Re: Changes in base

[David Chisnall]

On 13 Mar 2010, at 18:58, Stef Bidi wrote:

>  I really was missing that point, most because I did not know that was the 
> case!  Let me make sure I got this right, are all non-toll-free bridged CF 
> classes also a __NSCFType?  If that's the case, it works for me.

Yup, there are two issues.  All CoreFoundation types are (things that are not 
but look and behave exactly like) Objective-C objects.  They all inherit from 
__NSCFType.  A more interesting case is this:

        NSLog(@"%@", [[(id)s class] superclass]);

Which produces...

2010-03-13 19:01:09.460 a.out[44586:903] NSObject

So, __NSCFType is a class, which inherits from NSObject.  No, coincidentally, 
one of the examples from my upcoming book is a program that dumps the ivar and 
instance method names from all of the classes in the system.  What does this 
tell us:

NSCFType
        Ivars:
NSCFType
        Methods:
                finalize
                retainCount
                release
                retain
                hash
                isEqual:

These are largely irrelevant to us.  CoreFoundation has to exist when 
Foundation does not, so it uses its own reference counting mechanism.  CoreBase 
assumes the existence of base, so does not.  This means that we don't really 
need to bother with the __CFType class.  We can make CF classes inherit 
directly from NSObject.

For things that are not toll-free bridged, I suggest that you write them as 
Objective-C classes and use make_c_interface to dump the CFType version 
automatically.

David

> I guess that patch is invalid.
> 
> Stefan
> 
> On Sat, Mar 13, 2010 at 12:22 PM, David Chisnall <address@hidden> wrote:
> Hi Stef,
> 
> I believe you are missing something important here.  Although some 
> CoreFoundation types are not toll-free bridged with specific Objective-C 
> classes, they are all __attribute__((NSObject)) and all implement (at least 
> some of) the NSObject protocol.  You can send -retain / -release to all of 
> them, including things like CFSocket that are not toll-free bridged:
> 
> $ cat cf.m
> #import <Cocoa/Cocoa.h>
> #import <CoreFoundation/CoreFoundation.h>
> 
> int main(void)
> {
>        CFSocketRef s = CFSocketCreate(NULL, 0, 0, 0, 0, 0, 0);
>        printf("%d\n", (int)CFGetRetainCount(s));
>        [(id)s retain];
>        printf("%d\n", (int)CFGetRetainCount(s));
>        printf("%d\n", (int)[(id)s retainCount]);
>        NSLog(@"%@", s);
>        NSLog(@"%@", [(id)s class]);
>        return 0;
> }
> Liberator:tmp theraven$ gcc -framework Cocoa cf.m && ./a.out
> 2
> 3
> 3
> 2010-03-13 18:20:07.933 a.out[44021:903] <CFSocket 0x100111290 
> [0x7fff704fbf20]>{valid = Yes, type = 1, socket = 3, socket set count = 0,
>    callback types = 0x0, callout = ??? (0x0), source = 0x0,
>    run loops = <CFArray 0x100111360 [0x7fff704fbf20]>{type = mutable-small, 
> count = 0, values = ()},
>    context = <CFSocket context 0x0>}
> 2010-03-13 18:20:07.978 a.out[44021:903] __NSCFType
> 
> 
> As you can see, the object responds to -retainCount, -retain, -description 
> and -class.  It will also respond to things like -hash - you can store all 
> CoreFoundation types in Objective-C collection classes (they are also Core 
> Foundation collection types) transparently, for example.
> 
> David
> _______________________________________________
> Gnustep-dev mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/gnustep-dev


-- Sent from my IBM 1620