Re: [Chromium-dev] (no subject)

[Brian P. <br...@tu...>]

Joel Welling wrote:
> Hi folks;
>   My student Christiaan reported a problem with the psubmit demo a while ago 
> under the heading "readback/pack SPUs and two psubmit clients".  We've 
> clarified the problem description, but we still can't figure out what's going 
> on.  Has anyone else seen the following behavior?  I think it's a 
> manifestation of the state tracking problems reported recently.
>   The configuration is: 2 clients -> rendering server -> display server .  The 
> clients run the pack SPU, the rendering server runs the readback SPU and the 
> pack SPU, and the display server runs the nop SPU.
>   To do the experiment, watch the display on the *rendering server*, not the 
> display server.  As it renders frames, they'll get drawn there before they are 
> read back.
>   If the configuration is run as described, the geometry rendered from the 
> second client is drawn incorrectly- wrong lighting and wrong transformation.  
> This incorrect image gets read back, and ends up getting passed on to the 
> display server where it is discarded.
>   If the configuration is changed to replace the pack SPU on the downstream 
> end of the rendering server with the nop SPU (so that the display server gets 
> disconnected from the data stream), the images that appear on the rendering 
> server's display are correct.
>   I understand that this should be impossible, but, well, try it and see.  
> I've appended a configuration script which demonstrates this.

I had to do a lot of hacking on your config script to make it work:

1. CRNode doesn't have a Rank() method.
2. CR doesn't have a NodeRange() or ContextRange() method.
3. What is the 'quadrics' protocol?
4. The first server node has a render SPU followed by a pack SPU.
    That's incorrect since a render SPU has to be the last SPU in a chain.
    The pack SPU was being ignored.
    I replaced the render SPU with a readback SPU.
5. I modified all nodes to run on the local host, with assigned port numbers.
6. I replaced the nop SPU with a render SPU.

After all that I have the config running and it looks fine to me.
I'm using Linux.

Attached is the modified script.

-Brian