ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libev/ev.pod

Comparing libev/ev.pod (file contents):
Revision 1.35 by root, Fri Nov 23 19:35:09 2007 UTC vs.
Revision 1.285 by root, Tue Mar 16 00:20:17 2010 UTC

…		…
2		2
3	libev - a high performance full-featured event loop written in C	3	libev - a high performance full-featured event loop written in C
4		4
5	=head1 SYNOPSIS	5	=head1 SYNOPSIS
6		6
7	#include <ev.h>	7	#include <ev.h>
8		8
9	=head1 DESCRIPTION	9	=head2 EXAMPLE PROGRAM
		10
		11	// a single header file is required
		12	#include <ev.h>
		13
		14	#include <stdio.h> // for puts
		15
		16	// every watcher type has its own typedef'd struct
		17	// with the name ev_TYPE
		18	ev_io stdin_watcher;
		19	ev_timer timeout_watcher;
		20
		21	// all watcher callbacks have a similar signature
		22	// this callback is called when data is readable on stdin
		23	static void
		24	stdin_cb (EV_P_ ev_io *w, int revents)
		25	{
		26	puts ("stdin ready");
		27	// for one-shot events, one must manually stop the watcher
		28	// with its corresponding stop function.
		29	ev_io_stop (EV_A_ w);
		30
		31	// this causes all nested ev_loop's to stop iterating
		32	ev_unloop (EV_A_ EVUNLOOP_ALL);
		33	}
		34
		35	// another callback, this time for a time-out
		36	static void
		37	timeout_cb (EV_P_ ev_timer *w, int revents)
		38	{
		39	puts ("timeout");
		40	// this causes the innermost ev_loop to stop iterating
		41	ev_unloop (EV_A_ EVUNLOOP_ONE);
		42	}
		43
		44	int
		45	main (void)
		46	{
		47	// use the default event loop unless you have special needs
		48	struct ev_loop *loop = ev_default_loop (0);
		49
		50	// initialise an io watcher, then start it
		51	// this one will watch for stdin to become readable
		52	ev_io_init (&stdin_watcher, stdin_cb, /*STDIN_FILENO*/ 0, EV_READ);
		53	ev_io_start (loop, &stdin_watcher);
		54
		55	// initialise a timer watcher, then start it
		56	// simple non-repeating 5.5 second timeout
		57	ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
		58	ev_timer_start (loop, &timeout_watcher);
		59
		60	// now wait for events to arrive
		61	ev_loop (loop, 0);
		62
		63	// unloop was called, so exit
		64	return 0;
		65	}
		66
		67	=head1 ABOUT THIS DOCUMENT
		68
		69	This document documents the libev software package.
		70
		71	The newest version of this document is also available as an html-formatted
		72	web page you might find easier to navigate when reading it for the first
		73	time: L<http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod>.
		74
		75	While this document tries to be as complete as possible in documenting
		76	libev, its usage and the rationale behind its design, it is not a tutorial
		77	on event-based programming, nor will it introduce event-based programming
		78	with libev.
		79
		80	Familarity with event based programming techniques in general is assumed
		81	throughout this document.
		82
		83	=head1 ABOUT LIBEV
10		84
11	Libev is an event loop: you register interest in certain events (such as a	85	Libev is an event loop: you register interest in certain events (such as a
12	file descriptor being readable or a timeout occuring), and it will manage	86	file descriptor being readable or a timeout occurring), and it will manage
13	these event sources and provide your program with events.	87	these event sources and provide your program with events.
14		88
15	To do this, it must take more or less complete control over your process	89	To do this, it must take more or less complete control over your process
16	(or thread) by executing the I<event loop> handler, and will then	90	(or thread) by executing the I<event loop> handler, and will then
17	communicate events via a callback mechanism.	91	communicate events via a callback mechanism.
…		…
19	You register interest in certain events by registering so-called I<event	93	You register interest in certain events by registering so-called I<event
20	watchers>, which are relatively small C structures you initialise with the	94	watchers>, which are relatively small C structures you initialise with the
21	details of the event, and then hand it over to libev by I<starting> the	95	details of the event, and then hand it over to libev by I<starting> the
22	watcher.	96	watcher.
23		97
24	=head1 FEATURES	98	=head2 FEATURES
25		99
26	Libev supports select, poll, the linux-specific epoll and the bsd-specific	100	Libev supports C<select>, C<poll>, the Linux-specific C<epoll>, the
27	kqueue mechanisms for file descriptor events, relative timers, absolute	101	BSD-specific C<kqueue> and the Solaris-specific event port mechanisms
28	timers with customised rescheduling, signal events, process status change	102	for file descriptor events (C<ev_io>), the Linux C<inotify> interface
		103	(for C<ev_stat>), Linux eventfd/signalfd (for faster and cleaner
		104	inter-thread wakeup (C<ev_async>)/signal handling (C<ev_signal>)) relative
		105	timers (C<ev_timer>), absolute timers with customised rescheduling
		106	(C<ev_periodic>), synchronous signals (C<ev_signal>), process status
29	events (related to SIGCHLD), and event watchers dealing with the event	107	change events (C<ev_child>), and event watchers dealing with the event
30	loop mechanism itself (idle, prepare and check watchers). It also is quite	108	loop mechanism itself (C<ev_idle>, C<ev_embed>, C<ev_prepare> and
		109	C<ev_check> watchers) as well as file watchers (C<ev_stat>) and even
		110	limited support for fork events (C<ev_fork>).
		111
		112	It also is quite fast (see this
31	fast (see this L<benchmark|http://libev.schmorp.de/bench.html> comparing	113	L<benchmark|http://libev.schmorp.de/bench.html> comparing it to libevent
32	it to libevent for example).	114	for example).
33		115
34	=head1 CONVENTIONS	116	=head2 CONVENTIONS
35		117
36	Libev is very configurable. In this manual the default configuration	118	Libev is very configurable. In this manual the default (and most common)
37	will be described, which supports multiple event loops. For more info	119	configuration will be described, which supports multiple event loops. For
38	about various configuration options please have a look at the file	120	more info about various configuration options please have a look at
39	F<README.embed> in the libev distribution. If libev was configured without	121	B<EMBED> section in this manual. If libev was configured without support
40	support for multiple event loops, then all functions taking an initial	122	for multiple event loops, then all functions taking an initial argument of
41	argument of name C<loop> (which is always of type C<struct ev_loop *>)	123	name C<loop> (which is always of type C<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 (somewhere
48	the beginning of 1970, details are complicated, don't ask). This type is	130	near the beginning of 1970, details are complicated, don't ask). This
49	called C<ev_tstamp>, which is what you should use too. It usually aliases	131	type is called C<ev_tstamp>, which is what you should use too. It usually
50	to the C<double> type in C, and when you need to do any calculations on	132	aliases to the C<double> type in C. When you need to do any calculations
51	it, you should treat it as such.	133	on it, you should treat it as some floating point value. Unlike the name
		134	component C<stamp> might indicate, it is also used for time differences
		135	throughout libev.
		136
		137	=head1 ERROR HANDLING
		138
		139	Libev knows three classes of errors: operating system errors, usage errors
		140	and internal errors (bugs).
		141
		142	When libev catches an operating system error it cannot handle (for example
		143	a system call indicating a condition libev cannot fix), it calls the callback
		144	set via C<ev_set_syserr_cb>, which is supposed to fix the problem or
		145	abort. The default is to print a diagnostic message and to call C<abort
		146	()>.
		147
		148	When libev detects a usage error such as a negative timer interval, then
		149	it will print a diagnostic message and abort (via the C<assert> mechanism,
		150	so C<NDEBUG> will disable this checking): these are programming errors in
		151	the libev caller and need to be fixed there.
		152
		153	Libev also has a few internal error-checking C<assert>ions, and also has
		154	extensive consistency checking code. These do not trigger under normal
		155	circumstances, as they indicate either a bug in libev or worse.
52		156
53		157
54	=head1 GLOBAL FUNCTIONS	158	=head1 GLOBAL FUNCTIONS
55		159
56	These functions can be called anytime, even before initialising the	160	These functions can be called anytime, even before initialising the
…		…
62		166
63	Returns the current time as libev would use it. Please note that the	167	Returns the current time as libev would use it. Please note that the
64	C<ev_now> function is usually faster and also often returns the timestamp	168	C<ev_now> function is usually faster and also often returns the timestamp
65	you actually want to know.	169	you actually want to know.
66		170
		171	=item ev_sleep (ev_tstamp interval)
		172
		173	Sleep for the given interval: The current thread will be blocked until
		174	either it is interrupted or the given time interval has passed. Basically
		175	this is a sub-second-resolution C<sleep ()>.
		176
67	=item int ev_version_major ()	177	=item int ev_version_major ()
68		178
69	=item int ev_version_minor ()	179	=item int ev_version_minor ()
70		180
71	You can find out the major and minor version numbers of the library	181	You can find out the major and minor ABI version numbers of the library
72	you linked against by calling the functions C<ev_version_major> and	182	you linked against by calling the functions C<ev_version_major> and
73	C<ev_version_minor>. If you want, you can compare against the global	183	C<ev_version_minor>. If you want, you can compare against the global
74	symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the	184	symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the
75	version of the library your program was compiled against.	185	version of the library your program was compiled against.
76		186
		187	These version numbers refer to the ABI version of the library, not the
		188	release version.
		189
77	Usually, it's a good idea to terminate if the major versions mismatch,	190	Usually, it's a good idea to terminate if the major versions mismatch,
78	as this indicates an incompatible change. Minor versions are usually	191	as this indicates an incompatible change. Minor versions are usually
79	compatible to older versions, so a larger minor version alone is usually	192	compatible to older versions, so a larger minor version alone is usually
80	not a problem.	193	not a problem.
81		194
82	Example: make sure we haven't accidentally been linked against the wrong	195	Example: Make sure we haven't accidentally been linked against the wrong
83	version:	196	version.
84		197
85	assert (("libev version mismatch",	198	assert (("libev version mismatch",
86	ev_version_major () == EV_VERSION_MAJOR	199	ev_version_major () == EV_VERSION_MAJOR
87	&& ev_version_minor () >= EV_VERSION_MINOR));	200	&& ev_version_minor () >= EV_VERSION_MINOR));
88		201
89	=item unsigned int ev_supported_backends ()	202	=item unsigned int ev_supported_backends ()
90		203
91	Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>	204	Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
92	value) compiled into this binary of libev (independent of their	205	value) compiled into this binary of libev (independent of their
…		…
94	a description of the set values.	207	a description of the set values.
95		208
96	Example: make sure we have the epoll method, because yeah this is cool and	209	Example: make sure we have the epoll method, because yeah this is cool and
97	a must have and can we have a torrent of it please!!!11	210	a must have and can we have a torrent of it please!!!11
98		211
99	assert (("sorry, no epoll, no sex",	212	assert (("sorry, no epoll, no sex",
100	ev_supported_backends () & EVBACKEND_EPOLL));	213	ev_supported_backends () & EVBACKEND_EPOLL));
101		214
102	=item unsigned int ev_recommended_backends ()	215	=item unsigned int ev_recommended_backends ()
103		216
104	Return the set of all backends compiled into this binary of libev and also	217	Return the set of all backends compiled into this binary of libev and also
105	recommended for this platform. This set is often smaller than the one	218	recommended for this platform. This set is often smaller than the one
106	returned by C<ev_supported_backends>, as for example kqueue is broken on	219	returned by C<ev_supported_backends>, as for example kqueue is broken on
107	most BSDs and will not be autodetected unless you explicitly request it	220	most BSDs and will not be auto-detected unless you explicitly request it
108	(assuming you know what you are doing). This is the set of backends that	221	(assuming you know what you are doing). This is the set of backends that
109	libev will probe for if you specify no backends explicitly.	222	libev will probe for if you specify no backends explicitly.
110		223
111	=item unsigned int ev_embeddable_backends ()	224	=item unsigned int ev_embeddable_backends ()
112		225
…		…
116	C<ev_embeddable_backends () & ev_supported_backends ()>, likewise for	229	C<ev_embeddable_backends () & ev_supported_backends ()>, likewise for
117	recommended ones.	230	recommended ones.
118		231
119	See the description of C<ev_embed> watchers for more info.	232	See the description of C<ev_embed> watchers for more info.
120		233
121	=item ev_set_allocator (void *(*cb)(void *ptr, long size))	234	=item ev_set_allocator (void *(*cb)(void *ptr, long size)) [NOT REENTRANT]
122		235
123	Sets the allocation function to use (the prototype is similar to the	236	Sets the allocation function to use (the prototype is similar - the
124	realloc C function, the semantics are identical). It is used to allocate	237	semantics are identical to the C<realloc> C89/SuS/POSIX function). It is
125	and free memory (no surprises here). If it returns zero when memory	238	used to allocate and free memory (no surprises here). If it returns zero
126	needs to be allocated, the library might abort or take some potentially	239	when memory needs to be allocated (C<size != 0>), the library might abort
127	destructive action. The default is your system realloc function.	240	or take some potentially destructive action.
		241
		242	Since some systems (at least OpenBSD and Darwin) fail to implement
		243	correct C<realloc> semantics, libev will use a wrapper around the system
		244	C<realloc> and C<free> functions by default.
128		245
129	You could override this function in high-availability programs to, say,	246	You could override this function in high-availability programs to, say,
130	free some memory if it cannot allocate memory, to use a special allocator,	247	free some memory if it cannot allocate memory, to use a special allocator,
131	or even to sleep a while and retry until some memory is available.	248	or even to sleep a while and retry until some memory is available.
132		249
133	Example: replace the libev allocator with one that waits a bit and then	250	Example: Replace the libev allocator with one that waits a bit and then
134	retries: better than mine).	251	retries (example requires a standards-compliant C<realloc>).
135		252
136	static void *	253	static void *
137	persistent_realloc (void *ptr, long size)	254	persistent_realloc (void *ptr, size_t size)
138	{	255	{
139	for (;;)	256	for (;;)
140	{	257	{
141	void *newptr = realloc (ptr, size);	258	void *newptr = realloc (ptr, size);
142		259
…		…
148	}	265	}
149		266
150	...	267	...
151	ev_set_allocator (persistent_realloc);	268	ev_set_allocator (persistent_realloc);
152		269
153	=item ev_set_syserr_cb (void (*cb)(const char *msg));	270	=item ev_set_syserr_cb (void (*cb)(const char *msg)); [NOT REENTRANT]
154		271
155	Set the callback function to call on a retryable syscall error (such	272	Set the callback function to call on a retryable system call error (such
156	as failed select, poll, epoll_wait). The message is a printable string	273	as failed select, poll, epoll_wait). The message is a printable string
157	indicating the system call or subsystem causing the problem. If this	274	indicating the system call or subsystem causing the problem. If this
158	callback is set, then libev will expect it to remedy the sitution, no	275	callback is set, then libev will expect it to remedy the situation, no
159	matter what, when it returns. That is, libev will generally retry the	276	matter what, when it returns. That is, libev will generally retry the
160	requested operation, or, if the condition doesn't go away, do bad stuff	277	requested operation, or, if the condition doesn't go away, do bad stuff
161	(such as abort).	278	(such as abort).
162		279
163	Example: do the same thing as libev does internally:	280	Example: This is basically the same thing that libev does internally, too.
164		281
165	static void	282	static void
166	fatal_error (const char *msg)	283	fatal_error (const char *msg)
167	{	284	{
168	perror (msg);	285	perror (msg);
…		…
174		291
175	=back	292	=back
176		293
177	=head1 FUNCTIONS CONTROLLING THE EVENT LOOP	294	=head1 FUNCTIONS CONTROLLING THE EVENT LOOP
178		295
179	An event loop is described by a C<struct ev_loop *>. The library knows two	296	An event loop is described by a C<struct ev_loop *> (the C<struct>
180	types of such loops, the I<default> loop, which supports signals and child	297	is I<not> optional in this case, as there is also an C<ev_loop>
181	events, and dynamically created loops which do not.	298	I<function>).
182		299
183	If you use threads, a common model is to run the default event loop	300	The library knows two types of such loops, the I<default> loop, which
184	in your main thread (or in a separate thread) and for each thread you	301	supports signals and child events, and dynamically created loops which do
185	create, you also create another event loop. Libev itself does no locking	302	not.
186	whatsoever, so if you mix calls to the same event loop in different
187	threads, make sure you lock (this is usually a bad idea, though, even if
188	done correctly, because it's hideous and inefficient).
189		303
190	=over 4	304	=over 4
191		305
192	=item struct ev_loop *ev_default_loop (unsigned int flags)	306	=item struct ev_loop *ev_default_loop (unsigned int flags)
193		307
…		…
197	flags. If that is troubling you, check C<ev_backend ()> afterwards).	311	flags. If that is troubling you, check C<ev_backend ()> afterwards).
198		312
199	If you don't know what event loop to use, use the one returned from this	313	If you don't know what event loop to use, use the one returned from this
200	function.	314	function.
201		315
		316	Note that this function is I<not> thread-safe, so if you want to use it
		317	from multiple threads, you have to lock (note also that this is unlikely,
		318	as loops cannot be shared easily between threads anyway).
		319
		320	The default loop is the only loop that can handle C<ev_signal> and
		321	C<ev_child> watchers, and to do this, it always registers a handler
		322	for C<SIGCHLD>. If this is a problem for your application you can either
		323	create a dynamic loop with C<ev_loop_new> that doesn't do that, or you
		324	can simply overwrite the C<SIGCHLD> signal handler I<after> calling
		325	C<ev_default_init>.
		326
202	The flags argument can be used to specify special behaviour or specific	327	The flags argument can be used to specify special behaviour or specific
203	backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).	328	backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
204		329
205	The following flags are supported:	330	The following flags are supported:
206		331
…		…
211	The default flags value. Use this if you have no clue (it's the right	336	The default flags value. Use this if you have no clue (it's the right
212	thing, believe me).	337	thing, believe me).
213		338
214	=item C<EVFLAG_NOENV>	339	=item C<EVFLAG_NOENV>
215		340
216	If this flag bit is ored into the flag value (or the program runs setuid	341	If this flag bit is or'ed into the flag value (or the program runs setuid
217	or setgid) then libev will I<not> look at the environment variable	342	or setgid) then libev will I<not> look at the environment variable
218	C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will	343	C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
219	override the flags completely if it is found in the environment. This is	344	override the flags completely if it is found in the environment. This is
220	useful to try out specific backends to test their performance, or to work	345	useful to try out specific backends to test their performance, or to work
221	around bugs.	346	around bugs.
222		347
		348	=item C<EVFLAG_FORKCHECK>
		349
		350	Instead of calling C<ev_default_fork> or C<ev_loop_fork> manually after
		351	a fork, you can also make libev check for a fork in each iteration by
		352	enabling this flag.
		353
		354	This works by calling C<getpid ()> on every iteration of the loop,
		355	and thus this might slow down your event loop if you do a lot of loop
		356	iterations and little real work, but is usually not noticeable (on my
		357	GNU/Linux system for example, C<getpid> is actually a simple 5-insn sequence
		358	without a system call and thus I<very> fast, but my GNU/Linux system also has
		359	C<pthread_atfork> which is even faster).
		360
		361	The big advantage of this flag is that you can forget about fork (and
		362	forget about forgetting to tell libev about forking) when you use this
		363	flag.
		364
		365	This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
		366	environment variable.
		367
		368	=item C<EVFLAG_NOINOTIFY>
		369
		370	When this flag is specified, then libev will not attempt to use the
		371	I<inotify> API for it's C<ev_stat> watchers. Apart from debugging and
		372	testing, this flag can be useful to conserve inotify file descriptors, as
		373	otherwise each loop using C<ev_stat> watchers consumes one inotify handle.
		374
		375	=item C<EVFLAG_SIGNALFD>
		376
		377	When this flag is specified, then libev will attempt to use the
		378	I<signalfd> API for it's C<ev_signal> (and C<ev_child>) watchers. This API
		379	delivers signals synchronously, which makes it both faster and might make
		380	it possible to get the queued signal data. It can also simplify signal
		381	handling with threads, as long as you properly block signals in your
		382	threads that are not interested in handling them.
		383
		384	Signalfd will not be used by default as this changes your signal mask, and
		385	there are a lot of shoddy libraries and programs (glib's threadpool for
		386	example) that can't properly initialise their signal masks.
		387
223	=item C<EVBACKEND_SELECT> (value 1, portable select backend)	388	=item C<EVBACKEND_SELECT> (value 1, portable select backend)
224		389
225	This is your standard select(2) backend. Not I<completely> standard, as	390	This is your standard select(2) backend. Not I<completely> standard, as
226	libev tries to roll its own fd_set with no limits on the number of fds,	391	libev tries to roll its own fd_set with no limits on the number of fds,
227	but if that fails, expect a fairly low limit on the number of fds when	392	but if that fails, expect a fairly low limit on the number of fds when
228	using this backend. It doesn't scale too well (O(highest_fd)), but its usually	393	using this backend. It doesn't scale too well (O(highest_fd)), but its
229	the fastest backend for a low number of fds.	394	usually the fastest backend for a low number of (low-numbered :) fds.
		395
		396	To get good performance out of this backend you need a high amount of
		397	parallelism (most of the file descriptors should be busy). If you are
		398	writing a server, you should C<accept ()> in a loop to accept as many
		399	connections as possible during one iteration. You might also want to have
		400	a look at C<ev_set_io_collect_interval ()> to increase the amount of
		401	readiness notifications you get per iteration.
		402
		403	This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
		404	C<writefds> set (and to work around Microsoft Windows bugs, also onto the
		405	C<exceptfds> set on that platform).
230		406
231	=item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)	407	=item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
232		408
233	And this is your standard poll(2) backend. It's more complicated than	409	And this is your standard poll(2) backend. It's more complicated
234	select, but handles sparse fds better and has no artificial limit on the	410	than select, but handles sparse fds better and has no artificial
235	number of fds you can use (except it will slow down considerably with a	411	limit on the number of fds you can use (except it will slow down
236	lot of inactive fds). It scales similarly to select, i.e. O(total_fds).	412	considerably with a lot of inactive fds). It scales similarly to select,
		413	i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
		414	performance tips.
		415
		416	This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
		417	C<EV_WRITE> to C<POLLOUT | POLLERR | POLLHUP>.
237		418
238	=item C<EVBACKEND_EPOLL> (value 4, Linux)	419	=item C<EVBACKEND_EPOLL> (value 4, Linux)
239		420
		421	Use the linux-specific epoll(7) interface (for both pre- and post-2.6.9
		422	kernels).
		423
240	For few fds, this backend is a bit little slower than poll and select,	424	For few fds, this backend is a bit little slower than poll and select,
241	but it scales phenomenally better. While poll and select usually scale like	425	but it scales phenomenally better. While poll and select usually scale
242	O(total_fds) where n is the total number of fds (or the highest fd), epoll scales	426	like O(total_fds) where n is the total number of fds (or the highest fd),
243	either O(1) or O(active_fds).	427	epoll scales either O(1) or O(active_fds).
244		428
		429	The epoll mechanism deserves honorable mention as the most misdesigned
		430	of the more advanced event mechanisms: mere annoyances include silently
		431	dropping file descriptors, requiring a system call per change per file
		432	descriptor (and unnecessary guessing of parameters), problems with dup and
		433	so on. The biggest issue is fork races, however - if a program forks then
		434	I<both> parent and child process have to recreate the epoll set, which can
		435	take considerable time (one syscall per file descriptor) and is of course
		436	hard to detect.
		437
		438	Epoll is also notoriously buggy - embedding epoll fds I<should> work, but
		439	of course I<doesn't>, and epoll just loves to report events for totally
		440	I<different> file descriptors (even already closed ones, so one cannot
		441	even remove them from the set) than registered in the set (especially
		442	on SMP systems). Libev tries to counter these spurious notifications by
		443	employing an additional generation counter and comparing that against the
		444	events to filter out spurious ones, recreating the set when required.
		445
245	While stopping and starting an I/O watcher in the same iteration will	446	While stopping, setting and starting an I/O watcher in the same iteration
246	result in some caching, there is still a syscall per such incident	447	will result in some caching, there is still a system call per such
247	(because the fd could point to a different file description now), so its	448	incident (because the same I<file descriptor> could point to a different
248	best to avoid that. Also, dup()ed file descriptors might not work very	449	I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
249	well if you register events for both fds.	450	file descriptors might not work very well if you register events for both
		451	file descriptors.
250		452
251	Please note that epoll sometimes generates spurious notifications, so you	453	Best performance from this backend is achieved by not unregistering all
252	need to use non-blocking I/O or other means to avoid blocking when no data	454	watchers for a file descriptor until it has been closed, if possible,
253	(or space) is available.	455	i.e. keep at least one watcher active per fd at all times. Stopping and
		456	starting a watcher (without re-setting it) also usually doesn't cause
		457	extra overhead. A fork can both result in spurious notifications as well
		458	as in libev having to destroy and recreate the epoll object, which can
		459	take considerable time and thus should be avoided.
		460
		461	All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
		462	faster than epoll for maybe up to a hundred file descriptors, depending on
		463	the usage. So sad.
		464
		465	While nominally embeddable in other event loops, this feature is broken in
		466	all kernel versions tested so far.
		467
		468	This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
		469	C<EVBACKEND_POLL>.
254		470
255	=item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)	471	=item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
256		472
257	Kqueue deserves special mention, as at the time of this writing, it	473	Kqueue deserves special mention, as at the time of this writing, it
258	was broken on all BSDs except NetBSD (usually it doesn't work with	474	was broken on all BSDs except NetBSD (usually it doesn't work reliably
259	anything but sockets and pipes, except on Darwin, where of course its	475	with anything but sockets and pipes, except on Darwin, where of course
260	completely useless). For this reason its not being "autodetected"	476	it's completely useless). Unlike epoll, however, whose brokenness
		477	is by design, these kqueue bugs can (and eventually will) be fixed
		478	without API changes to existing programs. For this reason it's not being
261	unless you explicitly specify it explicitly in the flags (i.e. using	479	"auto-detected" unless you explicitly specify it in the flags (i.e. using
262	C<EVBACKEND_KQUEUE>).	480	C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
		481	system like NetBSD.
		482
		483	You still can embed kqueue into a normal poll or select backend and use it
		484	only for sockets (after having made sure that sockets work with kqueue on
		485	the target platform). See C<ev_embed> watchers for more info.
263		486
264	It scales in the same way as the epoll backend, but the interface to the	487	It scales in the same way as the epoll backend, but the interface to the
265	kernel is more efficient (which says nothing about its actual speed, of	488	kernel is more efficient (which says nothing about its actual speed, of
266	course). While starting and stopping an I/O watcher does not cause an	489	course). While stopping, setting and starting an I/O watcher does never
267	extra syscall as with epoll, it still adds up to four event changes per	490	cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
268	incident, so its best to avoid that.	491	two event changes per incident. Support for C<fork ()> is very bad (but
		492	sane, unlike epoll) and it drops fds silently in similarly hard-to-detect
		493	cases
		494
		495	This backend usually performs well under most conditions.
		496
		497	While nominally embeddable in other event loops, this doesn't work
		498	everywhere, so you might need to test for this. And since it is broken
		499	almost everywhere, you should only use it when you have a lot of sockets
		500	(for which it usually works), by embedding it into another event loop
		501	(e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
		502	also broken on OS X)) and, did I mention it, using it only for sockets.
		503
		504	This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
		505	C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
		506	C<NOTE_EOF>.
269		507
270	=item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)	508	=item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
271		509
272	This is not implemented yet (and might never be).	510	This is not implemented yet (and might never be, unless you send me an
		511	implementation). According to reports, C</dev/poll> only supports sockets
		512	and is not embeddable, which would limit the usefulness of this backend
		513	immensely.
273		514
274	=item C<EVBACKEND_PORT> (value 32, Solaris 10)	515	=item C<EVBACKEND_PORT> (value 32, Solaris 10)
275		516
276	This uses the Solaris 10 port mechanism. As with everything on Solaris,	517	This uses the Solaris 10 event port mechanism. As with everything on Solaris,
277	it's really slow, but it still scales very well (O(active_fds)).	518	it's really slow, but it still scales very well (O(active_fds)).
