model-writtenrun 8642 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Oct 2018 - Feb 2019 · 5 entries
- Surafel Temesgen proposed SQL-standard FETCH FIRST WITH TIES in #1, extending the Limit executor to return additional rows equal to the boundary row under the required ORDER BY clause.
- Tomas Vondra's first review supplied fixes for an incorrectly scoped else block and duplicated planner calls, requested regression tests and raised possible LIMIT syntax support; Surafel incorporated the review and tests in #9 following #2.
- Andrew Gierth advocated implicit window functions and a boolean stopping condition to support both WITH TIES and PERCENT in #3 and #5, while Tomas Vondra and Robert Haas questioned the performance implications in #4 and #7.
- After the rebase in #10, Tomas found nondeterministic regression output in #11, and Surafel made the tests stable by selecting only the ordering column in #12.
- Tomas proposed estimating the extra tied rows from average group size in #13 and #15, which Surafel implemented in #16 before subsequent compatibility updates in #18 and #20.
Mar - Apr 2019 · 2 entries
- Tomas's next review corrected documentation and serialization and proposed avoiding unnecessary tuple copies in #22 and #23; Surafel incorporated those changes and standardized names in #24.
- After Tomas questioned the estimate calculation in #25, Tom Lane argued that uncertain cardinality corrections were unnecessary for the initial feature in #26, and Tomas removed them while fixing tuple-bound notification and descriptor handling in #28.
Jul 2019 · 1 entry
- Following the July rebase in #31, Erik Rijkers demonstrated an assertion failure with FETCH FIRST 1 WITH TIES in #32, and Surafel added the missing initial boundary-tuple copy and a regression test in #33.
Sep 2019 · 1 entry
- Álvaro Herrera requested better grammar representation, clearer enum names and SQL feature reporting in #34; Surafel delivered those changes and deparser support in #37.
Nov 2019 - Jan 2020 · 5 entries
- Álvaro refreshed the patch and corrected documentation and a deparser typo in #40, while explaining that adding WITH TIES to the LIMIT spelling caused grammar conflicts and would require more extensive changes.
- Álvaro supplied a grammar cleanup delta in #43 and an expanded full series in #45, eliminating redundant structures and allocations and requesting a dedicated executor state for processing ties.
- Surafel introduced the new executor state in #46, but Álvaro's cursor tests exposed incorrect backward scanning in #47, leading to boundary-position fixes and cursor regression tests in #48.
- Andrew announced a separate alternative implementation in #50, and Tomas called for a decision and performance comparisons in #51; the eventual commit explicitly deferred that unfinished alternative from thread #41585 in favor of incremental development.
- Surafel adopted Álvaro's remaining control-flow cleanup and rebased in #52, and Álvaro refreshed the declaration conflict in #53 before identifying missing deparser test coverage in #54.
Apr - May 2020 · 3 entries
- Álvaro committed 357889eb17 from #53 with deparser restructuring, view tests and a parse-analysis rejection of bare NULL row counts, explaining in #55 that testing had exposed view definitions that could not be restored.
- After landing, Andres Freund reported a GCC 10 implicit-fallthrough warning in #56, and Álvaro confirmed the fallthrough was intentional and considered extracting a subroutine in #57; no warning-fix patch appears in the supplied thread.
- Andrew argued that the NULL problem should be fixed in ruleutils rather than by rejecting an otherwise accepted value in #58 and #60; the feature remained committed, but the last message, #61, reopened discussion of that unresolved behavior.