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

[Mark Mitchell <mark at codesourcery dot com>]

Eric Botcazou wrote:
> Hi,
>
> As explained in http://gcc.gnu.org/ml/gcc/2005-04/msg00299.html, the change
>
> 2004-03-17  Eric Christopher  <echristo@redhat.com>
>
> 	* builtins.c (apply_args_size): Use reg_raw_mode.
> 	(apply_result_size): Ditto.
>
> replaced the heuristics that had been used since the very beginning to compute 
> the mode of arguments and return value, by reg_raw_mode. 

I've read through the various threads, and, from what I can see, it does 
indeed look to me like the change above was a mistake.  I agree with 
Richard Earnshaw's comment that this is a change that would, at the very 
lease, require changing a number of back ends.

However, I don't think it's appropriate to revert the patch on the 4.0 
branch now, even as the prerelease bits are being built.  As the patch 
went in a year ago, and the reversion suggestion has only been recently 
raised, it seems unlikely that this SPARC bug is causing incredibly 
severe problems.  Because this change does affect all platforms, 
changing it could have unforseen consequences.  Making any change now 
will require a second prerelease, and inevitably delay the release, 
which I'm not eager to do.

For mainline, we need to use the usual reversion policy.  I believe the 
patch should be reverted.  If a second person steps up, then we can 
start the 48-hour clock, obligating Eric C. to either fix the SPARC 
problem or agree to have the patch reverted.  Assuming the patch is 
indeed reverted on mainline, then that change is also pre-approved for 
4.0.1.

--
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304