Node ID parts vs. argument order with stacked parametrize / permutations

[The-Compiler]

Someone in a pytest training of mine was surprised that:

import pytest

@pytest.mark.parametrize("a", ["a1", "a2"])
@pytest.mark.parametrize("b", ["b1", "b2"])
def test_permutations(a, b):
    ...

results in pytest collecting:

• test_permutations[b1-a1]
• test_permutations[b1-a2]
• test_permutations[b2-a1]
• test_permutations[b2-a2]

Where the arguments are represented as b-a despite the test argument order being a, b.

I suspect this is because the decorators are applied from bottom to top (and indeed swapping them also generates "nicer" IDs). But I wonder if we could/should do better in cases where the parameters directly correspond to test arguments.

[nicoddemus]

I agree it is a bit surprsing, I guess it is probably simple to just reverse the marks when evaluating them?