model-writtenrun 9625 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Nov - Dec 2017 · 3 entries
- Sergei Kornilov proposed avoiding long ALTER TABLE SET NOT NULL validation scans by proving non-nullness from CHECK constraints or querying suitable indexes through SPI in #1.
- Robert Haas questioned index-scan costs and multi-column validation in #2, while Tom Lane objected to SPI operating on partially altered tables in #4; Sergei consequently removed index validation in #10, retaining a workflow of adding an unvalidated CHECK constraint, validating it separately, and then setting NOT NULL.
- Robert Haas's naming, formatting and helper-interface review in #12 produced #14, and Stephen Frost's request for multi-column tests and clearer scan-state comments in #15 produced the documented and tested version in #16.
Jan - Apr 2018 · 4 entries
- Ildar Musin's review in #17 led to documentation improvements in #18 and a debate over proving that scans were actually skipped, prompting INFO-level test output in #20.
- Alvaro Herrera and Tom Lane rejected unsolicited INFO output in #21 and #23, while Sergei explained in #22 that validation scans do not fire table-rewrite events, eliminating that proposed testing alternative.
- Sergei restored DEBUG1 messages without checking them in regression output in #27, and Ildar reported successful compilation and tests in #28, accepting coverage-based verification for the time being.
- Peter Eisentraut advocated properly catalogued NOT NULL constraints and SET NOT NULL NOT VALID in #29, but Sergei defended the smaller immediately usable optimization in #30 and maintained it through rebases ending at #33.
Nov - Dec 2018 · 1 entry
- After several quiet commitfests, Dmitry Dolgov reopened the scope discussion in #34, Alvaro supported proceeding without waiting for a nonexistent alternative patch in #37, and Dmitry identified unsupported related cases involving domains and redundant CHECK constraints in #39.
Mar - Jun 2019 · 8 entries
- David Rowley resumed detailed review in #40, supplying a fix-up that avoided redundant proof work once a scan was required and improved comments and documentation; Sergei folded it into #41 while placing controversial DEBUG1 regression testing in a separate patch.
- David argued in #42 for separating partition-specific NOT NULL assumptions from the shared proof helper, and Sergei implemented that smaller, more efficient interface in #43.
- Robert Haas requested clearer wording and safer handling of the caller's constraint list in #45, leading to #46, which copied the list before extending it.
- Robert committed both parts of #46 as bbb96c3704, with only additional wording edits, and announced the landing in #47.
- Tom reported buildfarm trouble in #49, and Sergei diagnosed unexpected statement-log output under DEBUG1 and proposed reverting that test setting in #50.
- Robert reversed only the debug-output test patch in 5655565c07, explaining in #55 that the optimization remained useful but reliable tests and optional diagnostic output still needed a solution.
- Sergei moved the proposal to lower ATTACH PARTITION messages to a separate thread in #59, and Robert marked the original feature committed in #61.
- David clarified in #67 that SET NOT NULL already used DEBUG1 and the unresolved inconsistency concerned ATTACH PARTITION, then emphasized in #71 that lowering verbosity would also remove existing scan-path test coverage.
Aug 2019 · 2 entries
- At the thread's end, Tomas Vondra sought agreement to lower verbosity in #72, but Tom stated in #73 that a replacement method for verifying the executed code path was still needed; the feature remained landed with its debug-output regression patch reverted.
- The subsequent resolution landed from thread #40288 as db43831899, using the later submission #432659 to lower partition messages to DEBUG1 and introduce dedicated TAP tests that inspect the diagnostic output.