model-writtenrun 7553 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut opened the thread with the SQL-standard function-body landing announcement in #1: unquoted RETURN or BEGIN ATOMIC bodies are parsed at definition time, track dependencies, exclude polymorphic arguments, and require psql to recognize internal semicolons.
- Andres Freund reported buildfarm failures under forced parallel execution in #2, and Tom Lane reproduced the crash and identified missing query text in #3.
- David Rowley proposed running regression tests with forced parallelism in #4; Andres Freund suggested reusing the pg_upgrade test run in #5, and Michael Paquier supported that proposal in #6.
- Andres Freund committed b3ee4c5038 without an attached patch, making ExecInitParallelPlan tolerate NULL query text as an immediate response to prolonged buildfarm failures, while acknowledging uncertainty about the long-term approach.
- Peter Eisentraut defended NULL tolerance as consistent with other executor code in #7, while Robert Haas argued in #8 for retaining query strings for debugging or using a different fallback; the thread shows no further resolution of that policy question.
- Andrew Dunstan proposed eliminating pg_upgrade's duplicate regression run in #9, Tom Lane raised coverage concerns in #10, Andres Freund identified fsync-related timing differences in #11, and Michael Paquier suggested moving pg_upgrade tests to TAP in #12; no attached implementation followed.
- Laurenz Albe reported that an FDW option named case prevented psql from recognizing statement termination and supplied a parenthesis guard in #13, then extended it to END handling in #14.
- Peter Eisentraut expanded that fix in #15 to track BEGIN and END only within CREATE FUNCTION or CREATE PROCEDURE statements and to count CASE only inside BEGIN blocks; Laurenz Albe endorsed the stronger false-positive protection in #16.
- The final lexer patch in #15 landed unchanged apart from whitespace as 029c5ac03d, confirmed in #17; Laurenz Albe's thanks in #18 closed the thread with both fixes landed and the broader query-text and testing proposals unresolved.