Sparkles

Appearance

Manual-run queue ​

The single ordered checklist of every Tier C item from the windowing demo matrix — grouped by machine/session so each environment is visited once. Each entry carries: demo path, build command, steps, the expected observation, and where to paste results (the demo's findings doc). Agents append entries here; Petar checks them off.

Last reviewed: June 11, 2026

NOTE

On the Mac, build with nix develop -c ldc2 … (not dub — it fork-ENOMEMs there); the verified command pattern is in the AppKit scaffold notes. [wine]-flagged Win32 items are re-confirmations of behavior already observed under Wine.

Windows box ​

  • [ ] Win32 F03 on real Windows: run win32/examples/f03-modal-loop (build per win32/scaffold.md) without WSI_AUTO_EXIT, with and without WSI_MODAL_FIX=1; drag titlebar + corner ~2 s each; expect freeze vs ~15.6 ms src=timer ticks. Also Alt+Space → S → arrows to check whether Wine's keyboard-mode early-exit reproduces. Paste the modal_summary lines into win32/f03-modal-loop.md.

  • [ ] Win32 F04 on real Windows: run win32/examples/f04-frame-pacing with WSI_AUTO_EXIT=1 (expect path=dwm, true vblank stats) and again with WSI_FORCE_TIMER=1; note monitor Hz and whether minimize changes the DwmFlush cadence. Paste both stats blocks into win32/f04-frame-pacing.md.

  • [ ] Win32 F05 on real Windows: re-measure the wakeup-latency table and the waitable-timer tick distribution at 142 ms period (Wine's ±1 ms is tighter than Windows' default 15.6 ms timer-resolution quantization). Paste into win32/f05-loop-wakeup.md.

  • [ ] Win32 F06 on real Windows: LoadKeyboardLayoutW("00000407") with real KBDGR tables — capture the Y/Z swap on injected scan 0x15/0x2c and the dead-acute (scan 0x0d) + E sequence (WM_DEADCHAR 0x00b4 then WM_CHAR 0x00e9; the demo logs it as-is). Also confirm the VK_PACKET scancode field shows the UTF-16 unit's low byte. Paste into win32/f06-keyboard.md.

  • [ ] Win32 F07 with Microsoft Pinyin: run win32/examples/f07-text-input without WSI_AUTO_EXIT and execute the 6-step script in win32/f07-text-input.md. Record: composition keydowns swallowed as VK_PROCESSKEY?, the cancel shape (zero-flag WM_IME_COMPOSITION vs Wine's bare ENDCOMPOSITION), focus-loss pre-edit fate, candidate-window position vs the logged candidate_anchor x=. Paste into win32/f07-text-input.md.

  • [ ] Win32 F08 on mixed-DPI monitors: run win32/examples/f08-dpi-scaling without WSI_AUTO_EXIT, drag between 100% and 150%/200% monitors. Record the live dpi_changed sequence + ordering vs WM_GETDPISCALEDSIZE/WM_SIZE, hairline sharpness after the suggested-rect move, and whether the late SetProcessDpiAwarenessContext fails ERROR_ACCESS_DENIED (succeeds-but-noops under Wine). Paste into win32/f08-dpi-scaling.md.

  • [ ] Win32 F09 on multi-monitor Windows: run win32/examples/f09-outputs with WSI_AUTO_EXIT=1 WSI_RUN_MS=30000; hot-plug (or disable/enable in Display settings) a monitor mid-run. Expect output_added/removed with via=WM_DISPLAYCHANGE (vs Wine's via=poll), a WM_SETTINGCHANGE burst, WM_DEVICECHANGE arrivals, real per-monitor rcMonitor offsets, and true dmDisplayFrequency. Paste into win32/f09-outputs.md.

  • [ ] Win32 F12 on real Windows: run win32/examples/f12-cursors; visually confirm the 3×3 zone cursors, the custom bullseye + hotspot; confirm the precedence_result phase=set_then_def cursor_after=IDC_CROSS reproduces; expect ani_probe non-null (C:\Windows\Cursors ships .ani); check cursor_metrics under large-cursor accessibility / mixed DPI. Paste into win32/f12-cursors.md.

