model-writtenrun 7240 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Amit Langote revived the previously reverted SQL/JSON implementation in #1, preserving its feature divisions while replacing subtransaction-based error trapping with ErrorSaveContext support.
- Elena Indrupskaya supplied documentation corrections in #3; Andrew Dunstan questioned replacing parameter tags in #4, and Elena explained in #5 that these SQL constructs use replaceable expressions rather than named function parameters.
- Amit replaced subsidiary recursive expression evaluation with ExprEvalSteps in #7, addressing the earlier executor-design objections and retaining Elena's documentation changes as a separate patch.
- After rebasing in #9, Amit adapted to the queryjumble overhaul and merged common clauses into constructors in #10 because the standalone grammar additions produced unused-production warnings.
- Erik Rijkers reported a JSON_QUERY array-conversion hang in #12, and Amit corrected its infinite executor loop in #14; his correction in #15 clarified that the documentation edits had not yet been merged.
- Andres Freund challenged runtime constants, confusing evaluator names, grammar complexity and JSON_TABLE expression-state handling in #13; Amit answered with initialization-time constants, consolidated evaluation functions and revised scan/join structures in #16.
- Amit concluded that JSON_TABLE could not be treated as one scalar expression and moved column and PASSING ExprStates into TableFuncScanState across #18 and #19, avoiding repeated initialization inside opaque JSON_TABLE state.
- Andrew proposed landing the foundational features before JSON_TABLE in #20, while Peter Eisentraut requested fewer grammar symbols, removal of commented-out code and documentation accompanying each feature in #21.
- Amit redistributed documentation and combined JSON_TABLE with PLAN support in #22, then Andrew supplied grammar simplifications in #23.
- Alvaro Herrera isolated constructors for commitment in #24, added ECPG coverage and deferred IS JSON grammar machinery; #25 further reorganized documentation and removed obsolete skipped-key string handling.
- Peter explained in #28 that optional KEY syntax has an ambiguity in the SQL standard, supporting Alvaro's decision in #25 to leave that syntax unsupported.
- Justin Pryzby's sqlsmith and Valgrind testing in #26 exposed invalid JSONB data when all entries were skipped, and Amit supplied the uniqueifyJsonbObject() correction in #27.
- Erik reported successful constructor testing and documentation issues in #29; Alvaro announced the constructor landing in #30, and Erik supplied a follow-up clarification of NULL omission in #31.
- Alexander Lakhin identified naming, spelling and unused-assignment problems in #32, leading to 60966f56c3, which also cleaned up parser style and aggregate initialization.
- Alvaro resumed IS JSON and query functions in #33, deferred the other functionality, and recognized that WITHOUT_LA needed to distinguish WITHOUT TIME rather than WITHOUT UNIQUE to accommodate ARRAY WRAPPER syntax.
- Alvaro posted the polished IS JSON patch in #34 and confirmed its landing in #35, leaving the lookahead correction and query functions for further work.
- The lookahead correction attached in #35 landed in 71bfd1543f, together with additional node traversal, typmod and location fixes and a catalog-version bump.
- Alvaro requested cleaner ON ERROR/ON EMPTY grammar, better soft-error commentary, ECPG tests and documentation organization in #35; Alexander identified further stale references in #36, and Amit supplied soft-error propagation fixes and comment corrections in #37.
- Alvaro elaborated the unresolved grammar problems in #38, while Jonathan Katz directed Nikita Malakhov's dot-notation proposal to a separate discussion in #40.
- Andrew concluded in #41 that the remaining features would miss feature freeze, and Alvaro marked the partially landed CommitFest entry committed in #42; Erik subsequently proposed documenting VALUE as optional noise in #43.
- After Matthias Kurz raised the uncommitted remainder in #44, Alvaro requested rebased patches in a new thread in #45, and Amit linked that continuation and its CommitFest entry in #47.
- The expression soft-error machinery descended from #35, but its final source was #551965 in thread #48026, whose dedicated safe-coercion opcode and initialization-time error contexts landed in aaaf9449ec.
- The record-population work developed in #35 and #37 continued in thread #48026, where #551965 supplied the expanded composite and domain soft-error handling and jsonb_populate_record_valid() committed with documentation and comment edits in 1edb3b491b.
- The variable-fetch callback work in #35 also continued in thread #48026, whose newer #551965 supplied the matching lookup and variable-counting implementation for faa2b953ba.
- The query functions descended from #35, but 6185c9737c came from the newer #557196 in thread #48026, with executor bookkeeping, scalar-extraction and comment changes.
- This thread ends at the handoff in #47, with constructors and IS JSON landed and the remaining features continuing in thread #48026; the four later commits are attributed to that continuation's newer patchsets.