On 26.10.21 20:30, John Snow wrote:
>
>
> On Tue, Oct 26, 2021 at 6:51 AM Hanna Reitz <hreitz@redhat.com> wrote:
>
> On 19.10.21 16:49, John Snow wrote:
> > Now, 297 is just the iotests-specific incantations and
> linters.py is as
> > minimal as I can think to make it. The only remaining element in
> here
> > that ought to be configuration and not code is the list of skip
> files,
> > but they're still numerous enough that repeating them for mypy and
> > pylint configurations both would be ... a hassle.
> >
> > Signed-off-by: John Snow <jsnow@redhat.com>
> > ---
> > tests/qemu-iotests/297 | 72
> +++++----------------------------
> > tests/qemu-iotests/linters.py | 76
> +++++++++++++++++++++++++++++++++++
> > 2 files changed, 87 insertions(+), 61 deletions(-)
> > create mode 100644 tests/qemu-iotests/linters.py
>
> Reviewed-by: Hanna Reitz <hreitz@redhat.com>
>
>
> Thanks!
>
> I wonder about `check_linter()`, though. By not moving it to
> linters.py, we can’t use it in its entry point, and so the Python
> test
> infrastructure will have a strong dependency on these linters. Though
> then again, it probably already does, and I suppose that’s one of the
> points hindering us from running this from make check?
>
>
> That one is left behind because it uses iotests API to skip a test.
> Environment setup that guarantees we won't *need* to skip the test is
> handled by the virtual environment setup magic in qemu/python/Makefile.
>
> Hanna
>
>
> More info than you require:
>
> There's maybe about four ways you could run the python tests that
> might make sense depending on who you are and what you're trying to
> accomplish; they're documented in "make help" and again in
> qemu/python/README.rst;
>
> (1) make check-dev -- makes a venv with whatever python you happen to
> have, installs the latest packages, runs the tests
> (2) make check-pipenv -- requires python 3.6 specifically, installs
> the *oldest* packages, runs the tests
> (3) make check-tox -- requires python 3.6 through 3.10, installs the
> newest packages, runs the tests per each python version
> (4) make check -- perform no setup at all, just run the tests in the
> current environment. (Used directly by all three prior invocations)
AFAIU these are all to be run in build/python? Isn’t any of them hooked
up to the global `make check` in build? Because...
None of them are on make check, correct. Not yet. I'll try to make that happen soon to drop 297.
They run out of the source tree directly, since they're checks on the source and aren't actually related to a build of QEMU in any way.
(Y'know, yet. I'm not saying never.)
> In general, I assume that human beings that aren't working on Python
> stuff actively will be using (1) at their interactive console, if they
> decide to run any of these at all.
...I’m just running `make check` in the build directory.
Yeah, that's OK. I mean, I don't expect you to run them unless you were submitting a series to specifically me.
("If they decide to run any of these at all" - I'm aware that very few people would or are doing so. I consider the CI to be mostly for me as the maintainer to make sure nothing regressed.)
I would have hoped that this is hooked up to something that won’t fail
because I don’t have some necessary tools installed or perhaps even
because I have the wrong version of some tools installed (although the
latter would be kind of questionable...). Would be nice if the global
`make check` had a summary on what was skipped...
I mean. These targets shouldn't fail unless you're missing some really basic requirements. They're just not hooked in to the big "make check" yet.
In terms of environment probing and skipping the tests, though, I do need another layer somewhere to manage that. Stuff I'll need to put in configure/meson etc. I have to look into it.
> It imposes the least pre-requirements while still endeavoring to be a
> target that will "just work".
> Options (2) and (3) are what power the CI tests 'check-python-pipenv'
> and 'check-python-tox', respectively; with(2) being the one that
> actually gates the CI.
> Option (4) is only really a convenience for bypassing the venv-setup
> stuff if you want to construct your own (or not use one at all). So
> the tests just assume that the tools they have available will work.
> It's kind of a 'power user' target.
>
> The way the CI works at the moment is that it uses a "fedora:latest"
> base image and installs python interpreters 3.6 through 3.10
> inclusive, pipenv, and tox. From there, it has enough juice to run the
> makefile targets which take care of all of the actual linting
> dependencies and their different versions to get a wider matrix on the
> version testing to ensure we're still keeping compatibility with the
> 3.6 platform while also checking for new problems that show up on the
> bleeding edge.
Perhaps it’s unreasonable, but I’d prefer if a local `make check` would
already run most tests in some form or another and that I don’t have to
push to gitlab and wait for the CI there.
Yep, understand. That's a requirement for me as well in order to drop 297. On the list, I promise.
I mean, I can of course also integrate a `cd python && make check-dev`
invocation into my test scripts, but it doesn’t feel right to
special-case one test area.
Don't worry, I don't expect that. It just took a lot of work to standardize even that much of the test, so I went for the smaller thing instead of the perfect thing. I'm still inching along to the perfect thing, but considered the iotest cleanups I've done a requisite on that path.
> iotests 297 right now doesn't do any python environment setup at all,
> so we can't guarantee that the linters are present, so we decide to
> allow the test to be skipped. My big hesitation of adding this test
> directly into 'make check' is that I will need to do some environment
> probing to make sure that the 'pylint' version isn't too old -- or
> otherwise set up a venv in the build directory that can be used to run
> tests. I know we already set one up for the acceptance tests, so maybe
> there's an opportunity to re-use that venv, but I need to make sure
> that the dependencies between the two sets of tests are aligned. I
> don't know if they agree, currently.
I see.
> I think I probably want to minimize the number of different virtual
> python environments we create during the build, so I will look into
> what problems might exist in re-purposing the acceptance test venv. If
> that can get squared away easily, there's likely no real big barrier
> to adding more tests that rely on a python venv to get cooking during
> the normal build/check process, including iotest 297.
OK, thanks for the explanation!
Hanna
Yep. I'll start trying to integrate this into make check and see where the problems are. It should be safe to do during soft freeze, I think, since it's just testing ...