model-writtenrun 6961 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Erica Zhang proposed configurable TLS 1.3 cipher suites and colon-separated ECDH curves in #1 to meet product security requirements; Peter Eisentraut and Michael Paquier explained in #2 and #4 that new features would target PostgreSQL 18 because PostgreSQL 17 was already frozen.
- Jacob Champion requested separate patches, the newer OpenSSL group API, and caution about manually validating group names in #5, while Daniel Gustafsson argued for preserving OpenSSL defaults and questioned the need for another cipher setting in #6.
- Jelte Fennema-Nio defended separate settings because OpenSSL exposes separate APIs in #8, and Erica adopted that design and split the work in #11.
- Andres Freund asked for a list-appropriate name, documentation, and better diagnostics in #14; adding X25519 to the default was explicitly left for separate work.
- Daniel took over revision of Erica's patches in #16, adding documentation and test configuration, aggregating invalid curve names, and supporting a blank cipher-suite setting, while identifying OpenSSL 1.1.1 as a prerequisite.
- Peter challenged the semantics, portability, and complexity of custom string parsing in #17, leading Daniel to remove it and delegate validation to OpenSSL in #18.
- Jacob demonstrated unhelpful group errors and proposed an old-name compatibility mapping in #19, then explained why an empty cipher setting should preserve runtime OpenSSL defaults rather than install compile-time defaults in #20.
- After the summer gap, Daniel incorporated those changes in #21, renamed the setting to ssl_groups, retained ssl_ecdh_curve as an alias, and added a prerequisite patch raising the minimum OpenSSL version to 1.1.1.
- Jacob reported incompatibility with older LibreSSL releases and requested clearer names and error handling in #22; Peter supported a protocol-specific cipher name in #23, and Daniel responded with ssl_tls13_ciphers, SSLerrmessageExt(), and error-path tests in #24.
- The digit-bearing GUC name exposed a test harness limitation, so Daniel added a separate fix accepting alphanumeric GUC names in #25, expanding the series to four patches.
- Jacob refined group-error wording and documentation in #26 and #28, while Peter endorsed the series and proposed removing obsolete randomness initialization in #29.
- Daniel's final series in #30 incorporated those refinements and documented the LibreSSL minimum; Jacob approved it after spotting a missing closing parenthesis in #32.
- The final series in #30 supplied the unchanged GUC-test fix committed as f81855171f and the OpenSSL prerequisite committed with light build-comment and documentation edits as 6c66b7443c; although the prerequisite was also discussed in thread #50252, this thread contains the latest matching submission.
- The group-list patch from #30 landed unchanged as 3d1ef3a15c, and its TLS cipher-suite patch landed with corrected documentation anchors as 45188c2ea2.
- Daniel announced that the whole series had landed in #33 and reported an expected buildfarm failure caused by the raised library requirements, with upgrades to be coordinated with animal owners.
- Nathan Bossart reopened the version-specific naming question in #34; Jacob, Tom Lane, and Daniel favored established application conventions in #35, #36, and #37, while Nathan accepted retaining the chosen name in #38 and Jonathan S. Katz still preferred a version-independent name in #39, without a replacement patch.
- Dmitry Dolgov revived the thread in #40 after demonstrating hybrid post-quantum group negotiation on PostgreSQL 18 and requested a backport for older releases; Daniel declined on feature-backport policy grounds in #41, leaving all four changes landed for PostgreSQL 18 with no further patch pending in this thread.