model-writtenrun 7152 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jan - Apr 2023 · 12 entries
- Andres Freund proposed hiding manual IPC::Run pumping behind query helpers, using an echoed separator to detect completion even for empty results and keeping session buffers and timers together; Tom Lane supported the idea in #1 and #2.
- After finding helpers inside Cluster.pm awkward, Andres introduced a separate BackgroundPsql class with query, query_safe, query_until, and lifecycle methods, converting existing tests in #3.
- Daniel Gustafsson tested the prototype, proposed better result reporting and per-query timeout restarting, and took over polishing the submission with documentation and a timer-reset option in #4 and #7.
- Andrew Dunstan proposed context-sensitive return values, instance-local timer state, private-method naming, and constructor restrictions; Dagfinn Ilmari Mannsåker supplied an undef-capable setter idiom, while Daniel and Andres favored keeping the class in a separate file in #8, #9, #11, and #13.
- Daniel incorporated that API review in #15, then fixed lost psql parameters and documentation, removed unused imports, added the previously omitted SCRAM password test, and corrected recovery-test error handling across #16 and #17.
- Andres questioned an inherited subscription-test shutdown comment, and Daniel sought clarification from its earlier committer in #18 and #19; the eventual refactor retained the comment but removed the speculative XXX annotation in 664d757531.
- Missing IO::Pty initially led Daniel to postpone committing, but Andrew identified Windows availability problems, Andres explained that the dependency must remain optional, and Tom accepted conditional coverage in #21, #22, #23, and #25.
- Daniel wrapped the password test in an IO::Pty-dependent SKIP block in #28, while Andres left a shared helper for checking interactive-session availability as a possible later improvement in #27.
- After CI passed, Daniel committed the refactor from #28 with light documentation and comment edits as 664d757531, and committed its SCRAM test unchanged as bf5a894c55, announcing both in #29; although the SCRAM test originated in thread #47079, its latest matching submission was in this thread.
- Buildfarm animals then timed out in the new password test; Daniel and Tom chose to gather evidence instead of reverting, and Daniel noticed that failing animals used older IPC::Run versions in #30, #32, and #34.
- Tom reproduced the failure and established IPC::Run 0.98 as the minimum working version, recommending a test-specific skip instead of raising the general dependency requirement in #36 and #37.
- Daniel accepted that remedy in #40 and implemented it without an attached patch as 2e57ffe12f; Tom subsequently advised leaving older buildfarm installations unchanged in #42.
Jun - Jul 2024 · 6 entries
- In June 2024, Heikki Linnakangas revived the thread because a standby regression test needed BackgroundPsql on older branches, proposing either rewriting tests, replacing the old API, or retaining it alongside a new constructor in #43, with the original introduction version corrected to PostgreSQL 16 in #44.
- Andres supported backporting, Tom favored replacing the old API, Melanie warned about callers hidden behind skipped tests, and Daniel checked external-user compatibility while raising outstanding timer issues in #45, #46, #47, and #49.
- Andrew discussed a future libpq-based Session replacement, but Álvaro, Heikki, and Robert argued that the unfinished alternative should not delay making existing tests backportable in #48, #52, #54, and #55.
- Heikki implemented the discussion-driven backport as 187b8991f7 and its branch counterparts, bringing the facility down to PostgreSQL 12 and replacing the old constructor behavior as announced in #56; the PostgreSQL 16 commit principally supplies the newer timeout option and fatal-warning setting.
- Pavan Deolasee found that the backport omitted module installation and uninstallation rules, Daniel confirmed the omission, and Pavan supplied PostgreSQL 14 and 15 patch variants in #57, #58, and #59.
- Heikki committed the installation fix from #59 as 9d7506a2f7 and its branch counterparts, confirming completion for PostgreSQL 12 through 15 in the thread's final message #60; no submitted patch remained pending.