GeezerVison

What

Support ClickHouse's ORDER BY ... WITH FILL modifier and the query-level INTERPOLATE clause, following the ClickHouse grammar: per element [WITH FILL] [FROM expr] [TO expr] [STEP expr] [STALENESS expr] (the step may be an INTERVAL literal), and after the ORDER BY list an optional INTERPOLATE [(col [AS expr], ...)] where the bare form fills all allowed columns.

Why

SELECT d, v FROM t ORDER BY d WITH FILL FROM toDate('2024-01-01') TO toDate('2024-02-01') STEP INTERVAL 1 DAY INTERPOLATE (v AS v + 1) currently fails with a ParseException on WITH. WITH FILL is the standard ClickHouse construct for gap filling in time series, so statements using it cannot be parsed at all.

How

• OrderByElement carries a new WithFill holder (its presence marks the element as filled; from/to/step/staleness are optional expressions). This mirrors ClickHouse, where the bounds belong to the order by element.
• Select (shared by PlainSelect, ParenthesedSelect and SetOperationList) carries the List<InterpolateElement> as a sibling of orderByElements, mirroring ClickHouse, where INTERPOLATE is a query-level attribute. The existing union hoisting for orderByElements is extended so SELECT ... UNION SELECT ... ORDER BY x WITH FILL INTERPOLATE (...) lands on the SetOperationList.
• Rendering in both toString() and the deparser (bounds rendered through the expression visitor), and OrderByValidator/SelectValidator walk the new expressions.
• FILL, STEP, STALENESS and INTERPOLATE are declared as non-reserved keywords (inside the NonReservedWord range), so they remain usable as plain column names (SELECT fill, step FROM t ORDER BY interpolate still parses).

Root cause

The grammar had no production for the WITH FILL modifier family and no INTERPOLATE clause; the element-level bounds and the query-level column list are modeled after the ClickHouse reference parser (bounds only parsed after WITH FILL, in fixed FROM/TO/STEP/STALENESS order).

Testing

• ClickHouseTest: 9 new tests covering the issue statement, bare WITH FILL, FROM/TO/STEP, STEP INTERVAL plus STALENESS, DESC/NULLS LAST combinations, INTERPOLATE variants (multiple items, item without AS, bare INTERPOLATE), a UNION with hoisting, a subquery, and column names using the new keywords.
• ./gradlew test: 4793 tests, 0 failures (26 skipped, pre-existing).

Performance

gradle jmh, JSQLParserBenchmark.parseSQLStatements on performance.sql, version=latest, 10 forks × 10 iterations (100 samples) on a 32-core host:

The difference is within the confidence intervals: no regression.

Notes

• compileJavacc emits one additional choice-conflict warning (12 → 13) caused by the INTERPOLATE optional in PlainSelect's clause tail; the token has exactly one valid parse at that position, and the repo already carries such warnings (tracked in compileJavacc emits JavaCC warnings for shadowed literals and unreachable DATA_TYPE branches #2403).
• INTERPOLATE () with an empty list is accepted and normalized to the bare INTERPOLATE form (both fill all allowed columns in ClickHouse).

Fixes #2467