Re: [RFC] polymorphic arrays

[Mikael Morin <mikael dot morin at sfr dot fr>]

Le 06.08.2010 13:29, Janus Weil a écrit :
> Hi Thomas,
>
> > > 1) Use the class container also for polymorphic arrays, and have the
> > > "$data" field point to an array descriptor.
> >
> > I have to confess my ignorance here, but...
> >
> > Can different array elements have different types?  If so, this method
> > probably would not work.
>
> no, IIUC all array elements must have the same dynamic type. (Btw your
> comment applies in the same way to the second option I proposed.)
>
> To be honest, I have not found any direct statement on this in the
> standard. But I would not know how to set up an array which has
> different dynamic types, because:
> [...]
I came to the same conclusion.

About your question Janus, I agree with Daniel K. that (1)[class 
container around array descriptor] seems a better approach for the same 
reasons he gives (more code reuse and thus less invasive changes to 
introduce dynamic type arrays, and the possibility to keep the OOP stuff 
not too on the middle-end side).

If the extra dereferencing of the class container's data field is too 
much of a concern, one can think of making it a non-pointer (i.e. making 
it the array descriptor directly).

In any case [(1) or (2)], it seems to me that either (or both) of the 
array descriptor and the middle-end code generation have to be changed: 
the actual size of the array elements may not be known at compile time, 
and have to be kept either in the descriptor or the class container so 
that array elements accesses can use it.
One can keep the existing non-polymorphic code unchanged with (1) while 
with (2)[array descriptors as class containers] the array descriptor 
_will_ be altered (even slightly).

If we choose to go (2) I would like to see the change made to the 
existing scalar OOP code before introducing polymorphic arrays.

Be realistic, vote (1). [And don't forget to donate ;-)]

Cheers
Mikael