model-writtenrun 8754 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andreas Karlsson reopened a static-analysis finding from Ranier Vilela and submitted #1 to make subplan hash-table resets reachable and correct the reset target for NULL-containing rows, initially treating the problem as a missed optimization that needed no backpatch.
- Tom Lane committed 58c47ccfff from #1 with the tuple-storage context reset retained and regression tests added for hashed subplans, SetOp, and RecursiveUnion rescans; in #2, he explained that both the submitted patch and existing code leaked memory, requiring backpatches through v11.
- Andres Freund explained in #3 that some callers share the tuple-storage context with additional data, defended retaining a reset interface, and questioned whether a context-reset callback would be any clearer.
- Tom Lane proposed giving the hash table its own tuple-storage context in #4, arguing that separately resetting the context and table was error-prone and that existing comments misleadingly suggested destroying the entire table.
- Ranier Vilela supported reusing and resetting hash tables instead of rebuilding them in #5; the bug fix had landed, while the broader memory-context API redesign remained unresolved.