meshagent_dart_arrow 0.41.5

Apache Arrow type and IPC primitives for Meshagent Dart clients.

meshagent_dart_arrow

Apache Arrow type and IPC primitives for Meshagent Dart clients.

This package targets Arrow IPC metadata version V5 and is validated against Apache Arrow 21.0.0 in PyArrow, JavaScript apache-arrow, and .NET Apache.Arrow smoke interop tests.

IPC Support

Supported:

• Arrow IPC stream containers through ArrowIpcStreamReader and ArrowIpcStreamWriter.
• Arrow IPC file containers through ArrowIpcFileReader and ArrowIpcFileWriter.
• Schema messages, dictionary batches, record batches, and multi-batch tables.
• Full dictionary batches emitted as needed by the writer.

Not currently supported:

• IPC body compression.
• Dictionary replacement or delta semantics.
• Tensor IPC.
• Zero-copy buffer views over the original IPC body.

Type System

The Dart model exposes Arrow schema types directly instead of Meshagent legacy data type classes. ArrowSchema, ArrowField, and the ArrowDataType subclasses model the Arrow type tree, including nested, dictionary, union, run-end encoded, temporal, interval, decimal, and binary/view types.

Record batches expose typed Arrow arrays. toRows() is a convenience projection for UI and simple SDK cases; it is not the storage model.

Arrow extension types are represented as standard Arrow field metadata:

• ARROW:extension:name
• ARROW:extension:metadata

Use ArrowField.withExtension(...), extensionName, and extensionMetadata for this metadata. Extension types are not modeled as separate ArrowDataType subclasses because Arrow IPC stores the physical storage type plus extension metadata.

Dart Value Representation

• int8, int16, int32, uint8, uint16, and uint32 decode to Dart int.
• int64 and uint64 decode to BigInt so VM and dart2js behavior stay exact.
• Floating point values decode to double.
• UTF-8 values decode to String.
• Binary values decode to Uint8List.
• date32 and date64 decode to ArrowDateValue.
• timestamp decodes to ArrowTimestampValue, preserving unit, raw value, and timezone.
• decimal128 and decimal256 decode to ArrowDecimalValue, preserving precision, scale, bit width, and scaled integer value.
• Lists, structs, maps, unions, dictionaries, and run-end encoded arrays preserve their Arrow array classes and also project through operator [] / toRows().

Examples

Dart stream write/read:

final schema = ArrowSchema([
  const ArrowField(name: 'code', type: ArrowUtf8Type(), nullable: false),
  const ArrowField(
    name: 'count',
    type: ArrowIntType(bitWidth: 32, signed: true),
    nullable: false,
  ),
]);

final batch = ArrowRecordBatch.fromColumns(
  schema: schema,
  columns: [
    ArrowArray(field: schema.fields[0], values: ['SFO', 'LAX']),
    ArrowArray(field: schema.fields[1], values: [1, 2]),
  ],
);

final bytes = ArrowIpcStreamWriter(schema: schema, batches: [batch]).write();
final table = ArrowIpcStreamReader(bytes).readTable();

Dart file write/read:

final fileBytes = ArrowIpcFileWriter.fromTable(table).write();
final fileTable = ArrowIpcFileReader(fileBytes).readTable();

Python pyarrow equivalent:

import pyarrow as pa
import pyarrow.ipc as ipc

schema = pa.schema([
    pa.field("code", pa.utf8(), nullable=False),
    pa.field("count", pa.int32(), nullable=False),
])
batch = pa.record_batch(
    [pa.array(["SFO", "LAX"], type=pa.utf8()), pa.array([1, 2], type=pa.int32())],
    schema=schema,
)
sink = pa.BufferOutputStream()
with ipc.new_stream(sink, schema) as writer:
    writer.write_batch(batch)
bytes_ = sink.getvalue().to_pybytes()
table = ipc.open_stream(bytes_).read_all()

JavaScript apache-arrow equivalent:

import { tableFromArrays, tableFromIPC, tableToIPC } from 'apache-arrow';

const table = tableFromArrays({
  code: ['SFO', 'LAX'],
  count: Int32Array.from([1, 2]),
});
const bytes = tableToIPC(table, 'stream');
const decoded = tableFromIPC(bytes);

.NET Apache.Arrow equivalent:

using Apache.Arrow;
using Apache.Arrow.Ipc;
using Apache.Arrow.Types;

var schema = new Schema(
    new[]
    {
        new Field("code", StringType.Default, nullable: false),
        new Field("count", Int32Type.Default, nullable: false),
    },
    metadata: null
);
var batch = new RecordBatch(
    schema,
    new IArrowArray[]
    {
        new StringArray.Builder().Append("SFO").Append("LAX").Build(),
        new Int32Array.Builder().Append(1).Append(2).Build(),
    },
    length: 2
);
await using var stream = new MemoryStream();
using var writer = new ArrowStreamWriter(stream, schema);
await writer.WriteRecordBatchAsync(batch);
await writer.WriteEndAsync();
stream.Position = 0;
using var reader = new ArrowStreamReader(stream);
var decoded = await reader.ReadNextRecordBatchAsync();