Feat: transpile BQ APPROX_QUANTILES to DuckDB

[treysp]

Notes:

• Bigquery does not document the specific algorithm they use, so results may differ from DuckDB (which uses t-digest)
• DuckDB does not support the BQ RESPECT NULLS option

[georgesittas]

Looks good, only thing I'd like to understand a bit more is what the results look like in source/target. Do they mostly match, with the only difference being how the estimation happens? Types and all included?

[treysp]

Looks good, only thing I'd like to understand a bit more is what the results look like in source/target. Do they mostly match, with the only difference being how the estimation happens? Types and all included?

Yep, good match other than algorithm

test_type	input_values	BigQuery	Duckdb
BigQuery BIGNUMERIC → DuckDB DECIMAL/HUGEINT	[1, 50, 100]	[1,50,100]	[1.000,50.500,100.000]
BigQuery FLOAT64 → DuckDB DOUBLE	[1, 50, 100]	[1,50,100]	[1.0,50.5,100.0]
BigQuery INT64 → DuckDB BIGINT	[1, 50, 100]	[1,50,100]	[1,50,100]
Tests with negative numbers	[-100, -50, 0]	[-100,-51,-1]	[-100,-50,0]
Tests DISTINCT modifier	Removes duplicates	[0,4,9]	[0,4,9]
Tests NULL handling	Mixed with NULLs	[1,49,99]	[1,50,99]
BigQuery NUMERIC → DuckDB DECIMAL	[1, 50, 100]	[1,50,100]	[1.000,50.500,100.000]
BigQuery DATE → DuckDB DATE	[2024-01-01, 2024-02-19, 2024-04-10]	[2024-01-02, 2024-02-20, 2024-04-10]	[2024-01-02, 2024-02-20, 2024-04-10]
BigQuery DATETIME → DuckDB TIMESTAMP	[2024-01-01 00:00:00]	[2024-01-01T00:00:00, 2024-01-01T00:00:00, 2024-01-01T00:00:00]	[2024-01-01 00:00:00+00, 2024-01-01 00:00:00+00, 2024-01-01 00:00:00+00]
BigQuery TIME → DuckDB TIME	[00:00:00]	[00:00:00, 00:00:00, 00:00:00]	[00:00:00, 00:00:00, 01:27:49.182464]
BigQuery TIMESTAMP → DuckDB TIMESTAMP WITH TIME ZONE	[2024-01-01 00:00:00 UTC]	[2024-01-01T00:00:00Z, 2024-01-01T00:00:00Z, 2024-01-01T00:00:00Z]	[2024-01-01 00:00:00+00, 2024-01-01 00:00:00+00, 2024-01-01 00:00:00+00]

[georgesittas]

Great, thanks!

[georgesittas]

Great work