[regression-r3320][GLES] td plugin (3d Windows for the cube) has severe clipping/transformation glitches

[Daniel van Vugt]

Using the gles branch, the 'td' plugin has severe clipping/transformation glitches. They're visible on cube rotation (Ctrl+Alt+Drag)

[Jonathan Sambrook (jonathan-hmmn)]

Where "severe" means "makes compiz get completely locked in to a jittery loop".

[Launchpad Janitor (janitor)]

Status changed to 'Confirmed' because the bug affects multiple users.

[Luis Alvarado (luisalvarado)]

I am guessing this is the bug report where, if I activate rotate cube and 3d windows, when rotating the cube, the opened windows show 3d in the desktop and on the sides. http://imgur.com/yme8BLR

If it is, then yeah, I also suffer from this minor bug. It was one I wanted to use to attract users to the imagination found in Linux. But it is minor compared to other effects.

[MC Return (mc-return)]

If we do not manage to fix this for the Compiz 0.9.10 release, we should not include it at all, as it is completely useless in it's current state and just has a big potential to cause unwanted bugs.

Of course I would prefer to see it fixed.

[Luis Alvarado (luisalvarado)]

I understand MC Return, although it is one neat way of bringing windows users to Ubuntu, at least faster. When I do presentations using all the effects that do work, well, they tend to start using Ubuntu because is "they new toy". But I understand and yes, if this creates so many problems, plus the fact that Mir is coming then yeah it would be much better to just eliminate the option.

On the other hand it would be great to know if Mir will have some kind of custom settings where most Compiz "features" can also be done (cube 3d, 3d windows, and other effects).

[MC Return (mc-return)]

@Luis:
I think you misunderstood me ;)
I want this bug to be fixed of course, all I am saying is that we should disable plugins that obviously do not work when we ship/release a version of Compiz, as broken plugins won't help and be useful to anyone.

Regarding this bug here:
I am pretty sure it is a minor issue and I hope someone will soon have time to take a look @ it and finally eliminate it, so we won't have to exclude this awesome plugin once we release 0.9.10.0.

Regarding Mir/Unity8:
I would like to better not comment on that ;)

[MC Return (mc-return)]

Attaching a screenshot showing what is happening.
While all windows on all viewports are transformed and rendered correctly, the deformed cube itself, but also cube gears and the skydome have severe problems and are rendered completely incorrect.

[Erik Ackermann (domohawk)]

Another clue. The amount the sphere/cube is deformed/broken is directly linear to "Window Depth". Indeed setting the depth to 0 produces a perfectly formed sphere/cube (of course with no apparent window 3D effect).

It seems this Window Depth is somehow being used to offset the rendering of each workspace face of the cube/sphere, such that the larger the offset the more 'broken apart' the cube becomes.

With this known parameter to play with, I think I might take a stab at fixing this bug in the next few weeks since the code for this plugin is only one 600 line file. This neat effect is the one thing holding me back from my ideal desktop.

[Břulman Wee-Hool (weehool)]

some news ??

[Tank (digitalrelay)]

I have this bug. All my friends have this bug. My friends won't post because they tell me it's useless to post bugs anymore because Canonical won't do anything anyway. They have lost faith.

 People on other forums are saying that Canonical simply won't fix it.

 Does anyone know why is it being ignored?

[Eleni Maria Stea (hikiko)]

Thank you for the bug report. I have fixed this bug in the GL version and the branches above should fix the 3d windows problems in GLES except of the clipping. The 3d clipping is performed in viewing space (modelview matrix) therefore in GLES where the fixed function pipeline calls for clipping planes etc are ignored, we should set the clipping planes in the vertex shader. Compiz is designed to use some generic shaders for drawing that just do what the fixed function pipeline would do atm. The solution is to unload that shader program at the beginning of the 3d windows plugin drawing call, load another shader that performs the clipping in viewing space and when the drawing is done unload that shader, load the default and continue. I will fix it when there's some time.

[Eleni Maria Stea (hikiko)]

Basically, GLES is not supported officially anymore in Ubuntu, therefore I have to change the bug status to Won't Fix. But if someone wants to take the time to fix it the solution is here: https://bugs.launchpad.net/compiz/+bug/1024208/comments/11 Thank you and apologies.

[Eleni Maria Stea (hikiko)]

The cubeaddon-3d windows interactions are a different bug. The branch above (in progress) fixes the clipping in GLES2, although the deformation fixes are in the TODO list as well. In which device you run the GLES2 branch?

[Eleni Maria Stea (hikiko)]

Progress update:
In the linked branch above I can fake the clipping on the pixel shader as follows: instead of clipping polygons in view space I discard the pixels that would be clipped by the planes in the fragment shader. The shaders have been tested and work with the fixed function pipeline 3d windows implementation.

TODO:
In order to have clipping in GLES2 I need to modify the compiz shader system (see opengl plugin) to accept clipping planes and re-write the td function to use GLES2 compatible code. This is in progress but it's not a priority so, be patient...

The 3d windows work OK on the desktop.