The bug also exists when run the case with LIBGL_ALWAYS_SOFTWARE=1.
Some  mesa xdemos and  EGL cases such as opengles2/es2_info also has this bug.

The issue still exists with the follow commits:
Libdrm:         (master)2.4.33-20-gd72a44c7c4f5eea9c1e5bb0c36cb9e0224b9ca22
Mesa:           (8.0)fa68a8bae3961808288cfd84d5a7843f6fc0f317
Xserver:             (server-1.12-branch)xorg-server-1.12.1
Xf86_video_intel:    (master)2.19.0-36-ga3d37fb29f8dffb0e370ad95783994aaa7eccfaf
Kernel:         (drm-intel-fixes) 2e7a44814d802c8ba479164b8924070cd908d6b5

This bugs still exists. GLSL demos and Mesa_demos all have render error with only "X" but run well with "gnome".

The issue still exists with the follow commits:
---------------------------------
Kernel_version:     3.8
Libdrm:             2.4.42
Mesa:               (9.1)9.1-rc2
Xserver:            (server-1.13-branch)xorg-server-1.13.2
Xf86_video_intel:   (master)2.21.0
Cairo:              (master)1.12.12
Libva:              staging-20130205
Libva_intel_driver: staging-20130205

I don't understand how to reproduce this bug.  What do you mean by "only X"?  GNOME uses X.  I've also verified that they work fine in KDE (also using X).

Can you provide some instructions on how to reproduce it?  (Like, are you creating an .xinitrc that just runs the demo, or...?)

Sorry for my ambiguous explanation.
"only X" mean that only execute command "xinit" but not start gnome or KDE before run the demo.
for example:
1.xinit&
2../gloss

To the contrary, for example,
1.xinit&
2.gnome
2../gloss

I managed to bisect this down to:

commit 9a7f84d6b28e180ef79349b36de9a5d58a1e2dc9
Author: José Fonseca <jfonseca@vmware.com>
Date:   Thu Jul 14 17:28:52 2011 +0100

    Squashed commit of the following:
    ...

which is the commit that removed Mark Kilgard's GLUT from the Mesa tree.  After that, building Mesa no longer built libglut.  This meant that the demos started running with your distro-provided libglut, which is typically freeglut (not Kilgard's GLUT).

With freeglut and no window manager, the Mesa demos never get their Reshape() function called, so their initial frustum and such never get set.  (Kilgard's GLUT did cause Reshape() to get called.)  However, I don't believe this is a bug; I believe freeglut's behavior is valid.

I've filed #61640 against Mesa demos; they should probably be enhanced to set up their initial state and not rely on Reshape() getting automatically called.  This is just generally a good idea.

In the meantime, I would just run with a window manager.  Closing this as NOTOURBUG as it's a bug in the demos (or possibly freeglut), but not an actual Mesa regression.

(In reply to comment #7)
> I managed to bisect this down to:
> 
> commit 9a7f84d6b28e180ef79349b36de9a5d58a1e2dc9
> Author: José Fonseca <jfonseca@vmware.com>
> Date:   Thu Jul 14 17:28:52 2011 +0100
> 
>     Squashed commit of the following:
>     ...
> 
> which is the commit that removed Mark Kilgard's GLUT from the Mesa tree. 
> After that, building Mesa no longer built libglut.  This meant that the
> demos started running with your distro-provided libglut, which is typically
> freeglut (not Kilgard's GLUT).
> 
> With freeglut and no window manager, the Mesa demos never get their
> Reshape() function called, so their initial frustum and such never get set. 
> (Kilgard's GLUT did cause Reshape() to get called.)  However, I don't
> believe this is a bug; I believe freeglut's behavior is valid.
> 
> I've filed #61640 against Mesa demos; they should probably be enhanced to
> set up their initial state and not rely on Reshape() getting automatically
> called.  This is just generally a good idea.

I believe this is a freeglut bug.  Kilgard's GLUT came first and the behaviour of window reshaping is documented here:
http://www.opengl.org/resources/libraries/glut/spec3/node48.html#SECTION00083000000000000000

according to that, the reshape function should always get called at least once when the window first appears.

Ken, thanks for the investigation. I'm closing this on behalf of QA.