fix(spatialHandleNavigation): don't reset selection at grid edges

[chiefcll]

What

spatialHandleNavigation reset the container's selected to -1 whenever a cross-axis move failed at an edge, which made the next key press jump focus back to the first child.

Why

In the cross-axis branch (up/down in a row-wrapped container, left/right in a column-wrapped one), the handler called selectChild(el, closestIdx) unconditionally — including when the scan found nothing and left closestIdx === -1. selectChild treats an invalid index as "clear the selection":

if (child == null || child.skipFocus) {
  el.selected = -1;
  return false;
}

So a no-op edge move silently wiped selected. On the next press, the guard at the top of the handler saw selected === -1, treated it as "nothing selected", and fell back to findFirstFocusableChildIdx → the first child.

Reported symptoms, both reproduced: holding Down on the bottom row bounces focus to the first item, and holding Up from a first-row item does the same.

The in-flex-direction branch already returned false without touching selected, which is why only cross-axis moves were affected.

The change

Return false when no adjacent row/column exists, before selectChild can clobber the selection:

// No child in an adjacent column/row - keep the current selection so the
// next key press resumes from here instead of resetting to the first child.
if (closestIdx === -1) return false;

Returning false also lets the key bubble to a parent container, which is the desired behavior at a boundary.

Notes for reviewers

I deliberately kept the guard local to spatialHandleNavigation rather than removing the el.selected = -1 reset inside selectChild. That reset is still correct for navigableForwardFocus and spatialForwardFocus, which call it for the genuinely-no-focusable-children case.

Tests

New tests/spatialNavigation.test.tsx builds a 3×2 wrapped flex grid (0 1 2 / 3 4) and drives it through the real focus manager with dispatched keydown events — it asserts the flex layout first, then that repeated Down on the bottom row and repeated Up on the top row hold their selection and still navigate correctly afterward.

• Confirmed the new tests fail on main (expected -1 to equal 3, expected -1 to equal 1) and pass with the fix.
• Full suite: 174 passed / 17 files.
• npm run tsc clean; npm run lint 0 errors (pre-existing warnings unchanged).

One fixture note if these tests get extended: spatialForwardFocus selects the child closest to the previously active element, so rendering a fresh grid per test inherits the prior grid's position. The suite renders once in beforeAll and sets focus explicitly per case.

🤖 Generated with Claude Code