Why are the D3D12 samples using WM_PAINT?

Question

If you're trying to program a game, you really shouldn't wait around for WM_PAINT signals from the O/S to draw. You should repeatedly draw as quickly as possible.

Why then in the D3D sample code is rendering happening in WM_PAINT?

How can I rework the D3D12 sample to render as quickly as possible, without waiting for WM_PAINT signals from the O/S?

Answer 1

The "Hello, World" DirectX 12 sample on GitHub: DirectX-Graphics-Samples is relying on the fact that if you call PeekMessage and there is no other message available, it creates one WM_PAINT message:

MSG msg = {};
while (msg.message != WM_QUIT)
{
    // Process any messages in the queue.
    if (PeekMessage(&msg, NULL, 0, 0, PM_REMOVE))
    {
        TranslateMessage(&msg);
        DispatchMessage(&msg);
    }
}

See Microsoft Docs.

Typically game render loops don't rely on the "magic WM_PAINT" behavior, and that what I have in GitHub: directx-vs-templates:

MSG msg = {};
while (WM_QUIT != msg.message)
{
    if (PeekMessage(&msg, nullptr, 0, 0, PM_REMOVE))
    {
        TranslateMessage(&msg);
        DispatchMessage(&msg);
    }
    else
    {
        g_game->Tick();
    }
}

g_game.reset();

and the only thing WM_PAINT does is clear the screen--with special handling for resizing:

case WM_PAINT:
    if (s_in_sizemove && game)
    {
        game->Tick();
    }
    else
    {
        PAINTSTRUCT ps;
        (void)BeginPaint(hWnd, &ps);
        EndPaint(hWnd, &ps);
    }
    break;

Here Tick is a method which does the Update and Render, but depending on the mode of StepTimer it may call Update more than once per frame.

In the end, both Win32 'message pumps' achieve basically the same thing.