278		519
279	Please note that solaris ports can result in a lot of spurious	520	Please note that Solaris event ports can deliver a lot of spurious
280	notifications, so you need to use non-blocking I/O or other means to avoid	521	notifications, so you need to use non-blocking I/O or other means to avoid
281	blocking when no data (or space) is available.	522	blocking when no data (or space) is available.
		523
		524	While this backend scales well, it requires one system call per active
		525	file descriptor per loop iteration. For small and medium numbers of file
		526	descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
		527	might perform better.
		528
		529	On the positive side, with the exception of the spurious readiness
		530	notifications, this backend actually performed fully to specification
		531	in all tests and is fully embeddable, which is a rare feat among the
		532	OS-specific backends (I vastly prefer correctness over speed hacks).
		533
		534	This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
		535	C<EVBACKEND_POLL>.
282		536
283	=item C<EVBACKEND_ALL>	537	=item C<EVBACKEND_ALL>
284		538
285	Try all backends (even potentially broken ones that wouldn't be tried	539	Try all backends (even potentially broken ones that wouldn't be tried
286	with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as	540	with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
287	C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.	541	C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.
288		542
		543	It is definitely not recommended to use this flag.
		544
289	=back	545	=back
290		546
291	If one or more of these are ored into the flags value, then only these	547	If one or more of the backend flags are or'ed into the flags value,
292	backends will be tried (in the reverse order as given here). If none are	548	then only these backends will be tried (in the reverse order as listed
293	specified, most compiled-in backend will be tried, usually in reverse	549	here). If none are specified, all backends in C<ev_recommended_backends
294	order of their flag values :)	550	()> will be tried.
295		551
296	The most typical usage is like this:	552	Example: This is the most typical usage.
297		553
298	if (!ev_default_loop (0))	554	if (!ev_default_loop (0))
299	fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");	555	fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
300		556
301	Restrict libev to the select and poll backends, and do not allow	557	Example: Restrict libev to the select and poll backends, and do not allow
302	environment settings to be taken into account:	558	environment settings to be taken into account:
303		559
304	ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);	560	ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
305		561
306	Use whatever libev has to offer, but make sure that kqueue is used if	562	Example: Use whatever libev has to offer, but make sure that kqueue is
307	available (warning, breaks stuff, best use only with your own private	563	used if available (warning, breaks stuff, best use only with your own
308	event loop and only if you know the OS supports your types of fds):	564	private event loop and only if you know the OS supports your types of
		565	fds):
309		566
310	ev_default_loop (ev_recommended_backends () | EVBACKEND_KQUEUE);	567	ev_default_loop (ev_recommended_backends () | EVBACKEND_KQUEUE);
311		568
312	=item struct ev_loop *ev_loop_new (unsigned int flags)	569	=item struct ev_loop *ev_loop_new (unsigned int flags)
313		570
314	Similar to C<ev_default_loop>, but always creates a new event loop that is	571	Similar to C<ev_default_loop>, but always creates a new event loop that is
315	always distinct from the default loop. Unlike the default loop, it cannot	572	always distinct from the default loop. Unlike the default loop, it cannot
316	handle signal and child watchers, and attempts to do so will be greeted by	573	handle signal and child watchers, and attempts to do so will be greeted by
317	undefined behaviour (or a failed assertion if assertions are enabled).	574	undefined behaviour (or a failed assertion if assertions are enabled).
318		575
		576	Note that this function I<is> thread-safe, and the recommended way to use
		577	libev with threads is indeed to create one loop per thread, and using the
		578	default loop in the "main" or "initial" thread.
		579
319	Example: try to create a event loop that uses epoll and nothing else.	580	Example: Try to create a event loop that uses epoll and nothing else.
320		581
321	struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);	582	struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
322	if (!epoller)	583	if (!epoller)
323	fatal ("no epoll found here, maybe it hides under your chair");	584	fatal ("no epoll found here, maybe it hides under your chair");
324		585
325	=item ev_default_destroy ()	586	=item ev_default_destroy ()
326		587
327	Destroys the default loop again (frees all memory and kernel state	588	Destroys the default loop again (frees all memory and kernel state
328	etc.). This stops all registered event watchers (by not touching them in	589	etc.). None of the active event watchers will be stopped in the normal
329	any way whatsoever, although you cannot rely on this :).	590	sense, so e.g. C<ev_is_active> might still return true. It is your
		591	responsibility to either stop all watchers cleanly yourself I<before>
		592	calling this function, or cope with the fact afterwards (which is usually
		593	the easiest thing, you can just ignore the watchers and/or C<free ()> them
		594	for example).
		595
		596	Note that certain global state, such as signal state (and installed signal
		597	handlers), will not be freed by this function, and related watchers (such
		598	as signal and child watchers) would need to be stopped manually.
		599
		600	In general it is not advisable to call this function except in the
		601	rare occasion where you really need to free e.g. the signal handling
		602	pipe fds. If you need dynamically allocated loops it is better to use
		603	C<ev_loop_new> and C<ev_loop_destroy>.
330		604
331	=item ev_loop_destroy (loop)	605	=item ev_loop_destroy (loop)
332		606
333	Like C<ev_default_destroy>, but destroys an event loop created by an	607	Like C<ev_default_destroy>, but destroys an event loop created by an
334	earlier call to C<ev_loop_new>.	608	earlier call to C<ev_loop_new>.
335		609
336	=item ev_default_fork ()	610	=item ev_default_fork ()
337		611
		612	This function sets a flag that causes subsequent C<ev_loop> iterations
338	This function reinitialises the kernel state for backends that have	613	to reinitialise the kernel state for backends that have one. Despite the
339	one. Despite the name, you can call it anytime, but it makes most sense	614	name, you can call it anytime, but it makes most sense after forking, in
340	after forking, in either the parent or child process (or both, but that	615	the child process (or both child and parent, but that again makes little
341	again makes little sense).	616	sense). You I<must> call it in the child before using any of the libev
		617	functions, and it will only take effect at the next C<ev_loop> iteration.
342		618
343	You I<must> call this function in the child process after forking if and	619	On the other hand, you only need to call this function in the child
344	only if you want to use the event library in both processes. If you just	620	process if and only if you want to use the event library in the child. If
345	fork+exec, you don't have to call it.	621	you just fork+exec, you don't have to call it at all.
346		622
347	The function itself is quite fast and it's usually not a problem to call	623	The function itself is quite fast and it's usually not a problem to call
348	it just in case after a fork. To make this easy, the function will fit in	624	it just in case after a fork. To make this easy, the function will fit in
349	quite nicely into a call to C<pthread_atfork>:	625	quite nicely into a call to C<pthread_atfork>:
350		626
351	pthread_atfork (0, 0, ev_default_fork);	627	pthread_atfork (0, 0, ev_default_fork);
352		628
353	At the moment, C<EVBACKEND_SELECT> and C<EVBACKEND_POLL> are safe to use
354	without calling this function, so if you force one of those backends you
355	do not need to care.
356
357	=item ev_loop_fork (loop)	629	=item ev_loop_fork (loop)
358		630
359	Like C<ev_default_fork>, but acts on an event loop created by	631	Like C<ev_default_fork>, but acts on an event loop created by
360	C<ev_loop_new>. Yes, you have to call this on every allocated event loop	632	C<ev_loop_new>. Yes, you have to call this on every allocated event loop
361	after fork, and how you do this is entirely your own problem.	633	after fork that you want to re-use in the child, and how you do this is
		634	entirely your own problem.
		635
		636	=item int ev_is_default_loop (loop)
		637
		638	Returns true when the given loop is, in fact, the default loop, and false
		639	otherwise.
		640
		641	=item unsigned int ev_loop_count (loop)
		642
		643	Returns the count of loop iterations for the loop, which is identical to
		644	the number of times libev did poll for new events. It starts at C<0> and
		645	happily wraps around with enough iterations.
		646
		647	This value can sometimes be useful as a generation counter of sorts (it
		648	"ticks" the number of loop iterations), as it roughly corresponds with
		649	C<ev_prepare> and C<ev_check> calls.
		650
		651	=item unsigned int ev_loop_depth (loop)
		652
		653	Returns the number of times C<ev_loop> was entered minus the number of
		654	times C<ev_loop> was exited, in other words, the recursion depth.
		655
		656	Outside C<ev_loop>, this number is zero. In a callback, this number is
		657	C<1>, unless C<ev_loop> was invoked recursively (or from another thread),
		658	in which case it is higher.
		659
		660	Leaving C<ev_loop> abnormally (setjmp/longjmp, cancelling the thread
		661	etc.), doesn't count as exit.
362		662
363	=item unsigned int ev_backend (loop)	663	=item unsigned int ev_backend (loop)
364		664
365	Returns one of the C<EVBACKEND_*> flags indicating the event backend in	665	Returns one of the C<EVBACKEND_*> flags indicating the event backend in
366	use.	666	use.
…		…
369		669
370	Returns the current "event loop time", which is the time the event loop	670	Returns the current "event loop time", which is the time the event loop
371	received events and started processing them. This timestamp does not	671	received events and started processing them. This timestamp does not
372	change as long as callbacks are being processed, and this is also the base	672	change as long as callbacks are being processed, and this is also the base
373	time used for relative timers. You can treat it as the timestamp of the	673	time used for relative timers. You can treat it as the timestamp of the
374	event occuring (or more correctly, libev finding out about it).	674	event occurring (or more correctly, libev finding out about it).
		675
		676	=item ev_now_update (loop)
		677
		678	Establishes the current time by querying the kernel, updating the time
		679	returned by C<ev_now ()> in the progress. This is a costly operation and
		680	is usually done automatically within C<ev_loop ()>.
		681
		682	This function is rarely useful, but when some event callback runs for a
		683	very long time without entering the event loop, updating libev's idea of
		684	the current time is a good idea.
		685
		686	See also L<The special problem of time updates> in the C<ev_timer> section.
		687
		688	=item ev_suspend (loop)
		689
		690	=item ev_resume (loop)
		691
		692	These two functions suspend and resume a loop, for use when the loop is
		693	not used for a while and timeouts should not be processed.
		694
		695	A typical use case would be an interactive program such as a game: When
		696	the user presses C<^Z> to suspend the game and resumes it an hour later it
		697	would be best to handle timeouts as if no time had actually passed while
		698	the program was suspended. This can be achieved by calling C<ev_suspend>
		699	in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
		700	C<ev_resume> directly afterwards to resume timer processing.
		701
		702	Effectively, all C<ev_timer> watchers will be delayed by the time spend
		703	between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
		704	will be rescheduled (that is, they will lose any events that would have
		705	occured while suspended).
		706
		707	After calling C<ev_suspend> you B<must not> call I<any> function on the
		708	given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
		709	without a previous call to C<ev_suspend>.
		710
		711	Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
		712	event loop time (see C<ev_now_update>).
375		713
376	=item ev_loop (loop, int flags)	714	=item ev_loop (loop, int flags)
377		715
378	Finally, this is it, the event handler. This function usually is called	716	Finally, this is it, the event handler. This function usually is called
379	after you initialised all your watchers and you want to start handling	717	after you have initialised all your watchers and you want to start
380	events.	718	handling events.
381		719
382	If the flags argument is specified as C<0>, it will not return until	720	If the flags argument is specified as C<0>, it will not return until
383	either no event watchers are active anymore or C<ev_unloop> was called.	721	either no event watchers are active anymore or C<ev_unloop> was called.
384		722
385	Please note that an explicit C<ev_unloop> is usually better than	723	Please note that an explicit C<ev_unloop> is usually better than
386	relying on all watchers to be stopped when deciding when a program has	724	relying on all watchers to be stopped when deciding when a program has
387	finished (especially in interactive programs), but having a program that	725	finished (especially in interactive programs), but having a program
388	automatically loops as long as it has to and no longer by virtue of	726	that automatically loops as long as it has to and no longer by virtue
389	relying on its watchers stopping correctly is a thing of beauty.	727	of relying on its watchers stopping correctly, that is truly a thing of
		728	beauty.
390		729
391	A flags value of C<EVLOOP_NONBLOCK> will look for new events, will handle	730	A flags value of C<EVLOOP_NONBLOCK> will look for new events, will handle
392	those events and any outstanding ones, but will not block your process in	731	those events and any already outstanding ones, but will not block your
393	case there are no events and will return after one iteration of the loop.	732	process in case there are no events and will return after one iteration of
		733	the loop.
394		734
395	A flags value of C<EVLOOP_ONESHOT> will look for new events (waiting if	735	A flags value of C<EVLOOP_ONESHOT> will look for new events (waiting if
396	neccessary) and will handle those and any outstanding ones. It will block	736	necessary) and will handle those and any already outstanding ones. It
397	your process until at least one new event arrives, and will return after	737	will block your process until at least one new event arrives (which could
398	one iteration of the loop. This is useful if you are waiting for some	738	be an event internal to libev itself, so there is no guarantee that a
399	external event in conjunction with something not expressible using other	739	user-registered callback will be called), and will return after one
		740	iteration of the loop.
		741
		742	This is useful if you are waiting for some external event in conjunction
		743	with something not expressible using other libev watchers (i.e. "roll your
400	libev watchers. However, a pair of C<ev_prepare>/C<ev_check> watchers is	744	own C<ev_loop>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
401	usually a better approach for this kind of thing.	745	usually a better approach for this kind of thing.
402		746
403	Here are the gory details of what C<ev_loop> does:	747	Here are the gory details of what C<ev_loop> does:
404		748
405	* If there are no active watchers (reference count is zero), return.	749	- Before the first iteration, call any pending watchers.
406	- Queue prepare watchers and then call all outstanding watchers.	750	* If EVFLAG_FORKCHECK was used, check for a fork.
		751	- If a fork was detected (by any means), queue and call all fork watchers.
		752	- Queue and call all prepare watchers.
407	- If we have been forked, recreate the kernel state.	753	- If we have been forked, detach and recreate the kernel state
		754	as to not disturb the other process.
408	- Update the kernel state with all outstanding changes.	755	- Update the kernel state with all outstanding changes.
409	- Update the "event loop time".	756	- Update the "event loop time" (ev_now ()).
410	- Calculate for how long to block.	757	- Calculate for how long to sleep or block, if at all
		758	(active idle watchers, EVLOOP_NONBLOCK or not having
		759	any active watchers at all will result in not sleeping).
		760	- Sleep if the I/O and timer collect interval say so.
411	- Block the process, waiting for any events.	761	- Block the process, waiting for any events.
412	- Queue all outstanding I/O (fd) events.	762	- Queue all outstanding I/O (fd) events.
413	- Update the "event loop time" and do time jump handling.	763	- Update the "event loop time" (ev_now ()), and do time jump adjustments.
414	- Queue all outstanding timers.	764	- Queue all expired timers.
415	- Queue all outstanding periodics.	765	- Queue all expired periodics.
416	- If no events are pending now, queue all idle watchers.	766	- Unless any events are pending now, queue all idle watchers.
417	- Queue all check watchers.	767	- Queue all check watchers.
418	- Call all queued watchers in reverse order (i.e. check watchers first).	768	- Call all queued watchers in reverse order (i.e. check watchers first).
419	Signals and child watchers are implemented as I/O watchers, and will	769	Signals and child watchers are implemented as I/O watchers, and will
420	be handled here by queueing them when their watcher gets executed.	770	be handled here by queueing them when their watcher gets executed.
421	- If ev_unloop has been called or EVLOOP_ONESHOT or EVLOOP_NONBLOCK	771	- If ev_unloop has been called, or EVLOOP_ONESHOT or EVLOOP_NONBLOCK
422	were used, return, otherwise continue with step *.	772	were used, or there are no active watchers, return, otherwise
		773	continue with step *.
423		774
424	Example: queue some jobs and then loop until no events are outsanding	775	Example: Queue some jobs and then loop until no events are outstanding
425	anymore.	776	anymore.
426		777
427	... queue jobs here, make sure they register event watchers as long	778	... queue jobs here, make sure they register event watchers as long
428	... as they still have work to do (even an idle watcher will do..)	779	... as they still have work to do (even an idle watcher will do..)
429	ev_loop (my_loop, 0);	780	ev_loop (my_loop, 0);
430	... jobs done. yeah!	781	... jobs done or somebody called unloop. yeah!
431		782
432	=item ev_unloop (loop, how)	783	=item ev_unloop (loop, how)
433		784
434	Can be used to make a call to C<ev_loop> return early (but only after it	785	Can be used to make a call to C<ev_loop> return early (but only after it
435	has processed all outstanding events). The C<how> argument must be either	786	has processed all outstanding events). The C<how> argument must be either
436	C<EVUNLOOP_ONE>, which will make the innermost C<ev_loop> call return, or	787	C<EVUNLOOP_ONE>, which will make the innermost C<ev_loop> call return, or
437	C<EVUNLOOP_ALL>, which will make all nested C<ev_loop> calls return.	788	C<EVUNLOOP_ALL>, which will make all nested C<ev_loop> calls return.
438		789
		790	This "unloop state" will be cleared when entering C<ev_loop> again.
		791
		792	It is safe to call C<ev_unloop> from otuside any C<ev_loop> calls.
		793
439	=item ev_ref (loop)	794	=item ev_ref (loop)
440		795
441	=item ev_unref (loop)	796	=item ev_unref (loop)
442		797
443	Ref/unref can be used to add or remove a reference count on the event	798	Ref/unref can be used to add or remove a reference count on the event
444	loop: Every watcher keeps one reference, and as long as the reference	799	loop: Every watcher keeps one reference, and as long as the reference
445	count is nonzero, C<ev_loop> will not return on its own. If you have	800	count is nonzero, C<ev_loop> will not return on its own.
446	a watcher you never unregister that should not keep C<ev_loop> from	801
447	returning, ev_unref() after starting, and ev_ref() before stopping it. For	802	This is useful when you have a watcher that you never intend to
		803	unregister, but that nevertheless should not keep C<ev_loop> from
		804	returning. In such a case, call C<ev_unref> after starting, and C<ev_ref>
		805	before stopping it.
		806
448	example, libev itself uses this for its internal signal pipe: It is not	807	As an example, libev itself uses this for its internal signal pipe: It
449	visible to the libev user and should not keep C<ev_loop> from exiting if	808	is not visible to the libev user and should not keep C<ev_loop> from
450	no event watchers registered by it are active. It is also an excellent	809	exiting if no event watchers registered by it are active. It is also an
451	way to do this for generic recurring timers or from within third-party	810	excellent way to do this for generic recurring timers or from within
452	libraries. Just remember to I<unref after start> and I<ref before stop>.	811	third-party libraries. Just remember to I<unref after start> and I<ref
		812	before stop> (but only if the watcher wasn't active before, or was active
		813	before, respectively. Note also that libev might stop watchers itself
		814	(e.g. non-repeating timers) in which case you have to C<ev_ref>
		815	in the callback).
453		816
454	Example: create a signal watcher, but keep it from keeping C<ev_loop>	817	Example: Create a signal watcher, but keep it from keeping C<ev_loop>
455	running when nothing else is active.	818	running when nothing else is active.
456		819
457	struct dv_signal exitsig;	820	ev_signal exitsig;
458	ev_signal_init (&exitsig, sig_cb, SIGINT);	821	ev_signal_init (&exitsig, sig_cb, SIGINT);
459	ev_signal_start (myloop, &exitsig);	822	ev_signal_start (loop, &exitsig);
460	evf_unref (myloop);	823	evf_unref (loop);
461		824
462	Example: for some weird reason, unregister the above signal handler again.	825	Example: For some weird reason, unregister the above signal handler again.
463		826
464	ev_ref (myloop);	827	ev_ref (loop);
465	ev_signal_stop (myloop, &exitsig);	828	ev_signal_stop (loop, &exitsig);
		829
		830	=item ev_set_io_collect_interval (loop, ev_tstamp interval)
		831
		832	=item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
		833
		834	These advanced functions influence the time that libev will spend waiting
		835	for events. Both time intervals are by default C<0>, meaning that libev
		836	will try to invoke timer/periodic callbacks and I/O callbacks with minimum
		837	latency.
		838
		839	Setting these to a higher value (the C<interval> I<must> be >= C<0>)
		840	allows libev to delay invocation of I/O and timer/periodic callbacks
		841	to increase efficiency of loop iterations (or to increase power-saving
		842	opportunities).
		843
		844	The idea is that sometimes your program runs just fast enough to handle
		845	one (or very few) event(s) per loop iteration. While this makes the
		846	program responsive, it also wastes a lot of CPU time to poll for new
		847	events, especially with backends like C<select ()> which have a high
		848	overhead for the actual polling but can deliver many events at once.
		849
		850	By setting a higher I<io collect interval> you allow libev to spend more
		851	time collecting I/O events, so you can handle more events per iteration,
		852	at the cost of increasing latency. Timeouts (both C<ev_periodic> and
		853	C<ev_timer>) will be not affected. Setting this to a non-null value will
		854	introduce an additional C<ev_sleep ()> call into most loop iterations. The
		855	sleep time ensures that libev will not poll for I/O events more often then
		856	once per this interval, on average.
		857
		858	Likewise, by setting a higher I<timeout collect interval> you allow libev
		859	to spend more time collecting timeouts, at the expense of increased
		860	latency/jitter/inexactness (the watcher callback will be called
		861	later). C<ev_io> watchers will not be affected. Setting this to a non-null
		862	value will not introduce any overhead in libev.
		863
		864	Many (busy) programs can usually benefit by setting the I/O collect
		865	interval to a value near C<0.1> or so, which is often enough for
		866	interactive servers (of course not for games), likewise for timeouts. It
		867	usually doesn't make much sense to set it to a lower value than C<0.01>,
		868	as this approaches the timing granularity of most systems. Note that if
		869	you do transactions with the outside world and you can't increase the
		870	parallelity, then this setting will limit your transaction rate (if you
		871	need to poll once per transaction and the I/O collect interval is 0.01,
		872	then you can't do more than 100 transations per second).
		873
		874	Setting the I<timeout collect interval> can improve the opportunity for
		875	saving power, as the program will "bundle" timer callback invocations that
		876	are "near" in time together, by delaying some, thus reducing the number of
		877	times the process sleeps and wakes up again. Another useful technique to
		878	reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
		879	they fire on, say, one-second boundaries only.
		880
		881	Example: we only need 0.1s timeout granularity, and we wish not to poll
		882	more often than 100 times per second:
		883
		884	ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
		885	ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
		886
		887	=item ev_invoke_pending (loop)
		888
		889	This call will simply invoke all pending watchers while resetting their
		890	pending state. Normally, C<ev_loop> does this automatically when required,
		891	but when overriding the invoke callback this call comes handy.
		892
		893	=item int ev_pending_count (loop)
		894
		895	Returns the number of pending watchers - zero indicates that no watchers
		896	are pending.
		897
		898	=item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
		899
		900	This overrides the invoke pending functionality of the loop: Instead of
		901	invoking all pending watchers when there are any, C<ev_loop> will call
		902	this callback instead. This is useful, for example, when you want to
		903	invoke the actual watchers inside another context (another thread etc.).
		904
		905	If you want to reset the callback, use C<ev_invoke_pending> as new
		906	callback.
		907
		908	=item ev_set_loop_release_cb (loop, void (*release)(EV_P), void (*acquire)(EV_P))
		909
		910	Sometimes you want to share the same loop between multiple threads. This
		911	can be done relatively simply by putting mutex_lock/unlock calls around
		912	each call to a libev function.
		913
		914	However, C<ev_loop> can run an indefinite time, so it is not feasible to
		915	wait for it to return. One way around this is to wake up the loop via
		916	C<ev_unloop> and C<av_async_send>, another way is to set these I<release>
		917	and I<acquire> callbacks on the loop.
		918
		919	When set, then C<release> will be called just before the thread is
		920	suspended waiting for new events, and C<acquire> is called just
		921	afterwards.
		922
		923	Ideally, C<release> will just call your mutex_unlock function, and
		924	C<acquire> will just call the mutex_lock function again.
		925
		926	While event loop modifications are allowed between invocations of
		927	C<release> and C<acquire> (that's their only purpose after all), no
		928	modifications done will affect the event loop, i.e. adding watchers will
		929	have no effect on the set of file descriptors being watched, or the time
		930	waited. Use an C<ev_async> watcher to wake up C<ev_loop> when you want it
		931	to take note of any changes you made.
		932
		933	In theory, threads executing C<ev_loop> will be async-cancel safe between
		934	invocations of C<release> and C<acquire>.
		935
		936	See also the locking example in the C<THREADS> section later in this
		937	document.
		938
		939	=item ev_set_userdata (loop, void *data)
		940
		941	=item ev_userdata (loop)
		942
		943	Set and retrieve a single C<void *> associated with a loop. When
		944	C<ev_set_userdata> has never been called, then C<ev_userdata> returns
		945	C<0.>
		946
		947	These two functions can be used to associate arbitrary data with a loop,
		948	and are intended solely for the C<invoke_pending_cb>, C<release> and
		949	C<acquire> callbacks described above, but of course can be (ab-)used for
		950	any other purpose as well.
		951
		952	=item ev_loop_verify (loop)
		953
		954	This function only does something when C<EV_VERIFY> support has been
		955	compiled in, which is the default for non-minimal builds. It tries to go
		956	through all internal structures and checks them for validity. If anything
		957	is found to be inconsistent, it will print an error message to standard
		958	error and call C<abort ()>.
		959
		960	This can be used to catch bugs inside libev itself: under normal
		961	circumstances, this function will never abort as of course libev keeps its
		962	data structures consistent.
466		963
467	=back	964	=back
468		965
		966
469	=head1 ANATOMY OF A WATCHER	967	=head1 ANATOMY OF A WATCHER
		968
		969	In the following description, uppercase C<TYPE> in names stands for the
		970	watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
		971	watchers and C<ev_io_start> for I/O watchers.
470		972
471	A watcher is a structure that you create and register to record your	973	A watcher is a structure that you create and register to record your
472	interest in some event. For instance, if you want to wait for STDIN to	974	interest in some event. For instance, if you want to wait for STDIN to
473	become readable, you would create an C<ev_io> watcher for that:	975	become readable, you would create an C<ev_io> watcher for that:
474		976
475	static void my_cb (struct ev_loop *loop, struct ev_io *w, int revents)	977	static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
476	{	978	{
477	ev_io_stop (w);	979	ev_io_stop (w);
478	ev_unloop (loop, EVUNLOOP_ALL);	980	ev_unloop (loop, EVUNLOOP_ALL);
479	}	981	}
480		982
481	struct ev_loop *loop = ev_default_loop (0);	983	struct ev_loop *loop = ev_default_loop (0);
		984
482	struct ev_io stdin_watcher;	985	ev_io stdin_watcher;
		986
483	ev_init (&stdin_watcher, my_cb);	987	ev_init (&stdin_watcher, my_cb);
484	ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);	988	ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
485	ev_io_start (loop, &stdin_watcher);	989	ev_io_start (loop, &stdin_watcher);
		990
486	ev_loop (loop, 0);	991	ev_loop (loop, 0);
487		992
488	As you can see, you are responsible for allocating the memory for your	993	As you can see, you are responsible for allocating the memory for your
489	watcher structures (and it is usually a bad idea to do this on the stack,	994	watcher structures (and it is I<usually> a bad idea to do this on the
490	although this can sometimes be quite valid).	995	stack).
		996
		997	Each watcher has an associated watcher structure (called C<struct ev_TYPE>
		998	or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
491		999
492	Each watcher structure must be initialised by a call to C<ev_init	1000	Each watcher structure must be initialised by a call to C<ev_init
493	(watcher *, callback)>, which expects a callback to be provided. This	1001	(watcher *, callback)>, which expects a callback to be provided. This
494	callback gets invoked each time the event occurs (or, in the case of io	1002	callback gets invoked each time the event occurs (or, in the case of I/O
495	watchers, each time the event loop detects that the file descriptor given	1003	watchers, each time the event loop detects that the file descriptor given
496	is readable and/or writable).	1004	is readable and/or writable).
