Re: [Celestia-developers] More about GLSL Error.

[Pat S. <pa...@su...>]

NAILED it!

Chris Laurel wrote:
> In the default solarsys.ssc, the specular mask for Earth is in the alpha
> channel of the base texture, whereas Pluto has a specular mask in a
> separate texture. So, different shaders will be used for Earth and Pluto.

Well, uncommenting that default line in solarsys.ssc that has the 
specular texture instead works for Earth now too.

> I'm baffled. First of all, I think that texture2D should set all
> components of color even if the texture doesn't have an alpha channel. In
> older versions of OpenGL, there is a well-defined result when sampling
> from a texture with 'missing' channels. For example, when sampling from an
> RGB texture (no alpha channel), you'll always see an alpha value of 1.0.
> I've seen nothing in the spec suggesting that GLSL's texture2D()
> instruction behaves any differently.

After doing very rudimentary debugging by setting earth's colour to 
either red, green, or blue based on if statements (how lame!), your 
eMail sparked an idea...

> Second, even if some strange value was read out of the alpha channel, it
> should not cause a crash. An alpha greater than 1.0 should present no
> problems at all. Have you tried substituting a constant > 1.0 for color.a?
> Something like this:
> 
> gl_FragColor = color * diff + 2.0 * spec;

I thought that was the case, but trying this proved me wrong. The above 
ran fine, the highlight just got really bright.

> gl_FragColor.rgb = color.rgb * diff.rgb + color.a * spec.rgb;
> gl_FragColor.a = diff.a;

I had actually tried something really similar to this on my own.

First, I split the original:

	gl_FragColor = color * diff + color.a * spec;

into two lines thusly:

	gl_FragColor = color * diff
	gl_FragColor += color.a * spec;

That crashed. But running each of those lines separately (changing '+=' 
for '=') ran fine! WTF?!

Your two lines worked *perfectly*. I even duplicated the exact original 
meaning by changing the second line to:

	gl_FragColor.a = diff.a + spec.a;

That worked too!

Then I had the truly crazy idea. Remember how we had trouble in the past 
setting a vector with int values? I thought maybe it wasn't being 
typecast properly again. So, I went back to the original line, and made 
a very trivial addition.

	gl_FragColor = color * diff + float(color.a) * spec;

BINGO.

I tested with various colours and intensities in the .ssc file. It works.

> It sounds to me like something is very broken in the ATI driver.

Yes it does! I'm gonna give them shit over this one!

The output from texture2D() is very clearly a vec4 made of floats!

I'll check in this one-line change. Im theory, it shouldn't affect 
anything else, not even the binary output. Except on ATI cards.

I have meself a perfectly functional OpenGL2.0 Render Path. And all it 
took was one line, or 10 hours.

--Pat