Dock Tile Imaging using JDirect
Volume Number: 19 (2003)
Issue Number: 1
Column Tag: Java
Dock Tile Imaging using JDirect
A pure Java approach to changing a Java application's dock tile at runtime
by Andrew S. Downs
Overview
A previous article ("Dock Tile Imaging", MacTech June 2002) discussed using the Java Native Interface technology to dynamically change the dock tile for a Java application. This article illustrates an alternate approach: the use of Apple's JDirect technology to make the native API calls directly from Java without writing any C code. This eliminates the need for a shared library and the scaffolding to call functions in that library.
The application discussed in this article periodically fetches a new image from NASA's Kennedy Space Center video feed web page. The source page contains links to 15 channels (images). The application parses the source page to find the individual image links, then sequentially downloads each image. In addition to displaying each retrieved image, the application paints a countdown progress bar in the tile between fetches. This progress bar painting uses the QuickDraw API, called via JDirect.
Application startup
The JDirectDockTiler class contains main(), the entry point for the Java application. main() is used simply to instantiate the DockTiler class and invoke an instance method. DockTiler creates an instance of UrlDownloader, which fetches an image from a URL. UrlDownloader invokes a method in DockTiler, passing it the image bytes. Some of the code in this class is reused from the June 2002 article, but is not discussed again. Download the accompanying source files for this article, refer to the previous article, or send me email if you need more info.
The import statements include two Macintosh Runtime for Java (MRJ) packages, one of which is JDirect. The other package, macos.frameworks, contains declarations for the Carbon and ApplicationServices interfaces.
import com.apple.mrj.jdirect.*;
import com.apple.mrj.macos.frameworks.*;
public class JDirectDockTiler {
public static void main (String args[]) {
DockTiler dock = new DockTiler();
dock.run();
}
}
class DockTiler {
If we have trouble loading an image, the values of its width and height will remain -1. See loadImage().
int mWidth = -1, mHeight = -1;
The pixels are stored as a 1-dimensional array of integers. Three bytes of the int contain the RGB values, while the fourth byte contains the alpha (transparency) value.
int mPixels[];
public DockTiler() {
}
Instantiate a class that handles periodic image fetching and display.
public void run() {
UrlDownloader u = new UrlDownloader( this );
u.setUrl(
"http://spaceflight.nasa.gov/realdata/ksclive/index.html"
);
This substring will be located in the fetched page. This is a potential problem if the source page content changes, since then our substring search may stop working. This is a good candidate for a property in an external file.
u.setSearchString( "http://science.ksc.nasa.gov" );
The number of iterations indicates how many times the substring occurs in the page. This is another potential trouble spot that could result in something breaking.
u.setIterations( 15 );
Fortunately, the Thread sleep interval is entirely within our control. The value 5000 shown here indicates the Thread should sleep for 5 seconds between invocations.
u.setSleep( 5000 );
u.start();
}
static Frame f = null;
static PictureFrame pf = null;
This version of loadImage() differs from the JNI version in that it uses an array of bytes to create an image. Also, the grabPixels() completion checking is slightly more complicated in order to prevent partial or blank images from occurring.
protected void loadImage( byte [] array ) {
Image image =
Toolkit.getDefaultToolkit().createImage( array );
if ( f != null && pf != null )
f.remove( pf );
if ( f == null )
f = new Frame( "Image" );
// Offscreen windows move onscreen when you change the dock settings.
// Keeping it onscreen allows you to view the original image as well
// as the dock tile.
f.setBounds( 50, 50, 200, 200 ); // onscreen
// f.setBounds( -250, -250, 200, 200 ); // offscreen
f.setLayout( new BorderLayout() );
pf = new PictureFrame( image );
f.add( "Center", pf );
pf.setSize( 128, 128 );
f.pack();
f.setVisible( true );
f.repaint();
pf.repaint();
int width = image.getWidth( pf );
int height = image.getHeight( pf );
if ( width < 0 || height < 0 )
return;
mPixels = new int[ width * height ];
for ( int i = 0; i < width * height; i++ )
mPixels[ i ] = 0;
PixelGrabber pg = new PixelGrabber( image, 0, 0, width,
height, mPixels, 0, width );
try {
pg.grabPixels();
while ( ( pg.getStatus() & ImageObserver.ALLBITS )
== 0 ) {
if ( ( pg.getStatus() & ImageObserver.ABORT )
!= 0 )
return;
Thread.sleep( 3000 );
}
mWidth = width;
mHeight = height;
}
catch ( InterruptedException e ) {
return;
}
if ( mWidth > 0 && mHeight > 0 )
ImageUtils.setDockTile( mPixels, mWidth, mHeight );
}
}
Retrieving images
There are undoubtedly third party, or perhaps Sun-provided, classes that can fetch and parse HTML pages, but the one discussed here contains the necessary functionality for this app.
