rehsack has asked for the wisdom of the Perl Monks concerning the following question:

Hi,

I'm seeking perl wisdom against my "OO wisdom" ;-)
But maybe my OO wisdom needs to be updated, too.

I've written an XS speedup for Params::Util - and the author detects that my XS implementation of _INSTANCE differs to his perl implementation: The perl implementation simply calls $obj->isa($class) - and returns the result. The XS implementation checks first sv_derived_from and only in a negative result, it calls $obj->isa($class) (I think, I should even prove string equality of sv_reftype - but this is another story). My reason to implement it this way was: 

class A { ... };
class B : public A { ... };

[download]
B is a A

So what do you think? Should the XS implementation give the (overridable) isa method the precedence? Or is it just a limitation? Should the perl implementation better use UNIVERSAL::isa before trying the overridden method?

Replies are listed 'Best First'.
Re: sv_derived_from vs. sub ISA
by chromatic (Archbishop) on Apr 04, 2009 at 17:25 UTC
    Should the XS implementation give the (overridable) isa method the precedence?

    Only if you want Perl's OO to keep working. In all other cases -- every one -- Perl calls overridden methods which optionally call super methods.

    (Modern versions of UNIVERSAL explicitly recommend against calling UNIVERSAL::isa directly.)

[reply]
[d/l]
      Should the XS implementation give the (overridable) isa method the precedence?

      Only if you want Perl's OO to keep working. In all other cases -- every one -- Perl calls overridden methods which optionally call super methods.

      (Modern versions of UNIVERSAL explicitly recommend against calling UNIVERSAL::isa directly.)

      So if I understand you correctly, it's recommented to do: 
      package A;

sub foo {};

package B;

sub bar {};

package C;

use vars qw(@ISA);
@ISA = qw(B);

sub foo {};
sub isa {
  return $_[1] eq 'A' or $_[0]->SUPER::isa(@_);
}

      [download]
      against: 
      package A;

sub foo {};

package B;

sub bar {};

package C;

use vars qw(@ISA);
@ISA = qw(A B);

      [download]
      Or at least it's an allowed and wanted to work way? Further: It's allowed and wanted to work that: 
      package C;

use vars qw(@ISA);
@ISA = qw(B);

sub foo {};
sub isa {
  return $_[1] eq 'A';
}

      [download]
      To prevent, C is no B?
[reply]
[d/l]
[select]

        Yes, that's allowable. It's probably not a good idea from the OO design standpoint, but it's allowable.

        (In your first case, I'd override UNIVERSAL::DOES instead, but that's a 5.10 feature.)

[reply]
[d/l]

Back to Seekers of Perl Wisdom