ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/AnyEvent/lib/AnyEvent.pm

Comparing AnyEvent/lib/AnyEvent.pm (file contents):
Revision 1.3 by root, Thu Dec 1 21:19:58 2005 UTC vs.
Revision 1.349 by root, Mon Jul 4 21:14:42 2011 UTC

1	=head1 NAME	1	=head1 NAME
2		2
3	AnyEvent - provide framework for multiple event loops	3	AnyEvent - the DBI of event loop programming
4		4
5	Event, Coro, Glib, Tk - various supported event loops	5	EV, Event, Glib, Tk, Perl, Event::Lib, Irssi, rxvt-unicode, IO::Async, Qt
		6	and POE are various supported event loops/environments.
6		7
7	=head1 SYNOPSIS	8	=head1 SYNOPSIS
8		9
9	use AnyEvent;	10	use AnyEvent;
10		11
		12	# if you prefer function calls, look at the AE manpage for
		13	# an alternative API.
		14
		15	# file handle or descriptor readable
11	my $w = AnyEvent->timer (fh => ..., poll => "[rw]+", cb => sub {	16	my $w = AnyEvent->io (fh => $fh, poll => "r", cb => sub { ... });
12	my ($poll_got) = @_;	17
		18	# one-shot or repeating timers
		19	my $w = AnyEvent->timer (after => $seconds, cb => sub { ... });
		20	my $w = AnyEvent->timer (after => $seconds, interval => $seconds, cb => ...);
		21
		22	print AnyEvent->now; # prints current event loop time
		23	print AnyEvent->time; # think Time::HiRes::time or simply CORE::time.
		24
		25	# POSIX signal
		26	my $w = AnyEvent->signal (signal => "TERM", cb => sub { ... });
		27
		28	# child process exit
		29	my $w = AnyEvent->child (pid => $pid, cb => sub {
		30	my ($pid, $status) = @_;
13	...	31	...
14	});	32	});
		33
		34	# called when event loop idle (if applicable)
		35	my $w = AnyEvent->idle (cb => sub { ... });
		36
		37	my $w = AnyEvent->condvar; # stores whether a condition was flagged
		38	$w->send; # wake up current and all future recv's
		39	$w->recv; # enters "main loop" till $condvar gets ->send
		40	# use a condvar in callback mode:
		41	$w->cb (sub { $_[0]->recv });
		42
		43	=head1 INTRODUCTION/TUTORIAL
		44
		45	This manpage is mainly a reference manual. If you are interested
		46	in a tutorial or some gentle introduction, have a look at the
		47	L<AnyEvent::Intro> manpage.
		48
		49	=head1 SUPPORT
		50
		51	An FAQ document is available as L<AnyEvent::FAQ>.
		52
		53	There also is a mailinglist for discussing all things AnyEvent, and an IRC
		54	channel, too.
		55
		56	See the AnyEvent project page at the B<Schmorpforge Ta-Sa Software
		57	Repository>, at L<http://anyevent.schmorp.de>, for more info.
		58
		59	=head1 WHY YOU SHOULD USE THIS MODULE (OR NOT)
		60
		61	Glib, POE, IO::Async, Event... CPAN offers event models by the dozen
		62	nowadays. So what is different about AnyEvent?
		63
		64	Executive Summary: AnyEvent is I<compatible>, AnyEvent is I<free of
		65	policy> and AnyEvent is I<small and efficient>.
		66
		67	First and foremost, I<AnyEvent is not an event model> itself, it only
		68	interfaces to whatever event model the main program happens to use, in a
		69	pragmatic way. For event models and certain classes of immortals alike,
		70	the statement "there can only be one" is a bitter reality: In general,
		71	only one event loop can be active at the same time in a process. AnyEvent
		72	cannot change this, but it can hide the differences between those event
		73	loops.
		74
		75	The goal of AnyEvent is to offer module authors the ability to do event
		76	programming (waiting for I/O or timer events) without subscribing to a
		77	religion, a way of living, and most importantly: without forcing your
		78	module users into the same thing by forcing them to use the same event
		79	model you use.
		80
		81	For modules like POE or IO::Async (which is a total misnomer as it is
		82	actually doing all I/O I<synchronously>...), using them in your module is
		83	like joining a cult: After you join, you are dependent on them and you
		84	cannot use anything else, as they are simply incompatible to everything
		85	that isn't them. What's worse, all the potential users of your
		86	module are I<also> forced to use the same event loop you use.
		87
		88	AnyEvent is different: AnyEvent + POE works fine. AnyEvent + Glib works
		89	fine. AnyEvent + Tk works fine etc. etc. but none of these work together
		90	with the rest: POE + EV? No go. Tk + Event? No go. Again: if your module
		91	uses one of those, every user of your module has to use it, too. But if
		92	your module uses AnyEvent, it works transparently with all event models it
		93	supports (including stuff like IO::Async, as long as those use one of the
		94	supported event loops. It is easy to add new event loops to AnyEvent, too,
		95	so it is future-proof).
		96
		97	In addition to being free of having to use I<the one and only true event
		98	model>, AnyEvent also is free of bloat and policy: with POE or similar
		99	modules, you get an enormous amount of code and strict rules you have to
		100	follow. AnyEvent, on the other hand, is lean and to the point, by only
		101	offering the functionality that is necessary, in as thin as a wrapper as
		102	technically possible.
		103
		104	Of course, AnyEvent comes with a big (and fully optional!) toolbox
		105	of useful functionality, such as an asynchronous DNS resolver, 100%
		106	non-blocking connects (even with TLS/SSL, IPv6 and on broken platforms
		107	such as Windows) and lots of real-world knowledge and workarounds for
		108	platform bugs and differences.
		109
		110	Now, if you I<do want> lots of policy (this can arguably be somewhat
		111	useful) and you want to force your users to use the one and only event
		112	model, you should I<not> use this module.
		113
		114	=head1 DESCRIPTION
		115
		116	L<AnyEvent> provides a uniform interface to various event loops. This
		117	allows module authors to use event loop functionality without forcing
		118	module users to use a specific event loop implementation (since more
		119	than one event loop cannot coexist peacefully).
		120
		121	The interface itself is vaguely similar, but not identical to the L<Event>
		122	module.
		123
		124	During the first call of any watcher-creation method, the module tries
		125	to detect the currently loaded event loop by probing whether one of the
		126	following modules is already loaded: L<EV>, L<AnyEvent::Impl::Perl>,
		127	L<Event>, L<Glib>, L<Tk>, L<Event::Lib>, L<Qt>, L<POE>. The first one
		128	found is used. If none are detected, the module tries to load the first
		129	four modules in the order given; but note that if L<EV> is not
		130	available, the pure-perl L<AnyEvent::Impl::Perl> should always work, so
		131	the other two are not normally tried.
		132
		133	Because AnyEvent first checks for modules that are already loaded, loading
		134	an event model explicitly before first using AnyEvent will likely make
		135	that model the default. For example:
		136
		137	use Tk;
		138	use AnyEvent;
		139
		140	# .. AnyEvent will likely default to Tk
		141
		142	The I<likely> means that, if any module loads another event model and
		143	starts using it, all bets are off - this case should be very rare though,
		144	as very few modules hardcode event loops without announcing this very
		145	loudly.
		146
		147	The pure-perl implementation of AnyEvent is called
		148	C<AnyEvent::Impl::Perl>. Like other event modules you can load it
		149	explicitly and enjoy the high availability of that event loop :)
		150
		151	=head1 WATCHERS
		152
		153	AnyEvent has the central concept of a I<watcher>, which is an object that
		154	stores relevant data for each kind of event you are waiting for, such as
		155	the callback to call, the file handle to watch, etc.
		156
		157	These watchers are normal Perl objects with normal Perl lifetime. After
		158	creating a watcher it will immediately "watch" for events and invoke the
		159	callback when the event occurs (of course, only when the event model
		160	is in control).
		161
		162	Note that B<callbacks must not permanently change global variables>
		163	potentially in use by the event loop (such as C<$_> or C<$[>) and that B<<
		164	callbacks must not C<die> >>. The former is good programming practice in
		165	Perl and the latter stems from the fact that exception handling differs
		166	widely between event loops.
		167
		168	To disable a watcher you have to destroy it (e.g. by setting the
		169	variable you store it in to C<undef> or otherwise deleting all references
		170	to it).
		171
		172	All watchers are created by calling a method on the C<AnyEvent> class.
		173
		174	Many watchers either are used with "recursion" (repeating timers for
		175	example), or need to refer to their watcher object in other ways.
		176
		177	One way to achieve that is this pattern:
		178
		179	my $w; $w = AnyEvent->type (arg => value ..., cb => sub {
		180	# you can use $w here, for example to undef it
		181	undef $w;
		182	});
		183
		184	Note that C<my $w; $w => combination. This is necessary because in Perl,
		185	my variables are only visible after the statement in which they are
		186	declared.
		187
		188	=head2 I/O WATCHERS
		189
		190	$w = AnyEvent->io (
		191	fh => <filehandle_or_fileno>,
		192	poll => <"r" or "w">,
		193	cb => <callback>,
		194	);
		195
		196	You can create an I/O watcher by calling the C<< AnyEvent->io >> method
		197	with the following mandatory key-value pairs as arguments:
		198
		199	C<fh> is the Perl I<file handle> (or a naked file descriptor) to watch
		200	for events (AnyEvent might or might not keep a reference to this file
		201	handle). Note that only file handles pointing to things for which
		202	non-blocking operation makes sense are allowed. This includes sockets,
		203	most character devices, pipes, fifos and so on, but not for example files
		204	or block devices.
		205
		206	C<poll> must be a string that is either C<r> or C<w>, which creates a
		207	watcher waiting for "r"eadable or "w"ritable events, respectively.
		208
		209	C<cb> is the callback to invoke each time the file handle becomes ready.
		210
		211	Although the callback might get passed parameters, their value and
		212	presence is undefined and you cannot rely on them. Portable AnyEvent
		213	callbacks cannot use arguments passed to I/O watcher callbacks.
		214
		215	The I/O watcher might use the underlying file descriptor or a copy of it.
		216	You must not close a file handle as long as any watcher is active on the
		217	underlying file descriptor.
		218
		219	Some event loops issue spurious readiness notifications, so you should
		220	always use non-blocking calls when reading/writing from/to your file
		221	handles.
		222
		223	Example: wait for readability of STDIN, then read a line and disable the
		224	watcher.
		225
		226	my $w; $w = AnyEvent->io (fh => \*STDIN, poll => 'r', cb => sub {
		227	chomp (my $input = <STDIN>);
		228	warn "read: $input\n";
		229	undef $w;
		230	});
		231
		232	=head2 TIME WATCHERS
		233
		234	$w = AnyEvent->timer (after => <seconds>, cb => <callback>);
		235
		236	$w = AnyEvent->timer (
		237	after => <fractional_seconds>,
		238	interval => <fractional_seconds>,
		239	cb => <callback>,
		240	);
		241
		242	You can create a time watcher by calling the C<< AnyEvent->timer >>
		243	method with the following mandatory arguments:
		244
		245	C<after> specifies after how many seconds (fractional values are
		246	supported) the callback should be invoked. C<cb> is the callback to invoke
		247	in that case.
		248
		249	Although the callback might get passed parameters, their value and
		250	presence is undefined and you cannot rely on them. Portable AnyEvent
		251	callbacks cannot use arguments passed to time watcher callbacks.
		252
		253	The callback will normally be invoked only once. If you specify another
		254	parameter, C<interval>, as a strictly positive number (> 0), then the
		255	callback will be invoked regularly at that interval (in fractional
		256	seconds) after the first invocation. If C<interval> is specified with a
		257	false value, then it is treated as if it were not specified at all.
		258
		259	The callback will be rescheduled before invoking the callback, but no
		260	attempt is made to avoid timer drift in most backends, so the interval is
		261	only approximate.
		262
		263	Example: fire an event after 7.7 seconds.
		264
15	my $w = AnyEvent->io (after => $seconds, cb => sub {	265	my $w = AnyEvent->timer (after => 7.7, cb => sub {
		266	warn "timeout\n";
		267	});
		268
		269	# to cancel the timer:
		270	undef $w;
		271
		272	Example 2: fire an event after 0.5 seconds, then roughly every second.
		273
		274	my $w = AnyEvent->timer (after => 0.5, interval => 1, cb => sub {
		275	warn "timeout\n";
		276	};
		277
		278	=head3 TIMING ISSUES
		279
		280	There are two ways to handle timers: based on real time (relative, "fire
		281	in 10 seconds") and based on wallclock time (absolute, "fire at 12
		282	o'clock").
		283
		284	While most event loops expect timers to specified in a relative way, they
		285	use absolute time internally. This makes a difference when your clock
		286	"jumps", for example, when ntp decides to set your clock backwards from
		287	the wrong date of 2014-01-01 to 2008-01-01, a watcher that is supposed to
		288	fire "after a second" might actually take six years to finally fire.
		289
		290	AnyEvent cannot compensate for this. The only event loop that is conscious
		291	of these issues is L<EV>, which offers both relative (ev_timer, based
		292	on true relative time) and absolute (ev_periodic, based on wallclock time)
		293	timers.
		294
		295	AnyEvent always prefers relative timers, if available, matching the
		296	AnyEvent API.
		297
		298	AnyEvent has two additional methods that return the "current time":
		299
		300	=over 4
		301
		302	=item AnyEvent->time
		303
		304	This returns the "current wallclock time" as a fractional number of
		305	seconds since the Epoch (the same thing as C<time> or C<Time::HiRes::time>
		306	return, and the result is guaranteed to be compatible with those).
		307
		308	It progresses independently of any event loop processing, i.e. each call
		309	will check the system clock, which usually gets updated frequently.
		310
		311	=item AnyEvent->now
		312
		313	This also returns the "current wallclock time", but unlike C<time>, above,
		314	this value might change only once per event loop iteration, depending on
		315	the event loop (most return the same time as C<time>, above). This is the
		316	time that AnyEvent's timers get scheduled against.
		317
		318	I<In almost all cases (in all cases if you don't care), this is the
		319	function to call when you want to know the current time.>
		320
		321	This function is also often faster then C<< AnyEvent->time >>, and
		322	thus the preferred method if you want some timestamp (for example,
		323	L<AnyEvent::Handle> uses this to update its activity timeouts).
		324
		325	The rest of this section is only of relevance if you try to be very exact
		326	with your timing; you can skip it without a bad conscience.
		327
		328	For a practical example of when these times differ, consider L<Event::Lib>
		329	and L<EV> and the following set-up:
		330
		331	The event loop is running and has just invoked one of your callbacks at
		332	time=500 (assume no other callbacks delay processing). In your callback,
		333	you wait a second by executing C<sleep 1> (blocking the process for a
		334	second) and then (at time=501) you create a relative timer that fires
		335	after three seconds.
