Using mypy with an existing codebase#
This section explains how to get started using mypy with an existing, significant codebase that has little or no type annotations. If you are a beginner, you can skip this section.
If your codebase is large, pick a subset of your codebase (say, 5,000 to 50,000 lines) and get mypy to run successfully only on this subset at first, before adding annotations. This should be doable in a day or two. The sooner you get some form of mypy passing on your codebase, the sooner you benefit.
You’ll likely need to fix some mypy errors, either by inserting
annotations requested by mypy or by adding
# type: ignore
comments to silence errors you don’t want to fix now.
We’ll mention some tips for getting mypy passing on your codebase in various sections below.
Run mypy consistently and prevent regressions#
Make sure all developers on your codebase run mypy the same way.
One way to ensure this is adding a small script with your mypy
invocation to your codebase, or adding your mypy invocation to
existing tools you use to run tests, like
Make sure everyone runs mypy with the same options. Checking a mypy configuration file into your codebase can help with this.
Make sure everyone type checks the same set of files. See Specifying code to be checked for details.
Make sure everyone runs mypy with the same version of mypy, for instance by pinning mypy with the rest of your dev requirements.
In particular, you’ll want to make sure to run mypy as part of your Continuous Integration (CI) system as soon as possible. This will prevent new type errors from being introduced into your codebase.
A simple CI script could look something like this:
python3 -m pip install mypy==0.971
# Run your standardised mypy invocation, e.g.
# This could also look like `scripts/run_mypy.sh`, `tox run -e mypy`, `make mypy`, etc
Ignoring errors from certain modules#
By default mypy will follow imports in your code and try to check everything. This means even if you only pass in a few files to mypy, it may still process a large number of imported files. This could potentially result in lots of errors you don’t want to deal with at the moment.
One way to deal with this is to ignore errors in modules you aren’t yet ready to
type check. The
ignore_errors option is useful for this, for instance,
if you aren’t yet ready to deal with errors from
ignore_errors = True
You could even invert this, by setting
ignore_errors = True in your global
config section and only enabling error reporting with
ignore_errors = False
for the set of modules you are ready to type check.
Prioritise annotating widely imported modules#
Most projects have some widely imported modules, such as utilities or model classes. It’s a good idea to annotate these pretty early on, since this allows code using these modules to be type checked more effectively.
Mypy is designed to support gradual typing, i.e. letting you add annotations at your own pace, so it’s okay to leave some of these modules unannotated. The more you annotate, the more useful mypy will be, but even a little annotation coverage is useful.
Write annotations as you go#
Consider adding something like these in your code style conventions:
Developers should add annotations for any new code.
It’s also encouraged to write annotations when you modify existing code.
This way you’ll gradually increase annotation coverage in your codebase without much effort.
Automate annotation of legacy code#
A simple approach is to collect types from test runs. This may work well if your test coverage is good (and if your tests aren’t very slow).
Another approach is to enable type collection for a small, random fraction of production network requests. This clearly requires more care, as type collection could impact the reliability or the performance of your service.
Introduce stricter options#
Mypy is very configurable. Once you get started with static typing, you may want
to explore the various strictness options mypy provides to catch more bugs. For
example, you can ask mypy to require annotations for all functions in certain
modules to avoid accidentally introducing code that won’t be type checked using
disallow_untyped_defs. Refer to The mypy configuration file for the details.
An excellent goal to aim for is to have your codebase pass when run against
This basically ensures that you will never have a type related error without an explicit
circumvention somewhere (such as a
# type: ignore comment).
The following config is equivalent to
--strict (as of mypy 1.0):
# Start off with these
warn_unused_configs = True
warn_redundant_casts = True
warn_unused_ignores = True
# Getting these passing should be easy
strict_equality = True
strict_concatenate = True
# Strongly recommend enabling this one as soon as you can
check_untyped_defs = True
# These shouldn't be too much additional work, but may be tricky to
# get passing if you use a lot of untyped libraries
disallow_subclassing_any = True
disallow_untyped_decorators = True
disallow_any_generics = True
# These next few are various gradations of forcing use of type annotations
disallow_untyped_calls = True
disallow_incomplete_defs = True
disallow_untyped_defs = True
# This one isn't too hard to get passing, but return on investment is lower
no_implicit_reexport = True
# This one can be tricky to get passing if you use a lot of untyped libraries
warn_return_any = True
Note that you can also start with
--strict and subtract, for instance:
strict = True
warn_return_any = False
Remember that many of these options can be enabled on a per-module basis. For instance,
you may want to enable
disallow_untyped_defs for modules which you’ve completed
annotations for, in order to prevent new code from being added without annotations.
And if you want, it doesn’t stop at
--strict. Mypy has additional checks
that are not part of
--strict that can be useful. See the complete
The mypy command line reference and Error codes for optional checks.
Speed up mypy runs#
You can use mypy daemon to get much faster incremental mypy runs. The larger your project is, the more useful this will be. If your project has at least 100,000 lines of code or so, you may also want to set up remote caching for further speedups.