master fails on RH6/CentOS6

[doutriaux1]

see: https://open.cdash.org/viewTest.php?onlyfailed&buildid=3940803

[doutriaux1]

@sankhesh @jbeezley I'm going to try to get the buildbot server going here. Can you try in the meantime to get your CentOS VM talk to your buildbot server? Thx.

[doutriaux1]

@dlonie can you remind us what the various part of the picture supposed to be on CDASH?
I usually use left vs right, but in the test it seems that bottom/top is completely different. For example:
https://open.cdash.org/testDetails.php?test=360286757&build=3940803

[alliepiper]

Heh, yeah....CDash doesn't handle multiple images very well.

As I understand it, Top Left is the first test image, Top Right is the first reference image, and Bottom Left/Right are the same for the second image. Any others aren't shown in the fancy slider widget.

I usually just look at the image diffs that follow the slider when more than two images are present.

[alliepiper]

FWIW, the labeled isoline tests are failing due to the lack of a stencil buffer in the GLX visual. @durack1 and @jbeezley had some additional insights in the discussion on #1456.

[doutriaux1]

@jbeezley to follow up on your idea at #1456 here the xwininfo output:

xwininfo -root

xwininfo: Window id: 0x2e8 (the root window) (has no name)

  Absolute upper-left X:  0
  Absolute upper-left Y:  0
  Relative upper-left X:  0
  Relative upper-left Y:  0
  Width: 5040
  Height: 1920
  Depth: 24
  Visual: 0x56
  Visual Class: TrueColor
  Border width: 0
  Class: InputOutput
  Colormap: 0x55 (installed)
  Bit Gravity State: NorthWestGravity
  Window Gravity State: NorthWestGravity
  Backing Store State: NotUseful
  Save Under State: no
  Map State: IsViewable
  Override Redirect State: no
  Corners:  +0+0  -0+0  -0-0  +0-0
  -geometry 5040x1920+0+0

[alliepiper]

I setup a CentOS 6.7 VM using mesa-libGL 10.4.3 and can reproduce this. I focused on the stenciling issues with the labeled isolines.

What I've noticed:

• Only background mode has the issue. Plotting in a window stencils the lines out properly.
• At the time the stencil is created, OpenGL reports that GL_STENCIL_BITS is 8, meaning that the stencil buffer exists and has 8 bits (as it should).
• Dumping the stencil buffer with glReadPixels and writing this data out to a file shows that the stencil buffer is populated with the correct data (there is a quad stenciled out at each label's location).

From this, my next guess is that it's a mesa bug. For some reason, this GL implementation is ignoring the stencil buffer, or incorrectly performing the stencil test when using an FBO rather than a driver-provided visual (FBOs are used with offscreen rendering). I'll need to confirm this with a minimal example and if I can, I'll report the bug upstream to mesa / CentOS.

This will likely be a couple of weeks before I can get to it, as I've burned through my allocation for UVCDAT to track this down and other projects need my attention, but wanted to share that there is some progress being made on this issue.

[doutriaux1]

we commented these out for 2.4 but would be nice to be back in 3.0

[aashish24]

@doutriaux1 should we close this bug?

[doutriaux1]

hum... They are probably still commented out though.