Adopting in an existing codebase

Astrolabe checks only the frames that it can type. A plain pl.DataFrame stays plain, and the checker reports nothing about it. So you can add the plugin to a large codebase, and type it one frame at a time.

Steps

1. Install and set up the plugin. With the default level, the existing code gets few new errors.
2. Type the frames where data enters the program. Give each reader a schema, or bind the frame with check_schema. See Loading data and Runtime checks.
3. Declare a Columns class for each frame that several functions share, and use it in their signatures.
4. Use an open frame for a function that reads some columns of a larger frame. See Open frames.
5. Read the errors, and fix each fault. Each error has a code, and Error codes gives the cause and the fix.
6. When the code is clean at the default level, make the level stricter. See Configuration.

Silence one line

mypy keeps its usual controls. To silence one error on one line, add a comment with its code:

df.select("venue")  # type: ignore[undeclared-column]

To turn a code off for the whole project, use the disable_error_code setting of mypy.

Change the level of one operation

allow_unproven_overrides sets the level of one operation, and wins over the global level. For example, make the project stricter, and still accept a map_elements without return_dtype=:

pyproject.toml

[tool.astrolabe]
allow_unproven = "stated"

[tool.astrolabe.allow_unproven_overrides]
map_elements = "omitted"