At 11:26 PM 8/13/2001 -0700, Kurt D. Starsinic wrote:
>On Mon, Aug 13, 2001 at 06:45:30PM -0400, Dan Sugalski wrote:
> > At 02:59 PM 8/13/2001 -0700, Kurt D. Starsinic wrote:
> > > It doesn't seem Perlish to me for the interpreter to provide
> _special_
> > >support for HAS-A. HAS-A is too general of a design pattern to warrant
> > >pigeonholing into a particular object relationship.
> >
> > IS-A is in the same boat, as are object attributes. Given that those are
> > already interesting problems, it seemed reasonable to include interpreter
> > support for at least some HAS-A relationships for automatic delegation and
> > suchlike things. I'm not proposing we jam support for all HAS-A things,
> > just a few of the more common ones. Or if that's not feasable, then a
> > design that has a reasonable hook so classes can intercept this for
> > themselves somehow.
>
> ISA is a much more specific relationship than HAS-A, IMHO. It is,
>_per se_, about method inheritance -- and I think you'd be hard pressed
>to name a class-based OO language that implements method inheritance more
>generically or less restrictively than Perl. Now that we have NEXT.pm,
>the only thing I can _imagine_ asking for is the ability to walk the @ISA
>tree other than depth first (and I'm not losing sleep over that).
Even for ISA relationships it's not simple. You have to deal with:
search order
first found or all found methods
Whether multiply inherited base classes get methods run multiple times
It's also fun with properties, since in some scenarios, this inheritance
scheme:
A
/ \
B C
\ /
D
will have two separate, *independent* sets of A properties. (The set
inherited via B and the set inherited via C) Or so Damian tells me, and
I've no reason to doubt him.
> > >HAS-A can indicate
> > >delegation, brokering, snooping, containing, proxying . . . and there are
> > >many unique and appropriately distinct implementations of any of these
> > >relationships. I really like Perl's current support for HAS-A: instance
> > >variables. :^)
> >
> > Sure. But automatic delegation's faster than executing interpreter code to
> > provide the same functionality, or it ought to be at least.
>
> I don't think we can have automatic delegation, because the interesting
>part of delegation is the deciding-when-and-what-to-delegate and deciding-
>what-to-do-with-return-values part. If you think differently, I'd love to
>hear your ideas for a Generic Delegation Model.
I don't have one really--I do very little OO programming. I'm not looking
to have a Grand Unified Delegation Theory implemented here, as much as a
set of commonly used delegation things that we can provide the equivalent
of Hardware Support for.
There might be nothing beyond a hasa opcode that just returns true/false if
the class is defined as having something, and delegating to a class method
from there or something. That's OK. (OTOH, if we provide a generic,
inheritable HASA method, there ought to be a single opcode to dispatch it
for those classes that don't do magic)
>In Class::Delegate, I had
>come up with what I thought was a generic implementation (influenced heavily
>by NeXTstep's AppKit), but then I saw Damian's Class::Delegation talk at TPC.
I saw Damian's talk too. He scares me sometimes. :)
> > I'm not soliciting a comprehensive solution, nor even really a
> > perl-specific one, but more a general "Most HAS-A situations need X, Y,
> and
> > Z" sort of thing.
>
> HAS-A situations need pointers/references, probably the ability to
>twiddle the insides of the objects they HAS, and maybe two-way pointers
>that connect the HAS and IS-HAD objects. I think that's it.
>
> It's not clear to me whether you're asking about HAS-A relationships
>in general, or delegation relationships in particular, but I've already
>said my piece on delegation. As to the other HAS-A relationships I'd
>listed above, I think they're even less amenable to canonization than
>delegation relationships.
Delegation mainly. Other than that I don't much care, as I don't see
anything the interpreter can provide for support.
Dan
--------------------------------------"it's like this"-------------------
Dan Sugalski even samurai
[EMAIL PROTECTED] have teddy bears and even
teddy bears get drunk
