[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
Re: after-rounding tininess detection and directed rounding
Ahh - I missed the context for "fundamental change in tininess detection".
John Harrison is correct -- the tightening of the rule for decimal formats
is new since the 2006 drafts of 754R, as is the tightening of the rule for
loss-of-accuracy detection for binary formats.
The reason for allowing a choice is the need to accomodate existing
implementations -- and where this was not an issue, rules could be
tightened, and were. This also covers the new "reproducible mode"
attribute.
We should however be careful when choosing among alternatives, which
is why I wanted to compare the difficulties of emulating each mode
given a native implementation of the other, and given the variability
in the mechanisms that can be employed to do this efficiently (e.g.
trap handling).
Unless one of the alternatives would be impossible, or unaffordable
from a performance impact point of view, another consideration should
be aesthetics: let's try not to confuse programmers by using one rule
for decimal formats and another one for binary formats, when that is
avoidable. (For native methods it is unfortunately not avoidable.)
Michel.
Sent: 2007-10-15 21:58:04 UTC