model-writtenrun 2611 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Aug - Sep 2023 · 1 entry
- Rajendra Kumar Dangwal asked why pgoutput omitted generated-column values needed for Debezium and Kafka in #1, and Euler Taveira supported an optional, default-off way to send them in #2.
May - Jul 2024 · 7 entries
- Shubham Khanna’s initial patch #5 added pgoutput and test_decoding controls but deferred subscriptions because ordinary COPY could not transfer generated columns.
- Hayato Kuroda asked for subscription support, existing test_decoding defaults and correct column-list filtering in #6, leading Shlok Kyal to add a subscription option restricted to copy_data=false in #8.
- Masahiko Sawada insisted that initial synchronization agree with streaming in #9, while Peter Eisentraut emphasized stored-only support in #12.
- Shlok proposed general COPY changes in #23, but Peter Smith and Hayato challenged that scope in #25 and #28, so #31 used COPY (SELECT ...) within tablesync instead.
- Vignesh proposed separating virtual-column exclusions in #44, and #45 added that third patch while Shubham and Shlok maintained streaming and synchronization changes in parallel.
- Peter Smith’s reviews expanded coverage of missing and reordered columns, generated subscriber targets and older servers in #48, #58 and #78, while #61 restored test_decoding’s existing default.
- Peter proposed reusing the column bitmap instead of passing another boolean through protocol functions in #7 and #67; #72 implemented it, and #81 separated it into a fourth patch for review.
Aug - Sep 2024 · 6 entries
- TAP-test revisions exposed accidental acceptance of ordinary publisher columns targeting generated subscriber columns in #103, and #101 temporarily carried only the first patch to avoid repeatedly rebasing dependent work.
- Peter Smith supplied a revised test suite covering additional generated-column combinations and ALTER cases in #120, continuing the review of subscriber-side behavior.
- Amit Kapila argued for publication-level control in #121, Masahiko supported its flexibility in #122, and Shubham replaced subscription-level plumbing with publish_generated_columns in #124.
- Following Masahiko’s review #125, Shubham removed the test_decoding option and restored a separate initial-copy patch in #131.
- Amit settled that explicit column lists override the publication option in #142, and Masahiko proposed landing column-list support separately in #143.
- Peter Smith added a dedicated generated-column replication chapter in #161, and his objection to warning about a valid explicit column list in #167 led to removal of that warning in #173.
Oct 2024 · 10 entries
- Shubham merged streaming and initial-copy implementation and restored TAP tests in #177, while Peter grouped SGML changes into the documentation patch and incorporated Vignesh’s review in #185.
- Vignesh showed that explicit generated-column lists lost data during initial copy in #180, prompting the bitmap-preservation fix in #187.
- Masahiko reported an unpublished replica-identity problem in #184, but Amit showed it also existed without this patch in #201, and Masahiko agreed to handle it separately in #202.
- Amit objected to synthesized column lists and unnecessary per-tuple work in #204, so Vignesh reworked the implementation around relation-level generated-column state in #207 and simplified synchronization in #210.
- The split series #214 separated column-list support from the publication option, and Peter and Hayato found missing initial-copy coverage and incorrect test expectations in #216 and #220.
- Amit supplied a column-list simplification in #218, but Hou’s failing multi-publication example in #224 showed it was incorrect, and the revised implementation and initial-copy patches appeared in #226.
- Vignesh merged the column-list and tablesync patches in #235, then fixed Shubham’s older-publisher COPY failure #237 in #238, which landed unchanged as 745217a051.
- After the first landing, Vignesh rebased the remaining publication option, documentation and tests in #240.
- Peter identified stale generated-column documentation in #241, and Ajin proposed the pre-18 subscriber initial-copy caveat in #245; the revised correction #249 landed with a small wording edit as 5b0c46ea09.
- Ajin tested cross-version replication, dumping, upgrading and publication conflicts in #244 and #253, while Hayato reported a misleading generated-target error in #248 whose improvement was deferred in #264.
Nov 2024 - Feb 2025 · 13 entries
- Amit’s fix-up patches #267 and #273 helped consolidate column-list helper usage and comments, while Peter’s repeated-error test #274 was explained as existing historic-snapshot behavior in #275.
- The final option and TAP patches #276 landed together as 7054186c4e, with only the catalog-version bump added and the extended documentation still omitted.
- Peter Eisentraut reopened stored-versus-virtual ambiguity in #278, Noriyoshi Shinoda supplied missing catalog documentation in #279, and Amit acknowledged missing explicit stored-column handling in #280.
- After more than two months, Peter Smith revived the uncommitted feature chapter and catalog documentation in #283.
- Amit preferred extensible none/stored option values in #285, and Vignesh supplied the conversion in #286; the broader publish_column_types proposal in #288 was declined in #290 because column lists already handled selective publication and virtual replication remained future work.
- The combined follow-up series #291 also addressed broken psql output against older servers, and Peter’s SQL-alias approach in #299 produced the exact first patch of #303 committed as 991974bb48.
- Peter Eisentraut approved the conversion and reviewed chapter formatting in #308, and Vignesh merged the catalog documentation and test-name correction into the enum remainder in #309.
- The enum patch in #314 landed as e65dbc9927, with char parsing types retained and small comment, test and catalog-version edits described in #316.
- The same report #316 identified an older-server pg_dump buildfarm failure caused by stringifying an enum identifier, and b35434b134 exactly matches Vignesh’s fix attached shortly afterward in #317.
- Noriyoshi Shinoda corrected the catalog documentation’s column name and ordering in #319, which landed unchanged apart from whitespace as 117f9f328e.
- Peter Smith questioned the new catalog name in #320 and Tom Lane preferred pubgencols in #324, leading to Vignesh’s rename #326 and its landing with a catalog-version bump as 75eb9766ec.
- Amit considered the extra examples redundant with the behavior matrix in #329 and requested a split in #331, so Peter separated the chapter update from additional examples in #332.
- The first documentation patch from #332 landed as 6252b1eaf8; the final message #333 moved the remaining Examples proposal to a new thread, leaving the feature and principal documentation committed.