Re: [patch,fortran] Fix PR19310 unnecessary error for overflowing results

[Jerry DeLisle <jvdelisle at verizon dot net>]

Dominique Dhumieres wrote:
> A few comments about the test case:
>
> (1) is not 1.0/-0.0 illegal? If yes, it does not seem to be caught by gfortran;
> (2) in the first release of the test a = exp(2000.0)/exp(1000.0) should
> give a NaN and not +Infinity as in the following test:
>
> real a, b
> b=huge(a)
> a=exp(b)
> b=exp(b/2.0)
> print *, a, b, a/b, b/a
> end
>
> you can also consider the case a = exp(1000.0)/exp(2000.0).

I did look at that too. :)
> Dominique
This is interesting, I tried several combinations just to explore what the 
results would be and +Infinity is what mpfr gives in that case (2).  As far as 
(1), same thing, it is what mpfr gives.  We will have to explore this a bit.

Regards,

Jerry