Re: getting same output as gfortran, long_float

[robin.vowels@gmail.com]

On Monday, May 4, 2015 at 5:21:28 PM UTC+10, Dmitry A. Kazakov wrote:
> On Sun, 3 May 2015 17:15:15 -0700 (PDT), r.nospam@gmail.com wrote:
> 
> > On Friday, May 1, 2015 at 5:45:23 PM UTC+10, Dmitry A. Kazakov wrote:
> 
> >> Actually REAL*16 is exactly portable.
> > 
> > No it's not.
> > Some compilers treat that is an error.
> 
> Portability applies to compilable programs only. You could not compile
> 
>    type X is mod 2**64;
> 
> either. That does not make it non-portable, IMO.
> 
> Portable means: The program exposes same [logically relevant] behavior on
> all platforms [of interest].

Portable means that a program that adheres to the language standard
is capable of running on any system where there is a compiler for that language.

There might be different limits for such things as maximum size of number on
different machines, but nevertheless the program is capable of running on
any particular system provided that it does not exceed such limits.

> A program that does not compile on a platform of interest is just an
> illegal program. It is neither portable or non-portable. We don't know yet.

But in this case, we know that the program is not portable,
because it violates the standard.

> > That form is non-standard.
> 
> Maybe. Though "FORTRAN" and "standard" sound silly in the same context.
> 
> > SELECTED_REAL_KIND or a similar modern form is how
> > precision may be specified in a portable manner.
> 
> Maybe, but not for the reason of being standard or compilable. The point
> was about specifying precision [and accuracy] in the problem space terms
> rather than in machine-specific ones.