Previous Behavior

Popover had no focus trap. The surface rendered as a <div popover="auto"> and relied entirely on the browser's light-dismiss (Escape, click-outside, popover-stack peer dismissal).

New Behavior

Adds a trapFocus prop on Popover that delegates modal behavior. The surface always renders as a single element; the show mode is wired in usePopover:

• trapFocus={false} (default) — surface.showPopover(). Browser owns light dismiss; no focus trap or autofocus. Explicit role="group" overrides the dialog's implicit role="dialog" so assistive tech
  doesn't announce modal semantics that aren't present.
• trapFocus={true} — surface.showModal(). The platform supplies the focus trap, autofocus, trigger-restoration on close, and inert background — all spec-mandated by . The explicit role is
  dropped so the implicit role="dialog" + aria-modal="true" apply. The cancel event is intercepted (preventDefault + close via onOpenChange) so Escape flows through the same React state path as any other
  dismiss.

A single element supporting both modes means consumers flip one prop instead of swapping component variants, and there's no portal/stacking-context divergence between the two paths.

IMPORTANT NOTE:

There one notable behavioral difference compared to a custom focus trap implementation: By default with a native dialog, all elements within the same document, except the dialog and its descendants become inert, but it is scoped only to the containing document. If the dialog is rendered inside an iframe, the rest of the parent page remains interactive (focusable), which differs to v9 behaviour

Example: https://dialog-focus-trap-test.surge.sh/