++ed by:

1 non-PAUSE user.

Author image Felipe Gasper
and 1 contributors


Promise::ES6 - ES6-style promises in Perl


    use Promise::ES6;

    # OPTIONAL. And see below for other options.
    Promise::ES6::use_event('IO::Async', $loop);

    my $promise = Promise::ES6->new( sub {
        my ($resolve_cr, $reject_cr) = @_;

        # ..
    } );

    my $promise2 = $promise->then( sub { .. }, sub { .. } );

    my $promise3 = $promise->catch( sub { .. } );

    my $promise4 = $promise->finally( sub { .. } );

    my $resolved = Promise::ES6->resolve(5);
    my $rejected = Promise::ES6->reject('nono');

    my $all_promise = Promise::ES6->all( \@promises );

    my $race_promise = Promise::ES6->race( \@promises );

    my $allsettled_promise = Promise::ES6->allSettled( \@promises );


This module provides a Perl implementation of promises, a useful pattern for coordinating asynchronous tasks.

Unlike most other promise implementations on CPAN, this module mimics ECMAScript 6’s Promise interface. As the SYNOPSIS above shows, you can thus use patterns from JavaScript in Perl with only minimal changes needed to accommodate language syntax.

This is a rewrite of an earlier module, Promise::Tiny. It fixes several bugs and superfluous dependencies in the original.


This module is in use in production and, backed by a pretty extensive set of regression tests, may be considered stable.


  • Promise resolutions and rejections accept exactly one argument, not a list.

  • Unhandled rejections are reported via warn(). (See below for details.)

  • Undefined or empty rejection values trigger a warning. This provides the same value as Perl’s own warning on die(undef).

  • The Promises/A+ test suite avoids testing the case where an “executor” function’s resolve callback itself receives another promise, e.g.:

        my $p = Promise::ES6->new( sub ($res) {
            $res->( Promise::ES6->resolve(123) );
        } );

    What will $p’s resolution value be? 123, or the promise that wraps it?

    This module favors conformity with the ES6 standard, which indicates intent that $p’s resolution value be 123.


This module considers any object that has a then() method to be a promise. Note that, in the case of Future, this will yield a “false-positive”, as Future is not compatible with promises.

(See Promise::ES6::Future for more tools to interact with Future.)


This module’s handling of unhandled rejections has changed over time. The current behavior is: if any rejected promise is DESTROYed without first having received a catch callback, a warning is thrown.


In JavaScript, the following …

    Promise.resolve().then( () => console.log(1) );

… will log 2 then 1 because JavaScript’s then() defers execution of its callbacks until between iterations through JavaScript’s event loop.

Perl, of course, has no built-in event loop. This module accommodates that by implementing synchronous promises by default rather than asynchronous ones. This means that all promise callbacks run immediately rather than between iterations of an event loop. As a result, this:

    Promise::ES6->resolve(0)->then( sub { print 1 } );
    print 2;

… will print 12 instead of 21.

One effect of this is that Promise::ES6, in its default configuration, is agnostic regarding event loop interfaces: no special configuration is needed for any specific event loop. In fact, you don’t even need an event loop at all, which might be useful for abstracting over whether a given function works synchronously or asynchronously.

The disadvantage of synchronous promises—besides not being quite the same promises that we expect from JS—is that recursive promises can exceed call stack limits. For example, the following (admittedly contrived) code:

    my @nums = 1 .. 1000;

    sub _remove {
        if (@nums) {
            Promise::ES6->resolve(shift @nums)->then(\&_remove);


… will eventually fail because it will reach Perl’s call stack size limit.

That problem probably won’t affect most applications. The best way to avoid it, though, is to use asynchronous promises, à la JavaScript.

To do that, first choose one of the following event interfaces:

Then, before you start creating promises, do this:


… or:


… or:

    Promise::ES6::use_event('IO::Async', $loop);

That’s it! Promise::ES6 instances will now work asynchronously rather than synchronously.

Note that this changes Promise::ES6 globally. In IO::Async’s case, it won’t increase the passed-in IO::Async::Loop instance’s reference count, but if that loop object goes away, Promise::ES6 won’t work until you call use_event() again.

IMPORTANT: For the best long-term scalability and flexibility, your code should work with either synchronous or asynchronous promises.


Promises have never provided a standardized solution for cancellation—i.e., aborting an in-process operation. If you need this functionality, then, you’ll have to implement it yourself. Two ways of doing this are:

  • Subclass Promise::ES6 and provide cancellation logic in that subclass. See DNS::Unbound::AsyncQuery’s implementation for an example of this.

  • Implement the cancellation on a request object that your “promise-creator” also consumes. This is probably the more straightforward approach but requires that there be some object or ID besides the promise that uniquely identifies the action to be canceled. See Net::Curl::Promiser for an example of this approach.

You’ll need to decide if it makes more sense for your application to leave a canceled query in the “pending” state or to “settle” (i.e., resolve or reject) it. All things being equal, I feel the first approach is the most intuitive, while the latter ends up being “cleaner”.


It’s easy to create inadvertent memory leaks using promises in Perl. Here are a few “pointers” (heh) to bear in mind:

  • Any Promise::ES6 instances that are created while $Promise::ES6::DETECT_MEMORY_LEAKS is set to a truthy value are “leak-detect-enabled”, which means that if they survive until their original process’s global destruction, a warning is triggered. You should normally enable this flag in a development environment.

  • If your application needs recursive promises (e.g., to poll iteratively for completion of a task), the current_sub feature (i.e., __SUB__) may help you avoid memory leaks. In Perl versions that don’t support this feature (i.e., anything pre-5.16) you can imitate it thus:

        use constant _has_current_sub => eval "use feature 'current_sub'";
        use if _has_current_sub(), feature => 'current_sub';
        my $cb;
        $cb = sub {
            my $current_sub = do {
                no strict 'subs';
                _has_current_sub() ? __SUB__ : eval '$cb';

    Of course, it’s better if you can avoid doing that. :)

  • Garbage collection before Perl 5.18 seems to have been buggy. If you work with such versions and end up chasing leaks, try manually deleting as many references/closures as possible. See t/race_success.t for a notated example.

    You may also (counterintuitively, IMO) find that this:

        my ($resolve, $reject);
        my $promise = Promise::ES6->new( sub { ($resolve, $reject) = @_ } );
        # … etc.

    … works better than:

        my $promise = Promise::ES6->new( sub {
            my ($resolve, $reject) = @_;
            # … etc.
        } );


If you’re not sure of what promises are, there are several good introductions to the topic. You might start with this one.

Promise::XS is my refactor of AnyEvent::XSPromises. It’s a lot like this library but implemented mostly in XS for speed.

Promises is another pure-Perl Promise implementation.

Future fills a role similar to that of promises. Much of the IO::Async ecosystem assumes (or strongly encourages) its use.

CPAN contains a number of other modules that implement promises. I think mine are the nicest :), but YMMV. Enjoy!


Copyright 2019-2020 Gasper Software Consulting.

This library is licensed under the same terms as Perl itself.