model-writtenrun 5473 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane proposed removing trailing newlines from pg_upgrade's message strings and adding them centrally, improving translation reuse and preventing inconsistent caller conventions without migrating to common/logging.c; the attached patch also asserted that format strings lack trailing newlines and intentionally removed some extra blank lines #1.
- Kyotaro Horiguchi supported the newline convention and removal of double newlines, questioned the PG_STATUS comment, and suggested a separate line-spacing function to avoid embedding leading newlines in translatable strings #2.
- Kyotaro also demonstrated that a missing postgres executable produced a misleading 'not a regular file' error and supplied an inline correction distinguishing missing or non-executable files from unreadable files #3 #4.
- Peter Eisentraut recommended targeting PostgreSQL 16 and using the patch as a starting point for further pg_upgrade cleanup #5.
- Tom explored tracking whether previous output ended with a newline to eliminate leading newlines, while preserving status-line completion; Greg Stark raised possible interference from subprocess output, which Tom considered unlikely because subprocess output generally goes to separate logs #6 #7 #8.
- Peter tested output before and after the patch in several scenarios, found no problematic differences, and accepted the disappearing blank lines while suggesting direct output for any necessary additional spacing #9.
- Tom committed the attached patch as 7652353d87, preserving its design while rebasing, normalizing newly added messages, and incorporating Kyotaro's executable-check wording corrections; he announced the push in #10.
- Tom then committed 920072339f, replacing the executable-check wording workaround with appropriate errno values from validate_exec() and detailed %m diagnostics in its callers, explaining this broader solution in #11.
- Both changes had landed by the thread's end, while removal of leading newlines and a possible blank-line logging helper remained future work that Tom explicitly declined to pursue at that time #10 #11.