ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/AnyEvent/lib/AnyEvent.pm

Comparing AnyEvent/lib/AnyEvent.pm (file contents):
Revision 1.28 by root, Sat Oct 27 15:10:09 2007 UTC vs.
Revision 1.168 by root, Tue Jul 8 23:53:37 2008 UTC

1	=head1 NAME	1	=head1 NAME
2		2
3	AnyEvent - provide framework for multiple event loops	3	AnyEvent - provide framework for multiple event loops
4		4
5	Event, Coro, Glib, Tk, Perl - various supported event loops	5	EV, Event, Glib, Tk, Perl, Event::Lib, Qt, POE - various supported event loops
6		6
7	=head1 SYNOPSIS	7	=head1 SYNOPSIS
8		8
9	use AnyEvent;	9	use AnyEvent;
10		10
…		…
14		14
15	my $w = AnyEvent->timer (after => $seconds, cb => sub {	15	my $w = AnyEvent->timer (after => $seconds, cb => sub {
16	...	16	...
17	});	17	});
18		18
19	my $w = AnyEvent->condvar; # stores wether a condition was flagged	19	my $w = AnyEvent->condvar; # stores whether a condition was flagged
		20	$w->send; # wake up current and all future recv's
20	$w->wait; # enters "main loop" till $condvar gets ->broadcast	21	$w->recv; # enters "main loop" till $condvar gets ->send
21	$w->broadcast; # wake up current and all future wait's	22
		23	=head1 INTRODUCTION/TUTORIAL
		24
		25	This manpage is mainly a reference manual. If you are interested
		26	in a tutorial or some gentle introduction, have a look at the
		27	L<AnyEvent::Intro> manpage.
		28
		29	=head1 WHY YOU SHOULD USE THIS MODULE (OR NOT)
		30
		31	Glib, POE, IO::Async, Event... CPAN offers event models by the dozen
		32	nowadays. So what is different about AnyEvent?
		33
		34	Executive Summary: AnyEvent is I<compatible>, AnyEvent is I<free of
		35	policy> and AnyEvent is I<small and efficient>.
		36
		37	First and foremost, I<AnyEvent is not an event model> itself, it only
		38	interfaces to whatever event model the main program happens to use, in a
		39	pragmatic way. For event models and certain classes of immortals alike,
		40	the statement "there can only be one" is a bitter reality: In general,
		41	only one event loop can be active at the same time in a process. AnyEvent
		42	cannot change this, but it can hide the differences between those event
		43	loops.
		44
		45	The goal of AnyEvent is to offer module authors the ability to do event
		46	programming (waiting for I/O or timer events) without subscribing to a
		47	religion, a way of living, and most importantly: without forcing your
		48	module users into the same thing by forcing them to use the same event
		49	model you use.
		50
		51	For modules like POE or IO::Async (which is a total misnomer as it is
		52	actually doing all I/O I<synchronously>...), using them in your module is
		53	like joining a cult: After you joined, you are dependent on them and you
		54	cannot use anything else, as they are simply incompatible to everything
		55	that isn't them. What's worse, all the potential users of your
		56	module are I<also> forced to use the same event loop you use.
		57
		58	AnyEvent is different: AnyEvent + POE works fine. AnyEvent + Glib works
		59	fine. AnyEvent + Tk works fine etc. etc. but none of these work together
		60	with the rest: POE + IO::Async? No go. Tk + Event? No go. Again: if
		61	your module uses one of those, every user of your module has to use it,
		62	too. But if your module uses AnyEvent, it works transparently with all
		63	event models it supports (including stuff like IO::Async, as long as those
		64	use one of the supported event loops. It is trivial to add new event loops
		65	to AnyEvent, too, so it is future-proof).
		66
		67	In addition to being free of having to use I<the one and only true event
		68	model>, AnyEvent also is free of bloat and policy: with POE or similar
		69	modules, you get an enormous amount of code and strict rules you have to
		70	follow. AnyEvent, on the other hand, is lean and up to the point, by only
		71	offering the functionality that is necessary, in as thin as a wrapper as
		72	technically possible.
		73
		74	Of course, AnyEvent comes with a big (and fully optional!) toolbox
		75	of useful functionality, such as an asynchronous DNS resolver, 100%
		76	non-blocking connects (even with TLS/SSL, IPv6 and on broken platforms
		77	such as Windows) and lots of real-world knowledge and workarounds for
		78	platform bugs and differences.
		79
		80	Now, if you I<do want> lots of policy (this can arguably be somewhat
		81	useful) and you want to force your users to use the one and only event
		82	model, you should I<not> use this module.
22		83
23	=head1 DESCRIPTION	84	=head1 DESCRIPTION
24		85
25	L<AnyEvent> provides an identical interface to multiple event loops. This	86	L<AnyEvent> provides an identical interface to multiple event loops. This
26	allows module authors to utilise an event loop without forcing module	87	allows module authors to utilise an event loop without forcing module
27	users to use the same event loop (as only a single event loop can coexist	88	users to use the same event loop (as only a single event loop can coexist
28	peacefully at any one time).	89	peacefully at any one time).
29		90
30	The interface itself is vaguely similar but not identical to the Event	91	The interface itself is vaguely similar, but not identical to the L<Event>
31	module.	92	module.
32		93
33	On the first call of any method, the module tries to detect the currently	94	During the first call of any watcher-creation method, the module tries
34	loaded event loop by probing wether any of the following modules is	95	to detect the currently loaded event loop by probing whether one of the
35	loaded: L<Coro::Event>, L<Event>, L<Glib>, L<Tk>. The first one found is	96	following modules is already loaded: L<EV>,
36	used. If none is found, the module tries to load these modules in the	97	L<Event>, L<Glib>, L<AnyEvent::Impl::Perl>, L<Tk>, L<Event::Lib>, L<Qt>,
37	order given. The first one that could be successfully loaded will be	98	L<POE>. The first one found is used. If none are found, the module tries
38	used. If still none could be found, AnyEvent will fall back to a pure-perl	99	to load these modules (excluding Tk, Event::Lib, Qt and POE as the pure perl
39	event loop, which is also not very efficient.	100	adaptor should always succeed) in the order given. The first one that can
		101	be successfully loaded will be used. If, after this, still none could be
		102	found, AnyEvent will fall back to a pure-perl event loop, which is not
		103	very efficient, but should work everywhere.
40		104
41	Because AnyEvent first checks for modules that are already loaded, loading	105	Because AnyEvent first checks for modules that are already loaded, loading
42	an Event model explicitly before first using AnyEvent will likely make	106	an event model explicitly before first using AnyEvent will likely make
43	that model the default. For example:	107	that model the default. For example:
44		108
45	use Tk;	109	use Tk;
46	use AnyEvent;	110	use AnyEvent;
47		111
48	# .. AnyEvent will likely default to Tk	112	# .. AnyEvent will likely default to Tk
49		113
		114	The I<likely> means that, if any module loads another event model and
		115	starts using it, all bets are off. Maybe you should tell their authors to
		116	use AnyEvent so their modules work together with others seamlessly...
		117
50	The pure-perl implementation of AnyEvent is called	118	The pure-perl implementation of AnyEvent is called
51	C<AnyEvent::Impl::Perl>. Like other event modules you can load it	119	C<AnyEvent::Impl::Perl>. Like other event modules you can load it
52	explicitly.	120	explicitly and enjoy the high availability of that event loop :)
53		121
54	=head1 WATCHERS	122	=head1 WATCHERS
55		123
56	AnyEvent has the central concept of a I<watcher>, which is an object that	124	AnyEvent has the central concept of a I<watcher>, which is an object that
57	stores relevant data for each kind of event you are waiting for, such as	125	stores relevant data for each kind of event you are waiting for, such as
58	the callback to call, the filehandle to watch, etc.	126	the callback to call, the file handle to watch, etc.
59		127
60	These watchers are normal Perl objects with normal Perl lifetime. After	128	These watchers are normal Perl objects with normal Perl lifetime. After
61	creating a watcher it will immediately "watch" for events and invoke	129	creating a watcher it will immediately "watch" for events and invoke the
		130	callback when the event occurs (of course, only when the event model
		131	is in control).
		132
62	the callback. To disable the watcher you have to destroy it (e.g. by	133	To disable the watcher you have to destroy it (e.g. by setting the
63	setting the variable that stores it to C<undef> or otherwise deleting all	134	variable you store it in to C<undef> or otherwise deleting all references
64	references to it).	135	to it).
65		136
66	All watchers are created by calling a method on the C<AnyEvent> class.	137	All watchers are created by calling a method on the C<AnyEvent> class.
67		138
		139	Many watchers either are used with "recursion" (repeating timers for
		140	example), or need to refer to their watcher object in other ways.
		141
		142	An any way to achieve that is this pattern:
		143
		144	my $w; $w = AnyEvent->type (arg => value ..., cb => sub {
		145	# you can use $w here, for example to undef it
		146	undef $w;
		147	});
		148
		149	Note that C<my $w; $w => combination. This is necessary because in Perl,
		150	my variables are only visible after the statement in which they are
		151	declared.
		152
68	=head2 IO WATCHERS	153	=head2 I/O WATCHERS
69		154
70	You can create I/O watcher by calling the C<< AnyEvent->io >> method with	155	You can create an I/O watcher by calling the C<< AnyEvent->io >> method
71	the following mandatory arguments:	156	with the following mandatory key-value pairs as arguments:
72		157
73	C<fh> the Perl I<filehandle> (not filedescriptor) to watch for	158	C<fh> the Perl I<file handle> (I<not> file descriptor) to watch for events
		159	(AnyEvent might or might not keep a reference to this file handle). C<poll>
74	events. C<poll> must be a string that is either C<r> or C<w>, that creates	160	must be a string that is either C<r> or C<w>, which creates a watcher
75	a watcher waiting for "r"eadable or "w"ritable events. C<cb> teh callback	161	waiting for "r"eadable or "w"ritable events, respectively. C<cb> is the
76	to invoke everytime the filehandle becomes ready.	162	callback to invoke each time the file handle becomes ready.
77		163
78	Only one io watcher per C<fh> and C<poll> combination is allowed (i.e. on	164	Although the callback might get passed parameters, their value and
79	a socket you can have one r + one w, not any more (limitation comes from	165	presence is undefined and you cannot rely on them. Portable AnyEvent
80	Tk - if you are sure you are not using Tk this limitation is gone).	166	callbacks cannot use arguments passed to I/O watcher callbacks.
81		167
82	Filehandles will be kept alive, so as long as the watcher exists, the	168	The I/O watcher might use the underlying file descriptor or a copy of it.
83	filehandle exists, too.	169	You must not close a file handle as long as any watcher is active on the
		170	underlying file descriptor.
