model-writtenrun 8368 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Roman Zharkov reported that pg_regress --outputdir failed because sql/ and expected/ were missing, and supplied an inline patch to create those subdirectories in #1.
- Daniel Gustafsson confirmed that Roman's fix worked and attached a complementary cleanup replacing an explicit stat() check with directory_exists() in #2.
- Michael Paquier combined both changes and removed redundant subdirectory creation from the Unix and MSVC pg_upgrade test drivers in #3, arguing that backpatching was unnecessary.
- Tom Lane pointed out that the combined patch would create unwanted directories when the input directory was absent in #4; Daniel supported moving creation after input processing and also questioned MSVC testtablespace creation in #5.
- Michael moved directory creation after input validation in #6 but retained the MSVC testtablespace setup for separate work already underway; Daniel accepted that scope in #7.
- Michael committed #6 with a behavior-preserving statement reorder as e78900afd2, then reported in #8 that check-world generated the same directory counts before and after the change; the fix landed on HEAD without backpatches.