model-writtenrun 151 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- While reviewing separate pg_upgrade --initdb work, Heikki Linnakangas discovered that a configuration-only new data directory without --new-bindir could invoke "(null)/postgres", and submitted a patch to resolve binary directories first in #1.
- Chao Li agreed with the fix but reported that the updated test passed against unpatched master, questioning whether it covered the bug in #2.
- Heikki explained that no regression test for the reported bug was included: the test adjustment accommodated invalid binary paths failing before creation of pg_upgrade_output.d in #3.
- Bohyun Lee endorsed the sequence of resolving binary directories, locating actual data directories, and then checking them in #4.
- Heikki committed #1 with an identical diff as d733fcaf5c and announced completion in #5; the commit deferred backpatching because the issue seemed minor and explicitly supplying the binary directory provided a workaround.
- After the commit, Tom Lane pointed to an earlier report of the same problem and questioned whether it warranted reconsidering backpatching in #6; the thread ends with the development-branch fix landed and that question unanswered.