ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/AnyEvent/lib/AnyEvent.pm

Comparing AnyEvent/lib/AnyEvent.pm (file contents):
Revision 1.1 by root, Wed Apr 27 01:26:44 2005 UTC vs.
Revision 1.209 by root, Wed May 13 13:36:49 2009 UTC

1	=head1 NAME	1	=head1 NAME
2		2
3	AnyEvent - ???	3	AnyEvent - provide framework for multiple event loops
		4
		5	EV, Event, Glib, Tk, Perl, Event::Lib, Qt, POE - various supported event loops
4		6
5	=head1 SYNOPSIS	7	=head1 SYNOPSIS
6		8
		9	use AnyEvent;
		10
		11	# file descriptor readable
		12	my $w = AnyEvent->io (fh => $fh, poll => "r", cb => sub { ... });
		13
		14	# one-shot or repeating timers
		15	my $w = AnyEvent->timer (after => $seconds, cb => sub { ... });
		16	my $w = AnyEvent->timer (after => $seconds, interval => $seconds, cb => ...
		17
		18	print AnyEvent->now; # prints current event loop time
		19	print AnyEvent->time; # think Time::HiRes::time or simply CORE::time.
		20
		21	# POSIX signal
		22	my $w = AnyEvent->signal (signal => "TERM", cb => sub { ... });
		23
		24	# child process exit
		25	my $w = AnyEvent->child (pid => $pid, cb => sub {
		26	my ($pid, $status) = @_;
		27	...
		28	});
		29
		30	# called when event loop idle (if applicable)
		31	my $w = AnyEvent->idle (cb => sub { ... });
		32
		33	my $w = AnyEvent->condvar; # stores whether a condition was flagged
		34	$w->send; # wake up current and all future recv's
		35	$w->recv; # enters "main loop" till $condvar gets ->send
		36	# use a condvar in callback mode:
		37	$w->cb (sub { $_[0]->recv });
		38
		39	=head1 INTRODUCTION/TUTORIAL
		40
		41	This manpage is mainly a reference manual. If you are interested
		42	in a tutorial or some gentle introduction, have a look at the
		43	L<AnyEvent::Intro> manpage.
		44
		45	=head1 WHY YOU SHOULD USE THIS MODULE (OR NOT)
		46
		47	Glib, POE, IO::Async, Event... CPAN offers event models by the dozen
		48	nowadays. So what is different about AnyEvent?
		49
		50	Executive Summary: AnyEvent is I<compatible>, AnyEvent is I<free of
		51	policy> and AnyEvent is I<small and efficient>.
		52
		53	First and foremost, I<AnyEvent is not an event model> itself, it only
		54	interfaces to whatever event model the main program happens to use, in a
		55	pragmatic way. For event models and certain classes of immortals alike,
		56	the statement "there can only be one" is a bitter reality: In general,
		57	only one event loop can be active at the same time in a process. AnyEvent
		58	cannot change this, but it can hide the differences between those event
		59	loops.
		60
		61	The goal of AnyEvent is to offer module authors the ability to do event
		62	programming (waiting for I/O or timer events) without subscribing to a
		63	religion, a way of living, and most importantly: without forcing your
		64	module users into the same thing by forcing them to use the same event
		65	model you use.
		66
		67	For modules like POE or IO::Async (which is a total misnomer as it is
		68	actually doing all I/O I<synchronously>...), using them in your module is
		69	like joining a cult: After you joined, you are dependent on them and you
		70	cannot use anything else, as they are simply incompatible to everything
		71	that isn't them. What's worse, all the potential users of your
		72	module are I<also> forced to use the same event loop you use.
		73
		74	AnyEvent is different: AnyEvent + POE works fine. AnyEvent + Glib works
		75	fine. AnyEvent + Tk works fine etc. etc. but none of these work together
		76	with the rest: POE + IO::Async? No go. Tk + Event? No go. Again: if
		77	your module uses one of those, every user of your module has to use it,
		78	too. But if your module uses AnyEvent, it works transparently with all
		79	event models it supports (including stuff like IO::Async, as long as those
		80	use one of the supported event loops. It is trivial to add new event loops
		81	to AnyEvent, too, so it is future-proof).
		82
		83	In addition to being free of having to use I<the one and only true event
		84	model>, AnyEvent also is free of bloat and policy: with POE or similar
		85	modules, you get an enormous amount of code and strict rules you have to
		86	follow. AnyEvent, on the other hand, is lean and up to the point, by only
		87	offering the functionality that is necessary, in as thin as a wrapper as
		88	technically possible.
		89
		90	Of course, AnyEvent comes with a big (and fully optional!) toolbox
		91	of useful functionality, such as an asynchronous DNS resolver, 100%
		92	non-blocking connects (even with TLS/SSL, IPv6 and on broken platforms
		93	such as Windows) and lots of real-world knowledge and workarounds for
		94	platform bugs and differences.
		95
		96	Now, if you I<do want> lots of policy (this can arguably be somewhat
		97	useful) and you want to force your users to use the one and only event
		98	model, you should I<not> use this module.
		99
7	=head1 DESCRIPTION	100	=head1 DESCRIPTION
8		101
		102	L<AnyEvent> provides an identical interface to multiple event loops. This
		103	allows module authors to utilise an event loop without forcing module
		104	users to use the same event loop (as only a single event loop can coexist
		105	peacefully at any one time).
		106
		107	The interface itself is vaguely similar, but not identical to the L<Event>
		108	module.
		109
		110	During the first call of any watcher-creation method, the module tries
		111	to detect the currently loaded event loop by probing whether one of the
		112	following modules is already loaded: L<EV>,
		113	L<Event>, L<Glib>, L<AnyEvent::Impl::Perl>, L<Tk>, L<Event::Lib>, L<Qt>,
		114	L<POE>. The first one found is used. If none are found, the module tries
		115	to load these modules (excluding Tk, Event::Lib, Qt and POE as the pure perl
		116	adaptor should always succeed) in the order given. The first one that can
		117	be successfully loaded will be used. If, after this, still none could be
		118	found, AnyEvent will fall back to a pure-perl event loop, which is not
		119	very efficient, but should work everywhere.
		120
		121	Because AnyEvent first checks for modules that are already loaded, loading
		122	an event model explicitly before first using AnyEvent will likely make
		123	that model the default. For example:
		124
		125	use Tk;
		126	use AnyEvent;
		127
		128	# .. AnyEvent will likely default to Tk
		129
		130	The I<likely> means that, if any module loads another event model and
		131	starts using it, all bets are off. Maybe you should tell their authors to
		132	use AnyEvent so their modules work together with others seamlessly...
		133
		134	The pure-perl implementation of AnyEvent is called
		135	C<AnyEvent::Impl::Perl>. Like other event modules you can load it
		136	explicitly and enjoy the high availability of that event loop :)
		137
		138	=head1 WATCHERS
		139
		140	AnyEvent has the central concept of a I<watcher>, which is an object that
		141	stores relevant data for each kind of event you are waiting for, such as
		142	the callback to call, the file handle to watch, etc.
		143
		144	These watchers are normal Perl objects with normal Perl lifetime. After
		145	creating a watcher it will immediately "watch" for events and invoke the
		146	callback when the event occurs (of course, only when the event model
		147	is in control).
		148
		149	Note that B<callbacks must not permanently change global variables>
		150	potentially in use by the event loop (such as C<$_> or C<$[>) and that B<<
		151	callbacks must not C<die> >>. The former is good programming practise in
		152	Perl and the latter stems from the fact that exception handling differs
		153	widely between event loops.
		154
		155	To disable the watcher you have to destroy it (e.g. by setting the
		156	variable you store it in to C<undef> or otherwise deleting all references
		157	to it).
		158
		159	All watchers are created by calling a method on the C<AnyEvent> class.
		160
		161	Many watchers either are used with "recursion" (repeating timers for
		162	example), or need to refer to their watcher object in other ways.
		163
		164	An any way to achieve that is this pattern:
		165
		166	my $w; $w = AnyEvent->type (arg => value ..., cb => sub {
		167	# you can use $w here, for example to undef it
		168	undef $w;
		169	});
		170
		171	Note that C<my $w; $w => combination. This is necessary because in Perl,
		172	my variables are only visible after the statement in which they are
		173	declared.
		174
		175	=head2 I/O WATCHERS
		176
		177	You can create an I/O watcher by calling the C<< AnyEvent->io >> method
		178	with the following mandatory key-value pairs as arguments:
		179
		180	C<fh> is the Perl I<file handle> (I<not> file descriptor) to watch
		181	for events (AnyEvent might or might not keep a reference to this file
		182	handle). Note that only file handles pointing to things for which
		183	non-blocking operation makes sense are allowed. This includes sockets,
		184	most character devices, pipes, fifos and so on, but not for example files
		185	or block devices.
		186
		187	C<poll> must be a string that is either C<r> or C<w>, which creates a
		188	watcher waiting for "r"eadable or "w"ritable events, respectively.
		189
		190	C<cb> is the callback to invoke each time the file handle becomes ready.
		191
		192	Although the callback might get passed parameters, their value and
		193	presence is undefined and you cannot rely on them. Portable AnyEvent
		194	callbacks cannot use arguments passed to I/O watcher callbacks.
		195
		196	The I/O watcher might use the underlying file descriptor or a copy of it.
		197	You must not close a file handle as long as any watcher is active on the
		198	underlying file descriptor.
		199
		200	Some event loops issue spurious readyness notifications, so you should
		201	always use non-blocking calls when reading/writing from/to your file
		202	handles.
		203
		204	Example: wait for readability of STDIN, then read a line and disable the
		205	watcher.
		206
		207	my $w; $w = AnyEvent->io (fh => \*STDIN, poll => 'r', cb => sub {
		208	chomp (my $input = <STDIN>);
		209	warn "read: $input\n";
		210	undef $w;
		211	});
		212
		213	=head2 TIME WATCHERS
		214
		215	You can create a time watcher by calling the C<< AnyEvent->timer >>
		216	method with the following mandatory arguments:
		217
		218	C<after> specifies after how many seconds (fractional values are
		219	supported) the callback should be invoked. C<cb> is the callback to invoke
		220	in that case.
		221
		222	Although the callback might get passed parameters, their value and
		223	presence is undefined and you cannot rely on them. Portable AnyEvent
		224	callbacks cannot use arguments passed to time watcher callbacks.
		225
		226	The callback will normally be invoked once only. If you specify another
		227	parameter, C<interval>, as a strictly positive number (> 0), then the
		228	callback will be invoked regularly at that interval (in fractional
		229	seconds) after the first invocation. If C<interval> is specified with a
		230	false value, then it is treated as if it were missing.
