model-writtenrun 1807 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- JiaoShuntian opened the thread by asking for GB18030-2022 support and suggesting that the conversion generator and mapping tables could implement it #1 #2.
- John Naylor initially suggested a separate encoding to contain compatibility risks, and Andrew Dunstan supported that direction, but Tom Lane argued that a new encoding ID introduced version-skew hazards and favored updating the existing client-only encoding; Chao Li adopted that approach and John accepted it #4 #5 #6 #8 #9.
- Peter Eisentraut pointed out that users needing the old behavior could provide custom conversions through CREATE CONVERSION #10.
- Chao Li submitted the initial upgrade, explaining that the 66 newly required characters already worked algorithmically, nine optional compatibility characters would remain supported, and 18 changed mappings caused the compatibility break; the patch bundled UCM data, regenerated maps, and added a regression test #11.
- After Chao Li refreshed the submission, John Naylor requested separate commits for switching input format and changing the standard, so reviewers could distinguish preparation from mapping changes #12 #13 #15.
- Discussion of the upstream data's copyright notices led Peter Eisentraut to propose downloading UCM inputs on demand, while John Naylor discovered that EUC_CN still needed the checked-in XML file and asked that its deletion be deferred #18 #20 #21.
- Chao Li produced preparation using a historical GB18030-2000 UCM file that regenerated identical maps, then reduced parser churn and clarified round-trip filtering after John's review; John verified both build systems and the new test, and Chao repackaged the two patches in their proper order #23 #24 #25 #26 #27.
- John Naylor supplied a revised series with a separate parser-cleanup patch, requested broader mapping comparisons, and proposed documenting the supported standard version; Chao Li added the documentation and confirmed that Zheng Tao had co-authored the encoding update #29 #30.
- John Naylor supplied a diagnostic test-output diff, and Chao Li's independent comparison explained the additional differences beyond the 18 remappings: the corrected swapped mappings for U+E7C7 and U+1E3F, and rejection of the former A3A0 round-trip mapping #31 #32.
- Chao Li revised the commit explanation and added references for algorithmic conversion ranges, then rebased the series without changing its patch bodies #33 #34.
- John Naylor landed preparation as cfa6cd2927 from #34, squashing the parser cleanup and moving a shortened range-reference comment into that commit; the GB18030-2022 mapping update remained pending #36.
- Chao Li resent the remaining upgrade, and John Naylor finalized its explanation and regression-test wording; Chao approved that version #37 #38 #39.
- John Naylor proposed a future release-note query to detect affected text, and Chao Li offered to prepare it before release-note work began, so compatibility guidance was deferred without blocking the commit #40 #41.
- The GB18030-2022 update landed with the exact patch diff from #38 as 5334620eef, after which John Naylor invited the EUC_CN follow-up needed to remove the XML input #42.
- Chao Li started that follow-up by converting both EUC_CN and UHC to UCM inputs and reporting unchanged generated maps, then added the XML deletion in a second submission #44 #45.
- John Naylor asked to share the existing 2022 input directly, and Chao Li updated EUC_CN accordingly while omitting the large XML deletion from the attachment and asking the committer to perform it #46 #47.
- John Naylor clarified that UHC used a different input and was unrelated to removing the GB18030 XML file, so Chao Li dropped the UHC changes and submitted an EUC_CN-only revision #48 #50 #51.
- John Naylor committed the EUC_CN revision as 48566180ef, correcting a filename in its comment and performing the requested XML deletion, then marked the Commitfest entry committed #51 #52.
- The thread ended with the GB18030 work and EUC_CN cleanup complete; Chao Li revisited the omitted UHC conversion and a separate Makefile rebuild issue, but John Naylor saw no maintenance benefit in changing UHC and left the build issue to its existing separate thread #53 #54.