Root cause: all accessibility line methods used count_lines (logical/
newline-counting), but VoiceOver compares line numbers before/after
cursor movement. Wrapped lines had same logical line number, so VO
fell back to reading single characters instead of full lines.
Fix: replaced count_lines/find_newline_no_quit with compute_motion/
vmotion (visual screen lines) in all three methods. Matches iTerm2's
approach of returning screen rows.
Changes from v11:
- accessibilityInsertionPointLineNumber: compute_motion visual lines
- accessibilityLineForIndex: compute_motion visual lines
- accessibilityRangeForLine: vmotion for line start/end
- SelectedRowsChanged only on visual line change (new ivar tracking)
- accessibilityLabel returns buffer name for window switch
- No double ValueChanged on edits (content_changed flag)
- Trailing newline stripped from line ranges
- compute_motion width=-1 for consistency with vmotion
3 pipeline iterations, scores: 78 → 92 → 95
Daneel
2c78f04089