feat: support blob computed column refresh (#4081)

Computed-column planning currently sees Blob v2 storage descriptors, so
expressions cannot consume payload bytes or preserve Blob semantics in
their outputs.

A computed declaration now derives its output field from its expression.
A direct projection of a Blob v2 field inherits the source field's Blob
metadata; other expressions retain their ordinary Arrow-inferred type.
Declarations remain ordered, so the same rule applies across chained
projections.

Refresh materializes referenced Blob inputs as `LargeBinary` payload
bytes and publishes inherited Blob outputs through Lance's Blob
conversion path. Remote requests remain within the shared namespace
contract as `{name, computed}`; the server planner is being updated in
tandem to implement the same Blob-aware planning semantics, and remote
enablement must be aligned with that server rollout.

The existing null-as-unfilled contract remains unchanged. Row-level
freshness and cell flags remain follow-up work.
This commit is contained in:
Xuanwo
2026-08-28 17:23:49 +08:00
committed by GitHub
parent 6ab3b9eb30
commit 0559108fa9
7 changed files with 821 additions and 50 deletions
+23
View File
@@ -4087,6 +4087,29 @@ def test_computed_column_rejects_transforms_and_computed_together(tmp_path):
table.add_columns({"a": "x + 1"}, computed={"b": "x * 2"})
def test_computed_column_blob_projection_inherits_semantics(tmp_path):
schema = pa.schema([pa.field("id", pa.int64()), lancedb.blob("image")])
db = lancedb.connect(tmp_path)
table = db.create_table("computed_column_blob", schema=schema)
table.add(
[
{"id": 1, "image": b"hello"},
{"id": 2, "image": b""},
{"id": 3, "image": None},
]
)
table.add_columns(computed={"image_copy": "image", "second_copy": "image_copy"})
assert table.refresh_column("image_copy").rows_filled == 2
assert table.refresh_column("second_copy").rows_filled == 2
assert table.blob_columns() == ["image", "image_copy", "second_copy"]
hits = table.search().with_row_id(True).limit(10).to_arrow()
rows = sorted(zip(hits["id"].to_pylist(), hits["_rowid"].to_pylist()))
copied = table.fetch_blobs("second_copy", [row_id for _, row_id in rows])
assert copied.to_pylist() == [b"hello", b"", None]
@pytest.mark.asyncio
async def test_computed_column_async(tmp_path):
db = await lancedb.connect_async(tmp_path)