model-writtenrun 4909 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed replacing pg_dump's historical blob terminology with large object or LO, retaining archive-format compatibility and adding new long-option aliases in #1.
- Daniel Gustafsson reviewed the patch in #2, identifying a commit-message typo, ambiguous archive-entry comments and a missed test comment; Peter incorporated those changes and restored legacy options in help and documentation in #3, which Daniel approved in #4.
- Tom Lane raised the separate problem of excessive memory use with many large objects and proposed grouping objects with identical owners and ACLs into shared TOC entries in #5; Daniel confirmed that the terminology patch would not obstruct that work in #6.
- Andrew Dunstan supported grouping in #7, but Tom explained in #8 that it could prevent selective restoration of individual objects through pg_restore -L, motivating an optional compatibility mode; Andrew favored batching by default in #9.
- Peter committed #3 unchanged as 35ce24c333 and announced completion in #10.
- Robert Haas argued for preserving selective restore and investigating reduced TOC memory residency in #11; Tom identified rereading or temporary-file costs in #12, and Robert suggested grouping entries for incremental reading in #13, leaving scalability unresolved when the supplied thread ended.
- A later regression reintroduced blob wording in three directory-archive error messages; following a report credited to Kyotaro Horiguchi, Daniel restored LO wording in 94851e4b90 on February 24, 2023, with no corresponding attachment or review exchange in the supplied thread.