model-writtenrun 9360 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
May - Jun 2019 ยท 14 entries
- Jonathan S. Katz opened the thread with proposed PostgreSQL 12 Beta 1 highlights and operational compatibility warnings, aiming both to inform users and encourage testing #1.
- David Rowley argued that partitioning improvements concerned broader scalability rather than faster pruning alone, while Amit Langote warned that queries scanning every partition would not receive the advertised gains #3 #5 #6.
- Oleg Bartunov clarified the SQL/JSON description, and Peter Eisentraut proposed highlighting case-insensitive and accent-insensitive collations while questioning the prominence of partition-bound expressions; Jonathan's revised list incorporated those changes #7 #9 #13 #24.
- After Bruce Momjian proposed a broad claim about handling thousands of partitions, Amit supplied wording restricting it to operations touching few partitions; Bruce adopted that sentence in 356c83795a and confirmed the change #20 #21 #22.
- Amit and David explained that the announcement should cover both pruning and tuple routing, and Jonathan subsequently accepted wording describing operations using a small number of partitions #25 #26 #46.
- The discussion expanded into release-note policy: Alvaro Herrera and Peter Geoghegan advocated enough detail for users to identify relevant improvements, while Bruce defended concise descriptions accessible to the expected reader #29 #31 #34 #43.
- Peter proposed taking the policy question to the developer meeting, and Michael Paquier suggested exposing the commit references already embedded in the SGML; Bruce agreed this was feasible, but no implementation was posted here #37 #40 #44.
- Amit identified ordered partition scans as a distinct optimization missing from the broad performance summary, prompting Bruce to add a separate sort-avoidance item in 05685897f0 without an attached patch #41 #48.
- Jonathan described announcements as an entry point into more detailed release notes and documentation, and Magnus Hagander distinguished the release announcement from the press release as another place to provide detail #47 #52.
- David resurfaced an omitted optimization, leading Bruce to propose an inline release-note item about additional optimization opportunities when only one partition remains; Bruce also reported tightening the broader partition wording to avoid implying that touching partitions meant writing them #53 #54 #55.
- David reviewed the proposed single-partition wording, explaining that eliminating single-child Append and MergeAppend nodes also benefits inheritance and UNION ALL queries, reduces overhead, and permits additional plans; he also requested greater underlying implementation credit for Tom Lane #56.
- Bruce questioned whether the item belonged in multiple sections and whether the benefit was limited to parallel plans; Amit recommended the Optimizer section, and David agreed while explaining that removing an Append node can also eliminate a Materialize node #58 #59 #60.
- After a three-week gap, Bruce posted the sole attached patch, placing a concise description of partition and UNION ALL single-child optimization in the Optimizer section #61.
- David acknowledged the final wording, and Bruce committed the attachment unchanged as b9a0724cf7; the thread ended with all three identified documentation changes landed, while broader release-note presentation ideas remained unimplemented here #62 #63.