Re: gc in GNUstep (Was: soname of gc variant of gnustep-base)

[Helge Hess]

Markus Hitter wrote:
> Each time you have two similar mechanisms for a goal, one of them is 
> obsolete.

Not to offend you, but this is complete nonsense ;-)

There are almost always multiple solutions to a problem with various 
pros and cons. GC has it's advantages (eg correct detection of circular 
structures) and RC has it's advantages (eg much better portability).

Personally I have decided that the advantages of RC are more important 
than the ones of GC (more specifically, compatibility with MacOSX 
Foundation)

> What if I retain a to-be-garbage-collected object? Can I release it? Or 
> do I have to look into the object's docs to find out what to do?

The semantics of retain/release are/should be exactly the same in GC 
mode. The semantic is "create a strong, self-contained reference to the 
receiver" (it isn't implemented that way in neither gstep-base nor 
libFoundation, but probably it should).

Because of this a valid RC program *always* runs in GC mode as well ! GC 
is just an addition which makes it a lot easier, especially for starters 
(they can learn to do assembler later on ;-).

> This is right. How about enhancing the debugging mechanisms around 
> NSZombie, -retainCount etc. ?

This is always useful but doesn't improve the initial implementation (eg 
of complex object graphs).

Greetings
  Helge