model-writtenrun 2548 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jan 2021 · 3 entries
- Vignesh C proposed publishing all tables belonging to selected schemas in #1, and Bharath Rupireddy's review in #4 and #5 established the need to handle future tables, schema rename/drop and unsupported relation types.
- The initial schema-name storage became a dependency-aware pg_publication_schema catalog in #19, allowing schema deletion to remove publication membership correctly.
- Rahila Syed reported that actual replication was missing in #21, prompting pgoutput support and subscription tests in #23, with tests and documentation separated for review.
May - Jul 2021 · 4 entries
- After a four-month gap, Vignesh returned with #24, combining a rebase with publication-kind tracking, CURRENT_SCHEMA support, psql displays and tab completion.
- A skip-table extension appeared in #31, but Vignesh removed it in #37 to concentrate on making schema publication committable.
- Tang's ADD/DROP schema failures were addressed in #47, and Hou's cache-action and invalidation fix in #54 was incorporated into #56 after Greg also found missing UPDATE and DELETE replication.
- July reviews corrected public-schema dump behavior, publication-kind transitions, duplicate schema handling and fmtId buffer reuse; Tang contributed regression, pg_dump and subscription-test improvements in #90, incorporated into #91.
Aug 2021 · 7 entries
- Sawada's cross-schema partition report in #102 and Mark Dilger's privilege-bypass demonstration in #112 drove descendant-table handling and superuser restrictions in #124.
- Peter Eisentraut challenged mutually exclusive table/schema publication kinds in #122, while Tom Lane argued in #130 that explicit ALL TABLES IN SCHEMA wording would preserve meaning if sequence publication were added later.
- Tom developed a single comma-separated object-list syntax in #134 and #137, and Vignesh implemented mixed table/schema publications with a new parser representation in #140.
- Sawada questioned inconsistent CREATE and ALTER capabilities in #139, leading to a separate matching ALTER-syntax patch in #143 that was subsequently folded into the feature.
- Amit's invalidation review led to a reusable prerequisite in #154, which also moved subscription tests into their own file following Hou's request in #150.
- Amit proposed a view exposing publication objects in #147, and Vignesh added pg_publication_objects as an optional patch in #170.
- Greg questioned overlapping explicit-table and schema membership in #161 and #165; Amit chose to reject that overlap in #178, and #185 implemented the restriction while renaming the catalog to pg_publication_namespace and separating client, tests and documentation.
Sep 2021 · 6 entries
- Hou's parser simplifications were incorporated into #200, after which Amit supplied qualified-name handling in #201 and Hou supplied a documentation delta in #203.
- Tang found missing ALTER ADD overlap checks and incomplete SET replacement in #202; Amit and Tom agreed that SET must replace both membership lists in #204 and #205, implemented in #208.
- Concurrency reviews led to stronger object locking and publication serialization in #217 and #226, protecting membership changes against concurrent deletion and alteration.
- After the prerequisite landed, Hou supplied the rebased remainder and two refactoring top-ups in #239; Vignesh integrated them in #242, separating table/schema mutation and checking inexpensive failures before locking.
- Vignesh incorporated Alvaro's parser-postprocessing suggestions and offered alternative grammar in #250, then integrated it and fixed ALTER TABLE SET SCHEMA overlap enforcement in #255.
- Further review distinguished explicitly published tables from implicitly included partitions, with #268 refining the restriction so legitimate partition moves remained possible.
Oct - Dec 2021 · 14 entries
- Amit described additional parser and node-comparison fixes in #272, which Vignesh incorporated into #273; subsequent work in #281 aligned unlogged-table behavior with ALL TABLES and introduced the dump-mask check later revisited in this thread.
- Partition testing exposed missing and duplicated data, leading Hou to supply a top-up in #302 and a cleaner integrated implementation in #305 that filtered publication tables centrally.
- Tang caught a superuser-check regression in #308, and Vignesh corrected it with regression coverage in #313.
- Amit narrowed the partition fix to schema-related cases in #317 while a general duplication issue remained under separate discussion; Greg then identified an unrelated-schema edge case in #321, fixed in #322 and tested successfully in #324.
- Following Amit's request in #317, #322 combined the core feature, client support, regression tests and documentation while retaining potentially timing-sensitive TAP tests and the optional view as separate patches.
- After final review and a compilation correction, Amit committed the first patch of #333 as 5a2832465f, with light formatting, frontend allocation and catalog-version adjustments; he announced the landing in #335.
- Vignesh resent the remaining tests and view in #336, incorporating offline comments from Hou and Greg, and the TAP patch landed unchanged as 6b0f6f79ee, credited to Vignesh and Haiying Tang.
- While working on sequence replication, Tomas Vondra questioned ambiguous REL and CURRENT_SCHEMA enum names in #340 and proposed separating object type from schema selection in #346; Amit favored limited renaming in #349, and Tomas agreed in #350.
- The optional pg_publication_objects view lacked satisfactory ALL TABLES behavior and its advantage over existing catalogs remained disputed; Sawada recommended waiting for sequence support in #347, and Amit agreed in #348.
- Vignesh posted the enum cleanup in #352, revised naming after Hou and Peter's reviews, and added Peter's CREATE-versus-ALTER error-message correction in #361.
- Peter also identified an unused SQL alias in #364, and Amit incorporated that one-line correction alongside #361 in b3812d0b9b, confirming the combined fix in #365.
- Vignesh closed the feature's Commitfest entry in #366, explicitly leaving the publication-object view for consideration after sequence support.
- Alvaro revisited the catalog design in #367, proposing one membership catalog; Amit defended separate catalogs on row, index and scan-cost grounds in #368, and Vignesh supported their normalization in #369.
- All four candidate changes landed, while the optional publication-object view remained deferred under the decision recorded in #366 and the earlier skip-table extension was left outside the completed feature.
Dec 2024 · 3 entries
- Tom revived the thread in December 2024 with #370, showing that pg_dump inconsistently preserved public and extension schemas, omitted information_schema and silently changed publication membership when objects were excluded.
- Amit and Vignesh supported removing both schema and table dump-mask filters in #371 and #372; Arthur warned about restore failures for excluded schemas in #373, while Vignesh argued for consistency with other retained dependencies in #374.
- Vignesh supplied the pg_dump fix in #375; Arthur demonstrated the changed restore behavior in #376, and Vignesh explained the intentional dependency-preserving policy in #377, which Amit endorsed in #378.
Feb 2025 · 1 entry
- The thread ends with revised tests in #381 after Amit's request in #380; the next day, that patch landed with light test and comment edits as 4aa6fa3cd0, without backpatching because it changes behavior.