Documentation Status

An analysis tool to compare handwritten mypy stubs to stubgen-generated stubs.

stubalyzer makes the process of crafting types for untyped dependencies easier. Some example inconsistencies which are found and reported by stubalyzer are listed below:

  • handcrafted and generated stubs have different method signatures

  • generated stubs are missing functions/class members

  • generated types are more specific than the handcrafted types

How to use?


Install stubalyzer with the following command:

$ pip install stubalyzer

How to run

Run stubalyzer with the following command:

$ stubalyzer -h
usage: stubalyzer [-h] -c CONFIG [-e EXPECTED_MISMATCHES] [-r REFERENCE_STUBS] [-x CHECKSTYLE_REPORT] [-s] [-p]

Analyze a set of (handcrafted) mypy stubs by comparing them to (generated)
reference stubs

required arguments:
-c CONFIG, --config CONFIG
                        Mypy config file

positional arguments:
  STUBS_HANDWRITTEN     Directory of handwritten stubs that need to be

optional arguments:
  -h, --help            show this help message and exit
                        A JSON file, which defines expected mismatching
                        symbols and their match results. If any symbol is
                        declared in an expected_mismatches JSON file,
                        stubalyzer will count it as an expected failure, and
                        ignore this inconsistency.

                        Example contents:
                            "my.module.function: "mismatch",
                            "another.module.Class: "not_found"

                        According to the example above, we expect the signature
                        of my.module.function to mismatch, and module.Class to
                        be missing in the generated stubs. stubalyzer will
                        ignore these inconsistencies.

                        Directory of reference stubs to compare against. If
                        not specified stubgen will be used to generate the
                        reference stubs.

                        Write an xml report in checkstyle format to the given file.
  -s, --silent
                        Suppress all non-error output.
  -p, --include-private

                        Include definitions stubgen would otherwise consider
                        private, when generating the reference stubs. (e.g.
                        names with a single leading underscore, like "_foo")


If the comparison ends successfully with zero inconsistencies, stubalyzer will print a success message to stdout, alongside with an ignore message.

The ignore message includes the number of failures ignored, which are declared as expected in the file for expected mismatches. If this file is not provided, the ignore message will not be printed.

Successfully validated 68 stubs.

If there are mismatches in the given types, stubalyzer will print a list of all inconsistencies with a result message, alongside with an ignore message -if there is any, similar to the following:

Symbol "vars.any_var" not found in generated stubs.

Types for functions.additional_args do not match:

   Handwritten type: def (foo:, bar: -> builtins.str

   Reference type  : def (foo: -> builtins.str

Failure: 33 of 68 stubs seem not to be valid.

2 more fail(s) were ignored, because they were defined in expected mismatches.


The following section contains instructions on how to set up and use the development environment for this project.

Development Setup

Requirements for development:

  • A recent Python version (we currently use 3.9)

For a development setup, run the following shell script:

$ ./dev/

This will create a virtual environment in the directory venv and install the project’s dependencies. To activate the virtual environment, run source ./venv/bin/activate from the project directory.


Tests are run using pytest:

$ pytest

Type Checking

Type checking is done with Mypy:

$ mypy stubalyzer

Code Formatting

To set up the pre-commit hook to automatically format files, run:

$ pre-commit install

The source code is formatted using black and isort. The following will format all files in the project:

$ pre-commit run -a


Linting is done using flake8, in the root directory run:

$ flake8

On commit, pre-commit automtically runs flake8 on changed files.

Dependency Management

If you need new dependencies, add them in pyproject.toml.


The documentation is written using Sphinx. Build the documentation using:

$ cd doc; make html

The output will be in docs/_build/html/index.html.

You can update the API documentation using the following:

$ ./dev/


Stubalyzer has no fixed release schedule. Instead releases are made when needed.

To prepare a new release, run the following and follow the instructions in the output:

$ ./dev/ <patch|minor|major>