arrow-extendr: 🏹 Polars support 🐻‍❄️

arrow-extendr
arrow
release
polars

The latest release of arrow-extendr adds support for Polars, making it straightforward to move data between R and Polars DataFrames via the Arrow C Stream interface.

Author

Josiah Parry

Published

March 25, 2026

TL;DR

The latest release of arrow-extendr adds support for Polars, bridging the gap between Polars DataFrames and R’s Arrow ecosystem.

What is arrow-extendr?

arrow-extendr is a Rust crate that makes it straightforward to pass Apache Arrow memory between R and Rust. Rather than serializing and deserializing at every boundary, Arrow lets R and Rust share the same in-memory representation. This means your extendr package can talk directly to {nanoarrow}, the {arrow} R package, DuckDB, DataFusion, and now Polars, without any copying or conversion.

Arrow is, in our view, the standard for cross-language data science. Supporting it natively from extendr, rather than routing through a single R package, is a deliberate choice.

Use Polars from extendr

Polars ships its own Arrow implementation, polars-arrow, which is separate from arrow-rs. That difference has historically made it awkward to pass data between Polars and the rest of the Arrow ecosystem. The new polars feature flag handles the translation.

Add it to your Cargo.toml:

[dependencies]
arrow_extendr = {
    version = "58",
    features = ["polars"],
    default-features = false
}
polars-core = "0.53.0"
anyhow = "1"

You can also find the latest version on crates.io.

This gives you the following conversions via the Arrow C Stream interface:

Type Direction R object
polars_core::frame::DataFrame IntoArrowRobj nanoarrow_array_stream
polars_core::frame::DataFrame FromArrowRobj nanoarrow_array_stream
polars_arrow::ffi::ArrowArrayStream IntoArrowRobj nanoarrow_array_stream
polars_arrow::ffi::ArrowArrayStreamReader FromArrowRobj nanoarrow_array_stream

Round-trip a Polars DataFrame through R

Accept a nanoarrow_array_stream from R, load it into a Polars DataFrame, and return it back to R as a stream.

use extendr_api::prelude::*;
use anyhow::anyhow;
use arrow_extendr::{FromArrowRobj, IntoArrowRobj};
use polars_core::frame::DataFrame;

#[extendr]
/// @export
fn polars_round_trip(x: Robj) -> anyhow::Result<Robj> {
    let df = DataFrame::from_arrow_robj(&x)?;
    rprintln!("{df:?}");
    df.into_arrow_robj().map_err(|e| anyhow!("{e:?}"))
}

On the R side, pass any Arrow-compatible object. Here we use {polars} directly:

library(polars)
library(nanoarrow)

df <- pl$DataFrame(a = 1:5, b = letters[1:5])
stream <- as_nanoarrow_array_stream(df)
result <- polars_round_trip(stream)

# convert back to a polars DataFrame
pl$DataFrame(as_arrow_table(result))

The return value is a nanoarrow_array_stream, so callers can convert to whatever they need: {arrow}, {polars}, DuckDB, or anything else that speaks Arrow.

What changed

Breaking changes

  • FromArrowRobj, ToArrowRobj, and IntoArrowRobj moved to the crate root (arrow_extendr). Update imports accordingly.
  • FromArrowRobj::from_arrow_robj now returns std::result::Result<Self, anyhow::Error> instead of Result<Self, ArrowError>, providing a uniform error type across both feature implementations.
  • arrow-rs is now an optional dependency behind the arrow feature flag, which is on by default. Add features = ["arrow"] explicitly if you depend on arrow-rs types.
  • ErrArrowRobj type alias removed. Use anyhow::Error directly.

New features

  • polars feature flag enabling interop with polars-core.
  • FromArrowRobj for polars_arrow::ffi::ArrowArrayStreamReader.
  • IntoArrowRobj for polars_arrow::ffi::ArrowArrayStream.
  • IntoArrowRobj for polars_core::frame::DataFrame, preserving chunking.
  • FromArrowRobj for polars_core::frame::DataFrame.

Community-driven development

This feature came directly out of a conversation with the Bioconductor community last August. The topic was SingleRust, a single-cell analysis library in Rust, and how to expose its results to R. SingleRust’s core data structure is a Polars DataFrame, which meant crossing the polars-arrow/R boundary. That thread, between myself, Mossa (@cgmossa), and Artür (@Artur-man), is what motivated Mossa to build this out.

The only way we know what matters to you is for you to tell us. We really do listen 👂🏼. We’re a small team, but when you need something, we’ll provide, or do our best at least!

Get involved:

The full changelog is on GitHub.