Prevent --update-functional-output from deleting or overwriting fallback versioned output files for other Python versions.

[mchris86]

Type of Changes

	Type
✓	🐛 Bug fix

Description

Closes #10844

Fix --update-functional-output so it does not delete or overwrite fallback versioned expected-output files.

When Pylint runs on a newer Python version (e.g. 3.14) and no matching output file exists (like .314.txt), it falls back to an older version (e.g. .313.txt). The problem is that --update-functional-output would then modify that fallback file, when it shouldn't.

The fix adds a way to detect when the expected output file is a fallback. In such cases, the update logic skips deleting/overwriting the file.

Regression tests are added to cover both cases:

• fallback versioned output is not deleted
• fallback versioned output is not overwritten

Testing

pytest tests/testutils/test_lint_module_output_update.py -v

Before this fix, the new tests fail because the fallback file is modified. After applying the fix, all tests pass.

[Pierre-Sassoulas]

Thank you for working on pylint; I was planning on using pytest-remaster for the functional tests (there's an annoying number of edge cases and test to do to make it right and pylint main purpose is not to create a testing framework). I can push my work in progress if it helps

[mchris86]

Thanks for the context!
I would definitely be interested in seeing your work in progress.

[Pierre-Sassoulas]

Pushed in https://github.com/pylint-dev/pylint/tree/pytest-remaster, it's pretty raw still

[mchris86]

Thanks, I looked into the pytest-remaster branch and was able to reproduce the issue there as well.

It looks like test_file.expected_output is being used both to select the comparison target and the update target.

In the remaster version, that shows up here:

golden_master.check(
    lint_test.serialize_output(actual_output), test_file.expected_output 
)

If expected_output is a fallback like .313.txt while running on 3.14, remaster still updates that file.

In my attempt I handled this by detecting fallback outputs and skipping updates in that case. I'd be happy to help handle this in the remaster approach if that would be useful.

[Pierre-Sassoulas]

Sure, a new look on the pytest-remaster implementation would be great !