497		1005
498	Each watcher type has its own C<< ev_<type>_set (watcher *, ...) >> macro	1006	Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
499	with arguments specific to this watcher type. There is also a macro	1007	macro to configure it, with arguments specific to the watcher type. There
500	to combine initialisation and setting in one call: C<< ev_<type>_init	1008	is also a macro to combine initialisation and setting in one call: C<<
501	(watcher *, callback, ...) >>.	1009	ev_TYPE_init (watcher *, callback, ...) >>.
502		1010
503	To make the watcher actually watch out for events, you have to start it	1011	To make the watcher actually watch out for events, you have to start it
504	with a watcher-specific start function (C<< ev_<type>_start (loop, watcher	1012	with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
505	*) >>), and you can stop watching for events at any time by calling the	1013	*) >>), and you can stop watching for events at any time by calling the
506	corresponding stop function (C<< ev_<type>_stop (loop, watcher *) >>.	1014	corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
507		1015
508	As long as your watcher is active (has been started but not stopped) you	1016	As long as your watcher is active (has been started but not stopped) you
509	must not touch the values stored in it. Most specifically you must never	1017	must not touch the values stored in it. Most specifically you must never
510	reinitialise it or call its set macro.	1018	reinitialise it or call its C<ev_TYPE_set> macro.
511
512	You can check whether an event is active by calling the C<ev_is_active
513	(watcher *)> macro. To see whether an event is outstanding (but the
514	callback for it has not been called yet) you can use the C<ev_is_pending
515	(watcher *)> macro.
516		1019
517	Each and every callback receives the event loop pointer as first, the	1020	Each and every callback receives the event loop pointer as first, the
518	registered watcher structure as second, and a bitset of received events as	1021	registered watcher structure as second, and a bitset of received events as
519	third argument.	1022	third argument.
520		1023
…		…
544	The signal specified in the C<ev_signal> watcher has been received by a thread.	1047	The signal specified in the C<ev_signal> watcher has been received by a thread.
545		1048
546	=item C<EV_CHILD>	1049	=item C<EV_CHILD>
547		1050
548	The pid specified in the C<ev_child> watcher has received a status change.	1051	The pid specified in the C<ev_child> watcher has received a status change.
		1052
		1053	=item C<EV_STAT>
		1054
		1055	The path specified in the C<ev_stat> watcher changed its attributes somehow.
549		1056
550	=item C<EV_IDLE>	1057	=item C<EV_IDLE>
551		1058
552	The C<ev_idle> watcher has determined that you have nothing better to do.	1059	The C<ev_idle> watcher has determined that you have nothing better to do.
553		1060
…		…
561	received events. Callbacks of both watcher types can start and stop as	1068	received events. Callbacks of both watcher types can start and stop as
562	many watchers as they want, and all of them will be taken into account	1069	many watchers as they want, and all of them will be taken into account
563	(for example, a C<ev_prepare> watcher might start an idle watcher to keep	1070	(for example, a C<ev_prepare> watcher might start an idle watcher to keep
564	C<ev_loop> from blocking).	1071	C<ev_loop> from blocking).
565		1072
		1073	=item C<EV_EMBED>
		1074
		1075	The embedded event loop specified in the C<ev_embed> watcher needs attention.
		1076
		1077	=item C<EV_FORK>
		1078
		1079	The event loop has been resumed in the child process after fork (see
		1080	C<ev_fork>).
		1081
		1082	=item C<EV_ASYNC>
		1083
		1084	The given async watcher has been asynchronously notified (see C<ev_async>).
		1085
		1086	=item C<EV_CUSTOM>
		1087
		1088	Not ever sent (or otherwise used) by libev itself, but can be freely used
		1089	by libev users to signal watchers (e.g. via C<ev_feed_event>).
		1090
566	=item C<EV_ERROR>	1091	=item C<EV_ERROR>
567		1092
568	An unspecified error has occured, the watcher has been stopped. This might	1093	An unspecified error has occurred, the watcher has been stopped. This might
569	happen because the watcher could not be properly started because libev	1094	happen because the watcher could not be properly started because libev
570	ran out of memory, a file descriptor was found to be closed or any other	1095	ran out of memory, a file descriptor was found to be closed or any other
		1096	problem. Libev considers these application bugs.
		1097
571	problem. You best act on it by reporting the problem and somehow coping	1098	You best act on it by reporting the problem and somehow coping with the
572	with the watcher being stopped.	1099	watcher being stopped. Note that well-written programs should not receive
		1100	an error ever, so when your watcher receives it, this usually indicates a
		1101	bug in your program.
573		1102
574	Libev will usually signal a few "dummy" events together with an error,	1103	Libev will usually signal a few "dummy" events together with an error, for
575	for example it might indicate that a fd is readable or writable, and if	1104	example it might indicate that a fd is readable or writable, and if your
576	your callbacks is well-written it can just attempt the operation and cope	1105	callbacks is well-written it can just attempt the operation and cope with
577	with the error from read() or write(). This will not work in multithreaded	1106	the error from read() or write(). This will not work in multi-threaded
578	programs, though, so beware.	1107	programs, though, as the fd could already be closed and reused for another
		1108	thing, so beware.
579		1109
580	=back	1110	=back
581		1111
		1112	=head2 GENERIC WATCHER FUNCTIONS
		1113
		1114	=over 4
		1115
		1116	=item C<ev_init> (ev_TYPE *watcher, callback)
		1117
		1118	This macro initialises the generic portion of a watcher. The contents
		1119	of the watcher object can be arbitrary (so C<malloc> will do). Only
		1120	the generic parts of the watcher are initialised, you I<need> to call
		1121	the type-specific C<ev_TYPE_set> macro afterwards to initialise the
		1122	type-specific parts. For each type there is also a C<ev_TYPE_init> macro
		1123	which rolls both calls into one.
		1124
		1125	You can reinitialise a watcher at any time as long as it has been stopped
		1126	(or never started) and there are no pending events outstanding.
		1127
		1128	The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
		1129	int revents)>.
		1130
		1131	Example: Initialise an C<ev_io> watcher in two steps.
		1132
		1133	ev_io w;
		1134	ev_init (&w, my_cb);
		1135	ev_io_set (&w, STDIN_FILENO, EV_READ);
		1136
		1137	=item C<ev_TYPE_set> (ev_TYPE *watcher, [args])
		1138
		1139	This macro initialises the type-specific parts of a watcher. You need to
		1140	call C<ev_init> at least once before you call this macro, but you can
		1141	call C<ev_TYPE_set> any number of times. You must not, however, call this
		1142	macro on a watcher that is active (it can be pending, however, which is a
		1143	difference to the C<ev_init> macro).
		1144
		1145	Although some watcher types do not have type-specific arguments
		1146	(e.g. C<ev_prepare>) you still need to call its C<set> macro.
		1147
		1148	See C<ev_init>, above, for an example.
		1149
		1150	=item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
		1151
		1152	This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
		1153	calls into a single call. This is the most convenient method to initialise
		1154	a watcher. The same limitations apply, of course.
		1155
		1156	Example: Initialise and set an C<ev_io> watcher in one step.
		1157
		1158	ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
		1159
		1160	=item C<ev_TYPE_start> (loop, ev_TYPE *watcher)
		1161
		1162	Starts (activates) the given watcher. Only active watchers will receive
		1163	events. If the watcher is already active nothing will happen.
		1164
		1165	Example: Start the C<ev_io> watcher that is being abused as example in this
		1166	whole section.
		1167
		1168	ev_io_start (EV_DEFAULT_UC, &w);
		1169
		1170	=item C<ev_TYPE_stop> (loop, ev_TYPE *watcher)
		1171
		1172	Stops the given watcher if active, and clears the pending status (whether
		1173	the watcher was active or not).
		1174
		1175	It is possible that stopped watchers are pending - for example,
		1176	non-repeating timers are being stopped when they become pending - but
		1177	calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
		1178	pending. If you want to free or reuse the memory used by the watcher it is
		1179	therefore a good idea to always call its C<ev_TYPE_stop> function.
		1180
		1181	=item bool ev_is_active (ev_TYPE *watcher)
		1182
		1183	Returns a true value iff the watcher is active (i.e. it has been started
		1184	and not yet been stopped). As long as a watcher is active you must not modify
		1185	it.
		1186
		1187	=item bool ev_is_pending (ev_TYPE *watcher)
		1188
		1189	Returns a true value iff the watcher is pending, (i.e. it has outstanding
		1190	events but its callback has not yet been invoked). As long as a watcher
		1191	is pending (but not active) you must not call an init function on it (but
		1192	C<ev_TYPE_set> is safe), you must not change its priority, and you must
		1193	make sure the watcher is available to libev (e.g. you cannot C<free ()>
		1194	it).
		1195
		1196	=item callback ev_cb (ev_TYPE *watcher)
		1197
		1198	Returns the callback currently set on the watcher.
		1199
		1200	=item ev_cb_set (ev_TYPE *watcher, callback)
		1201
		1202	Change the callback. You can change the callback at virtually any time
		1203	(modulo threads).
		1204
		1205	=item ev_set_priority (ev_TYPE *watcher, int priority)
		1206
		1207	=item int ev_priority (ev_TYPE *watcher)
		1208
		1209	Set and query the priority of the watcher. The priority is a small
		1210	integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
		1211	(default: C<-2>). Pending watchers with higher priority will be invoked
		1212	before watchers with lower priority, but priority will not keep watchers
		1213	from being executed (except for C<ev_idle> watchers).
		1214
		1215	If you need to suppress invocation when higher priority events are pending
		1216	you need to look at C<ev_idle> watchers, which provide this functionality.
		1217
		1218	You I<must not> change the priority of a watcher as long as it is active or
		1219	pending.
		1220
		1221	Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
		1222	fine, as long as you do not mind that the priority value you query might
		1223	or might not have been clamped to the valid range.
		1224
		1225	The default priority used by watchers when no priority has been set is
		1226	always C<0>, which is supposed to not be too high and not be too low :).
		1227
		1228	See L<WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
		1229	priorities.
		1230
		1231	=item ev_invoke (loop, ev_TYPE *watcher, int revents)
		1232
		1233	Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
		1234	C<loop> nor C<revents> need to be valid as long as the watcher callback
		1235	can deal with that fact, as both are simply passed through to the
		1236	callback.
		1237
		1238	=item int ev_clear_pending (loop, ev_TYPE *watcher)
		1239
		1240	If the watcher is pending, this function clears its pending status and
		1241	returns its C<revents> bitset (as if its callback was invoked). If the
		1242	watcher isn't pending it does nothing and returns C<0>.
		1243
		1244	Sometimes it can be useful to "poll" a watcher instead of waiting for its
		1245	callback to be invoked, which can be accomplished with this function.
		1246
		1247	=item ev_feed_event (loop, ev_TYPE *watcher, int revents)
		1248
		1249	Feeds the given event set into the event loop, as if the specified event
		1250	had happened for the specified watcher (which must be a pointer to an
		1251	initialised but not necessarily started event watcher). Obviously you must
		1252	not free the watcher as long as it has pending events.
		1253
		1254	Stopping the watcher, letting libev invoke it, or calling
		1255	C<ev_clear_pending> will clear the pending event, even if the watcher was
		1256	not started in the first place.
		1257
		1258	See also C<ev_feed_fd_event> and C<ev_feed_signal_event> for related
		1259	functions that do not need a watcher.
		1260
		1261	=back
		1262
		1263
582	=head2 ASSOCIATING CUSTOM DATA WITH A WATCHER	1264	=head2 ASSOCIATING CUSTOM DATA WITH A WATCHER
583		1265
584	Each watcher has, by default, a member C<void *data> that you can change	1266	Each watcher has, by default, a member C<void *data> that you can change
585	and read at any time, libev will completely ignore it. This can be used	1267	and read at any time: libev will completely ignore it. This can be used
586	to associate arbitrary data with your watcher. If you need more data and	1268	to associate arbitrary data with your watcher. If you need more data and
587	don't want to allocate memory and store a pointer to it in that data	1269	don't want to allocate memory and store a pointer to it in that data
588	member, you can also "subclass" the watcher type and provide your own	1270	member, you can also "subclass" the watcher type and provide your own
589	data:	1271	data:
590		1272
591	struct my_io	1273	struct my_io
592	{	1274	{
593	struct ev_io io;	1275	ev_io io;
594	int otherfd;	1276	int otherfd;
595	void *somedata;	1277	void *somedata;
596	struct whatever *mostinteresting;	1278	struct whatever *mostinteresting;
597	}	1279	};
		1280
		1281	...
		1282	struct my_io w;
		1283	ev_io_init (&w.io, my_cb, fd, EV_READ);
598		1284
599	And since your callback will be called with a pointer to the watcher, you	1285	And since your callback will be called with a pointer to the watcher, you
600	can cast it back to your own type:	1286	can cast it back to your own type:
601		1287
602	static void my_cb (struct ev_loop *loop, struct ev_io *w_, int revents)	1288	static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
603	{	1289	{
604	struct my_io *w = (struct my_io *)w_;	1290	struct my_io *w = (struct my_io *)w_;
605	...	1291	...
606	}	1292	}
607		1293
608	More interesting and less C-conformant ways of catsing your callback type	1294	More interesting and less C-conformant ways of casting your callback type
609	have been omitted....	1295	instead have been omitted.
		1296
		1297	Another common scenario is to use some data structure with multiple
		1298	embedded watchers:
		1299
		1300	struct my_biggy
		1301	{
		1302	int some_data;
		1303	ev_timer t1;
		1304	ev_timer t2;
		1305	}
		1306
		1307	In this case getting the pointer to C<my_biggy> is a bit more
		1308	complicated: Either you store the address of your C<my_biggy> struct
		1309	in the C<data> member of the watcher (for woozies), or you need to use
		1310	some pointer arithmetic using C<offsetof> inside your watchers (for real
		1311	programmers):
		1312
		1313	#include <stddef.h>
		1314
		1315	static void
		1316	t1_cb (EV_P_ ev_timer *w, int revents)
		1317	{
		1318	struct my_biggy big = (struct my_biggy *)
		1319	(((char *)w) - offsetof (struct my_biggy, t1));
		1320	}
		1321
		1322	static void
		1323	t2_cb (EV_P_ ev_timer *w, int revents)
		1324	{
		1325	struct my_biggy big = (struct my_biggy *)
		1326	(((char *)w) - offsetof (struct my_biggy, t2));
		1327	}
		1328
		1329	=head2 WATCHER PRIORITY MODELS
		1330
		1331	Many event loops support I<watcher priorities>, which are usually small
		1332	integers that influence the ordering of event callback invocation
		1333	between watchers in some way, all else being equal.
		1334
		1335	In libev, Watcher priorities can be set using C<ev_set_priority>. See its
		1336	description for the more technical details such as the actual priority
		1337	range.
		1338
		1339	There are two common ways how these these priorities are being interpreted
		1340	by event loops:
		1341
		1342	In the more common lock-out model, higher priorities "lock out" invocation
		1343	of lower priority watchers, which means as long as higher priority
		1344	watchers receive events, lower priority watchers are not being invoked.
		1345
		1346	The less common only-for-ordering model uses priorities solely to order
		1347	callback invocation within a single event loop iteration: Higher priority
		1348	watchers are invoked before lower priority ones, but they all get invoked
		1349	before polling for new events.
		1350
		1351	Libev uses the second (only-for-ordering) model for all its watchers
		1352	except for idle watchers (which use the lock-out model).
		1353
		1354	The rationale behind this is that implementing the lock-out model for
		1355	watchers is not well supported by most kernel interfaces, and most event
		1356	libraries will just poll for the same events again and again as long as
		1357	their callbacks have not been executed, which is very inefficient in the
		1358	common case of one high-priority watcher locking out a mass of lower
		1359	priority ones.
		1360
		1361	Static (ordering) priorities are most useful when you have two or more
		1362	watchers handling the same resource: a typical usage example is having an
		1363	C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
		1364	timeouts. Under load, data might be received while the program handles
		1365	other jobs, but since timers normally get invoked first, the timeout
		1366	handler will be executed before checking for data. In that case, giving
		1367	the timer a lower priority than the I/O watcher ensures that I/O will be
		1368	handled first even under adverse conditions (which is usually, but not
		1369	always, what you want).
		1370
		1371	Since idle watchers use the "lock-out" model, meaning that idle watchers
		1372	will only be executed when no same or higher priority watchers have
		1373	received events, they can be used to implement the "lock-out" model when
		1374	required.
		1375
		1376	For example, to emulate how many other event libraries handle priorities,
		1377	you can associate an C<ev_idle> watcher to each such watcher, and in
		1378	the normal watcher callback, you just start the idle watcher. The real
		1379	processing is done in the idle watcher callback. This causes libev to
		1380	continously poll and process kernel event data for the watcher, but when
		1381	the lock-out case is known to be rare (which in turn is rare :), this is
		1382	workable.
		1383
		1384	Usually, however, the lock-out model implemented that way will perform
		1385	miserably under the type of load it was designed to handle. In that case,
		1386	it might be preferable to stop the real watcher before starting the
		1387	idle watcher, so the kernel will not have to process the event in case
		1388	the actual processing will be delayed for considerable time.
		1389
		1390	Here is an example of an I/O watcher that should run at a strictly lower
		1391	priority than the default, and which should only process data when no
		1392	other events are pending:
		1393
		1394	ev_idle idle; // actual processing watcher
		1395	ev_io io; // actual event watcher
		1396
		1397	static void
		1398	io_cb (EV_P_ ev_io *w, int revents)
		1399	{
		1400	// stop the I/O watcher, we received the event, but
		1401	// are not yet ready to handle it.
		1402	ev_io_stop (EV_A_ w);
		1403
		1404	// start the idle watcher to ahndle the actual event.
		1405	// it will not be executed as long as other watchers
		1406	// with the default priority are receiving events.
		1407	ev_idle_start (EV_A_ &idle);
		1408	}
		1409
		1410	static void
		1411	idle_cb (EV_P_ ev_idle *w, int revents)
		1412	{
		1413	// actual processing
		1414	read (STDIN_FILENO, ...);
		1415
		1416	// have to start the I/O watcher again, as
		1417	// we have handled the event
		1418	ev_io_start (EV_P_ &io);
		1419	}
		1420
		1421	// initialisation
		1422	ev_idle_init (&idle, idle_cb);
		1423	ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
		1424	ev_io_start (EV_DEFAULT_ &io);
		1425
		1426	In the "real" world, it might also be beneficial to start a timer, so that
		1427	low-priority connections can not be locked out forever under load. This
		1428	enables your program to keep a lower latency for important connections
		1429	during short periods of high load, while not completely locking out less
		1430	important ones.
610		1431
611		1432
612	=head1 WATCHER TYPES	1433	=head1 WATCHER TYPES
613		1434
614	This section describes each watcher in detail, but will not repeat	1435	This section describes each watcher in detail, but will not repeat
615	information given in the last section.	1436	information given in the last section. Any initialisation/set macros,
		1437	functions and members specific to the watcher type are explained.
616		1438
		1439	Members are additionally marked with either I<[read-only]>, meaning that,
		1440	while the watcher is active, you can look at the member and expect some
		1441	sensible content, but you must not modify it (you can modify it while the
		1442	watcher is stopped to your hearts content), or I<[read-write]>, which
		1443	means you can expect it to have some sensible content while the watcher
		1444	is active, but you can also modify it. Modifying it may not do something
		1445	sensible or take immediate effect (or do anything at all), but libev will
		1446	not crash or malfunction in any way.
617		1447
		1448
618	=head2 C<ev_io> - is this file descriptor readable or writable	1449	=head2 C<ev_io> - is this file descriptor readable or writable?
619		1450
620	I/O watchers check whether a file descriptor is readable or writable	1451	I/O watchers check whether a file descriptor is readable or writable
621	in each iteration of the event loop (This behaviour is called	1452	in each iteration of the event loop, or, more precisely, when reading
622	level-triggering because you keep receiving events as long as the	1453	would not block the process and writing would at least be able to write
623	condition persists. Remember you can stop the watcher if you don't want to	1454	some data. This behaviour is called level-triggering because you keep
624	act on the event and neither want to receive future events).	1455	receiving events as long as the condition persists. Remember you can stop
		1456	the watcher if you don't want to act on the event and neither want to
		1457	receive future events.
625		1458
626	In general you can register as many read and/or write event watchers per	1459	In general you can register as many read and/or write event watchers per
627	fd as you want (as long as you don't confuse yourself). Setting all file	1460	fd as you want (as long as you don't confuse yourself). Setting all file
628	descriptors to non-blocking mode is also usually a good idea (but not	1461	descriptors to non-blocking mode is also usually a good idea (but not
629	required if you know what you are doing).	1462	required if you know what you are doing).
630		1463
631	You have to be careful with dup'ed file descriptors, though. Some backends	1464	If you cannot use non-blocking mode, then force the use of a
632	(the linux epoll backend is a notable example) cannot handle dup'ed file	1465	known-to-be-good backend (at the time of this writing, this includes only
633	descriptors correctly if you register interest in two or more fds pointing	1466	C<EVBACKEND_SELECT> and C<EVBACKEND_POLL>). The same applies to file
634	to the same underlying file/socket etc. description (that is, they share	1467	descriptors for which non-blocking operation makes no sense (such as
635	the same underlying "file open").	1468	files) - libev doesn't guarentee any specific behaviour in that case.
636		1469
637	If you must do this, then force the use of a known-to-be-good backend	1470	Another thing you have to watch out for is that it is quite easy to
638	(at the time of this writing, this includes only C<EVBACKEND_SELECT> and	1471	receive "spurious" readiness notifications, that is your callback might
		1472	be called with C<EV_READ> but a subsequent C<read>(2) will actually block
		1473	because there is no data. Not only are some backends known to create a
		1474	lot of those (for example Solaris ports), it is very easy to get into
		1475	this situation even with a relatively standard program structure. Thus
		1476	it is best to always use non-blocking I/O: An extra C<read>(2) returning
		1477	C<EAGAIN> is far preferable to a program hanging until some data arrives.
		1478
		1479	If you cannot run the fd in non-blocking mode (for example you should
		1480	not play around with an Xlib connection), then you have to separately
		1481	re-test whether a file descriptor is really ready with a known-to-be good
		1482	interface such as poll (fortunately in our Xlib example, Xlib already
		1483	does this on its own, so its quite safe to use). Some people additionally
		1484	use C<SIGALRM> and an interval timer, just to be sure you won't block
		1485	indefinitely.
		1486
		1487	But really, best use non-blocking mode.
		1488
		1489	=head3 The special problem of disappearing file descriptors
		1490
		1491	Some backends (e.g. kqueue, epoll) need to be told about closing a file
		1492	descriptor (either due to calling C<close> explicitly or any other means,
		1493	such as C<dup2>). The reason is that you register interest in some file
		1494	descriptor, but when it goes away, the operating system will silently drop
		1495	this interest. If another file descriptor with the same number then is
		1496	registered with libev, there is no efficient way to see that this is, in
		1497	fact, a different file descriptor.
		1498
		1499	To avoid having to explicitly tell libev about such cases, libev follows
		1500	the following policy: Each time C<ev_io_set> is being called, libev
		1501	will assume that this is potentially a new file descriptor, otherwise
		1502	it is assumed that the file descriptor stays the same. That means that
		1503	you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
		1504	descriptor even if the file descriptor number itself did not change.
		1505
		1506	This is how one would do it normally anyway, the important point is that
		1507	the libev application should not optimise around libev but should leave
		1508	optimisations to libev.
		1509
		1510	=head3 The special problem of dup'ed file descriptors
		1511
		1512	Some backends (e.g. epoll), cannot register events for file descriptors,
		1513	but only events for the underlying file descriptions. That means when you
		1514	have C<dup ()>'ed file descriptors or weirder constellations, and register
		1515	events for them, only one file descriptor might actually receive events.
		1516
		1517	There is no workaround possible except not registering events
		1518	for potentially C<dup ()>'ed file descriptors, or to resort to
		1519	C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
		1520
		1521	=head3 The special problem of fork
		1522
		1523	Some backends (epoll, kqueue) do not support C<fork ()> at all or exhibit
		1524	useless behaviour. Libev fully supports fork, but needs to be told about
		1525	it in the child.
		1526
		1527	To support fork in your programs, you either have to call
		1528	C<ev_default_fork ()> or C<ev_loop_fork ()> after a fork in the child,
		1529	enable C<EVFLAG_FORKCHECK>, or resort to C<EVBACKEND_SELECT> or
639	C<EVBACKEND_POLL>).	1530	C<EVBACKEND_POLL>.
		1531
		1532	=head3 The special problem of SIGPIPE
		1533
		1534	While not really specific to libev, it is easy to forget about C<SIGPIPE>:
		1535	when writing to a pipe whose other end has been closed, your program gets
		1536	sent a SIGPIPE, which, by default, aborts your program. For most programs
		1537	this is sensible behaviour, for daemons, this is usually undesirable.
		1538
		1539	So when you encounter spurious, unexplained daemon exits, make sure you
		1540	ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
		1541	somewhere, as that would have given you a big clue).
		1542
		1543	=head3 The special problem of accept()ing when you can't
		1544
		1545	Many implementations of the POSIX C<accept> function (for example,
		1546	found in port-2004 Linux) have the peculiar behaviour of not removing a
		1547	connection from the pending queue in all error cases.
		1548
		1549	For example, larger servers often run out of file descriptors (because
		1550	of resource limits), causing C<accept> to fail with C<ENFILE> but not
		1551	rejecting the connection, leading to libev signalling readiness on
		1552	the next iteration again (the connection still exists after all), and
		1553	typically causing the program to loop at 100% CPU usage.
		1554
		1555	Unfortunately, the set of errors that cause this issue differs between
		1556	operating systems, there is usually little the app can do to remedy the
		1557	situation, and no known thread-safe method of removing the connection to
		1558	cope with overload is known (to me).
		1559
		1560	One of the easiest ways to handle this situation is to just ignore it
		1561	- when the program encounters an overload, it will just loop until the
		1562	situation is over. While this is a form of busy waiting, no OS offers an
		1563	event-based way to handle this situation, so it's the best one can do.
		1564
		1565	A better way to handle the situation is to log any errors other than
		1566	C<EAGAIN> and C<EWOULDBLOCK>, making sure not to flood the log with such
		1567	messages, and continue as usual, which at least gives the user an idea of
		1568	what could be wrong ("raise the ulimit!"). For extra points one could stop
		1569	the C<ev_io> watcher on the listening fd "for a while", which reduces CPU
		1570	usage.
		1571
		1572	If your program is single-threaded, then you could also keep a dummy file
		1573	descriptor for overload situations (e.g. by opening F</dev/null>), and
		1574	when you run into C<ENFILE> or C<EMFILE>, close it, run C<accept>,
		1575	close that fd, and create a new dummy fd. This will gracefully refuse
		1576	clients under typical overload conditions.
		1577
		1578	The last way to handle it is to simply log the error and C<exit>, as
		1579	is often done with C<malloc> failures, but this results in an easy
		1580	opportunity for a DoS attack.
		1581
		1582	=head3 Watcher-Specific Functions
640		1583
641	=over 4	1584	=over 4
642		1585
643	=item ev_io_init (ev_io *, callback, int fd, int events)	1586	=item ev_io_init (ev_io *, callback, int fd, int events)
644		1587
645	=item ev_io_set (ev_io *, int fd, int events)	1588	=item ev_io_set (ev_io *, int fd, int events)
646		1589
647	Configures an C<ev_io> watcher. The fd is the file descriptor to rceeive	1590	Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
648	events for and events is either C<EV_READ>, C<EV_WRITE> or C<EV_READ |	1591	receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
649	EV_WRITE> to receive the given events.	1592	C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
650		1593
651	Please note that most of the more scalable backend mechanisms (for example	1594	=item int fd [read-only]
652	epoll and solaris ports) can result in spurious readyness notifications	1595
653	for file descriptors, so you practically need to use non-blocking I/O (and	1596	The file descriptor being watched.
654	treat callback invocation as hint only), or retest separately with a safe	1597
655	interface before doing I/O (XLib can do this), or force the use of either	1598	=item int events [read-only]
656	C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>, which don't suffer from this	1599
657	problem. Also note that it is quite easy to have your callback invoked	1600	The events being watched.