		231
		232	The callback will be rescheduled before invoking the callback, but no
		233	attempt is done to avoid timer drift in most backends, so the interval is
		234	only approximate.
		235
		236	Example: fire an event after 7.7 seconds.
		237
		238	my $w = AnyEvent->timer (after => 7.7, cb => sub {
		239	warn "timeout\n";
		240	});
		241
		242	# to cancel the timer:
		243	undef $w;
		244
		245	Example 2: fire an event after 0.5 seconds, then roughly every second.
		246
		247	my $w = AnyEvent->timer (after => 0.5, interval => 1, cb => sub {
		248	warn "timeout\n";
		249	};
		250
		251	=head3 TIMING ISSUES
		252
		253	There are two ways to handle timers: based on real time (relative, "fire
		254	in 10 seconds") and based on wallclock time (absolute, "fire at 12
		255	o'clock").
		256
		257	While most event loops expect timers to specified in a relative way, they
		258	use absolute time internally. This makes a difference when your clock
		259	"jumps", for example, when ntp decides to set your clock backwards from
		260	the wrong date of 2014-01-01 to 2008-01-01, a watcher that is supposed to
		261	fire "after" a second might actually take six years to finally fire.
		262
		263	AnyEvent cannot compensate for this. The only event loop that is conscious
		264	about these issues is L<EV>, which offers both relative (ev_timer, based
		265	on true relative time) and absolute (ev_periodic, based on wallclock time)
		266	timers.
		267
		268	AnyEvent always prefers relative timers, if available, matching the
		269	AnyEvent API.
		270
		271	AnyEvent has two additional methods that return the "current time":
		272
9	=over 4	273	=over 4
10		274
		275	=item AnyEvent->time
		276
		277	This returns the "current wallclock time" as a fractional number of
		278	seconds since the Epoch (the same thing as C<time> or C<Time::HiRes::time>
		279	return, and the result is guaranteed to be compatible with those).
		280
		281	It progresses independently of any event loop processing, i.e. each call
		282	will check the system clock, which usually gets updated frequently.
		283
		284	=item AnyEvent->now
		285
		286	This also returns the "current wallclock time", but unlike C<time>, above,
		287	this value might change only once per event loop iteration, depending on
		288	the event loop (most return the same time as C<time>, above). This is the
		289	time that AnyEvent's timers get scheduled against.
		290
		291	I<In almost all cases (in all cases if you don't care), this is the
		292	function to call when you want to know the current time.>
		293
		294	This function is also often faster then C<< AnyEvent->time >>, and
		295	thus the preferred method if you want some timestamp (for example,
		296	L<AnyEvent::Handle> uses this to update it's activity timeouts).
		297
		298	The rest of this section is only of relevance if you try to be very exact
		299	with your timing, you can skip it without bad conscience.
		300
		301	For a practical example of when these times differ, consider L<Event::Lib>
		302	and L<EV> and the following set-up:
		303
		304	The event loop is running and has just invoked one of your callback at
		305	time=500 (assume no other callbacks delay processing). In your callback,
		306	you wait a second by executing C<sleep 1> (blocking the process for a
		307	second) and then (at time=501) you create a relative timer that fires
		308	after three seconds.
		309
		310	With L<Event::Lib>, C<< AnyEvent->time >> and C<< AnyEvent->now >> will
		311	both return C<501>, because that is the current time, and the timer will
		312	be scheduled to fire at time=504 (C<501> + C<3>).
		313
		314	With L<EV>, C<< AnyEvent->time >> returns C<501> (as that is the current
		315	time), but C<< AnyEvent->now >> returns C<500>, as that is the time the
		316	last event processing phase started. With L<EV>, your timer gets scheduled
		317	to run at time=503 (C<500> + C<3>).
		318
		319	In one sense, L<Event::Lib> is more exact, as it uses the current time
		320	regardless of any delays introduced by event processing. However, most
		321	callbacks do not expect large delays in processing, so this causes a
		322	higher drift (and a lot more system calls to get the current time).
		323
		324	In another sense, L<EV> is more exact, as your timer will be scheduled at
		325	the same time, regardless of how long event processing actually took.
		326
		327	In either case, if you care (and in most cases, you don't), then you
		328	can get whatever behaviour you want with any event loop, by taking the
		329	difference between C<< AnyEvent->time >> and C<< AnyEvent->now >> into
		330	account.
		331
		332	=item AnyEvent->now_update
		333
		334	Some event loops (such as L<EV> or L<AnyEvent::Impl::Perl>) cache
		335	the current time for each loop iteration (see the discussion of L<<
		336	AnyEvent->now >>, above).
		337
		338	When a callback runs for a long time (or when the process sleeps), then
		339	this "current" time will differ substantially from the real time, which
		340	might affect timers and time-outs.
		341
		342	When this is the case, you can call this method, which will update the
		343	event loop's idea of "current time".
		344
		345	Note that updating the time I<might> cause some events to be handled.
		346
		347	=back
		348
		349	=head2 SIGNAL WATCHERS
		350
		351	You can watch for signals using a signal watcher, C<signal> is the signal
		352	I<name> in uppercase and without any C<SIG> prefix, C<cb> is the Perl
		353	callback to be invoked whenever a signal occurs.
		354
		355	Although the callback might get passed parameters, their value and
		356	presence is undefined and you cannot rely on them. Portable AnyEvent
		357	callbacks cannot use arguments passed to signal watcher callbacks.
		358
		359	Multiple signal occurrences can be clumped together into one callback
		360	invocation, and callback invocation will be synchronous. Synchronous means
		361	that it might take a while until the signal gets handled by the process,
		362	but it is guaranteed not to interrupt any other callbacks.
		363
		364	The main advantage of using these watchers is that you can share a signal
		365	between multiple watchers.
		366
		367	This watcher might use C<%SIG>, so programs overwriting those signals
		368	directly will likely not work correctly.
		369
		370	Example: exit on SIGINT
		371
		372	my $w = AnyEvent->signal (signal => "INT", cb => sub { exit 1 });
		373
		374	=head2 CHILD PROCESS WATCHERS
		375
		376	You can also watch on a child process exit and catch its exit status.
		377
		378	The child process is specified by the C<pid> argument (if set to C<0>, it
		379	watches for any child process exit). The watcher will triggered only when
		380	the child process has finished and an exit status is available, not on
		381	any trace events (stopped/continued).
		382
		383	The callback will be called with the pid and exit status (as returned by
		384	waitpid), so unlike other watcher types, you I<can> rely on child watcher
		385	callback arguments.
		386
		387	This watcher type works by installing a signal handler for C<SIGCHLD>,
		388	and since it cannot be shared, nothing else should use SIGCHLD or reap
		389	random child processes (waiting for specific child processes, e.g. inside
		390	C<system>, is just fine).
		391
		392	There is a slight catch to child watchers, however: you usually start them
		393	I<after> the child process was created, and this means the process could
		394	have exited already (and no SIGCHLD will be sent anymore).
		395
		396	Not all event models handle this correctly (POE doesn't), but even for
		397	event models that I<do> handle this correctly, they usually need to be
		398	loaded before the process exits (i.e. before you fork in the first place).
		399
		400	This means you cannot create a child watcher as the very first thing in an
		401	AnyEvent program, you I<have> to create at least one watcher before you
		402	C<fork> the child (alternatively, you can call C<AnyEvent::detect>).
		403
		404	Example: fork a process and wait for it
		405
		406	my $done = AnyEvent->condvar;
		407
		408	my $pid = fork or exit 5;
		409
		410	my $w = AnyEvent->child (
		411	pid => $pid,
		412	cb => sub {
		413	my ($pid, $status) = @_;
		414	warn "pid $pid exited with status $status";
		415	$done->send;
		416	},
		417	);
		418
		419	# do something else, then wait for process exit
		420	$done->recv;
		421
		422	=head2 IDLE WATCHERS
		423
		424	Sometimes there is a need to do something, but it is not so important
		425	to do it instantly, but only when there is nothing better to do. This
		426	"nothing better to do" is usually defined to be "no other events need
		427	attention by the event loop".
		428
		429	Idle watchers ideally get invoked when the event loop has nothing
		430	better to do, just before it would block the process to wait for new
		431	events. Instead of blocking, the idle watcher is invoked.
		432
		433	Most event loops unfortunately do not really support idle watchers (only
		434	EV, Event and Glib do it in a usable fashion) - for the rest, AnyEvent
		435	will simply call the callback "from time to time".
		436
		437	Example: read lines from STDIN, but only process them when the
		438	program is otherwise idle:
		439
		440	my @lines; # read data
		441	my $idle_w;
		442	my $io_w = AnyEvent->io (fh => \*STDIN, poll => 'r', cb => sub {
		443	push @lines, scalar <STDIN>;
		444
		445	# start an idle watcher, if not already done
		446	$idle_w ||= AnyEvent->idle (cb => sub {
		447	# handle only one line, when there are lines left
		448	if (my $line = shift @lines) {
		449	print "handled when idle: $line";
		450	} else {
		451	# otherwise disable the idle watcher again
		452	undef $idle_w;
		453	}
		454	});
		455	});
		456
		457	=head2 CONDITION VARIABLES
		458
		459	If you are familiar with some event loops you will know that all of them
		460	require you to run some blocking "loop", "run" or similar function that
		461	will actively watch for new events and call your callbacks.
		462
		463	AnyEvent is different, it expects somebody else to run the event loop and
		464	will only block when necessary (usually when told by the user).
		465
		466	The instrument to do that is called a "condition variable", so called
		467	because they represent a condition that must become true.
		468
		469	Condition variables can be created by calling the C<< AnyEvent->condvar
		470	>> method, usually without arguments. The only argument pair allowed is
		471
		472	C<cb>, which specifies a callback to be called when the condition variable
		473	becomes true, with the condition variable as the first argument (but not
		474	the results).
		475
		476	After creation, the condition variable is "false" until it becomes "true"
		477	by calling the C<send> method (or calling the condition variable as if it
		478	were a callback, read about the caveats in the description for the C<<
		479	->send >> method).
		480
		481	Condition variables are similar to callbacks, except that you can
		482	optionally wait for them. They can also be called merge points - points
		483	in time where multiple outstanding events have been processed. And yet
		484	another way to call them is transactions - each condition variable can be
		485	used to represent a transaction, which finishes at some point and delivers
		486	a result.
		487
		488	Condition variables are very useful to signal that something has finished,
		489	for example, if you write a module that does asynchronous http requests,
		490	then a condition variable would be the ideal candidate to signal the
		491	availability of results. The user can either act when the callback is
		492	called or can synchronously C<< ->recv >> for the results.
