model-writtenrun 7393 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jun - Jul 2020 · 3 entries
- Peter Eisentraut proposed unquoted RETURN and BEGIN ATOMIC function bodies in #1, parsing them at definition time to track dependencies while initially storing their trees in probin; pg_dump support and additional testing were still unfinished.
- Robert Haas, Tom Lane, and Andres Freund explored the consequences of early binding, including search_path behavior, column changes, polymorphic arguments, and dump dependency cycles in #2, #5, #6, and #7, while Vik Fearing clarified the distinction from a full procedural language in #15 and #17.
- Tom Lane identified creation-time binding as useful for earthdistance functions but emphasized preserving existing late-binding use cases in #25; the feature retained quoted SQL bodies alongside the new syntax.
Aug - Nov 2020 · 2 entries
- Peter added pg_get_function_sqlbody() and pg_dump integration in #26, and extracted the basic function-execution tests into 11b80d900f before adding set-returning-function inlining in #27.
- After rebasing, Peter moved parsed bodies into a dedicated prosqlbody column and improved block preservation and deparsing in #30; following a longer gap, #33 completed argument-name qualification to avoid ambiguous output.
Feb - Jun 2021 · 16 entries
- Jaime Casanova's utility-statement test exposed missing serialization support in #35, so Peter explicitly prohibited utility statements in #36 and deferred their full support.
- Jaime found that CASE expressions confused psql's function-body boundary detection in #37, and Julien Rouhaud diagnosed and supplied an inline scanner correction in #38 and #39, which Peter incorporated in #40.
- Jaime's deparser assertion report in #41 and Julien's assertion and duplicate-body test requests in #42 produced #43; Julien then reviewed that version and identified remaining formatting issues in #44.
- Peter landed the feature as e717a9a18b from #43, adding missing-body checks, limiting the implicit SQL language to unquoted bodies, and refining deparser output before announcing completion in #45.
- The landing immediately exposed parallel-execution failures in #46 and #47; Andres Freund supplied a temporary NULL-query-string workaround, discussed in #48 and #49, while Tom argued in #50 for restoring source-text availability instead.
- Investigation also exposed missing buildfarm coverage for some contrib tests: Julien demonstrated the gap in #59, and Andrew Dunstan reported adding coverage in #61.
- Tom posted three follow-up patches in #64 to restore non-NULL prosrc, provide source text during parse analysis, and optionally retain the original CREATE FUNCTION text.
- Noah Misch showed in #66 that stored CREATE FUNCTION text becomes obsolete after DDL renames, and the ensuing discussion in #67 and #68 questioned its usefulness for accurate runtime error cursors.
- Tom committed the first part of #64 with documentation and test additions as 1111b2668d, then removed the temporary parallel-query workaround through 83efce7a1e.
- The parse-analysis source-text patch from #64 landed unchanged as the separate commit 409723365b; Tom confirmed in #69 that the original-CREATE-text storage proposal was left uncommitted.
- Noah demonstrated mutually dependent functions that pg_dump could not restore in #70; Tom initially favored dummy declarations in #71, but Peter and Tom later concluded that the required additional dump objects were substantial machinery for a rare case in #76 and #77.
- Jeff Janes reported broken psql parenthesis continuation prompts in #74, Justin Pryzby supplied the missing else inline in #75, and Peter committed that exact correction as d9a9f4b4b9, confirming it in #78.
- Peter deferred dependency-cycle work in #79, Tom agreed to classify it as a release won't-fix item in #80, and Noah accepted that disposition in #81.
- Noah reported a crash on doubled semicolons in #82, and Julien proposed skipping empty statements during transformation in #83; Peter countered with grammar-level rejection in #84.
- Julien defended ignoring redundant semicolons in #85, and Tom proposed discarding empty statements directly in the grammar in #86, posting that replacement with Julien's tests in #88.
- Tom landed #88 unchanged as bfeede9fa4 and announced the push in the final message #91, leaving the feature and these corrective fixes committed while utility-statement support, dependency-cycle handling, and original-source storage remained deferred.