The methods in this class:
- fetch a page at a given URL
- locate image links with a page
- fetch an image at a specified URL
public class UrlDownloader {
The following attributes have corresponding accessors (not shown):
String mUrl = "";
int mSleep = 0;
String searchString = "";
static int mCount = 0;
This corresponds to the number of iterations.
static int max = 1;
This attribute is set from the constructor, but could also be done with an accessor:
DockTiler tiler;
UrlDownloader( DockTiler dt ) {
tiler = dt;
}
public void start() {
downloadPage();
Timer is the actual Thread, and is discussed further down.
Timer t = new Timer( this, mSleep );
t.run();
}
This buffer holds the raw downloaded page content.
StringBuffer sb = new StringBuffer();
This method uses a URL string to open a connection and download the content of that page.
protected void downloadPage() {
try {
URL url = new URL( mUrl );
URLConnection conn = url.openConnection();
int length = conn.getContentLength();
BufferedInputStream is =
new BufferedInputStream( url.openStream() );
int total = 0;
int numRead = 1;
Rather than read the page as one huge chunk of data, read 1k pieces and append them to the buffer.
while ( numRead > 0 && total < length ) {
byte array[] = new byte[ 1024 ];
numRead = is.read( array, 0, 1024 );
total += numRead;
String s = new String( array );
sb.append( s );
}
}
catch ( MalformedURLException e ) {
System.out.println( e );
}
catch ( IOException e ) {
System.out.println( e );
}
}
This method gets called by the timer Thread. It dispatches to methods that locate the next occurrence of the search string on the page, fetch the appropriate image, then call back to DockTiler to display to image.
protected void fetch() {
String s = parseUrl();
byte array[] = fetchImage( s );
tiler.loadImage( array );
}
protected String parseUrl() {
String url = new String();
Update the current image number. Reset after reaching the total number of image links on the page.
mCount++;
if ( mCount > max )
mCount = 1;
int occurence = 0;
int last = 0;
String page = sb.toString();
int index = 0, prevIndex = 0;
Find the beginning of the next image link on the page.
while ( occurence < mCount ) {
index = page.indexOf( searchString, prevIndex );
prevIndex = index + 1;
occurence++;
}
Find the end of the link: a quote character.
if ( index > 0 )
url = page.substring( index,
page.indexOf( "\"", index + 1 ) );
return url;
}
protected byte[] fetchImage( String imageUrl ) {
Return a 1-byte array in lieu of an error code in case of failure.
byte array[] = new byte[ 1 ];
if ( !imageUrl.startsWith( "http" ) ) {
System.out.println( "Bad URL: " + imageUrl );
return array;
}
try {
This is similar to how we fetched the page contents above, except that readFully() eliminates the loop.
URL url = new URL( imageUrl );
URLConnection conn = url.openConnection();
int length = conn.getContentLength();
array = new byte[ length ];
DataInputStream is =
new DataInputStream( url.openStream() );
int total = 0;
int numRead = 1;
is.readFully( array );
}
catch ( MalformedURLException e ) {
System.out.println( "Bad URL: " + imageUrl );
array = new byte[ 1 ];
}
catch ( IOException e ) {
System.out.println( e );
}
catch ( NegativeArraySizeException e ) {
System.out.println( "Bad array size: " + e );
array = new byte[ 1 ];
}
return array;
}
}
This simple timer class calls the fetch() dispatch function, then sleeps for 1/100th of the specified Thread sleep time. In between, it updates the progress bar at the bottom of the dock tile image. Upon reaching 100% of the sleep interval, start another fetch cycle. To remove the progress bar updates, uncomment the specified Thread.sleep() call and remove the while {} loop.
class Timer implements Runnable {
UrlDownloader mUrlDownloader = null;
int mSleep = 5000;
Timer( UrlDownloader u, int sleep ) {
mUrlDownloader = u;
mSleep = sleep;
}
public void run() {
while ( true ) {
try {
mUrlDownloader.fetch();
// Replace following block with this line to avoid drawing of progress bar.