		336
		337	With L<Event::Lib>, C<< AnyEvent->time >> and C<< AnyEvent->now >> will
		338	both return C<501>, because that is the current time, and the timer will
		339	be scheduled to fire at time=504 (C<501> + C<3>).
		340
		341	With L<EV>, C<< AnyEvent->time >> returns C<501> (as that is the current
		342	time), but C<< AnyEvent->now >> returns C<500>, as that is the time the
		343	last event processing phase started. With L<EV>, your timer gets scheduled
		344	to run at time=503 (C<500> + C<3>).
		345
		346	In one sense, L<Event::Lib> is more exact, as it uses the current time
		347	regardless of any delays introduced by event processing. However, most
		348	callbacks do not expect large delays in processing, so this causes a
		349	higher drift (and a lot more system calls to get the current time).
		350
		351	In another sense, L<EV> is more exact, as your timer will be scheduled at
		352	the same time, regardless of how long event processing actually took.
		353
		354	In either case, if you care (and in most cases, you don't), then you
		355	can get whatever behaviour you want with any event loop, by taking the
		356	difference between C<< AnyEvent->time >> and C<< AnyEvent->now >> into
		357	account.
		358
		359	=item AnyEvent->now_update
		360
		361	Some event loops (such as L<EV> or L<AnyEvent::Impl::Perl>) cache
		362	the current time for each loop iteration (see the discussion of L<<
		363	AnyEvent->now >>, above).
		364
		365	When a callback runs for a long time (or when the process sleeps), then
		366	this "current" time will differ substantially from the real time, which
		367	might affect timers and time-outs.
		368
		369	When this is the case, you can call this method, which will update the
		370	event loop's idea of "current time".
		371
		372	A typical example would be a script in a web server (e.g. C<mod_perl>) -
		373	when mod_perl executes the script, then the event loop will have the wrong
		374	idea about the "current time" (being potentially far in the past, when the
		375	script ran the last time). In that case you should arrange a call to C<<
		376	AnyEvent->now_update >> each time the web server process wakes up again
		377	(e.g. at the start of your script, or in a handler).
		378
		379	Note that updating the time I<might> cause some events to be handled.
		380
		381	=back
		382
		383	=head2 SIGNAL WATCHERS
		384
		385	$w = AnyEvent->signal (signal => <uppercase_signal_name>, cb => <callback>);
		386
		387	You can watch for signals using a signal watcher, C<signal> is the signal
		388	I<name> in uppercase and without any C<SIG> prefix, C<cb> is the Perl
		389	callback to be invoked whenever a signal occurs.
		390
		391	Although the callback might get passed parameters, their value and
		392	presence is undefined and you cannot rely on them. Portable AnyEvent
		393	callbacks cannot use arguments passed to signal watcher callbacks.
		394
		395	Multiple signal occurrences can be clumped together into one callback
		396	invocation, and callback invocation will be synchronous. Synchronous means
		397	that it might take a while until the signal gets handled by the process,
		398	but it is guaranteed not to interrupt any other callbacks.
		399
		400	The main advantage of using these watchers is that you can share a signal
		401	between multiple watchers, and AnyEvent will ensure that signals will not
		402	interrupt your program at bad times.
		403
		404	This watcher might use C<%SIG> (depending on the event loop used),
		405	so programs overwriting those signals directly will likely not work
		406	correctly.
		407
		408	Example: exit on SIGINT
		409
		410	my $w = AnyEvent->signal (signal => "INT", cb => sub { exit 1 });
		411
		412	=head3 Restart Behaviour
		413
		414	While restart behaviour is up to the event loop implementation, most will
		415	not restart syscalls (that includes L<Async::Interrupt> and AnyEvent's
		416	pure perl implementation).
		417
		418	=head3 Safe/Unsafe Signals
		419
		420	Perl signals can be either "safe" (synchronous to opcode handling) or
		421	"unsafe" (asynchronous) - the former might get delayed indefinitely, the
		422	latter might corrupt your memory.
		423
		424	AnyEvent signal handlers are, in addition, synchronous to the event loop,
		425	i.e. they will not interrupt your running perl program but will only be
		426	called as part of the normal event handling (just like timer, I/O etc.
		427	callbacks, too).
		428
		429	=head3 Signal Races, Delays and Workarounds
		430
		431	Many event loops (e.g. Glib, Tk, Qt, IO::Async) do not support attaching
		432	callbacks to signals in a generic way, which is a pity, as you cannot
		433	do race-free signal handling in perl, requiring C libraries for
		434	this. AnyEvent will try to do its best, which means in some cases,
		435	signals will be delayed. The maximum time a signal might be delayed is
		436	specified in C<$AnyEvent::MAX_SIGNAL_LATENCY> (default: 10 seconds). This
		437	variable can be changed only before the first signal watcher is created,
		438	and should be left alone otherwise. This variable determines how often
		439	AnyEvent polls for signals (in case a wake-up was missed). Higher values
		440	will cause fewer spurious wake-ups, which is better for power and CPU
		441	saving.
		442
		443	All these problems can be avoided by installing the optional
		444	L<Async::Interrupt> module, which works with most event loops. It will not
		445	work with inherently broken event loops such as L<Event> or L<Event::Lib>
		446	(and not with L<POE> currently, as POE does its own workaround with
		447	one-second latency). For those, you just have to suffer the delays.
		448
		449	=head2 CHILD PROCESS WATCHERS
		450
		451	$w = AnyEvent->child (pid => <process id>, cb => <callback>);
		452
		453	You can also watch for a child process exit and catch its exit status.
		454
		455	The child process is specified by the C<pid> argument (on some backends,
		456	using C<0> watches for any child process exit, on others this will
		457	croak). The watcher will be triggered only when the child process has
		458	finished and an exit status is available, not on any trace events
		459	(stopped/continued).
		460
		461	The callback will be called with the pid and exit status (as returned by
		462	waitpid), so unlike other watcher types, you I<can> rely on child watcher
		463	callback arguments.
		464
		465	This watcher type works by installing a signal handler for C<SIGCHLD>,
		466	and since it cannot be shared, nothing else should use SIGCHLD or reap
		467	random child processes (waiting for specific child processes, e.g. inside
		468	C<system>, is just fine).
		469
		470	There is a slight catch to child watchers, however: you usually start them
		471	I<after> the child process was created, and this means the process could
		472	have exited already (and no SIGCHLD will be sent anymore).
		473
		474	Not all event models handle this correctly (neither POE nor IO::Async do,
		475	see their AnyEvent::Impl manpages for details), but even for event models
		476	that I<do> handle this correctly, they usually need to be loaded before
		477	the process exits (i.e. before you fork in the first place). AnyEvent's
		478	pure perl event loop handles all cases correctly regardless of when you
		479	start the watcher.
		480
		481	This means you cannot create a child watcher as the very first
		482	thing in an AnyEvent program, you I<have> to create at least one
		483	watcher before you C<fork> the child (alternatively, you can call
		484	C<AnyEvent::detect>).
		485
		486	As most event loops do not support waiting for child events, they will be
		487	emulated by AnyEvent in most cases, in which the latency and race problems
		488	mentioned in the description of signal watchers apply.
		489
		490	Example: fork a process and wait for it
		491
		492	my $done = AnyEvent->condvar;
		493
		494	my $pid = fork or exit 5;
		495
		496	my $w = AnyEvent->child (
		497	pid => $pid,
		498	cb => sub {
		499	my ($pid, $status) = @_;
		500	warn "pid $pid exited with status $status";
		501	$done->send;
		502	},
		503	);
		504
		505	# do something else, then wait for process exit
		506	$done->recv;
		507
		508	=head2 IDLE WATCHERS
		509
		510	$w = AnyEvent->idle (cb => <callback>);
		511
		512	This will repeatedly invoke the callback after the process becomes idle,
		513	until either the watcher is destroyed or new events have been detected.
		514
		515	Idle watchers are useful when there is a need to do something, but it
		516	is not so important (or wise) to do it instantly. The callback will be
		517	invoked only when there is "nothing better to do", which is usually
		518	defined as "all outstanding events have been handled and no new events
		519	have been detected". That means that idle watchers ideally get invoked
		520	when the event loop has just polled for new events but none have been
		521	detected. Instead of blocking to wait for more events, the idle watchers
		522	will be invoked.
		523
		524	Unfortunately, most event loops do not really support idle watchers (only
		525	EV, Event and Glib do it in a usable fashion) - for the rest, AnyEvent
		526	will simply call the callback "from time to time".
		527
		528	Example: read lines from STDIN, but only process them when the
		529	program is otherwise idle:
		530
		531	my @lines; # read data
		532	my $idle_w;
		533	my $io_w = AnyEvent->io (fh => \*STDIN, poll => 'r', cb => sub {
		534	push @lines, scalar <STDIN>;
		535
		536	# start an idle watcher, if not already done
		537	$idle_w ||= AnyEvent->idle (cb => sub {
		538	# handle only one line, when there are lines left
		539	if (my $line = shift @lines) {
		540	print "handled when idle: $line";
		541	} else {
		542	# otherwise disable the idle watcher again
		543	undef $idle_w;
		544	}
		545	});
		546	});
		547
		548	=head2 CONDITION VARIABLES
		549
		550	$cv = AnyEvent->condvar;
		551
		552	$cv->send (<list>);
		553	my @res = $cv->recv;
		554
		555	If you are familiar with some event loops you will know that all of them
		556	require you to run some blocking "loop", "run" or similar function that
		557	will actively watch for new events and call your callbacks.
		558
		559	AnyEvent is slightly different: it expects somebody else to run the event
		560	loop and will only block when necessary (usually when told by the user).
		561
		562	The tool to do that is called a "condition variable", so called because
		563	they represent a condition that must become true.
		564
		565	Now is probably a good time to look at the examples further below.
		566
		567	Condition variables can be created by calling the C<< AnyEvent->condvar
		568	>> method, usually without arguments. The only argument pair allowed is
		569	C<cb>, which specifies a callback to be called when the condition variable
		570	becomes true, with the condition variable as the first argument (but not
		571	the results).
		572
		573	After creation, the condition variable is "false" until it becomes "true"
		574	by calling the C<send> method (or calling the condition variable as if it
		575	were a callback, read about the caveats in the description for the C<<
		576	->send >> method).
		577
		578	Since condition variables are the most complex part of the AnyEvent API, here are
		579	some different mental models of what they are - pick the ones you can connect to:
		580
		581	=over 4
		582
		583	=item * Condition variables are like callbacks - you can call them (and pass them instead
		584	of callbacks). Unlike callbacks however, you can also wait for them to be called.
		585
		586	=item * Condition variables are signals - one side can emit or send them,
		587	the other side can wait for them, or install a handler that is called when
		588	the signal fires.
		589
		590	=item * Condition variables are like "Merge Points" - points in your program
		591	where you merge multiple independent results/control flows into one.
		592
		593	=item * Condition variables represent a transaction - functions that start
		594	some kind of transaction can return them, leaving the caller the choice
		595	between waiting in a blocking fashion, or setting a callback.
		596
		597	=item * Condition variables represent future values, or promises to deliver
		598	some result, long before the result is available.
		599
		600	=back
		601
		602	Condition variables are very useful to signal that something has finished,
		603	for example, if you write a module that does asynchronous http requests,
		604	then a condition variable would be the ideal candidate to signal the
		605	availability of results. The user can either act when the callback is
		606	called or can synchronously C<< ->recv >> for the results.
		607
		608	You can also use them to simulate traditional event loops - for example,
		609	you can block your main program until an event occurs - for example, you
		610	could C<< ->recv >> in your main program until the user clicks the Quit
		611	button of your app, which would C<< ->send >> the "quit" event.
		612
		613	Note that condition variables recurse into the event loop - if you have
		614	two pieces of code that call C<< ->recv >> in a round-robin fashion, you
		615	lose. Therefore, condition variables are good to export to your caller, but
		616	you should avoid making a blocking wait yourself, at least in callbacks,
		617	as this asks for trouble.
		618
		619	Condition variables are represented by hash refs in perl, and the keys
		620	used by AnyEvent itself are all named C<_ae_XXX> to make subclassing
		621	easy (it is often useful to build your own transaction class on top of
		622	AnyEvent). To subclass, use C<AnyEvent::CondVar> as base class and call
		623	its C<new> method in your own C<new> method.
		624
		625	There are two "sides" to a condition variable - the "producer side" which
		626	eventually calls C<< -> send >>, and the "consumer side", which waits
		627	for the send to occur.
		628
		629	Example: wait for a timer.
		630
		631	# condition: "wait till the timer is fired"
		632	my $timer_fired = AnyEvent->condvar;
		633
		634	# create the timer - we could wait for, say
		635	# a handle becomign ready, or even an
		636	# AnyEvent::HTTP request to finish, but
		637	# in this case, we simply use a timer:
		638	my $w = AnyEvent->timer (
		639	after => 1,
		640	cb => sub { $timer_fired->send },
		641	);
		642
		643	# this "blocks" (while handling events) till the callback
		644	# calls ->send
		645	$timer_fired->recv;
		646
		647	Example: wait for a timer, but take advantage of the fact that condition
		648	variables are also callable directly.
		649
		650	my $done = AnyEvent->condvar;
		651	my $delay = AnyEvent->timer (after => 5, cb => $done);
		652	$done->recv;
		653
		654	Example: Imagine an API that returns a condvar and doesn't support
		655	callbacks. This is how you make a synchronous call, for example from
		656	the main program:
		657
		658	use AnyEvent::CouchDB;
		659
		660	...
		661
		662	my @info = $couchdb->info->recv;
		663
		664	And this is how you would just set a callback to be called whenever the
		665	results are available:
		666
		667	$couchdb->info->cb (sub {
		668	my @info = $_[0]->recv;
		669	});
		670
		671	=head3 METHODS FOR PRODUCERS
		672
		673	These methods should only be used by the producing side, i.e. the
		674	code/module that eventually sends the signal. Note that it is also
		675	the producer side which creates the condvar in most cases, but it isn't
		676	uncommon for the consumer to create it as well.
		677
		678	=over 4
		679
		680	=item $cv->send (...)
		681
		682	Flag the condition as ready - a running C<< ->recv >> and all further
		683	calls to C<recv> will (eventually) return after this method has been
		684	called. If nobody is waiting the send will be remembered.
		685
		686	If a callback has been set on the condition variable, it is called
		687	immediately from within send.
		688
		689	Any arguments passed to the C<send> call will be returned by all
		690	future C<< ->recv >> calls.
		691
		692	Condition variables are overloaded so one can call them directly (as if
		693	they were a code reference). Calling them directly is the same as calling
		694	C<send>.
		695
		696	=item $cv->croak ($error)
		697
		698	Similar to send, but causes all calls to C<< ->recv >> to invoke
		699	C<Carp::croak> with the given error message/object/scalar.
		700
		701	This can be used to signal any errors to the condition variable
		702	user/consumer. Doing it this way instead of calling C<croak> directly
		703	delays the error detection, but has the overwhelming advantage that it
		704	diagnoses the error at the place where the result is expected, and not
		705	deep in some event callback with no connection to the actual code causing
		706	the problem.
