Strange CPP failures

[Neil Booth]

Jason R Thorpe wrote:-

> These failures happen for me on i386 native, as well.  It seems to
> be related to building the new compiler with the native NetBSD
> 2.95.3-based compiler; a native stage3 compiler from "make boostrap"
> doesn't have these failures.

Sigh.  There's been a bunch of problems recently with GCC compiled
with earlier GCCs, but not with existing GCCs.  *shrug*.  There
must be a bug somewhere, or a long-standing code generation bug
in older GCCs that new ones have exposed.  It has also been
suggested it might be binutils related.

> A bunch of these tests that fail are testing multiple things in one
> test, e.g.:
> 
>  > FAIL: gcc.dg/cpp/arith-1.c  (test for bogus messages, line 92)
> 
> is:
> 
> #if (+1 - 2) > 0 || (+1U - 2) < 0
> # error         /* { dg-bogus "error" }  */
> #endif
> 
> In addition to the combined test, I think it'd be useful to have two
> additional tests that check each part separately; it certainly might
> help figure out what the failure is.
> 
> In any case, I'm going to try and start hunting this down today (with
> any luck) ... but if you have any advance insight about what the problem
> might be before I dive in, I'm all ears :-)

No idea, sorry.  Let me know how you get on.

Neil.