the small 2-heap for timer management by a faster 4-heap, larger
		4039	hash table sizes and so on. This will usually further increase code size
		4040	and can additionally have an effect on the size of data structures at
		4041	runtime.
		4042
		4043	=item C<4> - full API configuration
		4044
		4045	This enables priorities (sets C<EV_MAXPRI>=2 and C<EV_MINPRI>=-2), and
		4046	enables multiplicity (C<EV_MULTIPLICITY>=1).
		4047
		4048	=item C<8> - full API
		4049
		4050	This enables a lot of the "lesser used" API functions. See C<ev.h> for
		4051	details on which parts of the API are still available without this
		4052	feature, and do not complain if this subset changes over time.
		4053
		4054	=item C<16> - enable all optional watcher types
		4055
		4056	Enables all optional watcher types. If you want to selectively enable
		4057	only some watcher types other than I/O and timers (e.g. prepare,
		4058	embed, async, child...) you can enable them manually by defining
		4059	C<EV_watchertype_ENABLE> to C<1> instead.
		4060
		4061	=item C<32> - enable all backends
		4062
		4063	This enables all backends - without this feature, you need to enable at
		4064	least one backend manually (C<EV_USE_SELECT> is a good choice).
		4065
		4066	=item C<64> - enable OS-specific "helper" APIs
		4067
		4068	Enable inotify, eventfd, signalfd and similar OS-specific helper APIs by
		4069	default.
		4070
		4071	=back
		4072
		4073	Compiling with C<gcc -Os -DEV_STANDALONE -DEV_USE_EPOLL=1 -DEV_FEATURES=0>
		4074	reduces the compiled size of libev from 24.7Kb code/2.8Kb data to 6.5Kb
		4075	code/0.3Kb data on my GNU/Linux amd64 system, while still giving you I/O
		4076	watchers, timers and monotonic clock support.
		4077
		4078	With an intelligent-enough linker (gcc+binutils are intelligent enough
		4079	when you use C<-Wl,--gc-sections -ffunction-sections>) functions unused by
		4080	your program might be left out as well - a binary starting a timer and an
		4081	I/O watcher then might come out at only 5Kb.
		4082
		4083	=item EV_AVOID_STDIO
		4084
		4085	If this is set to C<1> at compiletime, then libev will avoid using stdio
		4086	functions (printf, scanf, perror etc.). This will increase the code size
		4087	somewhat, but if your program doesn't otherwise depend on stdio and your
		4088	libc allows it, this avoids linking in the stdio library which is quite
		4089	big.
		4090
		4091	Note that error messages might become less precise when this option is
		4092	enabled.
		4093
		4094	=item EV_NSIG
		4095
		4096	The highest supported signal number, +1 (or, the number of
		4097	signals): Normally, libev tries to deduce the maximum number of signals
		4098	automatically, but sometimes this fails, in which case it can be
		4099	specified. Also, using a lower number than detected (C<32> should be
		4100	good for about any system in existence) can save some memory, as libev
		4101	statically allocates some 12-24 bytes per signal number.
		4102
		4103	=item EV_PID_HASHSIZE
		4104
		4105	C<ev_child> watchers use a small hash table to distribute workload by
		4106	pid. The default size is C<16> (or C<1> with C<EV_FEATURES> disabled),
		4107	usually more than enough. If you need to manage thousands of children you
		4108	might want to increase this value (I<must> be a power of two).
		4109
		4110	=item EV_INOTIFY_HASHSIZE
		4111
		4112	C<ev_stat> watchers use a small hash table to distribute workload by
		4113	inotify watch id. The default size is C<16> (or C<1> with C<EV_FEATURES>
		4114	disabled), usually more than enough. If you need to manage thousands of
		4115	C<ev_stat> watchers you might want to increase this value (I<must> be a
		4116	power of two).
		4117
		4118	=item EV_USE_4HEAP
		4119
		4120	Heaps are not very cache-efficient. To improve the cache-efficiency of the
		4121	timer and periodics heaps, libev uses a 4-heap when this symbol is defined
		4122	to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
		4123	faster performance with many (thousands) of watchers.
		4124
		4125	The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
		4126	will be C<0>.
		4127
		4128	=item EV_HEAP_CACHE_AT
		4129
		4130	Heaps are not very cache-efficient. To improve the cache-efficiency of the
		4131	timer and periodics heaps, libev can cache the timestamp (I<at>) within
		4132	the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
		4133	which uses 8-12 bytes more per watcher and a few hundred bytes more code,
		4134	but avoids random read accesses on heap changes. This improves performance
		4135	noticeably with many (hundreds) of watchers.
		4136
		4137	The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
		4138	will be C<0>.