658	when the readyness condition is no longer valid even when employing
659	typical ways of handling events, so its a good idea to use non-blocking
660	I/O unconditionally.
661		1601
662	=back	1602	=back
663		1603
		1604	=head3 Examples
		1605
664	Example: call C<stdin_readable_cb> when STDIN_FILENO has become, well	1606	Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
665	readable, but only once. Since it is likely line-buffered, you could	1607	readable, but only once. Since it is likely line-buffered, you could
666	attempt to read a whole line in the callback:	1608	attempt to read a whole line in the callback.
667		1609
668	static void	1610	static void
669	stdin_readable_cb (struct ev_loop *loop, struct ev_io *w, int revents)	1611	stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
670	{	1612	{
671	ev_io_stop (loop, w);	1613	ev_io_stop (loop, w);
672	.. read from stdin here (or from w->fd) and haqndle any I/O errors	1614	.. read from stdin here (or from w->fd) and handle any I/O errors
673	}	1615	}
674		1616
675	...	1617	...
676	struct ev_loop *loop = ev_default_init (0);	1618	struct ev_loop *loop = ev_default_init (0);
677	struct ev_io stdin_readable;	1619	ev_io stdin_readable;
678	ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);	1620	ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
679	ev_io_start (loop, &stdin_readable);	1621	ev_io_start (loop, &stdin_readable);
680	ev_loop (loop, 0);	1622	ev_loop (loop, 0);
681		1623
682		1624
683	=head2 C<ev_timer> - relative and optionally recurring timeouts	1625	=head2 C<ev_timer> - relative and optionally repeating timeouts
684		1626
685	Timer watchers are simple relative timers that generate an event after a	1627	Timer watchers are simple relative timers that generate an event after a
686	given time, and optionally repeating in regular intervals after that.	1628	given time, and optionally repeating in regular intervals after that.
687		1629
688	The timers are based on real time, that is, if you register an event that	1630	The timers are based on real time, that is, if you register an event that
689	times out after an hour and you reset your system clock to last years	1631	times out after an hour and you reset your system clock to January last
690	time, it will still time out after (roughly) and hour. "Roughly" because	1632	year, it will still time out after (roughly) one hour. "Roughly" because
691	detecting time jumps is hard, and some inaccuracies are unavoidable (the	1633	detecting time jumps is hard, and some inaccuracies are unavoidable (the
692	monotonic clock option helps a lot here).	1634	monotonic clock option helps a lot here).
		1635
		1636	The callback is guaranteed to be invoked only I<after> its timeout has
		1637	passed (not I<at>, so on systems with very low-resolution clocks this
		1638	might introduce a small delay). If multiple timers become ready during the
		1639	same loop iteration then the ones with earlier time-out values are invoked
		1640	before ones of the same priority with later time-out values (but this is
		1641	no longer true when a callback calls C<ev_loop> recursively).
		1642
		1643	=head3 Be smart about timeouts
		1644
		1645	Many real-world problems involve some kind of timeout, usually for error
		1646	recovery. A typical example is an HTTP request - if the other side hangs,
		1647	you want to raise some error after a while.
		1648
		1649	What follows are some ways to handle this problem, from obvious and
		1650	inefficient to smart and efficient.
		1651
		1652	In the following, a 60 second activity timeout is assumed - a timeout that
		1653	gets reset to 60 seconds each time there is activity (e.g. each time some
		1654	data or other life sign was received).
		1655
		1656	=over 4
		1657
		1658	=item 1. Use a timer and stop, reinitialise and start it on activity.
		1659
		1660	This is the most obvious, but not the most simple way: In the beginning,
		1661	start the watcher:
		1662
		1663	ev_timer_init (timer, callback, 60., 0.);
		1664	ev_timer_start (loop, timer);
		1665
		1666	Then, each time there is some activity, C<ev_timer_stop> it, initialise it
		1667	and start it again:
		1668
		1669	ev_timer_stop (loop, timer);
		1670	ev_timer_set (timer, 60., 0.);
		1671	ev_timer_start (loop, timer);
		1672
		1673	This is relatively simple to implement, but means that each time there is
		1674	some activity, libev will first have to remove the timer from its internal
		1675	data structure and then add it again. Libev tries to be fast, but it's
		1676	still not a constant-time operation.
		1677
		1678	=item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
		1679
		1680	This is the easiest way, and involves using C<ev_timer_again> instead of
		1681	C<ev_timer_start>.
		1682
		1683	To implement this, configure an C<ev_timer> with a C<repeat> value
		1684	of C<60> and then call C<ev_timer_again> at start and each time you
		1685	successfully read or write some data. If you go into an idle state where
		1686	you do not expect data to travel on the socket, you can C<ev_timer_stop>
		1687	the timer, and C<ev_timer_again> will automatically restart it if need be.
		1688
		1689	That means you can ignore both the C<ev_timer_start> function and the
		1690	C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
		1691	member and C<ev_timer_again>.
		1692
		1693	At start:
		1694
		1695	ev_init (timer, callback);
		1696	timer->repeat = 60.;
		1697	ev_timer_again (loop, timer);
		1698
		1699	Each time there is some activity:
		1700
		1701	ev_timer_again (loop, timer);
		1702
		1703	It is even possible to change the time-out on the fly, regardless of
		1704	whether the watcher is active or not:
		1705
		1706	timer->repeat = 30.;
		1707	ev_timer_again (loop, timer);
		1708
		1709	This is slightly more efficient then stopping/starting the timer each time
		1710	you want to modify its timeout value, as libev does not have to completely
		1711	remove and re-insert the timer from/into its internal data structure.
		1712
		1713	It is, however, even simpler than the "obvious" way to do it.
		1714
		1715	=item 3. Let the timer time out, but then re-arm it as required.
		1716
		1717	This method is more tricky, but usually most efficient: Most timeouts are
		1718	relatively long compared to the intervals between other activity - in
		1719	our example, within 60 seconds, there are usually many I/O events with
		1720	associated activity resets.
		1721
		1722	In this case, it would be more efficient to leave the C<ev_timer> alone,
		1723	but remember the time of last activity, and check for a real timeout only
		1724	within the callback:
		1725
		1726	ev_tstamp last_activity; // time of last activity
		1727
		1728	static void
		1729	callback (EV_P_ ev_timer *w, int revents)
		1730	{
		1731	ev_tstamp now = ev_now (EV_A);
		1732	ev_tstamp timeout = last_activity + 60.;
		1733
		1734	// if last_activity + 60. is older than now, we did time out
		1735	if (timeout < now)
		1736	{
		1737	// timeout occured, take action
		1738	}
		1739	else
		1740	{
		1741	// callback was invoked, but there was some activity, re-arm
		1742	// the watcher to fire in last_activity + 60, which is
		1743	// guaranteed to be in the future, so "again" is positive:
		1744	w->repeat = timeout - now;
		1745	ev_timer_again (EV_A_ w);
		1746	}
		1747	}
		1748
		1749	To summarise the callback: first calculate the real timeout (defined
		1750	as "60 seconds after the last activity"), then check if that time has
		1751	been reached, which means something I<did>, in fact, time out. Otherwise
		1752	the callback was invoked too early (C<timeout> is in the future), so
		1753	re-schedule the timer to fire at that future time, to see if maybe we have
		1754	a timeout then.
		1755
		1756	Note how C<ev_timer_again> is used, taking advantage of the
		1757	C<ev_timer_again> optimisation when the timer is already running.
		1758
		1759	This scheme causes more callback invocations (about one every 60 seconds
		1760	minus half the average time between activity), but virtually no calls to
		1761	libev to change the timeout.
		1762
		1763	To start the timer, simply initialise the watcher and set C<last_activity>
		1764	to the current time (meaning we just have some activity :), then call the
		1765	callback, which will "do the right thing" and start the timer:
		1766
		1767	ev_init (timer, callback);
		1768	last_activity = ev_now (loop);
		1769	callback (loop, timer, EV_TIMEOUT);
		1770
		1771	And when there is some activity, simply store the current time in
		1772	C<last_activity>, no libev calls at all:
		1773
		1774	last_actiivty = ev_now (loop);
		1775
		1776	This technique is slightly more complex, but in most cases where the
		1777	time-out is unlikely to be triggered, much more efficient.
		1778
		1779	Changing the timeout is trivial as well (if it isn't hard-coded in the
		1780	callback :) - just change the timeout and invoke the callback, which will
		1781	fix things for you.
		1782
		1783	=item 4. Wee, just use a double-linked list for your timeouts.
		1784
		1785	If there is not one request, but many thousands (millions...), all
		1786	employing some kind of timeout with the same timeout value, then one can
		1787	do even better:
		1788
		1789	When starting the timeout, calculate the timeout value and put the timeout
		1790	at the I<end> of the list.
		1791
		1792	Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
		1793	the list is expected to fire (for example, using the technique #3).
		1794
		1795	When there is some activity, remove the timer from the list, recalculate
		1796	the timeout, append it to the end of the list again, and make sure to
		1797	update the C<ev_timer> if it was taken from the beginning of the list.
		1798
		1799	This way, one can manage an unlimited number of timeouts in O(1) time for
		1800	starting, stopping and updating the timers, at the expense of a major
		1801	complication, and having to use a constant timeout. The constant timeout
		1802	ensures that the list stays sorted.
		1803
		1804	=back
		1805
		1806	So which method the best?
		1807
		1808	Method #2 is a simple no-brain-required solution that is adequate in most
		1809	situations. Method #3 requires a bit more thinking, but handles many cases
		1810	better, and isn't very complicated either. In most case, choosing either
		1811	one is fine, with #3 being better in typical situations.
		1812
		1813	Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
		1814	rather complicated, but extremely efficient, something that really pays
		1815	off after the first million or so of active timers, i.e. it's usually
		1816	overkill :)
		1817
		1818	=head3 The special problem of time updates
		1819
		1820	Establishing the current time is a costly operation (it usually takes at
		1821	least two system calls): EV therefore updates its idea of the current
		1822	time only before and after C<ev_loop> collects new events, which causes a
		1823	growing difference between C<ev_now ()> and C<ev_time ()> when handling
		1824	lots of events in one iteration.
693		1825
694	The relative timeouts are calculated relative to the C<ev_now ()>	1826	The relative timeouts are calculated relative to the C<ev_now ()>
695	time. This is usually the right thing as this timestamp refers to the time	1827	time. This is usually the right thing as this timestamp refers to the time
696	of the event triggering whatever timeout you are modifying/starting. If	1828	of the event triggering whatever timeout you are modifying/starting. If
697	you suspect event processing to be delayed and you I<need> to base the timeout	1829	you suspect event processing to be delayed and you I<need> to base the
698	on the current time, use something like this to adjust for this:	1830	timeout on the current time, use something like this to adjust for this:
699		1831
700	ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);	1832	ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);
701		1833
702	The callback is guarenteed to be invoked only when its timeout has passed,	1834	If the event loop is suspended for a long time, you can also force an
703	but if multiple timers become ready during the same loop iteration then	1835	update of the time returned by C<ev_now ()> by calling C<ev_now_update
704	order of execution is undefined.	1836	()>.
		1837
		1838	=head3 The special problems of suspended animation
		1839
		1840	When you leave the server world it is quite customary to hit machines that
		1841	can suspend/hibernate - what happens to the clocks during such a suspend?
		1842
		1843	Some quick tests made with a Linux 2.6.28 indicate that a suspend freezes
		1844	all processes, while the clocks (C<times>, C<CLOCK_MONOTONIC>) continue
		1845	to run until the system is suspended, but they will not advance while the
		1846	system is suspended. That means, on resume, it will be as if the program
		1847	was frozen for a few seconds, but the suspend time will not be counted
		1848	towards C<ev_timer> when a monotonic clock source is used. The real time
		1849	clock advanced as expected, but if it is used as sole clocksource, then a
		1850	long suspend would be detected as a time jump by libev, and timers would
		1851	be adjusted accordingly.
		1852
		1853	I would not be surprised to see different behaviour in different between
		1854	operating systems, OS versions or even different hardware.
		1855
		1856	The other form of suspend (job control, or sending a SIGSTOP) will see a
		1857	time jump in the monotonic clocks and the realtime clock. If the program
		1858	is suspended for a very long time, and monotonic clock sources are in use,
		1859	then you can expect C<ev_timer>s to expire as the full suspension time
		1860	will be counted towards the timers. When no monotonic clock source is in
		1861	use, then libev will again assume a timejump and adjust accordingly.
		1862
		1863	It might be beneficial for this latter case to call C<ev_suspend>
		1864	and C<ev_resume> in code that handles C<SIGTSTP>, to at least get
		1865	deterministic behaviour in this case (you can do nothing against
		1866	C<SIGSTOP>).
		1867
		1868	=head3 Watcher-Specific Functions and Data Members
705		1869
706	=over 4	1870	=over 4
707		1871
708	=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)	1872	=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
709		1873
710	=item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)	1874	=item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
711		1875
712	Configure the timer to trigger after C<after> seconds. If C<repeat> is	1876	Configure the timer to trigger after C<after> seconds. If C<repeat>
713	C<0.>, then it will automatically be stopped. If it is positive, then the	1877	is C<0.>, then it will automatically be stopped once the timeout is
714	timer will automatically be configured to trigger again C<repeat> seconds	1878	reached. If it is positive, then the timer will automatically be
715	later, again, and again, until stopped manually.	1879	configured to trigger again C<repeat> seconds later, again, and again,
		1880	until stopped manually.
716		1881
717	The timer itself will do a best-effort at avoiding drift, that is, if you	1882	The timer itself will do a best-effort at avoiding drift, that is, if
718	configure a timer to trigger every 10 seconds, then it will trigger at	1883	you configure a timer to trigger every 10 seconds, then it will normally
719	exactly 10 second intervals. If, however, your program cannot keep up with	1884	trigger at exactly 10 second intervals. If, however, your program cannot
720	the timer (because it takes longer than those 10 seconds to do stuff) the	1885	keep up with the timer (because it takes longer than those 10 seconds to
721	timer will not fire more than once per event loop iteration.	1886	do stuff) the timer will not fire more than once per event loop iteration.
722		1887
723	=item ev_timer_again (loop)	1888	=item ev_timer_again (loop, ev_timer *)
724		1889
725	This will act as if the timer timed out and restart it again if it is	1890	This will act as if the timer timed out and restart it again if it is
726	repeating. The exact semantics are:	1891	repeating. The exact semantics are:
727		1892
		1893	If the timer is pending, its pending status is cleared.
		1894
728	If the timer is started but nonrepeating, stop it.	1895	If the timer is started but non-repeating, stop it (as if it timed out).
729		1896
730	If the timer is repeating, either start it if necessary (with the repeat	1897	If the timer is repeating, either start it if necessary (with the
731	value), or reset the running timer to the repeat value.	1898	C<repeat> value), or reset the running timer to the C<repeat> value.
732		1899
733	This sounds a bit complicated, but here is a useful and typical	1900	This sounds a bit complicated, see L<Be smart about timeouts>, above, for a
734	example: Imagine you have a tcp connection and you want a so-called idle	1901	usage example.
735	timeout, that is, you want to be called when there have been, say, 60	1902
736	seconds of inactivity on the socket. The easiest way to do this is to	1903	=item ev_tstamp ev_timer_remaining (loop, ev_timer *)
737	configure an C<ev_timer> with after=repeat=60 and calling ev_timer_again each	1904
738	time you successfully read or write some data. If you go into an idle	1905	Returns the remaining time until a timer fires. If the timer is active,
739	state where you do not expect data to travel on the socket, you can stop	1906	then this time is relative to the current event loop time, otherwise it's
740	the timer, and again will automatically restart it if need be.	1907	the timeout value currently configured.
		1908
		1909	That is, after an C<ev_timer_set (w, 5, 7)>, C<ev_timer_remaining> returns
		1910	C<5>. When the timer is started and one second passes, C<ev_timer_remaining>
		1911	will return C<4>. When the timer expires and is restarted, it will return
		1912	roughly C<7> (likely slightly less as callback invocation takes some time,
		1913	too), and so on.
		1914
		1915	=item ev_tstamp repeat [read-write]
		1916
		1917	The current C<repeat> value. Will be used each time the watcher times out
		1918	or C<ev_timer_again> is called, and determines the next timeout (if any),
		1919	which is also when any modifications are taken into account.
741		1920
742	=back	1921	=back
743		1922
		1923	=head3 Examples
		1924
744	Example: create a timer that fires after 60 seconds.	1925	Example: Create a timer that fires after 60 seconds.
745		1926
746	static void	1927	static void
747	one_minute_cb (struct ev_loop *loop, struct ev_timer *w, int revents)	1928	one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
748	{	1929	{
749	.. one minute over, w is actually stopped right here	1930	.. one minute over, w is actually stopped right here
750	}	1931	}
751		1932
752	struct ev_timer mytimer;	1933	ev_timer mytimer;
753	ev_timer_init (&mytimer, one_minute_cb, 60., 0.);	1934	ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
754	ev_timer_start (loop, &mytimer);	1935	ev_timer_start (loop, &mytimer);
755		1936
756	Example: create a timeout timer that times out after 10 seconds of	1937	Example: Create a timeout timer that times out after 10 seconds of
757	inactivity.	1938	inactivity.
758		1939
759	static void	1940	static void
760	timeout_cb (struct ev_loop *loop, struct ev_timer *w, int revents)	1941	timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
761	{	1942	{
762	.. ten seconds without any activity	1943	.. ten seconds without any activity
763	}	1944	}
764		1945
765	struct ev_timer mytimer;	1946	ev_timer mytimer;
766	ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */	1947	ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
767	ev_timer_again (&mytimer); /* start timer */	1948	ev_timer_again (&mytimer); /* start timer */
768	ev_loop (loop, 0);	1949	ev_loop (loop, 0);
769		1950
770	// and in some piece of code that gets executed on any "activity":	1951	// and in some piece of code that gets executed on any "activity":
771	// reset the timeout to start ticking again at 10 seconds	1952	// reset the timeout to start ticking again at 10 seconds
772	ev_timer_again (&mytimer);	1953	ev_timer_again (&mytimer);
773		1954
774		1955
775	=head2 C<ev_periodic> - to cron or not to cron	1956	=head2 C<ev_periodic> - to cron or not to cron?
776		1957
777	Periodic watchers are also timers of a kind, but they are very versatile	1958	Periodic watchers are also timers of a kind, but they are very versatile
778	(and unfortunately a bit complex).	1959	(and unfortunately a bit complex).
779		1960
780	Unlike C<ev_timer>'s, they are not based on real time (or relative time)	1961	Unlike C<ev_timer>, periodic watchers are not based on real time (or
781	but on wallclock time (absolute time). You can tell a periodic watcher	1962	relative time, the physical time that passes) but on wall clock time
782	to trigger "at" some specific point in time. For example, if you tell a	1963	(absolute time, the thing you can read on your calender or clock). The
783	periodic watcher to trigger in 10 seconds (by specifiying e.g. c<ev_now ()	1964	difference is that wall clock time can run faster or slower than real
784	+ 10.>) and then reset your system clock to the last year, then it will	1965	time, and time jumps are not uncommon (e.g. when you adjust your
785	take a year to trigger the event (unlike an C<ev_timer>, which would trigger	1966	wrist-watch).
786	roughly 10 seconds later and of course not if you reset your system time
787	again).
788		1967
789	They can also be used to implement vastly more complex timers, such as	1968	You can tell a periodic watcher to trigger after some specific point
		1969	in time: for example, if you tell a periodic watcher to trigger "in 10
		1970	seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
		1971	not a delay) and then reset your system clock to January of the previous
		1972	year, then it will take a year or more to trigger the event (unlike an
		1973	C<ev_timer>, which would still trigger roughly 10 seconds after starting
		1974	it, as it uses a relative timeout).
		1975
		1976	C<ev_periodic> watchers can also be used to implement vastly more complex
790	triggering an event on eahc midnight, local time.	1977	timers, such as triggering an event on each "midnight, local time", or
		1978	other complicated rules. This cannot be done with C<ev_timer> watchers, as
		1979	those cannot react to time jumps.
791		1980
792	As with timers, the callback is guarenteed to be invoked only when the	1981	As with timers, the callback is guaranteed to be invoked only when the
793	time (C<at>) has been passed, but if multiple periodic timers become ready	1982	point in time where it is supposed to trigger has passed. If multiple
794	during the same loop iteration then order of execution is undefined.	1983	timers become ready during the same loop iteration then the ones with
		1984	earlier time-out values are invoked before ones with later time-out values
		1985	(but this is no longer true when a callback calls C<ev_loop> recursively).
		1986
		1987	=head3 Watcher-Specific Functions and Data Members
795		1988
796	=over 4	1989	=over 4
797		1990
798	=item ev_periodic_init (ev_periodic *, callback, ev_tstamp at, ev_tstamp interval, reschedule_cb)	1991	=item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
799		1992
800	=item ev_periodic_set (ev_periodic *, ev_tstamp after, ev_tstamp repeat, reschedule_cb)	1993	=item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
801		1994
802	Lots of arguments, lets sort it out... There are basically three modes of	1995	Lots of arguments, let's sort it out... There are basically three modes of
803	operation, and we will explain them from simplest to complex:	1996	operation, and we will explain them from simplest to most complex:
804		1997
805	=over 4	1998	=over 4
806		1999
807	=item * absolute timer (interval = reschedule_cb = 0)	2000	=item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
808		2001
809	In this configuration the watcher triggers an event at the wallclock time	2002	In this configuration the watcher triggers an event after the wall clock
810	C<at> and doesn't repeat. It will not adjust when a time jump occurs,	2003	time C<offset> has passed. It will not repeat and will not adjust when a
811	that is, if it is to be run at January 1st 2011 then it will run when the	2004	time jump occurs, that is, if it is to be run at January 1st 2011 then it
812	system time reaches or surpasses this time.	2005	will be stopped and invoked when the system clock reaches or surpasses
		2006	this point in time.
813		2007
814	=item * non-repeating interval timer (interval > 0, reschedule_cb = 0)	2008	=item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
815		2009
816	In this mode the watcher will always be scheduled to time out at the next	2010	In this mode the watcher will always be scheduled to time out at the next
817	C<at + N * interval> time (for some integer N) and then repeat, regardless	2011	C<offset + N * interval> time (for some integer N, which can also be
818	of any time jumps.	2012	negative) and then repeat, regardless of any time jumps. The C<offset>
		2013	argument is merely an offset into the C<interval> periods.
819		2014
820	This can be used to create timers that do not drift with respect to system	2015	This can be used to create timers that do not drift with respect to the
821	time:	2016	system clock, for example, here is an C<ev_periodic> that triggers each
		2017	hour, on the hour (with respect to UTC):
822		2018
823	ev_periodic_set (&periodic, 0., 3600., 0);	2019	ev_periodic_set (&periodic, 0., 3600., 0);
824		2020
825	This doesn't mean there will always be 3600 seconds in between triggers,	2021	This doesn't mean there will always be 3600 seconds in between triggers,
826	but only that the the callback will be called when the system time shows a	2022	but only that the callback will be called when the system time shows a
827	full hour (UTC), or more correctly, when the system time is evenly divisible	2023	full hour (UTC), or more correctly, when the system time is evenly divisible
828	by 3600.	2024	by 3600.
829		2025
830	Another way to think about it (for the mathematically inclined) is that	2026	Another way to think about it (for the mathematically inclined) is that
831	C<ev_periodic> will try to run the callback in this mode at the next possible	2027	C<ev_periodic> will try to run the callback in this mode at the next possible
832	time where C<time = at (mod interval)>, regardless of any time jumps.	2028	time where C<time = offset (mod interval)>, regardless of any time jumps.
833		2029
		2030	For numerical stability it is preferable that the C<offset> value is near
		2031	C<ev_now ()> (the current time), but there is no range requirement for
		2032	this value, and in fact is often specified as zero.
		2033
		2034	Note also that there is an upper limit to how often a timer can fire (CPU
		2035	speed for example), so if C<interval> is very small then timing stability
		2036	will of course deteriorate. Libev itself tries to be exact to be about one
		2037	millisecond (if the OS supports it and the machine is fast enough).
		2038
834	=item * manual reschedule mode (reschedule_cb = callback)	2039	=item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
835		2040
836	In this mode the values for C<interval> and C<at> are both being	2041	In this mode the values for C<interval> and C<offset> are both being
837	ignored. Instead, each time the periodic watcher gets scheduled, the	2042	ignored. Instead, each time the periodic watcher gets scheduled, the
838	reschedule callback will be called with the watcher as first, and the	2043	reschedule callback will be called with the watcher as first, and the
839	current time as second argument.	2044	current time as second argument.
840		2045
841	NOTE: I<This callback MUST NOT stop or destroy any periodic watcher,	2046	NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, ever,
842	ever, or make any event loop modifications>. If you need to stop it,	2047	or make ANY other event loop modifications whatsoever, unless explicitly
843	return C<now + 1e30> (or so, fudge fudge) and stop it afterwards (e.g. by	2048	allowed by documentation here>.
844	starting a prepare watcher).
845		2049
		2050	If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
		2051	it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
		2052	only event loop modification you are allowed to do).
		2053
846	Its prototype is C<ev_tstamp (*reschedule_cb)(struct ev_periodic *w,	2054	The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
847	ev_tstamp now)>, e.g.:	2055	*w, ev_tstamp now)>, e.g.:
848		2056
		2057	static ev_tstamp
849	static ev_tstamp my_rescheduler (struct ev_periodic *w, ev_tstamp now)	2058	my_rescheduler (ev_periodic *w, ev_tstamp now)
850	{	2059	{
851	return now + 60.;	2060	return now + 60.;
852	}	2061	}
853		2062
854	It must return the next time to trigger, based on the passed time value	2063	It must return the next time to trigger, based on the passed time value
855	(that is, the lowest time value larger than to the second argument). It	2064	(that is, the lowest time value larger than to the second argument). It
856	will usually be called just before the callback will be triggered, but	2065	will usually be called just before the callback will be triggered, but
857	might be called at other times, too.	2066	might be called at other times, too.
858		2067
859	NOTE: I<< This callback must always return a time that is later than the	2068	NOTE: I<< This callback must always return a time that is higher than or
860	passed C<now> value >>. Not even C<now> itself will do, it I<must> be larger.	2069	equal to the passed C<now> value >>.
861		2070
862	This can be used to create very complex timers, such as a timer that	2071	This can be used to create very complex timers, such as a timer that
863	triggers on each midnight, local time. To do this, you would calculate the	2072	triggers on "next midnight, local time". To do this, you would calculate the
864	next midnight after C<now> and return the timestamp value for this. How	2073	next midnight after C<now> and return the timestamp value for this. How
865	you do this is, again, up to you (but it is not trivial, which is the main	2074	you do this is, again, up to you (but it is not trivial, which is the main
866	reason I omitted it as an example).	2075	reason I omitted it as an example).
867		2076
868	=back	2077	=back
…		…
872	Simply stops and restarts the periodic watcher again. This is only useful	2081	Simply stops and restarts the periodic watcher again. This is only useful
873	when you changed some parameters or the reschedule callback would return	2082	when you changed some parameters or the reschedule callback would return
874	a different time than the last time it was called (e.g. in a crond like	2083	a different time than the last time it was called (e.g. in a crond like
875	program when the crontabs have changed).	2084	program when the crontabs have changed).
876		2085
		2086	=item ev_tstamp ev_periodic_at (ev_periodic *)
		2087
		2088	When active, returns the absolute time that the watcher is supposed
		2089	to trigger next. This is not the same as the C<offset> argument to
		2090	C<ev_periodic_set>, but indeed works even in interval and manual
		2091	rescheduling modes.
		2092
		2093	=item ev_tstamp offset [read-write]
		2094
		2095	When repeating, this contains the offset value, otherwise this is the
		2096	absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
		2097	although libev might modify this value for better numerical stability).
		2098
		2099	Can be modified any time, but changes only take effect when the periodic
		2100	timer fires or C<ev_periodic_again> is being called.
		2101
		2102	=item ev_tstamp interval [read-write]
		2103
		2104	The current interval value. Can be modified any time, but changes only
		2105	take effect when the periodic timer fires or C<ev_periodic_again> is being
		2106	called.
