model-writtenrun 5630 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Apr - Sep 2019 · 7 entries
- Fabien COELHO proposed SHOW_ALL_RESULTS in #1, replacing PQexec with asynchronous submission and explicit processing so combined queries could display every result, while initially preserving the old default.
- Aya Iwata tested the first patch and found misplaced warnings and a regression-test whitespace error in #2; Fabien added buffered notice handling, cleaner timing boundaries, help text, and expanded tests in #3.
- Daniel Verite found that errverbose no longer retained errors in #4, leading Fabien to preserve error results in #6.
- Daniel and Peter argued for displaying every result without a configuration switch in #4, #9, and #11, and Fabien removed the switch in #12.
- The timing discussion in #10 and #21 established that combined-query timing includes display of earlier results because server execution overlaps client processing.
- Kyotaro Horiguchi reviewed documentation, FETCH_COUNT interaction, COPY return handling, and timing in #15 and #20; Fabien clarified comments, fixed failure propagation, and updated regression expectations through #21, leaving the pre-existing FETCH_COUNT issue separate.
- Alvaro Herrera rebased and formatted the patch in #22, while Fabien continued his own earlier branch to address vignesh C's function-comment and duplicate-comment findings in #24 and #26.
Jan - Feb 2020 · 1 entry
- Tom Lane challenged the compatibility cost of removing the old behavior in #29, and Alvaro recommended retaining an opt-out in #31; the unresolved design question delayed the feature.
Jun - Jul 2020 · 2 entries
- Fabien extended the implementation to display multiple results under watch in #39, but Thomas Munro reported a crash-restart test failure in #40, and progress stalled.
- Peter first extracted Fabien's distinct BEFORE/AFTER COPY trigger notices from #39 into b3a9e9897e, then revived the feature in #43 with SHOW_ALL_RESULTS restored and enabled by default, crash-test handling repaired, and test changes reduced.
Feb - Apr 2021 · 4 entries
- Fabien tested Peter's revised patch successfully in #44, and Peter landed it as 3a51306722, additionally correcting COPY notice ordering as described in #45.
- Shi yu reported that Ctrl-C no longer canceled queries in #47; Fabien supplied #48, and Michael Paquier's review led to revised cancellation-reset placement in #55 and #57.
- Tom Lane found broken savepoint handling and questionable notice-buffer code in #60 and #61, then duplicate crash messages in #66; Nathan Bossart reproduced an infinite START_REPLICATION error loop and offered an inline correction in #67.
- Despite Alvaro's preference to repair the feature during the current cycle in #68, Peter reverted the first landing with fae65629ce and announced postponement to the next development cycle in #73.
Jun - Oct 2021 · 8 entries
- Fabien resumed with #74, adding transaction tests, centralized cancellation cleanup, and indexed notice buffers, but Peter showed in #75 that the tests did not exercise the savepoint regression.
- Peter supplied an independent cancellation TAP test in #78, and Fabien incorporated it while adapting the feature to watch-pager output and improving cancellation behavior in #84.
- Pavel Stehule explained in #85 and #87 that pspg did not support multiple result tables without better synchronization markers, leaving that pager capability outside the feature.
- Peter extracted the cancellation test from #84 into 5b3f471ff2, adjusting its harness and comments, and reported initial buildfarm portability failures in #90.
- Peter's new transaction and ECHO tests in #91 exposed the still-broken COMMIT AND CHAIN path; Fabien incorporated them and added explicit transaction-status propagation in #92, after which the unchanged tests landed separately as 14d755b000.
- Reviewing the remaining feature, Peter found that replication error loops and duplicate crash messages still persisted and supplied additional tests and static-analysis cleanup in #94.
- The basic command-line and help tests from #94 landed as c0280bc3ed, and its unsupported-replication-response test landed separately as 67c069848a.
- The scan-build cleanup from #94 landed as 390edeeb57, stabilizing timing checks and removing dead stores so static analysis could better assess subsequent feature revisions.
Nov 2021 - Apr 2022 · 11 entries
- Fabien addressed the replication loop and duplicate errors in #97 with COPY-state handling and a proposed PQclearErrorMessage API, then moved the deliberate-disconnection test into TAP in #102 after Peter objected to a successful regression test exiting abnormally.
- Andres Freund identified a use-after-free behind Windows failures in #103, and Fabien moved status extraction before result disposal in #104.
- Peter questioned adding a public libpq API solely for an error-message workaround in #105 and proposed a substantially different PQexecMulti design in #107; Fabien raised asynchronous-delivery, buffering, COPY, and ownership concerns in #108, and that alternative did not land.
- Peter's independent TAP series in #110 landed as fbee60f6a4 for helper cleanup and 9028cce426 for timing, encoding, and notification tests, while watch and pager coverage remained incomplete.
- Fabien continued the asynchronous psql implementation in #112 and removed the proposed libpq workaround in #113, initially accepting duplicate error output in the adjusted test from #116.
- Peter continued challenging the duplicate crash output in #117 and #120; Fabien then identified the remaining mistake and switched from connection-level PQerrorMessage to PQresultErrorMessage in #122, also making transaction tests specifically exercise combined queries.
- Peter extracted the crash-behavior TAP test from #122 into 8cd7627c7b, with polished assertions and pre-feature expected output, leaving only the feature-specific output adjustment in #123.
- Peter extracted preparatory SendQuery restructuring from #123 as 8910a25fef, preserving synchronous execution while separating query processing and savepoint-status reporting.
- After Fabien rebased in #124, Peter separately extracted HandleCopyResult as 5519d5affd and explained in #125 that he was committing the work in pieces.
- The remaining feature from #124 landed as 7844c9918a, with existing naming and structure retained where possible; Peter announced completion and deferred optional stylistic changes in #127.
- At the end of the thread, the feature and its preparatory work were committed, but Andres reported noisier pg_regress database-setup output in #128; Peter and Andres favored suppressing that output with quiet psql execution in #129 and #130, without an attached or candidate fix here.