		707
		708	=item $cv->begin ([group callback])
		709
		710	=item $cv->end
		711
		712	These two methods can be used to combine many transactions/events into
		713	one. For example, a function that pings many hosts in parallel might want
		714	to use a condition variable for the whole process.
		715
		716	Every call to C<< ->begin >> will increment a counter, and every call to
		717	C<< ->end >> will decrement it. If the counter reaches C<0> in C<< ->end
		718	>>, the (last) callback passed to C<begin> will be executed, passing the
		719	condvar as first argument. That callback is I<supposed> to call C<< ->send
		720	>>, but that is not required. If no group callback was set, C<send> will
		721	be called without any arguments.
		722
		723	You can think of C<< $cv->send >> giving you an OR condition (one call
		724	sends), while C<< $cv->begin >> and C<< $cv->end >> giving you an AND
		725	condition (all C<begin> calls must be C<end>'ed before the condvar sends).
		726
		727	Let's start with a simple example: you have two I/O watchers (for example,
		728	STDOUT and STDERR for a program), and you want to wait for both streams to
		729	close before activating a condvar:
		730
		731	my $cv = AnyEvent->condvar;
		732
		733	$cv->begin; # first watcher
		734	my $w1 = AnyEvent->io (fh => $fh1, cb => sub {
		735	defined sysread $fh1, my $buf, 4096
		736	or $cv->end;
		737	});
		738
		739	$cv->begin; # second watcher
		740	my $w2 = AnyEvent->io (fh => $fh2, cb => sub {
		741	defined sysread $fh2, my $buf, 4096
		742	or $cv->end;
		743	});
		744
		745	$cv->recv;
		746
		747	This works because for every event source (EOF on file handle), there is
		748	one call to C<begin>, so the condvar waits for all calls to C<end> before
		749	sending.
		750
		751	The ping example mentioned above is slightly more complicated, as the
		752	there are results to be passwd back, and the number of tasks that are
		753	begun can potentially be zero:
		754
		755	my $cv = AnyEvent->condvar;
		756
		757	my %result;
		758	$cv->begin (sub { shift->send (\%result) });
		759
		760	for my $host (@list_of_hosts) {
		761	$cv->begin;
		762	ping_host_then_call_callback $host, sub {
		763	$result{$host} = ...;
		764	$cv->end;
		765	};
		766	}
		767
		768	$cv->end;
		769
		770	This code fragment supposedly pings a number of hosts and calls
		771	C<send> after results for all then have have been gathered - in any
		772	order. To achieve this, the code issues a call to C<begin> when it starts
		773	each ping request and calls C<end> when it has received some result for
		774	it. Since C<begin> and C<end> only maintain a counter, the order in which
		775	results arrive is not relevant.
		776
		777	There is an additional bracketing call to C<begin> and C<end> outside the
		778	loop, which serves two important purposes: first, it sets the callback
		779	to be called once the counter reaches C<0>, and second, it ensures that
		780	C<send> is called even when C<no> hosts are being pinged (the loop
		781	doesn't execute once).
		782
		783	This is the general pattern when you "fan out" into multiple (but
		784	potentially zero) subrequests: use an outer C<begin>/C<end> pair to set
		785	the callback and ensure C<end> is called at least once, and then, for each
		786	subrequest you start, call C<begin> and for each subrequest you finish,
		787	call C<end>.
		788
		789	=back
		790
		791	=head3 METHODS FOR CONSUMERS
		792
		793	These methods should only be used by the consuming side, i.e. the
		794	code awaits the condition.
		795
		796	=over 4
		797
		798	=item $cv->recv
		799
		800	Wait (blocking if necessary) until the C<< ->send >> or C<< ->croak
		801	>> methods have been called on C<$cv>, while servicing other watchers
		802	normally.
		803
		804	You can only wait once on a condition - additional calls are valid but
		805	will return immediately.
		806
		807	If an error condition has been set by calling C<< ->croak >>, then this
		808	function will call C<croak>.
		809
		810	In list context, all parameters passed to C<send> will be returned,
		811	in scalar context only the first one will be returned.
		812
		813	Note that doing a blocking wait in a callback is not supported by any
		814	event loop, that is, recursive invocation of a blocking C<< ->recv
		815	>> is not allowed, and the C<recv> call will C<croak> if such a
		816	condition is detected. This condition can be slightly loosened by using
		817	L<Coro::AnyEvent>, which allows you to do a blocking C<< ->recv >> from
		818	any thread that doesn't run the event loop itself.
		819
		820	Not all event models support a blocking wait - some die in that case
		821	(programs might want to do that to stay interactive), so I<if you are
		822	using this from a module, never require a blocking wait>. Instead, let the
		823	caller decide whether the call will block or not (for example, by coupling
		824	condition variables with some kind of request results and supporting
		825	callbacks so the caller knows that getting the result will not block,
		826	while still supporting blocking waits if the caller so desires).
		827
		828	You can ensure that C<< ->recv >> never blocks by setting a callback and
		829	only calling C<< ->recv >> from within that callback (or at a later
		830	time). This will work even when the event loop does not support blocking
		831	waits otherwise.
		832
		833	=item $bool = $cv->ready
		834
		835	Returns true when the condition is "true", i.e. whether C<send> or
		836	C<croak> have been called.
		837
		838	=item $cb = $cv->cb ($cb->($cv))
		839
		840	This is a mutator function that returns the callback set and optionally
		841	replaces it before doing so.
		842
		843	The callback will be called when the condition becomes "true", i.e. when
		844	C<send> or C<croak> are called, with the only argument being the
		845	condition variable itself. If the condition is already true, the
		846	callback is called immediately when it is set. Calling C<recv> inside
		847	the callback or at any later time is guaranteed not to block.
		848
		849	=back
		850
		851	=head1 SUPPORTED EVENT LOOPS/BACKENDS
		852
		853	The available backend classes are (every class has its own manpage):
		854
		855	=over 4
		856
		857	=item Backends that are autoprobed when no other event loop can be found.
		858
		859	EV is the preferred backend when no other event loop seems to be in
		860	use. If EV is not installed, then AnyEvent will fall back to its own
		861	pure-perl implementation, which is available everywhere as it comes with
		862	AnyEvent itself.
		863
		864	AnyEvent::Impl::EV based on EV (interface to libev, best choice).
		865	AnyEvent::Impl::Perl pure-perl implementation, fast and portable.
		866
		867	=item Backends that are transparently being picked up when they are used.
		868
		869	These will be used if they are already loaded when the first watcher
		870	is created, in which case it is assumed that the application is using
		871	them. This means that AnyEvent will automatically pick the right backend
		872	when the main program loads an event module before anything starts to
		873	create watchers. Nothing special needs to be done by the main program.
		874
		875	AnyEvent::Impl::Event based on Event, very stable, few glitches.
		876	AnyEvent::Impl::Glib based on Glib, slow but very stable.
		877	AnyEvent::Impl::Tk based on Tk, very broken.
		878	AnyEvent::Impl::EventLib based on Event::Lib, leaks memory and worse.
		879	AnyEvent::Impl::POE based on POE, very slow, some limitations.
		880	AnyEvent::Impl::Irssi used when running within irssi.
		881	AnyEvent::Impl::IOAsync based on IO::Async.
		882	AnyEvent::Impl::Cocoa based on Cocoa::EventLoop.
		883	AnyEvent::Impl::FLTK based on FLTK.
		884
		885	=item Backends with special needs.
		886
		887	Qt requires the Qt::Application to be instantiated first, but will
		888	otherwise be picked up automatically. As long as the main program
		889	instantiates the application before any AnyEvent watchers are created,
		890	everything should just work.
		891
		892	AnyEvent::Impl::Qt based on Qt.
		893
		894	=item Event loops that are indirectly supported via other backends.
		895
		896	Some event loops can be supported via other modules:
		897
		898	There is no direct support for WxWidgets (L<Wx>) or L<Prima>.
		899
		900	B<WxWidgets> has no support for watching file handles. However, you can
		901	use WxWidgets through the POE adaptor, as POE has a Wx backend that simply
		902	polls 20 times per second, which was considered to be too horrible to even
		903	consider for AnyEvent.
		904
		905	B<Prima> is not supported as nobody seems to be using it, but it has a POE
		906	backend, so it can be supported through POE.
		907
		908	AnyEvent knows about both L<Prima> and L<Wx>, however, and will try to
		909	load L<POE> when detecting them, in the hope that POE will pick them up,
		910	in which case everything will be automatic.
		911
		912	=back
		913
		914	=head1 GLOBAL VARIABLES AND FUNCTIONS
		915
		916	These are not normally required to use AnyEvent, but can be useful to
		917	write AnyEvent extension modules.
		918
		919	=over 4
		920
		921	=item $AnyEvent::MODEL
		922
		923	Contains C<undef> until the first watcher is being created, before the
		924	backend has been autodetected.
		925
		926	Afterwards it contains the event model that is being used, which is the
		927	name of the Perl class implementing the model. This class is usually one
		928	of the C<AnyEvent::Impl::xxx> modules, but can be any other class in the
		929	case AnyEvent has been extended at runtime (e.g. in I<rxvt-unicode> it
		930	will be C<urxvt::anyevent>).
		931
		932	=item AnyEvent::detect
		933
		934	Returns C<$AnyEvent::MODEL>, forcing autodetection of the event model
		935	if necessary. You should only call this function right before you would
		936	have created an AnyEvent watcher anyway, that is, as late as possible at
		937	runtime, and not e.g. during initialisation of your module.
		938
		939	If you need to do some initialisation before AnyEvent watchers are
		940	created, use C<post_detect>.
		941
		942	=item $guard = AnyEvent::post_detect { BLOCK }
		943
		944	Arranges for the code block to be executed as soon as the event model is
		945	autodetected (or immediately if that has already happened).
		946
		947	The block will be executed I<after> the actual backend has been detected
		948	(C<$AnyEvent::MODEL> is set), but I<before> any watchers have been
		949	created, so it is possible to e.g. patch C<@AnyEvent::ISA> or do
		950	other initialisations - see the sources of L<AnyEvent::Strict> or
		951	L<AnyEvent::AIO> to see how this is used.
		952
		953	The most common usage is to create some global watchers, without forcing
		954	event module detection too early, for example, L<AnyEvent::AIO> creates
		955	and installs the global L<IO::AIO> watcher in a C<post_detect> block to
		956	avoid autodetecting the event module at load time.
		957
		958	If called in scalar or list context, then it creates and returns an object
		959	that automatically removes the callback again when it is destroyed (or
		960	C<undef> when the hook was immediately executed). See L<AnyEvent::AIO> for
		961	a case where this is useful.
		962
		963	Example: Create a watcher for the IO::AIO module and store it in
		964	C<$WATCHER>, but do so only do so after the event loop is initialised.
		965
		966	our WATCHER;
		967
		968	my $guard = AnyEvent::post_detect {
		969	$WATCHER = AnyEvent->io (fh => IO::AIO::poll_fileno, poll => 'r', cb => \&IO::AIO::poll_cb);
		970	};
		971
		972	# the ||= is important in case post_detect immediately runs the block,
		973	# as to not clobber the newly-created watcher. assigning both watcher and
		974	# post_detect guard to the same variable has the advantage of users being
		975	# able to just C<undef $WATCHER> if the watcher causes them grief.
		976
		977	$WATCHER ||= $guard;
		978
		979	=item @AnyEvent::post_detect
		980
		981	If there are any code references in this array (you can C<push> to it
		982	before or after loading AnyEvent), then they will be called directly
		983	after the event loop has been chosen.
		984
		985	You should check C<$AnyEvent::MODEL> before adding to this array, though:
		986	if it is defined then the event loop has already been detected, and the
		987	array will be ignored.
		988
		989	Best use C<AnyEvent::post_detect { BLOCK }> when your application allows
		990	it, as it takes care of these details.
		991
		992	This variable is mainly useful for modules that can do something useful
		993	when AnyEvent is used and thus want to know when it is initialised, but do
		994	not need to even load it by default. This array provides the means to hook
		995	into AnyEvent passively, without loading it.
		996
		997	Example: To load Coro::AnyEvent whenever Coro and AnyEvent are used
		998	together, you could put this into Coro (this is the actual code used by
		999	Coro to accomplish this):
		1000
		1001	if (defined $AnyEvent::MODEL) {
		1002	# AnyEvent already initialised, so load Coro::AnyEvent
		1003	require Coro::AnyEvent;
		1004	} else {
		1005	# AnyEvent not yet initialised, so make sure to load Coro::AnyEvent
		1006	# as soon as it is
		1007	push @AnyEvent::post_detect, sub { require Coro::AnyEvent };
		1008	}
		1009
		1010	=back
		1011
		1012	=head1 WHAT TO DO IN A MODULE
		1013
		1014	As a module author, you should C<use AnyEvent> and call AnyEvent methods
		1015	freely, but you should not load a specific event module or rely on it.
		1016
		1017	Be careful when you create watchers in the module body - AnyEvent will
		1018	decide which event module to use as soon as the first method is called, so
		1019	by calling AnyEvent in your module body you force the user of your module
		1020	to load the event module first.
		1021
		1022	Never call C<< ->recv >> on a condition variable unless you I<know> that
		1023	the C<< ->send >> method has been called on it already. This is
		1024	because it will stall the whole program, and the whole point of using
		1025	events is to stay interactive.
		1026
		1027	It is fine, however, to call C<< ->recv >> when the user of your module
		1028	requests it (i.e. if you create a http request object ad have a method
		1029	called C<results> that returns the results, it may call C<< ->recv >>
		1030	freely, as the user of your module knows what she is doing. Always).
		1031
		1032	=head1 WHAT TO DO IN THE MAIN PROGRAM
		1033
		1034	There will always be a single main program - the only place that should
		1035	dictate which event model to use.
		1036
		1037	If the program is not event-based, it need not do anything special, even
		1038	when it depends on a module that uses an AnyEvent. If the program itself
		1039	uses AnyEvent, but does not care which event loop is used, all it needs
		1040	to do is C<use AnyEvent>. In either case, AnyEvent will choose the best
		1041	available loop implementation.
		1042
		1043	If the main program relies on a specific event model - for example, in
		1044	Gtk2 programs you have to rely on the Glib module - you should load the
		1045	event module before loading AnyEvent or any module that uses it: generally
		1046	speaking, you should load it as early as possible. The reason is that
		1047	modules might create watchers when they are loaded, and AnyEvent will
		1048	decide on the event model to use as soon as it creates watchers, and it
		1049	might choose the wrong one unless you load the correct one yourself.