		2107
		2108	=item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
		2109
		2110	The current reschedule callback, or C<0>, if this functionality is
		2111	switched off. Can be changed any time, but changes only take effect when
		2112	the periodic timer fires or C<ev_periodic_again> is being called.
		2113
877	=back	2114	=back
878		2115
		2116	=head3 Examples
		2117
879	Example: call a callback every hour, or, more precisely, whenever the	2118	Example: Call a callback every hour, or, more precisely, whenever the
880	system clock is divisible by 3600. The callback invocation times have	2119	system time is divisible by 3600. The callback invocation times have
881	potentially a lot of jittering, but good long-term stability.	2120	potentially a lot of jitter, but good long-term stability.
882		2121
883	static void	2122	static void
884	clock_cb (struct ev_loop *loop, struct ev_io *w, int revents)	2123	clock_cb (struct ev_loop *loop, ev_io *w, int revents)
885	{	2124	{
886	... its now a full hour (UTC, or TAI or whatever your clock follows)	2125	... its now a full hour (UTC, or TAI or whatever your clock follows)
887	}	2126	}
888		2127
889	struct ev_periodic hourly_tick;	2128	ev_periodic hourly_tick;
890	ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);	2129	ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
891	ev_periodic_start (loop, &hourly_tick);	2130	ev_periodic_start (loop, &hourly_tick);
892		2131
893	Example: the same as above, but use a reschedule callback to do it:	2132	Example: The same as above, but use a reschedule callback to do it:
894		2133
895	#include <math.h>	2134	#include <math.h>
896		2135
897	static ev_tstamp	2136	static ev_tstamp
898	my_scheduler_cb (struct ev_periodic *w, ev_tstamp now)	2137	my_scheduler_cb (ev_periodic *w, ev_tstamp now)
899	{	2138	{
900	return fmod (now, 3600.) + 3600.;	2139	return now + (3600. - fmod (now, 3600.));
901	}	2140	}
902		2141
903	ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);	2142	ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
904		2143
905	Example: call a callback every hour, starting now:	2144	Example: Call a callback every hour, starting now:
906		2145
907	struct ev_periodic hourly_tick;	2146	ev_periodic hourly_tick;
908	ev_periodic_init (&hourly_tick, clock_cb,	2147	ev_periodic_init (&hourly_tick, clock_cb,
909	fmod (ev_now (loop), 3600.), 3600., 0);	2148	fmod (ev_now (loop), 3600.), 3600., 0);
910	ev_periodic_start (loop, &hourly_tick);	2149	ev_periodic_start (loop, &hourly_tick);
911		2150
912		2151
913	=head2 C<ev_signal> - signal me when a signal gets signalled	2152	=head2 C<ev_signal> - signal me when a signal gets signalled!
914		2153
915	Signal watchers will trigger an event when the process receives a specific	2154	Signal watchers will trigger an event when the process receives a specific
916	signal one or more times. Even though signals are very asynchronous, libev	2155	signal one or more times. Even though signals are very asynchronous, libev
917	will try it's best to deliver signals synchronously, i.e. as part of the	2156	will try it's best to deliver signals synchronously, i.e. as part of the
918	normal event processing, like any other event.	2157	normal event processing, like any other event.
919		2158
		2159	If you want signals to be delivered truly asynchronously, just use
		2160	C<sigaction> as you would do without libev and forget about sharing
		2161	the signal. You can even use C<ev_async> from a signal handler to
		2162	synchronously wake up an event loop.
		2163
920	You can configure as many watchers as you like per signal. Only when the	2164	You can configure as many watchers as you like for the same signal, but
		2165	only within the same loop, i.e. you can watch for C<SIGINT> in your
		2166	default loop and for C<SIGIO> in another loop, but you cannot watch for
		2167	C<SIGINT> in both the default loop and another loop at the same time. At
		2168	the moment, C<SIGCHLD> is permanently tied to the default loop.
		2169
921	first watcher gets started will libev actually register a signal watcher	2170	When the first watcher gets started will libev actually register something
922	with the kernel (thus it coexists with your own signal handlers as long	2171	with the kernel (thus it coexists with your own signal handlers as long as
923	as you don't register any with libev). Similarly, when the last signal	2172	you don't register any with libev for the same signal).
924	watcher for a signal is stopped libev will reset the signal handler to	2173
925	SIG_DFL (regardless of what it was set to before).	2174	If possible and supported, libev will install its handlers with
		2175	C<SA_RESTART> (or equivalent) behaviour enabled, so system calls should
		2176	not be unduly interrupted. If you have a problem with system calls getting
		2177	interrupted by signals you can block all signals in an C<ev_check> watcher
		2178	and unblock them in an C<ev_prepare> watcher.
		2179
		2180	=head3 The special problem of inheritance over fork/execve/pthread_create
		2181
		2182	Both the signal mask (C<sigprocmask>) and the signal disposition
		2183	(C<sigaction>) are unspecified after starting a signal watcher (and after
		2184	stopping it again), that is, libev might or might not block the signal,
		2185	and might or might not set or restore the installed signal handler.
		2186
		2187	While this does not matter for the signal disposition (libev never
		2188	sets signals to C<SIG_IGN>, so handlers will be reset to C<SIG_DFL> on
		2189	C<execve>), this matters for the signal mask: many programs do not expect
		2190	certain signals to be blocked.
		2191
		2192	This means that before calling C<exec> (from the child) you should reset
		2193	the signal mask to whatever "default" you expect (all clear is a good
		2194	choice usually).
		2195
		2196	The simplest way to ensure that the signal mask is reset in the child is
		2197	to install a fork handler with C<pthread_atfork> that resets it. That will
		2198	catch fork calls done by libraries (such as the libc) as well.
		2199
		2200	In current versions of libev, the signal will not be blocked indefinitely
		2201	unless you use the C<signalfd> API (C<EV_SIGNALFD>). While this reduces
		2202	the window of opportunity for problems, it will not go away, as libev
		2203	I<has> to modify the signal mask, at least temporarily.
		2204
		2205	So I can't stress this enough: I<If you do not reset your signal mask when
		2206	you expect it to be empty, you have a race condition in your code>. This
		2207	is not a libev-specific thing, this is true for most event libraries.
		2208
		2209	=head3 Watcher-Specific Functions and Data Members
926		2210
927	=over 4	2211	=over 4
928		2212
929	=item ev_signal_init (ev_signal *, callback, int signum)	2213	=item ev_signal_init (ev_signal *, callback, int signum)
930		2214
931	=item ev_signal_set (ev_signal *, int signum)	2215	=item ev_signal_set (ev_signal *, int signum)
932		2216
933	Configures the watcher to trigger on the given signal number (usually one	2217	Configures the watcher to trigger on the given signal number (usually one
934	of the C<SIGxxx> constants).	2218	of the C<SIGxxx> constants).
935		2219
		2220	=item int signum [read-only]
		2221
		2222	The signal the watcher watches out for.
		2223
936	=back	2224	=back
937		2225
		2226	=head3 Examples
938		2227
		2228	Example: Try to exit cleanly on SIGINT.
		2229
		2230	static void
		2231	sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
		2232	{
		2233	ev_unloop (loop, EVUNLOOP_ALL);
		2234	}
		2235
		2236	ev_signal signal_watcher;
		2237	ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
		2238	ev_signal_start (loop, &signal_watcher);
		2239
		2240
939	=head2 C<ev_child> - wait for pid status changes	2241	=head2 C<ev_child> - watch out for process status changes
940		2242
941	Child watchers trigger when your process receives a SIGCHLD in response to	2243	Child watchers trigger when your process receives a SIGCHLD in response to
942	some child status changes (most typically when a child of yours dies).	2244	some child status changes (most typically when a child of yours dies or
		2245	exits). It is permissible to install a child watcher I<after> the child
		2246	has been forked (which implies it might have already exited), as long
		2247	as the event loop isn't entered (or is continued from a watcher), i.e.,
		2248	forking and then immediately registering a watcher for the child is fine,
		2249	but forking and registering a watcher a few event loop iterations later or
		2250	in the next callback invocation is not.
		2251
		2252	Only the default event loop is capable of handling signals, and therefore
		2253	you can only register child watchers in the default event loop.
		2254
		2255	Due to some design glitches inside libev, child watchers will always be
		2256	handled at maximum priority (their priority is set to C<EV_MAXPRI> by
		2257	libev)
		2258
		2259	=head3 Process Interaction
		2260
		2261	Libev grabs C<SIGCHLD> as soon as the default event loop is
		2262	initialised. This is necessary to guarantee proper behaviour even if the
		2263	first child watcher is started after the child exits. The occurrence
		2264	of C<SIGCHLD> is recorded asynchronously, but child reaping is done
		2265	synchronously as part of the event loop processing. Libev always reaps all
		2266	children, even ones not watched.
		2267
		2268	=head3 Overriding the Built-In Processing
		2269
		2270	Libev offers no special support for overriding the built-in child
		2271	processing, but if your application collides with libev's default child
		2272	handler, you can override it easily by installing your own handler for
		2273	C<SIGCHLD> after initialising the default loop, and making sure the
		2274	default loop never gets destroyed. You are encouraged, however, to use an
		2275	event-based approach to child reaping and thus use libev's support for
		2276	that, so other libev users can use C<ev_child> watchers freely.
		2277
		2278	=head3 Stopping the Child Watcher
		2279
		2280	Currently, the child watcher never gets stopped, even when the
		2281	child terminates, so normally one needs to stop the watcher in the
		2282	callback. Future versions of libev might stop the watcher automatically
		2283	when a child exit is detected (calling C<ev_child_stop> twice is not a
		2284	problem).
		2285
		2286	=head3 Watcher-Specific Functions and Data Members
943		2287
944	=over 4	2288	=over 4
945		2289
946	=item ev_child_init (ev_child *, callback, int pid)	2290	=item ev_child_init (ev_child *, callback, int pid, int trace)
947		2291
948	=item ev_child_set (ev_child *, int pid)	2292	=item ev_child_set (ev_child *, int pid, int trace)
949		2293
950	Configures the watcher to wait for status changes of process C<pid> (or	2294	Configures the watcher to wait for status changes of process C<pid> (or
951	I<any> process if C<pid> is specified as C<0>). The callback can look	2295	I<any> process if C<pid> is specified as C<0>). The callback can look
952	at the C<rstatus> member of the C<ev_child> watcher structure to see	2296	at the C<rstatus> member of the C<ev_child> watcher structure to see
953	the status word (use the macros from C<sys/wait.h> and see your systems	2297	the status word (use the macros from C<sys/wait.h> and see your systems
954	C<waitpid> documentation). The C<rpid> member contains the pid of the	2298	C<waitpid> documentation). The C<rpid> member contains the pid of the
955	process causing the status change.	2299	process causing the status change. C<trace> must be either C<0> (only
		2300	activate the watcher when the process terminates) or C<1> (additionally
		2301	activate the watcher when the process is stopped or continued).
		2302
		2303	=item int pid [read-only]
		2304
		2305	The process id this watcher watches out for, or C<0>, meaning any process id.
		2306
		2307	=item int rpid [read-write]
		2308
		2309	The process id that detected a status change.
		2310
		2311	=item int rstatus [read-write]
		2312
		2313	The process exit/trace status caused by C<rpid> (see your systems
		2314	C<waitpid> and C<sys/wait.h> documentation for details).
956		2315
957	=back	2316	=back
958		2317
959	Example: try to exit cleanly on SIGINT and SIGTERM.	2318	=head3 Examples
960		2319
		2320	Example: C<fork()> a new process and install a child handler to wait for
		2321	its completion.
		2322
		2323	ev_child cw;
		2324
961	static void	2325	static void
962	sigint_cb (struct ev_loop *loop, struct ev_signal *w, int revents)	2326	child_cb (EV_P_ ev_child *w, int revents)
963	{	2327	{
964	ev_unloop (loop, EVUNLOOP_ALL);	2328	ev_child_stop (EV_A_ w);
		2329	printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
965	}	2330	}
966		2331
967	struct ev_signal signal_watcher;	2332	pid_t pid = fork ();
968	ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
969	ev_signal_start (loop, &sigint_cb);
970		2333
		2334	if (pid < 0)
		2335	// error
		2336	else if (pid == 0)
		2337	{
		2338	// the forked child executes here
		2339	exit (1);
		2340	}
		2341	else
		2342	{
		2343	ev_child_init (&cw, child_cb, pid, 0);
		2344	ev_child_start (EV_DEFAULT_ &cw);
		2345	}
971		2346
		2347
		2348	=head2 C<ev_stat> - did the file attributes just change?
		2349
		2350	This watches a file system path for attribute changes. That is, it calls
		2351	C<stat> on that path in regular intervals (or when the OS says it changed)
		2352	and sees if it changed compared to the last time, invoking the callback if
		2353	it did.
		2354
		2355	The path does not need to exist: changing from "path exists" to "path does
		2356	not exist" is a status change like any other. The condition "path does not
		2357	exist" (or more correctly "path cannot be stat'ed") is signified by the
		2358	C<st_nlink> field being zero (which is otherwise always forced to be at
		2359	least one) and all the other fields of the stat buffer having unspecified
		2360	contents.
		2361
		2362	The path I<must not> end in a slash or contain special components such as
		2363	C<.> or C<..>. The path I<should> be absolute: If it is relative and
		2364	your working directory changes, then the behaviour is undefined.
		2365
		2366	Since there is no portable change notification interface available, the
		2367	portable implementation simply calls C<stat(2)> regularly on the path
		2368	to see if it changed somehow. You can specify a recommended polling
		2369	interval for this case. If you specify a polling interval of C<0> (highly
		2370	recommended!) then a I<suitable, unspecified default> value will be used
		2371	(which you can expect to be around five seconds, although this might
		2372	change dynamically). Libev will also impose a minimum interval which is
		2373	currently around C<0.1>, but that's usually overkill.
		2374
		2375	This watcher type is not meant for massive numbers of stat watchers,
		2376	as even with OS-supported change notifications, this can be
		2377	resource-intensive.
		2378
		2379	At the time of this writing, the only OS-specific interface implemented
		2380	is the Linux inotify interface (implementing kqueue support is left as an
		2381	exercise for the reader. Note, however, that the author sees no way of
		2382	implementing C<ev_stat> semantics with kqueue, except as a hint).
		2383
		2384	=head3 ABI Issues (Largefile Support)
		2385
		2386	Libev by default (unless the user overrides this) uses the default
		2387	compilation environment, which means that on systems with large file
		2388	support disabled by default, you get the 32 bit version of the stat
		2389	structure. When using the library from programs that change the ABI to
		2390	use 64 bit file offsets the programs will fail. In that case you have to
		2391	compile libev with the same flags to get binary compatibility. This is
		2392	obviously the case with any flags that change the ABI, but the problem is
		2393	most noticeably displayed with ev_stat and large file support.
		2394
		2395	The solution for this is to lobby your distribution maker to make large
		2396	file interfaces available by default (as e.g. FreeBSD does) and not
		2397	optional. Libev cannot simply switch on large file support because it has
		2398	to exchange stat structures with application programs compiled using the
		2399	default compilation environment.
		2400
		2401	=head3 Inotify and Kqueue
		2402
		2403	When C<inotify (7)> support has been compiled into libev and present at
		2404	runtime, it will be used to speed up change detection where possible. The
		2405	inotify descriptor will be created lazily when the first C<ev_stat>
		2406	watcher is being started.
		2407
		2408	Inotify presence does not change the semantics of C<ev_stat> watchers
		2409	except that changes might be detected earlier, and in some cases, to avoid
		2410	making regular C<stat> calls. Even in the presence of inotify support
		2411	there are many cases where libev has to resort to regular C<stat> polling,
		2412	but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
		2413	many bugs), the path exists (i.e. stat succeeds), and the path resides on
		2414	a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
		2415	xfs are fully working) libev usually gets away without polling.
		2416
		2417	There is no support for kqueue, as apparently it cannot be used to
		2418	implement this functionality, due to the requirement of having a file
		2419	descriptor open on the object at all times, and detecting renames, unlinks
		2420	etc. is difficult.
		2421
		2422	=head3 C<stat ()> is a synchronous operation
		2423
		2424	Libev doesn't normally do any kind of I/O itself, and so is not blocking
		2425	the process. The exception are C<ev_stat> watchers - those call C<stat
		2426	()>, which is a synchronous operation.
		2427
		2428	For local paths, this usually doesn't matter: unless the system is very
		2429	busy or the intervals between stat's are large, a stat call will be fast,
		2430	as the path data is usually in memory already (except when starting the
		2431	watcher).
		2432
		2433	For networked file systems, calling C<stat ()> can block an indefinite
		2434	time due to network issues, and even under good conditions, a stat call
		2435	often takes multiple milliseconds.
		2436
		2437	Therefore, it is best to avoid using C<ev_stat> watchers on networked
		2438	paths, although this is fully supported by libev.
		2439
		2440	=head3 The special problem of stat time resolution
		2441
		2442	The C<stat ()> system call only supports full-second resolution portably,
		2443	and even on systems where the resolution is higher, most file systems
		2444	still only support whole seconds.
		2445
		2446	That means that, if the time is the only thing that changes, you can
		2447	easily miss updates: on the first update, C<ev_stat> detects a change and
		2448	calls your callback, which does something. When there is another update
		2449	within the same second, C<ev_stat> will be unable to detect unless the
		2450	stat data does change in other ways (e.g. file size).
		2451
		2452	The solution to this is to delay acting on a change for slightly more
		2453	than a second (or till slightly after the next full second boundary), using
		2454	a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
		2455	ev_timer_again (loop, w)>).
		2456
		2457	The C<.02> offset is added to work around small timing inconsistencies
		2458	of some operating systems (where the second counter of the current time
		2459	might be be delayed. One such system is the Linux kernel, where a call to
		2460	C<gettimeofday> might return a timestamp with a full second later than
		2461	a subsequent C<time> call - if the equivalent of C<time ()> is used to
		2462	update file times then there will be a small window where the kernel uses
		2463	the previous second to update file times but libev might already execute
		2464	the timer callback).
		2465
		2466	=head3 Watcher-Specific Functions and Data Members
		2467
		2468	=over 4
		2469
		2470	=item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
		2471
		2472	=item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
		2473
		2474	Configures the watcher to wait for status changes of the given
		2475	C<path>. The C<interval> is a hint on how quickly a change is expected to
		2476	be detected and should normally be specified as C<0> to let libev choose
		2477	a suitable value. The memory pointed to by C<path> must point to the same
		2478	path for as long as the watcher is active.
		2479
		2480	The callback will receive an C<EV_STAT> event when a change was detected,
		2481	relative to the attributes at the time the watcher was started (or the
		2482	last change was detected).
		2483
		2484	=item ev_stat_stat (loop, ev_stat *)
		2485
		2486	Updates the stat buffer immediately with new values. If you change the
		2487	watched path in your callback, you could call this function to avoid
		2488	detecting this change (while introducing a race condition if you are not
		2489	the only one changing the path). Can also be useful simply to find out the
		2490	new values.
		2491
		2492	=item ev_statdata attr [read-only]
		2493
		2494	The most-recently detected attributes of the file. Although the type is
		2495	C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
		2496	suitable for your system, but you can only rely on the POSIX-standardised
		2497	members to be present. If the C<st_nlink> member is C<0>, then there was
		2498	some error while C<stat>ing the file.
		2499
		2500	=item ev_statdata prev [read-only]
		2501
		2502	The previous attributes of the file. The callback gets invoked whenever
		2503	C<prev> != C<attr>, or, more precisely, one or more of these members
		2504	differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
		2505	C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
		2506
		2507	=item ev_tstamp interval [read-only]
		2508
		2509	The specified interval.
		2510
		2511	=item const char *path [read-only]
		2512
		2513	The file system path that is being watched.
		2514
		2515	=back
		2516
		2517	=head3 Examples
		2518
		2519	Example: Watch C</etc/passwd> for attribute changes.
		2520
		2521	static void
		2522	passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
		2523	{
		2524	/* /etc/passwd changed in some way */
		2525	if (w->attr.st_nlink)
		2526	{
		2527	printf ("passwd current size %ld\n", (long)w->attr.st_size);
		2528	printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
		2529	printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
		2530	}
		2531	else
		2532	/* you shalt not abuse printf for puts */
		2533	puts ("wow, /etc/passwd is not there, expect problems. "
		2534	"if this is windows, they already arrived\n");
		2535	}
		2536
		2537	...
		2538	ev_stat passwd;
		2539
		2540	ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
		2541	ev_stat_start (loop, &passwd);
		2542
		2543	Example: Like above, but additionally use a one-second delay so we do not
		2544	miss updates (however, frequent updates will delay processing, too, so
		2545	one might do the work both on C<ev_stat> callback invocation I<and> on
		2546	C<ev_timer> callback invocation).
		2547
		2548	static ev_stat passwd;
		2549	static ev_timer timer;
		2550
		2551	static void
		2552	timer_cb (EV_P_ ev_timer *w, int revents)
		2553	{
		2554	ev_timer_stop (EV_A_ w);
		2555
		2556	/* now it's one second after the most recent passwd change */
		2557	}
		2558
		2559	static void
		2560	stat_cb (EV_P_ ev_stat *w, int revents)
		2561	{
		2562	/* reset the one-second timer */
		2563	ev_timer_again (EV_A_ &timer);
		2564	}
		2565
		2566	...
		2567	ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
		2568	ev_stat_start (loop, &passwd);
		2569	ev_timer_init (&timer, timer_cb, 0., 1.02);
		2570
		2571
972	=head2 C<ev_idle> - when you've got nothing better to do	2572	=head2 C<ev_idle> - when you've got nothing better to do...
973		2573
974	Idle watchers trigger events when there are no other events are pending	2574	Idle watchers trigger events when no other events of the same or higher
975	(prepare, check and other idle watchers do not count). That is, as long	2575	priority are pending (prepare, check and other idle watchers do not count
976	as your process is busy handling sockets or timeouts (or even signals,	2576	as receiving "events").
977	imagine) it will not be triggered. But when your process is idle all idle	2577
978	watchers are being called again and again, once per event loop iteration -	2578	That is, as long as your process is busy handling sockets or timeouts
		2579	(or even signals, imagine) of the same or higher priority it will not be
		2580	triggered. But when your process is idle (or only lower-priority watchers
		2581	are pending), the idle watchers are being called once per event loop
979	until stopped, that is, or your process receives more events and becomes	2582	iteration - until stopped, that is, or your process receives more events
980	busy.	2583	and becomes busy again with higher priority stuff.
981		2584
982	The most noteworthy effect is that as long as any idle watchers are	2585	The most noteworthy effect is that as long as any idle watchers are
983	active, the process will not block when waiting for new events.	2586	active, the process will not block when waiting for new events.
984		2587
985	Apart from keeping your process non-blocking (which is a useful	2588	Apart from keeping your process non-blocking (which is a useful
986	effect on its own sometimes), idle watchers are a good place to do	2589	effect on its own sometimes), idle watchers are a good place to do
987	"pseudo-background processing", or delay processing stuff to after the	2590	"pseudo-background processing", or delay processing stuff to after the
988	event loop has handled all outstanding events.	2591	event loop has handled all outstanding events.
989		2592
		2593	=head3 Watcher-Specific Functions and Data Members
		2594
990	=over 4	2595	=over 4
991		2596
992	=item ev_idle_init (ev_signal *, callback)	2597	=item ev_idle_init (ev_idle *, callback)
993		2598
994	Initialises and configures the idle watcher - it has no parameters of any	2599	Initialises and configures the idle watcher - it has no parameters of any
995	kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,	2600	kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
996	believe me.	2601	believe me.
997		2602
998	=back	2603	=back
999		2604
		2605	=head3 Examples
		2606
1000	Example: dynamically allocate an C<ev_idle>, start it, and in the	2607	Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
1001	callback, free it. Alos, use no error checking, as usual.	2608	callback, free it. Also, use no error checking, as usual.
1002		2609
1003	static void	2610	static void
1004	idle_cb (struct ev_loop *loop, struct ev_idle *w, int revents)	2611	idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
1005	{	2612	{
1006	free (w);	2613	free (w);
1007	// now do something you wanted to do when the program has	2614	// now do something you wanted to do when the program has
1008	// no longer asnything immediate to do.	2615	// no longer anything immediate to do.
1009	}	2616	}
1010		2617
1011	struct ev_idle *idle_watcher = malloc (sizeof (struct ev_idle));	2618	ev_idle *idle_watcher = malloc (sizeof (ev_idle));
1012	ev_idle_init (idle_watcher, idle_cb);	2619	ev_idle_init (idle_watcher, idle_cb);
1013	ev_idle_start (loop, idle_cb);	2620	ev_idle_start (loop, idle_watcher);
1014		2621
1015		2622
1016	=head2 C<ev_prepare> and C<ev_check> - customise your event loop	2623	=head2 C<ev_prepare> and C<ev_check> - customise your event loop!
1017		2624
1018	Prepare and check watchers are usually (but not always) used in tandem:	2625	Prepare and check watchers are usually (but not always) used in pairs:
1019	prepare watchers get invoked before the process blocks and check watchers	2626	prepare watchers get invoked before the process blocks and check watchers
1020	afterwards.	2627	afterwards.
1021		2628
		2629	You I<must not> call C<ev_loop> or similar functions that enter
		2630	the current event loop from either C<ev_prepare> or C<ev_check>
		2631	watchers. Other loops than the current one are fine, however. The
		2632	rationale behind this is that you do not need to check for recursion in
		2633	those watchers, i.e. the sequence will always be C<ev_prepare>, blocking,
		2634	C<ev_check> so if you have one watcher of each kind they will always be
		2635	called in pairs bracketing the blocking call.
		2636
1022	Their main purpose is to integrate other event mechanisms into libev and	2637	Their main purpose is to integrate other event mechanisms into libev and
1023	their use is somewhat advanced. This could be used, for example, to track	2638	their use is somewhat advanced. They could be used, for example, to track
1024	variable changes, implement your own watchers, integrate net-snmp or a	2639	variable changes, implement your own watchers, integrate net-snmp or a
1025	coroutine library and lots more.	2640	coroutine library and lots more. They are also occasionally useful if
		2641	you cache some data and want to flush it before blocking (for example,
		2642	in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
		2643	watcher).
1026		2644
1027	This is done by examining in each prepare call which file descriptors need	2645	This is done by examining in each prepare call which file descriptors
1028	to be watched by the other library, registering C<ev_io> watchers for	2646	need to be watched by the other library, registering C<ev_io> watchers
1029	them and starting an C<ev_timer> watcher for any timeouts (many libraries	2647	for them and starting an C<ev_timer> watcher for any timeouts (many
1030	provide just this functionality). Then, in the check watcher you check for	2648	libraries provide exactly this functionality). Then, in the check watcher,
1031	any events that occured (by checking the pending status of all watchers	2649	you check for any events that occurred (by checking the pending status
1032	and stopping them) and call back into the library. The I/O and timer	2650	of all watchers and stopping them) and call back into the library. The
1033	callbacks will never actually be called (but must be valid nevertheless,	2651	I/O and timer callbacks will never actually be called (but must be valid
1034	because you never know, you know?).	2652	nevertheless, because you never know, you know?).
1035		2653
1036	As another example, the Perl Coro module uses these hooks to integrate	2654	As another example, the Perl Coro module uses these hooks to integrate
1037	coroutines into libev programs, by yielding to other active coroutines	2655	coroutines into libev programs, by yielding to other active coroutines
1038	during each prepare and only letting the process block if no coroutines	2656	during each prepare and only letting the process block if no coroutines
1039	are ready to run (it's actually more complicated: it only runs coroutines	2657	are ready to run (it's actually more complicated: it only runs coroutines
1040	with priority higher than or equal to the event loop and one coroutine	2658	with priority higher than or equal to the event loop and one coroutine
1041	of lower priority, but only once, using idle watchers to keep the event	2659	of lower priority, but only once, using idle watchers to keep the event
1042	loop from blocking if lower-priority coroutines are active, thus mapping	2660	loop from blocking if lower-priority coroutines are active, thus mapping
1043	low-priority coroutines to idle/background tasks).	2661	low-priority coroutines to idle/background tasks).
