Re: [PATCH] New option to support asymmetric integer ranges in gfortran

[Scott Robert Ladd <coyote at coyotegulch dot com>]

Tobias Schlüter wrote:
> So what you're suggesting is that integer ABS should explicitly compare
> its argument against -2**31 everytime it's called? I think that's a
> really bad idea.

Well then, we're back to one of two situations:

1) Have the compiler allow -2**31 as a constant, even though some (but 
not all) Fortraners think this violates the Standard.

2) Have the compiler reject -2**31 as a constant, as it does now.

Under option 2, we have two choices:

  a) Reject compatibility with compilers and languages that allow -2**31

  b) Implement the switch from my patch, which allows a programmer to
     explicitly say "ignore the standard" for the sake of
     interoperability.

As we move toward Fortran 2003, we'll need to consider whether an 
integer(C_INT32_t) allows -2**31, given that such a value *is* valid in 
Standard C.

> If the compiler can detect something like this, have it issue an error.
> Otherwise, don't hurt the people that don't do such things. If someone
> thinks that this is an error that should be caught by the library, one
> can add a compiler flag that makes it link against a library with
> complete error checking. But this should by no means be the default
> behavior.

I still think my original solution (the -fasymmetric-integers switch) 
works just fine. It allows gfortran to follow a strict interpretation of 
the standard, while allowing those who need it to use constants like 
-2**31. No one gets hurt, no overhead is imposed on anyone, and we can 
all happily compute our way to happiness.

Or somethign like that. ;)

--
Scott Robert Ladd
Coyote Gulch Productions (http://www.coyotegulch.com)
Software Invention for High-Performance Computing