From @nwc10

warnings​::enabled is documented as this​:

=item warnings​::enabled($category)

Return TRUE if the warnings category, C<$category>, is enabled in the
calling module.
Otherwise returns FALSE.

=cut

However, it only works as documented in simple cases. All 3 of these should
warn​:

#!perl
use strict;
no warnings;

package Foo;

use warnings::register;

sub method {
my $self = shift;
warnings::warnif($self, 'Invocant');
}

sub subroutine {
my ($a, $b) = @_;
warnings::warnif($b, 'Second');
}

sub funky {
my $self = shift;
my $r = \@_;
warnings::warnif($self, 'Funky');
}

package main;

use warnings 'Foo';

my $var = bless [], 'Foo';

$var->method();
Foo::subroutine(undef, $var);
$var->funky();
__END__

However, only the first does;

$ perl warner.pl
Invocant at warner.pl line 31
$

This is because the implementation relies on @​DB​::args​:

if ($isobj) {
while (do { { package DB; $pkg = (caller($i++))[0] } } ) {
last unless @DB::args && $DB::args[0] =~ /^$category=/ ;
}
$i -= 2 ;
}

It simply can't work (ever) as documented for all but the simplest cases.
I don't like this state of affairs.

I suspect that the "useful" intent here is to warn based on the lexical state
at the point of the method call. That *can* be done reliably, using the
packages returned from caller.

Nicholas Clark

From @nwc10

On Wed, Feb 03, 2010 at 02​:30​:46AM -0800, Nicholas Clark wrote​:

warnings​::enabled is documented as this​:

=item warnings​::enabled($category)

Return TRUE if the warnings category, C<$category>, is enabled in the
calling module.
Otherwise returns FALSE.

=cut

Nick's a muppet. I meant to grab​:

=item warnings​::enabled($object)

Use the name of the class for the object reference, C<$object>, as the
warnings category.

Return TRUE if that warnings category is enabled in the first scope
where the object is used.
Otherwise returns FALSE.

=cut

Then this makes sense​:

However, it only works as documented in simple cases. All 3 of these should
warn​:

It simply can't work (ever) as documented for all but the simplest cases.
I don't like this state of affairs.

I suspect that the "useful" intent here is to warn based on the lexical state
at the point of the method call. That *can* be done reliably, using the
packages returned from caller.

On CPAN, it's only used by

Graphics-MNG/MNG.pm
Mac-iPod-GNUpod/GNUpod.pm
Object-Meta-Plugin/lib/Object/Meta/Plugin/Useful/Generic.pm
Text-Sprintf-Named/lib/Text/Sprintf/Named.pm

As best I can tell, none of the above would notice if the definition and
implementation changed as proposed.

accessors-fast/lib/accessors/fast.pm passes a parameter of $ME, but that's
set once in a BEGIN block to __PACKAGE__. (No, I don't understand this - $ME
isn't documented as something you can change, so why not keep the constant?)

perl_mlb/warnings.pm is a copy of the perl library.

warnings-compat/t/warnings.t tests warnings.pm

Nicholas Clark

From @cpansprout

Now (as of 5.10.0) that warnings.pm uses Carp to determine the call
site, should we not just switch objects over to using Carp, which does
more or less was I think the warnings code was intended to do?

The RT System itself - Status changed from 'new' to 'open'