Re: lib/59811: strpct round-off error

[Jose Luis Duran <jlduran%gmail.com@localhost>]

On Mon, Dec 1, 2025 at 9:35 PM Robert Elz via gnats
<gnats-admin%netbsd.org@localhost> wrote:
>
> The following reply was made to PR lib/59811; it has been noted by GNATS.
>
> From: Robert Elz <kre%munnari.OZ.AU@localhost>
> To: gnats-bugs%netbsd.org@localhost
> Cc:
> Subject: Re: lib/59811: strpct round-off error
> Date: Tue, 02 Dec 2025 07:33:44 +0700
>
>      Date:        Sun, 30 Nov 2025 21:10:00 +0000 (UTC)
>      From:        jlduran%gmail.com@localhost
>      Message-ID:  <20251130211000.A4F7E1A923C%mollari.NetBSD.org@localhost>
>
>
>    | While porting df tests to FreeBSD, I realized some percentages
>    | were incorrectly rounded.
>
>  Hmm.   If I do the same calculation in bc (whose manual does say that
>  it truncates the results from divide operations) I see the same thing:
>
>  scale=1
>  1443260007 * 100 / 2164890010
>  66.6
>  scale=20
>  1443260007 * 100 / 2164890010
>  66.66666668206390771787
>
>  I'm not certain that having df give answers different than what bc
>  would give were the user to use that to calculate the values, is
>  necessarily a good idea.

Interesting, but also note that df[1] reads:
The percentage value shall be expressed as a positive integer, with
any fractional result causing it to be rounded to the next highest
integer.
That is at least when using the -P flag.
[1]: https://pubs.opengroup.org/onlinepubs/9799919799/utilities/df.html

>    | Apparently, it has issues rounding 66.6666... to 66.7 or 67.
>
>  ["it" being strpct(3) of course]
>
>  That, or it simply implements "round towards 0" for all operations.
>
>  I will allow others to ponder a little upon what result we should be
>  producing here (in the grand scheme of things I doubt it matters a lot)
>  while I take a look to see how strpct() might be adapted to do "rounding
>  to nearest" which is what I assume you're expecting it to do.

Yes, I agree it is a low-priority issue.

>  Maybe we need variants of str[s]pct() with a flags arg to indicate which
>  kind of rounding is desired, or would that be way over the top?

My personal opinion is that it would be over the top. I like the
overall idea of strpct(3), if this is the expected behavior, a test or
a simple paragraph in the manual page will suffice.
Thank you for looking into it!

>  kre
>