Regression tests for typeshed

This directory contains code samples that act as a regression test for typeshed's stdlib stubs.

This directory should only contain test cases for functions and classes which are known to have caused problems in the past, where the stubs are difficult to get right. 100% test coverage for typeshed is neither necessary nor desirable, as it would lead to code duplication. Moreover, typeshed has multiple other mechanisms for spotting errors in the stubs.

Where are the third-party test cases?

Not all third-party stubs packages in typeshed have test cases, and not all of them need test cases. For those that do have test cases, however, the samples can be found in @tests/test_cases subdirectories for each stubs package. For example, the test cases for requests can be found in the stubs/requests/@tests/test_cases directory.

The purpose of these tests

Different test cases in this directory serve different purposes. For some stubs in typeshed, the type annotations are complex enough that it's useful to have sanity checks that test whether a type checker understands the intent of the annotations correctly. Examples of tests like these are stdlib/builtins/check_pow.py and stdlib/asyncio/check_gather.py.

Other test cases, such as the samples for ExitStack in stdlib/check_contextlib.py and the samples for LogRecord in stdlib/check_logging.py, do not relate to stubs where the annotations are particularly complex, but they do relate to stubs where decisions have been taken that might be slightly unusual. These test cases serve a different purpose: to check that type checkers do not emit false-positive errors for idiomatic usage of these classes.

How the tests work

The code in this directory is not intended to be directly executed. Instead, type checkers are run on the code, to check that typing errors are emitted at the correct places.

Some files in this directory simply contain samples of idiomatic Python, which should not (if the stubs are correct) cause a type checker to emit any errors.

Many test cases also make use of assert_type, a function which allows us to test whether a type checker's inferred type of an expression is what we'd like it be.

Finally, some tests make use of # type: ignore comments (in combination with mypy's --warn-unused-ignores setting and pyright's reportUnnecessaryTypeIgnoreComment setting) to test instances where a type checker should emit some kind of error, if the stubs are correct. Both settings are enabled by default for the entire subdirectory.

For more information on using assert_type and --warn-unused-ignores/reportUnnecessaryTypeIgnoreComment to test type annotations, this page provides a useful guide.

Naming convention

Use the same top-level name for the module / package you would like to test. Use the check_${thing}.py naming pattern for individual test files.

By default, test cases go into a file with the same name as the stub file, prefixed with check_. For example: stdlib/check_contextlib.py.

If that file becomes too big, we instead create a directory with files named after individual objects being tested. For example: stdlib/builtins/check_dict.py.

Differences to the rest of typeshed

Unlike the rest of typeshed, this directory largely contains .py files. This is because the purpose of this folder is to test the implications of typeshed changes for end users, who will mainly be using .py files rather than .pyi files.

Another difference to the rest of typeshed (which stems from the fact that the test-case files are all .py files rather than .pyi files) is that the test cases cannot always use modern syntax for type hints. While we can use from __future__ import annotations to enable the use of modern typing syntax wherever possible, type checkers may (correctly) emit errors if PEP 604 syntax or PEP 585 syntax is used in a runtime context on lower versions of Python. For example:

from __future__ import annotations

from typing_extensions import assert_type

x: str | int  # PEP 604 syntax: okay on Python >=3.7, due to __future__ annotations
assert_type(x, str | int)  # Will fail at runtime on Python <3.10 (use typing.Union instead)

y: dict[str, int]  # PEP 585 syntax: okay on Python >= 3.7, due to __future__ annotations
assert_type(y, dict[str, int])  # Will fail at runtime on Python <3.9 (use typing.Dict instead)

Version-dependent tests

Some tests will only pass on mypy with a specific Python version passed on the command line to the tests/regr_test.py script. To mark a test-case file as being skippable on lower versions of Python, append -py3* to the filename. For example, if foo is a stdlib feature that's new in Python 3.9, test cases for foo should be put in a file named test_cases/stdlib/check_foo-py39.py. This means that mypy will only run the test case if --python-version 3.9, --python-version 3.10 or --python-version 3.11 is passed on the command line to tests/regr_test.py, but it won't run the test case if --python-version 3.7 or --python-version 3.8 is passed on the command line.

However, if sys.version_info >= (3, target): is still required for pyright in the test file itself. Example: check_exception_group-py311.py

Uh oh!