model-writtenrun 330 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Kirill Reshke opened the thread with DOMAIN support inside CREATE SCHEMA, following advice to prioritize SQL-standard features, and explicitly left documentation and tests for later #1.
- Jian He corrected schema qualification and added dependency tests, documentation, and psql completion across #2, #3, and #10, while Kirill tested and formatted the intermediate version in #5.
- A missing error cursor led Tom Lane to identify absent ParseState plumbing in #6; Kirill supplied an independent DefineDomain patch in #7, then moved that work to another thread after Tom requested broader scope #8 #9.
- Tom argued that DOMAIN support should include other types and eventually routines, and that sorting statements by object kind could break valid dependencies; he proposed executing subcommands in written order #12.
- Tom supplied the separate ordering patch in #13, and Kirill added temporary-object error locations and tests demonstrating the changed ordering behavior in #14.
- The standards discussion compared other implementations and examined name-resolution ambiguity: Michael Paquier questioned schema qualification, Tom defended PostgreSQL's search path, and Vik Fearing explained the standard's routine-path rules while favoring written order #19 #20 #21 #23.
- Peter Eisentraut identified standard text permitting circular foreign-key references in #24; Tom acknowledged that requirement but initially considered it less valuable than expanding supported objects, and illustrated why object-kind sorting cannot handle domains over composites or base types with support functions #25 #27.
- After an eight-month gap, Jian revived the ordering approach, combined it with DOMAIN support, and added COLLATION support in #28.
- Kirill found an unrecognized-node error for CREATE TYPE and criticized the unsupported-object wording in #29; Jian addressed those issues, added TYPE support, and moved encoding-sensitive tests in #30.
- Tom proposed a compromise that executes statements in order but extracts foreign keys into ALTER TABLE commands at the end, satisfying circular-reference requirements without general dependency sorting #31.
- Jian demonstrated that compromise with a restricted prototype in #32, then integrated it into the full series with column- and table-level constraint handling in #33.
- The subsequent submissions fixed test concurrency, expanded DDL-deparse coverage, refreshed comments, and switched DOMAIN qualification to DeconstructQualifiedName; the last two contained these revisions despite being described as rebases #34 #35 #36.
- Tom resumed final review after the long consensus wait and found the explicit F311-01 circular-reference requirement, strengthening the case for retaining foreign-key postponement #37 #38.
- Tom reorganized the work into three patches, rewrote foreign-key extraction to avoid modifying reusable input trees, and broadened object support to routines, aggregates, operators, and text-search objects #39.
- The expanded version deliberately omitted casts, transforms, and roles because their naming and ownership do not fit schema-contained objects, and left some psql completion limitations unresolved #39.
- Tom corrected collation-sensitive tests and documentation in #40; Jian reviewed that version and requested a wrong-schema test, while Tom explained why cached parse trees must remain unmodified #41 #42.
- The final ordering patch landed with only a helper rename and comment adjustment as a9c350d9ee, and the foreign-key patch landed unchanged apart from surrounding context as 404db8f9ed, both from #40.
- Expanded object support from #40 landed with small naming and regression-test additions as d516974840, and Tom announced the completed push in #43.
- Months later, Noah Misch supplied an automated audit and reproducer tests showing lexer, PL/pgSQL, completion, and authorization issues, but emphasized the more consequential silent change in which table an unqualified view reference selected #44.
- Tom initially defended the intentional ordering semantics and discussed the smaller bugs, while Noah clarified that an existing statement could succeed with a different meaning rather than fail visibly #45 #46 #47.
- Noah recommended reverting because the modest syntax benefit did not justify compatibility and design risks; Tom reconsidered, and Heikki Linnakangas and Nathan Bossart supported reversion #48 #49 #50 #51.
- Tom proposed preserving circular foreign-key support and an escaped-semicolon correction, and Noah discussed the risks of retaining only part of the original series while agreeing with preserving that lexer correction #52 #53.
- Tom posted the object-support revert in #54, noting that later lexer fixes prevented a verbatim rollback, then supplied the complete two-patch proposal restoring old ordering while retaining foreign-key postponement and diagnostic plumbing in #55.
- The two patches from #55 subsequently landed unchanged as 3c5d28ba64 and e0fdc3f54b, removing the expanded object types and restoring legacy ordering.
- The unattached release-note follow-up 6351dadfcf removed the reverted feature announcements and documented the retained foreign-key forward-reference improvement.
- At the last message, the selective reversion proposal was complete and checked against both development and release branches; a compatibility-preserving return of expanded object support remained only a possible future direction #54 #55.