Pytype checks and infers types for your Python code - without requiring type annotations. Pytype can:
Pytype is a static analyzer; it does not execute the code it runs on.
Thousands of projects at Google rely on pytype to keep their Python code well-typed and error-free.
Pytype uses inference instead of gradual typing. This means it will infer types on code even when the code has no type hints on it. So it can detect issues with code like this, which other type checkers would miss:
def f(): return "PyCon" def g(): return f() + 2019 # pytype: line 4, in g: unsupported operand type(s) for +: 'str' # and 'int' [unsupported-operands]
Pytype is lenient instead of strict. That means it allows all operations that succeed at runtime and don't contradict annotations. For instance, this code will pass as safe in pytype, but fail in other type checkers, which assign types to variables as soon as they are initialized:
from typing import List def get_list() -> List[str]: lst = ["PyCon"] lst.append(2019) return [str(x) for x in lst] # mypy: line 4: error: Argument 1 to "append" of "list" has # incompatible type "int"; expected "str"
Also see the corresponding FAQ entry.
To quickly get started with type-checking a file or directory, run the
file_or_directory with your input:
pip install pytype pytype file_or_directory
To set up pytype on an entire package, add the following to a
file in the directory immediately above the package, replacing
with the package name:
[tool.pytype] inputs = ['package_name']
Now you can run the no-argument command
pytype to type-check the package. It's
also easy to add pytype to your automated testing; see this
example of a GitHub project that runs pytype on GitHub Actions.
Finally, pytype generates files of inferred type information, located by default
.pytype/pyi. You can use this information to type-annotate the
corresponding source file:
merge-pyi -i <filepath>.py .pytype/pyi/<filename>.pyi
You need a Python 3.7-3.10 interpreter to run pytype, as well as an
$PATH for the Python version of the code you're analyzing
On Alpine Linux, installation may fail due to issues with upstream
dependencies. See the details of this issue for a
** If the ninja dependency fails to install, make sure cmake is installed. See this issue for details.
Pytype can be installed via pip. Note that the installation requires
setuptools. (If you're working in a virtualenv, these two packages should
already be present.)
pip install pytype
Or from the source code on GitHub.
git clone --recurse-submodules https://github.com/google/pytype.git cd pytype pip install .
Instead of using
--recurse-submodules, you could also have run
git submodule init git submodule update
pytype directory. To edit the code and have your edits tracked live,
replace the pip install command with:
pip install -e .
Follow the steps above, but make sure you have the correct libraries first:
sudo apt install build-essential python3-dev libpython3-dev
usage: pytype [options] input [input ...] positional arguments: input file or directory to process
-V, --python-version: Python version (major.minor) of the target code. Defaults to the version that pytype is running under.
-o, --output: The directory into which all pytype output goes, including generated .pyi files. Defaults to
-d, --disable. Comma or space-separated list of error names to ignore. Detailed explanations of pytype's error names are in this doc. Defaults to empty.
For a full list of options, run
In addition to the above, you can direct pytype to use a custom typeshed
installation instead of its own bundled copy by setting
For convenience, you can save your pytype configuration in a file. The config
file can be a TOML-style file with a
[tool.pytype] section (preferred) or an
INI-style file with a
[pytype] section. If an explicit config file is not
supplied, pytype will look for a pytype section in the first
setup.cfg file found by walking upwards from the current working directory.
Start off by generating a sample config file:
$ pytype --generate-config pytype.toml
Now customize the file based on your local setup, keeping only the sections you need. Directories may be relative to the location of the config file, which is useful if you want to check in the config file as part of your project.
For example, suppose you have the following directory structure and want to
~/repo1/foo, which depends on package
~/ ├── repo1 │ └── foo │ ├── __init__.py │ └── file_to_check.py └── repo2 └── bar ├── __init__.py └── dependency.py
Here is the filled-in config file, which instructs pytype to type-check
~/repo1/foo as Python 3.9 code, look for packages in
and ignore attribute errors. Notice that the path to a package does not include
the package itself.
$ cat ~/repo1/pytype.toml # NOTE: All relative paths are relative to the location of this file. [tool.pytype] # Space-separated list of files or directories to process. inputs = [ 'foo', ] # Python version (major.minor) of the target code. python_version = '3.9' # Paths to source code directories, separated by ':'. pythonpath = .:~/repo2 # Space-separated list of error names to ignore. disable = [ 'attribute-error', ]
We could've discovered that
~/repo2 needed to be added to the pythonpath by
running pytype's broken dependency checker:
$ pytype --config=~/repo1/pytype.toml ~/repo1/foo/*.py --unresolved Unresolved dependencies: bar.dependency
Pytype ships with a few scripts in addition to
annotate-ast, an in-progress type annotator for ASTs.
merge-pyi, for merging type information from a .pyi file into a Python file.
pytd-tool, a parser for .pyi files.
pytype-single, a debugging tool for pytype developers, which analyzes a single Python file assuming that .pyi files have already been generated for all of its dependencies.
pyxref, a cross-references generator.
This is not an official Google product.