model-writtenrun 7392 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane opened the thread because SQL CALL required dummy arguments for output-only parameters despite documentation describing input arguments, proposing to omit OUT arguments in plain SQL while retaining PL/pgSQL output variables #1.
- Tom's investigation shifted toward restoring the traditional input-only definition of pg_proc.proargtypes; his first patch also repaired named-argument handling and introduced separate SQL and PL/pgSQL CALL parsing behavior #3 #6.
- Peter Eisentraut objected that omitting output arguments would undermine SQL-standard compatibility and migrations, while Robert Haas questioned context-dependent CALL semantics; Tom clarified that preventing procedures distinguished only by their output types was his central objective #7 #9 #11 #13.
- Tom demonstrated that output-based overloading made untyped NULL placeholders select procedures unexpectedly and offered to retain OUT placeholders while restoring input-only signatures; Peter additionally identified that the patch broke standard DROP PROCEDURE argument lists #12 #16.
- Tom's second patch abandoned omission of OUT arguments from SQL CALL, retained the catalog and argument-processing changes, and expanded the documentation #18.
- Peter continued to press for portable procedure identification, while Tom emphasized existing ALTER/DROP ROUTINE compatibility and catalog consumers; David G. Johnston requested a review of the reference pages' compatibility sections #19 #23 #27.
- Tom proposed using traditional input-only lookup when argument modes are explicit and trying both input-only and all-argument lookup otherwise, rejecting conflicting matches; Andrew Dunstan supported ambiguity detection and Peter accepted the compromise #27 #30 #34.
- Tom discovered that the compromise could make pg_dump --clean output ambiguous, then supplied a two-patch series that implemented the lookup rules and made ruleutils.c emit explicit IN markers to protect generated procedure definitions and identifiers #40 #41.
- Peter inspected the series but could not apply its second patch for exploratory testing, and asked whether SQL CALL should require literal NULL for OUT arguments; Tom preferred retaining arbitrary unevaluated expressions and discussed future variable support without adding that restriction #42 #43.
- Tom consolidated the series, expanded compatibility documentation, and raised whether LookupFuncName should exclude procedures; Peter's final review supported that restriction and requested argument-order consistency and preservation of SGML formatting needed by man pages #44 #45.
- The thread ended with Tom acknowledging Peter's review and agreeing to address the remaining details #46; he subsequently committed the consolidated patch as e56bce5d43, including the function-lookup restriction, review corrections, additional named-INOUT coverage, and catalog-version bump.
- The resulting change retained procedure OUT support and mandatory CALL placeholders, restored input-only catalog signatures, and supported both procedure-identification conventions with ambiguity checks; omission of SQL OUT arguments and mandatory literal-NULL enforcement were left out #18 #43 e56bce5d43.