Fix color leakage issue in shader rendering

[mathsvisualization]

Fix Color Leakage in Fragment Shader

Motivation

While rendering overlapping triangles with opposite orientations,
minor color leakage artifacts were observed.

The issue was caused by aggressive alpha scaling in the fragment shader:

float a = 0.95 * frag_color.a;

This resulted in excessive blending, especially when negatively oriented triangles were involved.

Proposed Changes

• Alpha scaling factor was reduced from 0.95 to 0.90:

float a = 0.90 * frag_color.a;

This lowers the blending strength and prevents unintended color bleed.

• Orientation-Based Alpha Correction
  For negatively oriented triangles, the following transformation is used:

if (orientation < 0)
    a = -a / (1 - a);

Mathematically, the modified alpha value is:
a' = -a / (1 - a)
This ensures correct cancellation behavior under the standard blending equation:
(1 - alpha) * dst + alpha * src
The transformation guarantees that blending a positively oriented triangle followed by a negatively oriented one restores the original fragment color.

Testing

• Render overlapping triangles with opposite orientations
• Verify no residual tint remains
• Confirm stable blending behavior

from manimlib import *

class GlyphTest1(InteractiveScene):
    def construct(self):
        char = "B"
        t = TexText(char)
        t.set_height(FRAME_HEIGHT - 2)
        self.add(t)

Result

• Cleaner blending
• No color leakage
• Stable fragment output

Before

After

[3b1b]

Interesting, I don't see that on my system. I'm happy to merge this, but a few notes:

This whole "0.95" is clearly hacky, so I'm not proud of this, but there are a few other places in the code where it is sneakily relevant. One is line 476 of manim/manimlib/shader_wrapper.py, where that correction is counteracted. The other is line 288 of that same file, where we make sure the fill border aligns with the conventions of fill.

I've contemplated where it makes sense to roll back from the winding fill approach. There's an elegance to it compared to finding an internal triangulation, both to make interpolation more natural and also to be able to avoid having to recompute that triangulation if ever shape changes, but it does have these weird side quirks.

The motivation for having a more aggressive scaling factor is that this whole tactic for fill has a side effect where opacity 1 objects overlapping end up brighter over the overlap, even when the same color. For example:

The effect gets amplified when that scaling factor is turned down. For example, here it is at 0.5, where the appropriate line 476 correction would be to multiply back by 2:

Of course, if 0.95 causes your artifacts, the lesser of two evils is to scale it down as you have, I'd just add the corresponding corrections.