Java - rounded corner panel with compositing in pa

Question

From the original question (below), I am now offering a bounty for the following:

An AlphaComposite based solution for rounded corners.

• Please demonstrate with a JPanel.
• Corners must be completely transparent.
• Must be able to support JPG painting, but still have rounded corners
• Must not use setClip (or any clipping)
• Must have decent performance

Hopefully someone picks this up quick, it seems easy.

I will also award the bounty if there is a well-explained reason why this can never be done, that others agree with.

Here is a sample image of what I have in mind (but usingAlphaComposite)

---

Original question

I've been trying to figure out a way to do rounded corners using compositing, very similar to How to make a rounded corner image in Java or http://weblogs.java.net/blog/campbell/archive/2006/07/java_2d_tricker.html.

However, my attempts without an intermediate BufferedImage don't work - the rounded destination composite apparently doesn't affect the source. I've tried different things but nothing works. Should be getting a rounded red rectangle, instead I'm getting a square one.

So, I have two questions, really:

1) Is there a way to make this work?

2) Will an intermediate image actually generate better performance?

SSCCE:

the test panel TPanel

import java.awt.AlphaComposite;
import java.awt.Color;
import java.awt.Dimension;
import java.awt.Graphics;
import java.awt.Graphics2D;

import javax.swing.JLabel;

public class TPanel extends JLabel {
int w = 300;
int h = 200;

public TPanel() {
    setOpaque(false);
    setPreferredSize(new Dimension(w, h));
        setMaximumSize(new Dimension(w, h));
        setMinimumSize(new Dimension(w, h));
}

@Override
public void paintComponent(Graphics g) {
    super.paintComponent(g);
    Graphics2D g2d = (Graphics2D) g.create();

    // Yellow is the clipped area.
    g2d.setColor(Color.yellow);
    g2d.fillRoundRect(0, 0, w, h, 20, 20);
    g2d.setComposite(AlphaComposite.Src);

    // Red simulates the image.
    g2d.setColor(Color.red);
    g2d.setComposite(AlphaComposite.SrcAtop);

    g2d.fillRect(0, 0, w, h);
    }
}

and its Sandbox

import java.awt.Dimension;
import java.awt.FlowLayout;

import javax.swing.JFrame;

public class Sandbox {
public static void main(String[] args) {
    JFrame f = new JFrame();
        f.setMinimumSize(new Dimension(800, 600));
        f.setLocationRelativeTo(null);
        f.setDefaultCloseOperation(JFrame.EXIT_ON_CLOSE);
        f.setLayout(new FlowLayout());

        TPanel pnl = new TPanel();
        f.getContentPane().add(pnl);

        f.setVisible(true);
    }
}

Answer 1

With respect to your performance concerns the Java 2D Trickery article contains a link to a very good explanation by Chet Haase on the usage of Intermediate Images.

I think the following excerpt from O'Reilly's Java Foundation Classes in a Nutshell could be helpful to you in order to make sense of the AlphaComposite behaviour and why intermediate images may be the necessary technique to use.

The AlphaComposite Compositing Rules

The SRC_OVER compositing rule draws a possibly translucent source color over the destination color. This is what we typically want to happen when we perform a graphics operation. But the AlphaComposite object actually allows colors to be combined according to seven other rules as well.

Before we consider the compositing rules in detail, there is an important point you need to understand. Colors displayed on the screen never have an alpha channel. If you can see a color, it is an opaque color. The precise color value may have been chosen based on a transparency calculation, but, once that color is chosen, the color resides in the memory of a video card somewhere and does not have an alpha value associated with it. In other words, with on-screen drawing, destination pixels always have alpha values of 1.0.

The situation is different when you are drawing into an off-screen image, however. As you'll see when we consider the Java 2D BufferedImage class later in this chapter, you can specify the desired color representation when you create an off-screen image. By default, a BufferedImage object represents an image as an array of RGB colors, but you can also create an image that is an array of ARGB colors. Such an image has alpha values associated with it, and when you draw into the images, the alpha values remain associated with the pixels you draw.

This distinction between on-screen and off-screen drawing is important because some of the compositing rules perform compositing based on the alpha values of the destination pixels, rather than the alpha values of the source pixels. With on-screen drawing, the destination pixels are always opaque (with alpha values of 1.0), but with off-screen drawing, this need not be the case. Thus, some of the compositing rules only are useful when you are drawing into off-screen images that have an alpha channel.

