svn commit: r211304 - head/lib/libutil

[Bruce Evans]

On Mon, 23 Aug 2010, [utf-8] Dag-Erling Smørgrav wrote:

> Dimitry Andric <dimitry at andric.com> writes:
>> Dag-Erling Smørgrav <des at des.no> writes:
>>> Bruce Cran <bruce at cran.org.uk> writes:
>>>> Somewhat related, there are overflow bugs in humanize_number - for
>>>> example df(1) fails to display space from a 100PB filesystem
>>>> correctly.
>>> Patch?  :)
>> Attached.  This makes humanize_number() work properly for all possible
>> int64_t values.
>
> That's awesome!  Any way I can convince you to fix expand_number() as
> well?  I got a detailed explanation of what's wrong with it (both before
> and after my commits) from bde@ (cc:ed) in private correspondence; I can
> forward it to you if he doesn't mind.

OK.

I think the final point was that it should go back to supporting signed
numbers (only), and that means int64_t ones until scientificize^dehumanize^W
humanize_number() is fixed to support intmax_t ones.

Bruce