model-writtenrun 7151 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
May - Jun 2022 · 2 entries
- Soumyadeep Chakraborty opened the thread with #1, adapting Justin Pryzby's earlier partition-root access-method work and proposing a second patch that recursively changes existing partitions unless ONLY is specified.
- Justin favored consistency with tablespace inheritance and corrected a misleading comment in #3, while Zhihong Yu identified a tuple dereference before its validity check in #2 and #5; Soumyadeep incorporated these corrections through #6.
Mar - Jul 2023 · 3 entries
- After a long gap, Michael Paquier reviewed the work in #7 and demonstrated in #8 that dump/restore lost the partitioned parent's access method, changing the behavior of subsequently created partitions.
- Justin resumed development in #9, combining his older implementation with Soumyadeep's partition-root changes, adding pg_dump support and a cache helper, and leaving recursive alteration out of the revised proposal.
- Justin extended inheritance to intermediate partitioned tables in #12, and Michael's detailed review and revised dump test in #15 led to further documentation, assertions, and nested-partition tests in #16, followed by comment cleanup in #21.
Jan - Jul 2024 · 17 entries
- In #23, Alvaro Herrera argued that a partitioned table should be allowed to have no configured access method so new partitions can use the current default; Michael, Robert Haas, and Peter Eisentraut supported this direction in #24, #25, and #28.
- Michael substantially reworked the feature in #29 to preserve an unset relam and manage dependencies when an access method is added, changed, or removed, initially resetting the setting when it matched the default GUC.
- Justin proposed explicit DEFAULT syntax in #30, which Michael split into an independently useful prerequisite in #31; Jelte Fennema-Nio and Justin argued in #32, #35, and #36 that an explicitly named access method must remain explicit even when it currently matches the default.
- Michael's revision in #37 adopted explicit-versus-default semantics, restored CREATE TABLE USING support, and tested multilevel inheritance, while retaining the decision that existing partitions remain unchanged.
- Michael committed the incidental AlterTableCmd.name comment correction as 4f8c1e7aaf and the DEFAULT prerequisite from #37 as d61a6cad64, adding completion and regression coverage to the latter.
- Justin found that the remaining implementation mishandled repeated DEFAULT subcommands and supplied a chgAccessMethod-based correction in #38; Michael folded it into the full remaining feature and reviewed the tests in #39.
- Alvaro supplied the relative cleanup in #40, simplifying comments, grouping the AM routines, removing an unnecessary command-counter increment, and identifying related tablespace documentation work.
- The tablespace documentation portion of #40 landed separately, with revised wording and a redundant comment removed, as a0390f6ca6; the candidate 58efabdc0c is its PostgreSQL 12 backport rather than another change.
- With Michael temporarily unavailable, Alvaro posted the integrated feature in #42 and committed it with light documentation edits as 374c7a2290, crediting Justin, Soumyadeep, and Michael as authors; although thread #43767 supplied earlier work, this thread contains the latest matching complete patch.
- Alexander Lakhin reported a new foreign-table assertion failure in #44, prompting Alvaro's attached fix in #45; Justin supplied an alternative inline implementation in #46, which Alvaro adopted as e2395cdbe8.
- Alexander then demonstrated that the new regression test deadlocked against concurrent catalog maintenance in #48, Justin advised against using pg_am in #49, and Tom Lane confirmed another failure in #50; 13b3b62746 switched the test to an ordinary table.
- Michael discovered in #52 that dump/restore still erased explicit partitioned-table AM settings because SET default_table_access_method does not configure partitioned parents, and proposed emitting ALTER commands in #53.
- The all-ALTER prototype in #54 exposed binary-upgrade problems caused by rewriting stored relations, leading Michael to consider reverting the feature; Alvaro instead proposed distinguishing partitioned tables and keeping separate output routines in #56 and #58.
- Michael's replacement in #60 stored relkind in archive TOC entries, retained SET for stored relations, and emitted ALTER only for partitioned tables; Alvaro's review in #61 led to the clearer regression tests in #62.
- The dump repair from #62 landed with comment and formatting changes as f46bee346c, and Michael marked the open item fixed in #64, so the proposed revert was unnecessary.
- Justin proposed showing access methods in psql's verbose partitioned-relation listing with an inline patch in #65, which Michael attached in #67 and later committed unchanged as 978f38c771 after PostgreSQL 18 development opened.
- Tom questioned displaying the new column against older servers in #69, but withdrew the concern in the final message #71 after Michael explained in #70 that partitioned indexes still provide useful AM information; the posted feature and follow-up repairs had landed, while recursive ALTER behavior had not been carried forward.