[PATCH] D79064: [test-suite] Register result codes with lit.

[Michael Kruse via llvm-commits]

See commit a8dbdca1 for the workaround.

Michael

Am Do., 14. Mai 2020 um 11:56 Uhr schrieb Jessica Paquette
<jpaquette at apple.com>:
>
> > This patch makes the test-suite master work with llvm-project's
> > master, which IMHO should take precedence over users using older
> > versions of lit with newer version of test-suite for whatever reason.
> > Not having the two masters consistent does break most buildbot setups,
> > including LLVM's own.
>
> This is true, but we do have public-facing bots which rely on installing lit from pip.
>
> e.g. llvm-zorg/tasks/utils/venv_lit.sh installs lit via pip. This is used for several test suite builds (such as CTMark)
>
> So, unfortunately, this isn't really just in the realm of non-public-facing CI outside of the general llvm project.
>
> In general, I agree with what you're saying though. :/
>
> > I could offer to commit a workaround which only calls
> > add_result_category if it exists:
> >
> > try:
> >  lit.main.add_result_category(NOCHANGE, "Executable Unchanged")
> >  lit.main.add_result_category(NOEXE, "Executable Missing")
> > except:
> >  # add_result_category was added in newer version of lit
> >  pass
> >
> > If you insist on reverting this, I suggest to revert
> > rGfbdcfcd4c392c487c411632b951f4aada351154e as well.
> >
> > Michael
>
> I don't really want to revert the commits if possible, so I'm happy with the workaround.
>
> I think it would be good in the comment to make it clear that the workaround should be removed once the pypi lit has support for the feature.
>
> - Jessica
>
> > On May 13, 2020, at 7:35 PM, Michael Kruse <llvm at meinersbur.de> wrote:
> >
> > This patch makes the test-suite master work with llvm-project's
> > master, which IMHO should take precedence over users using older
> > versions of lit with newer version of test-suite for whatever reason.
> > Not having the two masters consistent does break most buildbot setups,
> > including LLVM's own.
> >
> > I could offer to commit a workaround which only calls
> > add_result_category if it exists:
> >
> > try:
> >  lit.main.add_result_category(NOCHANGE, "Executable Unchanged")
> >  lit.main.add_result_category(NOEXE, "Executable Missing")
> > except:
> >  # add_result_category was added in newer version of lit
> >  pass
> >
> > If you insist on reverting this, I suggest to revert
> > rGfbdcfcd4c392c487c411632b951f4aada351154e as well.
> >
> > Michael
> >
> >
> > Am Mi., 13. Mai 2020 um 18:57 Uhr schrieb Jessica Paquette via
> > Phabricator <reviews at reviews.llvm.org>:
> >>
> >> paquette added subscribers: tstellar, ddunbar, paquette.
> >> paquette added a comment.
> >>
> >> Just as a heads up, it seems that this commit breaks anyone using the lit in pypi. (https://pypi.org/project/lit/)
> >>
> >> If you try using pypi's lit with this commit (say by running the test suite with LNT), you'll get
> >>
> >>  AttributeError: 'function' object has no attribute 'add_result_category'
> >>
> >> Would it be possible to temporarily back this out until that lit has been updated? (cc: @tstellar @ddunbar)
> >>
> >>
> >> Repository:
> >>  rT test-suite
> >>
> >> CHANGES SINCE LAST ACTION
> >>  https://reviews.llvm.org/D79064/new/
> >>
> >> https://reviews.llvm.org/D79064
> >>
> >>
> >>
>