<Swing Dev> How to resolve image tearing

[Artem Ananiev]

On 5/26/2011 10:16 PM, ximalaya wrote:
> Hi Anthony,
> Thanks for your replies in last month. I still still have some questions
> and need your help.
>
> 1. Method paint of Canvas(hw component) is not called when its container
> JPanel repaints
> I have a JFrame. Inside the JFrame, there is a JPanel. And inside the
> JPanel there is a Canvas. I found Canvas.paint was never called when I
> call JPanel.repaint(). Is it normal? Is it just another case we simply
> can't sync up with repainting of lw and hw components?
>
> 2. Thread safe of Componet.repaint//JComponent.repaint()/
> I ever read some articles about JComponet.repaint. It was said that it
> was thread safe. But I learn from an article that JComponet.repaint is
> no longer thread safe since JDK 6 update 22 -
> "/do not call JComponent.repaint() outside the Swing Event Dispatch
> Thread" ,
> / You can find the article here, http://stevensrmiller.com/wordpress/?p=567

It's the first time I've heard about it. Looks like a terrible Swing 
bug, doesn't it? Unfortunately, there is no information about why 
exactly repaint() is no longer thread-safe, so I'm very skeptic about 
this post. Let other Swing team members comment.

Thanks,

Artem

> Does it apply to Componet.repaint also?
>
> Thanks and Best Regards,
> Xmly
> On 4/18/2011 8:51 PM, ximalaya wrote:
>  > Sorry, I don't quite understand why we simply can't sync up with
>  > repainting of lw and hw components. Could you please give me more
>  > information?
>
> We can't determine when a third-party hw component repaints itself.
> Therefore we can't simply paint non-opaque lw components onto the
> surface of the hw component.
>
> The only way to implement mixing for non-opaque lw components, as we see
> it currently, is to turn them into hw components and recalculate their
> shapes each time the lw components paints anything onto itself.
> Alternatively, we could leave lw component being lw, and instead
> calculate the shape of a "ho! le" to cut off of underneath components.
> Either way, the performance of such solutions leaves much to be desired
> because lw components may repaint themselves very frequently.
>
> Note that currently mixing is implemented by cutting off hw components
> based on shapes of lw components (i.e. the second option above). And
> since we support opaque lw components only, their shape is always
> rectangular, which doesn't hurt the performance of the hw/lw mixing feature.
>
> In your case, you update the content of your own hw canvas yourself.
> Hence, as a final step you can simply paint your text for each frame of
> your graphics right onto the canvas, and (baring the vertical sync
> issues) this should provide reliable results.
>
>
>  > For the difference between openJDK 1.6.0_18 and latest openJDK, please
>  > pay attention that, openJDK was used in our test. I am not sure the code
>  > base of openJDK 1.6.0_18 is same as JDK 1.6.0 u18, per! haps you can have
>  > a try with openJDK?
>
> I'm still unable to identify any change that could break the hw/lw
> mixing in 6u18. Could you please provide more details of what exact
> issue you're experiencing? A small test application that demonstrates
> the issue would be very helpful. Thanks.
>
>
>