Re: libm

[Rin Okuyama <rokuyama%rk.phys.keio.ac.jp@localhost>]

On 2016/08/26 13:13, David Holland wrote:
> (1) libm is horribly non-KNF; my understanding is that at one time it
> had an upstream of sorts which is why it's the way it is, but it
> doesn't any more so there's no longer any reason to humour this. Is
> that correct?

I think so too. But there still remains benefit to keep its style, because
FreeBSD also keeps it. FreeBSD has heavily improved their libm. They have
many long double functions, that are missing in ours. Also, many bug fixes
have been made, cf., lib/50698.

> (2) does anyone who knows floating-point stuff well (which I don't)
> want to collaborate with me on a general cleanup and rototill? (a real
> one, not just reformatting everything)

Unfortunately, I don't have enough knowledge of floating-point arithmetic.
I guess that FreeBSD's libm should be useful code base to improve ours.

Thanks,
Rin