Feature request: support for transpiling "distinct on" (eg: postgres >> redshift)

This is a useful PostgreSQL construct that Redshift (sadly!) still does not have support for; for example, the following would return the most recent single row per id-group (as determined by the some_timestamp sort order).

-- PostgreSQL
SELECT DISTINCT ON (some_id) AS some_id, some_value, some_other_data
FROM some_table
ORDER BY some_timestamp DESC

Rewriting for Redshift (and/or any other backend that does not support “DISTINCT ON”) would typically be done as a select from a nested subquery that uses the row_number window function:

-- Redshift
SELECT some_id, some_value, some_other_data
FROM (
    SELECT ROW_NUMBER() OVER (PARTITION BY some_id ORDER BY some_timestamp DESC) as r, * FROM some_table
)
WHERE r = 1

(I imagine this one isn’t as straightforward as a function-rewrite, but it’s a query pattern I find myself coming across quite frequently, so good to have 😉

[barakalon]

Alright, this should just work now. No need to explicitly call the transform.

[barakalon]

Ah I think I did…