Use octet_length instead of length in ClickBench SQL queries

[alamb]

Is your feature request related to a problem or challenge?

DataFusion's ClickBench benchmark queries currently use length(...) for string-length aggregations:

• benchmarks/queries/clickbench/queries/q27.sql
• benchmarks/queries/clickbench/queries/q28.sql

In DataFusion, length is an alias of character_length, which returns the number of characters in a string. DataFusion's octet_length returns the length of a string in bytes.

This differs from the benchmark semantics used by ClickHouse and DuckDB in ClickBench:

• The upstream ClickBench ClickHouse queries use length(URL) / length(Referer) in Q27 and Q28: ClickHouse/ClickBench/clickhouse/queries.sql. In ClickHouse, length is byte-oriented; the ClickHouse string function docs distinguish this from lengthUTF8, which returns Unicode code points rather than bytes.
• The upstream ClickBench DuckDB queries use STRLEN(URL) / STRLEN(Referer) in Q27 and Q28: ClickHouse/ClickBench/duckdb/queries.sql, duckdb-parquet/queries.sql, and duckdb-parquet-partitioned/queries.sql. DuckDB documents strlen(string) as returning the number of bytes in a string, while length(string) returns the number of characters.

Because byte length can be computed from string offsets, while character length must inspect UTF-8 contents, using octet_length in these benchmark queries may avoid unnecessary work and improve ClickBench performance without changing the intended benchmark semantics.

This was pointed out by @gatesn

Describe the solution you'd like

Update DataFusion's ClickBench SQL benchmark queries to use octet_length(...) where the upstream ClickBench / DuckDB benchmark is measuring byte length.

Note we would need to change in the DataFusion repo AND the ClickBench repository

Likely changes:

• AVG(length("URL")) -> AVG(octet_length("URL")) in Q27
• AVG(length("Referer")) -> AVG(octet_length("Referer")) in Q28

After changing the benchmark SQL, compare ClickBench timings before and after, especially Q27 and Q28.

Describe alternatives you've considered

Leave the queries as-is. That preserves current DataFusion SQL semantics, but likely benchmarks character-counting work that ClickHouse and DuckDB are not doing for Q27 and Q28 in ClickBench.

Add a ClickHouse compatibility alias where length means byte length. I do not think this is appropriate globally because DataFusion currently documents length as an alias of character_length.

Additional context

Related ClickBench tracking:

• [EPIC] Make DataFusion the top of the ClickBench Parquet leaderboard #18489
• [EPIC] Add SQL based benchmarking #21706
• Add clickbench SQL benchmark #22633
• Add clickbench extended SQL benchmark #22804
• add clickbench sorted SQL benchmark #22807