unittest.TestLoader.discover() leaves _top_level_dir set if discovery fails

[christianaurichzm]

Bug report

Bug description:

TestLoader.discover() saves self._top_level_dir on entry and restores it at the end of the call (gh-59215), but only when discovery succeeds. If discovery raises, for example because the start directory is not importable, the loader keeps the top-level directory of the failed call. A later discover() on the same loader without an explicit top_level_dir then resolves against it and fails:

import os
import tempfile
import unittest

with tempfile.TemporaryDirectory() as tmp:
    tests = os.path.join(tmp, "tests")
    os.mkdir(tests)
    with open(os.path.join(tests, "test_x.py"), "w") as f:
        f.write("import unittest\n"
                "class T(unittest.TestCase):\n"
                "    def test_ok(self): pass\n")

    loader = unittest.TestLoader()
    try:
        loader.discover(os.path.join(tmp, "tset"))  # typo, no such directory
    except ImportError as e:
        print(e)
    print(loader._top_level_dir)  # .../tset, expected None

    print(loader.discover(tests).countTestCases())  # ImportError, expected 1

Start directory is not importable: '/tmp/tmpXXXX/tset'
/tmp/tmpXXXX/tset
Traceback (most recent call last):
  ...
ImportError: Start directory is not importable: '/tmp/tmpXXXX/tests'

The same second call on a fresh TestLoader finds the test. Since unittest.defaultTestLoader is a module-level instance, one failed discovery through it breaks later discoveries that do not pass top_level_dir.

Restoring _top_level_dir in a finally block would fix this.

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

Linked PRs

• gh-158377: Fix unittest.TestLoader.discover() leaking _top_level_dir on error #158378
• gh-158377: Restore unittest loader top-level dir when discovery fails #158379