Re: alloc/init - copy/mutableCopy

[David Ayers]

Richard Frith-Macdonald wrote:

> On Thursday, February 6, 2003, at 04:35  pm, Stéphane Corthésy wrote:
>
> > Hello,
> >
> > I'm the author of these modifications. Please keep them, as they 
> > protect us from changes in mutableCopy/copy implementation in 
> > gnustep/base.
Yes I know that's why I cc-ed you :-)

> I think perhaps you got left out of the conversation where David (I 
> think) pointed out that this policy of Apples is really broken. 

I generally try to keep these discussions public:
http://mail.gnu.org/archive/html/bug-gnustep/2003-01/msg00099.html

But this one kind of evolved out of side issue :-(..but I think it's 
time to hit the list.

> I'm not sure how to proceed ... At the moment our concrete collection 
> classes all perform shallow copies, but I think our abstract 
> implementations work as the Apple documentation says.  This means that 
> in practice, while we conform to the Apple documentation, we are 
> consistent ... however anyone subclassing a collection class will need 
> to override the copy methods to get consistent behavior.  Perhaps we 
> should abandon apple compatibility and make the abstract methods 
> implement shallow copies too?

I've also been tossing and turning on this one.  On the one hand we 
*force* GNUstep developers to write compatible code, on the otherhand, 
we confront people playing with subclassing class clusters with 
unnecessary issues due to a broken API.  And as in this case write 
overly complicated code.  I'm not sure which way to go.  But what ever 
we do, we should document it well and use consistent techniques to 
create mutable/immutable copies.

Cheers,
Dave