model-writtenrun 8312 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jul - Aug 2017 · 3 entries
- Michael Paquier opened the thread after Moshe Jacobson reported user-triggerable cache lookup failures, proposing NULL results for invalid object addresses in three SQL functions #1, and supplied a comprehensive initial patch with regression tests #3.
- Álvaro Herrera objected to spreading NULL checks across callers and advocated missing_ok arguments instead #4; Michael accepted that direction and separated attribute, FDW, publication/subscription, and main object-address work into four patches #9.
- Aleksander Alekseev reported successful testing of the series #10, but Michael considered it premature to commit while type, operator, and procedure formatting still returned inconsistent textual placeholders #11.
Nov 2017 · 1 entry
- After several months, Michael rebased the series #12; Thomas Munro then identified documentation build failures #14, which Michael fixed in the next submission #15.
Jan - Mar 2018 · 2 entries
- Álvaro committed the attribute-name prerequisite from #16 as 8237f27b50, simplifying sepgsql's error handling and the get_attname implementation, and requested similar error-path layout plus a flag-based type-formatting interface for the remaining work #17.
- Michael supplied the remainder with a separate modular format_type prerequisite #18; Álvaro reported applying that prerequisite while retaining the typmod flag and commonly used wrappers #19, and Michael removed it from the resubmitted series #22.
May 2018 · 1 entry
- Álvaro examined external FDW usage and argued that changing existing signatures would impose compatibility work without benefiting those callers #23, leading Michael to introduce separate extended lookup functions and document them #26.
Jul 2018 · 1 entry
- Michael refreshed the remaining three patches with missing-object handling corrections and proposed prioritizing the two small prerequisites #27; Andrew Dunstan disputed the need to defer the main patch solely because of its size #28.
Sep 2018 · 2 entries
- After reviewing the refreshed prerequisites #32, Álvaro also identified misleading missing-column output in the main patch #33; Michael fixed the column-existence ordering and committed the publication/subscription member of #31 as 1d6fbc38d9, retaining the rest #34.
- Michael changed the extended FDW APIs to accept extensible flags #34, then adopted Álvaro's preference for bit types and preprocessor defines #35 in #36; that prerequisite landed essentially unchanged as 8fb569e978, with application confirmed in #38.
Dec 2018 - Feb 2019 · 3 entries
- Discussion of inconsistent missing-object results established a common contract: preserve a generic object-type name where applicable and return NULL for the unavailable identifying fields #36, with Álvaro supporting consistent representation and the generic routine terminology #37.
- Michael substantially reworked the remainder to implement that contract, adding nullable type, procedure, and operator formatting helpers, checking large-object existence, and clearing relation fields when a requested column was absent #39.
- The three-patch series then remained pending through repeated rebases into July 2019 #40 #41 #42, before Dmitry Dolgov identified inaccurate return-value documentation and an operator call using the procedure flag #43; Michael corrected those issues in #44.
Sep - Oct 2019 · 2 entries
- Kyotaro Horiguchi requested tests that distinguish NULL from empty strings #45, while Álvaro proposed a compact VALUES-driven test query and clearer INVALID_AS_NULL flag names #47 #48.
- Dmitry also found obsolete heap_close calls in the refreshed main patch #50; Michael incorporated table_close, the renamed flags, and the consolidated NULL-visible tests in #52.
Mar 2020 · 1 entry
- Daniel Gustafsson tested and reviewed the series, finding only minor comment and brace issues #53; after a documentation rebase #55, he checked the update and reiterated the nullable-return documentation correction #56.
Jul 2020 · 4 entries
- Michael committed the type-formatting member of #55 essentially unchanged as 1185c78294 and its procedure/operator member as aa38434824, correcting comments and declaration parameter names in the latter #58.
- The final remainder incorporated formatting feedback, local flag variables, and comment cleanup, but temporarily changed the generic routine label to procedure #59; Álvaro explained why routine covered both functions and procedures #60, and Michael accepted restoring it #61.
- Daniel read the final attached remainder without further objections #62, and Michael's final review removed a redundant attribute lookup and made regression output less sensitive to column widths before committing 2a10fdc430 from #59.
- Michael confirmed the final landing in #64, closing the thread just under three years after it began, with all remaining work applied and no outstanding patch or revert.