model-writtenrun 9588 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- David Rowley reported an assertion failure during foreign-key validation using a different table owner, with a reproducer that reached the slow validation path in #1.
- Andres Freund initially suspected his own changes in #2, then traced the failure to requesting a materialized tuple that no longer retained the buffer required by the snapshot check in #3.
- Andres's patch in #3 disabled materialization and added a per-tuple memory context, an interrupt check, and initialization of tg_newslot; he also questioned the lack of regression coverage and cleanup in this path.
- Tom Lane committed the tuple-materialization correction alongside storage checks and regression coverage as 46e3442c9e, using the later matching submission in thread #40399.
- Andres committed the remaining cleanup and defensive checks as 4c9e1bd0a3, moving the memory-context switch outside the loop, and confirmed in #4 that both portions had landed.