Re: The orca, GNUstep's mascot

[Richard Frith-Macdonald]

On 29 Apr 2008, at 07:39, Graham J Lee wrote:

> Hi Gregory,
>
> On 28 Apr 2008, at 22:52, Gregory John Casamento wrote:
> > Damn... yes, I was thinking of TEST_RELEASE.
> [...]
> > Gregory John Casamento wrote:
> > > RELEASE is a #define which does the standard checking that needs to
> > > be done when doing a release.
>
> In what way needs to be done?  The scenarios for sending [obj  
> release] are:
> 1. obj is nil; nothing happens.
> 2. obj is not an id; crash (if you're lucky).
> 3. obj is an id which does not respond to @selector(release), and  
> has no forwarding mechanism; runtime exception.
> 4. obj is an id which responds to @selector(release) or consumes it  
> in some other way; it does its stuff.
>
> Case 1 is a special example of case 4, so case 2 needs to be tested  
> for and some might argue case 3.  But the macro is defined as:
>
> #define TEST_RELEASE(object)     ({ if (object) [object release]; })
>
> which doesn't provide that.
>
> Thanks,

RELEASE was introduced to completely remove the call to -release in  
the case when gc was enabled.  Apple have now effectively done the  
same thing by modifying the compiler to remove the call itsself... we  
went for a more portable solution, but as they control the compiler  
version being used, Apples decision makes a lot of sense.

TEST_RELEASE was added to optimise the common case of sending -release  
to nil when gc was not enabled ... the 'if (object)' test being much  
more efficient than a method call.

Neither optimisation is really important IMO, now that processors are  
*much* faster than they were when these macros were introduced.  The  
overheads of calling -release are generally a much smaller proportion  
of the overall running time of a program than they used to be.

I think the ASSIGN, ASSIGNCOPY, and DESTROY macros are still  
worthwhile having as conveniences (and because they tend to prevent  
some common coding errors).