Adding the following line at the end of sub BUILD in “MyClass.pm” seems to fix the problem:

Role::Tiny->apply_roles_to_object($self, $module);
[download]

So my guess is: roles are, by default, class-based, rather than object-based, so that once a role has been added to a class, the definition of that class doesn’t change. But with multiple threads, the problem doesn’t arise as the class definition is re-created for each new thread.

Hope that helps,

Ha, that's actually what I suspected, that the package was being permanently altered in Perl's brain the first time through. I just didn't think that with() operated on the package level, as does apply_roles_to_package (which I was eschewing because of that).

But it gets back to what I mentioned in an earlier discussion, which is that using apply_roles_to_object seems to break the whole point of Roles, in that using it in sub BUILD in the parent causes the child's method to be loaded *before* the parent's of the same name. Notably, placing the conditional code inside an after BUILD => sub { ... }; block makes no difference to the outcome.

package MyClass;
use Moo;
use Moo::Role();

sub BUILD {
    my ( $self, $args ) = @_;

    my $module = $args->{'num'} % 2 ? 'Odd' : 'Even';
    warn "$args->{'num'} $self loading $module";

    Moo::Role->apply_roles_to_object( $self, $module);

    return $self;
};

sub baz {
    return 'qux';
}

1;
[download]

And in each of the Roles:

sub baz {
    die "Died in ", __PACKAGE__;
}
[download]

Output:

1 MyClass=HASH(0x19c12c0) loading Odd at MyClass.pm line 14.
Loaded Odd at Odd.pm line 4.
1 MyClass__WITH__Odd=HASH(0x19c12c0)->bar at test.pl line 11.
Died in Odd at Odd.pm line 11.
[download]

I guess I will just code around that; indeed I don't reuse method names anywhere as a practise ... but it seems odd. Oh well. ++Athanasius for helping me figure out that I was just seeing another symptom of the same issue I was dealing with earlier!



The way forward always starts with a minimal test.