model-writtenrun 7861 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Justin Pryzby proposed integrated zstd compression in #1 after finding smaller and faster dumps than native zlib, while preserving archive recognition and transparent restoration.
- Tom Lane spotted an accidentally included core file in #2; Justin removed it and corrected suffixes, compression defaults, and initialization in #3.
- Justin reorganized the series in #4 to separate generic compression infrastructure from zstd, centralize algorithm metadata, and support plain-format compression; he also addressed MSVC compilation and unseekable-input errors, while leaving experimental tar compression unattached.
- Andrey Borodin reviewed the abstraction and build choices in #5; Justin supplied memory measurements and explained that zstd remained disabled by default because CI lacked the library in #6, while Daniil Zakhlystov suggested limiting memory through window-size restrictions in #7.
- Andreas Karlsson's code review in #8 identified unsupported read/write modes, questionable default-algorithm handling, suffix-discovery issues, and unrelated configure changes, with no subsequent patchset addressing those comments in this thread.
- Tomas Vondra argued in #9 that libpq's server-memory concerns did not justify equivalent restrictions here because pg_dump compression consumes memory on the client.
- David Steele marked the work Returned with Feedback in #10 because the patch had gone stale and no longer applied, leaving this thread without a landing.
- The work later continued in thread #47519, whose latest source #532330 led to 84adc8e20f; the commit explicitly credits this original discussion but uses the newer compression API and leaves threaded compression disabled.