// Thread.sleep( mSleep );
int total = 0;
int i = 0;
float interval = mSleep / 100;
while ( i <= 100 ) {
ImageUtils.updateProgressBar( ( int )i );
Thread.sleep( ( int )interval );
i++;
}
}
catch ( InterruptedException e ) {
}
}
}
}
Calling Quartz
The ImageUtils class provides Java access to various native Mac OS API calls, such as the Quartz 2D (Core Graphics) library. The imports before the class definition include the JDirect package (containing the Linker class that is used to register the native methods) and the macos.frameworks package (which defines the Carbon and ApplicationServices interfaces):
import com.apple.mrj.jdirect.*;
import com.apple.mrj.macos.frameworks.*;
public class ImageUtils implements Carbon, ApplicationServices {
The Linker constructor call registers the native methods declared in this class as JNI methods:
private static final Object linkage =
new Linker( ImageUtils.class );
The Mac OS calls must be declared explicitly. Argument names are not important here. I used letters as placeholders, but if you intend to reuse such a class it may make sense to provide more descriptive names and possibly comments or documentation so you don't have to revisit the API documentation each time. These two Core Graphics declarations illustrate native methods without and with arguments:
public static native int
BeginCGContextForApplicationDockTile();
public static native int CGDataProviderCreateWithData(
int a, int [] b, int c, int d );
The non-native method setDockTile() draws an image using the Core Graphics API. The arguments include an array of integers (4-bytes) representing the RGBA values for each pixel, and the image width and height.
static protected void setDockTile( int [] imagePixels,
int width, int height ) {
This value defines the number of bytes in each pixel:
int kNumComponents = 4;
This is somewhat annoying, having to redefine the constant values already defined by the API:
final int kCGImageAlphaFirst = 4;
final int kCGRenderingIntentDefault = 0;
If you look at the API docs, you will notice that many of the values used or returned by the native calls are pointers. In JDirect (as in JNI), pointers map to the int data type. This call gets the graphics context for the current application's dock tile:
int theContext =
BeginCGContextForApplicationDockTile();
if ( theContext != 0 ) {
The number of bytes comprising one row of the image may be calculated using the number of pixels per row and the number of bytes per pixel:
int bytesPerRow = width * kNumComponents;
Create a data source using the pixel array passed in as the image source. Then create an RGB color space object.
int theProvider = CGDataProviderCreateWithData( 0,
imagePixels, ( bytesPerRow * height ), 0 );
int theColorspace = CGColorSpaceCreateDeviceRGB();
Now create the image. This is similar to creating a PixMap. The symbol kCGImageAlphaFirst specifies that in the pixel data, the alpha channel (transparency) value is in the first byte of each int. More information can be found at: http://developer.apple.com/techpubs/macosx/CoreTechnologies/graphics/Quartz2D/Quartz_2D_Ref/index.html.
int theImage = CGImageCreate( width, height, 8, 32,
bytesPerRow, theColorspace,
kCGImageAlphaFirst, theProvider, 0, 0,
kCGRenderingIntentDefault );
Cleanup by explicitly releasing pointers to native objects. We could have left this until the end and cleaned up all resources at the same time.
CGDataProviderRelease( theProvider );
CGColorSpaceRelease( theColorspace );
Use the newly created image as the tile image:
int theError =
SetApplicationDockTileImage( theImage );
Flush to ensure the pixels get written:
CGContextFlush( theContext );
Finally, free the native image and the drawing context:
CGImageRelease( theImage );
EndCGContextForApplicationDockTile( theContext );
}
}
Calling QuickDraw
This portion of the ImageUtils class is used to paint a progress bar over bottom of the image in the dock tile while the application sleeps prior to fetching a new image.
The QuickDraw Toolbox declarations look the same as the Core Graphics method declarations just discussed. However, since QuickDraw uses non-opaque data structures in various places this set of declarations includes as arguments, for example, an array of shorts to represent a native Rect.
public static native int
BeginQDContextForApplicationDockTile();
public static native int GetPortBounds( int port,
short [] rect );
public static native void SetRect( short [] r,
short left, short top, short right, short bottom );
Several color constants defined in the Toolbox must be defined here as well before they can be used:
static final int blackColor = 33;
static final int redColor = 205;
static final int greenColor = 341;
The progress bar is a 10-pixel tall rectangle superimposed on the lower-edge of the dock tile. The background of the progress bar is red, with a black frame. Each update fills a larger percentage of the bar with green.
static protected void updateProgressBar(
int currPercent ) {
First obtain a reference to the tile drawing surface:
int thePort = BeginQDContextForApplicationDockTile();
if ( thePort != 0 ) {
Next, create an instance of a Rect class (discussed in the next section). That rectangle will hold the boundaries of the drawing surface.
