model-writtenrun 7115 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jun - Oct 2023 · 15 entries
- Amit Langote reopened the remaining SQL/JSON work at #1, beginning with JSON_EXISTS, JSON_QUERY, and JSON_VALUE; he retained JsonBehavior nodes because DEFAULT carries an expression and simplified grammar and timezone-dependency interfaces inherited from earlier work.
- Amit expanded the series at #2 to include JSON(), JSON_SCALAR(), JSON_SERIALIZE(), JSON_TABLE, and conformance declarations, placing the independent constructor functions before query functions.
- Peter Eisentraut rejected adding newer keywords to the SQL:2016 list and supplied an independent grammar-indentation correction at #3; Amit approved it at #4, and it landed as 01f1f789df.
- Álvaro Herrera's reviews at #6 and #8 favored a unified JSON/JSONB categorization API and clearer RETURNING grammar, and Amit removed the unsupported RETURNING extension from JSON() and JSON_SCALAR() in #14.
- The preparatory constructName and CaseTestExpr changes from #11 landed as 785480c953 and b6e1157e7d, allowing #12 to resend the remaining work.
- Álvaro's API corrections at #13 were incorporated into the unified categorization patch in #17, which landed with a declaration adjustment as 3c152a27b0.
- Further review of formatted_expr guarantees at #18 prompted a standalone cleanup at #19, revised within #23 and committed as 7c7412cae3.
- The constructor series in #26 separated conversion exports and grammar renaming from the feature, landing as b22391a2ff, 254ac5a7c3, and 03734a7fed, with light final constructor edits.
- Shinoda's documentation correction at #29 landed unchanged as e055b6be7e, while #33 retained query functions and JSON_TABLE with improved diagnostics and tests.
- After the summer gap, Amit reworked parsing, coercion representation, and LLVM support in #36, extracting a soft-error prerequisite needed to implement ON ERROR.
- Álvaro's review at #39 challenged error-context management and propagation, leading #42 to separate populate_record_field() soft-error support and simplify context ownership.
- Erik Rijkers exposed crashes with wrapper and return-type combinations at #43, #45, and #47; Jian's diagnosis at #50 and subsequent revisions led to safer population-based coercion in #51 and pseudotype rejection in #57.
- The soft-error expression prerequisite from #64 landed as 7fbc75b26e, but assert-enabled LLVM builds crashed as reported at #65, and Amit announced its revert at #66.
- Amit supplied LLVM pointer and boolean fixes at #69, but Andres Freund's review at #72 raised ordinary-cast performance and executor-design concerns, prompting the separate EEOP_IOCOERCE_SAFE implementation in #75.
- Nikita Malakhov forwarded Anton Melnikov's coercion-crash fixes at #76, and Anton explained further EMPTY ARRAY problems at #81 and #83; these reports fed the continuing query-function revisions.
Nov - Dec 2023 · 8 entries
- Nikita separately experimented with a pg_proc.proerrsafe attribute at #92, then withdrew that approach at #93 after finding earlier objections to the design.
- Amit's full rewrite at #96 replaced tangled runtime coercion state with compiled jumps and target arrays, but Erik found runaway evaluation at #99, which #101 repaired.
- Peter's grammar cleanup patches at #107 removed unnecessary intermediary nodes and moved checks to parse analysis, and #108 incorporated them while removing JSON_EXISTS RETURNING support.
- The parser-size debate continued through #122 and Andrew Dunstan's measurements at #133, while Jian's coercion tests and deltas at #116 and #124 drove the substantial JsonBehavior and JsonCoercion redesign in #138.
- Álvaro proposed precedence and lookahead alternatives for NESTED syntax at #139, Amit tried lookahead in #140, and Robert Haas examined its future ambiguity risks at #150 and #154.
- Álvaro independently simplified FORMAT JSON grammar at #151; Amit carried the adjusted patch in #153, and that component landed as d3fe6e90ba.
- Andres's review at #156 rejected mutable runtime jump targets and requested smaller review units, leading #159 to return jump targets directly, reorganize executor state, and split support APIs.
- Jian's sustained quote, wrapper, typmod, and domain testing, including #165, #167, #170, and #177, produced the revised coercion behavior in #173 and domain-error handling in #182.