To overgeneralize a bit, we can say that when you are drawing on-screen, you typically stick with the default SRC_OVER compositing rule, use opaque colors, and vary the alpha value used by the AlphaComposite object. When working with off-screen images that have alpha channels, however, you can make use of other compositing rules. In this case, you typically use translucent colors and translucent images and an AlphaComposite object with an alpha value of 1.0.

Answer 2

I have looked into this issue and cannot see how to do this in a single call to system classes.

Graphics2D is an abstract instance, implemented as SunGraphics2D. The source code is available at for example docjar, and so we could potentially just 'do the same, but different' by copy some code. However, methods to paint an image depends on some 'pipe' classes which are not available. Although you do stuff with classloading, you will probably hit some native, optimized class which cannot be manipulated to do the theoretically optimal approach; all you get is image painting as squares.

However we can make an approach in which our own, non-native (read:slow?), code runs as little as possible, and not depending on the image size but rather the (relatively) low area in the round rect. Also, without copying the images in memory, consuming a lot of memory. But if you have a lot of memory, then obviously, a cached image is faster after the instance has been created.

Alternative 1:

import java.awt.Composite;
import java.awt.CompositeContext;
import java.awt.Dimension;
import java.awt.Graphics;
import java.awt.Graphics2D;
import java.awt.RenderingHints;
import java.awt.image.BufferedImage;
import java.awt.image.ColorModel;
import java.awt.image.Raster;
import java.awt.image.WritableRaster;

import javax.swing.JLabel;

public class TPanel2 extends JLabel implements Composite, CompositeContext {
private int w = 300;
private int h = 200;

private int cornerRadius = 20;
private int[] roundRect; // first quadrant
private BufferedImage image;
private int[][] first = new int[cornerRadius][];
private int[][] second = new int[cornerRadius][];
private int[][] third = new int[cornerRadius][];
private int[][] forth = new int[cornerRadius][];

public TPanel2() {
    setOpaque(false);
    setPreferredSize(new Dimension(w, h));
    setMaximumSize(new Dimension(w, h));
    setMinimumSize(new Dimension(w, h));

    // calculate round rect     
    roundRect = new int[cornerRadius];
    for(int i = 0; i < roundRect.length; i++) {
        roundRect[i] = (int)(Math.cos(Math.asin(1 - ((double)i)/20))*20); // x for y
    }

    image = new BufferedImage(w, h, BufferedImage.TYPE_INT_ARGB); // all black
}

@Override
public void paintComponent(Graphics g) {
    // discussion:
    // We have to work with the passed Graphics object.

    if(g instanceof Graphics2D) {

        Graphics2D g2d = (Graphics2D) g;

        // draw the whole image and save the corners
        g2d.setComposite(this);
        g2d.drawImage(image, 0, 0, null);
    } else {
        super.paintComponent(g);
    }
}

@Override
public CompositeContext createContext(ColorModel srcColorModel,
        ColorModel dstColorModel, RenderingHints hints) {
    return this;
}

@Override
public void dispose() {

}

@Override
public void compose(Raster src, Raster dstIn, WritableRaster dstOut) {
    // lets assume image pixels >> round rect pixels
    // lets also assume bulk operations are optimized

    // copy current pixels
    for(int i = 0; i < cornerRadius; i++) {
        // quadrants

        // from top to buttom
        // first
        first[i] = (int[]) dstOut.getDataElements(src.getWidth() - (cornerRadius - roundRect[i]), i, cornerRadius - roundRect[i], 1, first[i]);

        // second
        second[i] = (int[]) dstOut.getDataElements(0, i, cornerRadius - roundRect[i], 1, second[i]);

        // from buttom to top
        // third
        third[i] = (int[]) dstOut.getDataElements(0, src.getHeight() - i - 1, cornerRadius - roundRect[i], 1, third[i]);

        // forth
        forth[i] = (int[]) dstOut.getDataElements(src.getWidth() - cornerRadius + roundRect[i], src.getHeight() - i - 1, cornerRadius - roundRect[i], 1, forth[i]);
    }

    // overwrite entire image as a square
    dstOut.setRect(src);

    // copy previous pixels back in corners
    for(int i = 0; i < cornerRadius; i++) {
        // first
        dstOut.setDataElements(src.getWidth() - cornerRadius + roundRect[i], i, first[i].length, 1, second[i]);

        // second
        dstOut.setDataElements(0, i, second[i].length, 1, second[i]);

        // third
        dstOut.setDataElements(0, src.getHeight() - i - 1, third[i].length, 1, third[i]);

