model-writtenrun 1279 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Mahendra Singh Thalor reported that PostgreSQL accepted quoted database names containing newlines but pg_dumpall failed on them in #1, and Srinath Reddy Sadipiralla reproduced the problem and traced it to shell-argument validation in #2.
- After finding an existing comment recommending future rejection of CR/LF in database and role names, Mahendra proposed enforcing that restriction in #3 and submitted three alternatives in #4: reject database creation, skip affected databases during dumping, or fail with a clearer diagnostic.
- Srinath favored rejecting invalid names at creation and supplied TAP tests in #5, while Andrew Dunstan favored rejection plus better dump diagnostics and recommended strpbrk() in #9; Mahendra adopted strpbrk() and covered database renaming in #11.
- Andrew ruled out backpatching the behavior change in #13, and Álvaro Herrera required pg_upgrade --check to identify incompatible existing names in #15; Mahendra expanded rejection to roles in #16 and added upgrade checking in #17.
- Nathan Bossart proposed sharing the repeated checks and questioned whether restrictions belonged in the grammar in #18, while Srinath supplied a delta to report all invalid database and role names before failing in #19.
- Mahendra incorporated those reviews in #21 through a shared server helper and a combined pg_upgrade query, then adjusted diagnostic details and wording in #24 following Andrew's review in #23.
- Discussion broadened to restricting control characters in all identifiers: Álvaro preferred a limited restriction and objected to constructed error messages in #27, while Tom Lane favored broader restrictions but supported deferring the design beyond the v18 freeze in #30 and clarified the limitations of the SQL-standard argument in #32.
- Mahendra proposed a database-only pg_upgrade check as a smaller v18 cleanup in #34, but Nathan considered even that v19 material in #35; no part of the series landed during this phase.
- Mahendra restarted the full series for v19 in #36 and eventually rebased it in #37, returning to the database-and-role restrictions rather than the standalone database-only proposal.
- Andrew considered the revived patches fundamentally sound in #38, and Tom requested inline translatable errors, tests that checked the failure reason, and omission of redundant checks for v19 source clusters in #39; Mahendra addressed the tests and version gate but retained a renamed helper in #40.
- Álvaro requested consistent treatment of tablespace names in #41 and reiterated the translation problem in #42, with Tom supporting tablespace coverage in #43; Mahendra added tablespace checks and removed the shared error helper in #45.
- Nathan discussed broader name validation and remaining obligations when dumping older servers in #44, but the resulting series in #45 remained limited to CR/LF in database, role, and tablespace names.
- Andrew squashed the two patches, corrected error-reporting syntax, cleaned up the upgrade-check implementation, and adjusted additional incompatible dump tests in #46.
- The complete restriction and pre-v19 upgrade check landed as b380a56a3f from #46, with identical code and test content and only two proposed executable-bit changes omitted.
- Nathan reported a post-commit buildfarm failure in #47, and Andrew confirmed it had been fixed in #48; the thread ends with the selected series committed and that failure reported resolved, without an attached follow-up fix or a broader identifier restriction.