Re: alloc/init - copy/mutableCopy

[David Ayers]

Richard Frith-Macdonald wrote:

> Perhaps we should abandon apple compatibility and make the abstract 
> methods implement shallow copies too?
I'm starting to think we should...  Anyone writing custom collection 
classes that don't implement shallow copies, will be subject to breakage 
of code using copy/mutableCopy as soon as objects that don't implement 
NSCopying are contained in the array or the code otherwise relies on 
standard shallow copying behavior.  It's not just an inconvenience for 
class cluster developers but for the "users" of class cluster objects 
when the semantics are undefined.  It's just not that simple to say "the 
exact nature of the copy is determined by the class" when the API 
suggests copy/mutableCopy are to be used to transform between mutable an 
immutable versions.  To not rely in shallow copies for copy/mutableCopy 
constrains the usefulness of these methods needlessly. 

I think GNUstep can take the lead here, and just maybe Apple will follow 
suit.  But most likely not because we did it, but because hopefully 
Apple developers will kindly inform them of the implecations.

Anyway, when we use our own classes explicitly (e.g. GC(Mutable)Array) I 
think it's perfectly fine to assume copy/mutableCopy to be shallow.  We 
just need to be sure that any arrays passed to us are converted to our 
classes before we send copy/mutableCopy, which in GDL2 and 
GC(Mutable)Array should generally already be the case.

Cheers,
Dave