Async::Defer - VM to write and run async code in usual sync-like way
This document describes Async::Defer version v1.0.1
use Async::Defer;
# ... CREATE
my $defer = Async::Defer->new();
my $defer2 = $defer->clone();
# ... SETUP
$defer->do(sub{
my ($d, @param) = @_;
# run sync/async code which MUST end with one of:
# $d->done(@result);
# $d->throw($error);
# $d->continue();
# $d->break();
});
$defer->if(sub{ my $d=shift; return 1 });
$defer->try();
$defer->do($defer2);
$defer->catch(
qr/^io:/ => sub{
my ($d,$err) = @_;
# end with $d->done/throw/continue/break
},
qr/.*/ => sub{ # WILL CATCH ALL EXCEPTIONS
my ($d,$err) = @_;
# end with $d->done/throw/continue/break
},
FINALLY => sub{
my ($d,$err,@result) = @_;
# end with $d->done/throw/continue/break
},
);
$defer->else();
$defer->while(sub{ my $d=shift; return $d->iter() <= 3 });
$defer->do(sub{
my ($d) = @_;
# may access $d->iter() here
# end with $d->done/throw/continue/break
});
$defer->end_while();
$defer->end_if();
$defer->{anyvar} = 'anyval';
# ... START
$defer->run();
WARNING: This is experimental code, public interface may change.
This module's goal is to simplify writing complex async event-based code, which usually mean huge amount of callback/errback functions, very hard to support. It was initially inspired by Python/Twisted's Deferred object, but go further and provide virtual machine which allow you to write/define complete async program (which consists of many callback/errback) in sync way, just like you write usual non-async programs.
Main idea is simple. For example, if you've this non-async code:
$var = fetch_val();
process_val( $var );
and want to make fetch_val()
async, you usually do something like this:
fetch_val( cb => \&value_fetched );
sub value_fetched {
my ($var) = @_;
process_val( $var );
}
With Async::Defer you will split initial non-async code in sync parts (usually this mean - split on assignment operator):
### 1
fetch_val();
### 2
$var =
process_val( $var );
then wrap each part in separate anon sub and add Defer object to join these parts together:
$d = Async::Defer->new();
$d->do(sub{
my ($d) = @_;
fetch_val( $d ); # will call $d->done('…result…') when done
});
$d->do(sub{
my ($d, $var) = @_;
process_val( $var );
$d->done(); # this sub is sync, it call done() immediately
});
$d->run();
These anon subs are similar to statements in perl. Between these
statements you can use flow control operators like if()
,
while()
and try()
/catch()
. And inside statements you can
control execution flow using done()
, throw()
, continue()
and break()
operators when current async function will finish and
will be ready to go to the continue step.
Finally, you can use Async::Defer object to keep your local variables -
this object is empty hash, and you can create any keys in it.
Single Defer object described this way is sort of single function.
And it's possible to call another functions by using another Defer
object as parameter for do()
instead of usual anon sub.
While you can use both sync and async sub in do()
, they all MUST
call one of done()
, throw()
, continue()
or break()
when they finish
their work, and do this ONLY ONCE. This is Defer's way to proceed from
one step to another, and if not done right Defer object's behaviour is
undefined!
There are several ways to implement this, and it's unclear yet which
way is the best. We can implement full-featured stack with local variables
similar to perl's local
using getter/setter methods; we can fill called
Defer objects with copy of all keys in parent Defer object (so called
object will have full read-only access to parent's scalar data, and read/write
access to parent's reference data types); we can do nothing and let user
manually send all needed data to called Defer object as params and get
data back using returned values (by done()
or throw()
).
In current implementation we do nothing, so here is some ways to go:
### @results = another_defer(@params)
$d->do(sub{
my ($d) = @_;
my @params_for_another_defer = (…);
$d->done(@params_for_another_defer);
});
$d->do($another_defer);
$d->do(sub{
my ($d, @results_from_another_defer) = @_;
...
$d->done();
});
### share some local variables with $another_defer
$d->do(sub{
my ($d) = @_;
$d->{readonly} = $scalar;
$d->{readwrite} = $ref_to_something;
$another_defer->{readonly} = $d->{readonly};
$another_defer->{readwrite} = $d->{readwrite};
$d->done();
});
$d->do($another_defer);
$d->do(sub{
my ($d) = @_;
# $d->{readwrite} here may be modifed by $another_defer
$d->done();
});
### share all variables with $another_defer (run it manually)
$d->do(sub{
my ($d) = @_;
%$another_defer = %$d;
$another_defer->run($d);
});
$d->do(sub{
my ($d) = @_;
# all reference-type keys in $d may be modifed by $another_defer
$d->done();
});
If you want to reuse same Defer object several times, then you should keep
in mind: keys created inside this object on first run won't be automatically
removed, so on second and continue runs it will see internal data left by
previous runs. This may or may not be desirable behaviour. In later case
you should use clone()
and run only clones of original object (clones are
created using %$clone=%$orig
, so they share only reference-type keys
which exists in original Defer):
$d->do( $another_defer->clone() );
$d->do( $another_defer->clone() );
Async::Defer objects can be nested, and there are two ways to do it.
