model-writtenrun 8577 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Michael Paquier opened the thread with #1, proposing explicit SSL protocol-bound errors at startup or reload and regression tests because OpenSSL's connection error was confusing; his implementation reused protocol conversion to support older OpenSSL versions and checked bounds after configuring the context.
- Daniel Gustafsson argued in #2 that configuration validation belonged in library-independent GUC code, prompting Michael to replace the OpenSSL checks with paired GUC hooks in #3.
- Daniel endorsed the revised patch in #4, and Michael landed it unchanged apart from whitespace as 41aadeeb12, announced in #6, with a backport to PostgreSQL 12.
- Peter Eisentraut warned in #7 that cross-variable GUC checks depend on assignment order, and Tom Lane explained in #8 how changing both bounds could compare new values against old ones and incorrectly accept or reject configurations.
- Daniel suggested returning to the original OpenSSL approach in #9; Michael reverted the GUC implementation and its tests with 414c2fd1e1, then posted #10 restoring context-time validation with clearer errors because a replacement was too late for the imminent release.
- Daniel reviewed the restored approach in #11 and requested consistent quoting of configuration names and values; Michael made those light edits and committed #10 as 79dfa8afb2, confirming the landing in #12.
- During release testing, Daniel discovered that missing braces caused valid configurations with both bounds set to enter error cleanup, and supplied the focused fix in #13; Michael applied it as e30b0b5cfa and explained in #14 that reload would retain the old SSL context.
- The thread ended with all proposed fixes landed: Michael and Daniel declined an additional restart test because of its cost in #14 and #15, leaving the existing test that restores an unset maximum bound.