model-writtenrun 6214 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Pavel Stehule reported that psql counted a two-column smiley as one column, misaligning query output #1; Laurenz Albe supplied an inline range extension #2, which Pavel confirmed fixed his example #3.
- Kyotaro Horiguchi questioned how emoji properties map to terminal widths #4, while Jacob Champion raised presentation selectors and configurable widths #6; Michael Paquier and Laurenz also questioned the performance and complexity of a broader solution #8 #10.
- Jacob showed that the guessed emoji range both missed wide characters and incorrectly widened narrow ones, and proposed updating the long-stale East Asian width handling from Unicode data; country-flag combinations and zero-width-joiner sequences remained outside this work #11.
- Jacob submitted a generated Unicode 13.0 wide/fullwidth table, integrated its regeneration into update-unicode, and added a second interval search after the combining-character search #13.
- Pavel confirmed the fix and measured small absolute performance costs #14; Jacob supplied additional profiling and microbenchmarks #15, and Pavel subsequently reported successful check-world testing and marked the patch ready for a committer #16.
- John Naylor initially saw no improvement, but a clean rebuild made the fix work #17 #20; Jacob traced the stale result to missing libpq dependencies on common libraries and identified that as an independent build issue #22.
- John proposed clearer return logic and removal of outdated width comments #21, and Jacob accepted the return-style change #22.
- Discussion of table size exposed the combining generator's BMP-only cutoff #23 #24 #25; Peter Eisentraut recalled no substantive reason for it beyond matching the old manual table #28.
- John proposed merging the generators and tables to use one binary search #27, then abandoned an additional ASCII fast path after Jacob explained the existing shortcut and caught its incorrect condition #29 #30.
- John posted four patches covering renaming, explicit interval widths, the merged-table emoji fix, and combining characters beyond the BMP; he also raised unresolved handling of U+200B ZERO WIDTH SPACE #30.
- Jacob approved the revised series and its benchmark results, identified a mixed-declarations warning, and requested an overlap sanity check #32; John fixed the warning and landed the rename as eb0d0d2c73 and the range-return change as 78ab944cd4, keeping them separate for history readability #33.
- The requested check revealed that official Unicode data legitimately marks some characters as both combining and wide, making the merged interval table ambiguous; John chose to restore separate searches rather than depend on property annotations in data-file comments #34.
- John reverted the range-return prerequisite with f8c8a8bccc and the rename with 1563ecbc1b, implementing the reversal proposed in #34.
- John then landed Jacob's original two-search patch #13 as bab982161e, adding the earlier style cleanup and a comment documenting combining-character precedence, and announced the successful landing #35.
- The thread ended with Jacob acknowledging the missed overlaps and Pavel thanking the contributors #36 #37; the main fix had landed, while John had announced the beyond-BMP follow-up for later that day #35.
- That follow-up subsequently landed as 5bc429aacb, preserving the fourth patch's cutoff removal and intervals from #30 while adapting them to the separate combining table; the independent build-dependency issue and U+200B question have no resolution recorded in this thread #22 #30.