		1050
		1051	You can chose to use a pure-perl implementation by loading the
		1052	C<AnyEvent::Impl::Perl> module, which gives you similar behaviour
		1053	everywhere, but letting AnyEvent chose the model is generally better.
		1054
		1055	=head2 MAINLOOP EMULATION
		1056
		1057	Sometimes (often for short test scripts, or even standalone programs who
		1058	only want to use AnyEvent), you do not want to run a specific event loop.
		1059
		1060	In that case, you can use a condition variable like this:
		1061
		1062	AnyEvent->condvar->recv;
		1063
		1064	This has the effect of entering the event loop and looping forever.
		1065
		1066	Note that usually your program has some exit condition, in which case
		1067	it is better to use the "traditional" approach of storing a condition
		1068	variable somewhere, waiting for it, and sending it when the program should
		1069	exit cleanly.
		1070
		1071
		1072	=head1 OTHER MODULES
		1073
		1074	The following is a non-exhaustive list of additional modules that use
		1075	AnyEvent as a client and can therefore be mixed easily with other AnyEvent
		1076	modules and other event loops in the same program. Some of the modules
		1077	come as part of AnyEvent, the others are available via CPAN.
		1078
		1079	=over 4
		1080
		1081	=item L<AnyEvent::Util>
		1082
		1083	Contains various utility functions that replace often-used blocking
		1084	functions such as C<inet_aton> with event/callback-based versions.
		1085
		1086	=item L<AnyEvent::Socket>
		1087
		1088	Provides various utility functions for (internet protocol) sockets,
		1089	addresses and name resolution. Also functions to create non-blocking tcp
		1090	connections or tcp servers, with IPv6 and SRV record support and more.
		1091
		1092	=item L<AnyEvent::Handle>
		1093
		1094	Provide read and write buffers, manages watchers for reads and writes,
		1095	supports raw and formatted I/O, I/O queued and fully transparent and
		1096	non-blocking SSL/TLS (via L<AnyEvent::TLS>).
		1097
		1098	=item L<AnyEvent::DNS>
		1099
		1100	Provides rich asynchronous DNS resolver capabilities.
		1101
		1102	=item L<AnyEvent::HTTP>, L<AnyEvent::IRC>, L<AnyEvent::XMPP>, L<AnyEvent::GPSD>, L<AnyEvent::IGS>, L<AnyEvent::FCP>
		1103
		1104	Implement event-based interfaces to the protocols of the same name (for
		1105	the curious, IGS is the International Go Server and FCP is the Freenet
		1106	Client Protocol).
		1107
		1108	=item L<AnyEvent::Handle::UDP>
		1109
		1110	Here be danger!
		1111
		1112	As Pauli would put it, "Not only is it not right, it's not even wrong!" -
		1113	there are so many things wrong with AnyEvent::Handle::UDP, most notably
		1114	its use of a stream-based API with a protocol that isn't streamable, that
		1115	the only way to improve it is to delete it.
		1116
		1117	It features data corruption (but typically only under load) and general
		1118	confusion. On top, the author is not only clueless about UDP but also
		1119	fact-resistant - some gems of his understanding: "connect doesn't work
		1120	with UDP", "UDP packets are not IP packets", "UDP only has datagrams, not
		1121	packets", "I don't need to implement proper error checking as UDP doesn't
		1122	support error checking" and so on - he doesn't even understand what's
		1123	wrong with his module when it is explained to him.
		1124
		1125	=item L<AnyEvent::DBI>
		1126
		1127	Executes L<DBI> requests asynchronously in a proxy process for you,
		1128	notifying you in an event-based way when the operation is finished.
		1129
		1130	=item L<AnyEvent::AIO>
		1131
		1132	Truly asynchronous (as opposed to non-blocking) I/O, should be in the
		1133	toolbox of every event programmer. AnyEvent::AIO transparently fuses
		1134	L<IO::AIO> and AnyEvent together, giving AnyEvent access to event-based
		1135	file I/O, and much more.
		1136
		1137	=item L<AnyEvent::HTTPD>
		1138
		1139	A simple embedded webserver.
		1140
		1141	=item L<AnyEvent::FastPing>
		1142
		1143	The fastest ping in the west.
		1144
		1145	=item L<Coro>
		1146
		1147	Has special support for AnyEvent via L<Coro::AnyEvent>.
		1148
		1149	=back
		1150
		1151	=cut
		1152
		1153	package AnyEvent;
		1154
		1155	# basically a tuned-down version of common::sense
		1156	sub common_sense {
		1157	# from common:.sense 3.4
		1158	${^WARNING_BITS} ^= ${^WARNING_BITS} ^ "\x3c\x3f\x33\x00\x0f\xf0\x0f\xc0\xf0\xfc\x33\x00";
		1159	# use strict vars subs - NO UTF-8, as Util.pm doesn't like this atm. (uts46data.pl)
		1160	$^H |= 0x00000600;
		1161	}
		1162
		1163	BEGIN { AnyEvent::common_sense }
		1164
		1165	use Carp ();
		1166
		1167	our $VERSION = '5.34';
		1168	our $MODEL;
		1169
		1170	our $AUTOLOAD;
		1171	our @ISA;
		1172
		1173	our @REGISTRY;
		1174
		1175	our $VERBOSE;
		1176
		1177	BEGIN {
		1178	require "AnyEvent/constants.pl";
		1179
		1180	eval "sub TAINT (){" . (${^TAINT}*1) . "}";
		1181
		1182	delete @ENV{grep /^PERL_ANYEVENT_/, keys %ENV}
		1183	if ${^TAINT};
		1184
		1185	$VERBOSE = $ENV{PERL_ANYEVENT_VERBOSE}*1;
		1186
		1187	}
		1188
		1189	our $MAX_SIGNAL_LATENCY = 10;
		1190
		1191	our %PROTOCOL; # (ipv4|ipv6) => (1|2), higher numbers are preferred
		1192
		1193	{
		1194	my $idx;
		1195	$PROTOCOL{$_} = ++$idx
		1196	for reverse split /\s*,\s*/,
		1197	$ENV{PERL_ANYEVENT_PROTOCOLS} || "ipv4,ipv6";
		1198	}
		1199
		1200	my @models = (
		1201	[EV:: => AnyEvent::Impl::EV:: , 1],
		1202	[AnyEvent::Impl::Perl:: => AnyEvent::Impl::Perl:: , 1],
		1203	# everything below here will not (normally) be autoprobed
		1204	# as the pureperl backend should work everywhere
		1205	# and is usually faster
		1206	[Event:: => AnyEvent::Impl::Event::, 1],
		1207	[Glib:: => AnyEvent::Impl::Glib:: , 1], # becomes extremely slow with many watchers
		1208	[Event::Lib:: => AnyEvent::Impl::EventLib::], # too buggy
		1209	[Irssi:: => AnyEvent::Impl::Irssi::], # Irssi has a bogus "Event" package
		1210	[Tk:: => AnyEvent::Impl::Tk::], # crashes with many handles
		1211	[Qt:: => AnyEvent::Impl::Qt::], # requires special main program
		1212	[POE::Kernel:: => AnyEvent::Impl::POE::], # lasciate ogni speranza
		1213	[Wx:: => AnyEvent::Impl::POE::],
		1214	[Prima:: => AnyEvent::Impl::POE::],
		1215	[IO::Async::Loop:: => AnyEvent::Impl::IOAsync::],
		1216	[Cocoa::EventLoop:: => AnyEvent::Impl::Cocoa::],
		1217	[FLTK:: => AnyEvent::Impl::FLTK::],
		1218	);
		1219
		1220	our %method = map +($_ => 1),
		1221	qw(io timer time now now_update signal child idle condvar one_event DESTROY);
		1222
		1223	our @post_detect;
		1224
		1225	sub post_detect(&) {
		1226	my ($cb) = @_;
		1227
		1228	push @post_detect, $cb;
		1229
		1230	defined wantarray
		1231	? bless \$cb, "AnyEvent::Util::postdetect"
		1232	: ()
		1233	}
		1234
		1235	sub AnyEvent::Util::postdetect::DESTROY {
		1236	@post_detect = grep $_ != ${$_[0]}, @post_detect;
		1237	}
		1238
		1239	sub detect() {
		1240	# free some memory
		1241	*detect = sub () { $MODEL };
		1242
		1243	local $!; # for good measure
		1244	local $SIG{__DIE__};
		1245
		1246	if ($ENV{PERL_ANYEVENT_MODEL} =~ /^([a-zA-Z]+)$/) {
		1247	my $model = "AnyEvent::Impl::$1";
		1248	if (eval "require $model") {
		1249	$MODEL = $model;
		1250	warn "AnyEvent: loaded model '$model' (forced by \$ENV{PERL_ANYEVENT_MODEL}), using it.\n" if $VERBOSE >= 2;
		1251	} else {
		1252	warn "AnyEvent: unable to load model '$model' (from \$ENV{PERL_ANYEVENT_MODEL}):\n$@" if $VERBOSE;
		1253	}
		1254	}
		1255
		1256	# check for already loaded models
		1257	unless ($MODEL) {
		1258	for (@REGISTRY, @models) {
		1259	my ($package, $model) = @$_;
		1260	if (${"$package\::VERSION"} > 0) {
		1261	if (eval "require $model") {
		1262	$MODEL = $model;
		1263	warn "AnyEvent: autodetected model '$model', using it.\n" if $VERBOSE >= 2;
		1264	last;
		1265	}
		1266	}
		1267	}
		1268
		1269	unless ($MODEL) {
		1270	# try to autoload a model
		1271	for (@REGISTRY, @models) {
		1272	my ($package, $model, $autoload) = @$_;
		1273	if (
		1274	$autoload
		1275	and eval "require $package"
		1276	and ${"$package\::VERSION"} > 0
		1277	and eval "require $model"
		1278	) {
		1279	$MODEL = $model;
		1280	warn "AnyEvent: autoloaded model '$model', using it.\n" if $VERBOSE >= 2;
		1281	last;
		1282	}
		1283	}
		1284
		1285	$MODEL
		1286	or die "AnyEvent: backend autodetection failed - did you properly install AnyEvent?\n";
		1287	}
		1288	}
		1289
		1290	@models = (); # free probe data
		1291
		1292	push @{"$MODEL\::ISA"}, "AnyEvent::Base";
		1293	unshift @ISA, $MODEL;
		1294
		1295	# now nuke some methods that are overridden by the backend.
		1296	# SUPER is not allowed.
		1297	for (qw(time signal child idle)) {
		1298	undef &{"AnyEvent::Base::$_"}
		1299	if defined &{"$MODEL\::$_"};
		1300	}
		1301
		1302	if ($ENV{PERL_ANYEVENT_STRICT}) {
		1303	eval { require AnyEvent::Strict };
		1304	warn "AnyEvent: cannot load AnyEvent::Strict: $@"
		1305	if $@ && $VERBOSE;
		1306	}
		1307
		1308	(shift @post_detect)->() while @post_detect;
		1309
		1310	*post_detect = sub(&) {
		1311	shift->();
		1312
		1313	undef
		1314	};
		1315
		1316	$MODEL
		1317	}
		1318
		1319	sub AUTOLOAD {
		1320	(my $func = $AUTOLOAD) =~ s/.*://;
		1321
		1322	$method{$func}
		1323	or Carp::croak "$func: not a valid AnyEvent class method";
		1324
		1325	detect;
		1326
		1327	my $class = shift;
		1328	$class->$func (@_);
		1329	}
		1330
		1331	# utility function to dup a filehandle. this is used by many backends
		1332	# to support binding more than one watcher per filehandle (they usually
		1333	# allow only one watcher per fd, so we dup it to get a different one).
		1334	sub _dupfh($$;$$) {
		1335	my ($poll, $fh, $r, $w) = @_;
		1336
		1337	# cygwin requires the fh mode to be matching, unix doesn't
		1338	my ($rw, $mode) = $poll eq "r" ? ($r, "<&") : ($w, ">&");
		1339
		1340	open my $fh2, $mode, $fh
		1341	or die "AnyEvent->io: cannot dup() filehandle in mode '$poll': $!,";
		1342
		1343	# we assume CLOEXEC is already set by perl in all important cases
		1344
		1345	($fh2, $rw)
		1346	}
		1347
		1348	=head1 SIMPLIFIED AE API
		1349
		1350	Starting with version 5.0, AnyEvent officially supports a second, much
		1351	simpler, API that is designed to reduce the calling, typing and memory
		1352	overhead by using function call syntax and a fixed number of parameters.
		1353
		1354	See the L<AE> manpage for details.
		1355
		1356	=cut
		1357
		1358	package AE;
		1359
		1360	our $VERSION = $AnyEvent::VERSION;
		1361
		1362	# fall back to the main API by default - backends and AnyEvent::Base
		1363	# implementations can overwrite these.
		1364
		1365	sub io($$$) {
		1366	AnyEvent->io (fh => $_[0], poll => $_[1] ? "w" : "r", cb => $_[2])
		1367	}
		1368
		1369	sub timer($$$) {
		1370	AnyEvent->timer (after => $_[0], interval => $_[1], cb => $_[2])
		1371	}
		1372
		1373	sub signal($$) {
		1374	AnyEvent->signal (signal => $_[0], cb => $_[1])
		1375	}
		1376
		1377	sub child($$) {
		1378	AnyEvent->child (pid => $_[0], cb => $_[1])
		1379	}
		1380
		1381	sub idle($) {
		1382	AnyEvent->idle (cb => $_[0])
		1383	}
		1384
		1385	sub cv(;&) {
		1386	AnyEvent->condvar (@_ ? (cb => $_[0]) : ())
		1387	}
		1388
		1389	sub now() {
		1390	AnyEvent->now
		1391	}
		1392
		1393	sub now_update() {
		1394	AnyEvent->now_update
		1395	}
		1396
		1397	sub time() {
		1398	AnyEvent->time
		1399	}
		1400
		1401	package AnyEvent::Base;
		1402
		1403	# default implementations for many methods
		1404
		1405	sub time {
		1406	eval q{ # poor man's autoloading {}
		1407	# probe for availability of Time::HiRes
		1408	if (eval "use Time::HiRes (); Time::HiRes::time (); 1") {
		1409	warn "AnyEvent: using Time::HiRes for sub-second timing accuracy.\n" if $VERBOSE >= 8;
		1410	*AE::time = \&Time::HiRes::time;
		1411	# if (eval "use POSIX (); (POSIX::times())...
