model-writtenrun 8960 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Dec 2018 · 2 entries
- Filip Rembiałkowski proposed dropdb --force in #1, initially disabling new connections through pg_database and terminating existing sessions from the client, while acknowledging unfinished failure handling and limited tests.
- Pavel Stehule supplied an independently developed SQL-level FORCE implementation in #2; Marti Raudsepp favored combining the approaches on the server in #3, while Tom Lane warned that background workers could bypass datallowconn in #5.
Mar - Nov 2019 · 22 entries
- Filip combined Pavel's server patch with the client option, documentation and a test in #8, then fixed a Windows syntax error and unused code after Thomas Munro's review in #9 and #10.
- Ryan Lambert, Ibrar Ahmed and Anthony Nowocien reported successful functional testing in #11, #16 and #17; Pavel subsequently refreshed the patch for newer sources in #19 and corrected the TAP test count in #21.
- Tom Lane's review in #23 identified missing permission checks, termination hidden inside CountOtherDBBackends, unsafe work under ProcArrayLock, layered timeouts and an insufficiently extensible SQL syntax.
- Pavel's rework in #29 introduced TerminateOtherDBBackends and a parenthesized option list, then adopted pg_terminate_backend permissions and corrected timeout documentation in #37.
- Vignesh C demonstrated that a permission failure could leave some sessions unnecessarily terminated in #38; Pavel responded with a complete permission precheck before signaling in #43, also incorporating client and test cleanup identified by Dilip Kumar and Thomas Munro.
- Amit Kapila requested separate server and client patches in #47, and Pavel began that split while moving option processing into DropDatabase in #51.
- Álvaro Herrera's trailing-option syntax was adopted in #54; after Peter Eisentraut questioned WITH in #55, the keyword became optional in #59.
- Amit's prepared-transaction testing in #58 showed that sessions could be terminated even though dropping the database would fail, so #59 added a precheck rather than rolling back prepared transactions and completed the removal of client code.
- Further review produced psql completion, grammar cleanup, pluralized errors and syntax regression tests in #66 and #71, with Pavel reporting successful tests under heavy connection load.
- Amit challenged the extra fixed sleep after signaling, and Pavel removed it in #75, retaining CountOtherDBBackends as the existing mechanism for waiting and checking remaining connections.
- Amit raised a PID-reuse race while polishing the patch in #77, prompting Pavel to try SendProcSignal and a new interrupt handler in #81.
- Tom Lane argued in #84 that the proposed complexity did not justify departing from established signal handling, so Amit returned to the direct-signal version and documented the race in #90.
- After a prerequisite subscription-check fix discussed in #95, Amit rebased and polished the server patch, retaining its syntax tests with revised naming in #97.
- The server feature landed as 1379fd537f from #97, with only test-comment edits, and Amit invited the deferred client patch in #98.
- Pavel restored dropdb --force in #100, acknowledging Filip's earlier work; Vignesh's reviews led to documentation, SQL-placement and help-text refinements through #108.
- Amit simplified the client documentation in #111 and recommended a separate asynchronous session-termination test based on the crash-restart test machinery, which Pavel supplied in #115.
- The client option landed unchanged from #111 as 80e05a088e, leaving the additional functional tests outstanding as noted in #116.
- Vignesh identified duplicated pump_until code in #118 and took over the shared-helper extraction in #120, producing a two-patch series containing Pavel's tests and a reusable TestLib.pm routine.
- Amit requested removal of duplicate SQL-versus-client coverage in #122; after #126 removed the wrong scenario, #128 retained the dropdb test and restored existing recovery-test formatting.
- The shared helper landed as 290acac92b, taking its constituent patch unchanged from #128; Amit then requested consolidation and renaming of the remaining force tests in #129.
- Vignesh submitted the remaining tests in #130, moving existing force coverage into 051_dropdb_force.pl; they landed with comment-only edits as 8a7e9e9dad.
- Juan José Santamaría Flecha reported Windows buildfarm failures in #132, and Vignesh proposed removing an unnecessary PID comparison affected by carriage returns in #134; the supplied candidates do not establish a landing for that follow-up.
Feb 2020 · 1 entry
- Amit closed the commitfest entry as committed in #136, explaining that the feature had already landed and that the separate discussion about Windows socket-closure behavior had not established whether further changes were needed.