cardoso-neto/software-testing

software-testing

• When writing tests, avoid mocking too much otherwise your tests will be unmaintainable.
  • Always prefer one or two simple integration tests over a bunch of fully mocked unit tests.
  • Use real models and data structures whenever possible.
• Print local variables to stdout on tests.
  • Except when they're going to be tested against a particular expected value.
    • e.g.: don't do print(f"{header_count=}")\nassert header_count == 1
  • Focus on inputs and outputs to make it easier to debug it when tests fail.
• Prefer running tests individually rather than the entire suite when facing errors.
• Validate using models (e.g.: pydantic) instead of countless field-by-field assertions.
  • data = MyDataModel(**obj) validates structure and types.
  • then assert data == expected_data.
• When testing file operations (reading/writing), use tmp_path fixture.
  • Files are automatically cleaned up after the test.
• When reading Excel/CSV files for comparison with Pydantic models:
  • Use pd.read_excel(..., dtype=str, keep_default_na=False) to avoid type coercion issues.
  • Pandas converts numeric strings to int/float and empty cells to NaN, breaking Pydantic validation.
• Do not bend prod code over backwards to make tests easier to write.
  • Unless that refactoring is beneficial for the prod code itself.