model-writtenrun 7396 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed adding client-side SNI so SSL-aware proxies could route connections, posting an initial implementation and asking whether the behavior should be configurable in #1.
- Matthias van de Meent found that the initial IP-address heuristic incorrectly excluded valid digit-prefixed hostnames in #2; Peter replaced it with a numeric-and-dot check in #6.
- Jesse Zhang asked whether SNI could enable hostname-specific server certificates in #3, and Peter explained in #4 that the patch supplied only the client side, with no known server-side implementation planned.
- Jacob Champion requested an opt-in privacy toggle because SNI exposes hostnames in #5; Peter cited existing driver behavior in #7, while Jacob clarified that Go's automatic SNI depended on hostname verification in #8.
- Stephen Frost explained that proxies would still need to understand PostgreSQL's startup handshake and requested both explicit documentation and an off switch in #11.
- Greg Stark questioned IPv6 detection in #9 and #12; Peter and Magnus Hagander explained that the colon check already handled it in #13 and #14, with Magnus requesting a clearer comment.
- Peter added the sslsni option, PGSSLSNI environment variable, proxy and privacy documentation, and clearer IP-address comments in #15; that version landed unchanged apart from rebasing as 5c55dc8b47, with SNI enabled by default as confirmed in #16.
- Jacob reported a crash when connecting with hostaddr but no hostname in #17; Tom Lane reproduced it in #18 and then identified the broader multi-host bug, recommending the selected connhost entry with NULL and empty-string guards in #20.
- Peter posted the corrective delta in #21, and Tom endorsed the explicit empty-string guard after inspecting the patch in #22.
- Michael Paquier tested the fix and identified a redundant SSL_CTX_free() call in #23; Peter removed that call and committed 37e1cce4dd, confirming completion in #24, so both the feature and its host-handling correction were landed by the end of the thread.