		493
		494	You can also use them to simulate traditional event loops - for example,
		495	you can block your main program until an event occurs - for example, you
		496	could C<< ->recv >> in your main program until the user clicks the Quit
		497	button of your app, which would C<< ->send >> the "quit" event.
		498
		499	Note that condition variables recurse into the event loop - if you have
		500	two pieces of code that call C<< ->recv >> in a round-robin fashion, you
		501	lose. Therefore, condition variables are good to export to your caller, but
		502	you should avoid making a blocking wait yourself, at least in callbacks,
		503	as this asks for trouble.
		504
		505	Condition variables are represented by hash refs in perl, and the keys
		506	used by AnyEvent itself are all named C<_ae_XXX> to make subclassing
		507	easy (it is often useful to build your own transaction class on top of
		508	AnyEvent). To subclass, use C<AnyEvent::CondVar> as base class and call
		509	it's C<new> method in your own C<new> method.
		510
		511	There are two "sides" to a condition variable - the "producer side" which
		512	eventually calls C<< -> send >>, and the "consumer side", which waits
		513	for the send to occur.
		514
		515	Example: wait for a timer.
		516
		517	# wait till the result is ready
		518	my $result_ready = AnyEvent->condvar;
		519
		520	# do something such as adding a timer
		521	# or socket watcher the calls $result_ready->send
		522	# when the "result" is ready.
		523	# in this case, we simply use a timer:
		524	my $w = AnyEvent->timer (
		525	after => 1,
		526	cb => sub { $result_ready->send },
		527	);
		528
		529	# this "blocks" (while handling events) till the callback
		530	# calls send
		531	$result_ready->recv;
		532
		533	Example: wait for a timer, but take advantage of the fact that
		534	condition variables are also code references.
		535
		536	my $done = AnyEvent->condvar;
		537	my $delay = AnyEvent->timer (after => 5, cb => $done);
		538	$done->recv;
		539
		540	Example: Imagine an API that returns a condvar and doesn't support
		541	callbacks. This is how you make a synchronous call, for example from
		542	the main program:
		543
		544	use AnyEvent::CouchDB;
		545
		546	...
		547
		548	my @info = $couchdb->info->recv;
		549
		550	And this is how you would just ste a callback to be called whenever the
		551	results are available:
		552
		553	$couchdb->info->cb (sub {
		554	my @info = $_[0]->recv;
		555	});
		556
		557	=head3 METHODS FOR PRODUCERS
		558
		559	These methods should only be used by the producing side, i.e. the
		560	code/module that eventually sends the signal. Note that it is also
		561	the producer side which creates the condvar in most cases, but it isn't
		562	uncommon for the consumer to create it as well.
		563
		564	=over 4
		565
		566	=item $cv->send (...)
		567
		568	Flag the condition as ready - a running C<< ->recv >> and all further
		569	calls to C<recv> will (eventually) return after this method has been
		570	called. If nobody is waiting the send will be remembered.
		571
		572	If a callback has been set on the condition variable, it is called
		573	immediately from within send.
		574
		575	Any arguments passed to the C<send> call will be returned by all
		576	future C<< ->recv >> calls.
		577
		578	Condition variables are overloaded so one can call them directly
		579	(as a code reference). Calling them directly is the same as calling
		580	C<send>. Note, however, that many C-based event loops do not handle
		581	overloading, so as tempting as it may be, passing a condition variable
		582	instead of a callback does not work. Both the pure perl and EV loops
		583	support overloading, however, as well as all functions that use perl to
		584	invoke a callback (as in L<AnyEvent::Socket> and L<AnyEvent::DNS> for
		585	example).
		586
		587	=item $cv->croak ($error)
		588
		589	Similar to send, but causes all call's to C<< ->recv >> to invoke
		590	C<Carp::croak> with the given error message/object/scalar.
		591
		592	This can be used to signal any errors to the condition variable
		593	user/consumer.
		594
		595	=item $cv->begin ([group callback])
		596
		597	=item $cv->end
		598
		599	These two methods are EXPERIMENTAL and MIGHT CHANGE.
		600
		601	These two methods can be used to combine many transactions/events into
		602	one. For example, a function that pings many hosts in parallel might want
		603	to use a condition variable for the whole process.
		604
		605	Every call to C<< ->begin >> will increment a counter, and every call to
		606	C<< ->end >> will decrement it. If the counter reaches C<0> in C<< ->end
		607	>>, the (last) callback passed to C<begin> will be executed. That callback
		608	is I<supposed> to call C<< ->send >>, but that is not required. If no
		609	callback was set, C<send> will be called without any arguments.
		610
		611	Let's clarify this with the ping example:
		612
		613	my $cv = AnyEvent->condvar;
		614
		615	my %result;
		616	$cv->begin (sub { $cv->send (\%result) });
		617
		618	for my $host (@list_of_hosts) {
		619	$cv->begin;
		620	ping_host_then_call_callback $host, sub {
		621	$result{$host} = ...;
		622	$cv->end;
		623	};
		624	}
		625
		626	$cv->end;
		627
		628	This code fragment supposedly pings a number of hosts and calls
		629	C<send> after results for all then have have been gathered - in any
		630	order. To achieve this, the code issues a call to C<begin> when it starts
		631	each ping request and calls C<end> when it has received some result for
		632	it. Since C<begin> and C<end> only maintain a counter, the order in which
		633	results arrive is not relevant.
		634
		635	There is an additional bracketing call to C<begin> and C<end> outside the
		636	loop, which serves two important purposes: first, it sets the callback
		637	to be called once the counter reaches C<0>, and second, it ensures that
		638	C<send> is called even when C<no> hosts are being pinged (the loop
		639	doesn't execute once).
		640
		641	This is the general pattern when you "fan out" into multiple subrequests:
		642	use an outer C<begin>/C<end> pair to set the callback and ensure C<end>
		643	is called at least once, and then, for each subrequest you start, call
		644	C<begin> and for each subrequest you finish, call C<end>.
		645
		646	=back
		647
		648	=head3 METHODS FOR CONSUMERS
		649
		650	These methods should only be used by the consuming side, i.e. the
		651	code awaits the condition.
		652
		653	=over 4
		654
		655	=item $cv->recv
		656
		657	Wait (blocking if necessary) until the C<< ->send >> or C<< ->croak
		658	>> methods have been called on c<$cv>, while servicing other watchers
		659	normally.
		660
		661	You can only wait once on a condition - additional calls are valid but
		662	will return immediately.
		663
		664	If an error condition has been set by calling C<< ->croak >>, then this
		665	function will call C<croak>.
		666
		667	In list context, all parameters passed to C<send> will be returned,
		668	in scalar context only the first one will be returned.
		669
		670	Not all event models support a blocking wait - some die in that case
		671	(programs might want to do that to stay interactive), so I<if you are
		672	using this from a module, never require a blocking wait>, but let the
		673	caller decide whether the call will block or not (for example, by coupling
		674	condition variables with some kind of request results and supporting
		675	callbacks so the caller knows that getting the result will not block,
		676	while still supporting blocking waits if the caller so desires).
		677
		678	Another reason I<never> to C<< ->recv >> in a module is that you cannot
		679	sensibly have two C<< ->recv >>'s in parallel, as that would require
		680	multiple interpreters or coroutines/threads, none of which C<AnyEvent>
		681	can supply.
		682
		683	The L<Coro> module, however, I<can> and I<does> supply coroutines and, in
		684	fact, L<Coro::AnyEvent> replaces AnyEvent's condvars by coroutine-safe
		685	versions and also integrates coroutines into AnyEvent, making blocking
		686	C<< ->recv >> calls perfectly safe as long as they are done from another
		687	coroutine (one that doesn't run the event loop).
		688
		689	You can ensure that C<< -recv >> never blocks by setting a callback and
		690	only calling C<< ->recv >> from within that callback (or at a later
		691	time). This will work even when the event loop does not support blocking
		692	waits otherwise.
		693
		694	=item $bool = $cv->ready
		695
		696	Returns true when the condition is "true", i.e. whether C<send> or
		697	C<croak> have been called.
		698
		699	=item $cb = $cv->cb ($cb->($cv))
		700
		701	This is a mutator function that returns the callback set and optionally
		702	replaces it before doing so.
		703
		704	The callback will be called when the condition becomes "true", i.e. when
		705	C<send> or C<croak> are called, with the only argument being the condition
		706	variable itself. Calling C<recv> inside the callback or at any later time
		707	is guaranteed not to block.
		708
		709	=back
		710
		711	=head1 GLOBAL VARIABLES AND FUNCTIONS
		712
		713	=over 4
		714
		715	=item $AnyEvent::MODEL
		716
		717	Contains C<undef> until the first watcher is being created. Then it
		718	contains the event model that is being used, which is the name of the
		719	Perl class implementing the model. This class is usually one of the
		720	C<AnyEvent::Impl:xxx> modules, but can be any other class in the case
		721	AnyEvent has been extended at runtime (e.g. in I<rxvt-unicode>).
		722
		723	The known classes so far are:
		724
		725	AnyEvent::Impl::EV based on EV (an interface to libev, best choice).
		726	AnyEvent::Impl::Event based on Event, second best choice.
		727	AnyEvent::Impl::Perl pure-perl implementation, fast and portable.
		728	AnyEvent::Impl::Glib based on Glib, third-best choice.
		729	AnyEvent::Impl::Tk based on Tk, very bad choice.
		730	AnyEvent::Impl::Qt based on Qt, cannot be autoprobed (see its docs).
		731	AnyEvent::Impl::EventLib based on Event::Lib, leaks memory and worse.
		732	AnyEvent::Impl::POE based on POE, not generic enough for full support.
		733
		734	There is no support for WxWidgets, as WxWidgets has no support for
		735	watching file handles. However, you can use WxWidgets through the
		736	POE Adaptor, as POE has a Wx backend that simply polls 20 times per
		737	second, which was considered to be too horrible to even consider for
		738	AnyEvent. Likewise, other POE backends can be used by AnyEvent by using
		739	it's adaptor.
		740
		741	AnyEvent knows about L<Prima> and L<Wx> and will try to use L<POE> when
		742	autodetecting them.
		743
		744	=item AnyEvent::detect
		745
		746	Returns C<$AnyEvent::MODEL>, forcing autodetection of the event model
		747	if necessary. You should only call this function right before you would
		748	have created an AnyEvent watcher anyway, that is, as late as possible at
		749	runtime.
