fix(display): end the scroll before the schedule-off blank and WiFi notice - #721

Why

#716 ends the scroll state at a scroller-to-static handover, but only for plugin screens: it hooks _dispatch_first_frame. The display controller draws two screens itself, so they never got it:

• The schedule-off blank. It was drawn while the last scroll's state was still set. On a scan-compensated panel it went out with the ticker's lagging rows (rows 24-47 on ledpi) and stayed up for its 60 s dwell. fix(display): schedule windows end at the end time; on-demand ending in off hours blanks at once #714 made this path direct for an on-demand ticker that ends in off hours.
• The WiFi notice. fix(display): a WiFi notice preempts the current screen and Vegas yields to it #712 now lets it preempt a running scroller or Vegas. Its first frame showed the ticker's rows, and its 0.5 s redraws were timed as 0.5-1 s freezes of the old scroll, with a Render stall ... mid-scroll logged.

Both were found by the adversarial review of #716 and listed in docs/SCROLL_PERFORMANCE.md as a follow-up.

What

A new helper, DisplayController._end_scroll_before_core_screen(), calls display_manager.set_scrolling_state(False) (guarded with getattr like #716's helpers). It runs:

• before _blank_while_scheduled_off clears the panel;
• before _show_wifi_notice draws a pending notice. It does nothing when there is no notice.

With the state ended, _scan_segments sends the frame out as drawn and drops the history, the hold goes back to 1, and the frame is timed as static. A scroller or Vegas that resumes after the notice sets the state again on its next frame, as every scroller already does.

docs/SCROLL_PERFORMANCE.md drops the follow-up note. The one screen still composed after a scroll is an older static-image forced into the high-FPS loop. There is a CHANGELOG entry under Fixes.

Tests

test/test_handover_scroll_state.py::TestTheControllersOwnScreens uses a real emulator DisplayManager with ledpi's lag band. It checks that:

• the schedule-off blank shows none of the ticker, and is timed as a static frame;
• the WiFi notice shows none of the ticker, and is timed as a static frame;
• with no notice pending, the scroll and its hold are left alone.

Mutation check: with the helper made a no-op, the first two fail (they see the ticker's red rows). With it, all pass.

Related suites pass: 419 tests, covering the run loop goldens, WiFi preemption, live takeover, controller, on-demand, pending changes, the exception breaker, dirty tracking, scan order and frame timing.

Not soaked: neither screen comes up in a normal rotation soak.

🤖 Generated with Claude Code