Rotate/Pan/Zoom

shawn, i would *strongly* advise you to look at the current CVS, and in particular, the "transformer" mechanism under samples/newgraphdraw.

we're still working out details of how the new graph draw system will work, but it is custom-designed for that sort of operation.

trying to do this in the classic GraphDraw is -- to say the least -- a challenge.

that said, i'm happy to have you work on the old system if you like. i'm not sure i understand, though, what problem you are running into. can you walk me through it in a little more detail?

please feel free to email me

danyel

Danyel,

I am in a position where switching to the newgraphdraw may be possible so I will start learning what I can about it.

Thanks. Shawn

danyel,

I have switched over to the newgraphdraw examples found in the samples. The most noticable difference that I see is that there is VisVertex and VisEdge functionality. This has worked sufficiently, except for one detail which I am having trouble figuring out how to resolve (mostly because there are many different was to do it, but none that I like).
I allow the user to "offset" visVertices through the graphLayout  panel, keeping track of the transformations that take place. I am currently using a Circle layout for no other reason than it is simple and static. My problem is that if I use EmittedLayout to grab the nearest Vertex to a mouse click (this is after a translation/offset of coordinate points) the visVertexMap in EmittedLayout has not been updated with the new coordinate, and thus getNearestVertex returns the wrong vertex.
I understand that I can update the EmittedLayout visVertexMap manually, but then I am doubling up on  my work, having to keep track of updating the visVertex in 2 locations (EmittedLayout and Coordinates). What am I missing here? Is there a way that I can use the Coordinate offset, setX, and setY through visVertex, without having to manually update the layout?
Also, part of my problem is that I have to get into the JUNG source code to figure out what's going on. The layouts are especially confusing. Do you expect to have documentation on this new stuff anytime soon?

Thanks in advance, Shawn

That makes sense. One more piece of info and I think I'll be in good shape. I have been fighting resizeLayouts() in GraphLayoutPanel, and eventually extended GLP to override it. I think my lack of understanding of the pipeline was the issue. I also think my previous problem with visVertexMap is tied to my problem with resizeLayouts(). I think I was updating the VisVertex Coordinates too far down the pike. That is, I had extended VisVertexRenderer and performed the transformations there. This appears incorrect, right? Now I see that there are only 4 lines in resizeLayouts(), but I'm a little confused on what it is doing. It appears that it is always modifying the ORIGINAL layout, and not the LAST layout. Is that true, or have I misunderstood it? Also, I am including zoom operations so that I user that has a lot of data that may be grouped together can focus in on a cluster of vertices. resizeLayouts() was always returning all vertices to the screen. Is it possible that, after a transformation, only a portion of the "domain" can be shown.

Thanks, Shawn

Ok, it's working now. Now that I understand a little more about the layouts, I like them much better than what I was trying to implement. One note for future developers. While the custom Offset transformation operations worked, the graph would always "reset" to a centered location at the beginning of the next drag operation. There were 2 reasons for this:
1. I was using LocalGraphDraw, which defines in it's constructor FitOnScreenTransformer, which is then always called when the screen is updated via resizeLayouts().
2. cleanOriginal must be updated with mLayout in GraphLayoutPanel (i.e. getGraphLayoutPanel().setLayoutDisplay(getGraphLayoutPanel().getGraphLayout())
otherwise the layout will always restart from the original layout settings. But setLayoutDisplay(...) calls resizeLayouts() which sends the layout through another transformation, so the transformation should be set to zero before this call, or the whole class can be extended and portions overwritten to do the same.

JUNG folks, thanks so much for the help.
Shawn

Danyel,

On the second point, I have a mouseListener on GraphLayoutPanel which looks like this,

class PanMouseListener extends MouseInputAdapter
{
    double lastx,lasty;

        public void mousePressed(MouseEvent e)
        {
        lastx = e.getX();
        lasty = e.getY();
    }

    public void mouseReleased(MouseEvent e)
        {
        panLayoutTransformer.setOffset(0,0);
        getGraphLayoutPanel().setLayoutDisplay(getGraphLayoutPanel().getGraphLayout());
    }

        public void mouseDragged(MouseEvent e)
        {
            double deltax = e.getX() - lastx;
        double deltay = e.getY() - lasty;
        panLayoutTransformer.setOffset(deltax,deltay);
        getGraphLayoutPanel().resizeLayouts();
    }
    }

The mousePressed method grabs the first x,y position. Next, mouseDragged calculates the change in position, notifies the transformer of the change, and kicks off resizeLayouts to iterate through the layout transformations. But resizeLayouts copies the "cleanOriginal" layout to perform the transformation on. This is fine if I only want to do one pan operation. But If I want to do more pans, then I am always starting from the original layout (cleanOriginal) on the next mousePressed operation, not the layout I just got done transforming, i.e. "mLayout". To fix this I include the mouseReleased method. When I finish my pan operation and release the mouse, my "cleanOriginal" becomes the last layout produced as a result of the transformation. Notice in mouseReleased that I set the offset to zero, because setLayoutDisplay calls resizeLayouts, which then goes through another transformation.

I understand that I could have extended and overwritten this, but I am trying to work with the base API as much as possible to reduce the amount of code in my own package.

The only other code that I think you might want is the custom LayoutTransformer that you suggested above. Methods and parameters are shown below.

    private double xOffset=0; //global offset parameter
    private double yOffset=0;

    private VisVertex singleOffsetVertex = null;
    private double singleOffsetx = 0;
    private double singleOffsety = 0;

    public void setOffset( double x, double y ) {
    this.xOffset = x;
    this.yOffset = y;
    }

    public void setVertexOffset( VisVertex vv, double x, double y ) {
    this.singleOffsetVertex = vv;
    this.singleOffsetx = x;
    this.singleOffsety = y;
    }

    public EmittedLayout transform( EmittedLayout el ) {

    /*There are two forms of offsets: 1. individual node offsets,
    and 2. global offsets for the pan operation. */
   
    //1. Once a vertex has been offset, null it
    if( singleOffsetVertex != null)
    {
       singleOffsetVertex.offset(singleOffsetx,singleOffsety);
       singleOffsetVertex = null;
    }

    //2.
    if(xOffset != 0 && yOffset != 0)
    {
       for (Iterator iter = el.visVertexMap.values().iterator(); iter
                .hasNext();)
       {
           VisVertex vv = (VisVertex) iter.next();
        vv.offset(xOffset,yOffset);
       }
    }
   
    return el;
    }

I am interested in 2 possibilities, 1. a global pan operation of all nodes, or a pan operation on a single node (I don't have a need for the hash map because I only want to transform a single node or all nodes, not any combination in between, yet). The pan operation on all nodes works fine. This is what I was referring to in my last e-mail. But the single node transformation does not. Now I've only been working on it for an hour or so, but it's nearly identical to the global pan operation so I can't see why it doesn't work. The transformations (offsets) are taking place and vertices seem to shift a bit, but they never leave the circle (I'm using CirlceLayout to initialize the vertices). Any idea what's happening here? Anyway, no pressure, I still have to put some more time into it.

That's all for now. I'll be back at it tomorrow morning.
Shawn

It's a great job you guys have done. And I'd like to know when these new features will be released?  Thx

TC