netbook-launcher crashed with SIGSEGV in glGetString()

StacktraceTop:glGetString () from /usr/lib/libGL.so.1
_cogl_features_init () at cogl.c:170
cogl_features_available (features=COGL_FEATURE_TEXTURE_NPOT)
_cogl_texture_slices_create (tex=0x998c5d8)
cogl_texture_new_from_data (width=1, height=1,

Same on ibm t23 with the 9.10 and also with the lucid-netbook live-cd (from 11.02.10).

So, there are two bugs here - an underlying bug in the 3D drivers which seem to be crashing in glGetString while we have a valid GL context - and that netbook-launcher is not catching the bad state from Clutter - clutter is failing to create valid stages & actors, but netbook-launcher is proceeding merrily on.

As it stands now all clutter apps crash when using the savage driver (on a T23). This is what the clutter test apps do:

(gdb) c
Continuing.
ClutterX11-Message: [BACKEND] ./clutter-backend-x11.c:143: XOpenDisplay on ':0.0'
ClutterX11-Message: [BACKEND] ./clutter-backend-x11.c:170: Getting the X screen
Clutter-Message: [BACKEND] ./clutter-backend.c:173: Units per em: 12.02
ClutterX11-Message: [BACKEND] ./clutter-device-manager-x11.c:70: XInput support not enabled
ClutterX11-Message: [BACKEND] ./clutter-device-manager-x11.c:167: Added core pointer device
ClutterX11-Message: [BACKEND] ./clutter-device-manager-x11.c:176: Added core keyboard device
ClutterX11-Message: [BACKEND] ./clutter-backend-x11.c:225: X Display ':0.0'[0x807d130] opened (screen:0, root:99, dpi:96.094581)
Clutter-Message: [MISC] ./clutter-feature.c:91: checking features
Clutter-Message: [MISC] ./clutter-feature.c:95: allocating features data
ClutterGLX-Message: [BACKEND] clutter-backend-glx.c:407: Retrieving GL fbconfig, dpy: 0x807d130, xscreen; 0x8086f70 (0)

Program received signal SIGSEGV, Segmentation fault.
0x00478336 in glGetString () from /usr/lib/mesa/libGL.so.1

OK, I have made some progress with Clutter and the savage. There are two things at play here:

1) clutter always requests non-zero alpha depth when looking for visuals. The savage driver does not support non-zero alpha depth so it does not return any visuals. Clutter uses RGB visuals by default but clutter-gtk wants RGBA. I patched clutter to not request non-zero alpha depth and can now run clutter-based apps.

2) clutter fails to catch the failure to acquire a visual - and hence an OpenGL context - and blindly tries to call OpenGL functions. This is not possible without a context, which leads to the segfaults seen here (and elsewhere).

I changed clutter to get rid of this request and suddenly it works - sort of. That alpha request is there for a reason so the savage driver should be changed to either support alpha (difficult if the hardware does not offer support) or ignore the alpha request.

For now clutter is semi-usable when patched. I'm running clutter 1.2.2 so the patch is not applicable to the version currently used in Ubuntu. I'll have a look at implementing alpha support in the savage driver which would be the real solution to this problem.

I have patched the savage driver so it exposes modes with non-zero alpha depth. The patch only changes the modes which the driver exposes, I have not verified that all exposed modes are actually functional. Using the patched driver clutter applications are functional (within the constraints of the hardware of course) and the segfaults disappear.

The patch applies to mesa-7.7 as used in Lucid.

Again, this patch makes clutter apps work but I have not verified that all exposed modes are actually supported by the driver and the hardware. This is the mode table after applying the patch according to glxinfo on a 24 bit framebuffer:

24 GLX Visuals
   visual x bf lv rg d st colorbuffer ax dp st accumbuffer ms cav
 id dep cl sp sz l ci b ro r g b a bf th cl r g b a ns b eat
----------------------------------------------------------------------
0x21 24 tc 0 32 0 r y . 8 8 8 8 0 24 8 0 0 0 0 0 0 None
0x22 24 dc 0 32 0 r y . 8 8 8 8 0 24 8 0 0 0 0 0 0 None
0x92 24 tc 0 32 0 r . . 8 8 8 8 0 24 8 0 0 0 0 0 0 None
0x93 24 tc 0 32 0 r . . 8 8 8 8 0 24 8 16 16 16 16 0 0 Slow
0x94 24 tc 0 32 0 r y . 8 8 8 8 0 24 8 16 16 16 16 0 0 Slow
0x95 24 tc 0 32 0 r y . 8 8 8 8 0 24 8 16 16 16 16 0 0 Slow
0xa2 24 tc 0 24 0 r . . 8 8 8 0 0 24 8 0 0 0 0 0 0 None
0xa3 24 tc 0 24 0 r . . 8 8 8 0 0 24 8 16 16 16 0 0 0 Slow
0xa4 24 tc 0 24 0 r y . 8 8 8 0 0 24 8 0 0 0 0 0 0 None
0xa5 24 tc 0 24 0 r y . 8 8 8 0 0 24 8 16 16 16 0 0 0 Slow
0xa6 24 tc 0 24 0 r y . 8 8 8 0 0 24 8 0 0 0 0 0 0 None
0xa7 24 tc 0 24 0 r y . 8 8 8 0 0 24 8 16 16 16 0 0 0 Slow
0xb4 24 dc 0 32 0 r . . 8 8 8 8 0 24 8 0 0 0 0 0 0 None
0xb5 24 dc 0 32 0 r . . 8 8 8 8 0 24 8 16 16 16 16 0 0 Slow
0xb6 24 dc 0 32 0 r y . 8 8 8 8 0 24 8 16 16 16 16 0 0 Slow
0xb7 24 dc 0 32 0 r y . 8 8 8 8 0 24 8 0 0 0 0 0 0 None
0xb8 24 dc 0 32 0 r y . 8 8 8 8 0 24 8 16 16 16 16 0 0 Slow
0xc5 24 dc 0 24 0 r . . 8 8 8 0 0 24 8 0 0 0 0 0 0 None
0xc6 24 dc 0 24 0 r . . 8 8 8 0 0 24 8 16 16 16 0 0 0 Slow
0xc7 24 dc 0 24 0 r y . 8 8 8 0 0 24 8 0 0 0 0 0 0 None
0xc8 24 dc 0 24 0 r y . 8 8 8 0 0 24 8 16 16 16 0 0 0 Slow
0xc9 24 dc 0 24 0 r y . 8 8 8 0 0 24 8 0 0 0 0 0 0 None
0xca 24 dc 0 24 0 r y . 8 8 8 0 0 24 8 16 16 16 0 0 0 Slow
0x43 32 tc 0 32 0 r y . 8 8 8 8 0 24 8 0 0 0 0 0 0 Ncon

24 GLXFBConfigs:
   visual x bf lv rg d st colorbuffer ax dp st accumbuffer ms cav
 id dep cl sp sz l ci b ro r g b a bf th cl r g b a ns b eat
----------------------------------------------------------------------
0x4a 0 tc 0 32 0 r . . 8 8 8 8 0 24 8 0 0 0 0 0 0 None
0x4b 0 tc 0 32 0 r . . 8 8 8 8 0 24 8 16 16 16 16 0 0 Slow
0x4c 0 tc 0 32 0 r y . 8 8 8 8 0 24 8 0 0 0 0 0 0 None
0x4d 0 tc 0 32 0 r y . 8 8 8 8 0 24 8 16 16 16 16 0 0 Slow
0x4e 0 tc 0 32 0 r y . 8 8 8 8 0 24 8 0 0 0 0 0 0 None
0x4f 0 tc 0 32 0 r y . 8 8 8 8 0 24 8 ...

I noticed that the original submitter runs the framebuffer at 16bpp instead of 24. The patch does NOT add RGBA modes for 16 bit visuals so to use clutter he'll have to switch to 24bpp.

Marking as invalid the netbook-launcher task as it impacts the mesa driver. RAOF, Bryce? can you have a look at it?

This bug was fixed in the package netbook-launcher - 1:2.1.16-0ubuntu1

---------------
netbook-launcher (1:2.1.16-0ubuntu1) lucid; urgency=low

  * New upstream release:
    - Prevent crash in keyboard navigation into System category. (Jan Arne
      Petersen) (LP: #465548)
    - Works around a strange behaviour in the delivery of the
      child-focus-changed signal.a
    - Wrong clutter handling by opengl driver fallback to -efl launcher
      (LP: #467474)
    - Add right-click in all category menus and show Change Desktop Background
      menu like in Favorites. (LaserJock) (LP: #455143)
    - Fix keyboard navication in favourite view (Jan Arne Petersen)
      (LP: #472999)
    - Put the home, destkop, download and examples folder first (LP: #538485)
 -- Didier Roche <email address hidden> Thu, 01 Apr 2010 17:59:12 +0200

knarf, thanks for developing the patch - I would recommend you also send this patch upstream, as they may be able to spot some ways to improve it, and if it's included upstream we won't have to worry about maintaining it uniquely in ubuntu going forward.

I'll send a message to the mesa-dev list to see what the reaction is. While the patch makes it possible to run software which depends on RGBA visuals it is clear that it does not always work as it should, eg. text in clutter shows up as black or white rectangles. This might be caused by a faulty implementation of mipmapping or some other bug which was hidden hitherto.

This bug was fixed in the package mesa - 7.7.1-1ubuntu2

---------------
mesa (7.7.1-1ubuntu2) lucid; urgency=low

  * Add 103_savage-expose_fbmodes_with_nonzero_alpha.patch: Expose
    fbmodes with non-zero alpha depth. Fixes issue where clutter apps
    crash when using the savage driver. Thanks to knarf for developing
    the fix.
    (LP: #467474)
  * mesa-common-dev should depend on libdrm-dev
    (LP: #490811)
 -- Bryce Harrington <email address hidden> Wed, 14 Apr 2010 12:06:00 -0700