84		171
85	Example:	172	Some event loops issue spurious readyness notifications, so you should
		173	always use non-blocking calls when reading/writing from/to your file
		174	handles.
86		175
87	# wait for readability of STDIN, then read a line and disable the watcher	176	Example: wait for readability of STDIN, then read a line and disable the
		177	watcher.
		178
88	my $w; $w = AnyEvent->io (fh => \*STDIN, poll => 'r', cb => sub {	179	my $w; $w = AnyEvent->io (fh => \*STDIN, poll => 'r', cb => sub {
89	chomp (my $input = <STDIN>);	180	chomp (my $input = <STDIN>);
90	warn "read: $input\n";	181	warn "read: $input\n";
91	undef $w;	182	undef $w;
92	});	183	});
…		…
94	=head2 TIME WATCHERS	185	=head2 TIME WATCHERS
95		186
96	You can create a time watcher by calling the C<< AnyEvent->timer >>	187	You can create a time watcher by calling the C<< AnyEvent->timer >>
97	method with the following mandatory arguments:	188	method with the following mandatory arguments:
98		189
99	C<after> after how many seconds (fractions are supported) should the timer	190	C<after> specifies after how many seconds (fractional values are
100	activate. C<cb> the callback to invoke.	191	supported) the callback should be invoked. C<cb> is the callback to invoke
		192	in that case.
101		193
102	The timer callback will be invoked at most once: if you want a repeating	194	Although the callback might get passed parameters, their value and
103	timer you have to create a new watcher (this is a limitation by both Tk	195	presence is undefined and you cannot rely on them. Portable AnyEvent
104	and Glib).	196	callbacks cannot use arguments passed to time watcher callbacks.
105		197
106	Example:	198	The callback will normally be invoked once only. If you specify another
		199	parameter, C<interval>, as a strictly positive number (> 0), then the
		200	callback will be invoked regularly at that interval (in fractional
		201	seconds) after the first invocation. If C<interval> is specified with a
		202	false value, then it is treated as if it were missing.
107		203
		204	The callback will be rescheduled before invoking the callback, but no
		205	attempt is done to avoid timer drift in most backends, so the interval is
		206	only approximate.
		207
108	# fire an event after 7.7 seconds	208	Example: fire an event after 7.7 seconds.
		209
109	my $w = AnyEvent->timer (after => 7.7, cb => sub {	210	my $w = AnyEvent->timer (after => 7.7, cb => sub {
110	warn "timeout\n";	211	warn "timeout\n";
111	});	212	});
112		213
113	# to cancel the timer:	214	# to cancel the timer:
114	undef $w	215	undef $w;
115		216
116	=head2 CONDITION WATCHERS	217	Example 2: fire an event after 0.5 seconds, then roughly every second.
117		218
118	Condition watchers can be created by calling the C<< AnyEvent->condvar >>	219	my $w = AnyEvent->timer (after => 0.5, interval => 1, cb => sub {
119	method without any arguments.	220	warn "timeout\n";
		221	};
120		222
121	A condition watcher watches for a condition - precisely that the C<<	223	=head3 TIMING ISSUES
122	->broadcast >> method has been called.
123		224
124	The watcher has only two methods:	225	There are two ways to handle timers: based on real time (relative, "fire
		226	in 10 seconds") and based on wallclock time (absolute, "fire at 12
		227	o'clock").
		228
		229	While most event loops expect timers to specified in a relative way, they
		230	use absolute time internally. This makes a difference when your clock
		231	"jumps", for example, when ntp decides to set your clock backwards from
		232	the wrong date of 2014-01-01 to 2008-01-01, a watcher that is supposed to
		233	fire "after" a second might actually take six years to finally fire.
		234
		235	AnyEvent cannot compensate for this. The only event loop that is conscious
		236	about these issues is L<EV>, which offers both relative (ev_timer, based
		237	on true relative time) and absolute (ev_periodic, based on wallclock time)
		238	timers.
		239
		240	AnyEvent always prefers relative timers, if available, matching the
		241	AnyEvent API.
		242
		243	AnyEvent has two additional methods that return the "current time":
125		244
126	=over 4	245	=over 4
127		246
128	=item $cv->wait	247	=item AnyEvent->time
129		248
130	Wait (blocking if necessary) until the C<< ->broadcast >> method has been	249	This returns the "current wallclock time" as a fractional number of
131	called on c<$cv>, while servicing other watchers normally.	250	seconds since the Epoch (the same thing as C<time> or C<Time::HiRes::time>
		251	return, and the result is guaranteed to be compatible with those).
132		252
133	Not all event models support a blocking wait - some die in that case, so	253	It progresses independently of any event loop processing, i.e. each call
134	if you are using this from a module, never require a blocking wait, but	254	will check the system clock, which usually gets updated frequently.
135	let the caller decide wether the call will block or not (for example,
136	by coupling condition variables with some kind of request results and
137	supporting callbacks so the caller knows that getting the result will not
138	block, while still suppporting blockign waits if the caller so desires).
139		255
140	You can only wait once on a condition - additional calls will return	256	=item AnyEvent->now
141	immediately.
142		257
143	=item $cv->broadcast	258	This also returns the "current wallclock time", but unlike C<time>, above,
		259	this value might change only once per event loop iteration, depending on
		260	the event loop (most return the same time as C<time>, above). This is the
		261	time that AnyEvent's timers get scheduled against.
144		262
145	Flag the condition as ready - a running C<< ->wait >> and all further	263	I<In almost all cases (in all cases if you don't care), this is the
146	calls to C<wait> will return after this method has been called. If nobody	264	function to call when you want to know the current time.>
147	is waiting the broadcast will be remembered..
148		265
149	Example:	266	This function is also often faster then C<< AnyEvent->time >>, and
		267	thus the preferred method if you want some timestamp (for example,
		268	L<AnyEvent::Handle> uses this to update it's activity timeouts).
		269
		270	The rest of this section is only of relevance if you try to be very exact
		271	with your timing, you can skip it without bad conscience.
		272
		273	For a practical example of when these times differ, consider L<Event::Lib>
		274	and L<EV> and the following set-up:
		275
		276	The event loop is running and has just invoked one of your callback at
		277	time=500 (assume no other callbacks delay processing). In your callback,
		278	you wait a second by executing C<sleep 1> (blocking the process for a
		279	second) and then (at time=501) you create a relative timer that fires
		280	after three seconds.
		281
		282	With L<Event::Lib>, C<< AnyEvent->time >> and C<< AnyEvent->now >> will
		283	both return C<501>, because that is the current time, and the timer will
		284	be scheduled to fire at time=504 (C<501> + C<3>).
		285
		286	With L<EV>, C<< AnyEvent->time >> returns C<501> (as that is the current
		287	time), but C<< AnyEvent->now >> returns C<500>, as that is the time the
		288	last event processing phase started. With L<EV>, your timer gets scheduled
		289	to run at time=503 (C<500> + C<3>).
		290
		291	In one sense, L<Event::Lib> is more exact, as it uses the current time
		292	regardless of any delays introduced by event processing. However, most
		293	callbacks do not expect large delays in processing, so this causes a
		294	higher drift (and a lot more system calls to get the current time).
		295
		296	In another sense, L<EV> is more exact, as your timer will be scheduled at
		297	the same time, regardless of how long event processing actually took.
		298
		299	In either case, if you care (and in most cases, you don't), then you
		300	can get whatever behaviour you want with any event loop, by taking the
		301	difference between C<< AnyEvent->time >> and C<< AnyEvent->now >> into
		302	account.
		303
		304	=back
		305
		306	=head2 SIGNAL WATCHERS
		307
		308	You can watch for signals using a signal watcher, C<signal> is the signal
		309	I<name> in uppercase and without any C<SIG> prefix, C<cb> is the Perl
		310	callback to be invoked whenever a signal occurs.
		311
		312	Although the callback might get passed parameters, their value and
		313	presence is undefined and you cannot rely on them. Portable AnyEvent
		314	callbacks cannot use arguments passed to signal watcher callbacks.
		315
		316	Multiple signal occurrences can be clumped together into one callback
		317	invocation, and callback invocation will be synchronous. Synchronous means
		318	that it might take a while until the signal gets handled by the process,
		319	but it is guaranteed not to interrupt any other callbacks.
		320
		321	The main advantage of using these watchers is that you can share a signal
		322	between multiple watchers.
		323
		324	This watcher might use C<%SIG>, so programs overwriting those signals
		325	directly will likely not work correctly.
		326
		327	Example: exit on SIGINT
		328
		329	my $w = AnyEvent->signal (signal => "INT", cb => sub { exit 1 });
		330
		331	=head2 CHILD PROCESS WATCHERS
		332
		333	You can also watch on a child process exit and catch its exit status.
		334
		335	The child process is specified by the C<pid> argument (if set to C<0>, it
		336	watches for any child process exit). The watcher will trigger as often
		337	as status change for the child are received. This works by installing a
		338	signal handler for C<SIGCHLD>. The callback will be called with the pid
		339	and exit status (as returned by waitpid), so unlike other watcher types,
		340	you I<can> rely on child watcher callback arguments.
		341
		342	There is a slight catch to child watchers, however: you usually start them
		343	I<after> the child process was created, and this means the process could
		344	have exited already (and no SIGCHLD will be sent anymore).
		345
		346	Not all event models handle this correctly (POE doesn't), but even for
		347	event models that I<do> handle this correctly, they usually need to be
		348	loaded before the process exits (i.e. before you fork in the first place).
		349
		350	This means you cannot create a child watcher as the very first thing in an
		351	AnyEvent program, you I<have> to create at least one watcher before you
		352	C<fork> the child (alternatively, you can call C<AnyEvent::detect>).
		353
		354	Example: fork a process and wait for it
		355
		356	my $done = AnyEvent->condvar;
		357
		358	my $pid = fork or exit 5;
		359
		360	my $w = AnyEvent->child (
		361	pid => $pid,
		362	cb => sub {
		363	my ($pid, $status) = @_;
		364	warn "pid $pid exited with status $status";
		365	$done->send;
		366	},
		367	);
		368
		369	# do something else, then wait for process exit
		370	$done->recv;
		371
		372	=head2 CONDITION VARIABLES
		373
		374	If you are familiar with some event loops you will know that all of them
		375	require you to run some blocking "loop", "run" or similar function that
		376	will actively watch for new events and call your callbacks.
		377
		378	AnyEvent is different, it expects somebody else to run the event loop and
		379	will only block when necessary (usually when told by the user).
		380
		381	The instrument to do that is called a "condition variable", so called
		382	because they represent a condition that must become true.
		383
		384	Condition variables can be created by calling the C<< AnyEvent->condvar
		385	>> method, usually without arguments. The only argument pair allowed is
		386	C<cb>, which specifies a callback to be called when the condition variable
		387	becomes true.