		4139
		4140	=item EV_VERIFY
		4141
		4142	Controls how much internal verification (see C<ev_verify ()>) will
		4143	be done: If set to C<0>, no internal verification code will be compiled
		4144	in. If set to C<1>, then verification code will be compiled in, but not
		4145	called. If set to C<2>, then the internal verification code will be
		4146	called once per loop, which can slow down libev. If set to C<3>, then the
		4147	verification code will be called very frequently, which will slow down
		4148	libev considerably.
		4149
		4150	The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
		4151	will be C<0>.
		4152
		4153	=item EV_COMMON
		4154
		4155	By default, all watchers have a C<void *data> member. By redefining
		4156	this macro to something else you can include more and other types of
		4157	members. You have to define it each time you include one of the files,
		4158	though, and it must be identical each time.
		4159
		4160	For example, the perl EV module uses something like this:
		4161
		4162	#define EV_COMMON \
		4163	SV *self; /* contains this struct */ \
		4164	SV *cb_sv, *fh /* note no trailing ";" */
		4165
		4166	=item EV_CB_DECLARE (type)
		4167
		4168	=item EV_CB_INVOKE (watcher, revents)
		4169
		4170	=item ev_set_cb (ev, cb)
		4171
		4172	Can be used to change the callback member declaration in each watcher,
		4173	and the way callbacks are invoked and set. Must expand to a struct member
		4174	definition and a statement, respectively. See the F<ev.h> header file for
		4175	their default definitions. One possible use for overriding these is to
		4176	avoid the C<struct ev_loop *> as first argument in all cases, or to use
		4177	method calls instead of plain function calls in C++.
		4178
		4179	=back
		4180
		4181	=head2 EXPORTED API SYMBOLS
		4182
		4183	If you need to re-export the API (e.g. via a DLL) and you need a list of
		4184	exported symbols, you can use the provided F<Symbol.*> files which list
		4185	all public symbols, one per line:
		4186
		4187	Symbols.ev for libev proper
		4188	Symbols.event for the libevent emulation
		4189
		4190	This can also be used to rename all public symbols to avoid clashes with
		4191	multiple versions of libev linked together (which is obviously bad in
		4192	itself, but sometimes it is inconvenient to avoid this).
		4193
		4194	A sed command like this will create wrapper C<#define>'s that you need to
		4195	include before including F<ev.h>:
		4196
		4197	<Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
		4198
		4199	This would create a file F<wrap.h> which essentially looks like this:
		4200
		4201	#define ev_backend myprefix_ev_backend
		4202	#define ev_check_start myprefix_ev_check_start
		4203	#define ev_check_stop myprefix_ev_check_stop
		4204	...
		4205
		4206	=head2 EXAMPLES
		4207
		4208	For a real-world example of a program the includes libev
		4209	verbatim, you can have a look at the EV perl module
		4210	(L<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
		4211	the F<libev/> subdirectory and includes them in the F<EV/EVAPI.h> (public
		4212	interface) and F<EV.xs> (implementation) files. Only the F<EV.xs> file
		4213	will be compiled. It is pretty complex because it provides its own header
		4214	file.
		4215
		4216	The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
		4217	that everybody includes and which overrides some configure choices:
		4218
		4219	#define EV_FEATURES 8
		4220	#define EV_USE_SELECT 1
		4221	#define EV_PREPARE_ENABLE 1
		4222	#define EV_IDLE_ENABLE 1
		4223	#define EV_SIGNAL_ENABLE 1
		4224	#define EV_CHILD_ENABLE 1
		4225	#define EV_USE_STDEXCEPT 0
		4226	#define EV_CONFIG_H <config.h>
		4227
		4228	#include "ev++.h"
		4229
		4230	And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
		4231
		4232	#include "ev_cpp.h"
		4233	#include "ev.c"
		4234
		4235	=head1 INTERACTION WITH OTHER PROGRAMS OR LIBRARIES
		4236
		4237	=head2 THREADS AND COROUTINES
		4238
		4239	=head3 THREADS
		4240
		4241	All libev functions are reentrant and thread-safe unless explicitly
		4242	documented otherwise, but libev implements no locking itself. This means
		4243	that you can use as many loops as you want in parallel, as long as there
		4244	are no concurrent calls into any libev function with the same loop
		4245	parameter (C<ev_default_*> calls have an implicit default loop parameter,
		4246	of course): libev guarantees that different event loops share no data
		4247	structures that need any locking.
		4248
		4249	Or to put it differently: calls with different loop parameters can be done
		4250	concurrently from multiple threads, calls with the same loop parameter
		4251	must be done serially (but can be done from different threads, as long as
		4252	only one thread ever is inside a call at any point in time, e.g. by using
		4253	a mutex per loop).
		4254
		4255	Specifically to support threads (and signal handlers), libev implements
		4256	so-called C<ev_async> watchers, which allow some limited form of
		4257	concurrency on the same event loop, namely waking it up "from the
		4258	outside".
