Strategy for creating a layered window with child windows (controls)

Question

I want to create an irregularily shaped/skinned window (just with rounded, alpha blended corners for now). Upon creating the top-level window i am processing the WM_CREATE message and do the following:

1. Create a compatible memory DC
2. Create a DIB section compatible with the window DC
3. Select DIB into memory DC
4. Do the drawing of my backdrop
5. Apply alpha channel and premultiply RGB values
6. Call UpdateLayeredWindow()

Later on I am planning on rounding of the edges by setting the alpha channel and premultiplying the bits in question to make that happen.

Now, if I create for instance a button in my window, it will not render itself. I know why, and I am trying to come up with an elegant solution to making my custom controls (child windows) here.

My initial approach was to ditch using child windows in the first place and just let the top level window do all the drawing and also input handling, hit testing, and so on. I found this to be way to tedious and instead I want to let Windows handle all this for me.

Now, I know if I create a child window, it of course behaves normally (e.g. reacting to user input), and I want to leverage this. I am planning on creating the child windows (custom controls) normally using CreateWindowEx() so they get a window handle, and recieve window messages without me having to worry about passing them manually.

Somehow I need to get these windows painted, and as I see it, the only possible way to do this is from the routine that paints the whole top level window. I need to invent some kind of logic to get the top level window's paint routine to paint my child windows whenever necessary. As far as I understand the UpdateLayeredWindow() function need to redraw the whole window.

Is it sensible to for instance have the child controls render an image of themselves that are sent to the top level window's paint routine? Like for instance the child window sending a user WM to the top level window passing pointer to its rendered bitmap as a WPARAM and pointer to a structure defining its position and size as a LPARAM.

Any better ideas? Does this make any sense at all?

Thanks, Eirik

Answer 1

I was trying to do a very similar thing. From reading this and other searching web. It seams the recommended mechanism for drawing a CWnd (or HWND) and it's children onto your own CDC (or HDC) is to use the printing API.

CWnd has methods Print and PrintClient and which send WM_PRINT correctly. There is also the Win32 methods: PrintWindow.

I had trouble getting this to work at first but I eventually got the right method and parameters. The code that worked for me was:

void Vg2pImageHeaderRibbon::Update() {
    // Get dimensions
    CRect window_rect;
    GetWindowRect(&window_rect);

    // Make mem DC + mem  bitmap
    CDC* screen_dc = GetDC(); // Get DC for the hwnd
    CDC dc;
    dc.CreateCompatibleDC(screen_dc);
    CBitmap dc_buffer;
    dc_buffer.CreateCompatibleBitmap(screen_dc, window_rect.Width(), window_rect.Height());
    auto hBmpOld = dc.SelectObject(dc_buffer);

    // Create a buffer for manipulating the raw bitmap pixels (per-pixel alpha).
    // Used by ClearBackgroundAndPrepForPerPixelTransparency and CorrectPerPixelAlpha.
    BITMAP raw_bitmap;
    dc_buffer.GetBitmap(&raw_bitmap);
    int bytes =  raw_bitmap.bmWidthBytes * raw_bitmap.bmHeight;
    std::unique_ptr<char> bits(new char[bytes]);

    // Clears the background black (I want semi-transparent black background).
    ClearBackgroundAndPrepForPerPixelTransparency(dc, raw_bitmap, bytes, bits.get(), dc_buffer);

    // To get the window and it's children to draw using print command
    Print(&dc, PRF_CLIENT | PRF_CHILDREN | PRF_OWNED);

    CorrectPerPixelAlpha(dc, raw_bitmap, bytes, bits.get(), dc_buffer);

    // Call UpdateLayeredWindow
    BLENDFUNCTION blend = {0};
    blend.BlendOp = AC_SRC_OVER;
    blend.SourceConstantAlpha = 255;
    blend.AlphaFormat = AC_SRC_ALPHA;
    CPoint ptSrc;
    UpdateLayeredWindow(
        screen_dc, 
        &window_rect.TopLeft(), 
        &window_rect.Size(), 
        &dc, 
        &ptSrc, 
        0, 
        &blend, 
        ULW_ALPHA
    );

    SelectObject(dc, hBmpOld);
    DeleteObject(dc_buffer);
    ReleaseDC(screen_dc);
}

This worked for me just as is. But incase you window or children don't support WM_PRINT I looked at how it was implemented for CView class I discovered that this class provides a virtual method called OnDraw(CDC* dc) that is provided with a DC to draw with. WM_PAINT is implemented something like this:

CPaintDC dc(this);
OnDraw(&dc);

And the WM_PAINT is implemented:

CDC* dc = CDC::FromHandle((HDC)wParam);
OnDraw(dc);

So the WM_PAINT and WM_PRINT results an OnDraw(), and the drawing code implemented once.

You can basically add this same logic your own CWnd derived class. This may not be possible using visual studio's class wizards. I had to add the following to message map block:

BEGIN_MESSAGE_MAP(MyButton, CButton)
    ...other messages
    ON_MESSAGE(WM_PRINT, OnPrint)
END_MESSAGE_MAP()

And my handler:

LRESULT MyButton::OnPrint(WPARAM wParam, LPARAM lParam) {
    CDC* dc = CDC::FromHandle((HDC)wParam);
    OnDraw(dc);
    return 0;
}

NOTE: If you add a custom WM_PRINT handler on a class that already supports this automatically then you loose the default implementation. There isn't a CWnd method for OnPrint so you have to use the Default() method to invoke the default handler.

