Re: I think I figured out how Apple did their toll-free bridging.

[Maxthon Chan]

That doesn't matter if we keep internal consistency.

That is, the memory management of CoreFoundation objects are all done via 
bridging to Objective-C code, no matter if it is toll-free or not (For 
non-toll-free objects, the private Objective-C class corresponding to the 
CoreFoundation object is not complete, and does not have a public interface.)

If we can do the toll-free bridging in this fashion, our code can be a lot more 
compatible to Apple code. However, this may require promoting CoreBase from 
dev-libs into core, merging ti with Base. (that is, building Base will result 
in two libraries, libCoreBase and libgnustep-base, libCoreBase linked to 
libobjc2, and libgnustep-base linked to libCoreBase.)

And about the missing CoreFoundation features that are already implemented in 
Base, how about moving code after this promotion?

在 2013-5-18，上午5:55，David Chisnall <address@hidden> 写道：

> On 17 May 2013, at 17:48, Chan Maxthon <address@hidden> wrote:
> 
>> All CoreFoundation objects places the field for retain count at the same 
>> offset as Cocoa objects do
> 
> Cocoa objects do not have an in-object retain count.  This is why retain / 
> release on GNUstep are so much faster than on OS X / iOS.  On iOS, they are 
> all stored in a C++ map table (in fact, the LLVM one).  In OS X, there is an 
> initial hash that selects one of 8 map tables, for greater parallelism in the 
> retain / release operations.
> 
> David
> 
> -- Send from my Jacquard Loom
>