		4259
		4260	If you want to know which design (one loop, locking, or multiple loops
		4261	without or something else still) is best for your problem, then I cannot
		4262	help you, but here is some generic advice:
		4263
		4264	=over 4
		4265
		4266	=item * most applications have a main thread: use the default libev loop
		4267	in that thread, or create a separate thread running only the default loop.
		4268
		4269	This helps integrating other libraries or software modules that use libev
		4270	themselves and don't care/know about threading.
		4271
		4272	=item * one loop per thread is usually a good model.
		4273
		4274	Doing this is almost never wrong, sometimes a better-performance model
		4275	exists, but it is always a good start.
		4276
		4277	=item * other models exist, such as the leader/follower pattern, where one
		4278	loop is handed through multiple threads in a kind of round-robin fashion.
		4279
		4280	Choosing a model is hard - look around, learn, know that usually you can do
		4281	better than you currently do :-)
		4282
		4283	=item * often you need to talk to some other thread which blocks in the
		4284	event loop.
		4285
		4286	C<ev_async> watchers can be used to wake them up from other threads safely
		4287	(or from signal contexts...).
		4288
		4289	An example use would be to communicate signals or other events that only
		4290	work in the default loop by registering the signal watcher with the
		4291	default loop and triggering an C<ev_async> watcher from the default loop
		4292	watcher callback into the event loop interested in the signal.
		4293
		4294	=back
		4295
		4296	=head4 THREAD LOCKING EXAMPLE
		4297
		4298	Here is a fictitious example of how to run an event loop in a different
		4299	thread than where callbacks are being invoked and watchers are
		4300	created/added/removed.
		4301
		4302	For a real-world example, see the C<EV::Loop::Async> perl module,
		4303	which uses exactly this technique (which is suited for many high-level
		4304	languages).
		4305
		4306	The example uses a pthread mutex to protect the loop data, a condition
		4307	variable to wait for callback invocations, an async watcher to notify the
		4308	event loop thread and an unspecified mechanism to wake up the main thread.
		4309
		4310	First, you need to associate some data with the event loop:
		4311
		4312	typedef struct {
		4313	mutex_t lock; /* global loop lock */
		4314	ev_async async_w;
		4315	thread_t tid;
		4316	cond_t invoke_cv;
		4317	} userdata;
		4318
		4319	void prepare_loop (EV_P)
		4320	{
		4321	// for simplicity, we use a static userdata struct.
		4322	static userdata u;
		4323
		4324	ev_async_init (&u->async_w, async_cb);
		4325	ev_async_start (EV_A_ &u->async_w);
		4326
		4327	pthread_mutex_init (&u->lock, 0);
		4328	pthread_cond_init (&u->invoke_cv, 0);
		4329
		4330	// now associate this with the loop
		4331	ev_set_userdata (EV_A_ u);
		4332	ev_set_invoke_pending_cb (EV_A_ l_invoke);
		4333	ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
		4334
		4335	// then create the thread running ev_loop
		4336	pthread_create (&u->tid, 0, l_run, EV_A);
		4337	}
		4338
		4339	The callback for the C<ev_async> watcher does nothing: the watcher is used
		4340	solely to wake up the event loop so it takes notice of any new watchers
		4341	that might have been added:
		4342
		4343	static void
		4344	async_cb (EV_P_ ev_async *w, int revents)
		4345	{
		4346	// just used for the side effects
		4347	}
		4348
		4349	The C<l_release> and C<l_acquire> callbacks simply unlock/lock the mutex
		4350	protecting the loop data, respectively.
