- Most tests are "characterization" tests, meaning they represent the current output under these guidelines. This does not mean that it's the ideal output.
- If you need to change an existing characterization, that's fine, so long as you're convinced it's no less correct than what was previously there.
- See TestConstants for an automated way to rewrite the expected output en masse
- Run hooks/enable-hooks.cmd to enable a pre-commit hook which stops you committing the flag when true
- Single file characterization tests (e.g. ExpressionTests.cs)
- When there are several test files that could accomodate a new test, pick the most specifically named one that covers what you're aiming to test.
- Multi file characterization tests (e.g. MultiFileSolutionAndProjectTests.cs)
- Single file self-verifying tests (e.g. EnumTests.vb)
- These are the only tests where the output is not in the repository.
- They convert a test method, and ensure it still passes.
- These can give great confidence that the conversion is correct.
- It's also possible for some conversion bugs to make the test pass for the wrong reason (in the extreme, imagine that the entire test method body became empty)
- In general, the source and target of all test cases should compile.
- Exception: Things in the MissingSemanticModelInfo folder which specifically test common cases of incomplete input.
- Some older tests have compile errors in the input or output. These will be gradually fixed over time.