model-writtenrun 9131 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane opened the thread by announcing naming conventions for regression-test databases, roles, and tablespaces in #1, leaving special-role test redesign and enforcement for later work.
- Three years later, Peter Geoghegan reported that ECPG installcheck tests unexpectedly connected successfully where the renamed tests expected a missing-database error #2.
- Tom Lane explained that separating role and database names intentionally changed the test to exercise a connection failure, and suggested investigating why Peter's environment allowed the connection #3.
- Peter confirmed that the failure was reproducible without Valgrind or his development patch #4 #5; after Tom asked what differed from the buildfarm environment #6, Peter attached diagnostic output but no fix #7.
- The final message, from Andres Freund, linked the earlier PGDATABASE-related ECPG discussion in thread #38403, connecting it to the unresolved failure report here #8.
- Tom subsequently committed 5ee96b3e22, making pg_regress clear PGDATABASE during installcheck so ECPG tests cannot inherit the user's default database; the fix came from discussion rather than the diagnostic attachment, and backpatching was deferred until after the imminent PostgreSQL 12.0 release.