fix(ios): improve VoiceOver reading order for ColumnSet elements

[hggzm]

Fork-based PR (no Azure Pipelines CI). When ready for CI, push branch to upstream as users/hggzm/clean/fix-columnset-reading-order-42 and recreate.

History: upstream PRs #622 (closed)

Summary

VoiceOver was traversing ColumnSet content column-by-column instead of row-by-row. For example, in a flight passenger list with (Name, Seat) columns, VoiceOver would read all names first, then all seat numbers, rather than reading each passenger's name followed by their seat.

This happened because accessibilityElements was set to the column views themselves (via getArrangedSubviews), causing VoiceOver to drill into each column sequentially.

Changes

• ACRColumnSetRenderer.mm: Instead of setting accessibility elements to the column views, flatten all children from all columns into a single array. VoiceOver then uses geometric frame ordering on these flat elements, naturally producing left-to-right, top-to-bottom (row-by-row) reading order.

Before

VoiceOver reads: "Sarah Hum" → "Jeremy Goldberg" → "Evan Litvak" → "14A" → "14B" → "14C"

After

VoiceOver reads: "Sarah Hum" → "14A" → "Jeremy Goldberg" → "14B" → "Evan Litvak" → "14C"

Testing

1. Open a card with a ColumnSet that has multiple columns with corresponding rows (e.g. FlightUpdate.json with passenger names and seat numbers)
2. Enable VoiceOver
3. Swipe through the ColumnSet content
4. Verify elements are read row-by-row (left-to-right across columns at each level) rather than column-by-column

[hggzm]

VoiceOver Accessibility Validation

Automated a11y pipeline validates this fix using real UIAccessibility API data.

Annotated A11y Overlay

• 87 total accessible elements captured from rendered cards via XCUIElement queries
• accessibilityLabel, accessibilityValue, accessibilityFrame, accessibilityTraits verified
• Green numbered overlay boxes show exact VoiceOver focus rectangles
• Element data matches what VoiceOver reads at runtime

Full transcript | Gallery