model-writtenrun 5709 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Oct - Nov 2021 · 3 entries
- Andres Freund proposed dropping Python 2 support from the development branch because Python 2 was already unsupported and its compatibility machinery complicated the Meson work; Tom Lane agreed, while released PostgreSQL branches would retain support #1 #2 #3.
- Reviewing an initial implementation posted in the Meson thread, Peter Eisentraut recommended searching for python3 before python, removing compatibility macros, and cleaning up documentation and Windows infrastructure #6; the naming discussion favored retaining plpython3u and leaving the old names available for possible external Python 2 extensions #8 #9.
- Tom Lane delayed the removal because half the relevant buildfarm members still used Python 2 and wanted to coordinate Python requirements with Meson #10; Peter demonstrated that PL/Python still worked with Python 3.2.6, while participants debated whether the build system and procedural language needed the same minimum version #22 #23.
Jan - Mar 2022 · 14 entries
- In January, Peter revived the removal for PostgreSQL 15 independently of the unfinished Meson conversion #27, and Tom notified buildfarm owners of a mid-February transition deadline #33.
- Andres and Tom agreed to reject Python 2 during configuration before removing its implementation, allowing buildfarm dependency failures to be distinguished from cleanup bugs #38 #39; Andres then posted the monolithic patch in #40 and split it into configuration, extension, regression, and C-code patches in #41.
- Peter requested documentation corrections and noticed that transform-extension deletions had been placed in the regression patch #42, while Chapman Flack clarified the redundant minimum-version wording #44; the configuration portion landed with the documentation fixes as 19252e8ec9, announced in #46.
- The configuration change triggered a buildfarm migration phase involving missing Python development packages, executable names, and stale configure caches #50 #58 #65 #68; Mark Wong eventually reported most of his animals operational again, with some disabled after system-package changes removed their compilers #72.
- Once the buildfarm stabilized, Andres posted the remaining work in #73, expanding removal of the Python 2/3 compatibility layer, regrouping transform-extension deletions, and adding the documentation patch.
- A Windows crash exposed incorrect Unicode API substitutions in the expanded cleanup #74; Tom independently supplied a small Makefile and documentation delta, recommended moving the historical Python-version section, and argued that conflict detection must remain for external Python 2 extensions #75.
- Andres incorporated Tom's delta, corrected the Unicode-wrapper substitutions, and documented the retained conflict guard in #76; Tom accepted that version and advised retaining python_majorversion references for possible future Python versions #77.
- The final series' extension and regression components landed unchanged as 76a29adee7 and db23464715, both drawn from #76.
- The C-code component of #76 landed with additional PLyString-to-PLyUnicode function renames as 9b7e24a2cb, and its documentation component landed unchanged as 4228cabb72; Andres announced the completed series in #78.
- Post-commit testing revealed two problems: Tom reported that wrasse still needed the deleted alternative error output #79, and Andres identified a missing plpython.h include exposed by crake's header checks #80.
- The missing include was fixed directly from that diagnosis by 54c72eb5e5, without an attached patch; Andres confirmed the fix in #81.
- Using Debian 8 to reproduce the older-Python behavior, Andres also reported pushing an expected-output repair and explained that the deleted alternative distinguished Python 3 versions rather than Python 2 from Python 3 #81.
- Tom questioned retaining a large alternative output file for one test difference #82, but Andres noted the test guarded against a historical traceback crash #83, and Peter recommended leaving it until Python 3.5 support naturally aged out #85.
- By the final message, the removal series and reported follow-up repairs had landed, and Andrew Dunstan explained that the remaining cross-version upgrade alarm was his own buildfarm testing error and was already fixed #84 #86.