Resolve param indices using param values, not parameterset index

[sadra-barikbin]

To compute param indices smarter when user passes multiple parameter sets with duplicate values to parametrize in order to have a better reordering experience. Currently we have :

@pytest.fixture(scope='module')
def fixture1(request):
    pass

@pytest.fixture(scope='module')
def fixture2(request):
    pass

@pytest.mark.parametrize("fixture1, fixture2", [("a", 0), ("b", 1), ("a", 2)], indirect=True)
def test(fixture1, fixture2):
    pass

resulting in inefficient ordering i.e.

<Function test[a-0]>
<Function test[b-1]>
<Function test[a-2]>

It's a follow-up to #11220 .

[RonnyPfannschmidt]

@bluetech after some review of #519 and #2207 i believe it looks fair to consider the ordering more correct

i think we need a crosscheck of how we can observe interactions if people have multiple tests with different sets of parameters and their setup/teardown

[bluetech]

@sadra-barikbin Can you please rebase this on latest main?

[RonnyPfannschmidt]

Apologies but I'll be delayed in reviewing this, I'm still down with a cold

It would have been nice to land this in 8.x

[sadra-barikbin]

Apologies but I'll be delayed in reviewing this, I'm still down with a cold

It would have been nice to land this in 8.x

It's alright. I hope you recover soon.

[RonnyPfannschmidt]

i think we have to document this better

from my pov we assign parameters that are used multiple times in a expanded/multiplied parameter set the index of the firs occurrence

this can be seen as approximation of a "multi dimensional" index from filtered cross-product

[sadra-barikbin]

@bluetech , could you please review the PR too?

[cyw233]

Hi @sadra-barikbin, do we have any plan to merge this PR, please? cc. @RonnyPfannschmidt

[obestwalter]

Just resolved a minor conflict. Let's see if we can get this merged as it is already approved.

[obestwalter]

Our shiny new bot also enforces changelog entries now, so this needs to be added before it can be merged - could you please add that @sadra-barikbin?

[webknjaz]

@sadra-barikbin there's another merge conflict to be resolved in addition to the need of having a changelog fragment.

[sadra-barikbin]

I resolved the conflict but this PR introduces is a breaking change with regard to how indices are assigned to the direct params of the test functions (hence one of the tests fail). Currently, when whole parametrizations of a function take place ,the indices of a direct param is assigned to its occurrences sequentially all at once:

pytest/src/_pytest/python.py

Lines 1452 to 1456 in f373974

	def _recompute_direct_params_indices(self) -> None:
	for argname, param_type in self._params_directness.items():
	if param_type == "direct":
	for i, callspec in enumerate(self._calls):
	callspec.indices[argname] = i

To be able to merge this PR, I should bring the resolve_values_indices_in_parametersets logic in the method above which introduces change to the whole reordering.