[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [Axiom-developer] differences

From: Bill Page
Subject: RE: [Axiom-developer] differences
Date: Sun, 1 Apr 2007 23:38:23 -0400

On April 1, 2007 11:16 PM Tim Daly wrote:
> ... 
> (cos (acos 0.7)) on Redhat is 0.69999999999....96
> (cos (acos 0.7)) on Fedora is 0.70000000000...07
> ...

Earlier Gaby wrote:

> It will help me if you can say more about exactly what is
> being tested.  I see DoubleFloat computations, from what I
> discussed two or three days ago, the underlying representation
> is Lisp' long-float -- which I think is bogus.  That
> representation tends to differ from platform to platform
> along with the accuracy.  I believe double-float is more
> portable with predictable accuracy across platforms.  Also,
> there are platform-dependent libm variations to factor in
> (plus C compiler optimizations).  
> ...

I think Gaby has a point. Isn't this difference within the
accuracy expected for the representation used by Axiom?

Perhaps this demonstrates a potential problem for regression
tests that are based only on text file diffs. In many cases
it would be better to write specific comparison test code
in Axiom itself that would compare the result to the expected
values using methods appropriate to the domain.

Bill Page.

reply via email to

[Prev in Thread] Current Thread [Next in Thread]