		388
		389	After creation, the condition variable is "false" until it becomes "true"
		390	by calling the C<send> method (or calling the condition variable as if it
		391	were a callback, read about the caveats in the description for the C<<
		392	->send >> method).
		393
		394	Condition variables are similar to callbacks, except that you can
		395	optionally wait for them. They can also be called merge points - points
		396	in time where multiple outstanding events have been processed. And yet
		397	another way to call them is transactions - each condition variable can be
		398	used to represent a transaction, which finishes at some point and delivers
		399	a result.
		400
		401	Condition variables are very useful to signal that something has finished,
		402	for example, if you write a module that does asynchronous http requests,
		403	then a condition variable would be the ideal candidate to signal the
		404	availability of results. The user can either act when the callback is
		405	called or can synchronously C<< ->recv >> for the results.
		406
		407	You can also use them to simulate traditional event loops - for example,
		408	you can block your main program until an event occurs - for example, you
		409	could C<< ->recv >> in your main program until the user clicks the Quit
		410	button of your app, which would C<< ->send >> the "quit" event.
		411
		412	Note that condition variables recurse into the event loop - if you have
		413	two pieces of code that call C<< ->recv >> in a round-robin fashion, you
		414	lose. Therefore, condition variables are good to export to your caller, but
		415	you should avoid making a blocking wait yourself, at least in callbacks,
		416	as this asks for trouble.
		417
		418	Condition variables are represented by hash refs in perl, and the keys
		419	used by AnyEvent itself are all named C<_ae_XXX> to make subclassing
		420	easy (it is often useful to build your own transaction class on top of
		421	AnyEvent). To subclass, use C<AnyEvent::CondVar> as base class and call
		422	it's C<new> method in your own C<new> method.
		423
		424	There are two "sides" to a condition variable - the "producer side" which
		425	eventually calls C<< -> send >>, and the "consumer side", which waits
		426	for the send to occur.
		427
		428	Example: wait for a timer.
150		429
151	# wait till the result is ready	430	# wait till the result is ready
152	my $result_ready = AnyEvent->condvar;	431	my $result_ready = AnyEvent->condvar;
153		432
154	# do something such as adding a timer	433	# do something such as adding a timer
155	# or socket watcher the calls $result_ready->broadcast	434	# or socket watcher the calls $result_ready->send
156	# when the "result" is ready.	435	# when the "result" is ready.
		436	# in this case, we simply use a timer:
		437	my $w = AnyEvent->timer (
		438	after => 1,
		439	cb => sub { $result_ready->send },
		440	);
157		441
		442	# this "blocks" (while handling events) till the callback
		443	# calls send
158	$result_ready->wait;	444	$result_ready->recv;
		445
		446	Example: wait for a timer, but take advantage of the fact that
		447	condition variables are also code references.
		448
		449	my $done = AnyEvent->condvar;
		450	my $delay = AnyEvent->timer (after => 5, cb => $done);
		451	$done->recv;
		452
		453	=head3 METHODS FOR PRODUCERS
		454
		455	These methods should only be used by the producing side, i.e. the
		456	code/module that eventually sends the signal. Note that it is also
		457	the producer side which creates the condvar in most cases, but it isn't
		458	uncommon for the consumer to create it as well.
		459
		460	=over 4
		461
		462	=item $cv->send (...)
		463
		464	Flag the condition as ready - a running C<< ->recv >> and all further
		465	calls to C<recv> will (eventually) return after this method has been
		466	called. If nobody is waiting the send will be remembered.
		467
		468	If a callback has been set on the condition variable, it is called
		469	immediately from within send.
		470
		471	Any arguments passed to the C<send> call will be returned by all
		472	future C<< ->recv >> calls.
		473
		474	Condition variables are overloaded so one can call them directly
		475	(as a code reference). Calling them directly is the same as calling
		476	C<send>. Note, however, that many C-based event loops do not handle
		477	overloading, so as tempting as it may be, passing a condition variable
		478	instead of a callback does not work. Both the pure perl and EV loops
		479	support overloading, however, as well as all functions that use perl to
		480	invoke a callback (as in L<AnyEvent::Socket> and L<AnyEvent::DNS> for
		481	example).
		482
		483	=item $cv->croak ($error)
		484
		485	Similar to send, but causes all call's to C<< ->recv >> to invoke
		486	C<Carp::croak> with the given error message/object/scalar.
		487
		488	This can be used to signal any errors to the condition variable
		489	user/consumer.
		490
		491	=item $cv->begin ([group callback])
		492
		493	=item $cv->end
		494
		495	These two methods are EXPERIMENTAL and MIGHT CHANGE.
		496
		497	These two methods can be used to combine many transactions/events into
		498	one. For example, a function that pings many hosts in parallel might want
		499	to use a condition variable for the whole process.
		500
		501	Every call to C<< ->begin >> will increment a counter, and every call to
		502	C<< ->end >> will decrement it. If the counter reaches C<0> in C<< ->end
		503	>>, the (last) callback passed to C<begin> will be executed. That callback
		504	is I<supposed> to call C<< ->send >>, but that is not required. If no
		505	callback was set, C<send> will be called without any arguments.
		506
		507	Let's clarify this with the ping example:
		508
		509	my $cv = AnyEvent->condvar;
		510
		511	my %result;
		512	$cv->begin (sub { $cv->send (\%result) });
		513
		514	for my $host (@list_of_hosts) {
		515	$cv->begin;
		516	ping_host_then_call_callback $host, sub {
		517	$result{$host} = ...;
		518	$cv->end;
		519	};
		520	}
		521
		522	$cv->end;
		523
		524	This code fragment supposedly pings a number of hosts and calls
		525	C<send> after results for all then have have been gathered - in any
		526	order. To achieve this, the code issues a call to C<begin> when it starts
		527	each ping request and calls C<end> when it has received some result for
		528	it. Since C<begin> and C<end> only maintain a counter, the order in which
		529	results arrive is not relevant.
		530
		531	There is an additional bracketing call to C<begin> and C<end> outside the
		532	loop, which serves two important purposes: first, it sets the callback
		533	to be called once the counter reaches C<0>, and second, it ensures that
		534	C<send> is called even when C<no> hosts are being pinged (the loop
		535	doesn't execute once).
		536
		537	This is the general pattern when you "fan out" into multiple subrequests:
		538	use an outer C<begin>/C<end> pair to set the callback and ensure C<end>
		539	is called at least once, and then, for each subrequest you start, call
		540	C<begin> and for each subrequest you finish, call C<end>.
159		541
160	=back	542	=back
161		543
162	=head2 SIGNAL WATCHERS	544	=head3 METHODS FOR CONSUMERS
163		545
164	You can listen for signals using a signal watcher, C<signal> is the signal	546	These methods should only be used by the consuming side, i.e. the
165	I<name> without any C<SIG> prefix. Multiple signals events can be clumped	547	code awaits the condition.
166	together into one callback invocation, and callback invocation might or
167	might not be asynchronous.
168		548
169	These watchers might use C<%SIG>, so programs overwriting those signals	549	=over 4
170	directly will likely not work correctly.
171		550
172	Example: exit on SIGINT	551	=item $cv->recv
173		552
174	my $w = AnyEvent->signal (signal => "INT", cb => sub { exit 1 });	553	Wait (blocking if necessary) until the C<< ->send >> or C<< ->croak
		554	>> methods have been called on c<$cv>, while servicing other watchers
		555	normally.
175		556
176	=head2 CHILD PROCESS WATCHERS	557	You can only wait once on a condition - additional calls are valid but
		558	will return immediately.
177		559
178	You can also listen for the status of a child process specified by the	560	If an error condition has been set by calling C<< ->croak >>, then this
179	C<pid> argument. The watcher will only trigger once. This works by	561	function will call C<croak>.
180	installing a signal handler for C<SIGCHLD>.
181		562
182	Example: wait for pid 1333	563	In list context, all parameters passed to C<send> will be returned,
		564	in scalar context only the first one will be returned.
183		565
184	my $w = AnyEvent->child (pid => 1333, cb => sub { warn "exit status $?" });	566	Not all event models support a blocking wait - some die in that case
		567	(programs might want to do that to stay interactive), so I<if you are
		568	using this from a module, never require a blocking wait>, but let the
		569	caller decide whether the call will block or not (for example, by coupling
		570	condition variables with some kind of request results and supporting
		571	callbacks so the caller knows that getting the result will not block,
		572	while still supporting blocking waits if the caller so desires).
185		573
186	=head1 GLOBALS	574	Another reason I<never> to C<< ->recv >> in a module is that you cannot
		575	sensibly have two C<< ->recv >>'s in parallel, as that would require
		576	multiple interpreters or coroutines/threads, none of which C<AnyEvent>
		577	can supply.
		578
		579	The L<Coro> module, however, I<can> and I<does> supply coroutines and, in
		580	fact, L<Coro::AnyEvent> replaces AnyEvent's condvars by coroutine-safe
		581	versions and also integrates coroutines into AnyEvent, making blocking
		582	C<< ->recv >> calls perfectly safe as long as they are done from another
		583	coroutine (one that doesn't run the event loop).
		584
		585	You can ensure that C<< -recv >> never blocks by setting a callback and
		586	only calling C<< ->recv >> from within that callback (or at a later
		587	time). This will work even when the event loop does not support blocking
		588	waits otherwise.
		589
		590	=item $bool = $cv->ready
		591
		592	Returns true when the condition is "true", i.e. whether C<send> or
		593	C<croak> have been called.
		594
		595	=item $cb = $cv->cb ([new callback])
		596
		597	This is a mutator function that returns the callback set and optionally
		598	replaces it before doing so.
		599
		600	The callback will be called when the condition becomes "true", i.e. when
		601	C<send> or C<croak> are called, with the only argument being the condition
		602	variable itself. Calling C<recv> inside the callback or at any later time
		603	is guaranteed not to block.
		604
		605	=back
		606
		607	=head1 GLOBAL VARIABLES AND FUNCTIONS
187		608
188	=over 4	609	=over 4
189		610
190	=item $AnyEvent::MODEL	611	=item $AnyEvent::MODEL
191		612
…		…
195	C<AnyEvent::Impl:xxx> modules, but can be any other class in the case	616	C<AnyEvent::Impl:xxx> modules, but can be any other class in the case
196	AnyEvent has been extended at runtime (e.g. in I<rxvt-unicode>).	617	AnyEvent has been extended at runtime (e.g. in I<rxvt-unicode>).
197		618
198	The known classes so far are:	619	The known classes so far are:
199		620
200	AnyEvent::Impl::Coro based on Coro::Event, best choice.
201	EV::AnyEvent based on EV (an interface to libevent)	621	AnyEvent::Impl::EV based on EV (an interface to libev, best choice).
