model-writtenrun 3808 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Nazir Bilal Yavuz reported SSL test crashes with OpenSSL 3.2.0 on both macOS and Debian in #1.
- Tristan Partin submitted a fix co-authored with Bo Anderson in #2, explaining that PostgreSQL stored connection pointers in BIO private storage now used by OpenSSL, and switched both backend and libpq code to application storage.
- Tristan restored an accidentally removed comment in #3, then removed the remaining Autoconf probe in #5 after Michael Paquier identified the omission in #4.
- Tom Lane initially requested a compatibility probe and fallback in #6, but Bo established that the application-data API predated OpenSSL in #7, Tristan explained why falling back to BIO private storage was invalid in #8, and Tom accepted removing the probe in #9 and #11.
- Testing the final submission against MacPorts OpenSSL 3.2.0, Tom confirmed that crashes stopped but found three failures caused by changed certificate-revocation error wording in #12; Michael proposed accepting both spellings in #15, and Tom proposed a broader matching pattern in #16.
- Daniel Gustafsson clarified that BoringSSL was unsupported in #17, and Tom said this resolved the last compatibility concern in #23; Tom then took responsibility for finishing the test adjustments in #25.
- Tom committed #5 with three SSL test regex adjustments as c82207a548; the other supplied candidate, 0bd682246a, is the same fix adapted to PostgreSQL 12, without those test changes.
- Tom confirmed the push and healthy buildfarm results in #27, noting that the indri animal now used OpenSSL 3.2.0 to provide ongoing coverage.
- After Tristan noted BoringSSL's addition of the application-data APIs in #28, Daniel and Tom reiterated that its API stability policy did not justify PostgreSQL support in #29 and #30, with Alvaro Herrera agreeing in #32.
- The fix had landed, but Jelte Fennema-Nio reported potentially related SSL crashes in an MSYS2/MINGW build and noted that the fix appeared unreleased in #33; Tristan considered the OpenSSL 3.2.0 connection plausible and requested a backtrace in #34, leaving that diagnosis unresolved.