Getting started

dartaframes gives Dart applications native Polars queries and an owned Apache Arrow data model. Queries stay lazy until collectSync() or collect() runs them.

The package and native binaries are not published yet. Follow Build from source and use Polars.open(path) today. Verified releases will use Polars.native() without a local Rust toolchain.

Scan CSV

Use the full path to the local .dylib, .so, or .dll:

import 'package:dartaframes_polars/polars.dart';

void main() {
  final polars = Polars.open('/path/to/libdartaframes_polars_ffi.dylib');

  final frame = polars.scanCsv('people.csv').head(5).collectSync();
  print(const OwnedBatchJsonCodec().encode(frame.exportSync()));
}

scanCsv supports headers, separators, row limits, skipped rows, and date parsing. Because it creates a LazyFrame, later filters and projections can be pushed into the scan by Polars.

final adults = polars
    .scanCsv('people.csv', tryParseDates: true)
    .filter(polars.col('age').gt(polars.lit(17)))
    .select([polars.col('name'), polars.col('age')])
    .collectSync();

Scan Parquet

Parquet uses the same lazy query API:

final totals = polars
    .scanParquet('sales.parquet')
    .filter(polars.col('year').eq(polars.lit(2026)))
    .select([polars.col('region'), polars.col('revenue')])
    .collectSync();

Use nRows for a bounded scan and parallel: false when a workload requires a serial Parquet reader.

Arrow results

exportSync() copies a frame into a pure Dart RecordBatch containing typed Arrow arrays, fields, and values. It needs no native handle and no close().

For native interoperability, frames and series also support the standard Arrow C Data Interface, and frames support the Arrow C Stream Interface. See Arrow interoperability.

Resource lifetime

Examples omit explicit cleanup because native-backed values have finalizers. This keeps ordinary scripts and application code readable.

Call the idempotent close() method when deterministic release matters, such as inside long-running services, large loops, benchmarks, or stream consumers:

final query = polars.scanParquet('events.parquet');
final frame = query.collectSync();

// Use frame...

frame.close();
query.close();

Derived Expr, LazyFrame, DataFrame, and Series values own independent native handles. Values from different Polars runtimes cannot be mixed. Close every streamed batch and the stream itself when using bounded batch streaming.

Next: Build from source · Arrow interoperability · API reference · API coverage