[kvlonge]

Why

SQLBuild now runs the SQLClean Rules by default. Two engine issues surfaced in its compiler corpus and cold compile benchmarks:

• SQLCLEAN053 reported DuckDB PIVOT ... ON status USING MAX(amount) as an aggregate inside a join condition. The ON clause scan did not stop at USING, so it ran into the pivot aggregates.
• The null comparison check scanned backwards from every comparison to find its clause. On long queries that is quadratic: 12.5 ms of one 83 KB file's lint time went to this check alone.

Changes

• USING ends the clause in the SQLCLEAN053 scan. A real aggregate in a join ON beside a pivot still reports.
• assignment_operators finds SET assignments in one forward pass that tracks the latest clause opener per depth. Same stop set and same-depth rule as before.

Verification

• New SQLCLEAN053 cases: the pivot case fails without the fix and passes with it; the join ON aggregate next to a pivot still reports.
• Across 3000 compiled SQLBuild benchmark models with every Rule enabled, diagnostics are byte-identical before and after the perf change. Lint time drops by about 17% (largest file 27.8 ms to 16.5 ms). Cost per KB is now flat from small to 83 KB files.
• cargo fmt --check, clippy -D warnings, cargo test --all, ruff, ty, fensu check (0 faults), the docs skill check and the hygiene scan pass.

BEGIN_COMMIT_OVERRIDE
fix(rules): end the SQLCLEAN053 ON scan at a PIVOT USING list
perf(rules): find SET assignments in one forward pass
END_COMMIT_OVERRIDE

[cloudflare-workers-and-pages]

Deploying sqlclean with    Cloudflare Pages

Latest commit:	3f844e4
Status:	⚡️  Build in progress...

View logs