Test data

testdata/golden/ holds IPC bytes written by PyArrow 25. The PHP tests decode each file. The checked-in bytes are the PyArrow bytes. They are not rewritten by this library. The test runner does not import PyArrow.

File	What it holds
pa-basic.arrows	int32, utf8, bool, and float64, including a null and a NaN.
pa-list.arrows	list<int32> with a null list and an empty list.
pa-struct.arrows	struct<x: int32, y: utf8> with a null struct.
pa-dict.arrows	Dictionary-encoded utf8 with int32 indexes and a null index.
pa-file.arrow	The IPC file format for a nullable int32 column, including the footer.
pa-tensor.arrows	A 2×3 int32 dense tensor message.
pa-lz4.arrows	Two hundred int8 sevens with an LZ4 frame body.

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

The test runner also builds, in memory, a batch for each logical type the writer supports: null, signed and unsigned integers, float16/32/64, decimal32/128/256, date, time, timestamp, duration, month-day-nano interval, binary and large binary, utf8 and large utf8, fixed-size binary, list, large list, list view, large list view, fixed-size list, struct, map, dense union, sparse union, dictionary, run-end encoding, utf8 view, and binary view. It round-trips an LZ4 body and a Zstd RLE body, an IPC file footer, a V4 stream, a big-endian fixed-width buffer, and dense, COO, CSX, and CSF tensors.

PyArrow is an oracle for local checks. Continuous integration does not install it.