Re: bignum branch

[Paul Eggert]

On 08/03/2018 01:17 PM, Tom Tromey wrote:
> I'm not sure what the semantics of NaN comparison are in Emacs.  In
> particular, this doesn't really make sense to me:
>
>     (> 0 0.0e+NaN) => nil
>     (< 0 0.0e+NaN) => nil
>     (min 0 0.0e+NaN) => 0.0e+NaN
>     (min 0.0e+NaN 0) => 0.0e+NaN
>     (max 0 0.0e+NaN) => 0.0e+NaN
>     (max 0.0e+NaN 0) => 0.0e+NaN

NaNs never compare numerically equal to, less than, or greater than any
other floating-point value, even other NaNs. This is part of the IEEE
standard.

min and max propagate any NaNs they find.

By the way, we've changed master so that eql now looks at NaN's
significands. That is, (eql x y) now returns t if x and y are NaNs
containing identical significands.