[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: NaNcodes -- a missed opportunity

Vincent Lefèvre a écrit:
What annoys me is that there is a specification, but AFAIK, not yet
applications.  IMHO specifications should be justified by applications.
Otherwise how can you tell whether a specification is good or not?

Well, DFP is rather new.  The NaN picture for BFP was so grossly
underspecified that only dedicated-platform applications were even
conceivable -- for a long time there was NOTHING at a language level,
and when C99 started to consider it, it was still effectively unusable.

Note however that Ivan Godard has mentioned an actual -- though clearly
platform-specific -- application.

My point was that we missed an opportunity to make a least one type of
application conceivable (on most if not all platforms), namely that of
programmer-interpreted payloads in the form a small integers, where "small"
means representable as a NaNcode in all formats under consideration:  up
to 22 bits or 6 decimal digits, if 32-bit formats are involved.

---Sent: 2011-01-24 14:28:11 UTC

754 | revision | FAQ | references | list archive