model-writtenrun 2356 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Bernd Helmle proposed sha256crypt and sha512crypt support for pgcrypto's crypt() and gen_salt(), using the existing OpenSSL digest infrastructure and Drepper's reference algorithm; yescrypt and benchmark-table updates were left outside this submission #1.
- Daniel Gustafsson questioned whether the patch duplicated OpenSSL functionality, and Bernd clarified that it already used pgcrypto's OpenSSL-backed digest API, resolving the concern #3 #4 #5.
- Japin Li found a crash with malformed salt input and hashing operations that could not be cancelled; Bernd's next revision corrected buffer sizing and copying, adopted mutually exclusive digest selection and interrupt checks, and rejected out-of-range rounds #6 #8 #9.
- Alvaro Herrera requested formatting fixes, reference tests, clearer error reporting, and StringInfo-based output construction; Bernd incorporated these and restored reference-compatible rounds clamping with NOTICE messages, while omitting low-level digest test vectors because OpenSSL supplied that functionality #10 #12 #14.
- Japin's next review exposed confusing malformed-salt behavior and inconsistent macros, leading Bernd and Alvaro to discuss whether compatibility justified accepting input that silently produced an empty salt #15 #16 #17.
- Bernd compared Drepper, passlib, and OpenSSL behavior, while Japin favored passlib-style strictness and Alvaro raised salt encoding and minimum-length concerns #19 #20 #21 #22 #23.
- Bernd's March revision added salt-character and reserved-token checks but retained short and empty salts for validating externally generated hashes, supported by compatibility experiments; gen_salt() continued to generate 16-character salts #24.
- Alvaro supplied two patches on top of that revision, combining formatting cleanup with portable rounds parsing and other local corrections; although he preferred rejecting empty salts, the attached changes did not implement that restriction #25.
- After successful cross-platform CI and Bernd's agreement, Alvaro committed the exact cumulative contents of the March patch and his review deltas as 749a9e20c9, then announced the landing #26 #27 #28.
- Tom Lane subsequently reported a buildfarm failure caused by passing plain char values to ctype functions and questioned locale-dependent salt validation, recommending an explicit allowed-character lookup instead #30.
- Andres Freund independently found a write through a pointer after pfree(), and Bernd supplied a follow-up patch that removed the stale write, made the character table const, and restricted salts using strchr() #31 #32 #33 #34.
- Andres confirmed that Valgrind had also detected the invalid write, and Tom committed the follow-up as 969ab9d4f5 with a small diagnostic adjustment to print one complete multibyte character #36 #38.
- At the final message, the SHA-2 feature and its reported portability and memory-safety fixes had landed; no attached patch remained pending, and the discussed minimum-salt-length restriction had not been implemented #25 #38.