One way is to add a child defer to the parent defer using do()
method.
my $cd = Async::Defer->new();
## setup the child defer.
$cd->do( ... );
## parent defer
my $pd = Async::Defer->new();
$pd->do( ... );
$pd->do(sub {
my $d = shift;
...;
$d->done( @arguments_for_child_defer );
});
## run the child defer
$pd->do($cd);
$pd->do(sub {
my ($d, @results_from_child_defer) = @_;
...;
});
The other way is to call run()
on the child defer with its first
argument being the parent defer. This is very useful when you dynamically
create the child defer in statements of the parent defer.
## parent defer
my $pd = Async::Defer->new();
$pd->do(sub {
my ($d, @args) = @_;
## create the child defer in the statement
my $cd = Async::Defer->new();
## setup the child defer
$cd->do( ... );
## run() the child.
## You do not have to call $d->done explicitly,
## because the flow continues from the child to the parent.
$cd->run($d, @argments_for_child_defer);
});
$pd->do(sub {
my ($d, @results_from_child_defer) = @_;
...;
});
Nothing.
$defer = Async::Defer->new();
Create and return Async::Defer object.
$defer2 = $defer->clone();
Clone existing Async::Defer object and return clone.
Clone will have same program (STATEMENTS and OPERATORS added to
original object) and same local variables (non-deep copy of orig object
keys using %$clone=%$orig
). After cloning these two objects can be
modified (by adding new STATEMENTS, OPERATORS and modifying variables)
independently.
It's possible to clone()
object which is running right now, cloned object
will not be in running state - this is safe way to run()
objects which may
or may not be already running.
$defer->run( $parent_defer, @params );
$defer->run( \&callback, @params );
Start executing object's current program, which must be defined first by
adding at least one STATEMENT (do()
or <catch(FINALLY=
sub{})>>)
to this object.
Usually while run()
only first STATEMENT will be executed (with optional
@params
in parameters). It will just start some async function and
returns, and run()
will returns immediately after this too. Actual
execution of this object will continue when started async function will
finish (usually after Timer or I/O event) and call this object's done()
,
break()
, continue()
or throw()
methods.
It's possible to make all STATEMENTS sync - in this case full program
will be executed before returning from run()
- but this has no real sense
because you don't need Defer object for sync programs.
If run()
used to start top-level program (i.e. without $parent_defer
parameter), then there will be no return value at end of program -
after break STATEMENT in this object will call done()
nothing else will
happens and any parameters of that break done()
call will be ignored.
If this Defer object was started as part of another program (i.e. it was
added there using do()
or just manually executed from some STATEMENT with
defined $parent_defer
parameter), then it return value will be delivered
to continue STATEMENT in $parent_defer
object (See "NESTED DEFERS").
The first argument for run()
may also be a subroutine reference (\&callback
).
In this case, the callback is called after break STATEMENT in this object.
The arguments for the callback are the results of the break STATEMENT.
Any value returned from \&callback
will be ignored.
This method available only inside while()
- both in while()
's
\&conditional
argument and while()
's body STATEMENTS. It return
current iteration number for nearest while()
, starting from 1.
# this loop will execute 3 times:
$d->while(sub{ shift->iter() <= 3 });
$d->do(sub{
my ($d) = @_;
printf "Iteration %d\n", $d->iter();
$d->done();
});
$d->end_while();
All STATEMENTS methods return the Async::Defer object, so that you can chain method calls.
$defer = $defer->do( \&sync_or_async_code, … );
$defer = $defer->do( $child_defer, … );
Add STATEMENT to this object's program.
When this STATEMENT should be executed, \&sync_or_async_code
(or $child_defer
's first STATEMENT) will be called with these params:
( $defer_object, @optional_results_from_previous_STATEMENT )
do()
accepts multiple arguments. Those STATEMENTS are added to the object
in that order, and can be mix of any types - i.e. it's same as call do()
sequentially multiple times providing these arguments one-by-one.
$defer->do(
\&code1,
$defer2,
[$defer3, $defer4, \&code5],
{
task1 => $defer6,
task2 => \&code7,
},
\&more_code,
...
);
$defer = $defer->do(
[\&sync_or_async_code, $child_defer, ...],
...
);
$defer = $defer->do(
{
task1 => \&sync_or_async_code,
task2 => $child_defer,
...
},
...
)
Add one STATEMENT to this object's program.
