speed-up-imports.md

Problem

For CLIs in large repos or using larger third-party libraries (ie, Django python manage.py and pandas), imports are eagerly loaded at runtime. This leads to slow startup times for even basic things like running a --help command, because a bunch of unrelated modules are loaded.

ie,

import pandas as pd # slow import on initial load

def transform_df(df: pd.DataFrame) -> pd.DataFrame:
    return df.groupby(["state", "organization"], as_index=False).size()

Solution

One workaround for this is to use "inline imports": move imports within the scope of a function where the library gets used. Then use linting rules to ban module-level imports of the offending libraries.

from typing import TYPE_CHECKING

if TYPE_CHECKING:
    import pandas as pd

def transform_df(df: "pd.DataFrame") -> "pd.DataFrame":
    import pandas as pd # noqa
    return df.groupby(["state", "organization"], as_index=False).size()

Python PEP 810 (discussion) introduces new syntax to avoid this workaround by explicitly declaring lazy imports like below.

lazy import pandas as pd

def transform_df(df: pd.DataFrame) -> pd.DataFrame:
    return df.groupby(["state", "organization"], as_index=False).size()