Test data

testdata/golden/ holds IPC streams written by PyArrow 19. The Mojo tests decode each file and, for most of them, encode it again and decode the result. The checked-in bytes are the PyArrow bytes. They are not rewritten by this library.

File	Columns
pa-basic.arrows, pa-basic.arrow	int32, utf8, bool, float64, including a null and a NaN. The .arrow file is the IPC file format.
pa-list.arrows	list<int32> with a null list.
pa-struct.arrows	struct<x: int32, y: utf8>.
pa-map.arrows	map<utf8, int32>.
pa-dict.arrows	Dictionary-encoded utf8 with int32 indexes.
pa-dec.arrows	decimal128(10, 2).
pa-d32.arrows, pa-d256.arrows	decimal32(5, 2) and decimal256(10, 2).
pa-time.arrows	date32, timestamp[us, UTC], duration[ms].
pa-interval.arrows	Month-day-nanosecond interval.
pa-union.arrows	Dense union of int32 and utf8.
pa-ree.arrows	Run-end encoded int32.
pa-view.arrows	String view, including one string longer than 12 bytes.
pa-lview.arrows	List view of int32.
pa-fsb.arrows	Fixed-size binary of 3 bytes.
tensor.arrows	A 2×3 int32 tensor message.

lz4-hello.lz4 and the zstd-*.zst files are codec frames. zstd-dict.zst is rejected because its dictionary id is not zero. zstd-skip.zst has a skippable frame in front of a Zstd frame. zstd-checksum.zst carries a content checksum.

testdata/schema/message.json is the JSON Schema document the code generator reads. tests/generated/Message.mojo is the checked-in output. pixi run check-generated fails if those two drift apart.

PyArrow is an oracle for local checks. The test programs in tests/ do not import it, so continuous integration does not install it.