When this STATEMENT should be executed, all these tasks will be started
simultaneously (Defer objects using clone()
and run()
, code by
transforming into new Defer object and then also run()
).
This program will continue only after all these tasks will be finished
(either with done()
or throw()
).
It's possible to provide params individually for each of these tasks and receive results/error returned by each of these tasks, but actual syntax depends on how these tasks was named - by id (ARRAY) or by name (HASH):
$d->do(sub{
my ($d) = @_;
$d->done(
['param1 for task1', 'param2 for task1'],
['param1 for task2'],
[undef, 'param2 for task3'],
# no params for task4,task5,…
);
});
$d->do([ $d_task1, $d_task2, $d_task3, $d_some, $d_some ]);
$d->do(sub{
my ($d, @taskresults) = @_;
my $id = 1;
if (ref $taskresults[$id-1]) {
print "task $id results:", @{ $taskresults[$id-1] };
} else {
print "task $id throw error:", $taskresults[$id-1];
}
});
$d->do(sub{
my ($d) = @_;
$d->done(
task1 => ['param1 for task1', 'param2 for task1'],
task2 => ['param1 for task2'],
task3 => [undef, 'param2 for task3'],
# no params for task4,task5,…
);
});
$d->do({
task1 => $d_task1,
task2 => $d_task2,
task3 => $d_task3,
task4 => $d_some,
task5 => $d_some,
});
$d->do(sub{
my ($d, %taskresults) = @_;
if (ref $taskresults{task1}) {
print "task1 results:", @{ $taskresults{task1} };
} else {
print "task1 throw error:", $taskresults{task1};
}
});
$defer = $defer->if( \&conditional );
$defer = $defer->else();
$defer = $defer->end_if();
Add conditional OPERATOR to this object's program.
When this OPERATOR should be executed, \&conditional
will be called
with single param:
( $defer_object )
The \&conditional
MUST be sync, and return true/false.
$defer = $defer->while( \&conditional );
$defer = $defer->end_while();
Add loop OPERATOR to this object's program.
When this OPERATOR should be executed, \&conditional
will be called with
single param:
( $defer_object )
The \&conditional
MUST be sync, and return true/false.
$defer = $defer->try();
$defer = $defer->catch(
$regex_or_FINALLY => \&sync_or_async_code,
...
);
Add exception handling to this object's program.
In general, try/catch/finally behaviour is same as in Java (and probably many other languages).
If some STATEMENTS inside try/catch block will throw()
, the thrown error
can be intercepted (using matching regexp in catch()
) and handled in any
way (blocked - if catch()
handler call done()
, continue()
or break()
or
replaced by another exception - if catch()
handler call throw()
).
If exception match more than one regexp, first successfully matched
regexp's handler will be used. Handler will be executed with params:
( $defer_object, $error )
In addition to exception handlers you can also define FINALLY handler
(by using string "FINALLY"
instead of regex). FINALLY handler will be
called in any case (with/without exception) and may handle this in any way
just like any other exception handler in catch()
. FINALLY handler will
be executed with different params:
# with exception
( $defer_object, $error)
# without exception
( $defer_object, @optional_results_from_previous_STATEMENT )
Unless you are nesting child defers, one and only one of these methods MUST be called at end of each STATEMENT, both sync and async! In the case of nested defers, see "NESTED DEFERS".
$defer->done( @optional_result );
Go to continue STATEMENT/OPERATOR. If continue is STATEMENT, it will receive
@optional_result
in it parameters.
$defer->throw( $error );
Throw exception. Nearest matching catch()
or FINALLY STATEMENT will be
executed and receive $error
in it parameter.
$defer->continue();
Move to beginning of nearest while()
(or to first STATEMENT if
called outside while()
) and continue with continue iteration (if while()
's
\&conditional
still returns true).
$defer->break();
Move to first STATEMENT/OPERATOR after nearest while()
(or finish this
program if called outside while()
- returning to parent's Defer object
if any).
Please report any bugs or feature requests through the issue tracker at https://github.com/powerman/perl-Async-Defer/issues. You will be notified automatically of any progress on your issue.
This is open source software. The code repository is available for public review and contribution under the terms of the license. Feel free to fork the repository and submit pull requests.
https://github.com/powerman/perl-Async-Defer
git clone https://github.com/powerman/perl-Async-Defer.git
-
MetaCPAN Search
-
CPAN Ratings
-
AnnoCPAN: Annotated CPAN documentation
-
CPAN Testers Matrix
-
CPANTS: A CPAN Testing Service (Kwalitee)
Alex Efros powerman@cpan.org
Toshio Ito toshioito [at] cpan.org
This software is Copyright (c) 2011- by Alex Efros powerman@cpan.org.
This is free software, licensed under:
The MIT (X11) License