		750
		751	=item $guard = AnyEvent::post_detect { BLOCK }
		752
		753	Arranges for the code block to be executed as soon as the event model is
		754	autodetected (or immediately if this has already happened).
		755
		756	If called in scalar or list context, then it creates and returns an object
		757	that automatically removes the callback again when it is destroyed. See
		758	L<Coro::BDB> for a case where this is useful.
		759
		760	=item @AnyEvent::post_detect
		761
		762	If there are any code references in this array (you can C<push> to it
		763	before or after loading AnyEvent), then they will called directly after
		764	the event loop has been chosen.
		765
		766	You should check C<$AnyEvent::MODEL> before adding to this array, though:
		767	if it contains a true value then the event loop has already been detected,
		768	and the array will be ignored.
		769
		770	Best use C<AnyEvent::post_detect { BLOCK }> instead.
		771
		772	=back
		773
		774	=head1 WHAT TO DO IN A MODULE
		775
		776	As a module author, you should C<use AnyEvent> and call AnyEvent methods
		777	freely, but you should not load a specific event module or rely on it.
		778
		779	Be careful when you create watchers in the module body - AnyEvent will
		780	decide which event module to use as soon as the first method is called, so
		781	by calling AnyEvent in your module body you force the user of your module
		782	to load the event module first.
		783
		784	Never call C<< ->recv >> on a condition variable unless you I<know> that
		785	the C<< ->send >> method has been called on it already. This is
		786	because it will stall the whole program, and the whole point of using
		787	events is to stay interactive.
		788
		789	It is fine, however, to call C<< ->recv >> when the user of your module
		790	requests it (i.e. if you create a http request object ad have a method
		791	called C<results> that returns the results, it should call C<< ->recv >>
		792	freely, as the user of your module knows what she is doing. always).
		793
		794	=head1 WHAT TO DO IN THE MAIN PROGRAM
		795
		796	There will always be a single main program - the only place that should
		797	dictate which event model to use.
		798
		799	If it doesn't care, it can just "use AnyEvent" and use it itself, or not
		800	do anything special (it does not need to be event-based) and let AnyEvent
		801	decide which implementation to chose if some module relies on it.
		802
		803	If the main program relies on a specific event model - for example, in
		804	Gtk2 programs you have to rely on the Glib module - you should load the
		805	event module before loading AnyEvent or any module that uses it: generally
		806	speaking, you should load it as early as possible. The reason is that
		807	modules might create watchers when they are loaded, and AnyEvent will
		808	decide on the event model to use as soon as it creates watchers, and it
		809	might chose the wrong one unless you load the correct one yourself.
		810
		811	You can chose to use a pure-perl implementation by loading the
		812	C<AnyEvent::Impl::Perl> module, which gives you similar behaviour
		813	everywhere, but letting AnyEvent chose the model is generally better.
		814
		815	=head2 MAINLOOP EMULATION
		816
		817	Sometimes (often for short test scripts, or even standalone programs who
		818	only want to use AnyEvent), you do not want to run a specific event loop.
		819
		820	In that case, you can use a condition variable like this:
		821
		822	AnyEvent->condvar->recv;
		823
		824	This has the effect of entering the event loop and looping forever.
		825
		826	Note that usually your program has some exit condition, in which case
		827	it is better to use the "traditional" approach of storing a condition
		828	variable somewhere, waiting for it, and sending it when the program should
		829	exit cleanly.
		830
		831
		832	=head1 OTHER MODULES
		833
		834	The following is a non-exhaustive list of additional modules that use
		835	AnyEvent and can therefore be mixed easily with other AnyEvent modules
		836	in the same program. Some of the modules come with AnyEvent, some are
		837	available via CPAN.
		838
		839	=over 4
		840
		841	=item L<AnyEvent::Util>
		842
		843	Contains various utility functions that replace often-used but blocking
		844	functions such as C<inet_aton> by event-/callback-based versions.
		845
		846	=item L<AnyEvent::Socket>
		847
		848	Provides various utility functions for (internet protocol) sockets,
		849	addresses and name resolution. Also functions to create non-blocking tcp
		850	connections or tcp servers, with IPv6 and SRV record support and more.
		851
		852	=item L<AnyEvent::Handle>
		853
		854	Provide read and write buffers, manages watchers for reads and writes,
		855	supports raw and formatted I/O, I/O queued and fully transparent and
		856	non-blocking SSL/TLS.
		857
		858	=item L<AnyEvent::DNS>
		859
		860	Provides rich asynchronous DNS resolver capabilities.
		861
		862	=item L<AnyEvent::HTTP>
		863
		864	A simple-to-use HTTP library that is capable of making a lot of concurrent
		865	HTTP requests.
		866
		867	=item L<AnyEvent::HTTPD>
		868
		869	Provides a simple web application server framework.
		870
		871	=item L<AnyEvent::FastPing>
		872
		873	The fastest ping in the west.
		874
		875	=item L<AnyEvent::DBI>
		876
		877	Executes L<DBI> requests asynchronously in a proxy process.
		878
		879	=item L<AnyEvent::AIO>
		880
		881	Truly asynchronous I/O, should be in the toolbox of every event
		882	programmer. AnyEvent::AIO transparently fuses L<IO::AIO> and AnyEvent
		883	together.
		884
		885	=item L<AnyEvent::BDB>
		886
		887	Truly asynchronous Berkeley DB access. AnyEvent::BDB transparently fuses
		888	L<BDB> and AnyEvent together.
		889
		890	=item L<AnyEvent::GPSD>
		891
		892	A non-blocking interface to gpsd, a daemon delivering GPS information.
		893
		894	=item L<AnyEvent::IGS>
		895
		896	A non-blocking interface to the Internet Go Server protocol (used by
		897	L<App::IGS>).
		898
		899	=item L<AnyEvent::IRC>
		900
		901	AnyEvent based IRC client module family (replacing the older Net::IRC3).
		902
		903	=item L<Net::XMPP2>
		904
		905	AnyEvent based XMPP (Jabber protocol) module family.
		906
		907	=item L<Net::FCP>
		908
		909	AnyEvent-based implementation of the Freenet Client Protocol, birthplace
		910	of AnyEvent.
		911
		912	=item L<Event::ExecFlow>
		913
		914	High level API for event-based execution flow control.
		915
		916	=item L<Coro>
		917
		918	Has special support for AnyEvent via L<Coro::AnyEvent>.
		919
		920	=item L<IO::Lambda>
		921
		922	The lambda approach to I/O - don't ask, look there. Can use AnyEvent.
		923
		924	=back
		925
11	=cut	926	=cut
12		927
13	package AnyEvent;	928	package AnyEvent;
14		929
		930	no warnings;
		931	use strict qw(vars subs);
		932
15	use Carp;	933	use Carp;
16		934
17	$VERSION = 0.1;	935	our $VERSION = 4.41;
		936	our $MODEL;
18		937
19	no warnings;	938	our $AUTOLOAD;
		939	our @ISA;
		940
		941	our @REGISTRY;
		942
		943	our $WIN32;
		944
		945	BEGIN {
		946	my $win32 = ! ! ($^O =~ /mswin32/i);
		947	eval "sub WIN32(){ $win32 }";
		948	}
		949
		950	our $verbose = $ENV{PERL_ANYEVENT_VERBOSE}*1;
		951
		952	our %PROTOCOL; # (ipv4|ipv6) => (1|2), higher numbers are preferred
		953
		954	{
		955	my $idx;
		956	$PROTOCOL{$_} = ++$idx
		957	for reverse split /\s*,\s*/,
		958	$ENV{PERL_ANYEVENT_PROTOCOLS} || "ipv4,ipv6";
		959	}
20		960
21	my @models = (	961	my @models = (
22	[Coro => Coro::Event::],	962	[EV:: => AnyEvent::Impl::EV::],
23	[Event => Event::],	963	[Event:: => AnyEvent::Impl::Event::],
24	[Glib => Glib::],	964	[AnyEvent::Impl::Perl:: => AnyEvent::Impl::Perl::],
25	[Tk => Tk::],	965	# everything below here will not be autoprobed
		966	# as the pureperl backend should work everywhere
		967	# and is usually faster
		968	[Tk:: => AnyEvent::Impl::Tk::], # crashes with many handles
		969	[Glib:: => AnyEvent::Impl::Glib::], # becomes extremely slow with many watchers
		970	[Event::Lib:: => AnyEvent::Impl::EventLib::], # too buggy
		971	[Qt:: => AnyEvent::Impl::Qt::], # requires special main program
		972	[POE::Kernel:: => AnyEvent::Impl::POE::], # lasciate ogni speranza
		973	[Wx:: => AnyEvent::Impl::POE::],
		974	[Prima:: => AnyEvent::Impl::POE::],
26	);	975	);
27		976
		977	our %method = map +($_ => 1),
		978	qw(io timer time now now_update signal child idle condvar one_event DESTROY);
		979
		980	our @post_detect;
		981
		982	sub post_detect(&) {
		983	my ($cb) = @_;
		984
		985	if ($MODEL) {
		986	$cb->();
		987
		988	1
		989	} else {
		990	push @post_detect, $cb;
		991
		992	defined wantarray
		993	? bless \$cb, "AnyEvent::Util::postdetect"
		994	: ()
		995	}
		996	}
		997
		998	sub AnyEvent::Util::postdetect::DESTROY {
		999	@post_detect = grep $_ != ${$_[0]}, @post_detect;
		1000	}
		1001
		1002	sub detect() {
		1003	unless ($MODEL) {
		1004	no strict 'refs';
		1005	local $SIG{__DIE__};
		1006
		1007	if ($ENV{PERL_ANYEVENT_MODEL} =~ /^([a-zA-Z]+)$/) {
		1008	my $model = "AnyEvent::Impl::$1";
		1009	if (eval "require $model") {
		1010	$MODEL = $model;
		1011	warn "AnyEvent: loaded model '$model' (forced by \$PERL_ANYEVENT_MODEL), using it.\n" if $verbose > 1;
		1012	} else {
		1013	warn "AnyEvent: unable to load model '$model' (from \$PERL_ANYEVENT_MODEL):\n$@" if $verbose;
		1014	}
		1015	}
		1016
		1017	# check for already loaded models
		1018	unless ($MODEL) {
		1019	for (@REGISTRY, @models) {
		1020	my ($package, $model) = @$_;
		1021	if (${"$package\::VERSION"} > 0) {
		1022	if (eval "require $model") {
		1023	$MODEL = $model;
		1024	warn "AnyEvent: autodetected model '$model', using it.\n" if $verbose > 1;
		1025	last;
		1026	}
		1027	}
		1028	}
		1029
		1030	unless ($MODEL) {
		1031	# try to load a model
		1032
		1033	for (@REGISTRY, @models) {
		1034	my ($package, $model) = @$_;
		1035	if (eval "require $package"
		1036	and ${"$package\::VERSION"} > 0
		1037	and eval "require $model") {
		1038	$MODEL = $model;
		1039	warn "AnyEvent: autoprobed model '$model', using it.\n" if $verbose > 1;
		1040	last;
		1041	}
		1042	}
		1043
		1044	$MODEL
		1045	or die "No event module selected for AnyEvent and autodetect failed. Install any one of these modules: EV, Event or Glib.\n";
		1046	}
		1047	}
		1048
		1049	push @{"$MODEL\::ISA"}, "AnyEvent::Base";
		1050
		1051	unshift @ISA, $MODEL;
		1052
		1053	require AnyEvent::Strict if $ENV{PERL_ANYEVENT_STRICT};
		1054
		1055	(shift @post_detect)->() while @post_detect;
		1056	}
		1057
		1058	$MODEL
		1059	}
		1060
28	sub AUTOLOAD {	1061	sub AUTOLOAD {
29	$AUTOLOAD =~ s/.*://;	1062	(my $func = $AUTOLOAD) =~ s/.*://;
30		1063
31	for (@models) {	1064	$method{$func}
32	my ($model, $package) = @$_;	1065	or croak "$func: not a valid method for AnyEvent objects";
33	if (defined ${"$package\::VERSION"}) {	1066
34	$EVENT = "AnyEvent::Impl::$model";	1067	detect unless $MODEL;
35	eval "require $EVENT"; die if $@;	1068
36	goto &{"$EVENT\::$AUTOLOAD"};	1069	my $class = shift;
		1070	$class->$func (@_);
		1071	}
		1072
		1073	# utility function to dup a filehandle. this is used by many backends
		1074	# to support binding more than one watcher per filehandle (they usually
		1075	# allow only one watcher per fd, so we dup it to get a different one).
