Re: sort human readable suffixes

[Paul Eggert]

Cornelis de Gier <address@hidden> writes:

> Can you explain why strtod introduces errors?

Rounding errors.  For example, IEEE double can't represent the number
10.3 exactly, so it will miscalculate a number like "10.3P".

> The -n algorithm is fairly complicated (for me). I don't see a way to
> introduce the multiplication factors.

-n scales, so for powers of 10 you just need to adjust the scaling
factor.  Powers of 2 are harder; you're going to need multiple
precision arithmetic to solve that problem in general.  This still
isn't so bad, but I admit it's nontrivial.

> The goal in my opinion is to be able to sort the -h output from df, du
> and ls.

But "sort" shouldn't assume the input comes from df etc; it should
work in the general case.