At 12:15 AM 1/10/2002 +0000, Tim Bunce wrote:
>On Wed, Jan 09, 2002 at 04:42:51PM +0000, Graham Barr wrote:
> > On Tue, Jan 08, 2002 at 06:38:02PM -0500, Dan Sugalski wrote:
> > > ># Attributes are done as a hash of hashes. Each interpreter has a
> > > ># pointer to an attribute hash, whose keys are the attribute names. The
> > > ># values will be hash pointers. Those hashes will each have a key which
> > > ># is a PMC pointer (hashed up somehow) and the value is the attribute
> > > ># value.
> > > >
> > > >If you're talking about 'is'-style attributes, why not have them be
> > > >attached to the PMCs themselves?
> > >
> > > D'you want an attribute pointer for each PMC? (I've considered it,
> but for
> > > the moment I think it'll be too expensive. Might be wrong)
>
>This sounds rather like perl5's optional list of arbitrary 'magic'.
>Can't PMC's be 'upgraded' so only carry the overhead if needed?
Not easily. PMC structs are fixed size, unlike perl's variable-size SVs. I
was hoping to keep attributes as OOB data, since I'm not sure it'll be used
enough to be worth being in-band. However...
> > On the other hand when a variable is garbage collected you have to walk
> the hash
> > of hashes to clear out the attributes. Also what happens when you move
> the PMC
> > during garbage collection, will the key in the hash change ?
>
>Good points.
This is a darned good point. Good enough to dump my OOB argument, I think.
> > A PMC may have a flag to show if it has any entries in the global table,
> > but it is probably just a trade off between space and performance.
>
>Tim [who's not really been paying attention, so ignore me if I'm being daft].
Nah, you're making sense. Besides, vtables are all your fault in the first
place, so I ought to be listening... :)
Dan
--------------------------------------"it's like this"-------------------
Dan Sugalski even samurai
[EMAIL PROTECTED] have teddy bears and even
teddy bears get drunk
