model-writtenrun 9438 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jan - Apr 2018 · 5 entries
- Andrew Dunstan opened the thread by forwarding Nikita Glukhov's jsonpath engine, GIN and plain-json patches in #1, identifying timezone-format support as a prerequisite.
- Nikita separated datetime prerequisites in #7, and Andrew reported the initial encoding refactoring committed in #9; their representation discussion in #10 and #11 explained temporary datetime values and lazy variable evaluation.
- Nikita's expanded proposal in #14 added arithmetic-error suppression, grammar corrections, GIN changes and optional construction syntax, while preserving input timezone offsets to keep operations immutable.
- Robert Haas challenged exception-based arithmetic handling in #16, Alexander proposed internal error-returning numeric functions in #18, and Tom Lane's objections to the resulting fragile implementation in #26 led Nikita to separate that experiment in #27.
- Tomas Vondra's valgrind testing in #19 exposed uninitialized state, while later work added ECMAScript escapes in #29 and repaired Windows parser generation following Thomas Munro's report in #30 and Nikita's update in #31.
Aug 2018 - Feb 2019 · 8 entries
- After the summer rebase in #32, Stas Kelvich flattened the series and supplied compliance tests in #33; Nikita answered with object identifiers for keyvalue(), interrupt checks and cleanup in #34.
- Tomas's autumn review in #36 exposed regression failures and missing documentation; although #44 described memory-lifetime fixes, its supplied patch files are identical to #38.
- Nikita added Liudmila Mantrova's SGML documentation, separated optional numeric suppression and substantially cleaned up GIN support in #48, then removed unsupported fractional-second formats and unused code in #52.
- Alexander narrowed the deliverable in #57, removing the cumbersome plain-json implementation and extra operators while retaining a documented jsonb API that could ship independently of the larger SQL/JSON project.
- The API discussion in #58, #60 and #61 produced jsonb_path_* names, always-array results and a first-result function in #62; subsequent revisions corrected syntax, documentation and memory lifetime through #68.
- Alexander replaced saved ErrorData with immediate errors or jperError returns in #71 and documented the deliberately permissive match behavior in #74.
- Andres Freund and Tom Lane rejected catching backend errors without transaction cleanup in #75, #78 and #84, while Andres's detailed implementation review in #85 requested binary I/O, clearer comments and simpler infrastructure.
- Alexander responded with a reduced implementation in #90, excluding datetime and numeric suppression while adding binary I/O and implementation documentation; timezone-dependent operations remained incompatible with the intended immutable API.
Mar 2019 · 11 entries
- Nikita restored numeric suppression through explicit error-returning APIs in #93, keeping datetime separate and redesigning optional plain-json support around runtime access flags and JsonItem values.
- Tomas's testing and review in #98 and #100, Andrew's NaN question in #99 and Robert's diagnostic wording in #101 shaped #102, followed by the final core and numeric submission in #103.
- After final review was confirmed in #104, the core and numeric components of #103 landed as 72b6460336 and 16d489b0fe, with only catalog and documentation adjustments to the core.
- Jeff Janes and Tom Lane exposed generated-file cleanup and rebuild problems in #106, #108 and #109, leading directly to 142c400d72; the attachment in #113 proposed a separate Windows follow-up after that commit.
- Pavel tested the remaining GIN patch in #111, and Jonathan Katz's larger benchmarks in #117 and #119 showed that key-only extraction could produce expensive, unselective bitmap scans.
- Tom identified the generic scanner typedef problem in #125 and reviewed Alexander's proposed rename in #127, producing the unique JsonPathKeyword name in 75c57058b0 from #126.
- Andrew questioned installation of private parser headers in #114, and Tom proposed combined lexer/parser compilation in #115; Alexander implemented that in #132, which landed after a trivial rebase as 550b9d26f8.
- John Naylor's non-backtracking scanner patch in #129 developed into Alexander's parser-cleanup series in #133, with John's review in #135 prompting simpler hexadecimal handling and Unicode-surrogate tests in #138.
- Nikita separately proposed errors for invalid non-silent match results in #134 and stricter numeric-token syntax in #137; the accompanying suggestion to rename jsonb_path_match() did not reach a resolution here.
- Andres reported SPARC failures in #139, while Andrew's Windows investigation implicated the combined parser/scanner build in #159, opening a platform-debugging phase after the initial landing.
- Alexander used Jonathan's GIN findings to restrict extraction to accessor-chain equality predicates in #161, bundling the remaining GIN work with Nikita's two fixes; the GIN component landed with a catalog-version bump as 0a02e2ae02.
Apr - May 2019 · 7 entries
- Alexander's investigation in #168 implicated compiler behavior, and Andrew eliminated the Windows failure by downgrading GCC in #173; Tom reported the SPARC problem remained unreproduced in #174, with no revert recorded in this thread.
- John restored scanner clean rules and corrected an error-message typo in #177, which Tom committed with an explanatory comment as b4f96d69ad.
- Tom's diagnostic criticism in #178 led Alexander to start #181; reviews in #183 and #190 drove removal of message macros, inconsistent quotations and boilerplate primary errors.
- Further review in #192 removed unexplained singleton terminology and separated its documentation correction, producing the final two-part submission in #194.
- After Tom identified remaining diagnostic mistakes in #195, Alexander committed the corrected error-reporting component as 29ceacc3f9 and the unchanged documentation component as 53ae0b16d6.
- The deferred datetime work later acquired newer matching submissions in thread #40681, which owns 6dda292d4d, bffe1bd684, 66c74f8b6e and d589f94460, despite their historical links to this thread.
- At this thread's close, the core, numeric suppression, GIN and listed follow-up fixes had landed, while broader plain-json and syntax-extension work remained outside that landing; Alexander's documentation pair in #197 and proposal to consolidate the language description in #198 had no recorded resolution.