model-writtenrun 8262 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Daniel Gustafsson explained that OpenSSL 1.1.0 made RAND_cleanup() a no-op and submitted two patches: use RAND_poll() after fork, and simplify the random-seeding retry loop by relying on RAND_status() instead of stopping immediately when RAND_poll() fails #1.
- David Steele questioned whether callers expected rapid failure and whether immediate retries helped #2 #4; Daniel discussed the historical rationale #5, and Michael Paquier supported preserving retries for older OpenSSL versions while simplifying their failure handling and limiting the changes to HEAD #6.
- Noah Misch accepted the patches and considered, then rejected, calling both RAND_cleanup() and RAND_poll() because discarding the existing seed would slightly increase predictability #7; Daniel also raised possible FIPS problems with RAND_cleanup() and favored following upstream's RAND_poll() recommendation #9.
- Noah Misch committed both parts of the original submission #1: the fork change landed unchanged as ce4939ff70, and the retry cleanup landed with only comment polishing as 15e4419722.
- Michael Paquier confirmed that both patches had landed on HEAD and elaborated on RAND_cleanup() failures with FIPS configurations, reinforcing the choice to follow upstream's recommendation #10.
- After landing, Daniel concluded that the explicit reseeding call appeared unnecessary but harmless on OpenSSL 1.1.1 and proposed exploring RAND_priv_bytes() for sensitive randomness such as SCRAM salts #11.
- Noah asked whether private randomness needed a separate API #12, Michael favored keeping one API #13, and Daniel agreed because non-OpenSSL builds might lack an equivalent separate implementation; the thread ended with both original patches committed and Daniel promising further investigation and a follow-up patch #14.