AnyEvent::Subprocess - flexible, OO, asynchronous process spawning and management


version 1.102912


    use AnyEvent::Subprocess;

    # prepare the job
    my $job = AnyEvent::Subprocess->new(
        delegates     => ['StandardHandles'],
        on_completion => sub { die 'bad exit status' unless $_[0]->is_success },
        code          => sub {
            my %args = %{$_[0]};
                print "Got line: $_";
            exit 0;

    # start the child
    my $run = $job->run;

    # add watcher to print the next line we see on the child's stdout
    $run->delegate('stdout')->handle->push_read( line => sub {
        my ($h, $line) = @_;
        say "The child said: $line";

    # write to the child's stdin
    $run->delegate('stdin')->handle->push_write("Hello, world!\n");

    # close stdin after it has been written to the child
    $run->delegate('stdin')->handle->on_drain(sub { $_[0]->close_fh });

    # kill the child if it takes too long to produce a result
    my $killer = AnyEvent->timer( after => 42, interval => 0, cb => sub {
       $run->kill(2); # SIGINT.

    # ensure the event loop runs until the on_completion handler dies
    EV::loop(); # you can use any AnyEvent-compatible event loop, including POE

    # eventually prints "The child said: Got line: Hello, world!", or
    # perhaps dies if your system is really really overloaded.


There are so many possible ways to use this module that a tutorial would take me months to write. You should definitely read the test suite to see what possibilities exist. (There is also an examples directory in the dist.)

The basic "flow" is like in the SYNOPSIS section; create a job, call run, wait for your callback to be called with the exit status of the subprocess.

The fun comes when you add delegates.

Delegates are technically instances of classes. Typing:

   my $stdin = AnyEvent::Subprocess::Job::Delegate::Handle->new(
       name      => 'stdin',
       direction => 'w',
       replace   => \*STDIN,

Every time you want to be able to write to STDIN is going to become tiring after a while. When you load AnyEvent::Subprocess, you also load AnyEvent::Subprocess::DefaultDelegate. This registers short names for each delegate and will cause AnyEvent::Subprocess::Job to build the actual instances automatically. This means you can say 'StandardHandles' to get a delegate for each of STDIN, STDOUT, and STDERR. If you want to know how all the sugary names work, just open and take a look. (The documentation for that module also covers that, as well as how to define your own delegate builders.)

If you are too lazy to look -- there are delegates for giving the child arbitrary sockets or pipes opened to arbitrary file descriptors (so you can deal with more than stdin/stdout/stderr and communicate bidirectionally between the parent and child), there is a delegate for giving the child a pseudo-tty (which can run complicatged programs, like emacs!), there is a delegate for capturing any input automatically, and passing it back to the parent via the Done object, and there is a delegate for calling functions in the parent when certain events are received.

Once you have decided what delegates your job needs, you need to create a job object:

   my $proc = AnyEvent::Subprocess->new(
       delegates     => [qw/List them here/],
       code          => sub { code to run in the child },
       on_completion => sub { code to run in the parent when the child is done },

Then you can run it:

   my $running = $proc->run;
   my $another = $proc->run({ with => 'args' }); # a separate process

The code coderef receives a hashref as an argument; delegates typically populate this for you, but you can also pass a hashref of args to run (that will be merged with any arguments the delegates create; the delegates' arguments "win" if there is a conflict). The code then runs in a child process until it stops running for some reason. The on_completion hook is then run in the parent, with the AnyEvent::Subprocess::Done object passed in as an argument.

You can, of course, have as many children running concurrently as you desire. The event loop handles managing the any IO with the child, and the notifications of the child dying. (You don't need to deal with SIGCHLD or anything like that.)


AnyEvent::Subprocess is a set of modules for running external processes, and interacting with them in the context of an event-driven program. It is similar to POE::Wheel::Run, but much more customizable (and Moose-based). It is also similar to modules that really want to be event-based, but aren't for some reason; this includes IPC::Run and IPC::Open3, Expect, and even the built-in qx// operator. You can replace all those modules with this one, and have the ability to write much more flexible applications and libraries.

AnyEvent::Subprocess is based on three classes; AnyEvent::Subprocess::Job, which represents a job that can be run at a later time, AnyEvent::Subprocess::Running, which represents a running child process, and AnyEvent::Subprocess::Done, which represents a completed job. The Job object contains the command to run, information about its environment (which handles to capture, which plugins to run, what to do when the job is done, etc.). Then Run object is returned by $job->run, and lets you interact with the running subprocess. This includes things like writing to its pipes/sockets, reading from its pipes, sending it signals, and so on. When the running job exits, the on_completion handler provided by the Job object is called with a Done object. This contains the exit status, output that the process produced (if requested), and so on.

What makes this more interesting is the ability to add delegates to any of these classes. These delegates are called into at various points and allow you to add more features. By default, you just get a callback when the process exits. You can also kill the running process. That's it. From there, you can add delegates to add more features. You can add a pipe to share between the parent and the child. Instead of sharing a pipe, you can have an fd opened to an arbitrary file descriptor number in the child. You have an infinite number of these, so you can capture the child's stdout and stderr, write to its stdin, and also share a socket for out-of-band communication. You can also open a pipe to the child's fd #5 and write to it. (This is nice if you are invoking something like gpg that wants the password written on an arbitrary fd other than 1.)

(This is all done with the included Handle delegate. See AnyEvent::Subprocess::Job::Delegate::Handle.)

You can then build upon this; instead of writing your own code to reading the handles when they become readable and accumulate input, you can write a delegate that saves all the data coming from a given handle and gives it to your program after the child exits (via the Done instance).

(This is also included via the CaptureHandle delegate. See AnyEvent::Subprocess::Job::Delegate::CaptureHandle.)

All of this integrates into your existing event-based app; waiting for IO from the child (or waiting for the child to exit) is asynchronous, and lets your app do other work while waiting for the child. (It can integrate nicely into Coro, for example, unlike the default qx//.)


The parent's event loop still exists in the child process, which means you can't safely use it in the child. I have tried to work around this in a few event loops; my AnyEventX::Cancel module on github is an early attempt. When that becomes stable, I will remove this restriction.


This is not a problem if you are running external processes, but is a problem if you are running a code block and you want to do event-ful things in there. (Note that EV is designed to allow the child to handle events that the parent created watchers for. You can do that just fine. It's if you want a fresh event loop with no existing watchers that doesn't work well yet.)


Jonathan Rockway <>


This software is copyright (c) 2011 by Jonathan Rockway.

This is free software; you can redistribute it and/or modify it under the same terms as the Perl 5 programming language system itself.