		1412	} else {
		1413	warn "AnyEvent: using built-in time(), WARNING, no sub-second resolution!\n" if $VERBOSE;
		1414	*AE::time = sub (){ time }; # epic fail
		1415	}
		1416
		1417	*time = sub { AE::time }; # different prototypes
		1418	};
		1419	die if $@;
		1420
		1421	&time
		1422	}
		1423
		1424	*now = \&time;
		1425
		1426	sub now_update { }
		1427
		1428	# default implementation for ->condvar
		1429
		1430	sub condvar {
		1431	eval q{ # poor man's autoloading {}
		1432	*condvar = sub {
		1433	bless { @_ == 3 ? (_ae_cb => $_[2]) : () }, "AnyEvent::CondVar"
		1434	};
		1435
		1436	*AE::cv = sub (;&) {
		1437	bless { @_ ? (_ae_cb => shift) : () }, "AnyEvent::CondVar"
		1438	};
		1439	};
		1440	die if $@;
		1441
		1442	&condvar
		1443	}
		1444
		1445	# default implementation for ->signal
		1446
		1447	our $HAVE_ASYNC_INTERRUPT;
		1448
		1449	sub _have_async_interrupt() {
		1450	$HAVE_ASYNC_INTERRUPT = 1*(!$ENV{PERL_ANYEVENT_AVOID_ASYNC_INTERRUPT}
		1451	&& eval "use Async::Interrupt 1.02 (); 1")
		1452	unless defined $HAVE_ASYNC_INTERRUPT;
		1453
		1454	$HAVE_ASYNC_INTERRUPT
		1455	}
		1456
		1457	our ($SIGPIPE_R, $SIGPIPE_W, %SIG_CB, %SIG_EV, $SIG_IO);
		1458	our (%SIG_ASY, %SIG_ASY_W);
		1459	our ($SIG_COUNT, $SIG_TW);
		1460
		1461	# install a dummy wakeup watcher to reduce signal catching latency
		1462	# used by Impls
		1463	sub _sig_add() {
		1464	unless ($SIG_COUNT++) {
		1465	# try to align timer on a full-second boundary, if possible
		1466	my $NOW = AE::now;
		1467
		1468	$SIG_TW = AE::timer
		1469	$MAX_SIGNAL_LATENCY - ($NOW - int $NOW),
		1470	$MAX_SIGNAL_LATENCY,
		1471	sub { } # just for the PERL_ASYNC_CHECK
		1472	;
		1473	}
		1474	}
		1475
		1476	sub _sig_del {
		1477	undef $SIG_TW
		1478	unless --$SIG_COUNT;
		1479	}
		1480
		1481	our $_sig_name_init; $_sig_name_init = sub {
		1482	eval q{ # poor man's autoloading {}
		1483	undef $_sig_name_init;
		1484
		1485	if (_have_async_interrupt) {
		1486	*sig2num = \&Async::Interrupt::sig2num;
		1487	*sig2name = \&Async::Interrupt::sig2name;
		1488	} else {
		1489	require Config;
		1490
		1491	my %signame2num;
		1492	@signame2num{ split ' ', $Config::Config{sig_name} }
		1493	= split ' ', $Config::Config{sig_num};
		1494
		1495	my @signum2name;
		1496	@signum2name[values %signame2num] = keys %signame2num;
		1497
		1498	*sig2num = sub($) {
		1499	$_[0] > 0 ? shift : $signame2num{+shift}
		1500	};
		1501	*sig2name = sub ($) {
		1502	$_[0] > 0 ? $signum2name[+shift] : shift
		1503	};
		1504	}
		1505	};
		1506	die if $@;
		1507	};
		1508
		1509	sub sig2num ($) { &$_sig_name_init; &sig2num }
		1510	sub sig2name($) { &$_sig_name_init; &sig2name }
		1511
		1512	sub signal {
		1513	eval q{ # poor man's autoloading {}
		1514	# probe for availability of Async::Interrupt
		1515	if (_have_async_interrupt) {
		1516	warn "AnyEvent: using Async::Interrupt for race-free signal handling.\n" if $VERBOSE >= 8;
		1517
		1518	$SIGPIPE_R = new Async::Interrupt::EventPipe;
		1519	$SIG_IO = AE::io $SIGPIPE_R->fileno, 0, \&_signal_exec;
		1520
		1521	} else {
		1522	warn "AnyEvent: using emulated perl signal handling with latency timer.\n" if $VERBOSE >= 8;
		1523
		1524	if (AnyEvent::WIN32) {
		1525	require AnyEvent::Util;
		1526
		1527	($SIGPIPE_R, $SIGPIPE_W) = AnyEvent::Util::portable_pipe ();
		1528	AnyEvent::Util::fh_nonblocking ($SIGPIPE_R, 1) if $SIGPIPE_R;
		1529	AnyEvent::Util::fh_nonblocking ($SIGPIPE_W, 1) if $SIGPIPE_W; # just in case
		1530	} else {
		1531	pipe $SIGPIPE_R, $SIGPIPE_W;
		1532	fcntl $SIGPIPE_R, AnyEvent::F_SETFL, AnyEvent::O_NONBLOCK if $SIGPIPE_R;
		1533	fcntl $SIGPIPE_W, AnyEvent::F_SETFL, AnyEvent::O_NONBLOCK if $SIGPIPE_W; # just in case
		1534
		1535	# not strictly required, as $^F is normally 2, but let's make sure...
		1536	fcntl $SIGPIPE_R, AnyEvent::F_SETFD, AnyEvent::FD_CLOEXEC;
		1537	fcntl $SIGPIPE_W, AnyEvent::F_SETFD, AnyEvent::FD_CLOEXEC;
		1538	}
		1539
		1540	$SIGPIPE_R
		1541	or Carp::croak "AnyEvent: unable to create a signal reporting pipe: $!\n";
		1542
		1543	$SIG_IO = AE::io $SIGPIPE_R, 0, \&_signal_exec;
		1544	}
		1545
		1546	*signal = $HAVE_ASYNC_INTERRUPT
		1547	? sub {
		1548	my (undef, %arg) = @_;
		1549
		1550	# async::interrupt
		1551	my $signal = sig2num $arg{signal};
		1552	$SIG_CB{$signal}{$arg{cb}} = $arg{cb};
		1553
		1554	$SIG_ASY{$signal} ||= new Async::Interrupt
		1555	cb => sub { undef $SIG_EV{$signal} },
		1556	signal => $signal,
		1557	pipe => [$SIGPIPE_R->filenos],
		1558	pipe_autodrain => 0,
		1559	;
		1560
		1561	bless [$signal, $arg{cb}], "AnyEvent::Base::signal"
		1562	}
		1563	: sub {
		1564	my (undef, %arg) = @_;
		1565
		1566	# pure perl
		1567	my $signal = sig2name $arg{signal};
		1568	$SIG_CB{$signal}{$arg{cb}} = $arg{cb};
		1569
		1570	$SIG{$signal} ||= sub {
		1571	local $!;
		1572	syswrite $SIGPIPE_W, "\x00", 1 unless %SIG_EV;
		1573	undef $SIG_EV{$signal};
		1574	};
		1575
		1576	# can't do signal processing without introducing races in pure perl,
		1577	# so limit the signal latency.
		1578	_sig_add;
		1579
		1580	bless [$signal, $arg{cb}], "AnyEvent::Base::signal"
		1581	}
		1582	;
		1583
		1584	*AnyEvent::Base::signal::DESTROY = sub {
		1585	my ($signal, $cb) = @{$_[0]};
		1586
		1587	_sig_del;
		1588
		1589	delete $SIG_CB{$signal}{$cb};
		1590
		1591	$HAVE_ASYNC_INTERRUPT
		1592	? delete $SIG_ASY{$signal}
		1593	: # delete doesn't work with older perls - they then
		1594	# print weird messages, or just unconditionally exit
		1595	# instead of getting the default action.
		1596	undef $SIG{$signal}
		1597	unless keys %{ $SIG_CB{$signal} };
		1598	};
		1599
		1600	*_signal_exec = sub {
		1601	$HAVE_ASYNC_INTERRUPT
		1602	? $SIGPIPE_R->drain
		1603	: sysread $SIGPIPE_R, (my $dummy), 9;
		1604
		1605	while (%SIG_EV) {
		1606	for (keys %SIG_EV) {
		1607	delete $SIG_EV{$_};
		1608	$_->() for values %{ $SIG_CB{$_} || {} };
		1609	}
		1610	}
		1611	};
		1612	};
		1613	die if $@;
		1614
		1615	&signal
		1616	}
		1617
		1618	# default implementation for ->child
		1619
		1620	our %PID_CB;
		1621	our $CHLD_W;
		1622	our $CHLD_DELAY_W;
		1623
		1624	# used by many Impl's
		1625	sub _emit_childstatus($$) {
		1626	my (undef, $rpid, $rstatus) = @_;
		1627
		1628	$_->($rpid, $rstatus)
		1629	for values %{ $PID_CB{$rpid} || {} },
		1630	values %{ $PID_CB{0} || {} };
		1631	}
		1632
		1633	sub child {
		1634	eval q{ # poor man's autoloading {}
		1635	*_sigchld = sub {
		1636	my $pid;
		1637
		1638	AnyEvent->_emit_childstatus ($pid, $?)
		1639	while ($pid = waitpid -1, WNOHANG) > 0;
		1640	};
		1641
		1642	*child = sub {
		1643	my (undef, %arg) = @_;
		1644
		1645	defined (my $pid = $arg{pid} + 0)
		1646	or Carp::croak "required option 'pid' is missing";
		1647
		1648	$PID_CB{$pid}{$arg{cb}} = $arg{cb};
		1649
		1650	unless ($CHLD_W) {
		1651	$CHLD_W = AE::signal CHLD => \&_sigchld;
		1652	# child could be a zombie already, so make at least one round
		1653	&_sigchld;
		1654	}
		1655
		1656	bless [$pid, $arg{cb}], "AnyEvent::Base::child"
		1657	};
		1658
		1659	*AnyEvent::Base::child::DESTROY = sub {
		1660	my ($pid, $cb) = @{$_[0]};
		1661
		1662	delete $PID_CB{$pid}{$cb};
		1663	delete $PID_CB{$pid} unless keys %{ $PID_CB{$pid} };
		1664
		1665	undef $CHLD_W unless keys %PID_CB;
		1666	};
		1667	};
		1668	die if $@;
		1669
		1670	&child
		1671	}
		1672
		1673	# idle emulation is done by simply using a timer, regardless
		1674	# of whether the process is idle or not, and not letting
		1675	# the callback use more than 50% of the time.
		1676	sub idle {
		1677	eval q{ # poor man's autoloading {}
		1678	*idle = sub {
		1679	my (undef, %arg) = @_;
		1680
		1681	my ($cb, $w, $rcb) = $arg{cb};
		1682
		1683	$rcb = sub {
		1684	if ($cb) {
		1685	$w = _time;
		1686	&$cb;
		1687	$w = _time - $w;
		1688
		1689	# never use more then 50% of the time for the idle watcher,
		1690	# within some limits
		1691	$w = 0.0001 if $w < 0.0001;
		1692	$w = 5 if $w > 5;
		1693
		1694	$w = AE::timer $w, 0, $rcb;
		1695	} else {
		1696	# clean up...
		1697	undef $w;
		1698	undef $rcb;
		1699	}
		1700	};
		1701
		1702	$w = AE::timer 0.05, 0, $rcb;
		1703
		1704	bless \\$cb, "AnyEvent::Base::idle"
		1705	};
		1706
		1707	*AnyEvent::Base::idle::DESTROY = sub {
		1708	undef $${$_[0]};
		1709	};
		1710	};
		1711	die if $@;
		1712
		1713	&idle
		1714	}
		1715
		1716	package AnyEvent::CondVar;
		1717
		1718	our @ISA = AnyEvent::CondVar::Base::;
		1719
		1720	# only to be used for subclassing
		1721	sub new {
		1722	my $class = shift;
		1723	bless AnyEvent->condvar (@_), $class
		1724	}
		1725
		1726	package AnyEvent::CondVar::Base;
		1727
		1728	#use overload
		1729	# '&{}' => sub { my $self = shift; sub { $self->send (@_) } },
		1730	# fallback => 1;
		1731
		1732	# save 300+ kilobytes by dirtily hardcoding overloading
		1733	${"AnyEvent::CondVar::Base::OVERLOAD"}{dummy}++; # Register with magic by touching.
		1734	*{'AnyEvent::CondVar::Base::()'} = sub { }; # "Make it findable via fetchmethod."
		1735	*{'AnyEvent::CondVar::Base::(&{}'} = sub { my $self = shift; sub { $self->send (@_) } }; # &{}
		1736	${'AnyEvent::CondVar::Base::()'} = 1; # fallback
		1737
		1738	our $WAITING;
		1739
		1740	sub _send {
		1741	# nop
		1742	}
		1743
		1744	sub send {
		1745	my $cv = shift;
		1746	$cv->{_ae_sent} = [@_];
		1747	(delete $cv->{_ae_cb})->($cv) if $cv->{_ae_cb};
		1748	$cv->_send;
		1749	}
		1750
		1751	sub croak {
		1752	$_[0]{_ae_croak} = $_[1];
		1753	$_[0]->send;
		1754	}
		1755
		1756	sub ready {
		1757	$_[0]{_ae_sent}
		1758	}
		1759
		1760	sub _wait {
		1761	$WAITING
		1762	and !$_[0]{_ae_sent}
		1763	and Carp::croak "AnyEvent::CondVar: recursive blocking wait detected";
		1764
		1765	local $WAITING = 1;
		1766	AnyEvent->one_event while !$_[0]{_ae_sent};
		1767	}
		1768
		1769	sub recv {
		1770	$_[0]->_wait;
		1771
		1772	Carp::croak $_[0]{_ae_croak} if $_[0]{_ae_croak};
		1773	wantarray ? @{ $_[0]{_ae_sent} } : $_[0]{_ae_sent}[0]
		1774	}
		1775
		1776	sub cb {
		1777	my $cv = shift;
		1778
		1779	@_
		1780	and $cv->{_ae_cb} = shift
		1781	and $cv->{_ae_sent}
		1782	and (delete $cv->{_ae_cb})->($cv);
		1783
		1784	$cv->{_ae_cb}
		1785	}
		1786
		1787	sub begin {
		1788	++$_[0]{_ae_counter};
		1789	$_[0]{_ae_end_cb} = $_[1] if @_ > 1;
		1790	}
		1791
		1792	sub end {
		1793	return if --$_[0]{_ae_counter};
		1794	&{ $_[0]{_ae_end_cb} || sub { $_[0]->send } };
		1795	}
		1796
		1797	# undocumented/compatibility with pre-3.4
		1798	*broadcast = \&send;
		1799	*wait = \&_wait;
		1800
		1801	=head1 ERROR AND EXCEPTION HANDLING
		1802
		1803	In general, AnyEvent does not do any error handling - it relies on the
		1804	caller to do that if required. The L<AnyEvent::Strict> module (see also
		1805	the C<PERL_ANYEVENT_STRICT> environment variable, below) provides strict
		1806	checking of all AnyEvent methods, however, which is highly useful during
		1807	development.