202	AnyEvent::Impl::Event based on Event, also best choice :)	622	AnyEvent::Impl::Event based on Event, second best choice.
		623	AnyEvent::Impl::Perl pure-perl implementation, fast and portable.
203	AnyEvent::Impl::Glib based on Glib, second-best choice.	624	AnyEvent::Impl::Glib based on Glib, third-best choice.
204	AnyEvent::Impl::Tk based on Tk, very bad choice.	625	AnyEvent::Impl::Tk based on Tk, very bad choice.
205	AnyEvent::Impl::Perl pure-perl implementation, inefficient.	626	AnyEvent::Impl::Qt based on Qt, cannot be autoprobed (see its docs).
		627	AnyEvent::Impl::EventLib based on Event::Lib, leaks memory and worse.
		628	AnyEvent::Impl::POE based on POE, not generic enough for full support.
		629
		630	There is no support for WxWidgets, as WxWidgets has no support for
		631	watching file handles. However, you can use WxWidgets through the
		632	POE Adaptor, as POE has a Wx backend that simply polls 20 times per
		633	second, which was considered to be too horrible to even consider for
		634	AnyEvent. Likewise, other POE backends can be used by AnyEvent by using
		635	it's adaptor.
		636
		637	AnyEvent knows about L<Prima> and L<Wx> and will try to use L<POE> when
		638	autodetecting them.
206		639
207	=item AnyEvent::detect	640	=item AnyEvent::detect
208		641
209	Returns C<$AnyEvent::MODEL>, forcing autodetection of the event model if	642	Returns C<$AnyEvent::MODEL>, forcing autodetection of the event model
210	necessary. You should only call this function right before you would have	643	if necessary. You should only call this function right before you would
211	created an AnyEvent watcher anyway, that is, very late at runtime.	644	have created an AnyEvent watcher anyway, that is, as late as possible at
		645	runtime.
		646
		647	=item $guard = AnyEvent::post_detect { BLOCK }
		648
		649	Arranges for the code block to be executed as soon as the event model is
		650	autodetected (or immediately if this has already happened).
		651
		652	If called in scalar or list context, then it creates and returns an object
		653	that automatically removes the callback again when it is destroyed. See
		654	L<Coro::BDB> for a case where this is useful.
		655
		656	=item @AnyEvent::post_detect
		657
		658	If there are any code references in this array (you can C<push> to it
		659	before or after loading AnyEvent), then they will called directly after
		660	the event loop has been chosen.
		661
		662	You should check C<$AnyEvent::MODEL> before adding to this array, though:
		663	if it contains a true value then the event loop has already been detected,
		664	and the array will be ignored.
		665
		666	Best use C<AnyEvent::post_detect { BLOCK }> instead.
212		667
213	=back	668	=back
214		669
215	=head1 WHAT TO DO IN A MODULE	670	=head1 WHAT TO DO IN A MODULE
216		671
217	As a module author, you should "use AnyEvent" and call AnyEvent methods	672	As a module author, you should C<use AnyEvent> and call AnyEvent methods
218	freely, but you should not load a specific event module or rely on it.	673	freely, but you should not load a specific event module or rely on it.
219		674
220	Be careful when you create watchers in the module body - Anyevent will	675	Be careful when you create watchers in the module body - AnyEvent will
221	decide which event module to use as soon as the first method is called, so	676	decide which event module to use as soon as the first method is called, so
222	by calling AnyEvent in your module body you force the user of your module	677	by calling AnyEvent in your module body you force the user of your module
223	to load the event module first.	678	to load the event module first.
224		679
		680	Never call C<< ->recv >> on a condition variable unless you I<know> that
		681	the C<< ->send >> method has been called on it already. This is
		682	because it will stall the whole program, and the whole point of using
		683	events is to stay interactive.
		684
		685	It is fine, however, to call C<< ->recv >> when the user of your module
		686	requests it (i.e. if you create a http request object ad have a method
		687	called C<results> that returns the results, it should call C<< ->recv >>
		688	freely, as the user of your module knows what she is doing. always).
		689
225	=head1 WHAT TO DO IN THE MAIN PROGRAM	690	=head1 WHAT TO DO IN THE MAIN PROGRAM
226		691
227	There will always be a single main program - the only place that should	692	There will always be a single main program - the only place that should
228	dictate which event model to use.	693	dictate which event model to use.
229		694
230	If it doesn't care, it can just "use AnyEvent" and use it itself, or not	695	If it doesn't care, it can just "use AnyEvent" and use it itself, or not
231	do anything special and let AnyEvent decide which implementation to chose.	696	do anything special (it does not need to be event-based) and let AnyEvent
		697	decide which implementation to chose if some module relies on it.
232		698
233	If the main program relies on a specific event model (for example, in Gtk2	699	If the main program relies on a specific event model - for example, in
234	programs you have to rely on either Glib or Glib::Event), you should load	700	Gtk2 programs you have to rely on the Glib module - you should load the
235	it before loading AnyEvent or any module that uses it, generally, as early	701	event module before loading AnyEvent or any module that uses it: generally
236	as possible. The reason is that modules might create watchers when they	702	speaking, you should load it as early as possible. The reason is that
237	are loaded, and AnyEvent will decide on the event model to use as soon as	703	modules might create watchers when they are loaded, and AnyEvent will
238	it creates watchers, and it might chose the wrong one unless you load the	704	decide on the event model to use as soon as it creates watchers, and it
239	correct one yourself.	705	might chose the wrong one unless you load the correct one yourself.
240		706
241	You can chose to use a rather inefficient pure-perl implementation by	707	You can chose to use a pure-perl implementation by loading the
242	loading the C<AnyEvent::Impl::Perl> module, but letting AnyEvent chose is	708	C<AnyEvent::Impl::Perl> module, which gives you similar behaviour
243	generally better.	709	everywhere, but letting AnyEvent chose the model is generally better.
		710
		711	=head2 MAINLOOP EMULATION
		712
		713	Sometimes (often for short test scripts, or even standalone programs who
		714	only want to use AnyEvent), you do not want to run a specific event loop.
		715
		716	In that case, you can use a condition variable like this:
		717
		718	AnyEvent->condvar->recv;
		719
		720	This has the effect of entering the event loop and looping forever.
		721
		722	Note that usually your program has some exit condition, in which case
		723	it is better to use the "traditional" approach of storing a condition
		724	variable somewhere, waiting for it, and sending it when the program should
		725	exit cleanly.
		726
		727
		728	=head1 OTHER MODULES
		729
		730	The following is a non-exhaustive list of additional modules that use
		731	AnyEvent and can therefore be mixed easily with other AnyEvent modules
		732	in the same program. Some of the modules come with AnyEvent, some are
		733	available via CPAN.
		734
		735	=over 4
		736
		737	=item L<AnyEvent::Util>
		738
		739	Contains various utility functions that replace often-used but blocking
		740	functions such as C<inet_aton> by event-/callback-based versions.
		741
		742	=item L<AnyEvent::Socket>
		743
		744	Provides various utility functions for (internet protocol) sockets,
		745	addresses and name resolution. Also functions to create non-blocking tcp
		746	connections or tcp servers, with IPv6 and SRV record support and more.
		747
		748	=item L<AnyEvent::Handle>
		749
		750	Provide read and write buffers, manages watchers for reads and writes,
		751	supports raw and formatted I/O, I/O queued and fully transparent and
		752	non-blocking SSL/TLS.
		753
		754	=item L<AnyEvent::DNS>
		755
		756	Provides rich asynchronous DNS resolver capabilities.
		757
		758	=item L<AnyEvent::HTTP>
		759
		760	A simple-to-use HTTP library that is capable of making a lot of concurrent
		761	HTTP requests.
		762
		763	=item L<AnyEvent::HTTPD>
		764
		765	Provides a simple web application server framework.
		766
		767	=item L<AnyEvent::FastPing>
		768
		769	The fastest ping in the west.
		770
		771	=item L<AnyEvent::DBI>
		772
		773	Executes L<DBI> requests asynchronously in a proxy process.
		774
		775	=item L<AnyEvent::AIO>
		776
		777	Truly asynchronous I/O, should be in the toolbox of every event
		778	programmer. AnyEvent::AIO transparently fuses L<IO::AIO> and AnyEvent
		779	together.
		780
		781	=item L<AnyEvent::BDB>
		782
		783	Truly asynchronous Berkeley DB access. AnyEvent::BDB transparently fuses
		784	L<BDB> and AnyEvent together.
		785
		786	=item L<AnyEvent::GPSD>
		787
		788	A non-blocking interface to gpsd, a daemon delivering GPS information.
		789
		790	=item L<AnyEvent::IGS>
		791
		792	A non-blocking interface to the Internet Go Server protocol (used by
		793	L<App::IGS>).
		794
		795	=item L<Net::IRC3>
		796
		797	AnyEvent based IRC client module family.
		798
		799	=item L<Net::XMPP2>
		800
		801	AnyEvent based XMPP (Jabber protocol) module family.
		802
		803	=item L<Net::FCP>
		804
		805	AnyEvent-based implementation of the Freenet Client Protocol, birthplace
		806	of AnyEvent.
