model-writtenrun 5841 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Anders Kaseorg proposed preferring HOME for Unix home-directory lookup while retaining the password-database fallback in #1, motivated by shared UIDs and test environments that redirect HOME.
- Alvaro Herrera supported the convention in #2, while Michael Paquier recommended consolidating the duplicated lookup code in #3.
- Kyotaro Horiguchi questioned the security implications in #4; Anders explained SSH's historical setuid concerns in #5, and Kyotaro then endorsed the change in #6.
- Daniel Gustafsson questioned Unix portability in #7, and Anders answered with POSIX shell behavior and implementations across Linux and BSD in #8; Chapman Flack later endorsed changing the default rather than adding an option in #9.
- Tom Lane posted a revision in #10 that removed the fallback as supposedly unnecessary, replaced tilde notation with explicit $HOME documentation, and documented the existing Windows root-directory behavior of psql's directory-change command.
- Anders defended compatibility with cleared environments and the existing tilde documentation in #11 and #13, while Tom initially defended requiring HOME and changing the documentation in #12.
- Justin Pryzby reported init children without HOME in #14, and Christoph Moench-Tegeder identified affected PostgreSQL daemon environments and confirmed the behavior across Linux services in #15 and #17; Tom accepted restoring the fallback consistently in libpq and psql in #18.
- Tom committed Anders's original fallback design as 376ce3e404, with formatting and declaration cleanup plus nonfatal psql lookup errors; the proposed documentation rewrite and HOME-only behavior were omitted.
- After announcing that landing, Tom posted a new cleanup series in #19 to remove unnecessary system-header dependencies from port.h, move pqGethostbyname into its sole consumer, introduce simpler user lookup wrappers, and remove unused thread.c from the MSVC build.
- The thread ends with Tom requesting feedback on the cleanup API in #19, with no subsequent review shown; both cleanup patches later landed together as 98e93a1fc9, differing only by removal of an obsolete comment.