trailing comma at end of a parametrized test's arguments list changes test semantics

[MichaReiser]

Hi. I'm working on Ruff and we received a report that Ruff's formatter breaks pytests by adding a trailing comma at the end of a parametrized test's argument list (astral-sh/ruff#18437).

There test before formatting is:

class TestMonitoring:

    model_id = "test_model"
    run_date = datetime(2024, 5, 20)
    prediction_date = "2024-02-20"

    @pytest.mark.parametrize(
        "prod_run",
        [
            True,
            False,
        ]
    )
    def test_get_prediction_results(self, monkeypatch, spark, prod_run):
        database = "app_ard368_output" if prod_run else "discovery_ard368_output"

        prediction_df = spark.createDataFrame(TestDataMonitoring.prediction_data, TestDataMonitoring.prediction_schema)
        expected_df = spark.createDataFrame(TestDataMonitoring.prediction_data_specific_date, TestDataMonitoring.prediction_schema)

        monkeypatch.setattr(spark, "table", get_spark_tables({f"{database}.{self.model_id}_output": prediction_df}))

        monitoring = Monitoring(spark, self.model_id, self.run_date, prod_run)

        result_df = monitoring.get_prediction_results(self.prediction_date)

        assert_df_equality(result_df, expected_df, ignore_nullable=True)

Ruff's formatter then inserts a trailing comma after the [ True, False ] array. The formatted test is:

    ... 

    @pytest.mark.parametrize(
        "prod_run",
        [
            True,
            False,
        ],
    )
    def test_get_prediction_results(self, monkeypatch, spark, prod_run):
        ... # unchanged

Now, their test fails with FAILED tests/monitoring/test_unit_monitoring.py::TestMonitoring::test_get_prediction_results[True] - AttributeError: 'tuple' object has no attribute 'schema'.

Adding the trailing comma doesn't change semantics according to Python, but it seems to do so for Pytest. Is this intentional or a bug in pytest?

I tried to narrow this to a smaller repro but failed.

They use: pytest-8.2.2 pytest-metadata-3.1.1 pytest_html-4.1.1 pytest_mock-3.14.0 and run pytest with .venv/bin/python -m coverage run --source=src -m pytest -o log_cli=true --log-level=INFO --show-capture=log -W ignore:DeprecationWarning -vv

CC: @keita00 as you reported this issue with Ruff. What could help us repro this issue is if you could tell us for which object the .schema access fails. Is it the prod_run argument or is the access on the spark object? If so, how's the spark fixture defined?

[The-Compiler]

As you pointed out in the original issue:

I took your example and ran both the version with and without trailing comma through the python parser and they prodcue the same AST.

pytest does not parse the sources (other than for reporting perhaps), and the error @keita00 shows does not point at anything in pytest. Also:

Happened not only in tests, but in apllication as well. Usually changing tuple to tuple of tuples, or list/dist to tuple.

I don't think there is anything that pytest can do here without a clear reproducer, and a full stacktrace - but again, unless there is evidence of this being a pytest issue, I see nothing pointing towards that.

[RonnyPfannschmidt]

Closing as invalid
The original upstream poster avoids to tell the full story as of now