		807
		808	=item L<Event::ExecFlow>
		809
		810	High level API for event-based execution flow control.
		811
		812	=item L<Coro>
		813
		814	Has special support for AnyEvent via L<Coro::AnyEvent>.
		815
		816	=item L<IO::Lambda>
		817
		818	The lambda approach to I/O - don't ask, look there. Can use AnyEvent.
		819
		820	=back
244		821
245	=cut	822	=cut
246		823
247	package AnyEvent;	824	package AnyEvent;
248		825
249	no warnings;	826	no warnings;
250	use strict;	827	use strict;
251		828
252	use Carp;	829	use Carp;
253		830
254	our $VERSION = '2.55';	831	our $VERSION = 4.2;
255	our $MODEL;	832	our $MODEL;
256		833
257	our $AUTOLOAD;	834	our $AUTOLOAD;
258	our @ISA;	835	our @ISA;
259		836
		837	our @REGISTRY;
		838
		839	our $WIN32;
		840
		841	BEGIN {
		842	my $win32 = ! ! ($^O =~ /mswin32/i);
		843	eval "sub WIN32(){ $win32 }";
		844	}
		845
260	our $verbose = $ENV{PERL_ANYEVENT_VERBOSE}*1;	846	our $verbose = $ENV{PERL_ANYEVENT_VERBOSE}*1;
261		847
262	our @REGISTRY;	848	our %PROTOCOL; # (ipv4|ipv6) => (1|2), higher numbers are preferred
		849
		850	{
		851	my $idx;
		852	$PROTOCOL{$_} = ++$idx
		853	for reverse split /\s*,\s*/,
		854	$ENV{PERL_ANYEVENT_PROTOCOLS} || "ipv4,ipv6";
		855	}
263		856
264	my @models = (	857	my @models = (
265	[Coro::Event:: => AnyEvent::Impl::Coro::],
266	[EV:: => EV::AnyEvent::],	858	[EV:: => AnyEvent::Impl::EV::],
267	[Event:: => AnyEvent::Impl::Event::],	859	[Event:: => AnyEvent::Impl::Event::],
268	[Glib:: => AnyEvent::Impl::Glib::],
269	[Tk:: => AnyEvent::Impl::Tk::],
270	[AnyEvent::Impl::Perl:: => AnyEvent::Impl::Perl::],	860	[AnyEvent::Impl::Perl:: => AnyEvent::Impl::Perl::],
		861	# everything below here will not be autoprobed
		862	# as the pureperl backend should work everywhere
		863	# and is usually faster
		864	[Tk:: => AnyEvent::Impl::Tk::], # crashes with many handles
		865	[Glib:: => AnyEvent::Impl::Glib::], # becomes extremely slow with many watchers
		866	[Event::Lib:: => AnyEvent::Impl::EventLib::], # too buggy
		867	[Qt:: => AnyEvent::Impl::Qt::], # requires special main program
		868	[POE::Kernel:: => AnyEvent::Impl::POE::], # lasciate ogni speranza
		869	[Wx:: => AnyEvent::Impl::POE::],
		870	[Prima:: => AnyEvent::Impl::POE::],
271	);	871	);
272		872
273	our %method = map +($_ => 1), qw(io timer condvar broadcast wait signal one_event DESTROY);	873	our %method = map +($_ => 1), qw(io timer time now signal child condvar one_event DESTROY);
		874
		875	our @post_detect;
		876
		877	sub post_detect(&) {
		878	my ($cb) = @_;
		879
		880	if ($MODEL) {
		881	$cb->();
		882
		883	1
		884	} else {
		885	push @post_detect, $cb;
		886
		887	defined wantarray
		888	? bless \$cb, "AnyEvent::Util::PostDetect"
		889	: ()
		890	}
		891	}
		892
		893	sub AnyEvent::Util::PostDetect::DESTROY {
		894	@post_detect = grep $_ != ${$_[0]}, @post_detect;
		895	}
274		896
275	sub detect() {	897	sub detect() {
276	unless ($MODEL) {	898	unless ($MODEL) {
277	no strict 'refs';	899	no strict 'refs';
		900	local $SIG{__DIE__};
		901
		902	if ($ENV{PERL_ANYEVENT_MODEL} =~ /^([a-zA-Z]+)$/) {
		903	my $model = "AnyEvent::Impl::$1";
		904	if (eval "require $model") {
		905	$MODEL = $model;
		906	warn "AnyEvent: loaded model '$model' (forced by \$PERL_ANYEVENT_MODEL), using it.\n" if $verbose > 1;
		907	} else {
		908	warn "AnyEvent: unable to load model '$model' (from \$PERL_ANYEVENT_MODEL):\n$@" if $verbose;
		909	}
		910	}
278		911
279	# check for already loaded models	912	# check for already loaded models
		913	unless ($MODEL) {
280	for (@REGISTRY, @models) {	914	for (@REGISTRY, @models) {
281	my ($package, $model) = @$_;	915	my ($package, $model) = @$_;
282	if (${"$package\::VERSION"} > 0) {	916	if (${"$package\::VERSION"} > 0) {
283	if (eval "require $model") {	917	if (eval "require $model") {
284	$MODEL = $model;	918	$MODEL = $model;
285	warn "AnyEvent: found model '$model', using it.\n" if $verbose > 1;	919	warn "AnyEvent: autodetected model '$model', using it.\n" if $verbose > 1;
286	last;	920	last;
		921	}
287	}	922	}
288	}	923	}
		924
		925	unless ($MODEL) {
		926	# try to load a model
		927
		928	for (@REGISTRY, @models) {
		929	my ($package, $model) = @$_;
		930	if (eval "require $package"
		931	and ${"$package\::VERSION"} > 0
		932	and eval "require $model") {
		933	$MODEL = $model;
		934	warn "AnyEvent: autoprobed model '$model', using it.\n" if $verbose > 1;
		935	last;
		936	}
		937	}
		938
		939	$MODEL
		940	or die "No event module selected for AnyEvent and autodetect failed. Install any one of these modules: EV, Event or Glib.";
		941	}
289	}	942	}
290		943
291	unless ($MODEL) {	944	push @{"$MODEL\::ISA"}, "AnyEvent::Base";
292	# try to load a model
293
294	for (@REGISTRY, @models) {
295	my ($package, $model) = @$_;
296	if (eval "require $package"
297	and ${"$package\::VERSION"} > 0
298	and eval "require $model") {
299	$MODEL = $model;
300	warn "AnyEvent: autoprobed and loaded model '$model', using it.\n" if $verbose > 1;
301	last;
302	}
303	}
304
305	$MODEL
306	or die "No event module selected for AnyEvent and autodetect failed. Install any one of these modules: Event (or Coro+Event), Glib or Tk.";
307	}
308		945
309	unshift @ISA, $MODEL;	946	unshift @ISA, $MODEL;
310	push @{"$MODEL\::ISA"}, "AnyEvent::Base";	947
		948	require AnyEvent::Strict if $ENV{PERL_ANYEVENT_STRICT};
		949
		950	(shift @post_detect)->() while @post_detect;
311	}	951	}
312		952
313	$MODEL	953	$MODEL
314	}	954	}
315		955
…		…
325	$class->$func (@_);	965	$class->$func (@_);
326	}	966	}
327		967
328	package AnyEvent::Base;	968	package AnyEvent::Base;
329		969
		970	# default implementation for now and time
		971
		972	use Time::HiRes ();
		973
		974	sub time { Time::HiRes::time }
		975	sub now { Time::HiRes::time }
		976
330	# default implementation for ->condvar, ->wait, ->broadcast	977	# default implementation for ->condvar
331		978
332	sub condvar {	979	sub condvar {
333	bless \my $flag, "AnyEvent::Base::CondVar"	980	bless { @_ == 3 ? (_ae_cb => $_[2]) : () }, AnyEvent::CondVar::
334	}
335
336	sub AnyEvent::Base::CondVar::broadcast {
337	${$_[0]}++;
338	}
339
340	sub AnyEvent::Base::CondVar::wait {
341	AnyEvent->one_event while !${$_[0]};
342	}	981	}
343		982
344	# default implementation for ->signal	983	# default implementation for ->signal
345		984
346	our %SIG_CB;	985	our %SIG_CB;
…		…
362	sub AnyEvent::Base::Signal::DESTROY {	1001	sub AnyEvent::Base::Signal::DESTROY {
363	my ($signal, $cb) = @{$_[0]};	1002	my ($signal, $cb) = @{$_[0]};
364		1003
365	delete $SIG_CB{$signal}{$cb};	1004	delete $SIG_CB{$signal}{$cb};
366		1005
367	$SIG{$signal} = 'DEFAULT' unless keys %{ $SIG_CB{$signal} };	1006	delete $SIG{$signal} unless keys %{ $SIG_CB{$signal} };
368	}	1007	}
369		1008
370	# default implementation for ->child	1009	# default implementation for ->child
371		1010
372	our %PID_CB;	1011	our %PID_CB;
373	our $CHLD_W;	1012	our $CHLD_W;
		1013	our $CHLD_DELAY_W;
374	our $PID_IDLE;	1014	our $PID_IDLE;
375	our $WNOHANG;	1015	our $WNOHANG;
376		1016
377	sub _child_wait {	1017	sub _child_wait {
378	while (0 < (my $pid = waitpid -1, $WNOHANG)) {	1018	while (0 < (my $pid = waitpid -1, $WNOHANG)) {
379	$_->() for values %{ (delete $PID_CB{$pid}) || {} };	1019	$_->($pid, $?) for (values %{ $PID_CB{$pid} || {} }),
		1020	(values %{ $PID_CB{0} || {} });
380	}	1021	}
381		1022
382	undef $PID_IDLE;	1023	undef $PID_IDLE;
		1024	}
		1025
		1026	sub _sigchld {
		1027	# make sure we deliver these changes "synchronous" with the event loop.
		1028	$CHLD_DELAY_W ||= AnyEvent->timer (after => 0, cb => sub {
		1029	undef $CHLD_DELAY_W;
		1030	&_child_wait;
		1031	});
383	}	1032	}
384		1033
385	sub child {	1034	sub child {
386	my (undef, %arg) = @_;	1035	my (undef, %arg) = @_;
387		1036
388	my $pid = uc $arg{pid}	1037	defined (my $pid = $arg{pid} + 0)
389	or Carp::croak "required option 'pid' is missing";	1038	or Carp::croak "required option 'pid' is missing";
390		1039
391	$PID_CB{$pid}{$arg{cb}} = $arg{cb};	1040	$PID_CB{$pid}{$arg{cb}} = $arg{cb};
392		1041
393	unless ($WNOHANG) {	1042	unless ($WNOHANG) {
394	$WNOHANG = eval { require POSIX; &POSIX::WNOHANG } || 1;	1043	$WNOHANG = eval { local $SIG{__DIE__}; require POSIX; &POSIX::WNOHANG } || 1;
395	}	1044	}
396		1045
397	unless ($CHLD_W) {	1046	unless ($CHLD_W) {
398	$CHLD_W = AnyEvent->signal (signal => 'CHLD', cb => \&_child_wait);	1047	$CHLD_W = AnyEvent->signal (signal => 'CHLD', cb => \&_sigchld);
399	# child could be a zombie already	1048	# child could be a zombie already, so make at least one round
400	$PID_IDLE ||= AnyEvent->timer (after => 0, cb => \&_child_wait);	1049	&_sigchld;
401	}	1050	}
402		1051
403	bless [$pid, $arg{cb}], "AnyEvent::Base::Child"	1052	bless [$pid, $arg{cb}], "AnyEvent::Base::Child"
404	}	1053	}
405		1054
…		…
410	delete $PID_CB{$pid} unless keys %{ $PID_CB{$pid} };	1059	delete $PID_CB{$pid} unless keys %{ $PID_CB{$pid} };
411		1060
412	undef $CHLD_W unless keys %PID_CB;	1061	undef $CHLD_W unless keys %PID_CB;
413	}	1062	}
414		1063
		1064	package AnyEvent::CondVar;
		1065
		1066	our @ISA = AnyEvent::CondVar::Base::;
		1067
		1068	package AnyEvent::CondVar::Base;
		1069
		1070	use overload
		1071	'&{}' => sub { my $self = shift; sub { $self->send (@_) } },
		1072	fallback => 1;
		1073
		1074	sub _send {
		1075	# nop
		1076	}
		1077
		1078	sub send {
		1079	my $cv = shift;
		1080	$cv->{_ae_sent} = [@_];
		1081	(delete $cv->{_ae_cb})->($cv) if $cv->{_ae_cb};
		1082	$cv->_send;
		1083	}
		1084
		1085	sub croak {
		1086	$_[0]{_ae_croak} = $_[1];
		1087	$_[0]->send;
		1088	}
		1089
		1090	sub ready {
		1091	$_[0]{_ae_sent}
		1092	}
		1093
		1094	sub _wait {
		1095	AnyEvent->one_event while !$_[0]{_ae_sent};
		1096	}
		1097
		1098	sub recv {
		1099	$_[0]->_wait;
		1100
		1101	Carp::croak $_[0]{_ae_croak} if $_[0]{_ae_croak};
		1102	wantarray ? @{ $_[0]{_ae_sent} } : $_[0]{_ae_sent}[0]
		1103	}
		1104
		1105	sub cb {
		1106	$_[0]{_ae_cb} = $_[1] if @_ > 1;
		1107	$_[0]{_ae_cb}
		1108	}
		1109
		1110	sub begin {
		1111	++$_[0]{_ae_counter};
		1112	$_[0]{_ae_end_cb} = $_[1] if @_ > 1;
		1113	}
		1114
		1115	sub end {
		1116	return if --$_[0]{_ae_counter};
		1117	&{ $_[0]{_ae_end_cb} || sub { $_[0]->send } };
		1118	}
		1119
		1120	# undocumented/compatibility with pre-3.4
		1121	*broadcast = \&send;
		1122	*wait = \&_wait;
		1123
415	=head1 SUPPLYING YOUR OWN EVENT MODEL INTERFACE	1124	=head1 SUPPLYING YOUR OWN EVENT MODEL INTERFACE
		1125
		1126	This is an advanced topic that you do not normally need to use AnyEvent in
		1127	a module. This section is only of use to event loop authors who want to
		1128	provide AnyEvent compatibility.
