model-writtenrun 7208 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Aug 2020 · 3 entries
- Andres Freund opened the discussion in August 2020 by proposing regular automated pgindent runs and typedef updates to reduce unrelated changes during patch development and rebasing #1.
- Tom Lane supported more frequent, smaller formatting runs but warned that comment formatting and platform-dependent typedef extraction still needed human oversight #5 #6.
- Noah Misch proposed rejecting improperly formatted pushes, while Magnus Hagander favored local hooks and reviewable automation; Magnus subsequently implemented an experimental job publishing a separately formatted branch every fifteen minutes #8 #14 #35.
Jan - Feb 2023 · 15 entries
- Jelte Fennema-Nio revived the discussion in January 2023 after encountering formatting problems in libpq work, asking for easier tooling and stronger enforcement #36.
- Experiments with alternative formatters exposed extensive style differences, and concerns about version-dependent output kept attention on improving the existing tools #60 #63 #72.
- Andrew Dunstan's non-destructive pgindent patches gained help, documentation and improved diff output through Tom Lane's review, landing from #61 with spelling and presentation corrections as b90f0b5747.
- Mandatory server-side rejection remained disputed because developers sometimes intentionally separate formatting commits, and Andrew stated that project-wide enforcement lacked consensus #66 #71 #82.
- Andres proposed importing pg_bsd_indent to remove the separate-installation obstacle, resolved Tom's concern about the BSD advertising clause, and encouraged build integration while preserving the formatter's upstream indentation style #74 #79 #86 #87.
- Jelte's editorconfig and optional pre-commit-hook series moved Git file discovery from pgindent into the shell hook after Andrew's review; different staging workflows later led Andrew to publish hook examples on the developer wiki #81 #84 #85 #109.
- Andrew separately corrected exclusion handling for explicitly named files, with Jelte reviewing the fix before Andrew reported it committed #99 #100 #102.
- Andrew proposed rewriting pgperltidy with non-destructive modes, then demonstrated newer perltidy settings that reduce alignment-induced churn; the script rewrite has no demonstrated landing in this thread #120 #130.
- Robert Haas requested directory and commit-based file selection, and Jelte's testing exposed an accidental whole-tree fallback; Andrew's corrected #156 landed unchanged as 068a243b77 #131 #155 #158.
- Noriyoshi Shinoda supplied a follow-up correction to the --commit help indentation, which Andrew reported committed #160 #161.
- Shi Yu discovered that --commit tried to format deleted files, Jelte supplied filtering alongside dirty/staged options, and Andrew separated and strengthened the bug fix with existence checks #162 #163 #164 #165.
- The missing-file fix from #165 landed with additional duplicate suppression as dab07e8c68, while Andrew questioned further dedicated Git-selection options and suggested aliases instead #166.
- The formatter import proposed here was implemented in the separate thread thread #47416, whose attached series #527861 supplied the source import 4e831f4cee, copyright-notice cleanup b44e5fced3, and combined Make/Meson/TAP integration and portability fixes 156c049bee.
- With pg_bsd_indent in-tree, Andrew removed --build and --code-base and required explicit selection; #168 landed as b16259b3c1 with a source-root chdir added after Justin Pryzby's subdirectory failure report #169 #171.
- Dedicated dirty/staged switches remained unsettled as participants debated tracked files, untracked files and directory scope, while Shi Yu's separate FILE|DIR help correction was reported committed #173 #177 #180 #181 #182.
Apr - Jun 2023 · 4 entries
- The April restart shifted attention toward test-based checks and a buildfarm animal; Tom deferred bulk reformatting until pending feature decisions and perltidy settings were settled, while Magnus restored the existing formatted-branch mirror #183 #185 #191 #203.
- Andrew verified the newer perltidy version, and Alvaro Herrera tested and expanded the alignment exclusion list to preserve useful comment alignment; Tom rejected an additional comma-break option after experiments produced excessive unrelated changes #194 #211 #216.
- Tom's options patch #216, approved by Andrew in #217, landed as df6b19fbbc with one README invocation correction, followed by the freshly generated whole-tree formatting run 0245f8db36.
- Andrew deployed koel to check formatting, and its switch to whole-tree checks exposed existing differences; Tom's ensuing cleanup became b334612b8a #220 #221 #223 #224.
Aug 2023 · 2 entries
- Peter Geoghegan warned that frequent unignored formatting commits damaged blame history, prompting Nathan Bossart's #250, which Peter reviewed and which landed unchanged as 0df7d1da40 #226 #251.
- August discussion treated the committed typedef list as canonical while seeking easier local generation and an optional compile/update/indent workflow, with concerns about making ordinary tests expensive or disruptive #229 #240 #247 #248.
Oct 2023 · 6 entries
- Further koel failures revived the enforcement debate in October: Robert Haas emphasized preventing predictable human mistakes, while Tom preferred supervised reindent runs and resisted compulsory automated rewriting #253 #263 #264 #267.
- Jelte supplied a server-side update-hook example, but Magnus identified rebuild, branch-version and sandboxing requirements; David Rowley's proposal for opt-in indentation checks attracted support without establishing mandatory enforcement #280 #281 #283 #284 #290.
- Andres introduced Meson indent-tree, indent-head, indent-check and indent-diff targets in #292, then expanded the series with incremental local typedef extraction, merged lists and an optional nitpick test in #293.
- Tom's concerns about noisy generated typedef lists led Andres to adopt source-token filtering; #296 added that filtering and demonstration reformat output while remaining explicitly unfinished and limited to Linux/macOS extraction #294 #295.
- Andrew shared his pre-commit hook and proposed suppressing comment reflow in koel, but Tom questioned both the benefit and its consistency with keeping files clean for later edits #305 #306 #307.
- The thread ends with Andrew acknowledging that uncertainty in #308; the earlier tooling and maintenance commits have landed, while the Meson automation series, dedicated dirty/staged options and enforcement/comment-reflow policy remain unresolved.