Re: [MIT-Scheme-devel] possible bug in min function

[Chris Hanson]

Sean D'Epagnier wrote:
> I have been using mit-scheme for nearest neighbor algorithms, and often
> I have a default "distance" of ieee754 infinity for a point.
> 
> I am looking to find the minimum, but when I execute the following:
> 
> 1 ]=> (min 1 (/ 1.0 0.0))
> 
> ;Value: #[+inf]
> 
> 
> This seems wrong, I have worked around it for now using something like:
> 
> 1 ]=> (min (exact->inexact 1) (/ 1.0 0.0))
> 
> ;Value: 1.
> 
> 
> Does this make any sense? is it a bug?  This is on a 64bit linux
> system with the c backend.  I tried it on a 32bit linux system, and
> I get a division by zero error.  Shouldn't both versions behave the same?

It looks like you've found two separate bugs.  The first is in the
handling of a mixture of an infinity and a non-flonum.  The second is
that the 64-bit platform isn't initializing its floating-point unit
correctly.  The correct behavior in all situations should be that an
error is signalled.