model-writtenrun 7797 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Robert Haas reported that pgbench displayed an empty database name and misleadingly claimed a socket connection failure even though the server had returned a missing-database error #1.
- Tom Lane explained that identifying the target server matters for multi-host connections and proposed removing pgbench's redundant prefix #2; Robert distinguished transport failure from database connection failure, and they settled on connection-to-server wording #3 #4 #5.
- Alvaro Herrera traced the empty name to pgbench printing its input argument instead of libpq's resolved database name and posted a one-line PQdb() fix #6 #8; Tom supported retaining that approach for stable branches while simplifying development-branch messages #9.
- Tom finalized the libpq wording proposal in #11 and implemented it directly from the discussion in 27a48e5a16, updating the connection-message helper and affected tests before announcing completion in #12.
- Tom also posted a broader removal of redundant connection-error prefixes, citing psql's existing behavior as precedent #12; that patch landed unchanged across programs, examples, documentation and tests in 58cd8dca3d.
- Alvaro's original PQdb() patch #8 landed for branches 9.5 through 13 as f2dc962360, expanded to additional error sites and NULL-safe database-name handling; he confirmed the push in #13.
- While checking the remaining old-style message in pgbench, Alvaro found an unreachable CONNECTION_BAD block and supplied a deletion patch #13; Tom confirmed the analysis #14, and it landed unchanged as 6819b9042f, with completion reported in #15.
- In May, Peter Eisentraut reopened the usability question, arguing that long socket paths distracted from the actual server error in PostgreSQL 14 #16.
- Robert defended the additional server identity and conventional prefix ordering #17, and Tom demonstrated why they clarify failures involving multiple servers #18; the thread ended with all four changes landed and no further patch or reversal proposed.