		1076	sub _dupfh($$$$) {
		1077	my ($poll, $fh, $r, $w) = @_;
		1078
		1079	# cygwin requires the fh mode to be matching, unix doesn't
		1080	my ($rw, $mode) = $poll eq "r" ? ($r, "<")
		1081	: $poll eq "w" ? ($w, ">")
		1082	: Carp::croak "AnyEvent->io requires poll set to either 'r' or 'w'";
		1083
		1084	open my $fh2, "$mode&" . fileno $fh
		1085	or die "cannot dup() filehandle: $!,";
		1086
		1087	# we assume CLOEXEC is already set by perl in all important cases
		1088
		1089	($fh2, $rw)
		1090	}
		1091
		1092	package AnyEvent::Base;
		1093
		1094	# default implementations for many methods
		1095
		1096	BEGIN {
		1097	if (eval "use Time::HiRes (); Time::HiRes::time (); 1") {
		1098	*_time = \&Time::HiRes::time;
		1099	# if (eval "use POSIX (); (POSIX::times())...
		1100	} else {
		1101	*_time = sub { time }; # epic fail
		1102	}
		1103	}
		1104
		1105	sub time { _time }
		1106	sub now { _time }
		1107	sub now_update { }
		1108
		1109	# default implementation for ->condvar
		1110
		1111	sub condvar {
		1112	bless { @_ == 3 ? (_ae_cb => $_[2]) : () }, "AnyEvent::CondVar"
		1113	}
		1114
		1115	# default implementation for ->signal
		1116
		1117	our ($SIGPIPE_R, $SIGPIPE_W, %SIG_CB, %SIG_EV, $SIG_IO);
		1118
		1119	sub _signal_exec {
		1120	sysread $SIGPIPE_R, my $dummy, 4;
		1121
		1122	while (%SIG_EV) {
		1123	for (keys %SIG_EV) {
		1124	delete $SIG_EV{$_};
		1125	$_->() for values %{ $SIG_CB{$_} || {} };
37	}	1126	}
38	}	1127	}
		1128	}
39		1129
40	for (@models) {	1130	sub signal {
41	my ($model, $package) = @$_;	1131	my (undef, %arg) = @_;
42	$EVENT = "AnyEvent::Impl::$model";	1132
43	if (eval "require $EVENT") {	1133	unless ($SIGPIPE_R) {
44	goto &{"$EVENT\::$AUTOLOAD"};	1134	require Fcntl;
		1135
		1136	if (AnyEvent::WIN32) {
		1137	require AnyEvent::Util;
		1138
		1139	($SIGPIPE_R, $SIGPIPE_W) = AnyEvent::Util::portable_pipe ();
		1140	AnyEvent::Util::fh_nonblocking ($SIGPIPE_R) if $SIGPIPE_R;
		1141	AnyEvent::Util::fh_nonblocking ($SIGPIPE_W) if $SIGPIPE_W; # just in case
		1142	} else {
		1143	pipe $SIGPIPE_R, $SIGPIPE_W;
		1144	fcntl $SIGPIPE_R, &Fcntl::F_SETFL, &Fcntl::O_NONBLOCK if $SIGPIPE_R;
		1145	fcntl $SIGPIPE_W, &Fcntl::F_SETFL, &Fcntl::O_NONBLOCK if $SIGPIPE_W; # just in case
45	}	1146	}
		1147
		1148	$SIGPIPE_R
		1149	or Carp::croak "AnyEvent: unable to create a signal reporting pipe: $!\n";
		1150
		1151	# not strictly required, as $^F is normally 2, but let's make sure...
		1152	fcntl $SIGPIPE_R, &Fcntl::F_SETFD, &Fcntl::FD_CLOEXEC;
		1153	fcntl $SIGPIPE_W, &Fcntl::F_SETFD, &Fcntl::FD_CLOEXEC;
		1154
		1155	$SIG_IO = AnyEvent->io (fh => $SIGPIPE_R, poll => "r", cb => \&_signal_exec);
46	}	1156	}
47		1157
48	die "No event module selected for AnyEvent and autodetect failed. Install any of these: Coro, Event, Glib or Tk.";	1158	my $signal = uc $arg{signal}
49	}	1159	or Carp::croak "required option 'signal' is missing";
50		1160
51	1;	1161	$SIG_CB{$signal}{$arg{cb}} = $arg{cb};
		1162	$SIG{$signal} ||= sub {
		1163	local $!;
		1164	syswrite $SIGPIPE_W, "\x00", 1 unless %SIG_EV;
		1165	undef $SIG_EV{$signal};
		1166	};
52		1167
		1168	bless [$signal, $arg{cb}], "AnyEvent::Base::signal"
		1169	}
		1170
		1171	sub AnyEvent::Base::signal::DESTROY {
		1172	my ($signal, $cb) = @{$_[0]};
		1173
		1174	delete $SIG_CB{$signal}{$cb};
		1175
		1176	$SIG{$signal} = 'DEFAULT' unless keys %{ $SIG_CB{$signal} };
		1177	}
		1178
		1179	# default implementation for ->child
		1180
		1181	our %PID_CB;
		1182	our $CHLD_W;
		1183	our $CHLD_DELAY_W;
		1184	our $PID_IDLE;
		1185	our $WNOHANG;
		1186
		1187	sub _child_wait {
		1188	while (0 < (my $pid = waitpid -1, $WNOHANG)) {
		1189	$_->($pid, $?) for (values %{ $PID_CB{$pid} || {} }),
		1190	(values %{ $PID_CB{0} || {} });
		1191	}
		1192
		1193	undef $PID_IDLE;
		1194	}
		1195
		1196	sub _sigchld {
		1197	# make sure we deliver these changes "synchronous" with the event loop.
		1198	$CHLD_DELAY_W ||= AnyEvent->timer (after => 0, cb => sub {
		1199	undef $CHLD_DELAY_W;
		1200	&_child_wait;
		1201	});
		1202	}
		1203
		1204	sub child {
		1205	my (undef, %arg) = @_;
		1206
		1207	defined (my $pid = $arg{pid} + 0)
		1208	or Carp::croak "required option 'pid' is missing";
		1209
		1210	$PID_CB{$pid}{$arg{cb}} = $arg{cb};
		1211
		1212	unless ($WNOHANG) {
		1213	$WNOHANG = eval { local $SIG{__DIE__}; require POSIX; &POSIX::WNOHANG } || 1;
		1214	}
		1215
		1216	unless ($CHLD_W) {
		1217	$CHLD_W = AnyEvent->signal (signal => 'CHLD', cb => \&_sigchld);
		1218	# child could be a zombie already, so make at least one round
		1219	&_sigchld;
		1220	}
		1221
		1222	bless [$pid, $arg{cb}], "AnyEvent::Base::child"
		1223	}
		1224
		1225	sub AnyEvent::Base::child::DESTROY {
		1226	my ($pid, $cb) = @{$_[0]};
		1227
		1228	delete $PID_CB{$pid}{$cb};
		1229	delete $PID_CB{$pid} unless keys %{ $PID_CB{$pid} };
		1230
		1231	undef $CHLD_W unless keys %PID_CB;
		1232	}
		1233
		1234	# idle emulation is done by simply using a timer, regardless
		1235	# of whether the proces sis idle or not, and not letting
		1236	# the callback use more than 50% of the time.
		1237	sub idle {
		1238	my (undef, %arg) = @_;
		1239
		1240	my ($cb, $w, $rcb) = $arg{cb};
		1241
		1242	$rcb = sub {
		1243	if ($cb) {
		1244	$w = _time;
		1245	&$cb;
		1246	$w = _time - $w;
		1247
		1248	# never use more then 50% of the time for the idle watcher,
		1249	# within some limits
		1250	$w = 0.0001 if $w < 0.0001;
		1251	$w = 5 if $w > 5;
		1252
		1253	$w = AnyEvent->timer (after => $w, cb => $rcb);
		1254	} else {
		1255	# clean up...