		4351
		4352	static void
		4353	l_release (EV_P)
		4354	{
		4355	userdata *u = ev_userdata (EV_A);
		4356	pthread_mutex_unlock (&u->lock);
		4357	}
		4358
		4359	static void
		4360	l_acquire (EV_P)
		4361	{
		4362	userdata *u = ev_userdata (EV_A);
		4363	pthread_mutex_lock (&u->lock);
		4364	}
		4365
		4366	The event loop thread first acquires the mutex, and then jumps straight
		4367	into C<ev_run>:
		4368
		4369	void *
		4370	l_run (void *thr_arg)
		4371	{
		4372	struct ev_loop *loop = (struct ev_loop *)thr_arg;
		4373
		4374	l_acquire (EV_A);
		4375	pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
		4376	ev_run (EV_A_ 0);
		4377	l_release (EV_A);
		4378
		4379	return 0;
		4380	}
		4381
		4382	Instead of invoking all pending watchers, the C<l_invoke> callback will
		4383	signal the main thread via some unspecified mechanism (signals? pipe
		4384	writes? C<Async::Interrupt>?) and then waits until all pending watchers
		4385	have been called (in a while loop because a) spurious wakeups are possible
		4386	and b) skipping inter-thread-communication when there are no pending
		4387	watchers is very beneficial):
		4388
		4389	static void
		4390	l_invoke (EV_P)
		4391	{
		4392	userdata *u = ev_userdata (EV_A);
		4393
		4394	while (ev_pending_count (EV_A))
		4395	{
		4396	wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
		4397	pthread_cond_wait (&u->invoke_cv, &u->lock);
		4398	}
		4399	}
		4400
		4401	Now, whenever the main thread gets told to invoke pending watchers, it
		4402	will grab the lock, call C<ev_invoke_pending> and then signal the loop
		4403	thread to continue:
		4404
		4405	static void
		4406	real_invoke_pending (EV_P)
		4407	{
		4408	userdata *u = ev_userdata (EV_A);
		4409
		4410	pthread_mutex_lock (&u->lock);
		4411	ev_invoke_pending (EV_A);
		4412	pthread_cond_signal (&u->invoke_cv);
		4413	pthread_mutex_unlock (&u->lock);
		4414	}
		4415
		4416	Whenever you want to start/stop a watcher or do other modifications to an
		4417	event loop, you will now have to lock:
		4418
		4419	ev_timer timeout_watcher;
		4420	userdata *u = ev_userdata (EV_A);
		4421
		4422	ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
		4423
		4424	pthread_mutex_lock (&u->lock);
		4425	ev_timer_start (EV_A_ &timeout_watcher);
		4426	ev_async_send (EV_A_ &u->async_w);
		4427	pthread_mutex_unlock (&u->lock);
		4428
		4429	Note that sending the C<ev_async> watcher is required because otherwise
		4430	an event loop currently blocking in the kernel will have no knowledge
		4431	about the newly added timer. By waking up the loop it will pick up any new
		4432	watchers in the next event loop iteration.
		4433
		4434	=head3 COROUTINES
		4435
		4436	Libev is very accommodating to coroutines ("cooperative threads"):
		4437	libev fully supports nesting calls to its functions from different
		4438	coroutines (e.g. you can call C<ev_run> on the same loop from two
		4439	different coroutines, and switch freely between both coroutines running
		4440	the loop, as long as you don't confuse yourself). The only exception is
		4441	that you must not do this from C<ev_periodic> reschedule callbacks.
		4442
		4443	Care has been taken to ensure that libev does not keep local state inside
		4444	C<ev_run>, and other calls do not usually allow for coroutine switches as
		4445	they do not call any callbacks.
		4446
		4447	=head2 COMPILER WARNINGS
		4448
		4449	Depending on your compiler and compiler settings, you might get no or a
		4450	lot of warnings when compiling libev code. Some people are apparently
		4451	scared by this.
		4452
		4453	However, these are unavoidable for many reasons. For one, each compiler
		4454	has different warnings, and each user has different tastes regarding
		4455	warning options. "Warn-free" code therefore cannot be a goal except when
		4456	targeting a specific compiler and compiler-version.
		4457
		4458	Another reason is that some compiler warnings require elaborate
		4459	workarounds, or other changes to the code that make it less clear and less
		4460	maintainable.
		4461
		4462	And of course, some compiler warnings are just plain stupid, or simply
		4463	wrong (because they don't actually warn about the condition their message
		4464	seems to warn about). For example, certain older gcc versions had some
		4465	warnings that resulted in an extreme number of false positives. These have
		4466	been fixed, but some people still insist on making code warn-free with
		4467	such buggy versions.
		4468
		4469	While libev is written to generate as few warnings as possible,
		4470	"warn-free" code is not a goal, and it is recommended not to build libev
		4471	with any compiler warnings enabled unless you are prepared to cope with
		4472	them (e.g. by ignoring them). Remember that warnings are just that:
		4473	warnings, not errors, or proof of bugs.
		4474
		4475
		4476	=head2 VALGRIND
		4477
		4478	Valgrind has a special section here because it is a popular tool that is
		4479	highly useful. Unfortunately, valgrind reports are very hard to interpret.
		4480
		4481	If you think you found a bug (memory leak, uninitialised data access etc.)
		4482	in libev, then check twice: If valgrind reports something like:
		4483
		4484	==2274== definitely lost: 0 bytes in 0 blocks.
		4485	==2274== possibly lost: 0 bytes in 0 blocks.
		4486	==2274== still reachable: 256 bytes in 1 blocks.
		4487
		4488	Then there is no memory leak, just as memory accounted to global variables
		4489	is not a memleak - the memory is still being referenced, and didn't leak.
		4490
		4491	Similarly, under some circumstances, valgrind might report kernel bugs
		4492	as if it were a bug in libev (e.g. in realloc or in the poll backend,
		4493	although an acceptable workaround has been found here), or it might be
		4494	confused.
		4495
		4496	Keep in mind that valgrind is a very good tool, but only a tool. Don't
		4497	make it into some kind of religion.
		4498
		4499	If you are unsure about something, feel free to contact the mailing list
		4500	with the full valgrind report and an explanation on why you think this
		4501	is a bug in libev (best check the archives, too :). However, don't be
		4502	annoyed when you get a brisk "this is no bug" answer and take the chance
		4503	of learning how to interpret valgrind properly.
