model-writtenrun 1970 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Matheus Alcantara proposed reusing SCRAM ClientKey and ServerKey values from incoming authentication for outgoing postgres_fdw connections, avoiding stored plaintext passwords, and credited Peter Eisentraut's earlier prototype in #1.
- Jacob Champion questioned identical-verifier requirements and the security implications of sharing salts in #2, while Jelte Fennema-Nio argued for clustering and sharding uses and requested documentation in #3.
- Matheus expanded tests to cover both loopback and separate servers with copied verifiers, removed an unnecessary hash change in #6, and supplied initial documentation in #7.
- Jacob raised a more elaborate proxied-SCRAM alternative in #4; after Matheus reported successful channel-binding tests in #8, Jacob clarified in #9 that the channel-binding concerns belonged to that alternative and accepted identical verifiers as an explicit trust decision for this approach.
- Peter identified missing key-length validation, supplied allocation checks and cosmetic fixes, and requested fuller documentation and error guidance in #10; Matheus incorporated those changes and combined code with documentation in #11.
- Peter expanded the configuration requirements and clarified the libpq parameters in #12, and Matheus accepted and squashed those changes into #13.
- Peter committed #13 as 761c79508e with light formatting, test-text, and comment edits, announcing the landing in #14; the commit credits both Matheus and Peter as co-authors.
- Months later, Alexander Pyhalov reported that background workers can have a null MyProcPort and supplied checks for postgres_fdw and dblink in #16, followed by a reproducing extension in #18.
- Peter endorsed Alexander's fix in #19, and Matheus reproduced and tested it in #20, also proposing documentation that background workers need another authentication source because they have no incoming SCRAM keys.
- Peter committed Alexander's patch unchanged apart from rebasing as 138750dde4 and explained in #21 that the documentation should describe the general fallback behavior when incoming authentication does not use SCRAM.
- Matheus revised the documentation in #22 to explain that use_scram_passthrough is ignored without incoming SCRAM authentication and that a supplied user-mapping password is used instead; the feature and crash fix had landed, while this clarification remained proposed at the end of the thread.