[PATCH] Fix PR21595

[Roger Sayle]

On Mon, 16 May 2005, Richard Guenther wrote:
> 2005-05-16  Richard Guenther  <rguenth@gcc.gnu.org>
>
>	PR middle-end/21595
>	* builtins.c (fold_builtin_constant_p): Handle
>	&"string cst"[0] as constant.
>
>	* g++.dg/tree-ssa/builtin1.C: New testcase.

This is OK for mainline, and after a few days for the gcc-4_0 branch
as it is a regression from 3.4.

We should be able to consider array references other than 0 to also
be constant, but I appreciate there are middle-end representation
issues that may complicate things.  "foo" + 0 = "foo", but "foo" + 1
!= "oo", when expressed in terms of STRING_CST.


> int main()
> {
>	if (! __builtin_constant_p (&"Hello"[0]))
>		link_error();
> }

My maintainer-sense is tingling that it may be preferable to add
a "return 0;" at the end of main in this testcase.  I can't quite
put my finger on why, as this is a link rather than an execution
test, but given that exiting main without a return value isn't a
part of this test, there's no reason to anger the coding style gods.
This tweak to the testcase is preapproved.

Thanks,

Roger
--