model-writtenrun 8376 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Oct 2018 - Mar 2019 · 6 entries
- Alexey Kondratov proposed letting pg_rewind fetch missing target WAL from archives, supplying either a command-line restore command or one parsed from recovery.conf, with a copied frontend lexer and archive tests in #1.
- Alvaro Herrera suggested asking the backend for configuration values in #4, while Michael Paquier challenged parser duplication, copied restoration code, Windows tests, and %r semantics in #10; Alexey first addressed Windows support in #11.
- After recovery settings became GUCs, Alexey adapted the patch to postgresql.conf in #13 and then replaced the copied lexer with a shared frontend/backend parser in #14.
- Andrey Borodin's review in #15 led to documentation, conflicting-option checks, and rejection of %r during rewind in #16, followed by Windows and obsolete-state cleanup in #17.
- Andres Freund identified configuration precedence problems and demanded clearer separation in #21 and #23; Alexey first split the series in #22, then abandoned the lexer work and used postgres -C in #26, also following Alvaro Herrera's recommendation in #24.
- Andrey's buffer-size review in #27 produced the smaller cleanup in #29, while Michael's signal-handling concern in #31 led to explicit interruption handling in #32.
Jul - Aug 2019 · 1 entry
- After a several-month gap, Alexey rebased onto the unified frontend logging API in #34; Liudmila Mantrova then supplied documentation and wording improvements in #35, which Alexey selectively incorporated in #36.
Sep 2019 · 1 entry
- Alexey renamed the options to -c/--restore-target-wal and -C/--target-restore-command in #37 to reserve -R for recovery configuration, and later adapted archive tests to bypass automatic crash recovery after their WAL files had been moved in #39.
Jan - Apr 2020 · 15 entries
- Alexander Korotkov took up final cleanup in #40, but Michael Paquier's review in #41 prompted moving restoration into a frontend library and dropping the explicit command-line restore-command option in #44.
- Michael identified the remaining duplicated command-expansion code and custom test helpers in #45, leading Alexey to introduce a shared archive command builder and reuse RecursiveCopy and PostgresNode restoration helpers in #46.
- Alexander and Alexey refined permissions comments, documentation, and formatting through #47, #48, and #49, before Michael found missing frontend/backend guards and unquoted executable paths in #51.
- Alexey fixed those issues in #52, rebased in #53, and cleaned up diagnostics and option ordering in #55 following Michael's further review in #54.
- Michael supplied a broader cleanup in #58, extracting getRestoreCommand, reusing pipe_read_line and pg_strip_crlf, strengthening failure handling, and reporting tests on Windows and Linux.
- Michael took responsibility for committing in #62, and Alvaro's objection to introducing unused frontend code in #63 led to isolating the already-used shared command builder in #65.
- The builder API progressed from an explicit buffer length in #67 to an allocated return value in #72, following Alexander's buffer discussion in #68 and Alvaro's allocation proposal in #71.
- Michael restored the two-part series in #73, allowing unknown archive-file sizes and distinguishing missing files from fatal restoration failures so the frontend API could serve tools beyond pg_rewind.
- Alvaro requested removal of the fixed command-size limit in #75, and Michael implemented StringInfo-based expansion in #76; Alexey also raised the unresolved question of placing frontend-only code in src/fe_utils in #74.
- After Alvaro's final header-comment correction in #78, Michael committed the builder from #76 as e09ad07b21 and posted the unchanged remaining feature in #80.
- To permit frontend and backend restoration routines to share the same name without conflicting declarations, Michael proposed separating backend archive prototypes in #81, and Alexey supplied xlogarchive.h in #82.
- Alvaro assessed that header cleanup and suggested broader possible restructuring in #83, while Michael explained why the narrow split was sufficient for the pending feature in #84.
- Michael combined the header prerequisite with the remaining feature in #86, Alexander accepted the series in #87, and Alexey supplied final naming, indentation, and comment fixes in #88.
- Michael committed the header component of #88 as 616ae3d2b0, adjusting include ordering and declining an unnecessary frontend compilation guard as explained in #89.
- Michael then committed the feature component of #88 with light textual cleanup as a7e8ece41c and declared the original work complete in #90.
Jun 2020 · 3 entries
- Peter Eisentraut reopened the frontend-library placement issue in #92, and Michael supplied the move to src/fe_utils in #94, retaining the shared RestoreArchivedFile name rather than adopting the rename proposed in #93.
- Michael applied that relocation from #94 as a3b2bf1fe7 and confirmed it in #98; existing frontend-only APIs were left in place because moving them could break external users, as discussed in #96.
- The thread ended with all submitted work landed: Stephen Frost questioned external-library compatibility in #99, and Michael answered in #100 that these archive APIs were new to PostgreSQL 13 and could still be moved during beta.