model-writtenrun 5570 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Daniel Gustafsson reported that FIPS-enabled OpenSSL could queue two errors for a prohibited digest, causing successive calls to display stale errors, and proposed clearing the queue before cryptohash and HMAC context creation instead of changing the error-string API to concatenate errors #1.
- Tom Lane warned that clearing only before creation could leave errors for unrelated code and requested cleanup after error consumption plus comments #2; Daniel agreed to clear at both ends #3.
- Michael Paquier raised broader concerns about TLS and pgcrypto error queues and suggested centralizing cleanup in SSLerrmessage() #4, but Daniel chose a localized cleanup after cryptohash initialization failure while retaining the usual practice of clearing before OpenSSL operations #5.
- Daniel's revised series also replaced generic pgcrypto initialization failures with PXE_CIPHER_INIT, removed the unused generic error, and removed a question mark from the initialization-error text, explicitly leaving the corresponding test-output updates for later #5.
- Tom endorsed the revised patches #6; Michael requested a comment hyphenation fix, maintained his preference for broader cleanup, and suggested exposing OpenSSL diagnostics through pgcrypto #7.
- Daniel accepted the comment correction, defended clearing the queue at the start of each computation, and left richer pgcrypto diagnostics as a possible PostgreSQL 16 investigation #8.
- The revised series #5 landed on May 6: the queue cleanup received only the comment correction in 17ec5fa502, the pgcrypto error-code cleanup landed unchanged in 0250a167a0, and the punctuation cleanup gained matching regression expected-output changes in ee97d46cdb.
- Michael's final reply accepted clearing the queue when beginning a computation and confirmed that the entire submitted series had been applied, leaving no attached patches outstanding #9.