Some examples fail e.g. webgpu_materials_matcap and it seems it has to do with the new select() statement. For some reasons, the compiled GLSL ends up wrong.

From SpectorJS:

nodeVar1 = bool( f_nodeUniform3 );
if ( ( nodeVar1 == true ) ) {
    positionViewDirection = normalize( v_positionViewDirection );
    nodeVar2 = normalize( vec3( positionViewDirection.z, 0.0, ( - positionViewDirection.x ) ) );
    normalViewGeometry = normalize( v_normalViewGeometry );
    NORMAL_normalView = normalViewGeometry;
    normalView = NORMAL_normalView;
    matcapUV = ( ( vec2( dot( nodeVar2, normalView ), dot( cross( positionViewDirection, nodeVar2 ), normalView ) ) * vec2( 0.495 ) ) + vec2( 0.5 ) );
    nodeVar3 = ( f_nodeUniform4 * vec3( matcapUV, 1.0 ) ).xy;
    nodeVar0 = vec2( nodeVar3.x, 1.0 - nodeVar3.y );
}
else {
    matcapUV = ( ( vec2( dot( nodeVar2, normalView ), dot( cross( positionViewDirection, nodeVar2 ), normalView ) ) * vec2( 0.495 ) ) + vec2( 0.5 ) );
    nodeVar0 = ( f_nodeUniform4 * vec3( matcapUV, 1.0 ) ).xy;
}

Um, it seems the if/else statement from the select holds way too much code.

I think in the short-term this could potentially be resolved by using the uniformFlow context in your highlighted line of code, but the default behavior is certainly not correct.

I did a short test and got this as output in the fragment shader. There's probably some optimizations that can be done to only evaluate the f_nodeUniform4 * matcapUV operation once...

nodeVar17 = texture( nodeUniform2, ( nodeVar1 == true ) ? vec2( nodeVar16.x, 1.0 - nodeVar16.y ) : ( f_nodeUniform4 * vec3( matcapUV, 1.0 ) ).xy );

How did you update the TSL source to get this output?

this._flipYUniform.equal( true ).select( uvNode.flipY(), uvNode ).uniformFlow();

Thanks! I like the fact that the if/else blocks are replaced with a ternary operation.

Unfortunately, none of the suggested versions work right now. Both the uniformFlow() code as well as the uvNode.toVar() definitions break environment maps in different ways.

E.g. add uniformFlow() and then test webgpu_loader_gltf_transmission or webgpu_loader_gltf_sheen. The environment maps end up way too bright.

@Mugen87 I'm checking this bug, the caching and optimization system is failing at some point in the conditionals.