model-writtenrun 7328 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jan - Mar 2018 · 3 entries
- Nikita Glukhov opened the thread with SQL/JSON constructors, aggregates, query functions and IS JSON, explicitly excluding JSON_TABLE and depending on the separate jsonpath implementation #1.
- Early revisions corrected expression naming and expanded datetime tests, while Simon Riggs requested explanatory documentation and Oleg Bartunov pointed to his draft #3 #6 #7.
- Andres Freund objected to catching errors across expression execution, Alexander Korotkov proposed subtransactions, and Nikita made those unconditional rather than dropping standard ON ERROR behavior #9 #11 #13 #14.
Jun - Jul 2018 · 1 entry
- Nikita experimented with cached subtransactions to reduce overhead but acknowledged callback and parallelism problems; Pavel Stehule advocated exception-safe conversions, and the next revision removed an inappropriate implicit I/O cast #15 #16 #17.
Nov - Dec 2018 · 1 entry
- After maintenance rebases, Nikita bundled jsonpath for independent testing, then answered renewed safety concerns by replacing parser error-catching with throwErrors and avoiding transactions for no-op conversions #22 #23 #24.
Jul 2019 · 1 entry
- Once core jsonpath was available upstream, the remaining work was divided into constructors, IS JSON and query functions, with datetime/json extensions initially retained as prerequisites and common clauses later extracted #29 #34 #38.
Jan - Apr 2020 · 5 entries
- Pavel found malformed JSON being hidden by NULL ON ERROR; after discussing the standard and Oracle behavior, Nikita changed evaluation to parse through ordinary casts first, correcting an accidental resend of old attachments #39 #40 #41 #42 #44.
- Erik Rijkers found a pg_stat_statements compilation failure and Pavel supplied corrected regression output; Nikita fixed both while replacing function-format machinery with dedicated constructor and predicate executor nodes #45 #46 #47.
- Pavel drove removal of invisible coercions and adoption of conventional node structures, leading to separate coercion expressions and normal JsonFormat, JsonReturning and behavior nodes #48 #49 #50 #51.
- Nikita removed the immature jsonpath-for-json dependency and unnecessary public constructor helpers, received successful tests from Pavel, and split aggregate behavior into strict and uniqueness variants #53 #54 #55.
- Alexander identified quadratic key-uniqueness checks and timezone-sensitive immutability, and Nikita adopted a shared hash and automatic timezone analysis #56 #60.
Jul 2020 · 2 entries
- Andrew Dunstan resumed maintenance after a gap and found parallel-test and documentation problems; Nikita fixed these while adding JSON(), JSON_SCALAR(), JSON_SERIALIZE() and an experimental sql_json setting that remapped the JSON type #58 #60 #62.
- Andrew considered the type-mapping work excessive scope, while Pavel advocated landing simpler functions before staged JSON_TABLE support and Oleg argued for broader standard coverage #63 #66 #67 #69.
Dec 2020 - May 2021 · 2 entries
- Upstream grammar changes forced removal of the JSON TEXT pseudotype, and Zhihong Yu's review corrected encoding errors, executor null handling, control flow and enum naming #70 #71 #72 #73.
- Erik supplied documentation corrections and tested examples; Andrew folded in fixes and maintained the series, while Nikita repaired LLVM integration and Erik subsequently found additional JIT and jumbling problems #74 #76 #78 #79 #84 #86.
Dec 2021 - Apr 2022 · 16 entries
- Himanshu Upadhyaya tested constructors, scalar predicates and query examples, Peter Eisentraut explained optional KEY grammar ambiguity, and Andrew retained ordinary JSON scalar semantics while accepting naming and const-correctness cleanup #90 #91 #100 #101 #105.
- Andrew demonstrated that sql_json produced confusing type displays and function selection, deferred the GUC, and retained only independent RETURNING clauses for JSON() and JSON_SCALAR() #106 #107 #108 #109.
- The common-clause patch from #109 first landed unchanged as 865fe4d5df, but Tom Lane reported warnings-as-errors failures and Andres reported ecpg/bison problems, prompting a revert #111 #112 #113 #115.
- Tom recommended landing dependent grammar together, while Justin Pryzby exposed raw-expression walker failures and Andrew found related copying problems; the final series incorporated traversal, copy and equality repairs, including contributions from Masahiko Sawada #116 #117 #119 #120 #121.
- Andrew relanded unchanged common clauses from #121 as f79b803dcc together with its constructor code as f4fb45d15c, reserving documentation for a separate commit.
- The IS JSON code from #121 followed unchanged apart from deferred documentation as 33a377608f; continuing jabiru failures were traced to stale ccache state rather than a reproducible SQL/JSON defect #123 #138 #140 #141.
- The query-function code from #121 landed as 1a36bc9dba with enum names adjusted and documentation deferred.
- The JSON(), JSON_SCALAR() and JSON_SERIALIZE() code from #121 landed without documentation as 606948b058, and its RETURNING patch landed unchanged as 49082c2cc3.
- Andrew reported that the code series was committed and planned dependent JSON_TABLE work; consolidated functions documentation from #121, with aggregate editing and JSON_TABLE documentation from #503176 in thread #38010, then landed together as a6baa4badd #143.
- Justin's documentation review produced b46bcf7a4b, correcting wording and removing a misleading GROUP BY claim, while Tom's subsequent objection to the unusual section structure remained a separate cleanup issue #144 #146 #148.
- The supplied later candidates show renewed landings beyond the captured discussion: constructors and common infrastructure from #121 were substantially reorganized in 7081ac46ac, while IS JSON received grammar, validation and ecpg changes in 6ee30209a6.
- The later JSON(), JSON_SCALAR() and JSON_SERIALIZE() landing 03734a7fed comes from the newer submission #539031 in thread #48026, superseding this thread's ancestral implementation.
- The preparatory expression soft-error commit aaaf9449ec and record-population soft-error commit 1edb3b491b come from components of #551965 in thread #48026, replacing the previous attribution to historical discussion here.
- The variable-fetch callback refactor faa2b953ba matches #551965 in thread #48026, while the later query-function landing 6185c9737c comes from that thread's #557196 and its compiled evaluation steps and soft-error contexts.
- The basic JSON_TABLE implementation de3600452b and nested-path support bb766cde63 come from #558821 and #559360 respectively in thread #48026, consistent with JSON_TABLE being separate work in #1 and #143.
- The JSON_TABLE PLAN implementation 86ab7f4c72 exactly matches #1147204 in thread #50777, superseding both the older PLAN submission in thread #48026 and the historical discussion-only attribution here; no new local messages extend the thread beyond the confirmed JSON_SERIALIZE fix in #151.
Jul 2022 · 1 entry
- The captured thread ends with a.kozhemyakin reporting a JSON_SERIALIZE RETURNING jsonb crash introduced by 606948b058, and Andrew confirming a committed check that rejects the unsupported return type instead #149 #151.