416		1129
417	If you need to support another event library which isn't directly	1130	If you need to support another event library which isn't directly
418	supported by AnyEvent, you can supply your own interface to it by	1131	supported by AnyEvent, you can supply your own interface to it by
419	pushing, before the first watcher gets created, the package name of	1132	pushing, before the first watcher gets created, the package name of
420	the event module and the package name of the interface to use onto	1133	the event module and the package name of the interface to use onto
421	C<@AnyEvent::REGISTRY>. You can do that before and even without loading	1134	C<@AnyEvent::REGISTRY>. You can do that before and even without loading
422	AnyEvent.	1135	AnyEvent, so it is reasonably cheap.
423		1136
424	Example:	1137	Example:
425		1138
426	push @AnyEvent::REGISTRY, [urxvt => urxvt::anyevent::];	1139	push @AnyEvent::REGISTRY, [urxvt => urxvt::anyevent::];
427		1140
428	This tells AnyEvent to (literally) use the C<urxvt::anyevent::>	1141	This tells AnyEvent to (literally) use the C<urxvt::anyevent::>
429	package/class when it finds the C<urxvt> package/module is loaded. When	1142	package/class when it finds the C<urxvt> package/module is already loaded.
		1143
430	AnyEvent is loaded and asked to find a suitable event model, it will	1144	When AnyEvent is loaded and asked to find a suitable event model, it
431	first check for the presence of urxvt.	1145	will first check for the presence of urxvt by trying to C<use> the
		1146	C<urxvt::anyevent> module.
432		1147
433	The class should provide implementations for all watcher types (see	1148	The class should provide implementations for all watcher types. See
434	L<AnyEvent::Impl::Event> (source code), L<AnyEvent::Impl::Glib>	1149	L<AnyEvent::Impl::EV> (source code), L<AnyEvent::Impl::Glib> (Source code)
435	(Source code) and so on for actual examples, use C<perldoc -m	1150	and so on for actual examples. Use C<perldoc -m AnyEvent::Impl::Glib> to
436	AnyEvent::Impl::Glib> to see the sources).	1151	see the sources.
437		1152
		1153	If you don't provide C<signal> and C<child> watchers than AnyEvent will
		1154	provide suitable (hopefully) replacements.
		1155
438	The above isn't fictitious, the I<rxvt-unicode> (a.k.a. urxvt)	1156	The above example isn't fictitious, the I<rxvt-unicode> (a.k.a. urxvt)
439	uses the above line as-is. An interface isn't included in AnyEvent	1157	terminal emulator uses the above line as-is. An interface isn't included
440	because it doesn't make sense outside the embedded interpreter inside	1158	in AnyEvent because it doesn't make sense outside the embedded interpreter
441	I<rxvt-unicode>, and it is updated and maintained as part of the	1159	inside I<rxvt-unicode>, and it is updated and maintained as part of the
442	I<rxvt-unicode> distribution.	1160	I<rxvt-unicode> distribution.
443		1161
444	I<rxvt-unicode> also cheats a bit by not providing blocking access to	1162	I<rxvt-unicode> also cheats a bit by not providing blocking access to
445	condition variables: code blocking while waiting for a condition will	1163	condition variables: code blocking while waiting for a condition will
446	C<die>. This still works with most modules/usages, and blocking calls must	1164	C<die>. This still works with most modules/usages, and blocking calls must
447	not be in an interactive application, so it makes sense.	1165	not be done in an interactive application, so it makes sense.
448		1166
449	=head1 ENVIRONMENT VARIABLES	1167	=head1 ENVIRONMENT VARIABLES
450		1168
451	The following environment variables are used by this module:	1169	The following environment variables are used by this module:
452		1170
453	C<PERL_ANYEVENT_VERBOSE> when set to C<2> or higher, reports which event	1171	=over 4
454	model gets used.
455		1172
		1173	=item C<PERL_ANYEVENT_VERBOSE>
		1174
		1175	By default, AnyEvent will be completely silent except in fatal
		1176	conditions. You can set this environment variable to make AnyEvent more
		1177	talkative.
		1178
		1179	When set to C<1> or higher, causes AnyEvent to warn about unexpected
		1180	conditions, such as not being able to load the event model specified by
		1181	C<PERL_ANYEVENT_MODEL>.
		1182
		1183	When set to C<2> or higher, cause AnyEvent to report to STDERR which event
		1184	model it chooses.
		1185
		1186	=item C<PERL_ANYEVENT_STRICT>
		1187
		1188	AnyEvent does not do much argument checking by default, as thorough
		1189	argument checking is very costly. Setting this variable to a true value
		1190	will cause AnyEvent to thoroughly check the arguments passed to most
		1191	method calls and croaks if it finds any problems. In other words, enables
		1192	"strict" mode. Unlike C<use strict> it is definitely recommended ot keep
		1193	it off in production.
		1194
		1195	=item C<PERL_ANYEVENT_MODEL>
		1196
		1197	This can be used to specify the event model to be used by AnyEvent, before
		1198	auto detection and -probing kicks in. It must be a string consisting
		1199	entirely of ASCII letters. The string C<AnyEvent::Impl::> gets prepended
		1200	and the resulting module name is loaded and if the load was successful,
		1201	used as event model. If it fails to load AnyEvent will proceed with
		1202	auto detection and -probing.
		1203
		1204	This functionality might change in future versions.
		1205
		1206	For example, to force the pure perl model (L<AnyEvent::Impl::Perl>) you
		1207	could start your program like this:
		1208
		1209	PERL_ANYEVENT_MODEL=Perl perl ...
		1210
		1211	=item C<PERL_ANYEVENT_PROTOCOLS>
		1212
		1213	Used by both L<AnyEvent::DNS> and L<AnyEvent::Socket> to determine preferences
		1214	for IPv4 or IPv6. The default is unspecified (and might change, or be the result
		1215	of auto probing).
		1216
		1217	Must be set to a comma-separated list of protocols or address families,
		1218	current supported: C<ipv4> and C<ipv6>. Only protocols mentioned will be
		1219	used, and preference will be given to protocols mentioned earlier in the
		1220	list.
		1221
		1222	This variable can effectively be used for denial-of-service attacks
		1223	against local programs (e.g. when setuid), although the impact is likely
		1224	small, as the program has to handle connection errors already-
		1225
		1226	Examples: C<PERL_ANYEVENT_PROTOCOLS=ipv4,ipv6> - prefer IPv4 over IPv6,
		1227	but support both and try to use both. C<PERL_ANYEVENT_PROTOCOLS=ipv4>
		1228	- only support IPv4, never try to resolve or contact IPv6
		1229	addresses. C<PERL_ANYEVENT_PROTOCOLS=ipv6,ipv4> support either IPv4 or
		1230	IPv6, but prefer IPv6 over IPv4.
		1231
		1232	=item C<PERL_ANYEVENT_EDNS0>
		1233
		1234	Used by L<AnyEvent::DNS> to decide whether to use the EDNS0 extension
		1235	for DNS. This extension is generally useful to reduce DNS traffic, but
		1236	some (broken) firewalls drop such DNS packets, which is why it is off by
		1237	default.
		1238
		1239	Setting this variable to C<1> will cause L<AnyEvent::DNS> to announce
		1240	EDNS0 in its DNS requests.
		1241
		1242	=item C<PERL_ANYEVENT_MAX_FORKS>
		1243
		1244	The maximum number of child processes that C<AnyEvent::Util::fork_call>
		1245	will create in parallel.
