fix(status): clamp heatmap keyboard focus when the matrix shrinks

[kylebernhardy]

Fixes #1443 — HeatmapMatrix keyboard focus becomes unreachable when the matrix shrinks.

The roving-tabindex active state in HeatmapMatrix was never clamped when rows/cols shrank after a data refresh — if the stored active cell was out of range, no gridcell got tabIndex=0 and the grid dropped out of the tab order.

Change

• Derive the effective active cell at render time (Math.min(active[0], rows.length - 1), same for cols) — no effect, no extra render. Keyboard handlers stay consistent automatically since they receive the rendered (clamped) indices, and focusing the clamped cell re-syncs state via onFocus.
• Regression test: moves focus to the last row/col via keyboard, re-renders with a 1×2 matrix, asserts exactly one gridcell keeps tabIndex=0 (clamped to the last surviving row/col) and arrow keys still navigate from it. Fails on stage without the fix.

Cross-model review: Codex and Gemini both reviewed the diff with no findings.

Generated by kAIle (Claude Fable 5).

[gemini-code-assist]

Code Review

This pull request fixes a regression (#1443) where shrinking the heatmap matrix could cause the active cell in the roving tabindex state to go out of bounds, leaving the grid without any cell having tabIndex=0 and removing it from the tab order. This is resolved by clamping the active row and column indices to the current matrix dimensions during render. A corresponding unit test has been added to verify this behavior. I have no feedback to provide.

[github-actions]

Coverage Report

Status	Category	Percentage	Covered / Total
🔵	Lines	47.51%	4487 / 9444
🔵	Statements	47.85%	4785 / 10000
🔵	Functions	39.45%	1068 / 2707
🔵	Branches	39.65%	2897 / 7306

File Coverage

File	Stmts	Branches	Functions	Lines	Uncovered Lines
Changed Files
src/features/instance/status/analytics/primitives/HeatmapMatrix.tsx	88.99%	72.91%	93.33%	92.06%	242, 251, 252, 273, 275-281, 287, 328-329, 331-332, 336-341, 343, 353

Generated in workflow #1443 for commit c79b995 by the Vitest Coverage Report Action

[kriszyp]

Nice, clean a11y fix. The render-time clamp is correct: Math.min(active[0], rows.length - 1) caps at the true last index (no off-by-one), passes through unchanged on grid growth, and the empty-grid -1 is provably inert since the clamped indices are only ever compared (never used to index an array), an empty grid renders no gridcell to match, and focusCell guards its ref lookup. Deriving isActive from the clamped values guarantees exactly one surviving cell keeps tabIndex=0, so keyboard users can always re-enter the grid.

The regression test genuinely exercises the shrink path — it moves focus to the corner [1,2], re-renders as 1x2, and asserts exactly one tabbable cell that it's the clamped last-surviving cell (src-1 -> dest-b), plus that arrow-nav still works from it. Verifies direction, not just non-crash.

One non-blocking suggestion: Math.max(0, Math.min(...)) would normalize the empty-grid case to 0 and make the derived indices unconditionally valid array positions — reads more obviously safe to a future editor. Purely optional; the current -1 is safe as written.

LGTM.

— Claude (Sonnet 5), reviewed via review-queue