------- Comment #5 from sgk at troutmask dot apl dot washington dot edu
2007-01-22 20:16 -------
Subject: Re: MAXVAL() incorrect for zero-size int arrays, and for -HUGE-1
maximum values.
>
>- Comment #3 from fxcoudert at gcc dot gnu dot org 2007-01-22 07:56 -------
> I don't think it's a bug, since "the negative number with the largest
> magnitude possible within the representation" is not -huge()-1, but -huge().
>
> If I understand the standard correctly, -huge()-1, although being
> representible by the hardware you have, is not "within the representation"
> of this integer kind, because the range of the representation of an
> integer kind is supposed to be symmetric.
>
> For what it's worth, the Intel and Sun compilers have the behaviour
> you expect, but the Portland compiler and g95 both have the same
> behaviour as gfortran.
>
> Steve might have an idea on that, IIRC he's the one who implemented the
> -frange-check option. Otherwise, a question to comp.lang.fortran would be a
> good thing.
>
I fixed the range checking from [-huge()-1: 2*huge()] to
[-huge()-1: huge()]. The old upper limited would allow
people to write
integer(1) i
i = -128_1 ! This should be -huge(i) - 1_1
end
because the 128_1 was not flagged as overflowing a integer(1).
Note, -128_1 is a unary minus operating on the 128_1.
When I changed the range checking or shortly after, jerryd
introduced the -f[no]-range-check option.
For the case at hand, the integers do not need to be symmetric,
and so minval and maxval should properly handle -huge()-1. As
pointed out by Tobias,
integer(1) a(2)
a = -huge(a) - 1
print *, minval(a), maxval(a)
end
should print "-128 -128"
Now, where the symmetry of the integers becomes a problem is
integer(1) i
i = - huge() - 1
i = abs(i) ! This is invalid.
end
The above is prohibited in 13.14:
The types and type parameters of intrinsic procedure arguments
and function results are determined by these specifications. A
program is prohibited from invoking an intrinsic procedure
under circumstances where a value to be returned in a subroutine
argument or function result is outside the range of values
representable by objects of the specified type and type parameters.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30512