		1246
		1247	=back
		1248
456	=head1 EXAMPLE	1249	=head1 EXAMPLE PROGRAM
457		1250
458	The following program uses an io watcher to read data from stdin, a timer	1251	The following program uses an I/O watcher to read data from STDIN, a timer
459	to display a message once per second, and a condvar to exit the program	1252	to display a message once per second, and a condition variable to quit the
460	when the user enters quit:	1253	program when the user enters quit:
461		1254
462	use AnyEvent;	1255	use AnyEvent;
463		1256
464	my $cv = AnyEvent->condvar;	1257	my $cv = AnyEvent->condvar;
465		1258
466	my $io_watcher = AnyEvent->io (fh => \*STDIN, poll => 'r', cb => sub {	1259	my $io_watcher = AnyEvent->io (
		1260	fh => \*STDIN,
		1261	poll => 'r',
		1262	cb => sub {
467	warn "io event <$_[0]>\n"; # will always output <r>	1263	warn "io event <$_[0]>\n"; # will always output <r>
468	chomp (my $input = <STDIN>); # read a line	1264	chomp (my $input = <STDIN>); # read a line
469	warn "read: $input\n"; # output what has been read	1265	warn "read: $input\n"; # output what has been read
470	$cv->broadcast if $input =~ /^q/i; # quit program if /^q/i	1266	$cv->send if $input =~ /^q/i; # quit program if /^q/i
		1267	},
471	});	1268	);
472		1269
473	my $time_watcher; # can only be used once	1270	my $time_watcher; # can only be used once
474		1271
475	sub new_timer {	1272	sub new_timer {
476	$timer = AnyEvent->timer (after => 1, cb => sub {	1273	$timer = AnyEvent->timer (after => 1, cb => sub {
…		…
479	});	1276	});
480	}	1277	}
481		1278
482	new_timer; # create first timer	1279	new_timer; # create first timer
483		1280
484	$cv->wait; # wait until user enters /^q/i	1281	$cv->recv; # wait until user enters /^q/i
485		1282
486	=head1 REAL-WORLD EXAMPLE	1283	=head1 REAL-WORLD EXAMPLE
487		1284
488	Consider the L<Net::FCP> module. It features (among others) the following	1285	Consider the L<Net::FCP> module. It features (among others) the following
489	API calls, which are to freenet what HTTP GET requests are to http:	1286	API calls, which are to freenet what HTTP GET requests are to http:
…		…
539	syswrite $txn->{fh}, $txn->{request}	1336	syswrite $txn->{fh}, $txn->{request}
540	or die "connection or write error";	1337	or die "connection or write error";
541	$txn->{w} = AnyEvent->io (fh => $txn->{fh}, poll => 'r', cb => sub { $txn->fh_ready_r });	1338	$txn->{w} = AnyEvent->io (fh => $txn->{fh}, poll => 'r', cb => sub { $txn->fh_ready_r });
542		1339
543	Again, C<fh_ready_r> waits till all data has arrived, and then stores the	1340	Again, C<fh_ready_r> waits till all data has arrived, and then stores the
544	result and signals any possible waiters that the request ahs finished:	1341	result and signals any possible waiters that the request has finished:
545		1342
546	sysread $txn->{fh}, $txn->{buf}, length $txn->{$buf};	1343	sysread $txn->{fh}, $txn->{buf}, length $txn->{$buf};
547		1344
548	if (end-of-file or data complete) {	1345	if (end-of-file or data complete) {
549	$txn->{result} = $txn->{buf};	1346	$txn->{result} = $txn->{buf};
550	$txn->{finished}->broadcast;	1347	$txn->{finished}->send;
551	$txb->{cb}->($txn) of $txn->{cb}; # also call callback	1348	$txb->{cb}->($txn) of $txn->{cb}; # also call callback
552	}	1349	}
553		1350
554	The C<result> method, finally, just waits for the finished signal (if the	1351	The C<result> method, finally, just waits for the finished signal (if the
555	request was already finished, it doesn't wait, of course, and returns the	1352	request was already finished, it doesn't wait, of course, and returns the
556	data:	1353	data:
557		1354
558	$txn->{finished}->wait;	1355	$txn->{finished}->recv;
559	return $txn->{result};	1356	return $txn->{result};
560		1357
561	The actual code goes further and collects all errors (C<die>s, exceptions)	1358	The actual code goes further and collects all errors (C<die>s, exceptions)
562	that occured during request processing. The C<result> method detects	1359	that occurred during request processing. The C<result> method detects
563	wether an exception as thrown (it is stored inside the $txn object)	1360	whether an exception as thrown (it is stored inside the $txn object)
564	and just throws the exception, which means connection errors and other	1361	and just throws the exception, which means connection errors and other
565	problems get reported tot he code that tries to use the result, not in a	1362	problems get reported tot he code that tries to use the result, not in a
566	random callback.	1363	random callback.
567		1364
568	All of this enables the following usage styles:	1365	All of this enables the following usage styles:
569		1366
570	1. Blocking:	1367	1. Blocking:
571		1368
572	my $data = $fcp->client_get ($url);	1369	my $data = $fcp->client_get ($url);
573		1370
574	2. Blocking, but parallelizing:	1371	2. Blocking, but running in parallel:
575		1372
576	my @datas = map $_->result,	1373	my @datas = map $_->result,
577	map $fcp->txn_client_get ($_),	1374	map $fcp->txn_client_get ($_),
578	@urls;	1375	@urls;
579		1376
580	Both blocking examples work without the module user having to know	1377	Both blocking examples work without the module user having to know
581	anything about events.	1378	anything about events.
582		1379
583	3a. Event-based in a main program, using any support Event module:	1380	3a. Event-based in a main program, using any supported event module:
584		1381
585	use Event;	1382	use EV;
586		1383
587	$fcp->txn_client_get ($url)->cb (sub {	1384	$fcp->txn_client_get ($url)->cb (sub {
588	my $txn = shift;	1385	my $txn = shift;
589	my $data = $txn->result;	1386	my $data = $txn->result;
590	...	1387	...
591	});	1388	});
592		1389
593	Event::loop;	1390	EV::loop;
594		1391
595	3b. The module user could use AnyEvent, too:	1392	3b. The module user could use AnyEvent, too:
596		1393
597	use AnyEvent;	1394	use AnyEvent;
598		1395
599	my $quit = AnyEvent->condvar;	1396	my $quit = AnyEvent->condvar;
600		1397
601	$fcp->txn_client_get ($url)->cb (sub {	1398	$fcp->txn_client_get ($url)->cb (sub {
602	...	1399	...
603	$quit->broadcast;	1400	$quit->send;
604	});	1401	});
605		1402
606	$quit->wait;	1403	$quit->recv;
		1404
		1405
		1406	=head1 BENCHMARKS
		1407
		1408	To give you an idea of the performance and overheads that AnyEvent adds
		1409	over the event loops themselves and to give you an impression of the speed
		1410	of various event loops I prepared some benchmarks.
		1411
		1412	=head2 BENCHMARKING ANYEVENT OVERHEAD
		1413
		1414	Here is a benchmark of various supported event models used natively and
		1415	through AnyEvent. The benchmark creates a lot of timers (with a zero
		1416	timeout) and I/O watchers (watching STDOUT, a pty, to become writable,
		1417	which it is), lets them fire exactly once and destroys them again.
		1418
		1419	Source code for this benchmark is found as F<eg/bench> in the AnyEvent
		1420	distribution.
		1421
		1422	=head3 Explanation of the columns
		1423
		1424	I<watcher> is the number of event watchers created/destroyed. Since
		1425	different event models feature vastly different performances, each event
		1426	loop was given a number of watchers so that overall runtime is acceptable
		1427	and similar between tested event loop (and keep them from crashing): Glib
		1428	would probably take thousands of years if asked to process the same number
		1429	of watchers as EV in this benchmark.
		1430
		1431	I<bytes> is the number of bytes (as measured by the resident set size,
		1432	RSS) consumed by each watcher. This method of measuring captures both C
		1433	and Perl-based overheads.
		1434
		1435	I<create> is the time, in microseconds (millionths of seconds), that it
		1436	takes to create a single watcher. The callback is a closure shared between
		1437	all watchers, to avoid adding memory overhead. That means closure creation
		1438	and memory usage is not included in the figures.
		1439
		1440	I<invoke> is the time, in microseconds, used to invoke a simple
		1441	callback. The callback simply counts down a Perl variable and after it was
		1442	invoked "watcher" times, it would C<< ->send >> a condvar once to
		1443	signal the end of this phase.
		1444
		1445	I<destroy> is the time, in microseconds, that it takes to destroy a single
		1446	watcher.
		1447
		1448	=head3 Results
		1449
		1450	name watchers bytes create invoke destroy comment
		1451	EV/EV 400000 244 0.56 0.46 0.31 EV native interface
		1452	EV/Any 100000 244 2.50 0.46 0.29 EV + AnyEvent watchers
		1453	CoroEV/Any 100000 244 2.49 0.44 0.29 coroutines + Coro::Signal
		1454	Perl/Any 100000 513 4.92 0.87 1.12 pure perl implementation
		1455	Event/Event 16000 516 31.88 31.30 0.85 Event native interface
		1456	Event/Any 16000 590 35.75 31.42 1.08 Event + AnyEvent watchers
		1457	Glib/Any 16000 1357 98.22 12.41 54.00 quadratic behaviour
		1458	Tk/Any 2000 1860 26.97 67.98 14.00 SEGV with >> 2000 watchers
		1459	POE/Event 2000 6644 108.64 736.02 14.73 via POE::Loop::Event
		1460	POE/Select 2000 6343 94.13 809.12 565.96 via POE::Loop::Select
		1461
		1462	=head3 Discussion
		1463
		1464	The benchmark does I<not> measure scalability of the event loop very
		1465	well. For example, a select-based event loop (such as the pure perl one)
		1466	can never compete with an event loop that uses epoll when the number of
		1467	file descriptors grows high. In this benchmark, all events become ready at
		1468	the same time, so select/poll-based implementations get an unnatural speed
		1469	boost.
		1470
		1471	Also, note that the number of watchers usually has a nonlinear effect on
		1472	overall speed, that is, creating twice as many watchers doesn't take twice
		1473	the time - usually it takes longer. This puts event loops tested with a
		1474	higher number of watchers at a disadvantage.
		1475
		1476	To put the range of results into perspective, consider that on the
		1477	benchmark machine, handling an event takes roughly 1600 CPU cycles with
		1478	EV, 3100 CPU cycles with AnyEvent's pure perl loop and almost 3000000 CPU
		1479	cycles with POE.
		1480
		1481	C<EV> is the sole leader regarding speed and memory use, which are both
		1482	maximal/minimal, respectively. Even when going through AnyEvent, it uses
		1483	far less memory than any other event loop and is still faster than Event
		1484	natively.
		1485
		1486	The pure perl implementation is hit in a few sweet spots (both the
		1487	constant timeout and the use of a single fd hit optimisations in the perl
		1488	interpreter and the backend itself). Nevertheless this shows that it
		1489	adds very little overhead in itself. Like any select-based backend its
		1490	performance becomes really bad with lots of file descriptors (and few of
		1491	them active), of course, but this was not subject of this benchmark.
		1492
		1493	The C<Event> module has a relatively high setup and callback invocation
		1494	cost, but overall scores in on the third place.
		1495
		1496	C<Glib>'s memory usage is quite a bit higher, but it features a
		1497	faster callback invocation and overall ends up in the same class as
		1498	C<Event>. However, Glib scales extremely badly, doubling the number of
		1499	watchers increases the processing time by more than a factor of four,
		1500	making it completely unusable when using larger numbers of watchers
		1501	(note that only a single file descriptor was used in the benchmark, so
		1502	inefficiencies of C<poll> do not account for this).
