Re: ObjectiveC accessors

[Helge Hess]

On May 31, 2004, at 11:59, Marcel Weiher wrote:
> At an application-specific level, yes, just add Boehm's conservative 
> GC.  At a general library/language level:  not a good idea, and not 
> really possible while preserving existing language semantics.

I agree that this is not necessarily a good idea ;-), but notably 
Boehm-GC can be used as a drop-in replacement for the malloc library 
(without altering semantics!) and - which is often forgotten - it 
actually does provide a free() method, so for code which requires that, 
you can release memory at a specified time.

Some other things which are often "forgotten" to discourage Boehm-GC 
;-):
- autorelease also defers deallocation to some unknown time, pretty much
  comparable to GC, only that Boehm-GC is pretty smart and supports 
generational
  collection etc
- the GNU runtime provides typed object memory, so Boehm-GC is actually 
supported
  at the language runtime level! (so its non-conservative for objects)
- GC is no either or, eg for DO you would still use the regular RC 
(which of
  course is hidden to a GC environment by the proxy, just like in RC)
- you can still implement -retain in such a way that it adds a root 
reference,
  so GC/RC behaviour would be actually similiar to regular RC, ie, 
would also work
  for obscured pointers (like those passed to TCL as string references 
;-)

No intention to start one of the GC is good or not flame wars, feel 
free to look into gnustep-dev or macosx-dev archives for those, a lot 
of them should be available ;-)
Just wanted to point out some non-obvious things about the Boehm-GC 
implementation which actually can provide very good compatibility with 
standard C stuff.

Greets,
  Helge
--
http://docs.opengroupware.org/Members/helge/
OpenGroupware.org