IO::Async::Routine - execute code in an independent sub-process or thread


   use IO::Async::Routine;
   use IO::Async::Channel;

   use IO::Async::Loop;
   my $loop = IO::Async::Loop->new;

   my $nums_ch = IO::Async::Channel->new;
   my $ret_ch  = IO::Async::Channel->new;

   my $routine = IO::Async::Routine->new(
      channels_in  => [ $nums_ch ],
      channels_out => [ $ret_ch ],

      code => sub {
         my @nums = @{ $nums_ch->recv };
         my $ret = 0; $ret += $_ for @nums;

         # Can only send references
         $ret_ch->send( \$ret );

      on_finish => sub {
         say "The routine aborted early - $_[-1]";

   $loop->add( $routine );

   $nums_ch->send( [ 10, 20, 30 ] );
      on_recv => sub {
         my ( $ch, $totalref ) = @_;
         say "The total of 10, 20, 30 is: $$totalref";



This IO::Async::Notifier contains a body of code and executes it in a sub-process or thread, allowing it to act independently of the main program. Once set up, all communication with the code happens by values passed into or out of the Routine via IO::Async::Channel objects.

The code contained within the Routine is free to make blocking calls without stalling the rest of the program. This makes it useful for using existing code which has no option not to block within an IO::Async-based program.

To create asynchronous wrappers of functions that return a value based only on their arguments, and do not generally maintain state within the process it may be more convenient to use an IO::Async::Function instead, which uses an IO::Async::Routine to contain the body of the function and manages the Channels itself.


A choice of detachment model is available. Each has various advantages and disadvantages. Not all of them may be available on a particular system.

The fork model

The code in this model runs within its own process, created by calling fork() from the main process. It is isolated from the rest of the program in terms of memory, CPU time, and other resources. Because it is started using fork(), the initial process state is a clone of the main process.

This model performs well on UNIX-like operating systems which possess a true native fork() system call, but is not available on MSWin32 for example, because the operating system does not provide full fork-like semantics.

The thread model

The code in this model runs inside a separate thread within the main process. It therefore shares memory and other resources such as open filehandles with the main thread. As with the fork model, the initial thread state is cloned from the main controlling thread.

This model is only available on perls built to support threading.

The spawn model

Since version 0.79.

The code in this model runs within its own freshly-created process running another copy of the perl interpreter. Similar to the fork model it therefore has its own memory, CPU time, and other resources. However, since it is started freshly rather than by cloning the main process, it starts up in a clean state, without any shared resources from its parent.

Since this model creates a new fresh process rather than sharing existing state, it cannot use the code argument to specify the routine body; it must instead use only the module and func arguments.

In the current implementation this model requires exactly one input channel and exactly one output channel; both must be present, and there cannot be more than one of either.


on_finish $exitcode

For fork()-based Routines, this is invoked after the process has exited and is passed the raw exitcode status.

on_finish $type, @result

For thread-based Routines, this is invoked after the thread has returned from its code block and is passed the on_joined result.

As the behaviour of these events differs per model, it may be more convenient to use on_return and on_die instead.

on_return $result

Invoked if the code block returns normally. Note that fork()-based Routines can only transport an integer result between 0 and 255, as this is the actual exit() value.

on_die $exception

Invoked if the code block fails with an exception.


The following named parameters may be passed to new or configure:

model => "fork" | "thread" | "spawn"

Optional. Defines how the routine will detach itself from the main process. See the "Models" section above for more detail.

If the model is not specified, the environment variable IO_ASYNC_ROUTINE_MODEL is used to pick a default. If that isn't defined, fork is preferred if it is available, otherwise thread.

channels_in => ARRAY of IO::Async::Channel

ARRAY reference of IO::Async::Channel objects to set up for passing values in to the Routine.

channels_out => ARRAY of IO::Async::Channel

ARRAY reference of IO::Async::Channel objects to set up for passing values out of the Routine.

code => CODE

CODE reference to the body of the Routine, to execute once the channels are set up.

When using the spawn model, this is not permitted; you must use module and func instead.

module => STRING

func => STRING

Since version 0.79.

An alternative to the code argument, which names a module to load and a function to call within it. module should give a perl module name (i.e. Some::Name, not a filename like Some/, and func should give the basename of a function within that module (i.e. without the module name prefixed). It will be invoked as the main code body of the object, and passed in a list of all the channels; first the input ones then the output ones.

   module::func( @channels_in, @channels_out )

setup => ARRAY

Optional. For fork()-based Routines, gives a reference to an array to pass to the underlying Loop fork_child method. Ignored for thread-based Routines.



   $id = $routine->id

Returns an ID string that uniquely identifies the Routine out of all the currently-running ones. (The ID of already-exited Routines may be reused, however.)


   $model = $routine->model

Returns the detachment model in use by the Routine.


   $routine->kill( $signal )

Sends the specified signal to the routine code. This is either implemented by CORE::kill() or threads::kill as required. Note that in the thread case this has the usual limits of signal delivery to threads; namely, that it works at the Perl interpreter level, and cannot actually interrupt blocking system calls.


   $f = $routine->result_future

Since version 0.75.

Returns a new IO::Async::Future which will complete with the eventual return value or exception when the routine finishes.

If the routine finishes with a successful result then this will be the done result of the future. If the routine fails with an exception then this will be the fail result.


Paul Evans <>