hypotenuse

[Christopher Dimech <dimech@gmx.com>]

> Sent: Monday, July 22, 2024 at 4:12 AM
> From: "Eli Zaretskii" <eliz@gnu.org>
> To: "Max Nikulin" <manikulin@gmail.com>
> Cc: emacs-devel@gnu.org
> Subject: Re: hypotenuse
>
> > From: Max Nikulin <manikulin@gmail.com>
> > Date: Sun, 21 Jul 2024 22:54:53 +0700
> >
> > On 21/07/2024 22:12, Christopher Dimech wrote:
> > >>>> (defun hypotenuse (c1 c2)
> > >>>>    (sqrt (+ (* c1 c1) (* c2 c2))) )
> > >
> > > I have a solution to the problem in fortran, but not in elisp.  I could
> > > include it.  Its implementation required serious concentration.
> >
> > Underflow/overflow issues may be solved with a few lines of code and it
> > does matter if it is fortran, elisp, or another language (besides
> > pathological ones). The problem is to avoid error above 1 ULP.
>
> Right, overflow and underflow are just the tip of a very large
> iceberg.
>
> > In 2021 hypot implementation in glibc was replaced by a more precise one:
> > https://sourceware.org/git/?p=glibc.git;a=commit;h=6c848d70383e1dbe932ef41723ac0abfdeec7ca8
>
> Which is exactly why I said we should use the existing implementation
> in the standard library.

So we should.  I am in agreement.