Bug 8065

FWIW, I replaced the abort() with a _mesa_problem() call that should be more
informative.

(In reply to comment #1)
> FWIW, I replaced the abort() with a _mesa_problem() call that should be more
> informative.
No longer aborts, whole saturn is fully red, and the message is:
Mesa 6.5.1 implementation error: bad slang opcode 0x43
Please report at bugzilla.freedesktop.org

Looks like the slang compiler is generating opcodes (such as
slang_asm_vec4_copy) which the interpreter can't execute.

Michal Krol will probably have to look into that.

Fixed in CVS. Though I would like to see the particualr shader
causing the assertion if it is possible.

I have managed to extract the shader causing trouble and, as I expected, the fix
was not necessary - the shader is ill-formed, I am affraid. The Mesa's GLSL
compiler is under development and allows you to do some "extra" things that are
illegal. Other compilers from other vendors also have some "features", but they
are different from ours.

Particularly, Mesa GLSL currently accepts the following code:

const vec3 precalc_sin30 = vec3 (sin (30.0));
const float other_precalculated_constant = clamp (pow (cos (66.6), 16.0), 0.0, 1.0);
void main () {
gl_FragColor = vec4 (precalc_sin30, other_precalculated_constant);
}

It accepts it only because it does not accept lots of semantic errors.

On the other hand, celestia shader initializes a global variable with a uniform
which is illegal. The initializer must be a constant expression. Mesa will
accept that (see above), but will evaluate the initializer at compile-time,
meaning that it will be 0.0 all the time.

This is the celestia's shader.

uniform vec3 ambientColor;
vec4 diff = vec4(ambientColor, 1.0);
uniform vec3 lightcolor0;
varying vec2 diffTexCoord;
uniform sampler2D diffTex;
varying vec2 normTexCoord;
varying vec3 lightDir0;
uniform sampler2D normTex;
void main(void)
{
vec4 color;
vec3 n = texture2D(normTex, normTexCoord.st).xyz * vec3(2.0, 2.0, 2.0) -
vec3(1.0, 1.0, 1.0);
float l;
l = max(0.0, dot(lightDir0, n)) * clamp(lightDir0.z * 8.0, 0.0, 1.0);
diff.rgb += l * lightcolor0;
color = texture2D(diffTex, diffTexCoord.st);
gl_FragColor = color * diff;
}

Simply move the initialization part to the main() function body and it will run
on Mesa and other non-nVIDIA IHVs (I suspect nVIDIA is the development
environment for celestia team).

uniform vec3 ambientColor;
uniform vec3 lightcolor0;
varying vec2 diffTexCoord;
uniform sampler2D diffTex;
varying vec2 normTexCoord;
varying vec3 lightDir0;
uniform sampler2D normTex;
void main(void)
{
vec4 color;
vec3 n = texture2D(normTex, normTexCoord.st).xyz * vec3(2.0, 2.0, 2.0) -
vec3(1.0, 1.0, 1.0);
float l;
l = max(0.0, dot(lightDir0, n)) * clamp(lightDir0.z * 8.0, 0.0, 1.0);
vec4 diff = vec4(ambientColor, 1.0);
diff.rgb += l * lightcolor0;
color = texture2D(diffTex, diffTexCoord.st);
gl_FragColor = color * diff;
}

Roland, can you report the bug to the celestia project?

(In reply to comment #5)
> I have managed to extract the shader causing trouble and, as I expected, the fix
> was not necessary - the shader is ill-formed, I am affraid.
Well, usually crashing isn't how opengl user errors should be handled ;-). 

> Roland, can you report the bug to the celestia project?
Done: http://shatters.net/forum/viewtopic.php?p=76076

Mass version move, cvs -> git