Hi Mark,

Mark Claassen wrote:
Thanks for the reply.

Ok, so here is our theory:
1) Using the SVG view box and the scale factor, figure how many tiles
(at say 100 X 100) will fit into the image.
2) Create a JSVGComponent (workingComp) and set the preferred size to be
the size of our tiles (100 X 100)
3) Set the rendering transform of workingComp to have the scale factor
we want

The point of my message was that if you zoom the canvas with a static document it will automatically tile the image for you. Try it, bring up an image with a really complex background (try samples/batikBatik.svg) Zoom in a bunch Pan right notice it takes a second or two to update, Pan back left notice it repaints immediately.

    The biggest problem with the current scheme is that the tile
cache is extrememly limited in size it keeps ~50 128x128 tiles
in memory (around 4MB - which is like one screenful of data these
days).  This can be adjusted by making calls to:

org.apache.batik.ext.awt.image.rendered.TileCache.setSize(in sz);

    The cache actually does use SoftReferences to try and reclaim
tiles that age out of the cache but these are generally pretty quickly
reclaimed.  The number you provide is the number of tiles it will
hold hard references to.

---
In our paint method for our JComponent (myComp), we will paint the tiles
that are visible in our scrollpane.

For each visible tile
        4) translate the rendering transform so that the part of the
tile we want is on the workingComp
        5) Draw this to a buffered image and keep it.
        6) repaint this tile on myComp


Does this sound OK? Are there any obvious pitfalls or performance issues that might make this not a practical solution?

The current system goes to great lengths to rendering large 'patches' of the image at once and then split the data out into tiles after the fact, this is significantly more efficent then rendering each individual tile.

-----Original Message-----
From: Thomas DeWeese [mailto:[EMAIL PROTECTED]
Sent: Friday, August 08, 2003 6:29 PM
To: Batik Users
Subject: Re: Lazy rendering and zoom


Mark Claassen wrote:


Thanks for the reply.

As it turns out, we are using a static document already. We went
ahead and actually set the state to be static, but we still get the memory problem if we zoom enough. (It seems that we are

now able to


zoom a bit more, although this may just be a coincidence.)

From your


description, we thought it was futile to override the
createImageRenderer method as this point.

It seems to me that if the tiles are rendered lazily, we could zoom
indefinitely and use a relatively constant amount of

memory. Also, we


have never noticed the document being painted as we scroll.  If the
tiles truley were rendered lazily, I would except to see this.

Maybe we are just doing something wrong. We weren't quite

sure how to


get the zoom we wanted using the API directly, so we resize

the canvas


to zoom. To zoom 2X, we:

Dimension d = canvas.getSize();
d.width = d.width * 2;
d.height = d. height * 2;
canvas.setPreferredSize(d)
canvas.revalidate();

Is there a better way?

Yes change the rendering transform, take a look at
Zoom(In|Out)?Action in JSVGCanvas.java you are actually doubling the size of the 'visible' canvas so it creates an offscreen that size.






-----Original Message-----
From: Thomas DeWeese [mailto:[EMAIL PROTECTED]
Sent: Friday, August 08, 2003 12:10 PM
To: Batik Users
Subject: Re: Lazy rendering and zoom


Mark Claassen wrote:



I noticed there is a setProgressivePaint(boolean) method in JGVTComponent. This seems to display items on the canvas

as they are



rendered, and not just display the whole document at once.

What I was wondering was if there is an option that goes a step further, and would restrict rendering to just viewable

area. This is



a common thing in Swing, where, for instance, the rows of a

table that



are not currently visible in a JScrollPane are never rendered.

The default zooming behaviour in Batik seems to render a portion of


the document while not changing the canvas size. We would like to change the canvas size, so that the we can pan to see any

part of the



document we need simply by moving the scrollbars.

Unfortunately, this



causes OutOfMemory errors when the zoom factor gets large.

I know we can increase the JVM heap size, but this can only

go so far.



What I was hoping for was something in the JSVGCanvas that

would draw



just a portion of the document on the canvas, and then when

a scroll



occurred, it would noticed that this area had not been rendered and


draw that. One could even go so far as to anticipate the

loading of



sections and use SortReferences to hold now longer visible

sections.


Is there anything like this, or any plans to include something like


this?

Hi Mark,


This is already done for Static Documents. The Static Renderer


keeps a tiled view of the image and when translates happen it tries to populate the new view from the stored tiles. This is generally not done for dynamic documents as maintaining the tile store is expensive for many small changes (the system would widens rendering requests to tile
boundries) also there were a few bugs in what I will call odd
cases (the size/location of the root SVG changing). However depending on your context this still might be a win for your case. It is pretty easy to change this by overiding the following method in batik.swing.svg.JSVGComponent:


   protected ImageRenderer createImageRenderer() {
       if (isDynamicDocument) {
           return rendererFactory.createDynamicImageRenderer();
       } else {
           return rendererFactory.createStaticImageRenderer();
       }
   }

So it always returns a static ImageRenderer.




---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]



Reply via email to