My original question:

I'm creating a simple drawing application and need to be able to draw over existing, previously drawn content in my drawRect. What is the proper way to draw on top of existing content without entirely replacing it?

Based on answers received here and elsewhere, here is the deal.

1. You should be prepared to redraw the entire rectangle whenever drawRect is called.

2. You cannot prevent the contents from being erased by doing the following:

   [self setClearsContextBeforeDrawing: NO];

   This is merely a hint to the graphics engine that there is no point in having it pre-clear the view for you, since you will likely need to re-draw the whole area anyway. It may prevent your view from being automatically erased, but you cannot depend on it.

3. To draw on top of your view without erasing, do your drawing to an off-screen bitmap context (which is never cleared by the system.) Then in your drawRect, copy from this off-screen buffer to the view.

Example:

- (id) initWithCoder: (NSCoder*) coder {    
     if (self = [super initWithCoder: coder]) {
         self.backgroundColor = [UIColor clearColor];
         CGSize size = self.frame.size;
         drawingContext = [self createDrawingBufferContext: size];
     }

     return self;
 }

- (CGContextRef) createOffscreenContext: (CGSize) size  {
    CGColorSpaceRef colorSpace = CGColorSpaceCreateDeviceRGB();
    CGContextRef context = CGBitmapContextCreate(NULL, size.width, size.height, 8, size.width*4, colorSpace, kCGImageAlphaPremultipliedLast);
    CGColorSpaceRelease(colorSpace);

    CGContextTranslateCTM(context, 0, size.height);
    CGContextScaleCTM(context, 1.0, -1.0);
    return context;
}


- (void)drawRect:(CGRect) rect {    
    UIGraphicsPushContext(drawingContext);
    CGImageRef cgImage = CGBitmapContextCreateImage(drawingContext); 
    UIImage *uiImage = [[UIImage alloc] initWithCGImage:cgImage];
    UIGraphicsPopContext();
    CGImageRelease(cgImage);
    [uiImage drawInRect: rect];
    [uiImage release];
 }

TODO: can anyone optimize the drawRect so that only the (usually tiny) modified rectangle region is used for the copy?