[tae898]

The temporary database and server-root fixtures now create their directories with pytest's tmp_path_factory.mktemp (or tmp_path) instead of tempfile.mkdtemp. A killed run never reaches teardown, so the old directories stayed in /tmp for good (48 arcadedb_test_db_* directories on 2026-10-25). pytest keeps only its last three basetemp runs, so nothing permanent is left now.

Changed: temp_server_root, temp_db_path, temp_db and temp_dir_factory in tests/conftest.py; the module fixture db in tests/test_resultset_read_errors.py; the temp_db_path fixtures in tests/test_import_database.py and tests/test_exporter.py; cleanup_test_dirs in tests/test_server_patterns.py; cleanup_db in tests/test_concurrency.py; the tmp fixture in tests/test_free_threading.py; and test_async_executor_sql_command_insert_is_exact_at_parallel_one, which also leaked its mkdtemp parent on a normal exit. The explicit close and rmtree with the Windows retry stay, since pytest's cleanup is lazy and cannot retry a directory the JVM still holds. mktemp is used rather than the per-test tmp_path to keep the paths short on Windows.

Left as is: the __main__ stress path in tests/test_free_threading.py (a script, not a pytest fixture), the script strings run in a subprocess in tests/test_java_package_shadowing.py and tests/test_python_function_interrupt.py (no pytest fixtures there, and they remove their own directory), and tests/test_log_dir.py (it only reads gettempdir() to assert a default path). The NamedTemporaryFile files in tests/test_import_database.py are single CSV or XML files, not databases.

Checks: pre-commit (black, isort) passes on the changed files and the files compile. pyflakes reports only the existing f-string warning in tests/test_server_patterns.py, which is also on main. The suite was not run locally; CI covers it.

Closes #140

🤖 Generated with Claude Code