[Top][All Lists]

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

Re: bignum branch

From: Eli Zaretskii
Subject: Re: bignum branch
Date: Sat, 04 Aug 2018 14:07:36 +0300

> From: Achim Gratz <address@hidden>
> Date: Sat, 04 Aug 2018 12:49:24 +0200
> Eli Zaretskii writes:
> > There's a certain tension here between people who are used to do IEEE
> > compliant FP math in other languages, and the rest of us.  The former
> > will want the IEEE semantics of NaNs, which is what surprised Tom; the
> > latter will probably be surprised like Tom was.
> The semantics of NaN have not much to do with IEEE754 and a lot with how
> you do error handling, which shouldn't be a surprise to any programmer.

It won't surprise those of us who had to deal with FP calculations.
I'm not so sure about others, because the semantics of FP exceptions
is not necessarily known to them.

> > I don't see how we can fix this dilemma better than we already did,
> > with making sure eql compares NaNs as equal.  I do think we should
> > document the special behavior of NaNs, because many Emacs users will
> > not be aware of these subtleties.
> Again, comparing the representations of an NaN (binary or otherwise) is
> fair game.  The NaN itself, as long as it propagates through a chain of
> numerical computations, needs to be preserved; otherwise it'd be an
> exercise in futility to produce them in the first place.  If you don't
> want to deal with NaN at all, there are other methods of handling
> numerical domain errors, but they are usually worse (and often much more
> so) than the alternative.

Yes, I know.  Others may not be aware of all those subtleties, which
is exactly what I was saying in my original message.

reply via email to

[Prev in Thread] Current Thread [Next in Thread]