incomplete specification of the copysign function

[Michel Hack (1-914-784-7648) <hack@xxxxxxxxxxxxxx>]

Vincent LefÃ?vre wrote, aboy copySign() in draft 1.4.0 clause 6.5.1:
>   in the same format as x, but with the sign *bit* of y.

I agree.  Note that the cover paragraph does say "treating floating-point
numbers and NaNs alike" -- but you're right, NaNs have a sign bit, but no
sign (and are thus neither positive nor negative).

This also affects negate() and abs():

   negate(): ... reversing the sign bit.

   abs():    ... setting the sign bit to zero.
            (was: changing the sign to positive)

Michel.
Sent: 2007-07-31 15:58:23 UTC