Re: [RFC/PATCH] Fix PR middle-end/17909, ICE in verifiy_stms

[Andrew Pinski <pinskia at physics dot uc dot edu>]

On Oct 11, 2004, at 12:59 AM, Jason Merrill wrote:

> On Sat, 9 Oct 2004 23:57:02 -0400, Andrew Pinski 
> <pinskia@physics.uc.edu> wrote:
>
> > This patch fixes that by copying some code from builtins.c into 
> > gimplify.c.
> > We could make an fold_builtin_va_start which does this, if someone 
> > wants
> > me to do this, just ask.
>
> Please.


Actually there is already a function which does this,
fold_builtin_next_arg.  I exported it and made it return true if
a warning (undefined behavior) or error (fatal problem) so that
we don't get two or more warnings/errors for the same problem.
I also changed it so every place where we used fold_builtin_next_arg
before, we would just return const0_rtx (because these were in the
expander).  There is still some clean up possible in builtins.c
but that is for another day.

OK? Bootstrapped and tested on powerpc-darwin with no regressions.

Thanks,
Andrew Pinski

ChangeLog:

	* builtins.c (fold_builtin_next_arg): Export and return true
	when there is a warning or an error.
	(expand_builtin_va_start): When fold_builtin_next_arg returns true,
	return const0_rtx.
	(expand_builtin): Likewise.
	* gimplify.c (gimplify_call_expr): Error out if there is not
	enough arguments to va_start. Call fold_builtin_next_arg also
	on the second argument.

Attachment: temp.diff.txt
Description: Text document