[Toybox] diff not passing tests?

[enh]

On Thu, Jul 28, 2022 at 11:55 AM Rob Landley <rob at landley.net> wrote:
>
> On 7/27/22 18:32, enh wrote:
> >     Heh, interesting. This works:
> >
> >     $ ./diff <(echo -e '1\n2') <(echo -e '1\n3') --unchanged-line-format=U%l
> >     --old-line-format=D%l --new-line-format=A%l
> >     U1D2A3
> >
> >     But actually making the "aa" and "bb" files and sticking them at the end does
> >     not? Hmmm... (Gimme a couple hours for dinner, I'll try to fix it this evening.)
> >
> >
> > yeah, no hurry --- i've reverted us back to before the update, so no-one's
> > annoyed that their tests are failing. (we've also added one of the affected
> > tests to toybox's presubmit so we won't be able to submit a killall that broken
> > again, whether or not we don't come up with a toybox test to cover this specific
> > case.)
>
> Hopefully fixed now, and I added a skeleton.test for it (which is where I've
> been adding lib/args.c plumbing tests).

thanks, seems okay locally, so i've kicked off a presubmit run (which
now includes the killall-using test that broke before).

given that postsubmit takes a day or two to catch up with a culprit
change, i'll hold off submitting until monday [my time] though so that
people whose mondays come first don't have to deal with any fallout
:-)

> (The problem was --longopt= with nothing after it was advancing to eat the next
> argument instead of saving "" as its argument. The two modifier bits of ; and
> " " means there are four modes of argument parsing, and then each one has a
> -short and --long path, so eight codepaths to test and that's BEFORE you get
> into --longopt abc vs --longopt=abc having this corner case I just fixed...)
>
> Sigh, possibly I should make skeleton have an & prefix where the first argument
> has an imaginary dash because that's the OTHER mode of argument parsing that's
> not being tested by skeleton right now. (And I should wire it up to ps so "ps
> ax" and friends get parsed properly. It's there in "tar"...)
>
> Rob