Let's say we are going to test a module that is supposed to be run asynchronously - using threads, AnyEvent, or Coro, or some other means. And we need to check that certain events happen in certain sequence, because some of them depend on the others.

Probably the best way to achieve this would be of course to minimize interdependencies and use mathematically correct synchronization for whatever is left. Of course, that is not always achievable, due to limited time.

So I'm going to propose a primitive that I think should deal with a huge subclass of such tasks.

The code goes as follows:

use Test::AsyncSeq;

my $id  = Test::AsyncSeq->get_sequence_id;
my $id2 = Test::AsyncSeq->get_sequence_id( "frobnicate" );
    # would be "frobnicate1" or smth

# somewhere in threads/callbacks
is_after( $id, "start" );
# elsewhere

is_after( $id, "event2", "start" );

# more elsewhere
is_after( $id, "event3", "start" );

# finally
is_after( $id, "finish", "event2", "event3" );
[download]

The is_after( $id, $event, @dependencies ); passes if and only if:

• sequence named $id was created;
• $event was not seen in the sequence yet;
• all of the @dependencies have been seen at the moment of the call.

The id is just a string, and is required since Perl is not very good at passing blessed references across threads. And multiple tests MAY be needed in the same script, say to catch a race condition.

Does such interface make sense? Would it be of use to anyone?