new FAILs on HEAD

[Roger Sayle]

On Tue, 30 Dec 2003 gcc@ds217-115-141-84.dedicated.hosteurope.de wrote:
> These new FAILs
>
> FAIL: 23_containers/bitset/cons/1.cc (test for excess errors)
> FAIL: 23_containers/bitset/test/1.cc (test for excess errors)
>
> were caused by one of the following checkins

Damn!  This looks like its caused by my fold-const.c change.  I must
have tripped over a latent bug in the C++ front-end where the type of
a COND_EXPR didn't match the types of its second and/or third operands.
This is a documented requirement of GCC's COND_EXPR tree node.

Funny that this didn't show up during my bootstrapping/regression testing
last week, but I can confirm that I'm seeing it now.


I'm investigating further.  The fix from the middle-end is to either
check for VOID_TYPE_P specially, or locate where in the C++ parser
we're creating the dubious COND_EXPR.  Sorry for any inconvenience,

> 2003-12-29  Roger Sayle  <roger@eyesopen.com>
>
> 	PR fortran/12632
> 	* fold-const.c (fold) <COND_EXPR>: Don't fold a constant condition,
> 	if the type of the selected branch doesn't match its' parent.

Sorry again,

Roger
--