Jan - Mar 2024 · 14 entries
- Álvaro's EXPLAIN coverage investigation produced a proposed functional patch at #175, but he concluded at #184 that tests were sufficient and committed the XMLTABLE alias and FORMAT JSON coverage as 752533d40f.
- The helper organization was revised for testability in #182 and #186, from which the population soft-error changes and variable-fetch callback refactoring landed as 1edb3b491b and faa2b953ba.
- The rewritten expression soft-error prerequisite in #186 landed as aaaf9449ec with unchanged code, using EEOP_IOCOERCE_SAFE for input conversion and errsave() for domain constraints after the earlier implementation had been reverted.
- Amit reported the prerequisite landing, including the rewritten expression soft-error work, at #187 and resent the remainder with DEFAULT-expression restrictions prompted by Jian's findings at #183; typmod corrections followed in #188.
- The March refresh at #196 resumed the remaining feature work, and Tomas Vondra's large JSON_TABLE test at #202 exposed a coercion-cache memory leak fixed in #211 and successfully retested at #212.
- Himanshu Upadhyaya's overflow examples at #204 and #210 revealed inconsistent numeric and string coercion errors, helping motivate the input-function-based coercion rewrite and removal of JsonCoercion machinery in #239.
- Tomas questioned the required ordering of ON EMPTY and ON ERROR at #216, and Álvaro's standard-based explanation at #218 led Amit to retain that ordering at #219.
- Álvaro concluded at #231 that the separate TableFuncScan naming patch added little, so Amit dropped it and revised the JSON_TABLE presentation as described at #235.
- After Jian's final review, the query-functions portion of #243 landed with local cleanup as 6185c9737c, leaving the JSON_TABLE remainder in #244.
- Kyotaro Horiguchi identified a split error-message literal at #245, prompting the immediate discussion-derived correction 085e759e9d.
- Jian's PLAN-output concerns at #248 and request to divide the work at #249 led Amit to post basic JSON_TABLE alone in #251, a scope reduction Álvaro endorsed at #252 for the approaching release deadline.
- Amit restored NESTED support as a separate patch in #255 while leaving PLAN deferred, then reworked plan nodes, nested ordinality, row sources, and PASSING propagation in #262, incorporating Jian's proposals and tests from #260.
- Nikita revived the deferred PLAN portion of #244 in #298, but the thread ended at #300 with a newly found execution issue still unresolved and an intention to move the work to a new thread.
- PLAN support subsequently continued in thread #50777, where #1147204 supplied the exact patch committed as 86ab7f4c72; this supersedes the earlier attribution to the unfinished revival in #298.
Apr - Jun 2024 · 9 entries
- The basic JSON_TABLE portion of #266 separated table-level and column-level errors and landed with light cleanup as de3600452b; Amit explained at #271 that Jian's local view-validation helper needed more work before inclusion.
- Jian's duplicated-word report at #260 separately produced the small discussion-derived fix 2f6e78b061.
- Alexander Lakhin's assertion report at #272 was investigated and found reproducible without JSON_TABLE at #275, so it did not become a JSON_TABLE regression fix in this candidate set.
- Jian's view round-trip tests at #278 and #281 exposed lost QUOTES and WRAPPER semantics; the correction attached in #286 landed with a further deparser adjustment as f6a2529920.
- The NESTED remainder evolved from #277 to the clearer recursive executor in #286, and Jian's final review at #287, #288, and #289 preceded its lightly revised landing as bb766cde63.
- Amit closed the feature's commitfest entry at #290 with basic JSON_TABLE and NESTED support committed, explicitly deferring PLAN to another entry and potentially another thread.
- The DEFAULT domain-cast problem reported at #278 was subsequently fixed by c0fc075186, whose recursive expression validator came from the later patchset #560385 in thread #49423, rather than an attachment in this thread.
- Thom Brown's documentation-readability report at #291 prompted #292, which Jian reviewed at #293; the work continued in thread #49855, where the later rewrite #566294 supplied 42de72fa7b.
- Alexander Lakhin identified obsolete constructor comments at #294 and requested aggregate coverage at #296, producing #297, which landed unchanged as 3a8a1f3254.