		1503
		1504	The C<Tk> adaptor works relatively well. The fact that it crashes with
		1505	more than 2000 watchers is a big setback, however, as correctness takes
		1506	precedence over speed. Nevertheless, its performance is surprising, as the
		1507	file descriptor is dup()ed for each watcher. This shows that the dup()
		1508	employed by some adaptors is not a big performance issue (it does incur a
		1509	hidden memory cost inside the kernel which is not reflected in the figures
		1510	above).
		1511
		1512	C<POE>, regardless of underlying event loop (whether using its pure perl
		1513	select-based backend or the Event module, the POE-EV backend couldn't
		1514	be tested because it wasn't working) shows abysmal performance and
		1515	memory usage with AnyEvent: Watchers use almost 30 times as much memory
		1516	as EV watchers, and 10 times as much memory as Event (the high memory
		1517	requirements are caused by requiring a session for each watcher). Watcher
		1518	invocation speed is almost 900 times slower than with AnyEvent's pure perl
		1519	implementation.
		1520
		1521	The design of the POE adaptor class in AnyEvent can not really account
		1522	for the performance issues, though, as session creation overhead is
		1523	small compared to execution of the state machine, which is coded pretty
		1524	optimally within L<AnyEvent::Impl::POE> (and while everybody agrees that
		1525	using multiple sessions is not a good approach, especially regarding
		1526	memory usage, even the author of POE could not come up with a faster
		1527	design).
		1528
		1529	=head3 Summary
		1530
		1531	=over 4
		1532
		1533	=item * Using EV through AnyEvent is faster than any other event loop
		1534	(even when used without AnyEvent), but most event loops have acceptable
		1535	performance with or without AnyEvent.
		1536
		1537	=item * The overhead AnyEvent adds is usually much smaller than the overhead of
		1538	the actual event loop, only with extremely fast event loops such as EV
		1539	adds AnyEvent significant overhead.
		1540
		1541	=item * You should avoid POE like the plague if you want performance or
		1542	reasonable memory usage.
		1543
		1544	=back
		1545
		1546	=head2 BENCHMARKING THE LARGE SERVER CASE
		1547
		1548	This benchmark actually benchmarks the event loop itself. It works by
		1549	creating a number of "servers": each server consists of a socket pair, a
		1550	timeout watcher that gets reset on activity (but never fires), and an I/O
		1551	watcher waiting for input on one side of the socket. Each time the socket
		1552	watcher reads a byte it will write that byte to a random other "server".
		1553
		1554	The effect is that there will be a lot of I/O watchers, only part of which
		1555	are active at any one point (so there is a constant number of active
		1556	fds for each loop iteration, but which fds these are is random). The
		1557	timeout is reset each time something is read because that reflects how
		1558	most timeouts work (and puts extra pressure on the event loops).
		1559
		1560	In this benchmark, we use 10000 socket pairs (20000 sockets), of which 100
		1561	(1%) are active. This mirrors the activity of large servers with many
		1562	connections, most of which are idle at any one point in time.
		1563
		1564	Source code for this benchmark is found as F<eg/bench2> in the AnyEvent
		1565	distribution.
		1566
		1567	=head3 Explanation of the columns
		1568
		1569	I<sockets> is the number of sockets, and twice the number of "servers" (as
		1570	each server has a read and write socket end).
		1571
		1572	I<create> is the time it takes to create a socket pair (which is
		1573	nontrivial) and two watchers: an I/O watcher and a timeout watcher.
		1574
		1575	I<request>, the most important value, is the time it takes to handle a
		1576	single "request", that is, reading the token from the pipe and forwarding
		1577	it to another server. This includes deleting the old timeout and creating
		1578	a new one that moves the timeout into the future.
		1579
		1580	=head3 Results
		1581
		1582	name sockets create request
		1583	EV 20000 69.01 11.16
		1584	Perl 20000 73.32 35.87
		1585	Event 20000 212.62 257.32
		1586	Glib 20000 651.16 1896.30
		1587	POE 20000 349.67 12317.24 uses POE::Loop::Event
		1588
		1589	=head3 Discussion
		1590
		1591	This benchmark I<does> measure scalability and overall performance of the
		1592	particular event loop.
		1593
		1594	EV is again fastest. Since it is using epoll on my system, the setup time
		1595	is relatively high, though.
		1596
		1597	Perl surprisingly comes second. It is much faster than the C-based event
		1598	loops Event and Glib.
		1599
		1600	Event suffers from high setup time as well (look at its code and you will
		1601	understand why). Callback invocation also has a high overhead compared to
		1602	the C<< $_->() for .. >>-style loop that the Perl event loop uses. Event
		1603	uses select or poll in basically all documented configurations.
		1604
		1605	Glib is hit hard by its quadratic behaviour w.r.t. many watchers. It
		1606	clearly fails to perform with many filehandles or in busy servers.
		1607
		1608	POE is still completely out of the picture, taking over 1000 times as long
		1609	as EV, and over 100 times as long as the Perl implementation, even though
		1610	it uses a C-based event loop in this case.
		1611
		1612	=head3 Summary
		1613
		1614	=over 4
		1615
		1616	=item * The pure perl implementation performs extremely well.
		1617
		1618	=item * Avoid Glib or POE in large projects where performance matters.
		1619
		1620	=back
		1621
		1622	=head2 BENCHMARKING SMALL SERVERS
		1623
		1624	While event loops should scale (and select-based ones do not...) even to
		1625	large servers, most programs we (or I :) actually write have only a few
		1626	I/O watchers.
		1627
		1628	In this benchmark, I use the same benchmark program as in the large server
		1629	case, but it uses only eight "servers", of which three are active at any
		1630	one time. This should reflect performance for a small server relatively
		1631	well.
		1632
		1633	The columns are identical to the previous table.
		1634
		1635	=head3 Results
		1636
		1637	name sockets create request
		1638	EV 16 20.00 6.54
		1639	Perl 16 25.75 12.62
		1640	Event 16 81.27 35.86
		1641	Glib 16 32.63 15.48
		1642	POE 16 261.87 276.28 uses POE::Loop::Event
		1643
		1644	=head3 Discussion
		1645
		1646	The benchmark tries to test the performance of a typical small
		1647	server. While knowing how various event loops perform is interesting, keep
		1648	in mind that their overhead in this case is usually not as important, due
		1649	to the small absolute number of watchers (that is, you need efficiency and
		1650	speed most when you have lots of watchers, not when you only have a few of
		1651	them).
		1652
		1653	EV is again fastest.
		1654
		1655	Perl again comes second. It is noticeably faster than the C-based event
		1656	loops Event and Glib, although the difference is too small to really
		1657	matter.
		1658
		1659	POE also performs much better in this case, but is is still far behind the
		1660	others.
		1661
		1662	=head3 Summary
		1663
		1664	=over 4
		1665
		1666	=item * C-based event loops perform very well with small number of
		1667	watchers, as the management overhead dominates.
		1668
		1669	=back
		1670
		1671
		1672	=head1 FORK
		1673
		1674	Most event libraries are not fork-safe. The ones who are usually are
		1675	because they rely on inefficient but fork-safe C<select> or C<poll>
		1676	calls. Only L<EV> is fully fork-aware.
		1677
		1678	If you have to fork, you must either do so I<before> creating your first
		1679	watcher OR you must not use AnyEvent at all in the child.
		1680
		1681
		1682	=head1 SECURITY CONSIDERATIONS
		1683
		1684	AnyEvent can be forced to load any event model via
		1685	$ENV{PERL_ANYEVENT_MODEL}. While this cannot (to my knowledge) be used to
		1686	execute arbitrary code or directly gain access, it can easily be used to
		1687	make the program hang or malfunction in subtle ways, as AnyEvent watchers
		1688	will not be active when the program uses a different event model than
		1689	specified in the variable.
		1690
		1691	You can make AnyEvent completely ignore this variable by deleting it
		1692	before the first watcher gets created, e.g. with a C<BEGIN> block:
		1693
		1694	BEGIN { delete $ENV{PERL_ANYEVENT_MODEL} }
		1695
		1696	use AnyEvent;
		1697
		1698	Similar considerations apply to $ENV{PERL_ANYEVENT_VERBOSE}, as that can
		1699	be used to probe what backend is used and gain other information (which is
		1700	probably even less useful to an attacker than PERL_ANYEVENT_MODEL), and
		1701	$ENV{PERL_ANYEGENT_STRICT}.
		1702
		1703
		1704	=head1 BUGS
		1705
		1706	Perl 5.8 has numerous memleaks that sometimes hit this module and are hard
		1707	to work around. If you suffer from memleaks, first upgrade to Perl 5.10
		1708	and check wether the leaks still show up. (Perl 5.10.0 has other annoying
		1709	mamleaks, such as leaking on C<map> and C<grep> but it is usually not as
		1710	pronounced).
		1711
607		1712
608	=head1 SEE ALSO	1713	=head1 SEE ALSO
609		1714
610	Event modules: L<Coro::Event>, L<Coro>, L<Event>, L<Glib::Event>, L<Glib>.	1715	Utility functions: L<AnyEvent::Util>.
611		1716
612	Implementations: L<AnyEvent::Impl::Coro>, L<AnyEvent::Impl::Event>, L<AnyEvent::Impl::Glib>, L<AnyEvent::Impl::Tk>.	1717	Event modules: L<EV>, L<EV::Glib>, L<Glib::EV>, L<Event>, L<Glib::Event>,
		1718	L<Glib>, L<Tk>, L<Event::Lib>, L<Qt>, L<POE>.
613		1719
614	Nontrivial usage example: L<Net::FCP>.	1720	Implementations: L<AnyEvent::Impl::EV>, L<AnyEvent::Impl::Event>,
		1721	L<AnyEvent::Impl::Glib>, L<AnyEvent::Impl::Tk>, L<AnyEvent::Impl::Perl>,
		1722	L<AnyEvent::Impl::EventLib>, L<AnyEvent::Impl::Qt>,
		1723	L<AnyEvent::Impl::POE>.
615		1724
616	=head1	1725	Non-blocking file handles, sockets, TCP clients and
		1726	servers: L<AnyEvent::Handle>, L<AnyEvent::Socket>.
		1727
		1728	Asynchronous DNS: L<AnyEvent::DNS>.
		1729
		1730	Coroutine support: L<Coro>, L<Coro::AnyEvent>, L<Coro::EV>, L<Coro::Event>,
		1731
		1732	Nontrivial usage examples: L<Net::FCP>, L<Net::XMPP2>, L<AnyEvent::DNS>.
		1733
		1734
		1735	=head1 AUTHOR
		1736
		1737	Marc Lehmann <schmorp@schmorp.de>
		1738	http://home.schmorp.de/
617		1739
618	=cut	1740	=cut
619		1741
620	1	1742	1
621		1743

Diff Legend

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