model-writtenrun 5199 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jun - Sep 2021 · 6 entries
- Rahila Syed proposed publication column lists to reduce replicated data in #1, initially retaining subscriber defaults for excluded values and transmitting replica-identity columns automatically.
- Alvaro Herrera requested cached attribute bitmaps and existing lookup helpers in #4, while Tomas Vondra found inconsistent RangeVar/PublicationTable handling in #9, which Rahila corrected in #11.
- Peter Smith and Tomas Vondra argued that subscribers should need only the selected columns in #5 and #8; Rahila's #15 consequently omitted excluded attributes from protocol metadata and tuples and adapted initial synchronization.
- Alvaro objected to silently transmitting unspecified identity columns in #14, and Amit Kapila explained their operation-specific necessity in #16; explicit validation appeared in #20.
- Alvaro extracted the publication wrapper structs from #20 into 0c6828fa98 and posted the remaining feature work in #30.
- Shared parser requirements interrupted the feature: Vignesh supplied generic publication-grammar prototypes in #41 and a comma-conflict reproducer in #48, with typed fields and continuation handling discussed in #59 and agreement recorded in #73.
Dec 2021 - Feb 2022 · 11 entries
- After a two-month gap, Alvaro resumed with the rebase in #74 and restored the accidentally omitted TAP tests in #76.
- Vignesh identified missing dump, psql and validation support in #78; Peter Eisentraut supplied documentation and tests in #79 and recommended integer attribute storage and the name “column list.”
- The complete replacement in #84 adopted int2vector storage and experimented with publication-column dependency subobjects, but Alvaro rejected that machinery as too complicated in #86, and Tomas advocated ordinary dependencies in #87 and #89.
- Hou identified replica-identity mutation cases and proposed execution-time validation in #96; meanwhile, the discussion in #99, #101 and #106 established that subscriber-controlled initial COPY prevented column lists from serving as a security boundary.
- Alvaro corrected initial synchronization to consider only subscribed publications in #110, then introduced dedicated ALTER PUBLICATION ALTER TABLE SET COLUMNS syntax in #111.
- The preparatory cleanup patch in #112 landed as c9105dd366 with additional naming and diagnostic edits, leaving the reduced series in #114.
- The action-enum patch from #114 landed unchanged as 9623d89996, and #116 carried the remaining feature.
- Review revisions through #127 sorted attribute lists, rejected generated columns and limited replica-identity requirements to publications of UPDATE or DELETE, while #136 left subscriber schema compatibility to users.
- Alvaro proposed a publication-ID index in #129, Peter Eisentraut approved it in #130, and 025b920a3d added the index and used it in GetPublicationRelations.
- Peter Smith identified an unrelated completion change in his review #137; Alvaro extracted it from #127 and expanded ADD/SET and comma-list handling in 95787e849b.
- Tomas took over active development in #139, incorporating review and broader TAP tests, then expanded regression and pg_dump coverage and aligned partition behavior with row filters in #145.
Mar - May 2022 · 14 entries
- Amit's review #147 exposed DDL-validation gaps and locking hazards, leading #148 to move replica-identity checks into execution-time and relcache handling; Peter supplied further code and terminology corrections in #149.
- Testing exposed publication-order-dependent ancestor selection, so Tomas added a separate fix in #154; its tests also revealed incorrect row-filter publication selection, addressed in #158.
- Amit refined both fixes in #166, Wang Wei supplied a single-query column-list lookup in #162, and Tomas integrated these changes into #167 before Hou added regression tests in #169.
- The ancestor-selection code and tests from #169 landed as c91f71b9dc, with expanded tests covering both publication orders.
- The row-filter code and tests from #169 landed with light comment edits as 5a07966225, restricting filters to publications containing the chosen ancestor.
- The separate problem of overlapping ancestor and descendant initial synchronization was explicitly deferred in #172 and #173, while #181 reposted the main feature after the two preparatory fixes landed.
- Tomas simplified dependencies and dropped dedicated SET COLUMNS syntax in #186, preferring consistency with row filters; Amit clarified existing SET TABLE capabilities in #193, and Alvaro improved the corresponding example in #199.
- An intended warning fix was accidentally omitted from the unchanged resend #187 and actually appeared in #188; Peter's test-comment corrections in #203 were then integrated into #204.
- The main feature landed from #204 as 923def9a53, with unused helpers removed, old-update-tuple filtering corrected and tests adjusted, as announced in #206.
- Noriyoshi Shinoda reported misplaced prattrs catalog documentation and supplied #209, which landed unchanged as 41b00f8e60.
- Tom Lane reported intermittent column-list test failures in #212; Tomas diagnosed a missing tablesync wait in #213 and fixed it without an attached patch in 404f49338f.
- The ancestor-selection tests also exposed a preexisting worker-restart hang: Amit investigated it with diagnostic instrumentation in #190, while the discussion in #218 and #220 deferred a substantive fix rather than merely increasing worker limits.
- Masahiko Sawada reproduced the hang in #223, and Amit proposed restricting the sync-worker limit check to actual sync workers in #227; Tomas and Masahiko approved the change in #228 and #229, leading to dd4ab6fd65.
- Tomas confirmed that relevant buildfarm failures had stopped in #232, Jonathan Katz removed the issue from open items in #233, and Amit independently confirmed the resolution in #234.
Jul - Sep 2022 · 5 entries
- Peter Smith revived the missing chapter documentation in #235 and submitted a separate two-patch series in #238, covering replica-identity rules and a new Column Lists section.
- Review by Vignesh, Erik Rijkers and Amit refined operation-specific restrictions, examples and synchronization caveats across #240, #243, #246 and #248.
- The replica-identity correction from #248 landed as 376af68611, leaving only the chapter addition in #250.
- Shi yu tested the examples and requested corrections in #251, and Amit removed misleading refresh guidance in #253; the resulting #254 landed unchanged as f98d074245.
- The thread ends with the documentation landing announced in #256 and Peter's thanks in #257; no submitted feature or documentation patch remains pending, while dedicated alteration syntax and the separately identified overlapping-sync issue were left outside this work.