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

[Maíra Canal]

On 9/2/22 10:38, 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
> 
> 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).

Although I do like the idea of having a DRM-wide name convention for
test cases as it would ease the identification of test cases, I can see
that consensus on this matter might be hard. So I guess we can take case
by case.

But as it would be nice to remove the "igt_" prefix from the tests, a
first name convention might be needed. Currently, I'm not sure in which
direction to go for removing the "igt_" from the tests.

For me, any of the options mentioned above would be okay, with a slight
preference for the test_drm_ one.

Mentioning "test" on the test case really helps distinguish the API and
the test cases. Most of the KUnit tests that I saw have the "test" in
some part of the function (sometimes in the beginning, sometimes in the
end, or after the subsystem prefix). So, I guess using "test" on the
functions is not a bad practice, or redundant.

Best Regards,
- Maíra Canal

> 
> [1] https://docs.kernel.org/dev-tools/kunit/style.html#test-cases
> [2] https://docs.kernel.org/dev-tools/kunit/style.html#suites
> 
> -Michał
> 
>>
>>
>> BR,
>> Jani.
>>
>>
>> -- 
>> Jani Nikula, Intel Open Source Graphics Center