		1256	undef $w;
		1257	undef $rcb;
		1258	}
		1259	};
		1260
		1261	$w = AnyEvent->timer (after => 0.05, cb => $rcb);
		1262
		1263	bless \\$cb, "AnyEvent::Base::idle"
		1264	}
		1265
		1266	sub AnyEvent::Base::idle::DESTROY {
		1267	undef $${$_[0]};
		1268	}
		1269
		1270	package AnyEvent::CondVar;
		1271
		1272	our @ISA = AnyEvent::CondVar::Base::;
		1273
		1274	package AnyEvent::CondVar::Base;
		1275
		1276	use overload
		1277	'&{}' => sub { my $self = shift; sub { $self->send (@_) } },
		1278	fallback => 1;
		1279
		1280	sub _send {
		1281	# nop
		1282	}
		1283
		1284	sub send {
		1285	my $cv = shift;
		1286	$cv->{_ae_sent} = [@_];
		1287	(delete $cv->{_ae_cb})->($cv) if $cv->{_ae_cb};
		1288	$cv->_send;
		1289	}
		1290
		1291	sub croak {
		1292	$_[0]{_ae_croak} = $_[1];
		1293	$_[0]->send;
		1294	}
		1295
		1296	sub ready {
		1297	$_[0]{_ae_sent}
		1298	}
		1299
		1300	sub _wait {
		1301	AnyEvent->one_event while !$_[0]{_ae_sent};
		1302	}
		1303
		1304	sub recv {
		1305	$_[0]->_wait;
		1306
		1307	Carp::croak $_[0]{_ae_croak} if $_[0]{_ae_croak};
		1308	wantarray ? @{ $_[0]{_ae_sent} } : $_[0]{_ae_sent}[0]
		1309	}
		1310
		1311	sub cb {
		1312	$_[0]{_ae_cb} = $_[1] if @_ > 1;
		1313	$_[0]{_ae_cb}
		1314	}
		1315
		1316	sub begin {
		1317	++$_[0]{_ae_counter};
		1318	$_[0]{_ae_end_cb} = $_[1] if @_ > 1;
		1319	}
		1320
		1321	sub end {
		1322	return if --$_[0]{_ae_counter};
		1323	&{ $_[0]{_ae_end_cb} || sub { $_[0]->send } };
		1324	}
		1325
		1326	# undocumented/compatibility with pre-3.4
		1327	*broadcast = \&send;
		1328	*wait = \&_wait;
		1329
		1330	=head1 ERROR AND EXCEPTION HANDLING
		1331
		1332	In general, AnyEvent does not do any error handling - it relies on the
		1333	caller to do that if required. The L<AnyEvent::Strict> module (see also
		1334	the C<PERL_ANYEVENT_STRICT> environment variable, below) provides strict
		1335	checking of all AnyEvent methods, however, which is highly useful during
		1336	development.
		1337
		1338	As for exception handling (i.e. runtime errors and exceptions thrown while
		1339	executing a callback), this is not only highly event-loop specific, but
		1340	also not in any way wrapped by this module, as this is the job of the main
		1341	program.
		1342
		1343	The pure perl event loop simply re-throws the exception (usually
		1344	within C<< condvar->recv >>), the L<Event> and L<EV> modules call C<<
		1345	$Event/EV::DIED->() >>, L<Glib> uses C<< install_exception_handler >> and
		1346	so on.
		1347
		1348	=head1 ENVIRONMENT VARIABLES
		1349
		1350	The following environment variables are used by this module or its
		1351	submodules:
		1352
		1353	=over 4
		1354
		1355	=item C<PERL_ANYEVENT_VERBOSE>
		1356
		1357	By default, AnyEvent will be completely silent except in fatal
		1358	conditions. You can set this environment variable to make AnyEvent more
		1359	talkative.
		1360
		1361	When set to C<1> or higher, causes AnyEvent to warn about unexpected
		1362	conditions, such as not being able to load the event model specified by
		1363	C<PERL_ANYEVENT_MODEL>.
		1364
		1365	When set to C<2> or higher, cause AnyEvent to report to STDERR which event
		1366	model it chooses.
		1367
		1368	=item C<PERL_ANYEVENT_STRICT>
		1369
		1370	AnyEvent does not do much argument checking by default, as thorough
		1371	argument checking is very costly. Setting this variable to a true value
		1372	will cause AnyEvent to load C<AnyEvent::Strict> and then to thoroughly
		1373	check the arguments passed to most method calls. If it finds any problems
		1374	it will croak.
		1375
		1376	In other words, enables "strict" mode.
		1377
		1378	Unlike C<use strict>, it is definitely recommended ot keep it off in
		1379	production. Keeping C<PERL_ANYEVENT_STRICT=1> in your environment while
		1380	developing programs can be very useful, however.
		1381
		1382	=item C<PERL_ANYEVENT_MODEL>
		1383
		1384	This can be used to specify the event model to be used by AnyEvent, before
		1385	auto detection and -probing kicks in. It must be a string consisting
		1386	entirely of ASCII letters. The string C<AnyEvent::Impl::> gets prepended
		1387	and the resulting module name is loaded and if the load was successful,
		1388	used as event model. If it fails to load AnyEvent will proceed with
		1389	auto detection and -probing.
		1390
		1391	This functionality might change in future versions.
		1392
		1393	For example, to force the pure perl model (L<AnyEvent::Impl::Perl>) you
		1394	could start your program like this:
		1395
		1396	PERL_ANYEVENT_MODEL=Perl perl ...
		1397
		1398	=item C<PERL_ANYEVENT_PROTOCOLS>
		1399
		1400	Used by both L<AnyEvent::DNS> and L<AnyEvent::Socket> to determine preferences
		1401	for IPv4 or IPv6. The default is unspecified (and might change, or be the result
		1402	of auto probing).
		1403
		1404	Must be set to a comma-separated list of protocols or address families,
		1405	current supported: C<ipv4> and C<ipv6>. Only protocols mentioned will be
		1406	used, and preference will be given to protocols mentioned earlier in the
		1407	list.
		1408
		1409	This variable can effectively be used for denial-of-service attacks
		1410	against local programs (e.g. when setuid), although the impact is likely
		1411	small, as the program has to handle conenction and other failures anyways.
		1412
		1413	Examples: C<PERL_ANYEVENT_PROTOCOLS=ipv4,ipv6> - prefer IPv4 over IPv6,
		1414	but support both and try to use both. C<PERL_ANYEVENT_PROTOCOLS=ipv4>
		1415	- only support IPv4, never try to resolve or contact IPv6
		1416	addresses. C<PERL_ANYEVENT_PROTOCOLS=ipv6,ipv4> support either IPv4 or
		1417	IPv6, but prefer IPv6 over IPv4.
		1418
		1419	=item C<PERL_ANYEVENT_EDNS0>
		1420
		1421	Used by L<AnyEvent::DNS> to decide whether to use the EDNS0 extension
		1422	for DNS. This extension is generally useful to reduce DNS traffic, but
		1423	some (broken) firewalls drop such DNS packets, which is why it is off by
		1424	default.
		1425
		1426	Setting this variable to C<1> will cause L<AnyEvent::DNS> to announce
		1427	EDNS0 in its DNS requests.
		1428
		1429	=item C<PERL_ANYEVENT_MAX_FORKS>
		1430
		1431	The maximum number of child processes that C<AnyEvent::Util::fork_call>
		1432	will create in parallel.
		1433
		1434	=back
		1435
		1436	=head1 SUPPLYING YOUR OWN EVENT MODEL INTERFACE
		1437
		1438	This is an advanced topic that you do not normally need to use AnyEvent in
		1439	a module. This section is only of use to event loop authors who want to
		1440	provide AnyEvent compatibility.
		1441
		1442	If you need to support another event library which isn't directly
		1443	supported by AnyEvent, you can supply your own interface to it by
		1444	pushing, before the first watcher gets created, the package name of
		1445	the event module and the package name of the interface to use onto
		1446	C<@AnyEvent::REGISTRY>. You can do that before and even without loading
		1447	AnyEvent, so it is reasonably cheap.
		1448
		1449	Example:
		1450
		1451	push @AnyEvent::REGISTRY, [urxvt => urxvt::anyevent::];
		1452
		1453	This tells AnyEvent to (literally) use the C<urxvt::anyevent::>
		1454	package/class when it finds the C<urxvt> package/module is already loaded.
		1455
		1456	When AnyEvent is loaded and asked to find a suitable event model, it
		1457	will first check for the presence of urxvt by trying to C<use> the
		1458	C<urxvt::anyevent> module.
		1459
		1460	The class should provide implementations for all watcher types. See
		1461	L<AnyEvent::Impl::EV> (source code), L<AnyEvent::Impl::Glib> (Source code)
		1462	and so on for actual examples. Use C<perldoc -m AnyEvent::Impl::Glib> to
		1463	see the sources.
		1464
		1465	If you don't provide C<signal> and C<child> watchers than AnyEvent will
		1466	provide suitable (hopefully) replacements.
		1467
		1468	The above example isn't fictitious, the I<rxvt-unicode> (a.k.a. urxvt)
		1469	terminal emulator uses the above line as-is. An interface isn't included
		1470	in AnyEvent because it doesn't make sense outside the embedded interpreter
		1471	inside I<rxvt-unicode>, and it is updated and maintained as part of the
		1472	I<rxvt-unicode> distribution.
		1473
		1474	I<rxvt-unicode> also cheats a bit by not providing blocking access to
		1475	condition variables: code blocking while waiting for a condition will
		1476	C<die>. This still works with most modules/usages, and blocking calls must
		1477	not be done in an interactive application, so it makes sense.
		1478
		1479	=head1 EXAMPLE PROGRAM
		1480
		1481	The following program uses an I/O watcher to read data from STDIN, a timer
		1482	to display a message once per second, and a condition variable to quit the
		1483	program when the user enters quit:
		1484
		1485	use AnyEvent;
		1486
		1487	my $cv = AnyEvent->condvar;
		1488
		1489	my $io_watcher = AnyEvent->io (
		1490	fh => \*STDIN,
		1491	poll => 'r',
		1492	cb => sub {
		1493	warn "io event <$_[0]>\n"; # will always output <r>
		1494	chomp (my $input = <STDIN>); # read a line
		1495	warn "read: $input\n"; # output what has been read
		1496	$cv->send if $input =~ /^q/i; # quit program if /^q/i
		1497	},
		1498	);
		1499
		1500	my $time_watcher; # can only be used once
		1501
		1502	sub new_timer {
		1503	$timer = AnyEvent->timer (after => 1, cb => sub {
		1504	warn "timeout\n"; # print 'timeout' about every second
		1505	&new_timer; # and restart the time
		1506	});
		1507	}
		1508
		1509	new_timer; # create first timer
		1510
		1511	$cv->recv; # wait until user enters /^q/i
		1512
		1513	=head1 REAL-WORLD EXAMPLE
		1514
		1515	Consider the L<Net::FCP> module. It features (among others) the following
		1516	API calls, which are to freenet what HTTP GET requests are to http:
		1517
		1518	my $data = $fcp->client_get ($url); # blocks
		1519
		1520	my $transaction = $fcp->txn_client_get ($url); # does not block
		1521	$transaction->cb ( sub { ... } ); # set optional result callback
		1522	my $data = $transaction->result; # possibly blocks
		1523
		1524	The C<client_get> method works like C<LWP::Simple::get>: it requests the
		1525	given URL and waits till the data has arrived. It is defined to be:
		1526
		1527	sub client_get { $_[0]->txn_client_get ($_[1])->result }
		1528
		1529	And in fact is automatically generated. This is the blocking API of
		1530	L<Net::FCP>, and it works as simple as in any other, similar, module.
		1531
		1532	More complicated is C<txn_client_get>: It only creates a transaction
		1533	(completion, result, ...) object and initiates the transaction.
