model-writtenrun 8683 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Luis Carril proposed an opt-in switch for dumping all foreign-table data through pg_dump and pg_dumpall in #1, using COPY (SELECT ...) because direct COPY TO does not support foreign tables.
- Daniel Gustafsson raised restoration risks for non-writable FDWs in #3, while Tom Lane argued for selective dumping and noted the lack of a guaranteed consistent foreign-data snapshot in #5; Luis proposed explicit server selection in #7, which Daniel supported in #8.
- Luis implemented foreign-server pattern selection in #9, retaining ordinary table inclusion and exclusion controls and dropping the original pg_dumpall changes.
- Daniel's testing and code review in #10 led to documentation, explicit errors for empty or unmatched patterns, a schema-only incompatibility check and clearer naming in #11.
- Surafel Temesgen and vignesh C requested tests and examined query handling and redundant checks in #12 and #13; Luis added option-error tests and simplified checks in #15, retaining the option name and explaining the need for COPY (SELECT ...) in #17.
- Daniel supplied a functioning test FDW in #18; Luis incorporated it in #20 and restored accidentally omitted files in #22, after which Daniel reported successful application, compilation and tests in #24.
- Laurenz Albe warned about duplicate remote data on restore in #19, and Álvaro Herrera questioned whether locally stored FDW data should instead use table access methods in #23; Luis defended explicit opt-in and local-storage FDW use cases in #20.
- Tom accepted that use cases could remain but objected to the custom FDW tests' maintenance cost and incomplete restore coverage in #25, proposing handlerless dummy servers and simpler catalog lookups; Luis implemented the query simplifications in #27 while retaining the original tests.
- Vignesh reproduced a parallel-dump failure caused by attempting to lock foreign tables in #30; Luis confirmed it in #31 and added an explicit parallel-dump prohibition with a test in #33.
- Vignesh questioned missing server definitions when using --table in #34, but Luis explained in #35 that existing selective dumps also omit prerequisites, so that behavior remained unchanged.
- Peter Eisentraut proposed skipping foreign-table locks to retain parallel dumping in #36, but Luis preferred a separate improvement in #40; Álvaro investigated and documented the shared pg_dump/pg_restore linkage obstacle in #42, with a non-compiling exploratory inline patch in #43.
- Álvaro took over patch preparation in #44, moved foreign-server lookup into the main catalog query, removed the redundant option flag and explicit empty-pattern check, and separated the disputed tests; Daniel explained the original empty-pattern safeguard in #45.
- Álvaro replaced the custom C FDW infrastructure with compact handlerless-FDW tests in #46, checking the expected extraction failure for a selected table and successful dumping of a server without tables.
- After reporting CI failures in #47, Álvaro committed both patches from #46 as 2f9eb31320, correcting the test count and command argument spelling; his final announcement in #48 left any remaining CI diagnosis to the buildfarm, with parallel foreign-data dumping deferred.