		4504
		4505	If you need, for some reason, empty reports from valgrind for your project
		4506	I suggest using suppression lists.
		4507
		4508
		4509	=head1 PORTABILITY NOTES
		4510
		4511	=head2 GNU/LINUX 32 BIT LIMITATIONS
		4512
		4513	GNU/Linux is the only common platform that supports 64 bit file/large file
		4514	interfaces but I<disables> them by default.
		4515
		4516	That means that libev compiled in the default environment doesn't support
		4517	files larger than 2GiB or so, which mainly affects C<ev_stat> watchers.
		4518
		4519	Unfortunately, many programs try to work around this GNU/Linux issue
		4520	by enabling the large file API, which makes them incompatible with the
		4521	standard libev compiled for their system.
		4522
		4523	Likewise, libev cannot enable the large file API itself as this would
		4524	suddenly make it incompatible to the default compile time environment,
		4525	i.e. all programs not using special compile switches.
		4526
		4527	=head2 OS/X AND DARWIN BUGS
		4528
		4529	The whole thing is a bug if you ask me - basically any system interface
		4530	you touch is broken, whether it is locales, poll, kqueue or even the
		4531	OpenGL drivers.
		4532
		4533	=head3 C<kqueue> is buggy
		4534
		4535	The kqueue syscall is broken in all known versions - most versions support
		4536	only sockets, many support pipes.
		4537
		4538	Libev tries to work around this by not using C<kqueue> by default on this
		4539	rotten platform, but of course you can still ask for it when creating a
		4540	loop - embedding a socket-only kqueue loop into a select-based one is
		4541	probably going to work well.
		4542
		4543	=head3 C<poll> is buggy
		4544
		4545	Instead of fixing C<kqueue>, Apple replaced their (working) C<poll>
		4546	implementation by something calling C<kqueue> internally around the 10.5.6
		4547	release, so now C<kqueue> I<and> C<poll> are broken.
		4548
		4549	Libev tries to work around this by not using C<poll> by default on
		4550	this rotten platform, but of course you can still ask for it when creating
		4551	a loop.
		4552
		4553	=head3 C<select> is buggy
		4554
		4555	All that's left is C<select>, and of course Apple found a way to fuck this
		4556	one up as well: On OS/X, C<select> actively limits the number of file
		4557	descriptors you can pass in to 1024 - your program suddenly crashes when
		4558	you use more.
		4559
		4560	There is an undocumented "workaround" for this - defining
		4561	C<_DARWIN_UNLIMITED_SELECT>, which libev tries to use, so select I<should>
		4562	work on OS/X.
		4563
		4564	=head2 SOLARIS PROBLEMS AND WORKAROUNDS
		4565
		4566	=head3 C<errno> reentrancy
		4567
		4568	The default compile environment on Solaris is unfortunately so
		4569	thread-unsafe that you can't even use components/libraries compiled
		4570	without C<-D_REENTRANT> in a threaded program, which, of course, isn't
		4571	defined by default. A valid, if stupid, implementation choice.
		4572
		4573	If you want to use libev in threaded environments you have to make sure
		4574	it's compiled with C<_REENTRANT> defined.
		4575
		4576	=head3 Event port backend
		4577
		4578	The scalable event interface for Solaris is called "event
		4579	ports". Unfortunately, this mechanism is very buggy in all major
		4580	releases. If you run into high CPU usage, your program freezes or you get
		4581	a large number of spurious wakeups, make sure you have all the relevant
		4582	and latest kernel patches applied. No, I don't know which ones, but there
		4583	are multiple ones to apply, and afterwards, event ports actually work
		4584	great.
		4585
		4586	If you can't get it to work, you can try running the program by setting
		4587	the environment variable C<LIBEV_FLAGS=3> to only allow C<poll> and
		4588	C<select> backends.
		4589
		4590	=head2 AIX POLL BUG
		4591
		4592	AIX unfortunately has a broken C<poll.h> header. Libev works around
		4593	this by trying to avoid the poll backend altogether (i.e. it's not even
		4594	compiled in), which normally isn't a big problem as C<select> works fine
		4595	with large bitsets on AIX, and AIX is dead anyway.
		4596
		4597	=head2 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
		4598
		4599	=head3 General issues
		4600
		4601	Win32 doesn't support any of the standards (e.g. POSIX) that libev
		4602	requires, and its I/O model is fundamentally incompatible with the POSIX
		4603	model. Libev still offers limited functionality on this platform in
		4604	the form of the C<EVBACKEND_SELECT> backend, and only supports socket
		4605	descriptors. This only applies when using Win32 natively, not when using
		4606	e.g. cygwin. Actually, it only applies to the microsofts own compilers,
		4607	as every compielr comes with a slightly differently broken/incompatible
		4608	environment.
