Re: sqrtint() on non-integer arguments

[Kevin Ryde <user42_kevin@yahoo.com.au>]

Karim Belabas <Karim.Belabas@math.u-bordeaux.fr> writes:
>
> even floor(1.0) would not be
> well defined if 1.0 may represent any real number within 1ulp

Ah, on that one I'd be content that the mantissa extends down to the
radix point, so sensible within usual floating point accuracy.

The aim, yes?, of floor(1e40) giving an error is to notify of the worst
kinds of truncation trouble, where a mantissa is definitely too small
for what's asked of it ...