model-writtenrun 7747 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Kyotaro Horiguchi proposed accepting a CRL directory through ssl_crl_file in #1, and Stephen Frost confirmed that CRLs were used in practice and supported the feature in #3.
- Henry B. Hotz asked about delta CRLs in #2; Horiguchi explained directory lookups and concatenated CRL files in #4, then corrected his restart assumption in #6: existing CRL files already reload on SIGHUP, leaving selective on-demand loading as the directory method's advantage.
- Horiguchi posted the initial server and libpq patch in #7, but noted that per-backend verification prevents sharing the CRL cache across sessions; Frost encouraged proceeding and leaving cache improvements for future work in #8.
- Horiguchi emphasized easier file management and avoiding configuration reloads in #9, then corrected tests, revised the SSLServer.pm interface, and polished documentation in #10.
- After a several-month gap, Peter Eisentraut proposed separate file and directory settings so users could adopt directories gradually in #12; Horiguchi implemented ssl_crl_dir and sslcrldir in #15, while explicitly leaving analogous CA-directory support out because he saw less benefit in #13.
- Eisentraut's review in #16 identified unnecessary certificate regeneration, missing Makefile dependencies, stale postgres_fdw expected output, and style issues; Horiguchi addressed these in #17 and reported successful SSL tests with both existing and regenerated fixtures.
- Horiguchi rebased the patch after an SSL test infrastructure change, clarified the default CRL-file documentation, and added the sample configuration entry in #18.
- Eisentraut committed #18 as f5465fade9 with light edits, explaining in #19 that he consolidated documentation into the setting descriptions and referred readers to OpenSSL rehash tools instead of documenting internal directory formats.
- The feature was fully committed by the thread's end, with Horiguchi accepting the documentation changes in #20; Eisentraut noted in #19 that support for other SSL providers could require revisiting the documentation.