		1808
		1809	As for exception handling (i.e. runtime errors and exceptions thrown while
		1810	executing a callback), this is not only highly event-loop specific, but
		1811	also not in any way wrapped by this module, as this is the job of the main
		1812	program.
		1813
		1814	The pure perl event loop simply re-throws the exception (usually
		1815	within C<< condvar->recv >>), the L<Event> and L<EV> modules call C<<
		1816	$Event/EV::DIED->() >>, L<Glib> uses C<< install_exception_handler >> and
		1817	so on.
		1818
		1819	=head1 ENVIRONMENT VARIABLES
		1820
		1821	The following environment variables are used by this module or its
		1822	submodules.
		1823
		1824	Note that AnyEvent will remove I<all> environment variables starting with
		1825	C<PERL_ANYEVENT_> from C<%ENV> when it is loaded while taint mode is
		1826	enabled.
		1827
		1828	=over 4
		1829
		1830	=item C<PERL_ANYEVENT_VERBOSE>
		1831
		1832	By default, AnyEvent will be completely silent except in fatal
		1833	conditions. You can set this environment variable to make AnyEvent more
		1834	talkative.
		1835
		1836	When set to C<1> or higher, causes AnyEvent to warn about unexpected
		1837	conditions, such as not being able to load the event model specified by
		1838	C<PERL_ANYEVENT_MODEL>.
		1839
		1840	When set to C<2> or higher, cause AnyEvent to report to STDERR which event
		1841	model it chooses.
		1842
		1843	When set to C<8> or higher, then AnyEvent will report extra information on
		1844	which optional modules it loads and how it implements certain features.
		1845
		1846	=item C<PERL_ANYEVENT_STRICT>
		1847
		1848	AnyEvent does not do much argument checking by default, as thorough
		1849	argument checking is very costly. Setting this variable to a true value
		1850	will cause AnyEvent to load C<AnyEvent::Strict> and then to thoroughly
		1851	check the arguments passed to most method calls. If it finds any problems,
		1852	it will croak.
		1853
		1854	In other words, enables "strict" mode.
		1855
		1856	Unlike C<use strict> (or its modern cousin, C<< use L<common::sense>
		1857	>>, it is definitely recommended to keep it off in production. Keeping
		1858	C<PERL_ANYEVENT_STRICT=1> in your environment while developing programs
		1859	can be very useful, however.
		1860
		1861	=item C<PERL_ANYEVENT_MODEL>
		1862
		1863	This can be used to specify the event model to be used by AnyEvent, before
		1864	auto detection and -probing kicks in. It must be a string consisting
		1865	entirely of ASCII letters. The string C<AnyEvent::Impl::> gets prepended
		1866	and the resulting module name is loaded and if the load was successful,
		1867	used as event model. If it fails to load AnyEvent will proceed with
		1868	auto detection and -probing.
		1869
		1870	This functionality might change in future versions.
		1871
		1872	For example, to force the pure perl model (L<AnyEvent::Impl::Perl>) you
		1873	could start your program like this:
		1874
		1875	PERL_ANYEVENT_MODEL=Perl perl ...
		1876
		1877	=item C<PERL_ANYEVENT_PROTOCOLS>
		1878
		1879	Used by both L<AnyEvent::DNS> and L<AnyEvent::Socket> to determine preferences
		1880	for IPv4 or IPv6. The default is unspecified (and might change, or be the result
		1881	of auto probing).
		1882
		1883	Must be set to a comma-separated list of protocols or address families,
		1884	current supported: C<ipv4> and C<ipv6>. Only protocols mentioned will be
		1885	used, and preference will be given to protocols mentioned earlier in the
		1886	list.
		1887
		1888	This variable can effectively be used for denial-of-service attacks
		1889	against local programs (e.g. when setuid), although the impact is likely
		1890	small, as the program has to handle conenction and other failures anyways.
		1891
		1892	Examples: C<PERL_ANYEVENT_PROTOCOLS=ipv4,ipv6> - prefer IPv4 over IPv6,
		1893	but support both and try to use both. C<PERL_ANYEVENT_PROTOCOLS=ipv4>
		1894	- only support IPv4, never try to resolve or contact IPv6
		1895	addresses. C<PERL_ANYEVENT_PROTOCOLS=ipv6,ipv4> support either IPv4 or
		1896	IPv6, but prefer IPv6 over IPv4.
		1897
		1898	=item C<PERL_ANYEVENT_EDNS0>
		1899
		1900	Used by L<AnyEvent::DNS> to decide whether to use the EDNS0 extension
		1901	for DNS. This extension is generally useful to reduce DNS traffic, but
		1902	some (broken) firewalls drop such DNS packets, which is why it is off by
		1903	default.
		1904
		1905	Setting this variable to C<1> will cause L<AnyEvent::DNS> to announce
		1906	EDNS0 in its DNS requests.
		1907
		1908	=item C<PERL_ANYEVENT_MAX_FORKS>
		1909
		1910	The maximum number of child processes that C<AnyEvent::Util::fork_call>
		1911	will create in parallel.
		1912
		1913	=item C<PERL_ANYEVENT_MAX_OUTSTANDING_DNS>
		1914
		1915	The default value for the C<max_outstanding> parameter for the default DNS
		1916	resolver - this is the maximum number of parallel DNS requests that are
		1917	sent to the DNS server.
		1918
		1919	=item C<PERL_ANYEVENT_RESOLV_CONF>
		1920
		1921	The file to use instead of F</etc/resolv.conf> (or OS-specific
		1922	configuration) in the default resolver. When set to the empty string, no
		1923	default config will be used.
		1924
		1925	=item C<PERL_ANYEVENT_CA_FILE>, C<PERL_ANYEVENT_CA_PATH>.
		1926
		1927	When neither C<ca_file> nor C<ca_path> was specified during
		1928	L<AnyEvent::TLS> context creation, and either of these environment
		1929	variables exist, they will be used to specify CA certificate locations
		1930	instead of a system-dependent default.
		1931
		1932	=item C<PERL_ANYEVENT_AVOID_GUARD> and C<PERL_ANYEVENT_AVOID_ASYNC_INTERRUPT>
		1933
		1934	When these are set to C<1>, then the respective modules are not
		1935	loaded. Mostly good for testing AnyEvent itself.
		1936
		1937	=back
		1938
		1939	=head1 SUPPLYING YOUR OWN EVENT MODEL INTERFACE
		1940
		1941	This is an advanced topic that you do not normally need to use AnyEvent in
		1942	a module. This section is only of use to event loop authors who want to
		1943	provide AnyEvent compatibility.
		1944
		1945	If you need to support another event library which isn't directly
		1946	supported by AnyEvent, you can supply your own interface to it by
		1947	pushing, before the first watcher gets created, the package name of
		1948	the event module and the package name of the interface to use onto
		1949	C<@AnyEvent::REGISTRY>. You can do that before and even without loading
		1950	AnyEvent, so it is reasonably cheap.
		1951
		1952	Example:
		1953
		1954	push @AnyEvent::REGISTRY, [urxvt => urxvt::anyevent::];
		1955
		1956	This tells AnyEvent to (literally) use the C<urxvt::anyevent::>
		1957	package/class when it finds the C<urxvt> package/module is already loaded.
		1958
		1959	When AnyEvent is loaded and asked to find a suitable event model, it
		1960	will first check for the presence of urxvt by trying to C<use> the
		1961	C<urxvt::anyevent> module.
		1962
		1963	The class should provide implementations for all watcher types. See
		1964	L<AnyEvent::Impl::EV> (source code), L<AnyEvent::Impl::Glib> (Source code)
		1965	and so on for actual examples. Use C<perldoc -m AnyEvent::Impl::Glib> to
		1966	see the sources.
		1967
		1968	If you don't provide C<signal> and C<child> watchers than AnyEvent will
		1969	provide suitable (hopefully) replacements.
		1970
		1971	The above example isn't fictitious, the I<rxvt-unicode> (a.k.a. urxvt)
		1972	terminal emulator uses the above line as-is. An interface isn't included
		1973	in AnyEvent because it doesn't make sense outside the embedded interpreter
		1974	inside I<rxvt-unicode>, and it is updated and maintained as part of the
		1975	I<rxvt-unicode> distribution.
		1976
		1977	I<rxvt-unicode> also cheats a bit by not providing blocking access to
		1978	condition variables: code blocking while waiting for a condition will
		1979	C<die>. This still works with most modules/usages, and blocking calls must
		1980	not be done in an interactive application, so it makes sense.
		1981
		1982	=head1 EXAMPLE PROGRAM
		1983
		1984	The following program uses an I/O watcher to read data from STDIN, a timer
		1985	to display a message once per second, and a condition variable to quit the
		1986	program when the user enters quit:
		1987
		1988	use AnyEvent;
		1989
		1990	my $cv = AnyEvent->condvar;
		1991
		1992	my $io_watcher = AnyEvent->io (
		1993	fh => \*STDIN,
		1994	poll => 'r',
		1995	cb => sub {
		1996	warn "io event <$_[0]>\n"; # will always output <r>
		1997	chomp (my $input = <STDIN>); # read a line
		1998	warn "read: $input\n"; # output what has been read
		1999	$cv->send if $input =~ /^q/i; # quit program if /^q/i
		2000	},
		2001	);
		2002
		2003	my $time_watcher = AnyEvent->timer (after => 1, interval => 1, cb => sub {
		2004	warn "timeout\n"; # print 'timeout' at most every second
		2005	});
		2006
		2007	$cv->recv; # wait until user enters /^q/i
		2008
		2009	=head1 REAL-WORLD EXAMPLE
		2010
		2011	Consider the L<Net::FCP> module. It features (among others) the following
		2012	API calls, which are to freenet what HTTP GET requests are to http:
		2013
		2014	my $data = $fcp->client_get ($url); # blocks
		2015
		2016	my $transaction = $fcp->txn_client_get ($url); # does not block
		2017	$transaction->cb ( sub { ... } ); # set optional result callback
		2018	my $data = $transaction->result; # possibly blocks
		2019
		2020	The C<client_get> method works like C<LWP::Simple::get>: it requests the
		2021	given URL and waits till the data has arrived. It is defined to be:
		2022
		2023	sub client_get { $_[0]->txn_client_get ($_[1])->result }
		2024
		2025	And in fact is automatically generated. This is the blocking API of
		2026	L<Net::FCP>, and it works as simple as in any other, similar, module.
		2027
		2028	More complicated is C<txn_client_get>: It only creates a transaction
		2029	(completion, result, ...) object and initiates the transaction.
		2030
		2031	my $txn = bless { }, Net::FCP::Txn::;
		2032
		2033	It also creates a condition variable that is used to signal the completion
		2034	of the request:
		2035
		2036	$txn->{finished} = AnyAvent->condvar;
		2037
		2038	It then creates a socket in non-blocking mode.
		2039
		2040	socket $txn->{fh}, ...;
		2041	fcntl $txn->{fh}, F_SETFL, O_NONBLOCK;
		2042	connect $txn->{fh}, ...
		2043	and !$!{EWOULDBLOCK}
		2044	and !$!{EINPROGRESS}
		2045	and Carp::croak "unable to connect: $!\n";
		2046
		2047	Then it creates a write-watcher which gets called whenever an error occurs
		2048	or the connection succeeds:
		2049
		2050	$txn->{w} = AnyEvent->io (fh => $txn->{fh}, poll => 'w', cb => sub { $txn->fh_ready_w });
		2051
		2052	And returns this transaction object. The C<fh_ready_w> callback gets
		2053	called as soon as the event loop detects that the socket is ready for
		2054	writing.
		2055
		2056	The C<fh_ready_w> method makes the socket blocking again, writes the
		2057	request data and replaces the watcher by a read watcher (waiting for reply
		2058	data). The actual code is more complicated, but that doesn't matter for
		2059	this example:
		2060
		2061	fcntl $txn->{fh}, F_SETFL, 0;
		2062	syswrite $txn->{fh}, $txn->{request}
		2063	or die "connection or write error";
		2064	$txn->{w} = AnyEvent->io (fh => $txn->{fh}, poll => 'r', cb => sub { $txn->fh_ready_r });
		2065
		2066	Again, C<fh_ready_r> waits till all data has arrived, and then stores the
		2067	result and signals any possible waiters that the request has finished:
		2068
		2069	sysread $txn->{fh}, $txn->{buf}, length $txn->{$buf};
		2070
		2071	if (end-of-file or data complete) {
		2072	$txn->{result} = $txn->{buf};
		2073	$txn->{finished}->send;
		2074	$txb->{cb}->($txn) of $txn->{cb}; # also call callback
		2075	}
		2076
		2077	The C<result> method, finally, just waits for the finished signal (if the
		2078	request was already finished, it doesn't wait, of course, and returns the
		2079	data:
		2080
		2081	$txn->{finished}->recv;
		2082	return $txn->{result};
		2083
		2084	The actual code goes further and collects all errors (C<die>s, exceptions)
		2085	that occurred during request processing. The C<result> method detects
		2086	whether an exception as thrown (it is stored inside the $txn object)
		2087	and just throws the exception, which means connection errors and other
		2088	problems get reported to the code that tries to use the result, not in a
		2089	random callback.
		2090
		2091	All of this enables the following usage styles:
		2092
		2093	1. Blocking:
		2094
		2095	my $data = $fcp->client_get ($url);
		2096
		2097	2. Blocking, but running in parallel:
		2098
		2099	my @datas = map $_->result,
		2100	map $fcp->txn_client_get ($_),
		2101	@urls;
		2102
		2103	Both blocking examples work without the module user having to know
		2104	anything about events.
		2105
		2106	3a. Event-based in a main program, using any supported event module:
		2107
		2108	use EV;
		2109
		2110	$fcp->txn_client_get ($url)->cb (sub {
		2111	my $txn = shift;
		2112	my $data = $txn->result;
16	...	2113	...
17	});	2114	});
18		2115
19	# watchers get canceled whenever $w is destroyed	2116	EV::loop;
20	# only one watcher per $fh and $poll type is allowed
21	# (i.e. on a socket you cna have one r + one w or one rw
22	# watcher, not any more.
23	# timers can only be used once
24		2117
25	my $w = AnyEvent->condvar; # kind of main loop replacement	2118	3b. The module user could use AnyEvent, too:
26	# can only be used once
27	$w->wait; # enters main loop till $condvar gets ->send
28	$w->broadcast; # wake up waiting and future wait's
29		2119
30	=head1 DESCRIPTION	2120	use AnyEvent;
31		2121
32	L<AnyEvent> provides an identical interface to multiple event loops. This	2122	my $quit = AnyEvent->condvar;
33	allows module authors to utilizy an event loop without forcing module
34	users to use the same event loop (as only a single event loop can coexist
35	peacefully at any one time).
