#19857: Mistakenly including non .pm files when using require => 1

When using the "use Module::Pluggable require => 1", M::B mistakenly
returns directories from the plugins() method. This is a regression
from v2.97 and is causing test failures in Perl::Critic.

Here's a simplified test case that demonstrates the problem:

% mkdir -p lib/Foo/Plugin/Subdir
% cat > lib/Foo.pm
package Foo;
use Module::Pluggable;
print "$_\n" for plugins();
1;
% cat > lib/Foo/Plugin/Subdir/Bar.pm
package Foo::Plugin::Subdir::Bar;
1;
% perl -Ilib -e 'use Foo'
Foo::Plugin::Subdir::Bar
% cat >! lib/Foo.pm
package Foo;
use Module::Pluggable require => 1;
print "$_\n" for plugins();
1;
% perl -Ilib -e 'use Foo'
Foo::Plugin::Subdir
Foo::Plugin::Subdir::Bar

The failure is that plugins() returns "Foo::Plugin::Subdir" in the
latter case. I believe the bug is in Devel::InnerPackage. The
following example demonstrates the spurious return of the subdir:

% perl -Ilib -le'use Devel::InnerPackage;require
Foo::Plugin::Subdir::Bar;print for Devel::InnerPackage::list_packages
"Foo::Plugin";'
Foo::Plugin::Subdir
Foo::Plugin::Subdir::Bar

This should instead return just "Foo::Plugin::Subdir::Bar", I believe.

-- Chris

On Tue Jun 13 02:03:27 2006, guest wrote:
> When using the "use Module::Pluggable require => 1", M::B mistakenly
> returns directories from the plugins() method. This is a regression
> from v2.97 and is causing test failures in Perl::Critic.

The problem is caused by the change from 3.0 to 3.01

The big problem is that there's no way to tell (as far as I can see)
between a 'virtual' package and a real package.

So for example.

package No;

package No::Middle::Bit;

There is no actual package 'No::Middle'.

However, my fix for that breaks another test I have which is checking for

package TA::C::A::I;

sub foo { }

package TA::C::A::I::A;

sub foo { }

package TA::C::A::I::A::B;

sub foo { }

1;

In which case it only returns the last one TA::C::A::I::A::B, no ::I or
::I::A

I'm lookign at how to solve this now.

Sorry :(

Simon,

Thanks for the quick response. I discovered that setting "inner => 0"
works around the bug, so we'll release a new Perl::Critic and that
should make everything happy again on our end.

Perhaps inner => 0 should be the default? Or is it too late for an API
change of that magnitude?

Chris

Fixed in 3.10 I think