model-writtenrun 6312 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
May - Sep 2020 · 11 entries
- Daniel Gustafsson opened the thread with exploratory replacements for deprecated OpenSSL APIs and a separately applicable switch from DES to AES256 for SSL test keys in #1, while identifying pgcrypto as further work.
- After Laurenz Albe raised minimum-version compatibility in #2, Peter Eisentraut proposed declaring OPENSSL_API_COMPAT instead of replacing APIs immediately in #4 and #6; Daniel accepted that direction in #11.
- Provider policy was an early concern: Peter asked about existing DES keys and FIPS in #4, and Daniel proposed leaving provider selection to the user's OpenSSL configuration in #5.
- Peter's request for reproducible callback-test certificates in #9 led Andrew Dunstan to submit Makefile rules in #14; Tom Lane requested removal of the duplicate script comments in #15, and that lightly revised patch landed as b846091fd0.
- Peter and Michael Paquier subsequently requested consistent certificate-target names in #16 and #18, but Andrew defended keeping the module self-contained in #17, and these later suggestions are absent from the supplied commit.
- Andrew endorsed the AES256 test-data update in #7, and Peter committed that unchanged member of #1 as f0d2c65f17, announcing it in #20.
- Peter initially placed OPENSSL_API_COMPAT in c.h in #21, then moved it into configure and MSVC configuration generation in #29 so configure probes would see the same setting, following the reasoning in #27.
- Robert Haas questioned continued reliance on deprecated interfaces in #24, but Peter argued that OpenSSL's long compatibility history allowed deferring migration in #25; the compatibility definition landed as 4d3db13621 with a corrected version literal and Michael's requested minimum-version comment, as explained in #31.
- Michael revisited the initial API-migration sketch in #33, questioning the extra wrappers and recommending older EVP entry points and feature probes; those exploratory replacements were not part of the supplied landings.
- Daniel returned with pgcrypto provider loading and explicit padding control in #34; Michael's review in #35 led to checked padding calls and configure-based provider detection in #36, while the accompanying SSL error-message patch was withdrawn because upstream behavior was still changing.
- Peter questioned automatically enabling legacy algorithms in #37, prompting Daniel's redesign in #42 to respect openssl.cnf, document legacy-provider activation, and initialize configuration consistently on older OpenSSL versions.
Nov 2020 · 1 entry
- Continued upstream error-message changes reported in #43 delayed expected-output updates, and Daniel corrected the proposed documentation's cipher examples while recommending waiting for release in #47.
Mar 2021 · 2 entries
- Testing OpenSSL alpha12 exposed garbage decryption results and provider-configuration problems in #48; Daniel supplied a configuration enabling both default and legacy providers and demonstrated the padding fix in #49, while Michael confirmed older-version compatibility in #50.
- Peter isolated the ignored decryption-error return into #51, Daniel verified it with and without legacy algorithms in #52, and the patch landed unchanged as 22e1943f13, announced in #54.
Jul - Sep 2021 · 6 entries
- With OpenSSL beta1, Peter explained that EVP_DecryptUpdate() retained the final block when padding was enabled and supplied an alternative OpenSSL-managed-padding demonstration in #55, but endorsed explicit padding disabling as the lower-risk fix.
- Daniel retained the small padding fix and added alternative regression outputs and legacy-provider documentation in #56, declining the PX padding rewrite and abandoning older-version configuration initialization because its benefit did not justify the work.
- Michael approved the final series and reported that upstream had resolved the SSL-test error-message discrepancy in #57; its unchanged members landed separately as 318df80235 and 72bbff4cd6, announced in #63, with backpatching deferred pending buildfarm coverage.
- Daniel reported clean tests against the release-frozen OpenSSL tree in #64, and Michael's Fedora buildfarm failure report in #65 supplied the coverage needed to proceed with stable-branch fixes.
- Daniel determined that older branches also needed Peter's decryption-error check in #67; Robert Haas and Peter supported backpatching in #68 and #69, and the unchanged patch from #51 was applied as 841075a65c, with the broader stable-branch push announced in #71.
- PostgreSQL 9.6's older non-EVP implementation complicated full compatibility, and Michael and Tom opposed invasive changes near its final release in #72 and #74; Daniel instead proposed documenting the lack of support in #75, without a resulting patch in this thread.
Jun 2022 · 1 entry
- The thread reopened when Jelte Fennema-Nio reported PostgreSQL 13 deprecation warnings on Ubuntu 22.04 in #76; its last exchange distinguished the landed functional fixes from warning suppression, with Daniel explaining the testing cost and older minimum-version differences in #79, leaving broader warning-suppression backpatching unresolved here.