36		2123
37	The interface itself is vaguely similar but not identical to the Event	2124	$fcp->txn_client_get ($url)->cb (sub {
38	module.	2125	...
		2126	$quit->send;
		2127	});
39		2128
40	On the first call of any method, the module tries to detect the currently	2129	$quit->recv;
41	loaded event loop by probing wether any of the following modules is	2130
42	loaded: L<Coro::Event>, L<Event>, L<Glib>, L<Tk>. The first one found is	2131
43	used. If none is found, the module tries to load these modules in the	2132	=head1 BENCHMARKS
44	order given. The first one that could be successfully loaded will be	2133
45	used. If still none could be found, it will issue an error.	2134	To give you an idea of the performance and overheads that AnyEvent adds
		2135	over the event loops themselves and to give you an impression of the speed
		2136	of various event loops I prepared some benchmarks.
		2137
		2138	=head2 BENCHMARKING ANYEVENT OVERHEAD
		2139
		2140	Here is a benchmark of various supported event models used natively and
		2141	through AnyEvent. The benchmark creates a lot of timers (with a zero
		2142	timeout) and I/O watchers (watching STDOUT, a pty, to become writable,
		2143	which it is), lets them fire exactly once and destroys them again.
		2144
		2145	Source code for this benchmark is found as F<eg/bench> in the AnyEvent
		2146	distribution. It uses the L<AE> interface, which makes a real difference
		2147	for the EV and Perl backends only.
		2148
		2149	=head3 Explanation of the columns
		2150
		2151	I<watcher> is the number of event watchers created/destroyed. Since
		2152	different event models feature vastly different performances, each event
		2153	loop was given a number of watchers so that overall runtime is acceptable
		2154	and similar between tested event loop (and keep them from crashing): Glib
		2155	would probably take thousands of years if asked to process the same number
		2156	of watchers as EV in this benchmark.
		2157
		2158	I<bytes> is the number of bytes (as measured by the resident set size,
		2159	RSS) consumed by each watcher. This method of measuring captures both C
		2160	and Perl-based overheads.
		2161
		2162	I<create> is the time, in microseconds (millionths of seconds), that it
		2163	takes to create a single watcher. The callback is a closure shared between
		2164	all watchers, to avoid adding memory overhead. That means closure creation
		2165	and memory usage is not included in the figures.
		2166
		2167	I<invoke> is the time, in microseconds, used to invoke a simple
		2168	callback. The callback simply counts down a Perl variable and after it was
		2169	invoked "watcher" times, it would C<< ->send >> a condvar once to
		2170	signal the end of this phase.
		2171
		2172	I<destroy> is the time, in microseconds, that it takes to destroy a single
		2173	watcher.
		2174
		2175	=head3 Results
		2176
		2177	name watchers bytes create invoke destroy comment
		2178	EV/EV 100000 223 0.47 0.43 0.27 EV native interface
		2179	EV/Any 100000 223 0.48 0.42 0.26 EV + AnyEvent watchers
		2180	Coro::EV/Any 100000 223 0.47 0.42 0.26 coroutines + Coro::Signal
		2181	Perl/Any 100000 431 2.70 0.74 0.92 pure perl implementation
		2182	Event/Event 16000 516 31.16 31.84 0.82 Event native interface
		2183	Event/Any 16000 1203 42.61 34.79 1.80 Event + AnyEvent watchers
		2184	IOAsync/Any 16000 1911 41.92 27.45 16.81 via IO::Async::Loop::IO_Poll
		2185	IOAsync/Any 16000 1726 40.69 26.37 15.25 via IO::Async::Loop::Epoll
		2186	Glib/Any 16000 1118 89.00 12.57 51.17 quadratic behaviour
		2187	Tk/Any 2000 1346 20.96 10.75 8.00 SEGV with >> 2000 watchers
		2188	POE/Any 2000 6951 108.97 795.32 14.24 via POE::Loop::Event
		2189	POE/Any 2000 6648 94.79 774.40 575.51 via POE::Loop::Select
		2190
		2191	=head3 Discussion
		2192
		2193	The benchmark does I<not> measure scalability of the event loop very
		2194	well. For example, a select-based event loop (such as the pure perl one)
		2195	can never compete with an event loop that uses epoll when the number of
		2196	file descriptors grows high. In this benchmark, all events become ready at
		2197	the same time, so select/poll-based implementations get an unnatural speed
		2198	boost.
		2199
		2200	Also, note that the number of watchers usually has a nonlinear effect on
		2201	overall speed, that is, creating twice as many watchers doesn't take twice
		2202	the time - usually it takes longer. This puts event loops tested with a
		2203	higher number of watchers at a disadvantage.
		2204
		2205	To put the range of results into perspective, consider that on the
		2206	benchmark machine, handling an event takes roughly 1600 CPU cycles with
		2207	EV, 3100 CPU cycles with AnyEvent's pure perl loop and almost 3000000 CPU
		2208	cycles with POE.
		2209
		2210	C<EV> is the sole leader regarding speed and memory use, which are both
		2211	maximal/minimal, respectively. When using the L<AE> API there is zero
		2212	overhead (when going through the AnyEvent API create is about 5-6 times
		2213	slower, with other times being equal, so still uses far less memory than
		2214	any other event loop and is still faster than Event natively).
		2215
		2216	The pure perl implementation is hit in a few sweet spots (both the
		2217	constant timeout and the use of a single fd hit optimisations in the perl
		2218	interpreter and the backend itself). Nevertheless this shows that it
		2219	adds very little overhead in itself. Like any select-based backend its
		2220	performance becomes really bad with lots of file descriptors (and few of
		2221	them active), of course, but this was not subject of this benchmark.
		2222
		2223	The C<Event> module has a relatively high setup and callback invocation
		2224	cost, but overall scores in on the third place.
		2225
		2226	C<IO::Async> performs admirably well, about on par with C<Event>, even
		2227	when using its pure perl backend.
		2228
		2229	C<Glib>'s memory usage is quite a bit higher, but it features a
		2230	faster callback invocation and overall ends up in the same class as
		2231	C<Event>. However, Glib scales extremely badly, doubling the number of
		2232	watchers increases the processing time by more than a factor of four,
		2233	making it completely unusable when using larger numbers of watchers
		2234	(note that only a single file descriptor was used in the benchmark, so
		2235	inefficiencies of C<poll> do not account for this).
		2236
		2237	The C<Tk> adaptor works relatively well. The fact that it crashes with
		2238	more than 2000 watchers is a big setback, however, as correctness takes
		2239	precedence over speed. Nevertheless, its performance is surprising, as the
		2240	file descriptor is dup()ed for each watcher. This shows that the dup()
		2241	employed by some adaptors is not a big performance issue (it does incur a
		2242	hidden memory cost inside the kernel which is not reflected in the figures
		2243	above).
		2244
		2245	C<POE>, regardless of underlying event loop (whether using its pure perl
		2246	select-based backend or the Event module, the POE-EV backend couldn't
		2247	be tested because it wasn't working) shows abysmal performance and
		2248	memory usage with AnyEvent: Watchers use almost 30 times as much memory
		2249	as EV watchers, and 10 times as much memory as Event (the high memory
		2250	requirements are caused by requiring a session for each watcher). Watcher
		2251	invocation speed is almost 900 times slower than with AnyEvent's pure perl
		2252	implementation.
		2253
		2254	The design of the POE adaptor class in AnyEvent can not really account
		2255	for the performance issues, though, as session creation overhead is
		2256	small compared to execution of the state machine, which is coded pretty
		2257	optimally within L<AnyEvent::Impl::POE> (and while everybody agrees that
		2258	using multiple sessions is not a good approach, especially regarding
		2259	memory usage, even the author of POE could not come up with a faster
		2260	design).
		2261
		2262	=head3 Summary
46		2263
47	=over 4	2264	=over 4
48		2265
		2266	=item * Using EV through AnyEvent is faster than any other event loop
		2267	(even when used without AnyEvent), but most event loops have acceptable
		2268	performance with or without AnyEvent.
		2269
		2270	=item * The overhead AnyEvent adds is usually much smaller than the overhead of
		2271	the actual event loop, only with extremely fast event loops such as EV
		2272	adds AnyEvent significant overhead.
		2273
		2274	=item * You should avoid POE like the plague if you want performance or
		2275	reasonable memory usage.
		2276
		2277	=back
		2278
		2279	=head2 BENCHMARKING THE LARGE SERVER CASE
		2280
		2281	This benchmark actually benchmarks the event loop itself. It works by
		2282	creating a number of "servers": each server consists of a socket pair, a
		2283	timeout watcher that gets reset on activity (but never fires), and an I/O
		2284	watcher waiting for input on one side of the socket. Each time the socket
		2285	watcher reads a byte it will write that byte to a random other "server".
		2286
		2287	The effect is that there will be a lot of I/O watchers, only part of which
		2288	are active at any one point (so there is a constant number of active
		2289	fds for each loop iteration, but which fds these are is random). The
		2290	timeout is reset each time something is read because that reflects how
		2291	most timeouts work (and puts extra pressure on the event loops).
		2292
		2293	In this benchmark, we use 10000 socket pairs (20000 sockets), of which 100
		2294	(1%) are active. This mirrors the activity of large servers with many
		2295	connections, most of which are idle at any one point in time.
		2296
		2297	Source code for this benchmark is found as F<eg/bench2> in the AnyEvent
		2298	distribution. It uses the L<AE> interface, which makes a real difference
		2299	for the EV and Perl backends only.
		2300
		2301	=head3 Explanation of the columns
		2302
		2303	I<sockets> is the number of sockets, and twice the number of "servers" (as
		2304	each server has a read and write socket end).
		2305
		2306	I<create> is the time it takes to create a socket pair (which is
		2307	nontrivial) and two watchers: an I/O watcher and a timeout watcher.
		2308
		2309	I<request>, the most important value, is the time it takes to handle a
		2310	single "request", that is, reading the token from the pipe and forwarding
		2311	it to another server. This includes deleting the old timeout and creating
		2312	a new one that moves the timeout into the future.
		2313
		2314	=head3 Results
		2315
		2316	name sockets create request
		2317	EV 20000 62.66 7.99
		2318	Perl 20000 68.32 32.64
		2319	IOAsync 20000 174.06 101.15 epoll
		2320	IOAsync 20000 174.67 610.84 poll
		2321	Event 20000 202.69 242.91
		2322	Glib 20000 557.01 1689.52
		2323	POE 20000 341.54 12086.32 uses POE::Loop::Event
		2324
		2325	=head3 Discussion
		2326
		2327	This benchmark I<does> measure scalability and overall performance of the
		2328	particular event loop.
		2329
		2330	EV is again fastest. Since it is using epoll on my system, the setup time
		2331	is relatively high, though.
		2332
		2333	Perl surprisingly comes second. It is much faster than the C-based event
		2334	loops Event and Glib.
		2335
		2336	IO::Async performs very well when using its epoll backend, and still quite
		2337	good compared to Glib when using its pure perl backend.
		2338
		2339	Event suffers from high setup time as well (look at its code and you will
		2340	understand why). Callback invocation also has a high overhead compared to
		2341	the C<< $_->() for .. >>-style loop that the Perl event loop uses. Event
		2342	uses select or poll in basically all documented configurations.
		2343
		2344	Glib is hit hard by its quadratic behaviour w.r.t. many watchers. It
		2345	clearly fails to perform with many filehandles or in busy servers.
		2346
		2347	POE is still completely out of the picture, taking over 1000 times as long
		2348	as EV, and over 100 times as long as the Perl implementation, even though
		2349	it uses a C-based event loop in this case.
		2350
		2351	=head3 Summary
		2352
		2353	=over 4
		2354
		2355	=item * The pure perl implementation performs extremely well.
		2356
		2357	=item * Avoid Glib or POE in large projects where performance matters.
		2358
		2359	=back
		2360
		2361	=head2 BENCHMARKING SMALL SERVERS
		2362
		2363	While event loops should scale (and select-based ones do not...) even to
		2364	large servers, most programs we (or I :) actually write have only a few
		2365	I/O watchers.
		2366
		2367	In this benchmark, I use the same benchmark program as in the large server
		2368	case, but it uses only eight "servers", of which three are active at any
		2369	one time. This should reflect performance for a small server relatively
		2370	well.
		2371
		2372	The columns are identical to the previous table.
		2373
		2374	=head3 Results
		2375
		2376	name sockets create request
		2377	EV 16 20.00 6.54
		2378	Perl 16 25.75 12.62
		2379	Event 16 81.27 35.86
		2380	Glib 16 32.63 15.48
		2381	POE 16 261.87 276.28 uses POE::Loop::Event
		2382
		2383	=head3 Discussion
		2384
		2385	The benchmark tries to test the performance of a typical small
		2386	server. While knowing how various event loops perform is interesting, keep
		2387	in mind that their overhead in this case is usually not as important, due
		2388	to the small absolute number of watchers (that is, you need efficiency and
		2389	speed most when you have lots of watchers, not when you only have a few of
		2390	them).
		2391
		2392	EV is again fastest.
		2393
		2394	Perl again comes second. It is noticeably faster than the C-based event
		2395	loops Event and Glib, although the difference is too small to really
		2396	matter.
		2397
		2398	POE also performs much better in this case, but is is still far behind the
		2399	others.
		2400
		2401	=head3 Summary
		2402
		2403	=over 4
		2404
		2405	=item * C-based event loops perform very well with small number of
		2406	watchers, as the management overhead dominates.
		2407
		2408	=back
		2409
		2410	=head2 THE IO::Lambda BENCHMARK
		2411
		2412	Recently I was told about the benchmark in the IO::Lambda manpage, which
		2413	could be misinterpreted to make AnyEvent look bad. In fact, the benchmark
		2414	simply compares IO::Lambda with POE, and IO::Lambda looks better (which
		2415	shouldn't come as a surprise to anybody). As such, the benchmark is
		2416	fine, and mostly shows that the AnyEvent backend from IO::Lambda isn't
		2417	very optimal. But how would AnyEvent compare when used without the extra
		2418	baggage? To explore this, I wrote the equivalent benchmark for AnyEvent.
		2419
		2420	The benchmark itself creates an echo-server, and then, for 500 times,
		2421	connects to the echo server, sends a line, waits for the reply, and then
		2422	creates the next connection. This is a rather bad benchmark, as it doesn't
		2423	test the efficiency of the framework or much non-blocking I/O, but it is a
		2424	benchmark nevertheless.
