In flight

Work proposed to PostgreSQL that has not landed, and has not been rejected.

Getting a feature into PostgreSQL takes years and several attempts; self-join elimination was committed, reverted, and committed again a release later. This page is the part of the work that is still in the middle of that. Statuses were read from commitfest.postgresql.org on 5 August 2026 — it is the authority, not this page.

In a commitfest

Show hashed SAOP decision in EXPLAINReady for Committer

proposed June 2026 · targeting PostgreSQL 20 · PG20-2

When a ScalarArrayOpExpr is evaluated through a hash table rather than a linear scan of the array, EXPLAIN currently does not say so, which leaves no way to tell the two apart from the plan alone.

commitfest #6868 · thread

Lazy LEFT NestLoop JOIN: gate rescanning of the inner sideNeeds review

proposed July 2026 · targeting PostgreSQL 20 · PG20-2 · with M. Zhilin, P. Petrov

Skip rescanning the inner side of a nested loop when join clauses that reference only the outer side already decide that no inner row can match.

commitfest #6981 · thread

Vacuum statisticsNeeds review

proposed May 2024 · targeting PostgreSQL 19 · PG20-2 · with A. Zubkov, A. Rybakina

Per-relation and per-index statistics on what vacuum actually did, so that autovacuum behaviour can be diagnosed from the server rather than guessed at from log lines.

commitfest #5012 · thread

Give plugins control at the end of CHECKPOINT to store their stateNeeds review

proposed November 2025 · targeting PostgreSQL 19 · PG20-1 · with A. Rehman

A hook at the end of a checkpoint, so an extension can persist its own state at a point where the server guarantees a consistent on-disk image.

commitfest #6224 · thread

Incremental sort cost estimation instabilityWaiting on Author

proposed July 2024 · targeting PostgreSQL 18 · PG19-Final

The cost model for incremental sort swings on small changes in the input estimate, which makes the planner's choice between sort strategies less stable than the underlying data warrants.

commitfest #5093 · thread

On the mailing list

Proposed on pgsql-hackers, with no commitfest entry under a matching title.

Try a presorted outer path when referenced by an ORDER BY prefix

proposed April 2026

For "top-N with details" shapes — an ORDER BY and a LIMIT over a normalized schema reached through several optional LEFT JOINs — build a pre-sorted outer path for the nested loop even with no index on the sort expression, and propagate the tuple bound through the outer side.

thread

Carried in enterprise forks

Shipped to customers in a fork that carries patches core has not taken. Older than the commitfest entries above and not currently registered, so no status is quoted.

What did land is on the projects page, with the release each feature shipped in.