		1534
		1535	my $txn = bless { }, Net::FCP::Txn::;
		1536
		1537	It also creates a condition variable that is used to signal the completion
		1538	of the request:
		1539
		1540	$txn->{finished} = AnyAvent->condvar;
		1541
		1542	It then creates a socket in non-blocking mode.
		1543
		1544	socket $txn->{fh}, ...;
		1545	fcntl $txn->{fh}, F_SETFL, O_NONBLOCK;
		1546	connect $txn->{fh}, ...
		1547	and !$!{EWOULDBLOCK}
		1548	and !$!{EINPROGRESS}
		1549	and Carp::croak "unable to connect: $!\n";
		1550
		1551	Then it creates a write-watcher which gets called whenever an error occurs
		1552	or the connection succeeds:
		1553
		1554	$txn->{w} = AnyEvent->io (fh => $txn->{fh}, poll => 'w', cb => sub { $txn->fh_ready_w });
		1555
		1556	And returns this transaction object. The C<fh_ready_w> callback gets
		1557	called as soon as the event loop detects that the socket is ready for
		1558	writing.
		1559
		1560	The C<fh_ready_w> method makes the socket blocking again, writes the
		1561	request data and replaces the watcher by a read watcher (waiting for reply
		1562	data). The actual code is more complicated, but that doesn't matter for
		1563	this example:
		1564
		1565	fcntl $txn->{fh}, F_SETFL, 0;
		1566	syswrite $txn->{fh}, $txn->{request}
		1567	or die "connection or write error";
		1568	$txn->{w} = AnyEvent->io (fh => $txn->{fh}, poll => 'r', cb => sub { $txn->fh_ready_r });
		1569
		1570	Again, C<fh_ready_r> waits till all data has arrived, and then stores the
		1571	result and signals any possible waiters that the request has finished:
		1572
		1573	sysread $txn->{fh}, $txn->{buf}, length $txn->{$buf};
		1574
		1575	if (end-of-file or data complete) {
		1576	$txn->{result} = $txn->{buf};
		1577	$txn->{finished}->send;
		1578	$txb->{cb}->($txn) of $txn->{cb}; # also call callback
		1579	}
		1580
		1581	The C<result> method, finally, just waits for the finished signal (if the
		1582	request was already finished, it doesn't wait, of course, and returns the
		1583	data:
		1584
		1585	$txn->{finished}->recv;
		1586	return $txn->{result};
		1587
		1588	The actual code goes further and collects all errors (C<die>s, exceptions)
		1589	that occurred during request processing. The C<result> method detects
		1590	whether an exception as thrown (it is stored inside the $txn object)
		1591	and just throws the exception, which means connection errors and other
		1592	problems get reported tot he code that tries to use the result, not in a
		1593	random callback.
		1594
		1595	All of this enables the following usage styles:
		1596
		1597	1. Blocking:
		1598
		1599	my $data = $fcp->client_get ($url);
		1600
		1601	2. Blocking, but running in parallel:
		1602
		1603	my @datas = map $_->result,
		1604	map $fcp->txn_client_get ($_),
		1605	@urls;
		1606
		1607	Both blocking examples work without the module user having to know
		1608	anything about events.
		1609
		1610	3a. Event-based in a main program, using any supported event module:
		1611
		1612	use EV;
		1613
		1614	$fcp->txn_client_get ($url)->cb (sub {
		1615	my $txn = shift;
		1616	my $data = $txn->result;
		1617	...
		1618	});
		1619
		1620	EV::loop;
		1621
		1622	3b. The module user could use AnyEvent, too:
		1623
		1624	use AnyEvent;
		1625
		1626	my $quit = AnyEvent->condvar;
		1627
		1628	$fcp->txn_client_get ($url)->cb (sub {
		1629	...
		1630	$quit->send;
		1631	});
		1632
		1633	$quit->recv;
		1634
		1635
		1636	=head1 BENCHMARKS
		1637
		1638	To give you an idea of the performance and overheads that AnyEvent adds
		1639	over the event loops themselves and to give you an impression of the speed
		1640	of various event loops I prepared some benchmarks.
		1641
		1642	=head2 BENCHMARKING ANYEVENT OVERHEAD
		1643
		1644	Here is a benchmark of various supported event models used natively and
		1645	through AnyEvent. The benchmark creates a lot of timers (with a zero
		1646	timeout) and I/O watchers (watching STDOUT, a pty, to become writable,
		1647	which it is), lets them fire exactly once and destroys them again.
		1648
		1649	Source code for this benchmark is found as F<eg/bench> in the AnyEvent
		1650	distribution.
		1651
		1652	=head3 Explanation of the columns
		1653
		1654	I<watcher> is the number of event watchers created/destroyed. Since
		1655	different event models feature vastly different performances, each event
		1656	loop was given a number of watchers so that overall runtime is acceptable
		1657	and similar between tested event loop (and keep them from crashing): Glib
		1658	would probably take thousands of years if asked to process the same number
		1659	of watchers as EV in this benchmark.
		1660
		1661	I<bytes> is the number of bytes (as measured by the resident set size,
		1662	RSS) consumed by each watcher. This method of measuring captures both C
		1663	and Perl-based overheads.
		1664
		1665	I<create> is the time, in microseconds (millionths of seconds), that it
		1666	takes to create a single watcher. The callback is a closure shared between
		1667	all watchers, to avoid adding memory overhead. That means closure creation
		1668	and memory usage is not included in the figures.
		1669
		1670	I<invoke> is the time, in microseconds, used to invoke a simple
		1671	callback. The callback simply counts down a Perl variable and after it was
		1672	invoked "watcher" times, it would C<< ->send >> a condvar once to
		1673	signal the end of this phase.
		1674
		1675	I<destroy> is the time, in microseconds, that it takes to destroy a single
		1676	watcher.
		1677
		1678	=head3 Results
		1679
		1680	name watchers bytes create invoke destroy comment
		1681	EV/EV 400000 224 0.47 0.35 0.27 EV native interface
		1682	EV/Any 100000 224 2.88 0.34 0.27 EV + AnyEvent watchers
		1683	CoroEV/Any 100000 224 2.85 0.35 0.28 coroutines + Coro::Signal
		1684	Perl/Any 100000 452 4.13 0.73 0.95 pure perl implementation
		1685	Event/Event 16000 517 32.20 31.80 0.81 Event native interface
		1686	Event/Any 16000 590 35.85 31.55 1.06 Event + AnyEvent watchers
		1687	Glib/Any 16000 1357 102.33 12.31 51.00 quadratic behaviour
		1688	Tk/Any 2000 1860 27.20 66.31 14.00 SEGV with >> 2000 watchers
		1689	POE/Event 2000 6328 109.99 751.67 14.02 via POE::Loop::Event
		1690	POE/Select 2000 6027 94.54 809.13 579.80 via POE::Loop::Select
		1691
		1692	=head3 Discussion
		1693
		1694	The benchmark does I<not> measure scalability of the event loop very
		1695	well. For example, a select-based event loop (such as the pure perl one)
		1696	can never compete with an event loop that uses epoll when the number of
		1697	file descriptors grows high. In this benchmark, all events become ready at
		1698	the same time, so select/poll-based implementations get an unnatural speed
		1699	boost.
		1700
		1701	Also, note that the number of watchers usually has a nonlinear effect on
		1702	overall speed, that is, creating twice as many watchers doesn't take twice
		1703	the time - usually it takes longer. This puts event loops tested with a
		1704	higher number of watchers at a disadvantage.
		1705
		1706	To put the range of results into perspective, consider that on the
		1707	benchmark machine, handling an event takes roughly 1600 CPU cycles with
		1708	EV, 3100 CPU cycles with AnyEvent's pure perl loop and almost 3000000 CPU
		1709	cycles with POE.
		1710
		1711	C<EV> is the sole leader regarding speed and memory use, which are both
		1712	maximal/minimal, respectively. Even when going through AnyEvent, it uses
		1713	far less memory than any other event loop and is still faster than Event
		1714	natively.
		1715
		1716	The pure perl implementation is hit in a few sweet spots (both the
		1717	constant timeout and the use of a single fd hit optimisations in the perl
		1718	interpreter and the backend itself). Nevertheless this shows that it
		1719	adds very little overhead in itself. Like any select-based backend its
		1720	performance becomes really bad with lots of file descriptors (and few of
		1721	them active), of course, but this was not subject of this benchmark.
		1722
		1723	The C<Event> module has a relatively high setup and callback invocation
		1724	cost, but overall scores in on the third place.
		1725
		1726	C<Glib>'s memory usage is quite a bit higher, but it features a
		1727	faster callback invocation and overall ends up in the same class as
		1728	C<Event>. However, Glib scales extremely badly, doubling the number of
		1729	watchers increases the processing time by more than a factor of four,
		1730	making it completely unusable when using larger numbers of watchers
		1731	(note that only a single file descriptor was used in the benchmark, so
		1732	inefficiencies of C<poll> do not account for this).
		1733
		1734	The C<Tk> adaptor works relatively well. The fact that it crashes with
		1735	more than 2000 watchers is a big setback, however, as correctness takes
		1736	precedence over speed. Nevertheless, its performance is surprising, as the
		1737	file descriptor is dup()ed for each watcher. This shows that the dup()
		1738	employed by some adaptors is not a big performance issue (it does incur a
		1739	hidden memory cost inside the kernel which is not reflected in the figures
		1740	above).
		1741
		1742	C<POE>, regardless of underlying event loop (whether using its pure perl
		1743	select-based backend or the Event module, the POE-EV backend couldn't
		1744	be tested because it wasn't working) shows abysmal performance and
		1745	memory usage with AnyEvent: Watchers use almost 30 times as much memory
		1746	as EV watchers, and 10 times as much memory as Event (the high memory
		1747	requirements are caused by requiring a session for each watcher). Watcher
		1748	invocation speed is almost 900 times slower than with AnyEvent's pure perl
		1749	implementation.