		2425
		2426	name runtime
		2427	Lambda/select 0.330 sec
		2428	+ optimized 0.122 sec
		2429	Lambda/AnyEvent 0.327 sec
		2430	+ optimized 0.138 sec
		2431	Raw sockets/select 0.077 sec
		2432	POE/select, components 0.662 sec
		2433	POE/select, raw sockets 0.226 sec
		2434	POE/select, optimized 0.404 sec
		2435
		2436	AnyEvent/select/nb 0.085 sec
		2437	AnyEvent/EV/nb 0.068 sec
		2438	+state machine 0.134 sec
		2439
		2440	The benchmark is also a bit unfair (my fault): the IO::Lambda/POE
		2441	benchmarks actually make blocking connects and use 100% blocking I/O,
		2442	defeating the purpose of an event-based solution. All of the newly
		2443	written AnyEvent benchmarks use 100% non-blocking connects (using
		2444	AnyEvent::Socket::tcp_connect and the asynchronous pure perl DNS
		2445	resolver), so AnyEvent is at a disadvantage here, as non-blocking connects
		2446	generally require a lot more bookkeeping and event handling than blocking
		2447	connects (which involve a single syscall only).
		2448
		2449	The last AnyEvent benchmark additionally uses L<AnyEvent::Handle>, which
		2450	offers similar expressive power as POE and IO::Lambda, using conventional
		2451	Perl syntax. This means that both the echo server and the client are 100%
		2452	non-blocking, further placing it at a disadvantage.
		2453
		2454	As you can see, the AnyEvent + EV combination even beats the
		2455	hand-optimised "raw sockets benchmark", while AnyEvent + its pure perl
		2456	backend easily beats IO::Lambda and POE.
		2457
		2458	And even the 100% non-blocking version written using the high-level (and
		2459	slow :) L<AnyEvent::Handle> abstraction beats both POE and IO::Lambda
		2460	higher level ("unoptimised") abstractions by a large margin, even though
		2461	it does all of DNS, tcp-connect and socket I/O in a non-blocking way.
		2462
		2463	The two AnyEvent benchmarks programs can be found as F<eg/ae0.pl> and
		2464	F<eg/ae2.pl> in the AnyEvent distribution, the remaining benchmarks are
		2465	part of the IO::Lambda distribution and were used without any changes.
		2466
		2467
		2468	=head1 SIGNALS
		2469
		2470	AnyEvent currently installs handlers for these signals:
		2471
		2472	=over 4
		2473
		2474	=item SIGCHLD
		2475
		2476	A handler for C<SIGCHLD> is installed by AnyEvent's child watcher
		2477	emulation for event loops that do not support them natively. Also, some
		2478	event loops install a similar handler.
		2479
		2480	Additionally, when AnyEvent is loaded and SIGCHLD is set to IGNORE, then
		2481	AnyEvent will reset it to default, to avoid losing child exit statuses.
		2482
		2483	=item SIGPIPE
		2484
		2485	A no-op handler is installed for C<SIGPIPE> when C<$SIG{PIPE}> is C<undef>
		2486	when AnyEvent gets loaded.
		2487
		2488	The rationale for this is that AnyEvent users usually do not really depend
		2489	on SIGPIPE delivery (which is purely an optimisation for shell use, or
		2490	badly-written programs), but C<SIGPIPE> can cause spurious and rare
		2491	program exits as a lot of people do not expect C<SIGPIPE> when writing to
		2492	some random socket.
		2493
		2494	The rationale for installing a no-op handler as opposed to ignoring it is
		2495	that this way, the handler will be restored to defaults on exec.
		2496
		2497	Feel free to install your own handler, or reset it to defaults.
		2498
		2499	=back
		2500
49	=cut	2501	=cut
50		2502
51	package AnyEvent;	2503	undef $SIG{CHLD}
		2504	if $SIG{CHLD} eq 'IGNORE';
52		2505
53	no warnings;	2506	$SIG{PIPE} = sub { }
54	use strict 'vars';	2507	unless defined $SIG{PIPE};
55	use Carp;
56		2508
57	our $VERSION = 0.2;	2509	=head1 RECOMMENDED/OPTIONAL MODULES
58	our $MODEL;
59		2510
60	our $AUTOLOAD;	2511	One of AnyEvent's main goals is to be 100% Pure-Perl(tm): only perl (and
61	our @ISA;	2512	its built-in modules) are required to use it.
62		2513
63	my @models = (	2514	That does not mean that AnyEvent won't take advantage of some additional
64	[Coro => Coro::Event::],	2515	modules if they are installed.
65	[Event => Event::],
66	[Glib => Glib::],
67	[Tk => Tk::],
68	);
69		2516
70	our %method = map +($_ => 1), qw(io timer condvar broadcast wait cancel DESTROY);	2517	This section explains which additional modules will be used, and how they
		2518	affect AnyEvent's operation.
71		2519
72	sub AUTOLOAD {	2520	=over 4
73	$AUTOLOAD =~ s/.*://;
74		2521
75	$method{$AUTOLOAD}	2522	=item L<Async::Interrupt>
76	or croak "$AUTOLOAD: not a valid method for AnyEvent objects";
77		2523
78	unless ($MODEL) {	2524	This slightly arcane module is used to implement fast signal handling: To
79	# check for already loaded models	2525	my knowledge, there is no way to do completely race-free and quick
80	for (@models) {	2526	signal handling in pure perl. To ensure that signals still get
81	my ($model, $package) = @$_;	2527	delivered, AnyEvent will start an interval timer to wake up perl (and
82	if (scalar keys %{ *{"$package\::"} }) {	2528	catch the signals) with some delay (default is 10 seconds, look for
83	eval "require AnyEvent::Impl::$model"	2529	C<$AnyEvent::MAX_SIGNAL_LATENCY>).
84	or die;
85		2530
86	last if $MODEL;	2531	If this module is available, then it will be used to implement signal
87	}	2532	catching, which means that signals will not be delayed, and the event loop
88	}	2533	will not be interrupted regularly, which is more efficient (and good for
		2534	battery life on laptops).
89		2535
90	unless ($MODEL) {	2536	This affects not just the pure-perl event loop, but also other event loops
91	# try to load a model	2537	that have no signal handling on their own (e.g. Glib, Tk, Qt).
92		2538
93	for (@models) {	2539	Some event loops (POE, Event, Event::Lib) offer signal watchers natively,
94	my ($model, $package) = @$_;	2540	and either employ their own workarounds (POE) or use AnyEvent's workaround
95	eval "require AnyEvent::Impl::$model"	2541	(using C<$AnyEvent::MAX_SIGNAL_LATENCY>). Installing L<Async::Interrupt>
96	or die;	2542	does nothing for those backends.
97		2543
98	last if $MODEL;	2544	=item L<EV>
99	}
100		2545
101	$MODEL	2546	This module isn't really "optional", as it is simply one of the backend
102	or die "No event module selected for AnyEvent and autodetect failed. Install any one of these modules: Coro, Event, Glib or Tk.";	2547	event loops that AnyEvent can use. However, it is simply the best event
103	}	2548	loop available in terms of features, speed and stability: It supports
104	}	2549	the AnyEvent API optimally, implements all the watcher types in XS, does
		2550	automatic timer adjustments even when no monotonic clock is available,
		2551	can take avdantage of advanced kernel interfaces such as C<epoll> and
		2552	C<kqueue>, and is the fastest backend I<by far>. You can even embed
		2553	L<Glib>/L<Gtk2> in it (or vice versa, see L<EV::Glib> and L<Glib::EV>).
105		2554
106	@ISA = $MODEL;	2555	If you only use backends that rely on another event loop (e.g. C<Tk>),
		2556	then this module will do nothing for you.
107		2557
108	my $class = shift;	2558	=item L<Guard>
109	$class->$AUTOLOAD (@_);	2559
110	}	2560	The guard module, when used, will be used to implement
		2561	C<AnyEvent::Util::guard>. This speeds up guards considerably (and uses a
		2562	lot less memory), but otherwise doesn't affect guard operation much. It is
		2563	purely used for performance.
		2564
		2565	=item L<JSON> and L<JSON::XS>
		2566
		2567	One of these modules is required when you want to read or write JSON data
		2568	via L<AnyEvent::Handle>. L<JSON> is also written in pure-perl, but can take
		2569	advantage of the ultra-high-speed L<JSON::XS> module when it is installed.
		2570
		2571	=item L<Net::SSLeay>
		2572
		2573	Implementing TLS/SSL in Perl is certainly interesting, but not very
		2574	worthwhile: If this module is installed, then L<AnyEvent::Handle> (with
		2575	the help of L<AnyEvent::TLS>), gains the ability to do TLS/SSL.
		2576
		2577	=item L<Time::HiRes>
		2578
		2579	This module is part of perl since release 5.008. It will be used when the
		2580	chosen event library does not come with a timing source of its own. The
		2581	pure-perl event loop (L<AnyEvent::Impl::Perl>) will additionally use it to
		2582	try to use a monotonic clock for timing stability.
111		2583
112	=back	2584	=back
113		2585
114	=head1 EXAMPLE
115		2586
116	The following program uses an io watcher to read data from stdin, a timer	2587	=head1 FORK
117	to display a message once per second, and a condvar to exit the program
118	when the user enters quit:
119		2588
		2589	Most event libraries are not fork-safe. The ones who are usually are
		2590	because they rely on inefficient but fork-safe C<select> or C<poll> calls
		2591	- higher performance APIs such as BSD's kqueue or the dreaded Linux epoll
		2592	are usually badly thought-out hacks that are incompatible with fork in
		2593	one way or another. Only L<EV> is fully fork-aware and ensures that you
		2594	continue event-processing in both parent and child (or both, if you know
		2595	what you are doing).
		2596
		2597	This means that, in general, you cannot fork and do event processing in
		2598	the child if the event library was initialised before the fork (which
		2599	usually happens when the first AnyEvent watcher is created, or the library
		2600	is loaded).
		2601
		2602	If you have to fork, you must either do so I<before> creating your first
		2603	watcher OR you must not use AnyEvent at all in the child OR you must do
		2604	something completely out of the scope of AnyEvent.
		2605
		2606	The problem of doing event processing in the parent I<and> the child
		2607	is much more complicated: even for backends that I<are> fork-aware or
		2608	fork-safe, their behaviour is not usually what you want: fork clones all
		2609	watchers, that means all timers, I/O watchers etc. are active in both
		2610	parent and child, which is almost never what you want. USing C<exec>
		2611	to start worker children from some kind of manage rprocess is usually
		2612	preferred, because it is much easier and cleaner, at the expense of having
		2613	to have another binary.
		2614
		2615
		2616	=head1 SECURITY CONSIDERATIONS
		2617
		2618	AnyEvent can be forced to load any event model via
		2619	$ENV{PERL_ANYEVENT_MODEL}. While this cannot (to my knowledge) be used to
		2620	execute arbitrary code or directly gain access, it can easily be used to
		2621	make the program hang or malfunction in subtle ways, as AnyEvent watchers
		2622	will not be active when the program uses a different event model than
		2623	specified in the variable.
		2624
		2625	You can make AnyEvent completely ignore this variable by deleting it
		2626	before the first watcher gets created, e.g. with a C<BEGIN> block:
		2627
		2628	BEGIN { delete $ENV{PERL_ANYEVENT_MODEL} }
		2629
120	use AnyEvent;	2630	use AnyEvent;
121		2631
122	my $cv = AnyEvent->condvar;	2632	Similar considerations apply to $ENV{PERL_ANYEVENT_VERBOSE}, as that can
		2633	be used to probe what backend is used and gain other information (which is
		2634	probably even less useful to an attacker than PERL_ANYEVENT_MODEL), and
		2635	$ENV{PERL_ANYEVENT_STRICT}.
123		2636
124	my $io_watcher = AnyEvent->io (fh => \*STDIN, poll => 'r', cb => sub {	2637	Note that AnyEvent will remove I<all> environment variables starting with
125	warn "io event <$_[0]>\n"; # will always output <r>	2638	C<PERL_ANYEVENT_> from C<%ENV> when it is loaded while taint mode is
126	chomp (my $input = <STDIN>); # read a line	2639	enabled.
127	warn "read: $input\n"; # output what has been read
128	$cv->broadcast if $input =~ /^q/i; # quit program if /^q/i
129	});
130		2640
131	my $time_watcher; # can only be used once
132		2641
133	sub new_timer {	2642	=head1 BUGS
134	$timer = AnyEvent->timer (after => 1, cb => sub {
135	warn "timeout\n"; # print 'timeout' about every second
136	&new_timer; # and restart the time
137	});
138	}
139		2643
140	new_timer; # create first timer	2644	Perl 5.8 has numerous memleaks that sometimes hit this module and are hard
		2645	to work around. If you suffer from memleaks, first upgrade to Perl 5.10
		2646	and check wether the leaks still show up. (Perl 5.10.0 has other annoying
		2647	memleaks, such as leaking on C<map> and C<grep> but it is usually not as
		2648	pronounced).
141		2649
142	$cv->wait; # wait until user enters /^q/i
143		2650
144	=head1 SEE ALSO	2651	=head1 SEE ALSO
145		2652
146	L<Coro::Event>, L<Coro>, L<Event>, L<Glib::Event>, L<Glib>,	2653	Tutorial/Introduction: L<AnyEvent::Intro>.
147	L<AnyEvent::Impl::Coro>,	2654
148	L<AnyEvent::Impl::Event>,	2655	FAQ: L<AnyEvent::FAQ>.
149	L<AnyEvent::Impl::Glib>,	2656
		2657	Utility functions: L<AnyEvent::Util>.
		2658
		2659	Event modules: L<EV>, L<EV::Glib>, L<Glib::EV>, L<Event>, L<Glib::Event>,
		2660	L<Glib>, L<Tk>, L<Event::Lib>, L<Qt>, L<POE>.
		2661
		2662	Implementations: L<AnyEvent::Impl::EV>, L<AnyEvent::Impl::Event>,
		2663	L<AnyEvent::Impl::Glib>, L<AnyEvent::Impl::Tk>, L<AnyEvent::Impl::Perl>,
		2664	L<AnyEvent::Impl::EventLib>, L<AnyEvent::Impl::Qt>,
		2665	L<AnyEvent::Impl::POE>, L<AnyEvent::Impl::IOAsync>, L<Anyevent::Impl::Irssi>.
		2666
		2667	Non-blocking file handles, sockets, TCP clients and
		2668	servers: L<AnyEvent::Handle>, L<AnyEvent::Socket>, L<AnyEvent::TLS>.
		2669
		2670	Asynchronous DNS: L<AnyEvent::DNS>.
		2671
		2672	Thread support: L<Coro>, L<Coro::AnyEvent>, L<Coro::EV>, L<Coro::Event>.
		2673
		2674	Nontrivial usage examples: L<AnyEvent::GPSD>, L<AnyEvent::IRC>,
150	L<AnyEvent::Impl::Tk>.	2675	L<AnyEvent::HTTP>.
151		2676
152	=head1	2677
		2678	=head1 AUTHOR
		2679
		2680	Marc Lehmann <schmorp@schmorp.de>
		2681	http://home.schmorp.de/
153		2682
154	=cut	2683	=cut
155		2684
156	1	2685	1
157		2686

Diff Legend

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