  • [ ] Win32 F10 on real Windows: rerun win32/examples/f10-pointer-capture; expect WM_INPUT to report injected deltas verbatim while WM_MOUSEMOVE shows post-ballistics motion (Wine accelerated BOTH ×2). Does absolute SendInput produce MOUSE_MOVE_ABSOLUTE raw events? Does Alt-Tab to another process auto-clear ClipCursor? Paste into win32/f10-pointer-capture.md.

  • [ ] Win32 F11 on a precision touchpad: two-finger scroll + fling; confirm sub-120 wheelDelta stream and the inertia tail as decaying wheel messages (WSI_TRUNCATE=1 should visibly freeze the ruler); probe SPI_GETMOUSEWHEELROUTING (errored 1439 under Wine) and toggle "scroll inactive windows". Paste into win32/f11-scroll.md.

  • [ ] Win32 F14 on real Windows: keys M/N/R/F/D in win32/examples/f14-window-state; confirm a single WM_SIZE echo per transition (winewayland added a late compositor configure), where focus goes on minimize with other apps open, the fullscreen normal-rect corruption, and the ✕-click veto with D armed (src=external). Paste into win32/f14-window-state.md.

  • [ ] Win32 F15 on real Windows: right-click menu tour in win32/examples/f15-popup; confirm the parent survives submenu open, outside-click on ANOTHER process's window dismisses (headless Wine can't prove cross-process capture), edge placement matches the adjust= log. Paste into win32/f15-popup.md.

  • [ ] Win32 F16 on real Windows: re-run f16-clipboard-dnd.exe — delayed-render demand latencies, WM_CLIPBOARDUPDATE delivery, clipboard persistence after owner exit (Windows runs a session clipboard, Wine does not), and a real file-manager drag onto the window. Paste into win32/f16-clipboard-dnd.md.

  • [ ] Win32 F17 on real Windows: re-run f17-threading.exe — confirm probe 4a's mutual-send resolution and 4b's deadlock against real user32, plus probe 5's GDI behavior under DWM. Paste into win32/f17-threading.md.

Mac (mac-bsn, unlocked GUI session) ​

  • [ ] One-time: unlock the screen during a demo run so CGWindowListCopyWindowInfo reports onscreen=true, and grant Screen Recording TCC to screencapture for visual capture. Paste results into the AppKit scaffold findings.
  • [ ] AppKit F02 interactive live-resize: build docs/research/window-system-integration/os-apis/appkit/examples/f02-resize/ per the scaffold notes (nix develop -c ldc2 …, not dub), run ./demo with no env vars in an unlocked GUI session, drag the window border for a few seconds, then close the window. Expected on stderr: live_resize_start → repeated resize/frame_callback pairs at drag cadence → live_resize_end (all three are absent under the programmatic storm). Paste the sequence into appkit/f02-resize.md.
  • [ ] AppKit F03 interactive live-resize/title-drag: build …/appkit/examples/f03-modal-loop/ per the scaffold notes (binary also staged at /tmp/wsi-m2/f03-modal-loop/demo on the mac). Run ./demo with no env vars; drag the border, then the title bar, ~3 s each. Expect: color cycle never freezes (common-modes timer), modal_enter/modal_exit bracket the drag, the tick timer=default lines vanish during it with a post-drag gap_ms ≈ drag length, draw … live_resize=1. Paste into appkit/f03-modal-loop.md.
  • [ ] AppKit F04 display-link cadence on a real display: run …/appkit/examples/f04-frame-pacing/ (/tmp/wsi-m2/f04-frame-pacing/demo) in an unlocked GUI session. Expect path=displaylink at ~8.33 ms (120 Hz), fell_back=0; record whether fires pause during the 3 s orderOut: window. Paste into appkit/f04-frame-pacing.md.
  • [ ] AppKit F07 real Pinyin IME: add the Pinyin – Simplified input source, run …/appkit/examples/f07-text-input/ (/tmp/wsi-m4/f07-text-input/demo) with no env vars, click the window, type nihao + Space, x + Esc, then ni + focus switch — full keystroke script in appkit/f07-text-input.md § Tier C. Expect per-keystroke tic_set_marked_text (attributed string class?), system-driventic_first_rect re-queries anchoring the candidate window, tic_insert_text text=你好 on Space, and the Esc/focus-loss fate of the pre-edit under a real IME. Also re-do option-e + e from the physical keyboard on the US layout. Paste into appkit/f07-text-input.md.
  • [ ] AppKit F08 monitor drag across scales: attach a 1× external display, run …/appkit/examples/f08-dpi-scaling/ (/tmp/wsi-m4/f08-dpi-scaling/demo) with no env vars in an unlocked session. First read ctm when=drawRect on the Retina screen (expect a=2.00 d=-2.00, falsifying the headless identity-CTM artifact), then drag to the 1× display and back — expect window_backing_changed old_scale=…, backing_changed n=2/3, a CTM re-log, and buffer_alloc tracking the pixel size; note the crossing point and hairline crispness. Also flip the built-in "looks like" resolution and record which notification fires. Full script in appkit/f08-dpi-scaling.md § Tier C. Paste into appkit/f08-dpi-scaling.md.