		1750
		1751	The design of the POE adaptor class in AnyEvent can not really account
		1752	for the performance issues, though, as session creation overhead is
		1753	small compared to execution of the state machine, which is coded pretty
		1754	optimally within L<AnyEvent::Impl::POE> (and while everybody agrees that
		1755	using multiple sessions is not a good approach, especially regarding
		1756	memory usage, even the author of POE could not come up with a faster
		1757	design).
		1758
		1759	=head3 Summary
		1760
		1761	=over 4
		1762
		1763	=item * Using EV through AnyEvent is faster than any other event loop
		1764	(even when used without AnyEvent), but most event loops have acceptable
		1765	performance with or without AnyEvent.
		1766
		1767	=item * The overhead AnyEvent adds is usually much smaller than the overhead of
		1768	the actual event loop, only with extremely fast event loops such as EV
		1769	adds AnyEvent significant overhead.
		1770
		1771	=item * You should avoid POE like the plague if you want performance or
		1772	reasonable memory usage.
		1773
		1774	=back
		1775
		1776	=head2 BENCHMARKING THE LARGE SERVER CASE
		1777
		1778	This benchmark actually benchmarks the event loop itself. It works by
		1779	creating a number of "servers": each server consists of a socket pair, a
		1780	timeout watcher that gets reset on activity (but never fires), and an I/O
		1781	watcher waiting for input on one side of the socket. Each time the socket
		1782	watcher reads a byte it will write that byte to a random other "server".
		1783
		1784	The effect is that there will be a lot of I/O watchers, only part of which
		1785	are active at any one point (so there is a constant number of active
		1786	fds for each loop iteration, but which fds these are is random). The
		1787	timeout is reset each time something is read because that reflects how
		1788	most timeouts work (and puts extra pressure on the event loops).
		1789
		1790	In this benchmark, we use 10000 socket pairs (20000 sockets), of which 100
		1791	(1%) are active. This mirrors the activity of large servers with many
		1792	connections, most of which are idle at any one point in time.
		1793
		1794	Source code for this benchmark is found as F<eg/bench2> in the AnyEvent
		1795	distribution.
		1796
		1797	=head3 Explanation of the columns
		1798
		1799	I<sockets> is the number of sockets, and twice the number of "servers" (as
		1800	each server has a read and write socket end).
		1801
		1802	I<create> is the time it takes to create a socket pair (which is
		1803	nontrivial) and two watchers: an I/O watcher and a timeout watcher.
		1804
		1805	I<request>, the most important value, is the time it takes to handle a
		1806	single "request", that is, reading the token from the pipe and forwarding
		1807	it to another server. This includes deleting the old timeout and creating
		1808	a new one that moves the timeout into the future.
		1809
		1810	=head3 Results
		1811
		1812	name sockets create request
		1813	EV 20000 69.01 11.16
		1814	Perl 20000 73.32 35.87
		1815	Event 20000 212.62 257.32
		1816	Glib 20000 651.16 1896.30
		1817	POE 20000 349.67 12317.24 uses POE::Loop::Event
		1818
		1819	=head3 Discussion
		1820
		1821	This benchmark I<does> measure scalability and overall performance of the
		1822	particular event loop.
		1823
		1824	EV is again fastest. Since it is using epoll on my system, the setup time
		1825	is relatively high, though.
		1826
		1827	Perl surprisingly comes second. It is much faster than the C-based event
		1828	loops Event and Glib.
		1829
		1830	Event suffers from high setup time as well (look at its code and you will
		1831	understand why). Callback invocation also has a high overhead compared to
		1832	the C<< $_->() for .. >>-style loop that the Perl event loop uses. Event
		1833	uses select or poll in basically all documented configurations.
		1834
		1835	Glib is hit hard by its quadratic behaviour w.r.t. many watchers. It
		1836	clearly fails to perform with many filehandles or in busy servers.
		1837
		1838	POE is still completely out of the picture, taking over 1000 times as long
		1839	as EV, and over 100 times as long as the Perl implementation, even though
		1840	it uses a C-based event loop in this case.
		1841
		1842	=head3 Summary
		1843
		1844	=over 4
		1845
		1846	=item * The pure perl implementation performs extremely well.
		1847
		1848	=item * Avoid Glib or POE in large projects where performance matters.
		1849
		1850	=back
		1851
		1852	=head2 BENCHMARKING SMALL SERVERS
		1853
		1854	While event loops should scale (and select-based ones do not...) even to
		1855	large servers, most programs we (or I :) actually write have only a few
		1856	I/O watchers.
		1857
		1858	In this benchmark, I use the same benchmark program as in the large server
		1859	case, but it uses only eight "servers", of which three are active at any
		1860	one time. This should reflect performance for a small server relatively
		1861	well.
		1862
		1863	The columns are identical to the previous table.
		1864
		1865	=head3 Results
		1866
		1867	name sockets create request
		1868	EV 16 20.00 6.54
		1869	Perl 16 25.75 12.62
		1870	Event 16 81.27 35.86
		1871	Glib 16 32.63 15.48
		1872	POE 16 261.87 276.28 uses POE::Loop::Event
		1873
		1874	=head3 Discussion
		1875
		1876	The benchmark tries to test the performance of a typical small
		1877	server. While knowing how various event loops perform is interesting, keep
		1878	in mind that their overhead in this case is usually not as important, due
		1879	to the small absolute number of watchers (that is, you need efficiency and
		1880	speed most when you have lots of watchers, not when you only have a few of
		1881	them).
		1882
		1883	EV is again fastest.
		1884
		1885	Perl again comes second. It is noticeably faster than the C-based event
		1886	loops Event and Glib, although the difference is too small to really
		1887	matter.
		1888
		1889	POE also performs much better in this case, but is is still far behind the
		1890	others.
		1891
		1892	=head3 Summary
		1893
		1894	=over 4
		1895
		1896	=item * C-based event loops perform very well with small number of
		1897	watchers, as the management overhead dominates.
		1898
		1899	=back
		1900
		1901
		1902	=head1 SIGNALS
		1903
		1904	AnyEvent currently installs handlers for these signals:
		1905
		1906	=over 4
		1907
		1908	=item SIGCHLD
		1909
		1910	A handler for C<SIGCHLD> is installed by AnyEvent's child watcher
		1911	emulation for event loops that do not support them natively. Also, some
		1912	event loops install a similar handler.
		1913
		1914	=item SIGPIPE
		1915
		1916	A no-op handler is installed for C<SIGPIPE> when C<$SIG{PIPE}> is C<undef>
		1917	when AnyEvent gets loaded.
		1918
		1919	The rationale for this is that AnyEvent users usually do not really depend
		1920	on SIGPIPE delivery (which is purely an optimisation for shell use, or
		1921	badly-written programs), but C<SIGPIPE> can cause spurious and rare
		1922	program exits as a lot of people do not expect C<SIGPIPE> when writing to
		1923	some random socket.
		1924
		1925	The rationale for installing a no-op handler as opposed to ignoring it is
		1926	that this way, the handler will be restored to defaults on exec.
		1927
		1928	Feel free to install your own handler, or reset it to defaults.
		1929
		1930	=back
		1931
		1932	=cut
		1933
		1934	$SIG{PIPE} = sub { }
		1935	unless defined $SIG{PIPE};
		1936
		1937
		1938	=head1 FORK
		1939
		1940	Most event libraries are not fork-safe. The ones who are usually are
		1941	because they rely on inefficient but fork-safe C<select> or C<poll>
		1942	calls. Only L<EV> is fully fork-aware.
		1943
		1944	If you have to fork, you must either do so I<before> creating your first
		1945	watcher OR you must not use AnyEvent at all in the child.
		1946
		1947
		1948	=head1 SECURITY CONSIDERATIONS
		1949
		1950	AnyEvent can be forced to load any event model via
		1951	$ENV{PERL_ANYEVENT_MODEL}. While this cannot (to my knowledge) be used to
		1952	execute arbitrary code or directly gain access, it can easily be used to
		1953	make the program hang or malfunction in subtle ways, as AnyEvent watchers
		1954	will not be active when the program uses a different event model than
		1955	specified in the variable.
		1956
		1957	You can make AnyEvent completely ignore this variable by deleting it
		1958	before the first watcher gets created, e.g. with a C<BEGIN> block:
		1959
		1960	BEGIN { delete $ENV{PERL_ANYEVENT_MODEL} }
		1961
		1962	use AnyEvent;
		1963
		1964	Similar considerations apply to $ENV{PERL_ANYEVENT_VERBOSE}, as that can
		1965	be used to probe what backend is used and gain other information (which is
		1966	probably even less useful to an attacker than PERL_ANYEVENT_MODEL), and
		1967	$ENV{PERL_ANYEGENT_STRICT}.
		1968
		1969
		1970	=head1 BUGS
		1971
		1972	Perl 5.8 has numerous memleaks that sometimes hit this module and are hard
		1973	to work around. If you suffer from memleaks, first upgrade to Perl 5.10
		1974	and check wether the leaks still show up. (Perl 5.10.0 has other annoying
		1975	memleaks, such as leaking on C<map> and C<grep> but it is usually not as
		1976	pronounced).
		1977
		1978
		1979	=head1 SEE ALSO
		1980
		1981	Utility functions: L<AnyEvent::Util>.
		1982
		1983	Event modules: L<EV>, L<EV::Glib>, L<Glib::EV>, L<Event>, L<Glib::Event>,
		1984	L<Glib>, L<Tk>, L<Event::Lib>, L<Qt>, L<POE>.
		1985
		1986	Implementations: L<AnyEvent::Impl::EV>, L<AnyEvent::Impl::Event>,
		1987	L<AnyEvent::Impl::Glib>, L<AnyEvent::Impl::Tk>, L<AnyEvent::Impl::Perl>,
		1988	L<AnyEvent::Impl::EventLib>, L<AnyEvent::Impl::Qt>,
		1989	L<AnyEvent::Impl::POE>.
		1990
		1991	Non-blocking file handles, sockets, TCP clients and
		1992	servers: L<AnyEvent::Handle>, L<AnyEvent::Socket>.
		1993
		1994	Asynchronous DNS: L<AnyEvent::DNS>.
		1995
		1996	Coroutine support: L<Coro>, L<Coro::AnyEvent>, L<Coro::EV>, L<Coro::Event>,
		1997
		1998	Nontrivial usage examples: L<Net::FCP>, L<Net::XMPP2>, L<AnyEvent::DNS>.
		1999
		2000
		2001	=head1 AUTHOR
		2002
		2003	Marc Lehmann <schmorp@schmorp.de>
		2004	http://home.schmorp.de/
		2005
		2006	=cut
		2007
		2008	1
		2009

Diff Legend

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