Re: Forwarding and the GNU runtime one more time (was ANN: GNUstep 2.6.0 for openSUSE 11.4)

[David Chisnall]

On 22 Jun 2011, at 09:37, Nat! wrote:

> Well what I am trying to achieve, is just what I wrote (why doesn't anyone 
> seem to believe me :)). I want to lookup an implementation as fast as 
> possible and note if it is forwarding or not at the same time. If I have to 
> do it in two steps, it's just going to be a bit slower. But I'll probably 
> have to do that.

You can't do this with a single step in GNUstep, because the IMP that you 
actually get for forwarding is a closure generated with libffi based on the 
types of the arguments.  Note that this ignores the case where you're using a 
mechanism like -forwardingTargetForSelector: (which you can't currently with 
the GCC runtime, but can with the GNUstep runtime and the non-fragile ABI) that 
modifies the receiver but does a normal message send without constructing an 
NSInvocation.

The only reliable way of doing this is to first call 
class_respondsToSelector(), then call class_getMethodImplementation().  

Trying to combine the two steps is almost certainly a case of premature 
optimisation.  Why do you care if it's forwarded or not?  If it's likely to be 
invoking the forwarding mechanism, then it costs around 300 times as much as a 
normal message send, so avoiding one call is basically a waste of effort.  

David

-- Sent from my Apple II