model-writtenrun 2758 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Feb 2018 · 2 entries
- Rajkumar Raghuwanshi reported expression-reference errors when postgres_fdw combined FOR UPDATE with partitionwise joins, attaching reproducer queries in #1.
- Ashutosh Bapat traced both errors to child whole-row references being stripped out of their ConvertRowtypeExpr wrappers and proposed retaining and deparsing the conversions in #2; he supplied an archived three-patch implementation covering variable extraction, regular partitioned tables, and postgres_fdw in #5.
Apr - Oct 2018 · 11 entries
- Etsuro Fujita's review exposed additional inheritance, placeholder, and dropped-column failures in that implementation, leading to repeated corrections; by #26 he considered it ready for a committer apart from unresolved concerns about pull_var_clause() behavior and complexity.
- Etsuro proposed a different design in #27 and implemented it in #32: use plain child whole-row Vars while planning joins, restore parent rowtypes during plan creation, and derive child-join targetlists directly from parent targetlists instead of maintaining child attr_needed data.
- Ashutosh questioned targetlist consistency and projection assumptions in #33 and #39, while Robert Haas raised non-projecting nodes and Gather placement in #34; Etsuro revised topmost-relation handling and added tests in #62.
- Robert argued that changing targetlists during plan creation could invalidate costing and other derived properties in #65, and Tom Lane agreed in #71; Etsuro responded by moving the conversions into projection paths beneath Append and MergeAppend in #77.
- Robert still rejected the special targetlist semantics in #80, and Tom concluded that neither competing implementation was ready for PostgreSQL 11, proposing that partitionwise joins involving whole-row references be disabled instead in #85.
- Etsuro implemented that restriction in #86, retaining the parent-derived child targetlists and removing now-unnecessary ConvertRowtypeExpr handling; Robert accepted the release compromise in #92.
- Etsuro corrected multilevel-partition handling in #93, then incorporated Robert's assertion request from #94 and simplified eligibility checking by withholding join partition information in #95.
- Robert questioned possible partition-pruning dependencies in #96; Amit Langote clarified the existing base-relation restriction in #99, and Etsuro checked the relevant join and upper-relation callers in #100.
- The release management team warned that PostgreSQL 11 was running out of testing time in #97, and Etsuro posted the proposed commit message with the final patch in #103.
- After repository access problems delayed the push, Etsuro committed 7cfdc77023 from #103, polishing the message and removing tests made redundant by the planner simplification, as documented in #110; the fix was also backpatched to PostgreSQL 11.
- The PostgreSQL 11 open item was closed in #111, and the overlooked CommitFest entry was subsequently marked committed in #114, leaving full whole-row-reference support deferred.
Jul 2024 · 2 entries
- Nearly six years later, Alexander Pyhalov revived Ashutosh's approach in #116 as six patches intended to enable partitionwise foreign UPDATE and DELETE joins, including partial reversal of 7cfdc77023.
- Alexander's series preserved default pull_var_clause() behavior while adding opt-in conversion extraction, restored planner and postgres_fdw conversion handling, and extended DML targetlist translation and RETURNING tableoid handling; he explicitly left the correctness of ignoring varnullingrels during converted-row matching open in #116.
Oct 2024 · 1 entry
- Alexander rebased the series in #118 and refreshed it again in #120 for upstream planner and EXPLAIN changes after CI failures were reported.
Aug 2025 · 1 entry
- At the thread's last message, the six-patch restoration remained uncommitted and inactive: Alvaro Herrera moved it to PG19-Drafts and requested a fresh rebase before renewed review in #121.