        // forth
        dstOut.setDataElements(src.getWidth() - cornerRadius + roundRect[i], src.getHeight() - i - 1, forth[i].length, 1, forth[i]);
    }
}

}

Alternative 2:

import java.awt.AlphaComposite;
import java.awt.Composite;
import java.awt.CompositeContext;
import java.awt.Dimension;
import java.awt.Graphics;
import java.awt.Graphics2D;
import java.awt.RenderingHints;
import java.awt.image.BufferedImage;
import java.awt.image.ColorModel;
import java.awt.image.Raster;
import java.awt.image.WritableRaster;
import javax.swing.JLabel;

public class TPanel extends JLabel implements Composite, CompositeContext {
private int w = 300;
private int h = 200;

private int cornerRadius = 20;
private int[] roundRect; // first quadrant
private BufferedImage image;

private boolean initialized = false;
private int[][] first = new int[cornerRadius][];
private int[][] second = new int[cornerRadius][];
private int[][] third = new int[cornerRadius][];
private int[][] forth = new int[cornerRadius][];

public TPanel() {
    setOpaque(false);
    setPreferredSize(new Dimension(w, h));
    setMaximumSize(new Dimension(w, h));
    setMinimumSize(new Dimension(w, h));

    // calculate round rect     
    roundRect = new int[cornerRadius];
    for(int i = 0; i < roundRect.length; i++) {
        roundRect[i] = (int)(Math.cos(Math.asin(1 - ((double)i)/20))*20); // x for y
    }

    image = new BufferedImage(w, h, BufferedImage.TYPE_INT_ARGB); // all black
}

@Override
public void paintComponent(Graphics g) {
    if(g instanceof Graphics2D) {

        Graphics2D g2d = (Graphics2D) g;

        // draw 1 + 2 rectangles and copy pixels from image. could also be 1 rectangle + 4 edges
        g2d.setComposite(AlphaComposite.Src);

        g2d.drawImage(image, cornerRadius, 0, image.getWidth() - cornerRadius - cornerRadius, image.getHeight(), null);
        g2d.drawImage(image, 0, cornerRadius, cornerRadius, image.getHeight() - cornerRadius - cornerRadius, null);
        g2d.drawImage(image, image.getWidth() - cornerRadius, cornerRadius, image.getWidth(), image.getHeight() - cornerRadius, image.getWidth() - cornerRadius, cornerRadius, image.getWidth(), image.getHeight() - cornerRadius, null);

        // draw the corners using our own logic
        g2d.setComposite(this);

        g2d.drawImage(image, 0, 0, null);

    } else {
        super.paintComponent(g);
    }
}

@Override
public CompositeContext createContext(ColorModel srcColorModel,
        ColorModel dstColorModel, RenderingHints hints) {
    return this;
}

@Override
public void dispose() {

}

@Override
public void compose(Raster src, Raster dstIn, WritableRaster dstOut) {
    // assume only corners need painting

    if(!initialized) {
        // copy image pixels
        for(int i = 0; i < cornerRadius; i++) {
            // quadrants

            // from top to buttom
            // first
            first[i] = (int[]) src.getDataElements(src.getWidth() - cornerRadius, i, roundRect[i], 1, first[i]);

            // second
            second[i] = (int[]) src.getDataElements(cornerRadius - roundRect[i], i, roundRect[i], 1, second[i]);

            // from buttom to top
            // third
            third[i] = (int[]) src.getDataElements(cornerRadius - roundRect[i], src.getHeight() - i - 1, roundRect[i], 1, third[i]);

            // forth
            forth[i] = (int[]) src.getDataElements(src.getWidth() - cornerRadius, src.getHeight() - i - 1, roundRect[i], 1, forth[i]);
        }
        initialized = true;
    }       

    // copy image pixels into corners
    for(int i = 0; i < cornerRadius; i++) {
        // first
        dstOut.setDataElements(src.getWidth() - cornerRadius, i, first[i].length, 1, second[i]);

        // second
        dstOut.setDataElements(cornerRadius - roundRect[i], i, second[i].length, 1, second[i]);

        // third
        dstOut.setDataElements(cornerRadius - roundRect[i], src.getHeight() - i - 1, third[i].length, 1, third[i]);

        // forth
        dstOut.setDataElements(src.getWidth() - cornerRadius, src.getHeight() - i - 1, forth[i].length, 1, forth[i]);
    }
}

}

Hope this helps, this is somewhat of a second-best solution but that's life (this is when some graphics-guru comes and proves me wrong (??)..) ;-)