model-writtenrun 4925 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Feb - Mar 2022 · 7 entries
- Julien Rouhaud proposed authentication-file inclusion so automation could replace application rules without disturbing administrator or replication access, alongside a pg_ident_file_mappings view and an exploratory pg_hba_matches() function in #1.
- Michael Paquier requested moving SQL-facing authentication code out of the growing hba.c in #3, and Julien added that prerequisite in #9; further review led to preserving existing function order and adopting the new set-returning-function helper in #13.
- Aleksander Alekseev reviewed and tested the extraction, mapping view, and inclusion functionality in #14 and #15, while recommending separate treatment of the unfinished pg_hba_matches() feature.
- Michael supplied cosmetic naming changes in #17 and committed that standalone extraction unchanged as d4781d8873, announced in #20.
- Julien added authentication-view TAP tests in #23, but Windows and EXEC_BACKEND reload authentication files for new connections, making malformed-file tests problematic; after Michael's expanded test draft in #25, they removed TAP coverage and strengthened SQL error checks in #28.
- Michael first extracted the existing pg_hba_file_rules regression-query improvement from #28 as 091a971bb5, then committed the mapping view with documentation, comment, and formatting corrections as a2c84990be, explaining both landings in #29.
- Julien resent the remaining inclusion and diagnostic-function patches in #30, adding include_dir, and returned after a gap with the rebased series in #31.
May - Nov 2022 · 23 entries
- Michael identified missing directory sorting and opportunities to share configuration-file machinery in #32 and #34; the ensuing discussion settled on include, include_dir, and include_if_exists matching postgresql.conf conventions, including sorted .conf files, in #38.
- The testing discussion in #39 and #40 separated successful view checks, server-log error checks, and malformed-file view checks, with persistent connections and EXEC_BACKEND portability remaining obstacles.
- While building those tests, Julien discovered that fill_ident_view() used the HBA filename to resolve relative ident-file references and posted #41; the exact fix landed as 27e0ee57f6 and was backpatched.
- The expanded series in #43 introduced comprehensive TAP coverage and separated logging and numbering prerequisites; Michael committed its real-filename logging patch as 47ab1ac822 and merged its two ident diagnostic patches as 718fe0a14a, as reported in #46.
- Julien carried the remaining work through August-to-October rebases in #48, #49, and #51, also removing a shadowed error variable in #50.
- Michael resumed detailed review in #52 and extracted source-file tracking in #54; that component landed as 1b73d0b1c3 with additional auth.c diagnostics and without the proposed IdentLine source-file member.
- Julien proposed simplifying view-filling arguments in #57, but Michael explained in #58 that parsed line structures can be NULL on errors; he discarded that simplification and committed the numbering component, renamed to map_number and rebased onto the retained interfaces, as c591300a8f, announced in #59.
- The pg_hba_matches() proof of concept disappeared from the focused series in #54; earlier review in #32 had questioned its dependence on internal connection state, and no diagnostic-function implementation landed in this thread.
- Michael requested smaller, independently reviewable test stages in #59 and #64, while Julien explained in #65 that repeatedly splitting and reconstructing the existing test infrastructure would be costly and difficult to maintain.
- Michael extracted two mapping-view documentation corrections from #57 as e76502871e, backpatched them, and took over posting rebased inclusion patches with his own simplifications in #66.
- The shared directory-listing and path utilities from both attachments in #66 landed with light cleanup as a1a7bb8f16, reducing duplication between GUC and authentication-file handling.
- Michael introduced AbsoluteConfigLocation() reuse and a common open_auth_file() interface in #69; the path reuse from #70 landed as 6bbd8b7385, retaining MAXPGPATH rather than adopting the posted dynamically sized buffer.
- Michael added a tokenization error-context callback in #73, Julien approved its filename-and-line diagnostics in #74, and the opening helper and callback landed separately with light cleanup as 783e8c69cb and ad6c52846f.
- In #76 and #78, Julien defended collecting all inclusion errors for useful diagnostic views and keeping missing optional files out of the error column; Michael accepted continuing tokenization after failures in #81, while retaining the rule that any real error prevents configuration replacement.
- Ian Lawrence Barwick noticed that the new TAP test was absent from meson.build in #79, and Julien acknowledged that the registration needed fixing in #80.
- Michael substantially reworked memory ownership in #82, also identifying that include_if_exists must ignore only ENOENT; after repeated reloads exposed leaks, #83 separated a static token-storage context from per-call temporary contexts and supplied a dedicated cleanup interface.
- Julien approved that memory design with a comment clarification in #84, and its prerequisite landed with light edits as efc981627a.
- A missing inc_lines initialization, already present in the inclusion attachment of #83, was extracted immediately afterward as d5566fbfeb to repair the prerequisite split.
- The inclusion feature from #83 then landed without its TAP tests as a54b658ce7, adding all three include directives and file_name columns after naming and documentation cleanup; Michael announced the feature and memory-management landings in #85.
- Julien supplied shared recursion-depth constants and a Windows-normalized test remainder in #87; the constants landed with comment adjustments as d13b684117, confirmed in #88.
- After Julien resent the remaining tests in #92, Michael committed only their positive cases as cbe6e482d7, revising counter handling, using basenames for portability, and adding coverage for existing @ file expansion.
- Michael posted the rebased negative-test remainder in #95, requesting simpler error-reporting helpers and basename-based expectations and suggesting a separate thread for that work.
- At the final message, #96, Michael marked the CommitFest entry committed: authentication-file inclusion and positive tests had landed, while the negative-test attachment in #95 remained uncommitted.