		4609
		4610	Lifting these limitations would basically require the full
		4611	re-implementation of the I/O system. If you are into this kind of thing,
		4612	then note that glib does exactly that for you in a very portable way (note
		4613	also that glib is the slowest event library known to man).
		4614
		4615	There is no supported compilation method available on windows except
		4616	embedding it into other applications.
		4617
		4618	Sensible signal handling is officially unsupported by Microsoft - libev
		4619	tries its best, but under most conditions, signals will simply not work.
		4620
		4621	Not a libev limitation but worth mentioning: windows apparently doesn't
		4622	accept large writes: instead of resulting in a partial write, windows will
		4623	either accept everything or return C<ENOBUFS> if the buffer is too large,
		4624	so make sure you only write small amounts into your sockets (less than a
		4625	megabyte seems safe, but this apparently depends on the amount of memory
		4626	available).
		4627
		4628	Due to the many, low, and arbitrary limits on the win32 platform and
		4629	the abysmal performance of winsockets, using a large number of sockets
		4630	is not recommended (and not reasonable). If your program needs to use
		4631	more than a hundred or so sockets, then likely it needs to use a totally
		4632	different implementation for windows, as libev offers the POSIX readiness
		4633	notification model, which cannot be implemented efficiently on windows
		4634	(due to Microsoft monopoly games).
		4635
		4636	A typical way to use libev under windows is to embed it (see the embedding
		4637	section for details) and use the following F<evwrap.h> header file instead
		4638	of F<ev.h>:
		4639
		4640	#define EV_STANDALONE /* keeps ev from requiring config.h */
		4641	#define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
		4642
		4643	#include "ev.h"
		4644
		4645	And compile the following F<evwrap.c> file into your project (make sure
		4646	you do I<not> compile the F<ev.c> or any other embedded source files!):
		4647
		4648	#include "evwrap.h"
		4649	#include "ev.c"
		4650
		4651	=head3 The winsocket C<select> function
		4652
		4653	The winsocket C<select> function doesn't follow POSIX in that it
		4654	requires socket I<handles> and not socket I<file descriptors> (it is
		4655	also extremely buggy). This makes select very inefficient, and also
		4656	requires a mapping from file descriptors to socket handles (the Microsoft
		4657	C runtime provides the function C<_open_osfhandle> for this). See the
		4658	discussion of the C<EV_SELECT_USE_FD_SET>, C<EV_SELECT_IS_WINSOCKET> and
		4659	C<EV_FD_TO_WIN32_HANDLE> preprocessor symbols for more info.
		4660
		4661	The configuration for a "naked" win32 using the Microsoft runtime
		4662	libraries and raw winsocket select is:
		4663
		4664	#define EV_USE_SELECT 1
		4665	#define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
		4666
		4667	Note that winsockets handling of fd sets is O(n), so you can easily get a
		4668	complexity in the O(n²) range when using win32.
		4669
		4670	=head3 Limited number of file descriptors
		4671
		4672	Windows has numerous arbitrary (and low) limits on things.
		4673
		4674	Early versions of winsocket's select only supported waiting for a maximum
		4675	of C<64> handles (probably owning to the fact that all windows kernels
		4676	can only wait for C<64> things at the same time internally; Microsoft
		4677	recommends spawning a chain of threads and wait for 63 handles and the
		4678	previous thread in each. Sounds great!).
		4679
		4680	Newer versions support more handles, but you need to define C<FD_SETSIZE>
		4681	to some high number (e.g. C<2048>) before compiling the winsocket select
		4682	call (which might be in libev or elsewhere, for example, perl and many
		4683	other interpreters do their own select emulation on windows).
		4684
		4685	Another limit is the number of file descriptors in the Microsoft runtime
		4686	libraries, which by default is C<64> (there must be a hidden I<64>
		4687	fetish or something like this inside Microsoft). You can increase this
		4688	by calling C<_setmaxstdio>, which can increase this limit to C<2048>
		4689	(another arbitrary limit), but is broken in many versions of the Microsoft
		4690	runtime libraries. This might get you to about C<512> or C<2048> sockets
		4691	(depending on windows version and/or the phase of the moon). To get more,
		4692	you need to wrap all I/O functions and provide your own fd management, but
		4693	the cost of calling select (O(n²)) will likely make this unworkable.
		4694
		4695	=head2 PORTABILITY REQUIREMENTS
		4696
		4697	In addition to a working ISO-C implementation and of course the
		4698	backend-specific APIs, libev relies on a few additional extensions:
		4699
		4700	=over 4
		4701
		4702	=item C<void (*)(ev_watcher_type *, int revents)> must have compatible
		4703	calling conventions regardless of C<ev_watcher_type *>.