1044		2662
		2663	It is recommended to give C<ev_check> watchers highest (C<EV_MAXPRI>)
		2664	priority, to ensure that they are being run before any other watchers
		2665	after the poll (this doesn't matter for C<ev_prepare> watchers).
		2666
		2667	Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
		2668	activate ("feed") events into libev. While libev fully supports this, they
		2669	might get executed before other C<ev_check> watchers did their job. As
		2670	C<ev_check> watchers are often used to embed other (non-libev) event
		2671	loops those other event loops might be in an unusable state until their
		2672	C<ev_check> watcher ran (always remind yourself to coexist peacefully with
		2673	others).
		2674
		2675	=head3 Watcher-Specific Functions and Data Members
		2676
1045	=over 4	2677	=over 4
1046		2678
1047	=item ev_prepare_init (ev_prepare *, callback)	2679	=item ev_prepare_init (ev_prepare *, callback)
1048		2680
1049	=item ev_check_init (ev_check *, callback)	2681	=item ev_check_init (ev_check *, callback)
1050		2682
1051	Initialises and configures the prepare or check watcher - they have no	2683	Initialises and configures the prepare or check watcher - they have no
1052	parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>	2684	parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
1053	macros, but using them is utterly, utterly and completely pointless.	2685	macros, but using them is utterly, utterly, utterly and completely
		2686	pointless.
1054		2687
1055	=back	2688	=back
1056		2689
1057	Example: *TODO*.	2690	=head3 Examples
1058		2691
		2692	There are a number of principal ways to embed other event loops or modules
		2693	into libev. Here are some ideas on how to include libadns into libev
		2694	(there is a Perl module named C<EV::ADNS> that does this, which you could
		2695	use as a working example. Another Perl module named C<EV::Glib> embeds a
		2696	Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
		2697	Glib event loop).
1059		2698
		2699	Method 1: Add IO watchers and a timeout watcher in a prepare handler,
		2700	and in a check watcher, destroy them and call into libadns. What follows
		2701	is pseudo-code only of course. This requires you to either use a low
		2702	priority for the check watcher or use C<ev_clear_pending> explicitly, as
		2703	the callbacks for the IO/timeout watchers might not have been called yet.
		2704
		2705	static ev_io iow [nfd];
		2706	static ev_timer tw;
		2707
		2708	static void
		2709	io_cb (struct ev_loop *loop, ev_io *w, int revents)
		2710	{
		2711	}
		2712
		2713	// create io watchers for each fd and a timer before blocking
		2714	static void
		2715	adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
		2716	{
		2717	int timeout = 3600000;
		2718	struct pollfd fds [nfd];
		2719	// actual code will need to loop here and realloc etc.
		2720	adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
		2721
		2722	/* the callback is illegal, but won't be called as we stop during check */
		2723	ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
		2724	ev_timer_start (loop, &tw);
		2725
		2726	// create one ev_io per pollfd
		2727	for (int i = 0; i < nfd; ++i)
		2728	{
		2729	ev_io_init (iow + i, io_cb, fds [i].fd,
		2730	((fds [i].events & POLLIN ? EV_READ : 0)
		2731	| (fds [i].events & POLLOUT ? EV_WRITE : 0)));
		2732
		2733	fds [i].revents = 0;
		2734	ev_io_start (loop, iow + i);
		2735	}
		2736	}
		2737
		2738	// stop all watchers after blocking
		2739	static void
		2740	adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
		2741	{
		2742	ev_timer_stop (loop, &tw);
		2743
		2744	for (int i = 0; i < nfd; ++i)
		2745	{
		2746	// set the relevant poll flags
		2747	// could also call adns_processreadable etc. here
		2748	struct pollfd *fd = fds + i;
		2749	int revents = ev_clear_pending (iow + i);
		2750	if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
		2751	if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
		2752
		2753	// now stop the watcher
		2754	ev_io_stop (loop, iow + i);
		2755	}
		2756
		2757	adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
		2758	}
		2759
		2760	Method 2: This would be just like method 1, but you run C<adns_afterpoll>
		2761	in the prepare watcher and would dispose of the check watcher.
		2762
		2763	Method 3: If the module to be embedded supports explicit event
		2764	notification (libadns does), you can also make use of the actual watcher
		2765	callbacks, and only destroy/create the watchers in the prepare watcher.
		2766
		2767	static void
		2768	timer_cb (EV_P_ ev_timer *w, int revents)
		2769	{
		2770	adns_state ads = (adns_state)w->data;
		2771	update_now (EV_A);
		2772
		2773	adns_processtimeouts (ads, &tv_now);
		2774	}
		2775
		2776	static void
		2777	io_cb (EV_P_ ev_io *w, int revents)
		2778	{
		2779	adns_state ads = (adns_state)w->data;
		2780	update_now (EV_A);
		2781
		2782	if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
		2783	if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
		2784	}
		2785
		2786	// do not ever call adns_afterpoll
		2787
		2788	Method 4: Do not use a prepare or check watcher because the module you
		2789	want to embed is not flexible enough to support it. Instead, you can
		2790	override their poll function. The drawback with this solution is that the
		2791	main loop is now no longer controllable by EV. The C<Glib::EV> module uses
		2792	this approach, effectively embedding EV as a client into the horrible
		2793	libglib event loop.
		2794
		2795	static gint
		2796	event_poll_func (GPollFD *fds, guint nfds, gint timeout)
		2797	{
		2798	int got_events = 0;
		2799
		2800	for (n = 0; n < nfds; ++n)
		2801	// create/start io watcher that sets the relevant bits in fds[n] and increment got_events
		2802
		2803	if (timeout >= 0)
		2804	// create/start timer
		2805
		2806	// poll
		2807	ev_loop (EV_A_ 0);
		2808
		2809	// stop timer again
		2810	if (timeout >= 0)
		2811	ev_timer_stop (EV_A_ &to);
		2812
		2813	// stop io watchers again - their callbacks should have set
		2814	for (n = 0; n < nfds; ++n)
		2815	ev_io_stop (EV_A_ iow [n]);
		2816
		2817	return got_events;
		2818	}
		2819
		2820
1060	=head2 C<ev_embed> - when one backend isn't enough	2821	=head2 C<ev_embed> - when one backend isn't enough...
1061		2822
1062	This is a rather advanced watcher type that lets you embed one event loop	2823	This is a rather advanced watcher type that lets you embed one event loop
1063	into another.	2824	into another (currently only C<ev_io> events are supported in the embedded
		2825	loop, other types of watchers might be handled in a delayed or incorrect
		2826	fashion and must not be used).
1064		2827
1065	There are primarily two reasons you would want that: work around bugs and	2828	There are primarily two reasons you would want that: work around bugs and
1066	prioritise I/O.	2829	prioritise I/O.
1067		2830
1068	As an example for a bug workaround, the kqueue backend might only support	2831	As an example for a bug workaround, the kqueue backend might only support
1069	sockets on some platform, so it is unusable as generic backend, but you	2832	sockets on some platform, so it is unusable as generic backend, but you
1070	still want to make use of it because you have many sockets and it scales	2833	still want to make use of it because you have many sockets and it scales
1071	so nicely. In this case, you would create a kqueue-based loop and embed it	2834	so nicely. In this case, you would create a kqueue-based loop and embed
1072	into your default loop (which might use e.g. poll). Overall operation will	2835	it into your default loop (which might use e.g. poll). Overall operation
1073	be a bit slower because first libev has to poll and then call kevent, but	2836	will be a bit slower because first libev has to call C<poll> and then
1074	at least you can use both at what they are best.	2837	C<kevent>, but at least you can use both mechanisms for what they are
		2838	best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
1075		2839
1076	As for prioritising I/O: rarely you have the case where some fds have	2840	As for prioritising I/O: under rare circumstances you have the case where
1077	to be watched and handled very quickly (with low latency), and even	2841	some fds have to be watched and handled very quickly (with low latency),
1078	priorities and idle watchers might have too much overhead. In this case	2842	and even priorities and idle watchers might have too much overhead. In
1079	you would put all the high priority stuff in one loop and all the rest in	2843	this case you would put all the high priority stuff in one loop and all
1080	a second one, and embed the second one in the first.	2844	the rest in a second one, and embed the second one in the first.
1081		2845
1082	As long as the watcher is started it will automatically handle events. The	2846	As long as the watcher is active, the callback will be invoked every
1083	callback will be invoked whenever some events have been handled. You can	2847	time there might be events pending in the embedded loop. The callback
1084	set the callback to C<0> to avoid having to specify one if you are not	2848	must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
1085	interested in that.	2849	sweep and invoke their callbacks (the callback doesn't need to invoke the
		2850	C<ev_embed_sweep> function directly, it could also start an idle watcher
		2851	to give the embedded loop strictly lower priority for example).
1086		2852
1087	Also, there have not currently been made special provisions for forking:	2853	You can also set the callback to C<0>, in which case the embed watcher
1088	when you fork, you not only have to call C<ev_loop_fork> on both loops,	2854	will automatically execute the embedded loop sweep whenever necessary.
1089	but you will also have to stop and restart any C<ev_embed> watchers
1090	yourself.
1091		2855
		2856	Fork detection will be handled transparently while the C<ev_embed> watcher
		2857	is active, i.e., the embedded loop will automatically be forked when the
		2858	embedding loop forks. In other cases, the user is responsible for calling
		2859	C<ev_loop_fork> on the embedded loop.
		2860
1092	Unfortunately, not all backends are embeddable, only the ones returned by	2861	Unfortunately, not all backends are embeddable: only the ones returned by
1093	C<ev_embeddable_backends> are, which, unfortunately, does not include any	2862	C<ev_embeddable_backends> are, which, unfortunately, does not include any
1094	portable one.	2863	portable one.
1095		2864
1096	So when you want to use this feature you will always have to be prepared	2865	So when you want to use this feature you will always have to be prepared
1097	that you cannot get an embeddable loop. The recommended way to get around	2866	that you cannot get an embeddable loop. The recommended way to get around
1098	this is to have a separate variables for your embeddable loop, try to	2867	this is to have a separate variables for your embeddable loop, try to
1099	create it, and if that fails, use the normal loop for everything:	2868	create it, and if that fails, use the normal loop for everything.
1100		2869
1101	struct ev_loop *loop_hi = ev_default_init (0);	2870	=head3 C<ev_embed> and fork
1102	struct ev_loop *loop_lo = 0;
1103	struct ev_embed embed;
1104
1105	// see if there is a chance of getting one that works
1106	// (remember that a flags value of 0 means autodetection)
1107	loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
1108	? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
1109	: 0;
1110		2871
1111	// if we got one, then embed it, otherwise default to loop_hi	2872	While the C<ev_embed> watcher is running, forks in the embedding loop will
1112	if (loop_lo)	2873	automatically be applied to the embedded loop as well, so no special
1113	{	2874	fork handling is required in that case. When the watcher is not running,
1114	ev_embed_init (&embed, 0, loop_lo);	2875	however, it is still the task of the libev user to call C<ev_loop_fork ()>
1115	ev_embed_start (loop_hi, &embed);	2876	as applicable.
1116	}	2877
1117	else	2878	=head3 Watcher-Specific Functions and Data Members
1118	loop_lo = loop_hi;
1119		2879
1120	=over 4	2880	=over 4
1121		2881
1122	=item ev_embed_init (ev_embed *, callback, struct ev_loop *loop)	2882	=item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
1123		2883
1124	=item ev_embed_set (ev_embed *, callback, struct ev_loop *loop)	2884	=item ev_embed_set (ev_embed *, callback, struct ev_loop *embedded_loop)
1125		2885
1126	Configures the watcher to embed the given loop, which must be embeddable.	2886	Configures the watcher to embed the given loop, which must be
		2887	embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
		2888	invoked automatically, otherwise it is the responsibility of the callback
		2889	to invoke it (it will continue to be called until the sweep has been done,
		2890	if you do not want that, you need to temporarily stop the embed watcher).
		2891
		2892	=item ev_embed_sweep (loop, ev_embed *)
		2893
		2894	Make a single, non-blocking sweep over the embedded loop. This works
		2895	similarly to C<ev_loop (embedded_loop, EVLOOP_NONBLOCK)>, but in the most
		2896	appropriate way for embedded loops.
		2897
		2898	=item struct ev_loop *other [read-only]
		2899
		2900	The embedded event loop.
1127		2901
1128	=back	2902	=back
1129		2903
		2904	=head3 Examples
		2905
		2906	Example: Try to get an embeddable event loop and embed it into the default
		2907	event loop. If that is not possible, use the default loop. The default
		2908	loop is stored in C<loop_hi>, while the embeddable loop is stored in
		2909	C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
		2910	used).
		2911
		2912	struct ev_loop *loop_hi = ev_default_init (0);
		2913	struct ev_loop *loop_lo = 0;
		2914	ev_embed embed;
		2915
		2916	// see if there is a chance of getting one that works
		2917	// (remember that a flags value of 0 means autodetection)
		2918	loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
		2919	? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
		2920	: 0;
		2921
		2922	// if we got one, then embed it, otherwise default to loop_hi
		2923	if (loop_lo)
		2924	{
		2925	ev_embed_init (&embed, 0, loop_lo);
		2926	ev_embed_start (loop_hi, &embed);
		2927	}
		2928	else
		2929	loop_lo = loop_hi;
		2930
		2931	Example: Check if kqueue is available but not recommended and create
		2932	a kqueue backend for use with sockets (which usually work with any
		2933	kqueue implementation). Store the kqueue/socket-only event loop in
		2934	C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
		2935
		2936	struct ev_loop *loop = ev_default_init (0);
		2937	struct ev_loop *loop_socket = 0;
		2938	ev_embed embed;
		2939
		2940	if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
		2941	if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
		2942	{
		2943	ev_embed_init (&embed, 0, loop_socket);
		2944	ev_embed_start (loop, &embed);
		2945	}
		2946
		2947	if (!loop_socket)
		2948	loop_socket = loop;
		2949
		2950	// now use loop_socket for all sockets, and loop for everything else
		2951
		2952
		2953	=head2 C<ev_fork> - the audacity to resume the event loop after a fork
		2954
		2955	Fork watchers are called when a C<fork ()> was detected (usually because
		2956	whoever is a good citizen cared to tell libev about it by calling
		2957	C<ev_default_fork> or C<ev_loop_fork>). The invocation is done before the
		2958	event loop blocks next and before C<ev_check> watchers are being called,
		2959	and only in the child after the fork. If whoever good citizen calling
		2960	C<ev_default_fork> cheats and calls it in the wrong process, the fork
		2961	handlers will be invoked, too, of course.
		2962
		2963	=head3 The special problem of life after fork - how is it possible?
		2964
		2965	Most uses of C<fork()> consist of forking, then some simple calls to ste
		2966	up/change the process environment, followed by a call to C<exec()>. This
		2967	sequence should be handled by libev without any problems.
		2968
		2969	This changes when the application actually wants to do event handling
		2970	in the child, or both parent in child, in effect "continuing" after the
		2971	fork.
		2972
		2973	The default mode of operation (for libev, with application help to detect
		2974	forks) is to duplicate all the state in the child, as would be expected
		2975	when I<either> the parent I<or> the child process continues.
		2976
		2977	When both processes want to continue using libev, then this is usually the
		2978	wrong result. In that case, usually one process (typically the parent) is
		2979	supposed to continue with all watchers in place as before, while the other
		2980	process typically wants to start fresh, i.e. without any active watchers.
		2981
		2982	The cleanest and most efficient way to achieve that with libev is to
		2983	simply create a new event loop, which of course will be "empty", and
		2984	use that for new watchers. This has the advantage of not touching more
		2985	memory than necessary, and thus avoiding the copy-on-write, and the
		2986	disadvantage of having to use multiple event loops (which do not support
		2987	signal watchers).
		2988
		2989	When this is not possible, or you want to use the default loop for
		2990	other reasons, then in the process that wants to start "fresh", call
		2991	C<ev_default_destroy ()> followed by C<ev_default_loop (...)>. Destroying
		2992	the default loop will "orphan" (not stop) all registered watchers, so you
		2993	have to be careful not to execute code that modifies those watchers. Note
		2994	also that in that case, you have to re-register any signal watchers.
		2995
		2996	=head3 Watcher-Specific Functions and Data Members
		2997
		2998	=over 4
		2999
		3000	=item ev_fork_init (ev_signal *, callback)
		3001
		3002	Initialises and configures the fork watcher - it has no parameters of any
		3003	kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
		3004	believe me.
		3005
		3006	=back
		3007
		3008
		3009	=head2 C<ev_async> - how to wake up another event loop
		3010
		3011	In general, you cannot use an C<ev_loop> from multiple threads or other
		3012	asynchronous sources such as signal handlers (as opposed to multiple event
		3013	loops - those are of course safe to use in different threads).
		3014
		3015	Sometimes, however, you need to wake up another event loop you do not
		3016	control, for example because it belongs to another thread. This is what
		3017	C<ev_async> watchers do: as long as the C<ev_async> watcher is active, you
		3018	can signal it by calling C<ev_async_send>, which is thread- and signal
		3019	safe.
		3020
		3021	This functionality is very similar to C<ev_signal> watchers, as signals,
		3022	too, are asynchronous in nature, and signals, too, will be compressed
		3023	(i.e. the number of callback invocations may be less than the number of
		3024	C<ev_async_sent> calls).
		3025
		3026	Unlike C<ev_signal> watchers, C<ev_async> works with any event loop, not
		3027	just the default loop.
		3028
		3029	=head3 Queueing
		3030
		3031	C<ev_async> does not support queueing of data in any way. The reason
		3032	is that the author does not know of a simple (or any) algorithm for a
		3033	multiple-writer-single-reader queue that works in all cases and doesn't
		3034	need elaborate support such as pthreads or unportable memory access
		3035	semantics.
		3036
		3037	That means that if you want to queue data, you have to provide your own
		3038	queue. But at least I can tell you how to implement locking around your
		3039	queue:
		3040
		3041	=over 4
		3042
		3043	=item queueing from a signal handler context
		3044
		3045	To implement race-free queueing, you simply add to the queue in the signal
		3046	handler but you block the signal handler in the watcher callback. Here is
		3047	an example that does that for some fictitious SIGUSR1 handler:
		3048
		3049	static ev_async mysig;
		3050
		3051	static void
		3052	sigusr1_handler (void)
		3053	{
		3054	sometype data;
		3055
		3056	// no locking etc.
		3057	queue_put (data);
		3058	ev_async_send (EV_DEFAULT_ &mysig);
		3059	}
		3060
		3061	static void
		3062	mysig_cb (EV_P_ ev_async *w, int revents)
		3063	{
		3064	sometype data;
		3065	sigset_t block, prev;
		3066
		3067	sigemptyset (&block);
		3068	sigaddset (&block, SIGUSR1);
		3069	sigprocmask (SIG_BLOCK, &block, &prev);
		3070
		3071	while (queue_get (&data))
		3072	process (data);
		3073
		3074	if (sigismember (&prev, SIGUSR1)
		3075	sigprocmask (SIG_UNBLOCK, &block, 0);
		3076	}
		3077
		3078	(Note: pthreads in theory requires you to use C<pthread_setmask>
		3079	instead of C<sigprocmask> when you use threads, but libev doesn't do it
		3080	either...).
		3081
		3082	=item queueing from a thread context
		3083
		3084	The strategy for threads is different, as you cannot (easily) block
		3085	threads but you can easily preempt them, so to queue safely you need to
		3086	employ a traditional mutex lock, such as in this pthread example:
		3087
		3088	static ev_async mysig;
		3089	static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
		3090
		3091	static void
		3092	otherthread (void)
		3093	{
		3094	// only need to lock the actual queueing operation
		3095	pthread_mutex_lock (&mymutex);
		3096	queue_put (data);
		3097	pthread_mutex_unlock (&mymutex);
		3098
		3099	ev_async_send (EV_DEFAULT_ &mysig);
		3100	}
		3101
		3102	static void
		3103	mysig_cb (EV_P_ ev_async *w, int revents)
		3104	{
		3105	pthread_mutex_lock (&mymutex);
		3106
		3107	while (queue_get (&data))
		3108	process (data);
		3109
		3110	pthread_mutex_unlock (&mymutex);
		3111	}
		3112
		3113	=back
		3114
		3115
		3116	=head3 Watcher-Specific Functions and Data Members
		3117
		3118	=over 4
		3119
		3120	=item ev_async_init (ev_async *, callback)
		3121
		3122	Initialises and configures the async watcher - it has no parameters of any
		3123	kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
		3124	trust me.
		3125
		3126	=item ev_async_send (loop, ev_async *)
		3127
		3128	Sends/signals/activates the given C<ev_async> watcher, that is, feeds
		3129	an C<EV_ASYNC> event on the watcher into the event loop. Unlike
		3130	C<ev_feed_event>, this call is safe to do from other threads, signal or
		3131	similar contexts (see the discussion of C<EV_ATOMIC_T> in the embedding
		3132	section below on what exactly this means).
		3133
		3134	Note that, as with other watchers in libev, multiple events might get
		3135	compressed into a single callback invocation (another way to look at this
		3136	is that C<ev_async> watchers are level-triggered, set on C<ev_async_send>,
		3137	reset when the event loop detects that).
		3138
		3139	This call incurs the overhead of a system call only once per event loop
		3140	iteration, so while the overhead might be noticeable, it doesn't apply to
		3141	repeated calls to C<ev_async_send> for the same event loop.
		3142
		3143	=item bool = ev_async_pending (ev_async *)
		3144
		3145	Returns a non-zero value when C<ev_async_send> has been called on the
		3146	watcher but the event has not yet been processed (or even noted) by the
		3147	event loop.
		3148
		3149	C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
		3150	the loop iterates next and checks for the watcher to have become active,
		3151	it will reset the flag again. C<ev_async_pending> can be used to very
		3152	quickly check whether invoking the loop might be a good idea.
		3153
		3154	Not that this does I<not> check whether the watcher itself is pending,
		3155	only whether it has been requested to make this watcher pending: there
		3156	is a time window between the event loop checking and resetting the async
		3157	notification, and the callback being invoked.
		3158
		3159	=back
		3160
1130		3161
1131	=head1 OTHER FUNCTIONS	3162	=head1 OTHER FUNCTIONS
1132		3163
1133	There are some other functions of possible interest. Described. Here. Now.	3164	There are some other functions of possible interest. Described. Here. Now.
1134		3165
1135	=over 4	3166	=over 4
1136		3167
1137	=item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)	3168	=item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)
1138		3169
1139	This function combines a simple timer and an I/O watcher, calls your	3170	This function combines a simple timer and an I/O watcher, calls your
1140	callback on whichever event happens first and automatically stop both	3171	callback on whichever event happens first and automatically stops both
1141	watchers. This is useful if you want to wait for a single event on an fd	3172	watchers. This is useful if you want to wait for a single event on an fd
1142	or timeout without having to allocate/configure/start/stop/free one or	3173	or timeout without having to allocate/configure/start/stop/free one or
1143	more watchers yourself.	3174	more watchers yourself.
1144		3175
1145	If C<fd> is less than 0, then no I/O watcher will be started and events	3176	If C<fd> is less than 0, then no I/O watcher will be started and the
1146	is being ignored. Otherwise, an C<ev_io> watcher for the given C<fd> and	3177	C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
1147	C<events> set will be craeted and started.	3178	the given C<fd> and C<events> set will be created and started.
1148		3179
1149	If C<timeout> is less than 0, then no timeout watcher will be	3180	If C<timeout> is less than 0, then no timeout watcher will be
1150	started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and	3181	started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
1151	repeat = 0) will be started. While C<0> is a valid timeout, it is of	3182	repeat = 0) will be started. C<0> is a valid timeout.
1152	dubious value.
1153		3183
1154	The callback has the type C<void (*cb)(int revents, void *arg)> and gets	3184	The callback has the type C<void (*cb)(int revents, void *arg)> and gets
1155	passed an C<revents> set like normal event callbacks (a combination of	3185	passed an C<revents> set like normal event callbacks (a combination of
1156	C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMEOUT>) and the C<arg>	3186	C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMEOUT>) and the C<arg>
1157	value passed to C<ev_once>:	3187	value passed to C<ev_once>. Note that it is possible to receive I<both>
		3188	a timeout and an io event at the same time - you probably should give io
		3189	events precedence.
1158		3190
		3191	Example: wait up to ten seconds for data to appear on STDIN_FILENO.
		3192
1159	static void stdin_ready (int revents, void *arg)	3193	static void stdin_ready (int revents, void *arg)
1160	{	3194	{
1161	if (revents & EV_TIMEOUT)
1162	/* doh, nothing entered */;
1163	else if (revents & EV_READ)	3195	if (revents & EV_READ)
1164	/* stdin might have data for us, joy! */;	3196	/* stdin might have data for us, joy! */;
		3197	else if (revents & EV_TIMEOUT)
		3198	/* doh, nothing entered */;
1165	}	3199	}
1166		3200
1167	ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);	3201	ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
1168
1169	=item ev_feed_event (loop, watcher, int events)
1170
1171	Feeds the given event set into the event loop, as if the specified event
1172	had happened for the specified watcher (which must be a pointer to an
1173	initialised but not necessarily started event watcher).
1174		3202
1175	=item ev_feed_fd_event (loop, int fd, int revents)	3203	=item ev_feed_fd_event (loop, int fd, int revents)
1176		3204
1177	Feed an event on the given fd, as if a file descriptor backend detected	3205	Feed an event on the given fd, as if a file descriptor backend detected
1178	the given events it.	3206	the given events it.
1179		3207
1180	=item ev_feed_signal_event (loop, int signum)	3208	=item ev_feed_signal_event (loop, int signum)
1181		3209
1182	Feed an event as if the given signal occured (loop must be the default loop!).	3210	Feed an event as if the given signal occurred (C<loop> must be the default
		3211	loop!).
1183		3212
1184	=back	3213	=back
1185		3214
1186		3215
1187	=head1 LIBEVENT EMULATION	3216	=head1 LIBEVENT EMULATION
…		…
1202		3231
1203	=item * Priorities are not currently supported. Initialising priorities	3232	=item * Priorities are not currently supported. Initialising priorities
1204	will fail and all watchers will have the same priority, even though there	3233	will fail and all watchers will have the same priority, even though there
1205	is an ev_pri field.	3234	is an ev_pri field.
1206		3235
		3236	=item * In libevent, the last base created gets the signals, in libev, the
		3237	first base created (== the default loop) gets the signals.
		3238
1207	=item * Other members are not supported.	3239	=item * Other members are not supported.
1208		3240
1209	=item * The libev emulation is I<not> ABI compatible to libevent, you need	3241	=item * The libev emulation is I<not> ABI compatible to libevent, you need
1210	to use the libev header file and library.	3242	to use the libev header file and library.
1211		3243
1212	=back	3244	=back
1213		3245
1214	=head1 C++ SUPPORT	3246	=head1 C++ SUPPORT
1215		3247
1216	TBD.	3248	Libev comes with some simplistic wrapper classes for C++ that mainly allow
		3249	you to use some convenience methods to start/stop watchers and also change
		3250	the callback model to a model using method callbacks on objects.
		3251
		3252	To use it,
		3253
		3254	#include <ev++.h>
		3255
		3256	This automatically includes F<ev.h> and puts all of its definitions (many
		3257	of them macros) into the global namespace. All C++ specific things are
		3258	put into the C<ev> namespace. It should support all the same embedding
		3259	options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
		3260
		3261	Care has been taken to keep the overhead low. The only data member the C++
		3262	classes add (compared to plain C-style watchers) is the event loop pointer
		3263	that the watcher is associated with (or no additional members at all if
		3264	you disable C<EV_MULTIPLICITY> when embedding libev).
		3265
		3266	Currently, functions, and static and non-static member functions can be
		3267	used as callbacks. Other types should be easy to add as long as they only
		3268	need one additional pointer for context. If you need support for other
		3269	types of functors please contact the author (preferably after implementing
		3270	it).
		3271
		3272	Here is a list of things available in the C<ev> namespace:
		3273
		3274	=over 4
		3275
		3276	=item C<ev::READ>, C<ev::WRITE> etc.
		3277
		3278	These are just enum values with the same values as the C<EV_READ> etc.
		3279	macros from F<ev.h>.
