model-writtenrun 1696 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andrei Klychkov proposed suppressing empty-string quoting for list GUCs to prevent ALTER SYSTEM settings from causing restart failures #1; Jim Jones confirmed the intended behavior but found a function-deparsing regression #2, prompting Andrei to move the workaround into configuration-file writing #3.
- Fujii Masao argued that SET and ALTER SYSTEM should behave consistently and proposed handling a lone empty string in flatten_set_variable_args() #4; Jim Jones tested that proposal and requested regression coverage #6.
- Tom Lane explained that a list containing an empty string differs from an empty list and that SET lacked syntax for the latter #9; he rejected an omitted value because of grammar conflicts and explored both NULL syntax and redefining a lone empty string while forbidding empty elements #10.
- Tom implemented the empty-string interpretation across SET, list parsing, deparsing, and pg_dump #11; Jim tested it but expressed a preference for NULL syntax #12, while Andrei added documentation and ALTER SYSTEM TAP coverage #13.
- Responding to Jim's preference, Tom supplied an alternative using SET list_guc TO NULL, preserving existing empty-string elements and omitting the preceding documentation and TAP additions #14; Jim tested the new syntax, restart behavior, and invalid combinations successfully #15.
- Andrei endorsed the NULL approach but requested documentation and noted that assigning an empty string could still prevent restart #16; after nearly two months, Tom supplied documentation and a rebased patch, explaining that the change provides a valid empty-list syntax rather than making the old command safe #17.
- Tom treated checking preload-library files as separate work requiring further design and announced his intention to commit #18; Andreas Karlsson considered NULL the least problematic alternative #19, and Tom reiterated that introducing another fully reserved keyword was unjustified #20.
- Tom committed #17 as ff4597acd4, retaining its implementation and tests while adding an ALTER SYSTEM documentation example and polishing the commit message.
- After the commit, Maciek Sakrejda pointed to an earlier unsuccessful library-validation proposal #21, and Tom reopened the question of hard errors versus warnings or syntax-only validation to catch empty library names #22.
- The empty-list syntax had landed, but validation remained unresolved when Jacob Champion revived the thread in September 2026, proposing default SplitGUCList() validation for GUC_LIST_QUOTE parameters without check hooks and demonstrating malformed values reaching function deparsing #23.