Re: Is +[NSMutableArray initialize] completely implemented?

[David Ayers]

Richard Frith-Macdonald wrote:
> On 22 Feb 2004, at 11:56, Philip Mötteli wrote:
>
> > > > > and that [NSMutableArray+initialize] is called before any other 
> > > > > method of the class, so I don't know how -initWithCapacity: could 
> > > > > be called without NSMutableArrayClass being initialised.
> > > > > or step through in debug yourself to find out what is going on?
> > > >
> > > >
> > > > I set a breakpoint in both +initialize, in NSMutableArray and 
> > > > NSArray and NSArray's initialize is clearly not called.
> > >
> > >
> > > Then I guess for some reason the runtime does not think that 
> > > NSMutableArray inherits from NSArray ... why not?
> >
> >
> > I found the problem: I have a +initialize method in a category of 
> > NSArray. So the runtime system called only this one.
>
>
> Interesting ... it's worth trying to remember that problem because it's 
> not at all obvious.
>
> Perhaps we should have some coding hints somewhere to say things like 
> ... 'don't override +initilize in a category'

This sounds reasonable... I also fell into that trap before because I 
believed to have read that +initialize would be called for every 
category.  But I had confused that with +load which /does/ get called 
per category.

> Actually, it's not clear to me whether the runtime should try to detect 
> this situation and call the class implementation before calling the 
> category implementation ... probably not , as I think that categories 
> are generally supposed to be responsible for handling things like that. 
>  Also, what if two or more categories override +initialize etc.  As far 
> as I know there is no standard way to resolve the situation where 
> multiple categories override a method.

The +load mechanism works because the runtimes invokes it directly (i.e. 
via implementation pointer when loading the category and not via message 
lookup.)  I don't think we should try doing the same for the current 
runtimes wrt +initialize.  Now whether a future unified runtime had 
similar mechanism that invokes a standardized method per Class/Category 
after all runtime initialization has been done, is a different issue.

> Perhaps GNUstep should establish some conventions for this and perhaps 
> provide a few runtime helper functions to assist in managing this sort 
> of thing.
> On the other hand, it's rarely an issue ...

That's something I think is worth pondering about.

Cheers,
David