Re: [PATCH v2 2/2] drm/tests: Change "igt_" prefix to "test_drm_"

[Michał Winiarski]

On Mon, Sep 05, 2022 at 02:10:00PM +0200, Maxime Ripard wrote:
> On Fri, Sep 02, 2022 at 03:38:28PM +0200, Michał Winiarski wrote:
> > On Fri, Sep 02, 2022 at 04:03:20PM +0300, Jani Nikula wrote:
> > > On Fri, 02 Sep 2022, Maxime Ripard <maxime@xxxxxxxxxx> wrote:
> > > > On Fri, Sep 02, 2022 at 11:04:14AM +0300, Jani Nikula wrote:
> > > >> On Thu, 01 Sep 2022, Maíra Canal <mairacanal@xxxxxxxxxx> wrote:
> > > >> > Hi Maxime,
> > > >> >
> > > >> > On 9/1/22 09:55, Maxime Ripard wrote:
> > > >> >> Hi,
> > > >> >> 
> > > >> >> On Thu, Sep 01, 2022 at 09:42:10AM -0300, Maíra Canal wrote:
> > > >> >>> With the introduction of KUnit, IGT is no longer the only option to run
> > > >> >>> the DRM unit tests, as the tests can be run through kunit-tool or on
> > > >> >>> real hardware with CONFIG_KUNIT.
> > > >> >>>
> > > >> >>> Therefore, remove the "igt_" prefix from the tests and replace it with
> > > >> >>> the "test_drm_" prefix, making the tests' names independent from the tool
> > > >> >>> used.
> > > >> >>>
> > > >> >>> Signed-off-by: Maíra Canal <mairacanal@xxxxxxxxxx>
> > > >> >>>
> > > >> >>> ---
> > > >> >>> v1 -> v2: https://lore.kernel.org/dri-devel/20220830211603.191734-1-mairacanal@xxxxxxxxxx/
> > > >> >>> - Change "drm_" prefix to "test_drm_", as "drm_" can be a bit confusing (Jani Nikula).
> > > >> >> 
> > > >> >> I appreciate it's a bit of a bikeshed but I disagree with this. The
> > > >> >> majority of the kunit tests already out there start with the framework
> > > >> >> name, including *all* the examples in the kunit doc. Plus, it's fairly
> > > >> >> obvious that it's a test, kunit is only about running tests in the first
> > > >> >> place.
> > > >> >
> > > >> > Would it be better to keep it as "drm_"?
> > > >> 
> > > >> That's not "keeping". That's renaming igt to drm.
> > > >
> > > > Well, there's like half the tests that are prefixed with drm, the other
> > > > with igt, so it's both really
> > > >
> > > >> > Currently, I don't think it is appropriate to hold the "igt_" prefix, as
> > > >> > the tests are not IGT exclusive, but I don't have a strong opinion on
> > > >> > using the "drm_" or the "test_drm" prefixes.
> > > >> 
> > > >> I repeat my stance that "drm_" alone is confusing.
> > > >
> > > > What are you confusing it with?
> > > >
> > > >> For the reason alone that it pollutes the code tagging tools, mixing
> > > >> actual drm_ types and functions with unit test functions.
> > > >
> > > > I don't get it, I'm sorry. All these functions are static and not part
> > > > of any API, so I can't see how it would pollute a code tagging tool. Or
> > > > at least, not more than any driver does.
> > > >
> > > > And we're part of a larger project here, it's about consistency with the
> > > > rest of the ecosystem.
> > > 
> > > Okay, so I'm just going to explain what I mean, but say "whatever" right
> > > after and move on.
> > > 
> > > For example, drm_buddy_test.c includes drm_buddy.h so with the igt_ ->
> > > drm_ rename none of the test functions may clash with the drm_buddy_
> > > prefixed existing functions. Ditto for all tests similarly.
> > > 
> > > For example drm_buddy_alloc_range() as a name sounds like something that
> > > allocs a range, not something that tests range allocation. On the other
> > > hand, you have drm_buddy_alloc_blocks() which is actually a real
> > > drm_buddy function, not a test. What would you call a test that tests
> > > that? Here, we end up with names that are all prefixed drm_buddy and you
> > > won't know what's the actual function and what's the test unless you
> > > look it up.
> > > 
> > > I use code tagging that I can use for finding and completing
> > > e.g. functions. Currently I have the following completions, for
> > > igt_buddy_ and drm_buddy_, respectively:
> > > 
> > > Possible completions are:
> > > igt_buddy_alloc_limit 	igt_buddy_alloc_optimistic 	igt_buddy_alloc_pathological
> > > igt_buddy_alloc_pessimistic 	igt_buddy_alloc_range 	igt_buddy_alloc_smoke
> > > 
> > > Possible completions are:
> > > drm_buddy_alloc_blocks 	drm_buddy_block 	drm_buddy_block_is_allocated 	drm_buddy_block_is_free
> > > drm_buddy_block_is_split 	drm_buddy_block_offset 	drm_buddy_block_order 	drm_buddy_block_print
> > > drm_buddy_block_size 	drm_buddy_block_state 	drm_buddy_block_trim 	drm_buddy_fini
> > > drm_buddy_free_block 	drm_buddy_free_list 	drm_buddy_init 	drm_buddy_init_test
> > > drm_buddy_module_exit 	drm_buddy_module_init 	drm_buddy_print
> > > 
> > > With the patch at hand, they'll all be lumped under drm_buddy_
> > > completions, and some of them will be actual drm buddy functions and
> > > some not.
> > > 
> > > I just find it a very odd convention to name the tests in a way that's
> > > indistinguishable from the real things. Even *within* drm_buddy_test.c
> > > where you read the test code. Because currently you do have calls to
> > > igt_buddy_ prefixed functions from other igt_buddy_ prefixed functions,
> > > along with the drm_buddy_ prefixed calls. I think it's just going to be
> > > a mess.
> > > 
> > > /rant
> > > 
> > > Whatever. Moving on.
> > 
> > KUnit docs [1] state:
> > https://docs.kernel.org/dev-tools/kunit/style.html#test-cases
> > "As tests are themselves functions, their names cannot conflict with other
> > C identifiers in the kernel. This may require some creative naming."
> > And give examples of names. But this should be local to individual test suite -
> > as long as the test is readable, and the name describes what it is testing, we
> > should be fine. We don't even need to pass drm_* prefix, as this convention is
> > expected for test suites, not test cases [2].
> > 
> > Having said that - I do believe that igt_* prefix don't belong here (which is
> > why I'm progressively trying to get rid of in the patches that refactor some of
> > the tests).
> > I agree with Jani - can we take it on a case-by-case basis?
> > If the test name is too similar to the function that it is testing, we could go
> > with one of the following (taking igt_buddy_alloc_limit as example):
> > drm_buddy_test_alloc_limit
> > test_drm_buddy_alloc_limit
> > buddy_test_alloc_limit
> > test_buddy_alloc_limit
> 
> We also have drm_test_buddy_alloc_limit, or drm_buddy_test_alloc_limit
> 
> Both would be fine for me, with a small preference for the former, which
> should also address Jani's concerns?
> 
> > And either of those is fine in my opinion (I'd personally go with
> > test_buddy_alloc_limit in this case).
> > We don't really need a DRM-wide (or worse, kernel wide) convention for test case
> > names (it's only really needed for test suites).
> 
> Sure we do. kunit.py can take some filters too. Being able to only run
> DRM tests with a single filter is super convenient, and if we fail to
> provide a consistent naming we're pretty much sure people running the
> tests are going to miss some.

We have "test suite" - and we do have a naming convention for that.
Filter glob takes test_suite.test_name.

Let's use specific examples.

Starting with igt_buddy_alloc_limit, "full" test name is
"drm_buddy.igt_buddy_alloc_limit" - the way to run this single test would be:
./kunit.py run 'drm_buddy.igt_buddy_alloc_limit'

Say we want to run all drm_buddy tests, we can do it with:
./kunit.py run 'drm_buddy.*'

We can also run all DRM tests:
./kunit.py run 'drm*'

What you're arguing, is that we need to rename all test cases to have drm_test
prefix, so the full testname for igt_buddy_alloc_limit becomes
drm_buddy.drm_test_buddy_alloc_limit, because we want to be able to use this
specific glob for matching by test case names (not test suites):
./kunit.py run '*.drm*'

Why? We can already do that with 'drm*', that's what the test suites are for,
and the glob is simpler and more intuitive IMO.

-Michał

> 
> Maxime