Merge lp://qastaging/~smspillaz/unity/unity.fix_864784_868120_872625v2 into lp://qastaging/unity

This branch does break the repaint-cycle. While it compiles/works/runs screen-updates only happen when a window is moved. From the diff I currently don't see what you're missing.

I tried this on my laptop (intel graphics, mesa) and it worked ok. So something odd is happening with nvidia (binary driver). Might need to see how ATI/fglrx behaves with this.

Nvidia here and I can confirm MacSlow's findings.

> Nvidia here and I can confirm MacSlow's findings.

Damn, ok. Their driver is broken :( I will have to work around it.

Noted: the driver is not carrying over attributes when set using glPushAttrib (GL_CURRENT_BIT), Jay says that they should be, and this is what the mesa drivers do, so NVIDIA should know about this.

I can confirm that it properly repaints now, can't test multiple monitors though.

+1

Can someone with fglrx test this ?

On 11/10/11 12:18, Sam Spilsbury wrote:
> Noted: the driver is not carrying over attributes when set using glPushAttrib (GL_CURRENT_BIT), Jay says that they should be, and this is what the mesa drivers do, so NVIDIA should know about this.

merge approved

--
Gordon Allott
Canonical Ltd.
27 Floor, Millbank Tower
London SW1P 4QP
www.canonical.com

Testcase is to make sure that the screen actually redraws the way you'd expect it to. Bonus points if you can

 1) Test with multiple monitors to make sure the blur is correct on both
 2) No blur offsets
 3) No "flickering" when switching workspaces using expo
 4) No crashes when you hotplug a monitor on intel.

There is no unit test.

*cough* *cough* someone test this on fglrx *cough*

Instead of
  uScreen->_fbo = UnityFBO::Ptr (new UnityFBO (geometry));
you can go
  uScreen->_fbo.reset(new UnityFBO(geometry));

Saves a little typing (and it is more efficient in behaviour),

  UnityFBO::UnityFBO (CompRect geometry)
should be
  UnityFBO::UnityFBO(CompRect const& geometry)

Avoids an extra copy.

This requires testing for merge