model-writtenrun 5193 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Reid Thompson reported that PgStat contexts could be created with null parents before CacheMemoryContext existed, demonstrated the problem in autovacuum workers, and proposed creating the missing parent at the allocation sites #1.
- Zhang Mingli questioned whether these processes needed CacheMemoryContext and suggested TopMemoryContext #2; Kyotaro Horiguchi instead proposed creating the cache context during pgstat initialization #3, which Reid implemented in his revision #4.
- Kyotaro proposed eagerly allocating the statistics contexts during initialization #5, but Andres Freund questioned unnecessary allocations and recommended TopMemoryContext as a better parent #7; Kyotaro adapted his proposal to that parent #8, while Andres continued to object to eager allocation in processes that might never use the contexts #9.
- Bertrand Drouvot proposed changing only the three parent arguments #10 and retaining lazy allocation #12; after Andres emphasized the cost of reserving memory for an entire process lifetime #13, Kyotaro accepted that minimal approach #14, and Bertrand submitted the final patch #15.
- Kyotaro approved the final patch after make check-world passed with an additional assertion checking non-null parents #16, and Bertrand marked it ready for a committer #17.
- Andres committed Bertrand's patch unchanged apart from rebasing as 9d3ebba729, with a backpatch to PostgreSQL 15, and announced the push #18; Bertrand then marked the Commitfest entry committed, leaving no outstanding work in the thread #19.