[Buildroot] [PATCH v5 02/10] testing/infra: split runtime test from BRTest

[Matthew Weber]

Ricardo / Arnout,

On Mon, Feb 4, 2019 at 7:12 PM Matthew Weber
<matthew.weber at rockwellcollins.com> wrote:
>
> Ricardo,
> On Mon, Feb 4, 2019 at 7:00 PM Ricardo Martincoski
> <ricardo.martincoski at gmail.com> wrote:
> >
> > Hello,
> >
> > + Matt
> >
> > On Mon, Feb 04, 2019 at 01:55 PM, Arnout Vandecappelle wrote:
> >
> > > On 12/05/2018 04:58, Ricardo Martincoski wrote:
> > >> From: Ricardo Martincoski <ricardo.martincoski at datacom.ind.br>
> > >>
> > >> Move the setup of emulator to a new class, RuntimeTestBase, that bahaves
> > >> exactly like BRTest currently does.
> > >> It will avoid duplicating code when adding a common class to test the
> > >> git download infra.
> > >>
> > >> Change all current test cases to use the new class.
> > >> Do this by first using automatic replace:
> > >> $ find support/testing/ -name '*.py' | \
> > >>   xargs grep -l BRTest | \
> > >>   xargs sed -i \
> > >>     -e 's,import infra.basetest,\0\nimport infra.runtimetest,g' \
> > >>     -e 's,infra.basetest.BRTest,infra.runtimetest.RuntimeTestBase,g'
> > >> and then manually add code to import runtimetest in test_external.py to
> > >> avoid this error:
> > >>  AttributeError: 'module' object has no attribute 'LoadTestsFailure'
> > >> This explicit import was not need before because run-tests imports
> > >> BRTest and this is the only test file that do not use the defconfig
> > >> fragments from basetest.py in its code.
> > >>
> > >> Signed-off-by: Ricardo Martincoski <ricardo.martincoski at datacom.ind.br>
> > >> Cc: Arnout Vandecappelle <arnout at mind.be>
> > >> ---
> > >> WARNING: this patch changes all current test cases, so if a new test case was
> > >> applied after this patch was sent, 'git am' will apply this patch cleanly, yet
> > >> any testcase created in the meantime will be broken.
> > >> But I can refresh it if need. It will probably only need the automatic replace
> > >> from the commit log to be re-run.
> > >
> > >  This seems to be an optimisation rather than a requirement for the other
> > > patches, so I haven't applied this - it needs to be refreshed first.
> >
> > It's an optimisation, but I failed to describe why it should be done first...
> >
> > The download test cases do not need to build anything before they run, so by
> > splitting the two classes and making test_git to inherit from the new BRTest
> > instead of RuntimeTestBase makes the download test cases to run really fast.
> > The two test cases already on master, after applying and refreshing this patch,
> > run in 35 seconds total in my machine.
> > Less than 2 minutes each in the Gitlab CI auto-scale runners:
> > https://gitlab.com/RicardoMartincoski/buildroot/-/jobs/156194310
> > https://gitlab.com/RicardoMartincoski/buildroot/-/jobs/156194313
>
> If I understand correctly those are the only two at this point that
> should use the BRTest class instead of the RuntimeTestBase, right?
>
> > Unfortunately there is still some problem in my WIP refresh of the patch, so I
> > can't send it now.
> >
> > >
> > >  If you do so before Wednesday, we might still apply it.
> >
> > Unfortunately it's unlikely I can do in time.
> >
>
> I might be in the same situation after looking at the build failures.
> I'll spend a bit more on it tonight but there are a lot of tests to
> either spot check or run all.  I'm not sure yet if it is my build
> machine's OS and python version or qemu at this point.
>

Here's the v6 and I'll start looking through the differences in the CI
reports (looked similar enough I figured I'd get the patch out first).
http://patchwork.ozlabs.org/patch/1036559/