Some way of indicating XSUB would be nice

[ribasushi]

(*) perhaps?
An XSUB can be detected like this

[kentfredric]

Is there a specific way I should obtain references to subs for best results?

Given I'm presently just doing string mangling from mro:: results.

[kentfredric]

well, not just that, I'm using Package::Stash ... to get symbol names. I guess I could factor that better so I can just reuse that object. :)

[ribasushi]

Is there a specific way I should obtain references to subs for best results?

The canonical thing is ->can. That will unambiguously return the "thing perl would call". You can also rely on P::S, which will return the right thing almost in all cases, unless there is a stale mro cache entry (extremely rare).

[kentfredric]

Just I'm worried that with ->can, I'm going to wind up in a situation where Package::Stash reports "Namespace::sub" exists, but its a stale cache entry, so the data says "I have a sub", but the information resolved by ->can() instead refers to the parent classes sub due to can invoking Perl MRO lookup mechanics.

At least if I use Package::Stash consistently everywhere, I'm going to be consistent in what I'm reporting.

That way the stale MRO cache entry will resolve as "no" for an XSUB check, but its parent will return a "Yes" for an XSUB check ( which is desirable ), instead of both returning "Yes" for an XSUB check ( due to resolving to the same coderef )

[ribasushi]

Correct, this is why I recommended in the very beginning you use ->can throughout the entire stack. If Child->can("foo") == Parent->can("foo"), then either "foo" is inherited, or it has been inserted into the symtable via *Child::foo = \&Parent::foo. That difference is immaterial to isa-splain.

Additionally using ->can this way allows you to properly report the ancestry of these weird-ass cases.

[kentfredric]

Ah, so you mean perform secondary validation on the results of Package::Stash->get_all_symbols('CODE') and skip refs that are unmodified from their parent?

K. Simple enough.

[kentfredric]

This logic is confusing me:

Child->can("foo") == Parent->can("foo")

How do we determine what "Parent" is in this equation?

Do we:

a: Simply perform the same query on all @ISA, ie:

     foreach my $parent ( @{$child::ISA} ) {
             if ( $child->can('sub') == $parent->can('sub')  ) { 
                   ## sub is a cached node, skip it
             }
     }

?

Or do we have to probe linearized isa?

[ribasushi]

Forget directly looking at the @ISA variable entirely (it won't give you a sensible answer on MI codebases). mro::get_lineari_isa is the only way.

And yes - you check each class in a hierarchy independently, and then "recombine" the result: in essence replicating what the p5 OO dispatch does.

[kentfredric]

So is the correct approach:

     foreach my $parent ( linearised_isa($child) ) {
             next unless $parent->can('sub');
             last if $child->can('sub') != $parent->can('sub');
             skip('sub');                   ## sub is a cached node, skip it
     }

Sorry, I'm extra dumb today.

[ribasushi]

Yep, you got it (though the skip + comment confuses me, but that's likely a leftover from something else)

[ribasushi]

Though note that mro::get_linear_isa($class) always returns $class at the first position, so you can skip checking that

[kentfredric]

I mean "skip" means "sub foo does not actually exist (as far as our purposes are concerned) in $child, so don't show it in child" . Though I might have been side-tracked by the whole Package::Stash debacle.

[ribasushi]

I think you are getting things confused. Let me give you a bird-eye view of what the proper way would be taking in consideration all of MI, 5.8 mro, and mro in general:

You use P::S once in the entire program like so:

~$ perl -MPackage::Stash -MDevel::Dwarn -MDBIx::Class::Core -e '
  my $all_found;
  $all_found->{$_}++ for map
    { Package::Stash->new($_)->list_all_symbols("CODE") }
    @{mro::get_linear_isa("DBIx::Class::Core")}
  ;
  Dwarn $all_found
'

Then you forget P::S exists, and use the "->can crawl" to actually figure out who came from where.

Does this make more sense?

[kentfredric]

Not hugely, I think my brain has dieded.

use the "->can crawl"

^^^ this is still where I get lost.