gajus/slonik slonik@49.10.6

on GitHub

Patch Changes

• ff7a4ea Thanks @gajus! - Reduce heap allocations on hot query construction and result paths

  Several internal paths allocated short-lived intermediate arrays/objects that are now avoided, with no change in behavior:

  • sql.join/sql.list no longer re-validate and copy the assembled values array — the members are already narrowed to SQL tokens or primitive value expressions while the fragment is built.
  • sql.unnest no longer builds a separate values array and spreads the per-column bindings into it; the bindings array is returned directly.
  • pool.record() builds the result object in a single pass instead of Object.fromEntries(rows.map(...)), avoiding a [key, value] tuple allocation per row. The prototype-pollution safety for a __proto__ key is preserved via Object.defineProperty, and the result is still a normal Object.prototype object.
  • sql.and/sql.or reuse a hoisted filter predicate instead of allocating a new closure on every call.

  In an isolated benchmark the record transform produced roughly 3x fewer garbage-collection scavenges (and ran ~3.6x faster) on a 1,000-row result, and the sql.join value handling dropped from one array allocation per call to none. Real-world impact scales with how large and frequent these operations are; for I/O-bound workloads the gain is modest but free.

• Updated dependencies []:

  • @slonik/driver@49.10.6
  • @slonik/errors@49.10.6
  • @slonik/pg-driver@49.10.6
  • @slonik/sql-tag@49.10.6
  • @slonik/utilities@49.10.6