Win32 backend: can't hover over nonclient areas when window does not have focus

[thundercarrot]

Version: 1.88
Branch: master/viewport/docking

Back-ends: imgui_impl_win32.cpp
Compiler: vs2022
Operating System: Windows 11

imgui buttons that are part of the nonclient area of the window behave correctly only when the window has focus:

When creating a frameless window with owner draw captions you can use imgui buttons for min/max/close buttons, by appropriately processing WM_NCHITTEST on the button regions. This works fine when the window has focus (is foreground). In this case AddMousePosEvent is not called from ImGui_ImplWin32_WndProcHandler's WM_MOUSEMOVE handler since they are considered nonclient areas. Instead, AddMousePosEvent is called from
ImGui_ImplWin32_UpdateMouseData(). However, UpdateMouseData checks first that the window has focus, so that when the window does not have focus AddMousePosEvent is never called and hover tests fail.

ImGui_ImplWin32_WndProcHandler should probably handle WM_NCMOUSEMOVE as well as WM_MOUSEMOVE.

[ocornut]

ImGui_ImplWin32_WndProcHandler should probably handle WM_NCMOUSEMOVE as well as WM_MOUSEMOVE.

It doesn't seem trivial to handle seems how WM_MOUSELEAVE are occasionally triggered after WM_NCMOUSEMOVE.

I came up with this:

Add bool MouseInClientArea; in ImGui_ImplWin32_Data.

    case WM_MOUSEMOVE:
    case WM_NCMOUSEMOVE:
    {
        // We need to call TrackMouseEvent in order to receive WM_MOUSELEAVE events
        bd->MouseHwnd = hwnd;
        bd->MouseInClientArea = (msg == WM_MOUSEMOVE);
        if (!bd->MouseTracked)
        {
            TRACKMOUSEEVENT tme = { sizeof(tme), TME_LEAVE, hwnd, 0 };
            ::TrackMouseEvent(&tme);
            bd->MouseTracked = true;
        }
        POINT pos = { (LONG)GET_X_LPARAM(lParam), (LONG)GET_Y_LPARAM(lParam) };
        if (msg == WM_NCMOUSEMOVE && ::ScreenToClient(hwnd, &pos) == FALSE) // WM_NCMOUSEMOVE are provided in screen coordinates.
            break;
        io.AddMousePosEvent((float)pos.x, (float)pos.y);
        break;
    }
    case WM_MOUSELEAVE:
    case WM_NCMOUSELEAVE:
        if (bd->MouseInClientArea == (msg == WM_MOUSELEAVE))
        {
            if (bd->MouseHwnd == hwnd)
                bd->MouseHwnd = nullptr;
            bd->MouseTracked = false;
            io.AddMousePosEvent(-FLT_MAX, -FLT_MAX);
        }
        break;

I would however need to test it more thoroughly, and with multi-viewports.
Could you test on your side?
That stuff tends to be tricky and fragile so if it comes with other undesirable side-effects we might need to back-track.

[thundercarrot]

In handling WM_NCMOUSEMOVE I think you want TRACKMOUSEEVENT tme = { sizeof(tme), TME_LEAVE | TME_NONCLIENT, hwnd, 0 };

Getting WM_MOUSELEAVE after the first WM_NCMOUSEMOVE is unfortunate but not the end of the world. Otherwise I believe this works. One idea is to cancel the outstanding track mouse event (TME_CANCEL | TME_LEAVE) and issue a new one.

[thundercarrot]

@ocornut your code above works perfectly with the addition of

if (msg == WM_NCMOUSEMOVE) tme.dwFlags |= TME_NONCLIENT;

[ocornut]

your code above works perfectly with the addition of

That didn't look right. When unfocused we don't always get leave events there.

I think I worked it our right always sending a cancel request when changing area.
Pushed 5d74891
Merged in docking via e25e452

cc @BobbyAnguelov would be good if you can confirm it works with vanilla backend and without your custom msg handler.

[BobbyAnguelov]

Works in both the minimal example I created and in Esoterica.