Re: [PATCH] Revert controversial apply_{args,result}_size change

[Eric Botcazou <ebotcazou at libertysurf dot fr>]

> I think that's fine.  I know Richard E. (who weighed in before) is
> travelling, so he may not be able to comment.  I'd be interested in hear
> from Eric B., as to whether he's comfortable with your approach.  (I
> think he's already done the SPARC back-end changes, consistent with the
> new definition, as has Richard E. for ARM.)

Given that it's precisely the solution I suggested in
  http://gcc.gnu.org/ml/gcc/2005-04/msg00299.html
I can't really disagree. :-)

However, as I pointed out in the above message, the macro is not referenced 
only in builtins.c, but in lcm.c and, more importantly, in rtlanal.c so I 
think there is the risk of pessimizing the common case.  This would need to 
be audited beforehand.

Moreover, given Eric's recent clarification:

  It would be, and I have no preference (other than that the current
  definition makes no sense if you look at it). My change was originally
  just going to be for apply_args_size, and does match the definition
  there - I just took it across to the similar function.

it appears that the change to apply_result_size was not necessary at all.

I'm, therefore, now of the opinion that reverting the change to 
apply_result_size (and only apply_result_size) is the best solution.

[Note that I haven't installed any SPARC back-end changes as of this writing, 
I'm waiting for your approval.  The patch is only linked to from the Wiki.
Note also that my change is only a workaround, I certainly don't want to 
modify SPARC's FUNCTION_VALUE_REGNO_P at this point.]

-- 
Eric Botcazou