[vera-rubin]

A valid table with a scalar column and a two-item FixedSizeList<Struct> writes successfully in the default 2.2 format, but to_table() rejects it with cannot read columns of differing lengths together (id=1, points=2). We expect the original one-row table, as the read guide describes.

import lance
import pyarrow as pa

points = pa.list_(pa.struct([pa.field("x", pa.int32())]), 2)
table = pa.table({"id": [1], "points": pa.array([[{"x": 10}, {"x": 20}]], points)})
assert lance.write_dataset(table, "data.lance").to_table().equals(table)

We scale dense page counts by the structural list dimensions during projection validation and preserve sparse-page normalization.

The Python reproducer and all 260 Rust file tests pass, including dense 2.2/2.3 and sparse 2.3 regression cases.

Workspace Clippy is unverified; crate Clippy without --no-deps failed on existing lance-io dead-code warnings.

— ChipLab

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[lance-gatekeeper]

✅ Gate recommendation: approve.

Scalar and FixedSizeList<Struct> columns now round-trip together without changing stored encodings. Dense normalization rejects non-divisible row counts, while sparse normalization retains its existing semantics. The file crate suite and additional nested/null, multi-page projection checks pass.

[Xuanwo]

Verified the current revision after merging main. The fixed-size-list row scale now covers stable 2.2 Structural pages and dense 2.3 Planar pages while preserving sparse page-local normalization. Non-divisible dense counts return an invalid-input error, and the Python round trip plus full current-head CI pass.