Mypy Clean Slate

CLI tool for providing a clean slate for mypy usage within a project

Motivation

It can be difficult to get a large project to the point where mypy --strict can be run on it. Rather than incrementally increasing the severity of mypy, either overall or per module, mypy_clean_slate enables one to ignore all previous errors so that mypy --strict (or similar) can be used almost immediately. This enables all code written from that point on to be checked with mypy --strict (or whichever flags are preferred), gradually removing the type: ignore comments from that point onwards.

Often running mypy_clean_slate will cover all errors cleanly in a single pass, but there are cases when not all error output is generated first time, and it can be necessary to run a couple of times, checking the diffs. Example of this scenario is given.

mypy_clean_slate works by parsing the output of mypy --strict --show-error-codes, and adding the relevant type: ignore[code] to each line. Only errors from the report are considered, notes are not handled. Meaning something such as error: Function is missing a type annotation [no-untyped-def] will have # type: ignore[no-untyped-def] appended to the end of the line, whereas note: (Skipping most remaining errors due to unresolved imports or missing stubs; fix these first) will be ignored.

Usage

usage: mypy_clean_slate [-h] [-r] [-a] [-o MYPY_REPORT_OUTPUT]

CLI tool for providing a clean slate for mypy usage within a project.

optional arguments:
  -h, --help            show this help message and exit
  -r, --generate_mypy_error_report
                        Generate 'mypy_error_report.txt' in the cwd.
  -a, --add_type_ignore
                        Add "# type: ignore[<error-code>]" to suppress all
                        raised mypy errors.
  -o MYPY_REPORT_OUTPUT, --mypy_report_output MYPY_REPORT_OUTPUT
                        File to save report output to (default is
                        mypy_error_report.txt)

See ./tests/test_mypy_clean_slate.py for a basic, self contained, before/after example.

Examples

Simple example

Given a project with only:

➜  simple_example git:(master) ✗ tree
.
`-- simple.py

0 directories, 1 file

Containing:

# simple.py
def f(x):
    return x + 1

The report can be generated, and simple.py updated, using mypy_clean_slate -ra, resulting in:

def f(x):  # type: ignore[no-untyped-def]
    return x + 1

And mypy --strict will now pass.

Project example, using `pingouin`

Project pingouin is located at: https://github.com/raphaelvallat/pingouin, and commit ea8b5605a1776aaa0e89dd5c0e3df4320950fb38 is used for this example. mypy_clean_slate needs to be run a couple of times here.

First, generate report and apply type: ignore[<error code>]

mypy_clean_slate -ra

Looking at a subset of git diff:

(venv) ➜ pingouin git:(master) ✗ git diff | grep 'type' | head
+import sphinx_bootstrap_theme # type: ignore[import]
+from outdated import warn_if_outdated # type: ignore[import]
+import numpy as np # type: ignore[import]
+from scipy.integrate import quad # type: ignore[import]
+ from scipy.special import gamma, betaln, hyp2f1 # type: ignore[import]
+ from mpmath import hyp3f2 # type: ignore[import]
+ from scipy.stats import binom # type: ignore[import]
+import numpy as np # type: ignore[import]
+from scipy.stats import norm # type: ignore[import]
+import numpy as np # type: ignore[import]

Changes are added and committed with message 'mypy_clean_slate first pass' (commit message used makes no functional difference), and the report re-generated:

mypy_clean_slate -r

Which reports Found 1107 errors in 39 files (checked 42 source files). So, re-running mypy_clean_slate

mypy_clean_slate -a

And looking again at the diff:

(venv) ➜ pingouin git:(master) ✗ gd | grep 'type' | head
+latex_elements = { # type: ignore[var-annotated]
+def setup(app): # type: ignore[no-untyped-def]
@@ -27,4 +27,4 @@ from outdated import warn_if_outdated # type: ignore[import]
+set_default_options() # type: ignore[no-untyped-call]
+def _format_bf(bf, precision=3, trim='0'): # type: ignore[no-untyped-def]
if type(bf) == str:
+def bayesfactor_ttest(t, nx, ny=None, paired=False, tail='two-sided', r=.707): # type: ignore[no-untyped-def]
+ def fun(g, t, n, r, df): # type: ignore[no-untyped-def]
+def bayesfactor_pearson(r, n, tail='two-sided', method='ly', kappa=1.): # type: ignore[no-untyped-def]
+ def fun(g, r, n): # type: ignore[no-untyped-def]

Committing these with 'mypy_clean_slate second pass', and re-running mypy_clean_slate -r outputs the following:

(venv) ➜ pingouin git:(master) ✗ cat mypy_error_report.txt
Success: no issues found in 42 source files

Can now rebase / amend commits as necessary, but could now update CI/pre-commit or whatever to use mypy --strict (or a subset of its flags) going forwards.

Handling of existing comments and `pylint`

Lines which contain existing comments such as:

def ThisFunction(something): # pylint: disable=invalid-name
    return f"this is {something}"

Will be updated to:

def ThisFunction(something):   # type: ignore[no-untyped-def] # pylint: disable=invalid-name
    return f"this is {something}"

As the type: comment needs to precede pylints.

Issues

Generating report

The report generation is pretty straightforward, mypy . --strict --show-error-codes, so might not be worth having as part of this script. The user can generate the report to a text file and just pass the path to that as an argument.

Handling `-> None`

Report output for functions which don't return is pretty consistent, so these could be automated if considered worth it.

Integration with other tooling

I've tried to consider pylint comments, but no doubt there are many other arguments for different tools which aren't taken into consideration.

aucampia/mypy_clean_slate