Arnold Neumaier wrote:
But do you really support Van Snyder's view that conversion of floats
should treat the latter as inaccurate?
No, I do not. We have no idea what the user intends when
a float is presented to the conversion operator. I feel
the best course would be to treat it as exact. If it's
clear in the standard, then users can go from there,
and compensate if they know just how inexact it is.
If it's not treated as exact, we are forcing the user
to accept wider data than was possibly intended.
You can quote me to the entire list as you wish.
Best regards,
Baker