		4704
		4705	Libev assumes not only that all watcher pointers have the same internal
		4706	structure (guaranteed by POSIX but not by ISO C for example), but it also
		4707	assumes that the same (machine) code can be used to call any watcher
		4708	callback: The watcher callbacks have different type signatures, but libev
		4709	calls them using an C<ev_watcher *> internally.
		4710
		4711	=item C<sig_atomic_t volatile> must be thread-atomic as well
		4712
		4713	The type C<sig_atomic_t volatile> (or whatever is defined as
		4714	C<EV_ATOMIC_T>) must be atomic with respect to accesses from different
		4715	threads. This is not part of the specification for C<sig_atomic_t>, but is
		4716	believed to be sufficiently portable.
		4717
		4718	=item C<sigprocmask> must work in a threaded environment
		4719
		4720	Libev uses C<sigprocmask> to temporarily block signals. This is not
		4721	allowed in a threaded program (C<pthread_sigmask> has to be used). Typical
		4722	pthread implementations will either allow C<sigprocmask> in the "main
		4723	thread" or will block signals process-wide, both behaviours would
		4724	be compatible with libev. Interaction between C<sigprocmask> and
		4725	C<pthread_sigmask> could complicate things, however.
		4726
		4727	The most portable way to handle signals is to block signals in all threads
		4728	except the initial one, and run the default loop in the initial thread as
		4729	well.
		4730
		4731	=item C<long> must be large enough for common memory allocation sizes
		4732
		4733	To improve portability and simplify its API, libev uses C<long> internally
		4734	instead of C<size_t> when allocating its data structures. On non-POSIX
		4735	systems (Microsoft...) this might be unexpectedly low, but is still at
		4736	least 31 bits everywhere, which is enough for hundreds of millions of
		4737	watchers.
		4738
		4739	=item C<double> must hold a time value in seconds with enough accuracy
		4740
		4741	The type C<double> is used to represent timestamps. It is required to
		4742	have at least 51 bits of mantissa (and 9 bits of exponent), which is
		4743	good enough for at least into the year 4000 with millisecond accuracy
		4744	(the design goal for libev). This requirement is overfulfilled by
		4745	implementations using IEEE 754, which is basically all existing ones. With
		4746	IEEE 754 doubles, you get microsecond accuracy until at least 2200.
		4747
		4748	=back
		4749
		4750	If you know of other additional requirements drop me a note.
		4751
		4752
		4753	=head1 ALGORITHMIC COMPLEXITIES
		4754
		4755	In this section the complexities of (many of) the algorithms used inside
		4756	libev will be documented. For complexity discussions about backends see
		4757	the documentation for C<ev_default_init>.
		4758
		4759	All of the following are about amortised time: If an array needs to be
		4760	extended, libev needs to realloc and move the whole array, but this
		4761	happens asymptotically rarer with higher number of elements, so O(1) might
		4762	mean that libev does a lengthy realloc operation in rare cases, but on
		4763	average it is much faster and asymptotically approaches constant time.
		4764
		4765	=over 4
		4766
		4767	=item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
		4768
		4769	This means that, when you have a watcher that triggers in one hour and
		4770	there are 100 watchers that would trigger before that, then inserting will
		4771	have to skip roughly seven (C<ld 100>) of these watchers.
		4772
		4773	=item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
		4774
		4775	That means that changing a timer costs less than removing/adding them,
		4776	as only the relative motion in the event queue has to be paid for.
		4777
		4778	=item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
		4779
		4780	These just add the watcher into an array or at the head of a list.
		4781
		4782	=item Stopping check/prepare/idle/fork/async watchers: O(1)
		4783
		4784	=item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
		4785
		4786	These watchers are stored in lists, so they need to be walked to find the
		4787	correct watcher to remove. The lists are usually short (you don't usually
		4788	have many watchers waiting for the same fd or signal: one is typical, two
		4789	is rare).
		4790
		4791	=item Finding the next timer in each loop iteration: O(1)
		4792
		4793	By virtue of using a binary or 4-heap, the next timer is always found at a
		4794	fixed position in the storage array.
		4795
		4796	=item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
		4797
		4798	A change means an I/O watcher gets started or stopped, which requires
		4799	libev to recalculate its status (and possibly tell the kernel, depending
		4800	on backend and whether C<ev_io_set> was used).
		4801
		4802	=item Activating one watcher (putting it into the pending state): O(1)
		4803
		4804	=item Priority handling: O(number_of_priorities)
		4805
		4806	Priorities are implemented by allocating some space for each
		4807	priority. When doing priority-based operations, libev usually has to
		4808	linearly search all the priorities, but starting/stopping and activating
		4809	watchers becomes O(1) with respect to priority handling.
		4810
		4811	=item Sending an ev_async: O(1)
		4812
		4813	=item Processing ev_async_send: O(number_of_async_watchers)
		4814
		4815	=item Processing signals: O(max_signal_number)
		4816
		4817	Sending involves a system call I<iff> there were no other C<ev_async_send>
		4818	calls in the current loop iteration. Checking for async and signal events
		4819	involves iterating over all running async watchers or all signal numbers.
