ContextQMD
Back to Plate

Slate browser selectionchange proof must separate traceability from programmatic import

docs/solutions/test-failures/2026-04-22-slate-browser-selectionchange-proof-must-separate-traceability-from-programmatic-import.md

53.0.63.3 KB
Original Source

Slate browser selectionchange proof must separate traceability from programmatic import

Problem

A Batch 1 closure row tried to prove two different contracts at once: selectionchange traceability and pure programmatic DOM-selection import. That made the test red for the wrong reason.

Symptoms

  • selectDOMRange(...) placed the visible DOM caret at This is editable offset 4.
  • getSelectionWithHandle(...) kept returning Slate selection [0, 0]@0.
  • Kernel traces contained allowed selectionchange events, but they reflected the existing selection rather than the synthetic DOM range.
  • Mobile Playwright keyboard transport also mutated visible DOM differently than model selection, so semantic-handle transport remained the honest mobile proof.

What Didn't Work

  • Waiting only for the assertion did not fix the red row because the synthetic selectionchange could land inside the runtime's throttled selectionchange window.
  • Dispatching mousedown manually did not reliably model the full user selection authority transition.
  • A temporary runtime DOM-selection converter patch was rejected because it did not close the proof and would have been unverified patch debt.

Solution

Split the proof:

  • Assert selectionchange is emitted as an allowed kernel trace.
  • Use the existing semantic selection handle for deterministic setup before the repair/text mutation step.
  • Assert repair trace, model text, visible DOM text, model selection, and desktop DOM caret after the model-owned text path.
  • Keep mobile on semantic-handle text transport unless the row explicitly claims native mobile keyboard transport.
ts
await expect
  .poll(async () =>
    (await harness.get.kernelTrace()).some(
      entry => (entry as { eventFamily?: string }).eventFamily === 'selectionchange'
    )
  )
  .toBe(true)

await selectWithHandle(harness.root, {
  anchor: { path: [0, 0], offset: 4 },
  focus: { path: [0, 0], offset: 4 },
})

Why This Works

Selectionchange traceability and programmatic DOM-selection import are separate contracts. Batch 1 needed traceable kernel results and repair correctness. Programmatic import needs its own explicit API/proof in the next batch because synthetic selectionchange timing is not equivalent to a real user selection.

Prevention

  • Do not claim raw DOM-selection import from a test that uses synthetic selectionchange unless the import API itself is the contract.
  • For browser editing closure, label mobile semantic-handle transport explicitly instead of pretending it proves native keyboard transport.
  • If a test asserts a kernel event family, also assert illegal transitions stay empty.
  • Keep failed proof attempts in the architecture ledger so future work does not repeat the same false claim.