model-writtenrun 3717 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Joe Conway proposed moving the noninteractive portion of psql's password command into PQchangePassword(), giving libpq clients a convenient way to change passwords without sending plaintext to the server, with psql integration and documentation included #1.
- Tom Lane questioned the algorithm parameter, argument order, boolean return type, and nonblocking claims; Joe revised the API to take user before password, return PGresult, and use the server's password_encryption setting through PQencryptPasswordConn() #2 #4.
- Jonathan S. Katz proposed covering CREATE ROLE as well, but Tom Lane objected to maintaining an expanding C API, and Magnus Hagander argued that frameworks could combine role creation and password assignment; the series remained focused on password changes #5 #6 #15 #16.
- Peter Eisentraut questioned how the wrapper solved password leakage, while Joe and Tom explained its convenience and the need for adoption by other client libraries; Dave Cramer reported that comparable JDBC functionality was already underway and later available #10 #11 #12 #13 #20.
- Discussion explored client versus server encryption defaults and separate password-generation APIs; Joe retained existing psql behavior and deferred broader APIs, while Sehrope Sarkuni clarified that JDBC also used the server default for its end-to-end password-changing operation #18 #19 #21 #23 #25.
- Jim Nasby's suggestion to restrict SQL password setting was not incorporated, and the optional VALID UNTIL parameter raised initially likewise remained outside the submitted implementation #22 #1.
- Joe resent the unchanged series and announced plans to commit; Sehrope then suggested explicitly wiping password buffers, which Joe treated as an independent psql change #24 #26 #27 #28.
- Tom's full code review requested clearer documentation of encryption defaults and return values, removal of redundant initializations, a narrower buffer scope, and a simpler return statement; Joe incorporated those suggestions and ran pgindent in the final submission #29 #30.
- The supplied thread ends with that revised series awaiting commitment #30; Joe subsequently committed all three parts as a7be2a6c26, adding only a documentation cross-reference and crediting Tom Lane's review.