GNOME session (mutter) ​

  • [ ] Wayland F03 — interactive drag continuity (also run in the KDE and sway sections below): dub run --root=docs/research/window-system-integration/os-apis/wayland/examples/f03-modal-loop (no WSI_AUTO_EXIT). Drag the window by its title bar ≥5 s, then drag an edge to resize ≥5 s. Expect: uninterrupted ~2 Hz color cycling, the tick … dt_us=… stream bounded near the display's frame period throughout, resizing=1 in the configure lines during the edge drag, modal_enter=0 modal_exit=0 in the exit summary. Any tick gap far beyond one frame period during the drag falsifies the no-modal-loop finding. Paste into wayland/f03-modal-loop.md.
  • [ ] Wayland F06 on a real seat (any Wayland session): re-run wayland/examples/f06-keyboard with a physical keyboard — live layout switch via compositor config (virtual keyboards carry their own keymap, so the headless run could not observe it), AltGr/ISO_Level3 symbols, and per-device keymap behavior. Paste into wayland/f06-keyboard.md.
  • [ ] Wayland F07 candidate-window visuals (GNOME + kimpanel, or KDE): run wayland/examples/f07-text-input on the real session with fcitx5+pinyin active, type nihao + space at both ends of the line, Esc mid-composition, then switch focus away mid-composition. Expect the doc's logged choreography plus a candidate window anchored at the caret rect. Paste into wayland/f07-text-input.md.
  • [ ] X11 F03 interactive border/title drag (any real X11/XWayland session): build with nix develop -c dub build --root=docs/research/window-system-integration/os-apis/x11/examples/f03-modal-loop, run …/examples/f03-modal-loop/build/f03_modal_loop_x11 with no env vars. Border-drag-resize ~5 s, then title-bar-drag ~5 s; any key exits. Expected: the color cycle never freezes; tick … gap_us=… stays ~16 ms (nothing above ~50 ms) through both drags. Paste the max observed gap into x11/f03-modal-loop.md (findings table, "Interactive border/title drag" row).
  • [ ] X11 F05 under a real compositing X server + load: re-run x11/examples/f05-loop-wakeup and compare the latency distributions against the Xvfb numbers (no vsync/compositor contention there; the 808 µs ClientMessage max likely grows). Paste into x11/f05-loop-wakeup.md.
  • [ ] X11 F06 on real de hardware: AltGr/ISO_Level3 symbols, multi-layout group switching (grp: options, nonzero group= field), and whether real input drivers emit per-device XkbNewKeyboardNotify storms like Xvfb's 3x. Paste into x11/f06-keyboard.md.

KDE session (kwin) ​

(no entries yet)

sway session ​

(no entries yet)