A Polars expression plugin, generated from the polars-plugin-101 workshop template with cargo-generate.

Everything here is the boilerplate that the workshop's Step 1 and the scaffolding parts of Steps 2 and 3 would have you type out by hand. The expression logic is up to you.

• Cargo.toml — Rust dependencies: polars, pyo3, pyo3-polars, serde
• pyproject.toml — maturin build settings, with module-name wired to {{crate_name}}._internal
• src/lib.rs — the _internal PyO3 module and the Polars global allocator
• src/expressions.rs — empty; your #[polars_expr] functions go here
• {{crate_name}}/__init__.py — where you register each expression with Polars
• {{crate_name}}/_typing.py — IntoExpr / IntoExprColumn aliases, so we never import the private polars._typing
• {{crate_name}}/_internal.pyi — type stub for the compiled module

Set up a virtual environment and install the build tools:

uv venv .venv
source .venv/bin/activate
python -m ensurepip --default-pip
uv pip install polars maturin

The ensurepip line is needed because maturin develop cannot find pip otherwise.

Then build:

maturin develop

That compiles the Rust, drops _internal.abi3.so into {{crate_name}}/, and installs the package into the virtual environment.

Add a function to src/expressions.rs — the file starts out empty, so bring the imports with you:

use polars::prelude::*;
use pyo3_polars::derive::polars_expr;

#[polars_expr(output_type=Float64)]
fn to_farenheit(inputs: &[Series]) -> PolarsResult<Series> {
    let ca = inputs[0].f64()?;
    let out: Float64Chunked = ca.apply_values(|deg_c| deg_c * 9.0 / 5.0 + 32.0);
    Ok(out.into_series())
}

Register it in {{crate_name}}/__init__.py:

def to_farenheit(expr: IntoExpr) -> pl.Expr:
    """Converting Celcius to Farenheit."""
    return register_plugin_function(
        plugin_path=PLUGIN_PATH,
        function_name="to_farenheit",
        args=expr,
        is_elementwise=True,
    )

Run maturin develop again, and use it:

import polars as pl
from {{crate_name}} import to_farenheit

df = pl.DataFrame({"deg_c": [0.0, 21.5, 100.0]})
print(df.with_columns(deg_f=to_farenheit("deg_c")))

Note that #[polars_expr] functions are not added to the module in lib.rs. The macro exports each one as a C function in the compiled library, and Polars looks it up by name when register_plugin_function runs — which is why function_name has to match the Rust function name exactly.

Work through the workshop README for multi-column inputs, dtype dispatch, column-wise accumulation and kwargs.

If you want CI for wheel builds, maturin generate-ci github will write a workflow for you.