[ohmayr]

Yields one empty RecordBatch with the query's schema when Arrow results have no rows, so to_arrow() no longer returns a table without columns.

• Uses the Arrow schema BigQuery sent, else the schema from jobs.getQueryResults, else no columns, like _EmptyRowIterator.
• Covers results that fit in the first page, an empty stream, and an empty destination table.
• Adds unit tests and system tests for empty queries and DDL/DML statements.

[gemini-code-assist]

Code Review

This pull request ensures that Arrow queries returning zero rows still yield a single empty batch so that callers can retrieve the schema. It introduces a helper function _empty_batch and updates download_arrow_from_job to track whether any batches have been yielded, along with adding corresponding unit and system tests. A review comment suggests guarding the empty batch yield when reader is None with if not yielded to prevent appending an extra empty batch to a non-empty result set.

[gemini-code-assist]

If reader is None but yielded is already True (e.g., some batches were already yielded in Step 1), yielding an empty batch here would append an extra empty batch to a non-empty result set. We should guard this with if not yielded: to ensure we only yield an empty batch when no rows have been yielded at all.

Suggested change

	if reader is None:
	# An empty destination table has no streams.
	yield _empty_batch(pa_schema, row_iterator._schema)
	return
	if reader is None:
	# An empty destination table has no streams.
	if not yielded:
	yield _empty_batch(pa_schema, row_iterator._schema)
	return

[ohmayr]

reader is only None on the explicit-destination path, and explicit destinations always go through jobs.insert, where we wait with getQueryResults(maxResults=0) so there's no inline Arrow batch. yielded is always False there.

[tswast]

Is this necessary? Do we not get an arrow schema in the jobs.query response if the query completes with empty results.

[ohmayr]

I filed b/573768143. Until this is addressed, we'll need to ensure we return an empty table when the response is empty.

I believe the empty schema case (this function) is not relevant to jobs.query but to the case where ReadRows is called after jobs.insert.

Correction: you're right that jobs.query returns arrowSchema when the query completes, even with 0 rows, so this isn't used there. It's for empty results read from the job's _default stream: ReadRows sends no messages at all for 0 rows, so there's no schema. That happens after jobs.insert with an anonymous destination, and also after jobs.query when the query doesn't finish in time and we wait. Checked all of these against BigQuery. Explicit destinations get the schema from the read session.