		3280
		3281	=item C<ev::tstamp>, C<ev::now>
		3282
		3283	Aliases to the same types/functions as with the C<ev_> prefix.
		3284
		3285	=item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
		3286
		3287	For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
		3288	the same name in the C<ev> namespace, with the exception of C<ev_signal>
		3289	which is called C<ev::sig> to avoid clashes with the C<signal> macro
		3290	defines by many implementations.
		3291
		3292	All of those classes have these methods:
		3293
		3294	=over 4
		3295
		3296	=item ev::TYPE::TYPE ()
		3297
		3298	=item ev::TYPE::TYPE (loop)
		3299
		3300	=item ev::TYPE::~TYPE
		3301
		3302	The constructor (optionally) takes an event loop to associate the watcher
		3303	with. If it is omitted, it will use C<EV_DEFAULT>.
		3304
		3305	The constructor calls C<ev_init> for you, which means you have to call the
		3306	C<set> method before starting it.
		3307
		3308	It will not set a callback, however: You have to call the templated C<set>
		3309	method to set a callback before you can start the watcher.
		3310
		3311	(The reason why you have to use a method is a limitation in C++ which does
		3312	not allow explicit template arguments for constructors).
		3313
		3314	The destructor automatically stops the watcher if it is active.
		3315
		3316	=item w->set<class, &class::method> (object *)
		3317
		3318	This method sets the callback method to call. The method has to have a
		3319	signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
		3320	first argument and the C<revents> as second. The object must be given as
		3321	parameter and is stored in the C<data> member of the watcher.
		3322
		3323	This method synthesizes efficient thunking code to call your method from
		3324	the C callback that libev requires. If your compiler can inline your
		3325	callback (i.e. it is visible to it at the place of the C<set> call and
		3326	your compiler is good :), then the method will be fully inlined into the
		3327	thunking function, making it as fast as a direct C callback.
		3328
		3329	Example: simple class declaration and watcher initialisation
		3330
		3331	struct myclass
		3332	{
		3333	void io_cb (ev::io &w, int revents) { }
		3334	}
		3335
		3336	myclass obj;
		3337	ev::io iow;
		3338	iow.set <myclass, &myclass::io_cb> (&obj);
		3339
		3340	=item w->set (object *)
		3341
		3342	This is an B<experimental> feature that might go away in a future version.
		3343
		3344	This is a variation of a method callback - leaving out the method to call
		3345	will default the method to C<operator ()>, which makes it possible to use
		3346	functor objects without having to manually specify the C<operator ()> all
		3347	the time. Incidentally, you can then also leave out the template argument
		3348	list.
		3349
		3350	The C<operator ()> method prototype must be C<void operator ()(watcher &w,
		3351	int revents)>.
		3352
		3353	See the method-C<set> above for more details.
		3354
		3355	Example: use a functor object as callback.
		3356
		3357	struct myfunctor
		3358	{
		3359	void operator() (ev::io &w, int revents)
		3360	{
		3361	...
		3362	}
		3363	}
		3364
		3365	myfunctor f;
		3366
		3367	ev::io w;
		3368	w.set (&f);
		3369
		3370	=item w->set<function> (void *data = 0)
		3371
		3372	Also sets a callback, but uses a static method or plain function as
		3373	callback. The optional C<data> argument will be stored in the watcher's
		3374	C<data> member and is free for you to use.
		3375
		3376	The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
		3377
		3378	See the method-C<set> above for more details.
		3379
		3380	Example: Use a plain function as callback.
		3381
		3382	static void io_cb (ev::io &w, int revents) { }
		3383	iow.set <io_cb> ();
		3384
		3385	=item w->set (loop)
		3386
		3387	Associates a different C<struct ev_loop> with this watcher. You can only
		3388	do this when the watcher is inactive (and not pending either).
		3389
		3390	=item w->set ([arguments])
		3391
		3392	Basically the same as C<ev_TYPE_set>, with the same arguments. Must be
		3393	called at least once. Unlike the C counterpart, an active watcher gets
		3394	automatically stopped and restarted when reconfiguring it with this
		3395	method.
		3396
		3397	=item w->start ()
		3398
		3399	Starts the watcher. Note that there is no C<loop> argument, as the
		3400	constructor already stores the event loop.
		3401
		3402	=item w->stop ()
		3403
		3404	Stops the watcher if it is active. Again, no C<loop> argument.
		3405
		3406	=item w->again () (C<ev::timer>, C<ev::periodic> only)
		3407
		3408	For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
		3409	C<ev_TYPE_again> function.
		3410
		3411	=item w->sweep () (C<ev::embed> only)
		3412
		3413	Invokes C<ev_embed_sweep>.
		3414
		3415	=item w->update () (C<ev::stat> only)
		3416
		3417	Invokes C<ev_stat_stat>.
		3418
		3419	=back
		3420
		3421	=back
		3422
		3423	Example: Define a class with an IO and idle watcher, start one of them in
		3424	the constructor.
		3425
		3426	class myclass
		3427	{
		3428	ev::io io ; void io_cb (ev::io &w, int revents);
		3429	ev::idle idle; void idle_cb (ev::idle &w, int revents);
		3430
		3431	myclass (int fd)
		3432	{
		3433	io .set <myclass, &myclass::io_cb > (this);
		3434	idle.set <myclass, &myclass::idle_cb> (this);
		3435
		3436	io.start (fd, ev::READ);
		3437	}
		3438	};
		3439
		3440
		3441	=head1 OTHER LANGUAGE BINDINGS
		3442
		3443	Libev does not offer other language bindings itself, but bindings for a
		3444	number of languages exist in the form of third-party packages. If you know
		3445	any interesting language binding in addition to the ones listed here, drop
		3446	me a note.
		3447
		3448	=over 4
		3449
		3450	=item Perl
		3451
		3452	The EV module implements the full libev API and is actually used to test
		3453	libev. EV is developed together with libev. Apart from the EV core module,
		3454	there are additional modules that implement libev-compatible interfaces
		3455	to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
		3456	C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
		3457	and C<EV::Glib>).
		3458
		3459	It can be found and installed via CPAN, its homepage is at
		3460	L<http://software.schmorp.de/pkg/EV>.
		3461
		3462	=item Python
		3463
		3464	Python bindings can be found at L<http://code.google.com/p/pyev/>. It
		3465	seems to be quite complete and well-documented.
		3466
		3467	=item Ruby
		3468
		3469	Tony Arcieri has written a ruby extension that offers access to a subset
		3470	of the libev API and adds file handle abstractions, asynchronous DNS and
		3471	more on top of it. It can be found via gem servers. Its homepage is at
		3472	L<http://rev.rubyforge.org/>.
		3473
		3474	Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
		3475	makes rev work even on mingw.
		3476
		3477	=item Haskell
		3478
		3479	A haskell binding to libev is available at
		3480	L<http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
		3481
		3482	=item D
		3483
		3484	Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
		3485	be found at L<http://proj.llucax.com.ar/wiki/evd>.
		3486
		3487	=item Ocaml
		3488
		3489	Erkki Seppala has written Ocaml bindings for libev, to be found at
		3490	L<http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
		3491
		3492	=item Lua
		3493
		3494	Brian Maher has written a partial interface to libev for lua (at the
		3495	time of this writing, only C<ev_io> and C<ev_timer>), to be found at
		3496	L<http://github.com/brimworks/lua-ev>.
		3497
		3498	=back
		3499
		3500
		3501	=head1 MACRO MAGIC
		3502
		3503	Libev can be compiled with a variety of options, the most fundamental
		3504	of which is C<EV_MULTIPLICITY>. This option determines whether (most)
		3505	functions and callbacks have an initial C<struct ev_loop *> argument.
		3506
		3507	To make it easier to write programs that cope with either variant, the
		3508	following macros are defined:
		3509
		3510	=over 4
		3511
		3512	=item C<EV_A>, C<EV_A_>
		3513
		3514	This provides the loop I<argument> for functions, if one is required ("ev
		3515	loop argument"). The C<EV_A> form is used when this is the sole argument,
		3516	C<EV_A_> is used when other arguments are following. Example:
		3517
		3518	ev_unref (EV_A);
		3519	ev_timer_add (EV_A_ watcher);
		3520	ev_loop (EV_A_ 0);
		3521
		3522	It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
		3523	which is often provided by the following macro.
		3524
		3525	=item C<EV_P>, C<EV_P_>
		3526
		3527	This provides the loop I<parameter> for functions, if one is required ("ev
		3528	loop parameter"). The C<EV_P> form is used when this is the sole parameter,
		3529	C<EV_P_> is used when other parameters are following. Example:
		3530
		3531	// this is how ev_unref is being declared
		3532	static void ev_unref (EV_P);
		3533
		3534	// this is how you can declare your typical callback
		3535	static void cb (EV_P_ ev_timer *w, int revents)
		3536
		3537	It declares a parameter C<loop> of type C<struct ev_loop *>, quite
		3538	suitable for use with C<EV_A>.
		3539
		3540	=item C<EV_DEFAULT>, C<EV_DEFAULT_>
		3541
		3542	Similar to the other two macros, this gives you the value of the default
		3543	loop, if multiple loops are supported ("ev loop default").
		3544
		3545	=item C<EV_DEFAULT_UC>, C<EV_DEFAULT_UC_>
		3546
		3547	Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
		3548	default loop has been initialised (C<UC> == unchecked). Their behaviour
		3549	is undefined when the default loop has not been initialised by a previous
		3550	execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
		3551
		3552	It is often prudent to use C<EV_DEFAULT> when initialising the first
		3553	watcher in a function but use C<EV_DEFAULT_UC> afterwards.
		3554
		3555	=back
		3556
		3557	Example: Declare and initialise a check watcher, utilising the above
		3558	macros so it will work regardless of whether multiple loops are supported
		3559	or not.
		3560
		3561	static void
		3562	check_cb (EV_P_ ev_timer *w, int revents)
		3563	{
		3564	ev_check_stop (EV_A_ w);
		3565	}
		3566
		3567	ev_check check;
		3568	ev_check_init (&check, check_cb);
		3569	ev_check_start (EV_DEFAULT_ &check);
		3570	ev_loop (EV_DEFAULT_ 0);
		3571
		3572	=head1 EMBEDDING
		3573
		3574	Libev can (and often is) directly embedded into host
		3575	applications. Examples of applications that embed it include the Deliantra
		3576	Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
		3577	and rxvt-unicode.
		3578
		3579	The goal is to enable you to just copy the necessary files into your
		3580	source directory without having to change even a single line in them, so
		3581	you can easily upgrade by simply copying (or having a checked-out copy of
		3582	libev somewhere in your source tree).
		3583
		3584	=head2 FILESETS
		3585
		3586	Depending on what features you need you need to include one or more sets of files
		3587	in your application.
		3588
		3589	=head3 CORE EVENT LOOP
		3590
		3591	To include only the libev core (all the C<ev_*> functions), with manual
		3592	configuration (no autoconf):
		3593
		3594	#define EV_STANDALONE 1
		3595	#include "ev.c"
		3596
		3597	This will automatically include F<ev.h>, too, and should be done in a
		3598	single C source file only to provide the function implementations. To use
		3599	it, do the same for F<ev.h> in all files wishing to use this API (best
		3600	done by writing a wrapper around F<ev.h> that you can include instead and
		3601	where you can put other configuration options):
		3602
		3603	#define EV_STANDALONE 1
		3604	#include "ev.h"
		3605
		3606	Both header files and implementation files can be compiled with a C++
		3607	compiler (at least, that's a stated goal, and breakage will be treated
		3608	as a bug).
		3609
		3610	You need the following files in your source tree, or in a directory
		3611	in your include path (e.g. in libev/ when using -Ilibev):
		3612
		3613	ev.h
		3614	ev.c
		3615	ev_vars.h
		3616	ev_wrap.h
		3617
		3618	ev_win32.c required on win32 platforms only
		3619
		3620	ev_select.c only when select backend is enabled (which is enabled by default)
		3621	ev_poll.c only when poll backend is enabled (disabled by default)
		3622	ev_epoll.c only when the epoll backend is enabled (disabled by default)
		3623	ev_kqueue.c only when the kqueue backend is enabled (disabled by default)
		3624	ev_port.c only when the solaris port backend is enabled (disabled by default)
		3625
		3626	F<ev.c> includes the backend files directly when enabled, so you only need
		3627	to compile this single file.
		3628
		3629	=head3 LIBEVENT COMPATIBILITY API
		3630
		3631	To include the libevent compatibility API, also include:
		3632
		3633	#include "event.c"
		3634
		3635	in the file including F<ev.c>, and:
		3636
		3637	#include "event.h"
		3638
		3639	in the files that want to use the libevent API. This also includes F<ev.h>.
		3640
		3641	You need the following additional files for this:
		3642
		3643	event.h
		3644	event.c
		3645
		3646	=head3 AUTOCONF SUPPORT
		3647
		3648	Instead of using C<EV_STANDALONE=1> and providing your configuration in
		3649	whatever way you want, you can also C<m4_include([libev.m4])> in your
		3650	F<configure.ac> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
		3651	include F<config.h> and configure itself accordingly.
		3652
		3653	For this of course you need the m4 file:
		3654
		3655	libev.m4
		3656
		3657	=head2 PREPROCESSOR SYMBOLS/MACROS
		3658
		3659	Libev can be configured via a variety of preprocessor symbols you have to
		3660	define before including (or compiling) any of its files. The default in
		3661	the absence of autoconf is documented for every option.
		3662
		3663	Symbols marked with "(h)" do not change the ABI, and can have different
		3664	values when compiling libev vs. including F<ev.h>, so it is permissible
		3665	to redefine them before including F<ev.h> without breakign compatibility
		3666	to a compiled library. All other symbols change the ABI, which means all
		3667	users of libev and the libev code itself must be compiled with compatible
		3668	settings.
		3669
		3670	=over 4
		3671
		3672	=item EV_STANDALONE (h)
		3673
		3674	Must always be C<1> if you do not use autoconf configuration, which
		3675	keeps libev from including F<config.h>, and it also defines dummy
		3676	implementations for some libevent functions (such as logging, which is not
		3677	supported). It will also not define any of the structs usually found in
		3678	F<event.h> that are not directly supported by the libev core alone.
		3679
		3680	In standalone mode, libev will still try to automatically deduce the
		3681	configuration, but has to be more conservative.
		3682
		3683	=item EV_USE_MONOTONIC
		3684
		3685	If defined to be C<1>, libev will try to detect the availability of the
		3686	monotonic clock option at both compile time and runtime. Otherwise no
		3687	use of the monotonic clock option will be attempted. If you enable this,
		3688	you usually have to link against librt or something similar. Enabling it
		3689	when the functionality isn't available is safe, though, although you have
		3690	to make sure you link against any libraries where the C<clock_gettime>
		3691	function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
		3692
		3693	=item EV_USE_REALTIME
		3694
		3695	If defined to be C<1>, libev will try to detect the availability of the
		3696	real-time clock option at compile time (and assume its availability
		3697	at runtime if successful). Otherwise no use of the real-time clock
		3698	option will be attempted. This effectively replaces C<gettimeofday>
		3699	by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
		3700	correctness. See the note about libraries in the description of
		3701	C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
		3702	C<EV_USE_CLOCK_SYSCALL>.
		3703
		3704	=item EV_USE_CLOCK_SYSCALL
		3705
		3706	If defined to be C<1>, libev will try to use a direct syscall instead
		3707	of calling the system-provided C<clock_gettime> function. This option
		3708	exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
		3709	unconditionally pulls in C<libpthread>, slowing down single-threaded
		3710	programs needlessly. Using a direct syscall is slightly slower (in
		3711	theory), because no optimised vdso implementation can be used, but avoids
		3712	the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
		3713	higher, as it simplifies linking (no need for C<-lrt>).
		3714
		3715	=item EV_USE_NANOSLEEP
		3716
		3717	If defined to be C<1>, libev will assume that C<nanosleep ()> is available
		3718	and will use it for delays. Otherwise it will use C<select ()>.
		3719
		3720	=item EV_USE_EVENTFD
		3721
		3722	If defined to be C<1>, then libev will assume that C<eventfd ()> is
		3723	available and will probe for kernel support at runtime. This will improve
		3724	C<ev_signal> and C<ev_async> performance and reduce resource consumption.
		3725	If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
		3726	2.7 or newer, otherwise disabled.
		3727
		3728	=item EV_USE_SELECT
		3729
		3730	If undefined or defined to be C<1>, libev will compile in support for the
		3731	C<select>(2) backend. No attempt at auto-detection will be done: if no
		3732	other method takes over, select will be it. Otherwise the select backend
		3733	will not be compiled in.
		3734
		3735	=item EV_SELECT_USE_FD_SET
		3736
		3737	If defined to C<1>, then the select backend will use the system C<fd_set>
		3738	structure. This is useful if libev doesn't compile due to a missing
		3739	C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
		3740	on exotic systems. This usually limits the range of file descriptors to
		3741	some low limit such as 1024 or might have other limitations (winsocket
		3742	only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
		3743	configures the maximum size of the C<fd_set>.
		3744
		3745	=item EV_SELECT_IS_WINSOCKET
		3746
		3747	When defined to C<1>, the select backend will assume that
		3748	select/socket/connect etc. don't understand file descriptors but
		3749	wants osf handles on win32 (this is the case when the select to
		3750	be used is the winsock select). This means that it will call
		3751	C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
		3752	it is assumed that all these functions actually work on fds, even
		3753	on win32. Should not be defined on non-win32 platforms.
		3754
		3755	=item EV_FD_TO_WIN32_HANDLE(fd)
		3756
		3757	If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
		3758	file descriptors to socket handles. When not defining this symbol (the
		3759	default), then libev will call C<_get_osfhandle>, which is usually
		3760	correct. In some cases, programs use their own file descriptor management,
		3761	in which case they can provide this function to map fds to socket handles.
		3762
		3763	=item EV_WIN32_HANDLE_TO_FD(handle)
		3764
		3765	If C<EV_SELECT_IS_WINSOCKET> then libev maps handles to file descriptors
		3766	using the standard C<_open_osfhandle> function. For programs implementing
		3767	their own fd to handle mapping, overwriting this function makes it easier
		3768	to do so. This can be done by defining this macro to an appropriate value.
		3769
		3770	=item EV_WIN32_CLOSE_FD(fd)
		3771
		3772	If programs implement their own fd to handle mapping on win32, then this
		3773	macro can be used to override the C<close> function, useful to unregister
		3774	file descriptors again. Note that the replacement function has to close
		3775	the underlying OS handle.
		3776
		3777	=item EV_USE_POLL
		3778
		3779	If defined to be C<1>, libev will compile in support for the C<poll>(2)
		3780	backend. Otherwise it will be enabled on non-win32 platforms. It
		3781	takes precedence over select.
		3782
		3783	=item EV_USE_EPOLL
		3784
		3785	If defined to be C<1>, libev will compile in support for the Linux
		3786	C<epoll>(7) backend. Its availability will be detected at runtime,
		3787	otherwise another method will be used as fallback. This is the preferred
		3788	backend for GNU/Linux systems. If undefined, it will be enabled if the
		3789	headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
		3790
		3791	=item EV_USE_KQUEUE
		3792
		3793	If defined to be C<1>, libev will compile in support for the BSD style
		3794	C<kqueue>(2) backend. Its actual availability will be detected at runtime,
		3795	otherwise another method will be used as fallback. This is the preferred
		3796	backend for BSD and BSD-like systems, although on most BSDs kqueue only
		3797	supports some types of fds correctly (the only platform we found that
		3798	supports ptys for example was NetBSD), so kqueue might be compiled in, but
		3799	not be used unless explicitly requested. The best way to use it is to find
		3800	out whether kqueue supports your type of fd properly and use an embedded
		3801	kqueue loop.
		3802
		3803	=item EV_USE_PORT
		3804
		3805	If defined to be C<1>, libev will compile in support for the Solaris
		3806	10 port style backend. Its availability will be detected at runtime,
		3807	otherwise another method will be used as fallback. This is the preferred
		3808	backend for Solaris 10 systems.
		3809
		3810	=item EV_USE_DEVPOLL
		3811
		3812	Reserved for future expansion, works like the USE symbols above.
		3813
		3814	=item EV_USE_INOTIFY
		3815
		3816	If defined to be C<1>, libev will compile in support for the Linux inotify
		3817	interface to speed up C<ev_stat> watchers. Its actual availability will
		3818	be detected at runtime. If undefined, it will be enabled if the headers
		3819	indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
		3820
		3821	=item EV_ATOMIC_T
		3822
		3823	Libev requires an integer type (suitable for storing C<0> or C<1>) whose
		3824	access is atomic with respect to other threads or signal contexts. No such
		3825	type is easily found in the C language, so you can provide your own type
		3826	that you know is safe for your purposes. It is used both for signal handler "locking"
		3827	as well as for signal and thread safety in C<ev_async> watchers.
		3828
		3829	In the absence of this define, libev will use C<sig_atomic_t volatile>
		3830	(from F<signal.h>), which is usually good enough on most platforms.
		3831
		3832	=item EV_H (h)
		3833
		3834	The name of the F<ev.h> header file used to include it. The default if
		3835	undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
		3836	used to virtually rename the F<ev.h> header file in case of conflicts.
		3837
		3838	=item EV_CONFIG_H (h)
		3839
		3840	If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
		3841	F<ev.c>'s idea of where to find the F<config.h> file, similarly to
		3842	C<EV_H>, above.
		3843
		3844	=item EV_EVENT_H (h)
		3845
		3846	Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
		3847	of how the F<event.h> header can be found, the default is C<"event.h">.
		3848
		3849	=item EV_PROTOTYPES (h)
		3850
		3851	If defined to be C<0>, then F<ev.h> will not define any function
		3852	prototypes, but still define all the structs and other symbols. This is
		3853	occasionally useful if you want to provide your own wrapper functions
		3854	around libev functions.
		3855
		3856	=item EV_MULTIPLICITY
		3857
		3858	If undefined or defined to C<1>, then all event-loop-specific functions
		3859	will have the C<struct ev_loop *> as first argument, and you can create
		3860	additional independent event loops. Otherwise there will be no support
		3861	for multiple event loops and there is no first event loop pointer
		3862	argument. Instead, all functions act on the single default loop.
		3863
		3864	=item EV_MINPRI
		3865
		3866	=item EV_MAXPRI
		3867
		3868	The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
		3869	C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
		3870	provide for more priorities by overriding those symbols (usually defined
		3871	to be C<-2> and C<2>, respectively).
		3872
		3873	When doing priority-based operations, libev usually has to linearly search
		3874	all the priorities, so having many of them (hundreds) uses a lot of space
		3875	and time, so using the defaults of five priorities (-2 .. +2) is usually
		3876	fine.
		3877
		3878	If your embedding application does not need any priorities, defining these
		3879	both to C<0> will save some memory and CPU.
		3880
		3881	=item EV_PERIODIC_ENABLE, EV_IDLE_ENABLE, EV_EMBED_ENABLE, EV_STAT_ENABLE,
		3882	EV_PREPARE_ENABLE, EV_CHECK_ENABLE, EV_FORK_ENABLE, EV_SIGNAL_ENABLE,
		3883	EV_ASYNC_ENABLE, EV_CHILD_ENABLE.
		3884
		3885	If undefined or defined to be C<1> (and the platform supports it), then
		3886	the respective watcher type is supported. If defined to be C<0>, then it
		3887	is not. Disabling watcher types mainly saves codesize.
		3888
		3889	=item EV_FEATURES
		3890
		3891	If you need to shave off some kilobytes of code at the expense of some
		3892	speed (but with the full API), you can define this symbol to request
		3893	certain subsets of functionality. The default is to enable all features
		3894	that can be enabled on the platform.
		3895
		3896	Note that using autoconf will usually override most of the features, so
		3897	using this symbol makes sense mostly when embedding libev.
		3898
		3899	A typical way to use this symbol is to define it to C<0> (or to a bitset
		3900	with some broad features you want) and then selectively re-enable
		3901	additional parts you want, for example if you want everything minimal,
		3902	but multiple event loop support, async and child watchers and the poll
		3903	backend, use this:
		3904
		3905	#define EV_FEATURES 0
		3906	#define EV_MULTIPLICITY 1
		3907	#define EV_USE_POLL 1
		3908	#define EV_CHILD_ENABLE 1
		3909	#define EV_ASYNC_ENABLE 1
		3910
		3911	The actual value is a bitset, it can be a combination of the following
		3912	values:
		3913
		3914	=over 4
		3915
		3916	=item C<1> - faster/larger code
		3917
		3918	Use larger code to speed up some operations.
		3919
		3920	Currently this is used to override some inlining decisions (enlarging the roughly
		3921	30% code size on amd64.
		3922
		3923	Also disables C<assert>'s in the code, unless you define C<NDEBUG>
		3924	explicitly to C<0>.
		3925
		3926	Use of compiler flags such as C<-Os> with gcc that optimise for size are
		3927	recommended when disabling this feature.
		3928
		3929	=item C<2> - faster/larger data structures
		3930
		3931	Replaces the small 2-heap for timer management by a faster 4-heap, larger
		3932	hash table sizes and so on. This will usually further increase codesize
		3933	and can additionally have an effect on the size of data structures at
		3934	runtime.
		3935
		3936	=item C<4> - full API configuration
		3937
		3938	This enables priorities (sets C<EV_MAXPRI>=2 and C<EV_MINPRI>=-2), and
		3939	enables multiplicity (C<EV_MULTIPLICITY>=1).
		3940
		3941	It also enables a lot of the "lesser used" core API functions. See C<ev.h>
		3942	for details on which parts of the API are still available without this
		3943	feature, and do not complain if this subset changes over time.
		3944
		3945	=item C<8> - enable all optional watcher types
		3946
		3947	Enables all optional watcher types. If you want to selectively enable
		3948	only some watcher types other than I/O and timers (e.g. prepare,
		3949	embed, async, child...) you can enable them manually by defining
		3950	C<EV_watchertype_ENABLE> to C<1> instead.
		3951
		3952	=item C<16> - enable all backends
		3953
		3954	This enables all backends - without this feature, you need to enable at
		3955	least one backend manually (C<EV_USE_SELECT> is a good choice).
		3956
		3957	=item C<32> - enable OS-specific "helper" APIs
		3958
		3959	Enable inotify, eventfd, signalfd and similar OS-specific helper APIs by
		3960	default.
		3961
		3962	=back
		3963
		3964	Compiling with C<gcc -Os -DEV_STANDALONE -DEV_USE_EPOLL=1 -DEV_FEATURES=0>
		3965	reduces the compiled size of libev from 24.7Kb to 6.5Kb on my GNU/Linux
		3966	amd64 system, while still giving you I/O watchers, timers and monotonic
		3967	clock support.
		3968
		3969	With an intelligent-enough linker (gcc+binutils are intelligent enough
		3970	when you use C<-Wl,--gc-sections -ffunction-sections>) functions unused by
		3971	your program might be left out as well - a binary starting a timer and an
		3972	I/O watcher then might come out at only 5Kb.
		3973
		3974	=item EV_AVOID_STDIO
		3975
		3976	If this is set to C<1> at compiletime, then libev will avoid using stdio
		3977	functions (printf, scanf, perror etc.). This will increase the codesize
		3978	somewhat, but if your program doesn't otherwise depend on stdio and your
		3979	libc allows it, this avoids linking in the stdio library which is quite
		3980	big.
		3981
		3982	Note that error messages might become less precise when this option is
		3983	enabled.
		3984
		3985	=item EV_NSIG
		3986
		3987	The highest supported signal number, +1 (or, the number of
		3988	signals): Normally, libev tries to deduce the maximum number of signals
		3989	automatically, but sometimes this fails, in which case it can be
		3990	specified. Also, using a lower number than detected (C<32> should be
		3991	good for about any system in existance) can save some memory, as libev
		3992	statically allocates some 12-24 bytes per signal number.
		3993
		3994	=item EV_PID_HASHSIZE
		3995
		3996	C<ev_child> watchers use a small hash table to distribute workload by
		3997	pid. The default size is C<16> (or C<1> with C<EV_FEATURES> disabled),
		3998	usually more than enough. If you need to manage thousands of children you
		3999	might want to increase this value (I<must> be a power of two).
