[testsuite] fix C tests to use correct dg-error or dg-warning directive

[Mike Stump]

On May 15, 2007, at 11:34 AM, Janis Johnson wrote:
> Several C compiler tests use dg-error for warnings and vice versa.

Sounds about right.  Kinda surprising we get as close as your patch  
suggests.

> This currently makes no difference because the DejaGnu support  
> doesn't pay attention to the message prefix, but Manuel and I are  
> working on a patch to change that within the GCC testsuite support;  
> see PR 25241.

Wonderful.  I'd vote against dg-gcc-warning, once done, it would  
never be undone.  :-(   What's your current thinking, one language at  
a time and not doing dg-gcc-warning?

> I've looked at the options used to compile these tests and the  
> behavior of earlier versions of GCC, and the changes look  
> reasonable to me.

I reviewed them all, and I'd say that I think they are ok.

Does anyone remember the testsuite code before the expect code?  Did  
it handle the difference between warnings and errors?  For some  
strange reason I was thinking this was a regression when we switched  
to expect?  Rob or Julia might remember.

Thanks.