model-writtenrun 9635 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- hubert depesz lubaczewski reported a directory-format restore crash in #1; although he withdrew his report in #5, Justin Pryzby reproduced a crash with an older pg_restore in #3, and Tom Lane traced it to formerly empty TOC fields becoming NULL in #4.
- Michael Paquier raised an archive-version bump in #6, but Tom Lane argued in #7 that a notational refactoring should preserve file compatibility, and Michael agreed in #8.
- Dmitry Dolgov supplied an initial fix in #9 that substituted empty strings inside ArchiveEntry, then rejected doing the substitution in WriteToc as offering no consistency benefit in #10.
- Alvaro Herrera proposed restoring explicit empty-string arguments at ArchiveEntry call sites in #14, supported by Tom Lane in #15; Dmitry implemented that revision in #19 and explained why sanitize_line needed no accompanying change.
- Alvaro Herrera and Tom Lane agreed to defer consistent NULL handling until a substantive feature required an archive-version bump in #17 and #18, and Alvaro committed #19 as 19455c9f56 with an added reminder comment, announcing the push in #21.
- Andres Freund supplied separate table-access-method infrastructure and pg_dump patches in #23, explaining that switching the default access method during restore made dumps easier to adjust but required a new archive format.
- Tom Lane accepted a feature-driven format bump and suggested associating access methods with tablespaces in #24; Andres rejected that coupling in #25 because it would require additional superuser-created tablespaces and fit the catalogs poorly.
- Dmitry Dolgov posted #26 on top of the table-access-method patches, restoring NULL defaults and removing the explicit empty-string workaround now that the proposed archive version prevented older readers from accepting incompatible dumps.
- Alvaro Herrera revived the overlooked cleanup in #27 after the archive-version bump had landed; the subsequent responses in #28, #29 and #30 expressed timing approval or general agreement without substantive patch review.
- Alvaro Herrera committed #26 unchanged apart from rebasing as 7fcdb5e002 and confirmed in #31 that the open item was closed, completing both the compatibility repair and the deferred NULL cleanup.