		4000
		4001	=item EV_INOTIFY_HASHSIZE
		4002
		4003	C<ev_stat> watchers use a small hash table to distribute workload by
		4004	inotify watch id. The default size is C<16> (or C<1> with C<EV_FEATURES>
		4005	disabled), usually more than enough. If you need to manage thousands of
		4006	C<ev_stat> watchers you might want to increase this value (I<must> be a
		4007	power of two).
		4008
		4009	=item EV_USE_4HEAP
		4010
		4011	Heaps are not very cache-efficient. To improve the cache-efficiency of the
		4012	timer and periodics heaps, libev uses a 4-heap when this symbol is defined
		4013	to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
		4014	faster performance with many (thousands) of watchers.
		4015
		4016	The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
		4017	will be C<0>.
		4018
		4019	=item EV_HEAP_CACHE_AT
		4020
		4021	Heaps are not very cache-efficient. To improve the cache-efficiency of the
		4022	timer and periodics heaps, libev can cache the timestamp (I<at>) within
		4023	the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
		4024	which uses 8-12 bytes more per watcher and a few hundred bytes more code,
		4025	but avoids random read accesses on heap changes. This improves performance
		4026	noticeably with many (hundreds) of watchers.
		4027
		4028	The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
		4029	will be C<0>.
		4030
		4031	=item EV_VERIFY
		4032
		4033	Controls how much internal verification (see C<ev_loop_verify ()>) will
		4034	be done: If set to C<0>, no internal verification code will be compiled
		4035	in. If set to C<1>, then verification code will be compiled in, but not
		4036	called. If set to C<2>, then the internal verification code will be
		4037	called once per loop, which can slow down libev. If set to C<3>, then the
		4038	verification code will be called very frequently, which will slow down
		4039	libev considerably.
		4040
		4041	The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
		4042	will be C<0>.
		4043
		4044	=item EV_COMMON
		4045
		4046	By default, all watchers have a C<void *data> member. By redefining
		4047	this macro to a something else you can include more and other types of
		4048	members. You have to define it each time you include one of the files,
		4049	though, and it must be identical each time.
		4050
		4051	For example, the perl EV module uses something like this:
		4052
		4053	#define EV_COMMON \
		4054	SV *self; /* contains this struct */ \
		4055	SV *cb_sv, *fh /* note no trailing ";" */
		4056
		4057	=item EV_CB_DECLARE (type)
		4058
		4059	=item EV_CB_INVOKE (watcher, revents)
		4060
		4061	=item ev_set_cb (ev, cb)
		4062
		4063	Can be used to change the callback member declaration in each watcher,
		4064	and the way callbacks are invoked and set. Must expand to a struct member
		4065	definition and a statement, respectively. See the F<ev.h> header file for
		4066	their default definitions. One possible use for overriding these is to
		4067	avoid the C<struct ev_loop *> as first argument in all cases, or to use
		4068	method calls instead of plain function calls in C++.
		4069
		4070	=back
		4071
		4072	=head2 EXPORTED API SYMBOLS
		4073
		4074	If you need to re-export the API (e.g. via a DLL) and you need a list of
		4075	exported symbols, you can use the provided F<Symbol.*> files which list
		4076	all public symbols, one per line:
		4077
		4078	Symbols.ev for libev proper
		4079	Symbols.event for the libevent emulation
		4080
		4081	This can also be used to rename all public symbols to avoid clashes with
		4082	multiple versions of libev linked together (which is obviously bad in
		4083	itself, but sometimes it is inconvenient to avoid this).
		4084
		4085	A sed command like this will create wrapper C<#define>'s that you need to
		4086	include before including F<ev.h>:
		4087
		4088	<Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
		4089
		4090	This would create a file F<wrap.h> which essentially looks like this:
		4091
		4092	#define ev_backend myprefix_ev_backend
		4093	#define ev_check_start myprefix_ev_check_start
		4094	#define ev_check_stop myprefix_ev_check_stop
		4095	...
		4096
		4097	=head2 EXAMPLES
		4098
		4099	For a real-world example of a program the includes libev
		4100	verbatim, you can have a look at the EV perl module
		4101	(L<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
		4102	the F<libev/> subdirectory and includes them in the F<EV/EVAPI.h> (public
		4103	interface) and F<EV.xs> (implementation) files. Only the F<EV.xs> file
		4104	will be compiled. It is pretty complex because it provides its own header
		4105	file.
		4106
		4107	The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
		4108	that everybody includes and which overrides some configure choices:
		4109
		4110	#define EV_FEATURES 0
		4111	#define EV_USE_SELECT 1
		4112	#define EV_CONFIG_H <config.h>
		4113
		4114	#include "ev++.h"
		4115
		4116	And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
		4117
		4118	#include "ev_cpp.h"
		4119	#include "ev.c"
		4120
		4121	=head1 INTERACTION WITH OTHER PROGRAMS OR LIBRARIES
		4122
		4123	=head2 THREADS AND COROUTINES
		4124
		4125	=head3 THREADS
		4126
		4127	All libev functions are reentrant and thread-safe unless explicitly
		4128	documented otherwise, but libev implements no locking itself. This means
		4129	that you can use as many loops as you want in parallel, as long as there
		4130	are no concurrent calls into any libev function with the same loop
		4131	parameter (C<ev_default_*> calls have an implicit default loop parameter,
		4132	of course): libev guarantees that different event loops share no data
		4133	structures that need any locking.
		4134
		4135	Or to put it differently: calls with different loop parameters can be done
		4136	concurrently from multiple threads, calls with the same loop parameter
		4137	must be done serially (but can be done from different threads, as long as
		4138	only one thread ever is inside a call at any point in time, e.g. by using
		4139	a mutex per loop).
		4140
		4141	Specifically to support threads (and signal handlers), libev implements
		4142	so-called C<ev_async> watchers, which allow some limited form of
		4143	concurrency on the same event loop, namely waking it up "from the
		4144	outside".
		4145
		4146	If you want to know which design (one loop, locking, or multiple loops
		4147	without or something else still) is best for your problem, then I cannot
		4148	help you, but here is some generic advice:
		4149
		4150	=over 4
		4151
		4152	=item * most applications have a main thread: use the default libev loop
		4153	in that thread, or create a separate thread running only the default loop.
		4154
		4155	This helps integrating other libraries or software modules that use libev
		4156	themselves and don't care/know about threading.
		4157
		4158	=item * one loop per thread is usually a good model.
		4159
		4160	Doing this is almost never wrong, sometimes a better-performance model
		4161	exists, but it is always a good start.
		4162
		4163	=item * other models exist, such as the leader/follower pattern, where one
		4164	loop is handed through multiple threads in a kind of round-robin fashion.
		4165
		4166	Choosing a model is hard - look around, learn, know that usually you can do
		4167	better than you currently do :-)
		4168
		4169	=item * often you need to talk to some other thread which blocks in the
		4170	event loop.
		4171
		4172	C<ev_async> watchers can be used to wake them up from other threads safely
		4173	(or from signal contexts...).
		4174
		4175	An example use would be to communicate signals or other events that only
		4176	work in the default loop by registering the signal watcher with the
		4177	default loop and triggering an C<ev_async> watcher from the default loop
		4178	watcher callback into the event loop interested in the signal.
		4179
		4180	=back
		4181
		4182	=head4 THREAD LOCKING EXAMPLE
		4183
		4184	Here is a fictitious example of how to run an event loop in a different
		4185	thread than where callbacks are being invoked and watchers are
		4186	created/added/removed.
		4187
		4188	For a real-world example, see the C<EV::Loop::Async> perl module,
		4189	which uses exactly this technique (which is suited for many high-level
		4190	languages).
		4191
		4192	The example uses a pthread mutex to protect the loop data, a condition
		4193	variable to wait for callback invocations, an async watcher to notify the
		4194	event loop thread and an unspecified mechanism to wake up the main thread.
		4195
		4196	First, you need to associate some data with the event loop:
		4197
		4198	typedef struct {
		4199	mutex_t lock; /* global loop lock */
		4200	ev_async async_w;
		4201	thread_t tid;
		4202	cond_t invoke_cv;
		4203	} userdata;
		4204
		4205	void prepare_loop (EV_P)
		4206	{
		4207	// for simplicity, we use a static userdata struct.
		4208	static userdata u;
		4209
		4210	ev_async_init (&u->async_w, async_cb);
		4211	ev_async_start (EV_A_ &u->async_w);
		4212
		4213	pthread_mutex_init (&u->lock, 0);
		4214	pthread_cond_init (&u->invoke_cv, 0);
		4215
		4216	// now associate this with the loop
		4217	ev_set_userdata (EV_A_ u);
		4218	ev_set_invoke_pending_cb (EV_A_ l_invoke);
		4219	ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
		4220
		4221	// then create the thread running ev_loop
		4222	pthread_create (&u->tid, 0, l_run, EV_A);
		4223	}
		4224
		4225	The callback for the C<ev_async> watcher does nothing: the watcher is used
		4226	solely to wake up the event loop so it takes notice of any new watchers
		4227	that might have been added:
		4228
		4229	static void
		4230	async_cb (EV_P_ ev_async *w, int revents)
		4231	{
		4232	// just used for the side effects
		4233	}
		4234
		4235	The C<l_release> and C<l_acquire> callbacks simply unlock/lock the mutex
		4236	protecting the loop data, respectively.
		4237
		4238	static void
		4239	l_release (EV_P)
		4240	{
		4241	userdata *u = ev_userdata (EV_A);
		4242	pthread_mutex_unlock (&u->lock);
		4243	}
		4244
		4245	static void
		4246	l_acquire (EV_P)
		4247	{
		4248	userdata *u = ev_userdata (EV_A);
		4249	pthread_mutex_lock (&u->lock);
		4250	}
		4251
		4252	The event loop thread first acquires the mutex, and then jumps straight
		4253	into C<ev_loop>:
		4254
		4255	void *
		4256	l_run (void *thr_arg)
		4257	{
		4258	struct ev_loop *loop = (struct ev_loop *)thr_arg;
		4259
		4260	l_acquire (EV_A);
		4261	pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
		4262	ev_loop (EV_A_ 0);
		4263	l_release (EV_A);
		4264
		4265	return 0;
		4266	}
		4267
		4268	Instead of invoking all pending watchers, the C<l_invoke> callback will
		4269	signal the main thread via some unspecified mechanism (signals? pipe
		4270	writes? C<Async::Interrupt>?) and then waits until all pending watchers
		4271	have been called (in a while loop because a) spurious wakeups are possible
		4272	and b) skipping inter-thread-communication when there are no pending
		4273	watchers is very beneficial):
		4274
		4275	static void
		4276	l_invoke (EV_P)
		4277	{
		4278	userdata *u = ev_userdata (EV_A);
		4279
		4280	while (ev_pending_count (EV_A))
		4281	{
		4282	wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
		4283	pthread_cond_wait (&u->invoke_cv, &u->lock);
		4284	}
		4285	}
		4286
		4287	Now, whenever the main thread gets told to invoke pending watchers, it
		4288	will grab the lock, call C<ev_invoke_pending> and then signal the loop
		4289	thread to continue:
		4290
		4291	static void
		4292	real_invoke_pending (EV_P)
		4293	{
		4294	userdata *u = ev_userdata (EV_A);
		4295
		4296	pthread_mutex_lock (&u->lock);
		4297	ev_invoke_pending (EV_A);
		4298	pthread_cond_signal (&u->invoke_cv);
		4299	pthread_mutex_unlock (&u->lock);
		4300	}
		4301
		4302	Whenever you want to start/stop a watcher or do other modifications to an
		4303	event loop, you will now have to lock:
		4304
		4305	ev_timer timeout_watcher;
		4306	userdata *u = ev_userdata (EV_A);
		4307
		4308	ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
		4309
		4310	pthread_mutex_lock (&u->lock);
		4311	ev_timer_start (EV_A_ &timeout_watcher);
		4312	ev_async_send (EV_A_ &u->async_w);
		4313	pthread_mutex_unlock (&u->lock);
		4314
		4315	Note that sending the C<ev_async> watcher is required because otherwise
		4316	an event loop currently blocking in the kernel will have no knowledge
		4317	about the newly added timer. By waking up the loop it will pick up any new
		4318	watchers in the next event loop iteration.
		4319
		4320	=head3 COROUTINES
		4321
		4322	Libev is very accommodating to coroutines ("cooperative threads"):
		4323	libev fully supports nesting calls to its functions from different
		4324	coroutines (e.g. you can call C<ev_loop> on the same loop from two
		4325	different coroutines, and switch freely between both coroutines running
		4326	the loop, as long as you don't confuse yourself). The only exception is
		4327	that you must not do this from C<ev_periodic> reschedule callbacks.
		4328
		4329	Care has been taken to ensure that libev does not keep local state inside
		4330	C<ev_loop>, and other calls do not usually allow for coroutine switches as
		4331	they do not call any callbacks.
		4332
		4333	=head2 COMPILER WARNINGS
		4334
		4335	Depending on your compiler and compiler settings, you might get no or a
		4336	lot of warnings when compiling libev code. Some people are apparently
		4337	scared by this.
		4338
		4339	However, these are unavoidable for many reasons. For one, each compiler
		4340	has different warnings, and each user has different tastes regarding
		4341	warning options. "Warn-free" code therefore cannot be a goal except when
		4342	targeting a specific compiler and compiler-version.
		4343
		4344	Another reason is that some compiler warnings require elaborate
		4345	workarounds, or other changes to the code that make it less clear and less
		4346	maintainable.
		4347
		4348	And of course, some compiler warnings are just plain stupid, or simply
		4349	wrong (because they don't actually warn about the condition their message
		4350	seems to warn about). For example, certain older gcc versions had some
		4351	warnings that resulted an extreme number of false positives. These have
		4352	been fixed, but some people still insist on making code warn-free with
		4353	such buggy versions.
		4354
		4355	While libev is written to generate as few warnings as possible,
		4356	"warn-free" code is not a goal, and it is recommended not to build libev
		4357	with any compiler warnings enabled unless you are prepared to cope with
		4358	them (e.g. by ignoring them). Remember that warnings are just that:
		4359	warnings, not errors, or proof of bugs.
		4360
		4361
		4362	=head2 VALGRIND
		4363
		4364	Valgrind has a special section here because it is a popular tool that is
		4365	highly useful. Unfortunately, valgrind reports are very hard to interpret.
		4366
		4367	If you think you found a bug (memory leak, uninitialised data access etc.)
		4368	in libev, then check twice: If valgrind reports something like:
		4369
		4370	==2274== definitely lost: 0 bytes in 0 blocks.
		4371	==2274== possibly lost: 0 bytes in 0 blocks.
		4372	==2274== still reachable: 256 bytes in 1 blocks.
		4373
		4374	Then there is no memory leak, just as memory accounted to global variables
		4375	is not a memleak - the memory is still being referenced, and didn't leak.
		4376
		4377	Similarly, under some circumstances, valgrind might report kernel bugs
		4378	as if it were a bug in libev (e.g. in realloc or in the poll backend,
		4379	although an acceptable workaround has been found here), or it might be
		4380	confused.
		4381
		4382	Keep in mind that valgrind is a very good tool, but only a tool. Don't
		4383	make it into some kind of religion.
		4384
		4385	If you are unsure about something, feel free to contact the mailing list
		4386	with the full valgrind report and an explanation on why you think this
		4387	is a bug in libev (best check the archives, too :). However, don't be
		4388	annoyed when you get a brisk "this is no bug" answer and take the chance
		4389	of learning how to interpret valgrind properly.
		4390
		4391	If you need, for some reason, empty reports from valgrind for your project
		4392	I suggest using suppression lists.
		4393
		4394
		4395	=head1 PORTABILITY NOTES
		4396
		4397	=head2 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
		4398
		4399	Win32 doesn't support any of the standards (e.g. POSIX) that libev
		4400	requires, and its I/O model is fundamentally incompatible with the POSIX
		4401	model. Libev still offers limited functionality on this platform in
		4402	the form of the C<EVBACKEND_SELECT> backend, and only supports socket
		4403	descriptors. This only applies when using Win32 natively, not when using
		4404	e.g. cygwin.
		4405
		4406	Lifting these limitations would basically require the full
		4407	re-implementation of the I/O system. If you are into these kinds of
		4408	things, then note that glib does exactly that for you in a very portable
		4409	way (note also that glib is the slowest event library known to man).
		4410
		4411	There is no supported compilation method available on windows except
		4412	embedding it into other applications.
		4413
		4414	Sensible signal handling is officially unsupported by Microsoft - libev
		4415	tries its best, but under most conditions, signals will simply not work.
		4416
		4417	Not a libev limitation but worth mentioning: windows apparently doesn't
		4418	accept large writes: instead of resulting in a partial write, windows will
		4419	either accept everything or return C<ENOBUFS> if the buffer is too large,
		4420	so make sure you only write small amounts into your sockets (less than a
		4421	megabyte seems safe, but this apparently depends on the amount of memory
		4422	available).
		4423
		4424	Due to the many, low, and arbitrary limits on the win32 platform and
		4425	the abysmal performance of winsockets, using a large number of sockets
		4426	is not recommended (and not reasonable). If your program needs to use
		4427	more than a hundred or so sockets, then likely it needs to use a totally
		4428	different implementation for windows, as libev offers the POSIX readiness
		4429	notification model, which cannot be implemented efficiently on windows
		4430	(due to Microsoft monopoly games).
		4431
		4432	A typical way to use libev under windows is to embed it (see the embedding
		4433	section for details) and use the following F<evwrap.h> header file instead
		4434	of F<ev.h>:
		4435
		4436	#define EV_STANDALONE /* keeps ev from requiring config.h */
		4437	#define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
		4438
		4439	#include "ev.h"
		4440
		4441	And compile the following F<evwrap.c> file into your project (make sure
		4442	you do I<not> compile the F<ev.c> or any other embedded source files!):
		4443
		4444	#include "evwrap.h"
		4445	#include "ev.c"
		4446
		4447	=over 4
		4448
		4449	=item The winsocket select function
		4450
		4451	The winsocket C<select> function doesn't follow POSIX in that it
		4452	requires socket I<handles> and not socket I<file descriptors> (it is
		4453	also extremely buggy). This makes select very inefficient, and also
		4454	requires a mapping from file descriptors to socket handles (the Microsoft
		4455	C runtime provides the function C<_open_osfhandle> for this). See the
		4456	discussion of the C<EV_SELECT_USE_FD_SET>, C<EV_SELECT_IS_WINSOCKET> and
		4457	C<EV_FD_TO_WIN32_HANDLE> preprocessor symbols for more info.
		4458
		4459	The configuration for a "naked" win32 using the Microsoft runtime
		4460	libraries and raw winsocket select is:
		4461
		4462	#define EV_USE_SELECT 1
		4463	#define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
		4464
		4465	Note that winsockets handling of fd sets is O(n), so you can easily get a
		4466	complexity in the O(n²) range when using win32.
		4467
		4468	=item Limited number of file descriptors
		4469
		4470	Windows has numerous arbitrary (and low) limits on things.
		4471
		4472	Early versions of winsocket's select only supported waiting for a maximum
		4473	of C<64> handles (probably owning to the fact that all windows kernels
		4474	can only wait for C<64> things at the same time internally; Microsoft
		4475	recommends spawning a chain of threads and wait for 63 handles and the
		4476	previous thread in each. Sounds great!).
		4477
		4478	Newer versions support more handles, but you need to define C<FD_SETSIZE>
		4479	to some high number (e.g. C<2048>) before compiling the winsocket select
		4480	call (which might be in libev or elsewhere, for example, perl and many
		4481	other interpreters do their own select emulation on windows).
		4482
		4483	Another limit is the number of file descriptors in the Microsoft runtime
		4484	libraries, which by default is C<64> (there must be a hidden I<64>
		4485	fetish or something like this inside Microsoft). You can increase this
		4486	by calling C<_setmaxstdio>, which can increase this limit to C<2048>
		4487	(another arbitrary limit), but is broken in many versions of the Microsoft
		4488	runtime libraries. This might get you to about C<512> or C<2048> sockets
		4489	(depending on windows version and/or the phase of the moon). To get more,
		4490	you need to wrap all I/O functions and provide your own fd management, but
		4491	the cost of calling select (O(n²)) will likely make this unworkable.
		4492
		4493	=back
		4494
		4495	=head2 PORTABILITY REQUIREMENTS
		4496
		4497	In addition to a working ISO-C implementation and of course the
		4498	backend-specific APIs, libev relies on a few additional extensions:
		4499
		4500	=over 4
		4501
		4502	=item C<void (*)(ev_watcher_type *, int revents)> must have compatible
		4503	calling conventions regardless of C<ev_watcher_type *>.
		4504
		4505	Libev assumes not only that all watcher pointers have the same internal
		4506	structure (guaranteed by POSIX but not by ISO C for example), but it also
		4507	assumes that the same (machine) code can be used to call any watcher
		4508	callback: The watcher callbacks have different type signatures, but libev
		4509	calls them using an C<ev_watcher *> internally.
		4510
		4511	=item C<sig_atomic_t volatile> must be thread-atomic as well
		4512
		4513	The type C<sig_atomic_t volatile> (or whatever is defined as
		4514	C<EV_ATOMIC_T>) must be atomic with respect to accesses from different
		4515	threads. This is not part of the specification for C<sig_atomic_t>, but is
		4516	believed to be sufficiently portable.
		4517
		4518	=item C<sigprocmask> must work in a threaded environment
		4519
		4520	Libev uses C<sigprocmask> to temporarily block signals. This is not
		4521	allowed in a threaded program (C<pthread_sigmask> has to be used). Typical
		4522	pthread implementations will either allow C<sigprocmask> in the "main
		4523	thread" or will block signals process-wide, both behaviours would
		4524	be compatible with libev. Interaction between C<sigprocmask> and
		4525	C<pthread_sigmask> could complicate things, however.
		4526
		4527	The most portable way to handle signals is to block signals in all threads
		4528	except the initial one, and run the default loop in the initial thread as
		4529	well.
		4530
		4531	=item C<long> must be large enough for common memory allocation sizes
		4532
		4533	To improve portability and simplify its API, libev uses C<long> internally
		4534	instead of C<size_t> when allocating its data structures. On non-POSIX
		4535	systems (Microsoft...) this might be unexpectedly low, but is still at
		4536	least 31 bits everywhere, which is enough for hundreds of millions of
		4537	watchers.
		4538
		4539	=item C<double> must hold a time value in seconds with enough accuracy
		4540
		4541	The type C<double> is used to represent timestamps. It is required to
		4542	have at least 51 bits of mantissa (and 9 bits of exponent), which is good
		4543	enough for at least into the year 4000. This requirement is fulfilled by
		4544	implementations implementing IEEE 754, which is basically all existing
		4545	ones. With IEEE 754 doubles, you get microsecond accuracy until at least
		4546	2200.
		4547
		4548	=back
		4549
		4550	If you know of other additional requirements drop me a note.
		4551
		4552
		4553	=head1 ALGORITHMIC COMPLEXITIES
		4554
		4555	In this section the complexities of (many of) the algorithms used inside
		4556	libev will be documented. For complexity discussions about backends see
		4557	the documentation for C<ev_default_init>.
		4558
		4559	All of the following are about amortised time: If an array needs to be
		4560	extended, libev needs to realloc and move the whole array, but this
		4561	happens asymptotically rarer with higher number of elements, so O(1) might
		4562	mean that libev does a lengthy realloc operation in rare cases, but on
		4563	average it is much faster and asymptotically approaches constant time.
		4564
		4565	=over 4
		4566
		4567	=item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
		4568
		4569	This means that, when you have a watcher that triggers in one hour and
		4570	there are 100 watchers that would trigger before that, then inserting will
		4571	have to skip roughly seven (C<ld 100>) of these watchers.
		4572
		4573	=item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
		4574
		4575	That means that changing a timer costs less than removing/adding them,
		4576	as only the relative motion in the event queue has to be paid for.
		4577
		4578	=item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
		4579
		4580	These just add the watcher into an array or at the head of a list.
		4581
		4582	=item Stopping check/prepare/idle/fork/async watchers: O(1)
		4583
		4584	=item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
		4585
		4586	These watchers are stored in lists, so they need to be walked to find the
		4587	correct watcher to remove. The lists are usually short (you don't usually
		4588	have many watchers waiting for the same fd or signal: one is typical, two
		4589	is rare).
		4590
		4591	=item Finding the next timer in each loop iteration: O(1)
		4592
		4593	By virtue of using a binary or 4-heap, the next timer is always found at a
		4594	fixed position in the storage array.
		4595
		4596	=item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
		4597
		4598	A change means an I/O watcher gets started or stopped, which requires
		4599	libev to recalculate its status (and possibly tell the kernel, depending
		4600	on backend and whether C<ev_io_set> was used).
		4601
		4602	=item Activating one watcher (putting it into the pending state): O(1)
		4603
		4604	=item Priority handling: O(number_of_priorities)
		4605
		4606	Priorities are implemented by allocating some space for each
		4607	priority. When doing priority-based operations, libev usually has to
		4608	linearly search all the priorities, but starting/stopping and activating
		4609	watchers becomes O(1) with respect to priority handling.
		4610
		4611	=item Sending an ev_async: O(1)
		4612
		4613	=item Processing ev_async_send: O(number_of_async_watchers)
		4614
		4615	=item Processing signals: O(max_signal_number)
		4616
		4617	Sending involves a system call I<iff> there were no other C<ev_async_send>
		4618	calls in the current loop iteration. Checking for async and signal events
		4619	involves iterating over all running async watchers or all signal numbers.
		4620
		4621	=back
		4622
		4623
		4624	=head1 GLOSSARY
		4625
		4626	=over 4
		4627
		4628	=item active
		4629
		4630	A watcher is active as long as it has been started (has been attached to
		4631	an event loop) but not yet stopped (disassociated from the event loop).
		4632
		4633	=item application
		4634
		4635	In this document, an application is whatever is using libev.
		4636
		4637	=item callback
		4638
		4639	The address of a function that is called when some event has been
		4640	detected. Callbacks are being passed the event loop, the watcher that
		4641	received the event, and the actual event bitset.
		4642
		4643	=item callback invocation
		4644
		4645	The act of calling the callback associated with a watcher.
		4646
		4647	=item event
		4648
		4649	A change of state of some external event, such as data now being available
		4650	for reading on a file descriptor, time having passed or simply not having
		4651	any other events happening anymore.
		4652
		4653	In libev, events are represented as single bits (such as C<EV_READ> or
		4654	C<EV_TIMEOUT>).
		4655
		4656	=item event library
		4657
		4658	A software package implementing an event model and loop.
		4659
		4660	=item event loop
		4661
		4662	An entity that handles and processes external events and converts them
		4663	into callback invocations.
		4664
		4665	=item event model
		4666
		4667	The model used to describe how an event loop handles and processes
		4668	watchers and events.
		4669
		4670	=item pending
		4671
		4672	A watcher is pending as soon as the corresponding event has been detected,
		4673	and stops being pending as soon as the watcher will be invoked or its
		4674	pending status is explicitly cleared by the application.
		4675
		4676	A watcher can be pending, but not active. Stopping a watcher also clears
		4677	its pending status.
		4678
		4679	=item real time
		4680
		4681	The physical time that is observed. It is apparently strictly monotonic :)
		4682
		4683	=item wall-clock time
		4684
		4685	The time and date as shown on clocks. Unlike real time, it can actually
		4686	be wrong and jump forwards and backwards, e.g. when the you adjust your
		4687	clock.
		4688
		4689	=item watcher
		4690
		4691	A data structure that describes interest in certain events. Watchers need
		4692	to be started (attached to an event loop) before they can receive events.
		4693
		4694	=item watcher invocation
		4695
		4696	The act of calling the callback associated with a watcher.
		4697
		4698	=back
1217		4699
1218	=head1 AUTHOR	4700	=head1 AUTHOR
1219		4701
1220	Marc Lehmann <libev@schmorp.de>.	4702	Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael Magnusson.
1221		4703

Diff Legend

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