The coverage seems to have decreased. Does this not remove the lines from the percentage?

So I think it does remove the lines, but the percentage will depend on what was covered. Here, the only noticeable decrease is for the benchmarks. If we look at, e.g. glass/algorithm.py. I think it makes sense that the coverage has "decreased".

Before 8.25%

Manually counting I see:

• 4 import lines covered
• 4 non-import lines covered
• 75 lines not covered

8/75=10.7% 🤔 might be something to do with branch coverage

After 4.3%

Manually counting I see:

• 0 import lines covered
• 4 non-import lines covered
• 75 lines not covered

4/75=5.3% 🤔 might be something to do with branch coverage

@paddyroddy I'm not sure I follow. So the import lines are still counted in the total number of lines to be covered. Does that mean this ignore is purely for the highlighting in the coveralls UI?

@paddyroddy I'm not sure I follow. So the import lines are still counted in the total number of lines to be covered. Does that mean this ignore is purely for the highlighting in the coveralls UI?

Pretty much. The lines aren't counted in those to be covered. It's the same reason we're doing it for the others

See #279

So the import lines are still counted in the total number of lines to be covered.

This statement is incorrect

@connoraird failing with this. Think we shouldn't do import tests.conftest directly 🤔

@connoraird failing with this. Think we shouldn't do import tests.conftest directly 🤔

Weird, I originally had just import conftest but I noticed we were using import test.conftest in the type checking block.

Weird, I originally had just import conftest but I noticed we were using import test.conftest in the type checking block.

We're doing from tests.conftest, subtly different

Actually the issue is that it's not in a type checking block. You aren't really meant to import from test files

Related pytest-dev/pytest#349

Actually the issue is that it's not in a type checking block. You aren't really meant to import from test files

Yeah I know you're not supposed to but I didn't know how else to use xp_available_backends in the definition of the parameter like I do here.

import conftest
...
@pytest.mark.stable
@pytest.mark.parametrize(
    "xp",
    [
        xp
        for name, xp in conftest.xp_available_backends.items()
        if name != "jax.numpy"
    ],
)
def test_displacement(
    benchmark: BenchmarkFixture,
    urng: UnifiedGenerator,
    xp: ModuleType,  # noqa: ARG001
) -> None:
    """Benchmark for glass.displacement with all backends, but jax."""
    _benchmark_displacement(benchmark, urng)

@paddyroddy Unless I make a fixture which excludes jax from xp?

@paddyroddy Unless I make a fixture which excludes jax from xp?

Let's do that. I'll revert the changes in this PR and wait until we've got that in.

Broken due to changes in #864. Will merge as the fixes are unrelated.