		4820
		4821	=back
		4822
		4823
		4824	=head1 PORTING FROM LIBEV 3.X TO 4.X
		4825
		4826	The major version 4 introduced some minor incompatible changes to the API.
		4827
		4828	At the moment, the C<ev.h> header file tries to implement superficial
		4829	compatibility, so most programs should still compile. Those might be
		4830	removed in later versions of libev, so better update early than late.
		4831
		4832	=over 4
		4833
		4834	=item function/symbol renames
		4835
		4836	A number of functions and symbols have been renamed:
		4837
		4838	ev_loop => ev_run
		4839	EVLOOP_NONBLOCK => EVRUN_NOWAIT
		4840	EVLOOP_ONESHOT => EVRUN_ONCE
		4841
		4842	ev_unloop => ev_break
		4843	EVUNLOOP_CANCEL => EVBREAK_CANCEL
		4844	EVUNLOOP_ONE => EVBREAK_ONE
		4845	EVUNLOOP_ALL => EVBREAK_ALL
		4846
		4847	EV_TIMEOUT => EV_TIMER
		4848
		4849	ev_loop_count => ev_iteration
		4850	ev_loop_depth => ev_depth
		4851	ev_loop_verify => ev_verify
		4852
		4853	Most functions working on C<struct ev_loop> objects don't have an
		4854	C<ev_loop_> prefix, so it was removed; C<ev_loop>, C<ev_unloop> and
		4855	associated constants have been renamed to not collide with the C<struct
		4856	ev_loop> anymore and C<EV_TIMER> now follows the same naming scheme
		4857	as all other watcher types. Note that C<ev_loop_fork> is still called
		4858	C<ev_loop_fork> because it would otherwise clash with the C<ev_fork>
		4859	typedef.
		4860
		4861	=item C<EV_COMPAT3> backwards compatibility mechanism
		4862
		4863	The backward compatibility mechanism can be controlled by
		4864	C<EV_COMPAT3>. See L<PREPROCESSOR SYMBOLS/MACROS> in the L<EMBEDDING>
		4865	section.
		4866
		4867	=item C<EV_MINIMAL> mechanism replaced by C<EV_FEATURES>
		4868
		4869	The preprocessor symbol C<EV_MINIMAL> has been replaced by a different
		4870	mechanism, C<EV_FEATURES>. Programs using C<EV_MINIMAL> usually compile
		4871	and work, but the library code will of course be larger.
		4872
		4873	=back
		4874
		4875
		4876	=head1 GLOSSARY
		4877
		4878	=over 4
		4879
		4880	=item active
		4881
		4882	A watcher is active as long as it has been started and not yet stopped.
		4883	See L<WATCHER STATES> for details.
		4884
		4885	=item application
		4886
		4887	In this document, an application is whatever is using libev.
		4888
		4889	=item backend
		4890
		4891	The part of the code dealing with the operating system interfaces.
		4892
		4893	=item callback
		4894
		4895	The address of a function that is called when some event has been
		4896	detected. Callbacks are being passed the event loop, the watcher that
		4897	received the event, and the actual event bitset.
		4898
		4899	=item callback/watcher invocation
		4900
		4901	The act of calling the callback associated with a watcher.
		4902
		4903	=item event
		4904
		4905	A change of state of some external event, such as data now being available
		4906	for reading on a file descriptor, time having passed or simply not having
		4907	any other events happening anymore.
		4908
		4909	In libev, events are represented as single bits (such as C<EV_READ> or
		4910	C<EV_TIMER>).
		4911
		4912	=item event library
		4913
		4914	A software package implementing an event model and loop.
		4915
		4916	=item event loop
		4917
		4918	An entity that handles and processes external events and converts them
		4919	into callback invocations.
		4920
		4921	=item event model
		4922
		4923	The model used to describe how an event loop handles and processes
		4924	watchers and events.
		4925
		4926	=item pending
		4927
		4928	A watcher is pending as soon as the corresponding event has been
		4929	detected. See L<WATCHER STATES> for details.
		4930
		4931	=item real time
		4932
		4933	The physical time that is observed. It is apparently strictly monotonic :)
		4934
		4935	=item wall-clock time
		4936
		4937	The time and date as shown on clocks. Unlike real time, it can actually
		4938	be wrong and jump forwards and backwards, e.g. when the you adjust your
		4939	clock.
		4940
		4941	=item watcher
		4942
		4943	A data structure that describes interest in certain events. Watchers need
		4944	to be started (attached to an event loop) before they can receive events.
		4945
		4946	=back
878		4947
879	=head1 AUTHOR	4948	=head1 AUTHOR
880		4949
881	Marc Lehmann <libev@schmorp.de>.	4950	Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael Magnusson.
882		4951

Diff Legend

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