[gfortran] PATCH arith.c, another range checking problem

[Tobias Schlüter]

Steve Kargl wrote:
> This range checking problem is probably a benign bug, but I
> thought we should fix it away.  The fix is obvious.

I applied this as obvious.

I'm wondering if this can ever trigger. Is there a combination of 
integer and real kinds where an integer can overflow a real? Or is there 
something else which could happen? Nevertheless, having the correct 
check in place seems more appropriate.

- Tobi