model-writtenrun 9258 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Apr 2019 · 2 entries
- Jonathan S. Katz opened the thread with a documentation patch that moved authentication advice beside the initdb examples and prominently warned about trust authentication, while Peter Eisentraut argued for changing the defaults themselves and Magnus Hagander supported both safer defaults and a backpatched warning #1 #2 #4.
- Peter Eisentraut cautioned that SCRAM was not ready as an out-of-the-box recommendation, then proposed peer for local connections and password authentication for TCP/IP, with password authentication also serving platforms without peer support #6 #10 #11.
May - Jul 2019 · 12 entries
- Jonathan S. Katz revised the warning to account for Windows lacking peer authentication, retaining the SCRAM recommendation while acknowledging that the default password-encryption setting made it problematic #13.
- Tom Lane favored switching directly to SCRAM, but Magnus Hagander distinguished the relatively limited impact of abandoning trust from the broader client-compatibility consequences of changing password_encryption; Peter Eisentraut ruled out making the change for PostgreSQL 12 #14 #18 #22.
- The SCRAM discussion expanded into migration support: Heikki Linnakangas explained the existing md5-to-SCRAM transition mechanism, Stephen Frost identified remaining compatibility and observability limitations, and Jonathan S. Katz advocated clearer migration instructions #28 #29 #30 #31 #32.
- Noah Misch proposed implementing peer-like behavior on Windows through SSPI, while Magnus Hagander and Stephen Frost questioned calling TCP-based authentication peer; possible SSPI mappings remained a separate design discussion #33 #35 #36 #37.
- Peter Eisentraut replaced the documentation-only approach with a six-file implementation selecting peer where supported and md5 otherwise, explicitly leaving SCRAM for separate work and making regression tests request trust authentication #38.
- Julien Rouhaud tested the implementation and documentation successfully, requested an explicit explanation of failures on platforms without peer, and supplied that explanation on top of Peter Eisentraut's patch after agreement to place it in the initdb reference page #39 #41 #42.
- Peter Eisentraut committed the implementation from #38 together with the documentation addition from #42 as 09f08930f0, lightly rewriting the added paragraph and announcing the push in #43.
- Tom Lane reported widespread buildfarm failures because its forced database username often differed from the operating-system username, while Windows required a superuser password; Andrew Dunstan agreed to reverting, and Tom's subsequent HP-UX report referred to the revert having happened #44 #45 #46 #47 #49.
- Andrew Dunstan found that explicitly requesting trust fixed the main buildfarm initialization path, while Tom Lane corrected the assumption that AIX lacked peer and identified fragile header-dependent HAVE_AUTH_PEER detection that should instead be handled by configure #50 #52.
- Jonathan S. Katz proposed reviving the original documentation recommendations as an interim measure, while Peter Eisentraut endorsed explicit trust authentication for the test infrastructure #53 #54.
- Andrew Dunstan reported implementing buildfarm-client changes and supplied a separate patch synchronizing Windows upgrade-test initialization with the Unix script, including explicit trust authentication, WAL segment size, group permissions, and temporary configuration handling; the thread contains no landing announcement for that patch #56 #58.
- At the thread's end, the defaults change remained reverted and further work was open: Peter Eisentraut proposed allowing the data-directory owner to authenticate locally as any database user, while Tom Lane questioned the safety and user-facing consequences, and both treated that proposal as separate from the required buildfarm fixes #59 #60 #61.