model-writtenrun 5866 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Shinya Kato opened the thread with missing EmitWarningsOnPlaceholders calls for four contrib extensions so misspelled configuration settings would produce warnings when those modules loaded #1.
- Daniel Gustafsson identified missing Tcl coverage, and Bharath Rupireddy identified three test modules and requested usage guidance; Kato expanded the patch accordingly #2 #3 #4 #5.
- Bharath also demonstrated that bogus extension settings remained accepted after loading an extension and advocated errors, leading Kato to change WARNING to ERROR; Bharath subsequently suggested separate warning and error APIs for compatibility #4 #5 #6.
- After a month, Kato added checking during custom-name validation, but Fujii Masao showed that it rejected legitimate configuration entries before their extension loaded and failed compilation; Kato fixed the function name, while Fujii confirmed regression failures remained #7 #8 #9 #10.
- Peter Eisentraut requested separating the straightforward call-site additions from the behavior changes, and Kato submitted three patches while expressing doubts about the validation-time check #11 #12.
- Kyotaro Horiguchi caught the missing pltcl prefix and explained why unknown namespaces must remain usable before extension loading; Tom Lane committed the missing-call component with the Tcl correction and explanatory comments as 1fada5d81e #13 #14 #15.
- Tom rejected raising an error during extension initialization because it could leave initialization incomplete, and argued that a separately committed warning implementation should instead reject new placeholders at creation time once their prefix was reserved #15.
- Tom supplied a replacement that checks reserved prefixes in find_option and fixes incomplete prefix comparisons, while proposing the clearer MarkGUCPrefixReserved name with a compatibility macro; Kato withdrew his two remaining behavior patches and endorsed Tom's replacement and rename #16 #17 #18.
- Tom committed the replacement from #16 as 2ed8a8cc5b, relocating and expanding its regression tests, and separately implemented the discussed API rename as 5609cc01c6.
- The relocated tests exposed failures under force_parallel_mode because workers could load extensions and restore settings in a different order from their leader; Tom announced and performed the revert cab5b9ab2c, leaving the original missing-call additions intact #19.
- Tom then posted a delta to apply after undoing the revert, deleting existing invalid placeholders when reserving a namespace so they could not break subsequent parallel-worker startup #20.
- After the holiday gap, Florin Irion reported successful tests with and without parallel mode, and Tom concluded that catching configuration mistakes outweighed the compatibility cost for applications using extension namespaces as arbitrary variables #21 #22.
- Tom restored the reserved-prefix checks and API rename together with the posted deletion delta in 88103567cb, preserving its behavior while accommodating the newer source tree, and announced completion #20 #23.
- The thread ended with Florin's thanks: the warning-call additions and the revised namespace enforcement had landed, while Kato's original initialization-error and validation-time proposals remained withdrawn #17 #24.