model-writtenrun 5518 - 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 identifying inconsistent ERROR/FATAL usage and three incompatible ways to print supplementary frontend messages, asking for a uniform logging convention in #1.
- Kyotaro Horiguchi proposed explicit detail messages in #2, while Robert Haas advocated backend-like conventions and lowercase frontend labels in #3; Tom combined these ideas with an error-and-exit shortcut in #4.
- Tom's initial API draft #5 introduced separate primary, detail, and hint components and removed the distinct FATAL severity; after Kyotaro questioned verbosity behavior, Tom deferred independent suppression of supplementary messages because the user-facing control was unclear in #7.
- Peter Eisentraut argued that logging should remain separate from flow control in #10 and #15, while Michael Paquier highlighted pg_dump's cleanup requirements in #13; broader backend-style exception and cleanup facilities remained outside this patch.
- After a three-month gap, Tom supplied the full frontend conversion in #17, showing substantial code reduction and replacing inconsistent supplementary-message patterns with explicit detail/hint calls.
- Andres Freund proposed keeping an explicitly named pg_fatal shortcut separate from logging in #18, and Tom explored preserving FATAL severity in #21 before rejecting that approach in #27 because indirect exits made consistent severity assignment impractical and changed many test expectations.
- The revised patch #27 standardized pg_fatal as error logging followed by exit, retained only ERROR severity, expanded wrapper conversions, and updated translation extraction; Michael questioned the number of detail/hint macros in #28, but Tom retained their names to distinguish severity from message components in #29.
- Andres endorsed the revised approach in #32, and Tom rebased it in #33; because the widespread edits repeatedly conflicted with other work, Tom planned to land it after the commitfest in #40.
- Daniel Gustafsson reviewed the conversions for control-flow regressions in #34, while Peter endorsed the patch and proposed program prefixes for secondary messages in #35; broader message rewriting and optional color changes were left for follow-up work.
- Tom kept the main patch focused on mechanical conversions in #38, and Daniel agreed to pursue broader wording improvements separately while questioning whether rare internal errors should lose translation support in #41.
- Tom committed #33 as 9a374b77fb, with rebased call sites, removal of unhelpful installation hints, and a corrected close-error message, reporting those final edits in #42.
- Tom then proposed centralizing log-level filtering in #43, reporting about a 5 KB reduction in pg_dump code size; Peter approved the change in #44, and 2c9381840f landed it with simplified non-debug macros while retaining DEBUG guards.
- After that commit, Peter revisited argument-evaluation costs in #46, and Tom explained in #47 that frontend logging arguments were generally cheap enough to favor reduced code size.
- Peter implemented the deferred program-prefix change in #48 so detail and hint lines aligned and could be filtered by program; it landed unchanged as a8cca6026e, with confirmation in #49.
- All submitted changes had landed by the final exchange: Dmitry Koval questioned the main commit's removal of PGDLLIMPORT in #50, and Tom confirmed in #51 that the removal was correct because the header is frontend-only, requiring no additional fix.