model-writtenrun 9305 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Alex Aktsipetrov reported that SELECT FOR UPDATE referencing a trigger transition table failed with a missing named-tuplestore error on PostgreSQL 12beta2, while omitting row locking avoided the failure #1.
- Alex supplied a one-line fix making the EvalPlanQual executor state inherit its parent's QueryEnvironment #3; Thomas Munro reproduced the failure and reposted the same assignment beside the other inherited fields #5.
- Thomas connected the failure to the older report in thread #69045 and explained that unconditional EPQ initialization exposed a pre-existing bug without requiring concurrent updates #4 #6.
- Tom Lane approved the revised fix and requested credit for the related bug report, while raising a separate concern that eagerly initializing EPQ imposed unnecessary work when no update conflict existed #7.
- Thomas committed Alex's fix unchanged from #5 as f5825853e3, with backports through PostgreSQL 10, and confirmed the push while suggesting cheaper slot initialization #8.
- At the thread's final message, the transition-table fix had landed, while the overhead question remained open: Andres Freund explained why tuple-slot support reused EPQ infrastructure and proposed separating minimal slot setup from preparation to execute EPQ #9.
- The later implementation was developed in thread #69348 and landed as 27cc7cd2bc, separating slot availability from full EPQ initialization, fetching non-locking rowmarks on demand, and expanding isolation tests.