Bug 138951 – [GTK] Fix the broken edge-distance antialiasing

• Status: NEW
• Component: Platform

[JeongBong Seo]

There exist a wrong matrix multiply to get viewport space position.
---------------------------------------------------------------------------------
vec4 toViewportSpace(vec2 pos) { return vec4(pos, 0., 1.) * u_modelViewMatrix; }
---------------------------------------------------------------------------------
It generates wrong position and affects the inflation ratio to each vertices.
The transformation matrix should be left side.

After fix the multiplication, I got little strange result.
Anti-aliased layer's (inner) contents are expanded 1px.
But it shouldn't happen and only edge should be inflated.

Than I just erase the clamp function to create transformedTexCoord, and finally, got the good result.
But in theory clamping the position should be there, because we using GL_REPEAT in many case.

Please give me some idea or comment.

[JeongBong Seo]

Created attachment 242024 [details]
Patch

[JeongBong Seo]

Created attachment 242025 [details]
Patch

[JeongBong Seo]

Created attachment 242026 [details]
after fix

[JeongBong Seo]

Created attachment 242027 [details]
before fix

[JeongBong Seo]

TEST URL : https://bug-124653-attachments.webkit.org/attachment.cgi?id=217424

[JeongBong Seo]

Comment on attachment 242025 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=242025&action=review

> Source/WebCore/platform/graphics/texmap/TextureMapperShaderProgram.cpp:-177
> -            vec4 clampedPosition = clamp(vec4(position, 0., 1.), 0., 1.);

I'm not sure eliminating the clamp function. But with clamp, it shows expanded output.

[JeongBong Seo]

anybody?