"Empty id" mechanism in callspec

[smarie]

Hi dear pytest maintainers,

I noticed that starting in pytest 5.4.0, CallSpec2.id does not automatically filter out ids that are empty strings from the ids list. So if in the ids list there is ['a', '', 'b'] in old versions that would lead to id 'a-b' and now it leads to 'a--b'. I understand the rationale: pytest should not hide anything from users, so if there is an empty string id somewhere, it should appear.

• In 5.3.0: CallSpec2.id has a filter
• In 5.4.0 and later: CallSpec2.id has no filter

Unfortunately I was relying on this undocumented mechanism in pytest-cases, because there is a specific situation with fixture unions induced by the new @parametrize, in which I wish to remove a useless piece of id from a callspec. I found an ugly workaround: to use a very long and unique string constant as a marker for "id to remove" and then have a specific cleaning hook in the plugin removing such constants from callspec ids lists.

I was wondering if there could be a better way for the future, baked in the pytest API. For example we could introduce an explicit pytest.NO_ID constant that would be filtered out in CallSpec2.id.

What do you think ? I can propose a PR if needed

[RonnyPfannschmidt]

abd5fc8 from #6607 fixing #6597

i believe the union ids should be as a union parametrize call with explicit ids, im against a NO_ID

[smarie]

Thanks @RonnyPfannschmidt ! Yes I totally agree with you for fixture_union in general.

However the need arises with a parametrize containing fixture references:

from pytest_cases import parametrize, fixture, fixture_ref

@fixture
@parametrize("o", [1, 2])
def b(o):
    pass

@parametrize("a", [0, 1, 2, fixture_ref(b)])
def test_foo(a):
    pass

In that example, you probably agree that a good list of id would be:

test_blah.py::test_foo[0] PASSED                                         [ 20%]
test_blah.py::test_foo[1] PASSED                                         [ 40%]
test_blah.py::test_foo[2] PASSED                                         [ 60%]
test_blah.py::test_foo[b-1] PASSED                                       [ 80%]
test_blah.py::test_foo[b-2] PASSED                                       [100%]

However, behind the scenes, I do a little optimization so I create a fixture_ref for the first three parameters instead of creating a fixture for each of them. You can see it with the idstyle="explicit":

from pytest_cases import parametrize, fixture, fixture_ref

@fixture
@parametrize("o", [1, 2])
def b(o):
    pass

@parametrize("a", [0, 1, 2, fixture_ref(b)], idstyle="explicit")
def test_foo(a):
    pass

yields

test_blah.py::test_foo[a\P0:3-0] PASSED                                  [ 20%]
test_blah.py::test_foo[a\P0:3-1] PASSED                                  [ 40%]
test_blah.py::test_foo[a\P0:3-2] PASSED                                  [ 60%]
test_blah.py::test_foo[a\b-1] PASSED                                     [ 80%]
test_blah.py::test_foo[a\b-2] PASSED                                     [100%]

You can see that there are two hidden parameter controling the fixture union alternative "branch 1 = params 0:3" or "branch 2 = the fixture reference". With this id style we see explicit test ids a/P0:3 and a/b (supposed to be a slash but changed in the console, see #8157 )

So back to the ids of the first example above: in order to make them simple and straightforward, I need the "union alternative" parameter to completely disappear, not even remaining as a single dash. The best I can do without hack is:

test_blah.py::test_foo[-0] PASSED                                         [ 20%]
test_blah.py::test_foo[-1] PASSED                                         [ 40%]
test_blah.py::test_foo[-2] PASSED                                         [ 60%]
test_blah.py::test_foo[b-1] PASSED                                       [ 80%]
test_blah.py::test_foo[b-2] PASSED                                       [100%]

So I had to hack to remove the useless dash. This is where a NO_ID helper would help.

(Note: the above behaviour is in pytest-cases 3.0.0, about to be released)

[smarie]

For reference: the discussion about various styles of ids: smarie/python-pytest-cases#154 (comment)

[RonnyPfannschmidt]

That seems to me like the full parameterset with explicit ids would have to be generated in pytest-cases to support reasonably complete I'd control