(j3.2006) Interval arithmetic

[Malcolm Cohen]

Van Snyder wrote: A whole load of stuff that is complicated, irrelevant, 
and probably incorrect.

The stuff about intrinsic vs. known-to-the-compiler was not a trivial 
detail at the time; and continues to mislead.  Deliberately conflating 
stuff that misleads might be thought as an indication that the arguments 
are not in good faith.

But anyway, we started here with some grandiose claims about something 
being "impossible".  Those claims have not been substantiated.  When 
challenged, they were supported by claims about the x86 hardware being 
deficient that were factually incorrect.

Furthermore, this continues to fall back on the subsidiary claim about 
it being "impossible" *for us to do* because of an answer we gave to an 
interp for a previous revision of the standard.  That's obviously 
flat-out wrong.

I see no point in continuing a discussion with so many counterfactuals.  
Life is too short for me to spend it repeatedly deconstructing the same 
mistakes.

Sure go ahead, go around telling people that it's impossible for Fortran 
to do interval arithmetic except via an intrinsic type.  Just don't 
represent it as the considered view of the Fortran committee or of the 
Fortran vendors.

(As Keith pointed out there are advantages to do with "sharpness" that 
one can gain by compiler knowledge, and I would agree with that; but 
that is a far cry from the notion that "register spills make it 
impossible to obey 754".)

Cheers,
-- 
......................Malcolm Cohen, Nihon NAG, Tokyo.