Re: ffcall on Darwin/PPC

[Wolfgang Lux]

Lars Sonchocky-Helldorf wrote:

> Did you file a bug with gcc on this?
>
> http://gcc.gnu.org/bugzilla/

Andrew Pinski wrote:

> You know I have not see any bug reports from you about libffi.  In
> fact I rather have GNUStep always use libffi because it is more
> supported with more targets (and the semi official base code is inside
> GCC).  I am about to review the patch which used libffi instead of
> __builtin_apply (etc.) inside libobjc also.

Looks like you misunderstood my email (okay, that's what I deserve for
writing mails on saturday evening :-). I do not think there is a bug in
gcc or libffi. It is even worse than that: I was thinking that libffi
is inappropriate or at least is used inappropriately for implementing
message forwarding in GNUstep.

The point is that the implementation of message forwarding in  
GSFFIInvocation.m
copies the incoming message arguments into a malloc'ed buffer. In order
to do so, it somehow needs to know the size of the argument frame. In
order to get this information, GSFFIInvocation uses the message  
signature
supplied to the callback and falls back to the signature of a method  
with
no arguments and an id result if the selector passed to the callback  
cannot
provide type information.

Now, I'm not saying that it is the fault of either libobjc or gcc  
that the
exact signature type is not available. There are situations where it is
difficult or maybe even impossible to determine this information. Have a
careful look at the example from Issac Trotts' posting:
 http://lists.gnu.org/archive/html/discuss-gnustep/2006-04/ 
msg00107.html
The receiver of the message is a proxy and does not even declare the  
methods
it is forwarding. I may be wrong, but to me it looks hard to  
determine the
right message signature here (even with Andrew Ruder's patch).

Compare this with the implementation of message forwarding in
GSFFCallInvocation.m. The code appears to use only information about the
return type of a message, so it can handle missing type information more
gracefully in a lot more cases than GSFFIInvocation. This led me to the
(admittedly provocative) question, why GNUstep needs to maintain two
different implementations of message forwarding.

Regards
Wolfgang