I have't tried the following but I expect it works:

LRESULT MyCWnd::OnPrint(WPARAM wParam, LPARAM lParam) {
    CDC* dc = CDC::FromHandle((HDC)wParam);
    // Do my own drawing using custom drawing
    OnDraw(dc);

    // And get the default handler to draw children
    return Default();
}

Above I defined some strange methods: ClearBackgroundAndPrepForPerPixelTransparency and CorrectPerPixelAlpha. These allow me to set the background of my dialog be semi-transparent when having the child controls be full opaque (this is my per-pixel transparency).

// This method is not very efficient but the CBitmap class doens't
// give me a choice I have to copy all the pixel data out, process it and set it back again.
// For performance I recommend using another bitmap class
//
// This method makes every pixel have an opacity of 255 (fully opaque).
void Vg2pImageHeaderRibbon::ClearBackgroundAndPrepForPerPixelTransparency( 
    CDC& dc, const BITMAP& raw_bitmap, int bytes, char* bits, CBitmap& dc_buffer 
) {
    CRect rect;
    GetClientRect(&rect);
    dc.FillSolidRect(0, 0, rect.Width(), rect.Height(), RGB(0,0,0));
    dc_buffer.GetBitmapBits(bytes, bits);
    UINT* pixels = reinterpret_cast<UINT*>(bits);
    for (int c = 0; c < raw_bitmap.bmWidth * raw_bitmap.bmHeight; c++ ){
        pixels[c] |= 0xff000000;
    }
    dc_buffer.SetBitmapBits(bytes, bits);
}

// This method is not very efficient but the CBitmap class doens't
// give me a choice I have to copy all the pixel data out, process it and set it back again.
// For performance I recommend using another bitmap class
//
// This method modifies the opacity value because we know GDI drawing always sets
// the opacity to 0 we find all pixels that have been drawn on since we called 
// For performance I recommend using another bitmap class such as the IcfMfcRasterImage
// ClearBackgroundAndPrepForPerPixelTransparency. Anything that has been drawn on will get an 
// opacity of 255 and all untouched pixels will get an opacity of 100.
void Vg2pImageHeaderRibbon::CorrectPerPixelAlpha( 
    CDC& dc, const BITMAP& raw_bitmap, int bytes, char* bits, CBitmap& dc_buffer 
) {
    const unsigned char AlphaForBackground = 100; // (0 - 255)
    const int AlphaForBackgroundWord = AlphaForBackground << 24;
    dc_buffer.GetBitmapBits(bytes, bits);
    UINT* pixels = reinterpret_cast<UINT*>(bits);
    for (int c = 0; c < raw_bitmap.bmWidth * raw_bitmap.bmHeight; c++ ){
        if ((pixels[c] & 0xff000000) == 0) {
            pixels[c] |= 0xff000000;
        } else {
            pixels[c] = (pixels[c] & 0x00ffffff) | AlphaForBackgroundWord;
        }
    }
    dc_buffer.SetBitmapBits(bytes, bits);
}

Here is a screen shot of my test application. When the user hovers the mouse over the "more buttons" button the dialog box is created with a semi-transparent background. The buttons "B1" to "B16" are child controls derived from CButton and are being drawn using the Print() call show above. You can see the semi-transparent background at the right hand edge of the view and between the buttons.

Answer 2

To quote the MSDN page for WM_PAINT:

The WM_PAINT message is generated by the system and should not be sent by an application. To force a window to draw into a specific device context, use the WM_PRINT or WM_PRINTCLIENT message. Note that this requires the target window to support the WM_PRINTCLIENT message. Most common controls support the WM_PRINTCLIENT message.

So it looks like you can iterate through all the child windows with EnumChildWindows and send them WM_PRINT with your memory DC between steps 5 and 6.

It will look something like this:

static BOOL CALLBACK MyPaintCallback(HWND hChild,LPARAM lParam) {
    SendMessage(hChild,WM_PRINT,(WPARAM)lParam,(LPARAM)(PRF_CHECKVISIBLE|PRF_CHILDREN|PRF_CLIENT));
    return TRUE;
}

void MyPaintMethod(HWND hWnd) {
    //steps 1-5

    EnumChildWindows(hWnd,MyPaintCallback,MyMemoryDC);

    //step 6
}

Answer 3

I think I'm going to go for this solution:

Top level window

The top level window maintains two bitmaps. One which is the displayed window and one without any of the child controls rendered. The latter one will only need redrawing when the window changes size. The window will have a message handler that renders a child control on the displayed bitmap. The message handler will expect a pointer to either a DIB containing the child control, or to the actual bits (not sure which is best at the moment), as the WPARAM, and a pointer to a structure containing the rectangle that the child shall be drawn into as the LPARAM. A call to BitBlt() will be made to clear out the underlying surface (this is where the other bitmap comes in) prior to an AlphaBlend() call for rendering the child control bitmap onto the displayed bitmap surface.

The parent window will call the EnumChildWindows whenever it is resized or for some reason need to redraw its children. There could of course be some kind of invalidation regime enforced here to reduce unnecessary rendering of the child controls. Not sure if the speed increase is worth the effort, though.

Child windows

Upon creation of the child control instance, an internal bitmap compatible with that of the top-level window is created. The child renders itself into this internal bitmap and whenever it needs redrawing it notifies its parent window via the SendMessage() function, passing a pointer to its bitmap as the WPARAM, and a RECT as the LPARAM defining its position and dimensions. If the parent needs redrawing, it issues a message down to all its child windows requesting their bitmap. Childs will then respond with the same message that they normally would send when they decide they need redrawing themselves.

Eirik