PT022 gives a false positive with GDAL code

[jontwo]

Summary

This is a bit of a niche issue, sorry, but when testing a GDALDataset, you need to keep the dataset in scope to avoid segfaults. This means you have to construct a fixture like this to give you a vector layer for testing with.

    @pytest.fixture
    def lyr1(self, fields, srs_wgs84):
        """Return a polygon layer for testing."""
        ds = self.create_temp_vector(lyr_names=["lyr1"], fields=fields, srs=srs_wgs84)
        lyr = ds.GetLayer()
        yield lyr

Once the layer goes out of scope, the dataset (ds) can be closed. If you return lyr, the ds will close and be garbage collected and you will get a segfault when you try and use the layer.

Feel free to close this issue as a Won't Fix, but I thought it was worth raising just in case. We can always noqa that line to keep Ruff quiet.

Version

No response

[ntBre]

Thanks for the report! As you mentioned, I think this is probably beyond Ruff's capabilities to detect automatically, so a noqa is probably the best solution, or some kind of dummy usage of lyr after the yield like:

    @pytest.fixture
    def lyr1(self, fields, srs_wgs84):
        """Return a polygon layer for testing."""
        ds = self.create_temp_vector(lyr_names=["lyr1"], fields=fields, srs=srs_wgs84)
        lyr = ds.GetLayer()
        yield lyr
        del lyr
        # or
        _ = lyr

but I guess a noqa on the same line is actually shorter.

I'm pretty surprised that we offer a safe fix in this situation, though, given that it definitely changes the behavior in your case!

[ntBre]

Probably worth checking the precedent in other rules that rewrite returns and yields, but I suspect that this should be an unsafe fix in general.

ruff/crates/ruff_linter/src/rules/flake8_pytest_style/rules/fixture.rs

Lines 842 to 845 in e1198ba

	if let Some(return_type_edit) = return_type_edit {
	diagnostic.set_fix(Fix::safe_edits(yield_edit, [return_type_edit]));
	} else {
	diagnostic.set_fix(Fix::safe_edit(yield_edit));

[jontwo]

del lyr is a good shout - we often do that in other places as the layer/dataset isn't written until it goes out of scope but sometimes you want to read it back in the same code block.