model-writtenrun 7325 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jan - Mar 2018 · 2 entries
- Nikita Glukhov opened the thread with JSON_TABLE support for jsonb and json, built on the separately developed SQL/JSON functions, in #1.
- Early revisions corrected PLAN DEFAULT flags and invalid-plan validation in #4 and #5, then adjusted expression evaluation in #6.
Jun - Jul 2018 · 1 entry
- Because the standard's implicit root-path naming was unclear, Nikita required explicit path names when using explicit PLAN clauses in #7; coercion-related test expectations followed in #8.
Mar 2019 · 1 entry
- The implementation continued through dependency updates, with json and jsonb patches combined in #12 and a bundled prerequisite submission in #15; Nikita explained that version numbers followed the dependent SQL/JSON and jsonpath series in #14.
Jun - Nov 2019 · 7 entries
- By #17, the foundational jsonpath implementation had left the prerequisite bundle, leaving pending SQL/JSON functions and JSON_TABLE for review.
- Pavel Stehule reported regression failures, a suspicious cast condition, and the compatibility risk of reserving FORMAT in #18, while Alvaro Herrera requested independently reviewable pieces in #19.
- Nikita fixed those prerequisite issues in #20 and clarified in #21 that the bundled functions belonged to another review thread, with JSON_TABLE itself the focus here.
- Pavel's review of #22 requested a separate parser file and documentation in #23, then distinguished the standard's JSON_TABLE, PLAN, and PLAN DEFAULT features and requested separate patches in #24.
- Nikita implemented that split, moved parsing into parse_jsontable.c, and supplied documentation in #25, reducing the base JSON_TABLE patch substantially.
- Pavel traced the regression discrepancies to locale-dependent comparisons in #29 and identified missing EXISTS PATH support and surprising formatted-column behavior in #30.
- Nikita explained the distinction between ON EMPTY and ON ERROR and the standard's restrictions in #31, then added EXISTS PATH, formatted-column DEFAULT, and implicit FORMAT JSON for structured output types in #33.
Jan 2020 · 1 entry
- Pavel reported successful compilation, check-world, compatibility checks, and performance testing in #34, while asking for better examples and suggesting that the advanced PLAN features could wait.
Mar 2020 · 1 entry
- The series adopted revised SQL/JSON expression infrastructure and dropped json-input support in #36, then resumed after a long gap with refreshed prerequisites and documentation in #39.
Dec 2020 - Jan 2021 · 1 entry
- Zhihong Yu reviewed naming, comments, memory ownership, and PLAN details in #40, and Nikita incorporated the requested cleanups in #41.
Mar - May 2021 · 3 entries
- Andrew Dunstan took over rebasing in #43, but Erik Rijkers found omitted build files in #44; Nikita restored them and repaired the prerequisite patch in #46.
- Erik's JSON_TABLE testing exposed missing LLVM evaluator references in #47; Nikita proposed the fix in #48, and Erik corrected a function name and confirmed success in #49.
- Although Erik reported substantial successful functionality and speed testing, the feature missed PostgreSQL 14's freeze, as recorded in #50, and Andrew resumed maintenance with #51 and #52.
Jul - Sep 2021 · 3 entries
- The LLVM fix had not been integrated, so Erik supplied it separately in #53 and resent the full series with two LLVM corrections in #54; Andrew folded them in with #56.
- Daniel Gustafsson's review in #55 requested clearer conditionals, a jsonb casting hint, accurate comments, and removal of dead code, but several comments remained outstanding until the final review round.
- After the Value-node rebase in #58, Erik reported a pg_stat_statements crash in #59; Andrew corrected the ListCell handling in queryjumble.c in #62.
Jan - Apr 2022 · 12 entries
- Further base updates led to #66, which Himanshu Upadhyaya tested extensively in #67, raising error-behavior questions and supplying documentation, diagnostic, and code-cleanup comments.
- Andrew addressed those cleanups and removed the sql_json configuration prerequisite in #68, but Erik found that constructor RETURNING clauses had disappeared in #69, leading to their independent restoration in #71.
- Andrew explained the interaction of ERROR ON EMPTY, default error handling, and strict paths in #72, while acknowledging that the behavior needed better documentation.
- Alvaro requested staggered commits to isolate buildfarm problems in #77, and Andrew agreed in #80 before beginning prerequisite landings reported in #82.
- Andrew repaired missing node handling in #83 and addressed Daniel's renewed review in #78 with the final full series in #84.
- Erik supplied documentation corrections in #86, but Andrew decided in #88 to land the code first and consolidate documentation separately because editing it across the series caused merge difficulties.
- The base JSON_TABLE component of #84 landed as 4e34747c88, with documentation excluded and small naming and conditional-style edits.
- The two PLAN components of #84 landed together as fadb48b00e, preserving their combined code and tests while deferring documentation.
- Andres Freund measured excessive jsonb_sqljson runtime in #91, prompting 2ef6f11b0c to halve the million-row fixture and make it unlogged.
- Stephen Frost reported that the unlogged table broke recovery tests in #95, while Tom Lane objected to huge core regression fixtures in #98; Andres proposed planner settings in #99, which 14d3f24fa8 adopted alongside a 50,000-row fixture and table cleanup.
- At the last message, the feature and test fixes had landed, but Andres still wanted explicit retained-unlogged-table coverage and changes to recovery dump handling, asking for further input in #101.
- After the thread's last message, a6baa4badd consolidated the unchanged JSON_TABLE and PLAN documentation from #84 with the SQL/JSON functions documentation attributed to #502994 in thread #38009, including editorial changes to the functions and aggregate sections; its commit message acknowledged remaining copy editing, and many example corrections from #86 were still absent.