Limitations and roadmap

What v1 declines to do, and what is likely next.

Not in v1

• jev_* in CTEs, subqueries, HAVING, window functions, DISTINCT, set operations.
• OR jev(...). Split the query, or rewrite as jev() with a broader condition.
• Comparing probabilities in WHERE. Use jev(src, 'condition', 0.7).
• Aggregates other than count, sum, avg, min, max alongside jev.
• FETCH FIRST ... WITH TIES, FOR UPDATE.
• Judging a subquery or CTE alias. The source must be a base table in FROM.
• Cache expiry. Clear it by hand.

Known rough edges

• Column-list form needs two or more columns. jev((name), ...) parses as a plain column reference; write ROW(name).
• GROUP BY output ordering follows first appearance unless you add ORDER BY.
• The macOS Intel binary is built on GitHub’s Intel runners; report anything odd.

Likely next

• Expiry and size limits for the cache.
• OR jev(...) by judging the union and filtering in the client, behind an explicit flag because of cost.
• Streaming results for jevql serve.
• A Postgres wire proxy, so unmodified applications can send WHERE jev(...). Deliberately not in this repository yet.

Edit this page on GitHub ↗