Rect theRect = new Rect();
GetPortBounds( thePort, theRect.getArray() );
The initial call simply frames the empty red rectangle:
if ( currPercent == 0 ) {
SetRect( theRect.getArray(), theRect.getLeft(),
( short )( theRect.getBottom() - ( short )10 ),
theRect.getRight(), theRect.getBottom() );
ForeColor( redColor );
PaintRect( theRect.getArray() );
ForeColor( blackColor );
FrameRect( theRect.getArray() );
}
The right edge of the bar must be calculated using the current percent complete and the width of the bar in pixels. The right edge is treated as a float because of the fractions encountered during the calculation.
float right = 0;
if ( currPercent >= 100 )
right = ( float )theRect.getRight();
else
right = ( ( ( float )theRect.getRight() -
( float )theRect.getLeft() ) /
( float )100 ) * ( float )currPercent;
Fill the "percent complete" area with green. The right edge value calculated above gets cast to a short data type and used to define the rectangle to be painted:
ForeColor( greenColor );
SetRect( theRect.getArray(),
( short )( theRect.getLeft() + 1 ),
( short )( theRect.getBottom() - 9 ),
( short )right,
( short )( theRect.getBottom() - 1 ) );
PaintRect( theRect.getArray() );
Flush the bits to the screen and cleanup:
QDFlushPortBuffer( thePort, 0 );
EndQDContextForApplicationDockTile( thePort );
}
}
}
Rect Support Class
JDirect relies on various Toolbox structures that exist in native Mac OS libraries, but need to be explicitly defined for use with JDirect. Here is one way to define a class that corresponds to the Rect data structure:
public class Rect {
private short [] theRect = new short[ 4 ];
The four corners of the Rect are stored in an array whose sequence we arbitrarily define as:
theRect[ 0 ] = Left
theRect[ 1 ] = Top
theRect[ 2 ] = Right
theRect[ 3 ] = Bottom
The storage assignments are arbitrary because access to the actual values is provided through accessor methods only, such as:
public short getLeft() {
return theRect[ 0 ];
}
public void setLeft( short i ) {
theRect[ 0 ] = i;
}
The constructor initializes the fields. The explicit casts to the short (16-bit) data type convert each 0 value, which defaults to type int (32-bit).
public Rect() {
setLeft( ( short )0 );
setTop( ( short )0 );
setRight( ( short )0 );
setBottom( ( short )0 );
}
A more compact though implementation of this class could eliminate the accessor methods and allow direct access to the data values, perhaps as four short values. That exposure may lead to problems should you wish to change the internal data representation. Always try to hide the internals of a class and enforce access through an API.
A more complex get/set combination that hides the internal representation can be implemented using constant values to represent each corner, as in:
static final int kLeft = 0, kTop = 1, kRight = 2,
kBottom = 3;
public short get( short corner ) {
if ( corner >= kLeft && corner <= kRight )
return theRect[ corner ];
else
// Not good because 0 is a valid value for any corner.
return 0;
}
public void set( short corner, short i ) {
if ( corner >= kLeft && corner <= kRight )
theRect[ corner ] = i;
// What if the caller passes a bad corner identifier?
}
Although the variations just discussed have their merits, the Rect implementation for this project uses explicit get/set methods for each corner of the Rect. The class definition is slightly longer, but more forgiving, than an implementation requiring the caller to specify a numeric value for a corner as an additional argument to both get/set. Also, this implementation does not expose the internal data representation.
References
The book Early Adopter Mac OS X Java (Wrox Press, Inc., November 2001) contains an excellent chapter on integrating Java with native Mac OS calls using JDirect. TechNote 2002 (http://developer.apple.com/technotes/tn/tn2002.html) also discusses this issue.
Thank you to Eric Albert, Greg Parker, and Nicholas Riley for their assistance and advice while I wrote this program at MacHack.
Andrew has been a Java fan since his first encounter with the language at the WebEdge III conference in 1996. You can reach him at andrew@downs.ws.