feat: define DATA_VALUE_TYPE and DATA_VALUE_TYPE_IS_* for shaders

[LDeakin]

This is handy for writing generic user shaders that use getDataValue() etc

[chrisj]

Just my thoughts...

DATA_TYPE or dataType would be a preferable name.

dataType x = toNormalized(getDataValue());

Also, the several following ${getShaderType(dataType,)} could be replaced with the define.

[jbms]

Maybe DATA_VALUE_TYPE --- since DATA_TYPE might be expected to more correspond to the typescript DataType.

[jbms]

Can you also add this to the documentation (https://github.com/google/neuroglancer/blob/master/src/sliceview/image_layer_rendering.md)?

[jbms]

wait does that actually work?

I thought the GLSL preprocessor was similar to the C preprocessor except more limited? Without token pasting ## most of the "advanced C preprocessor" tricks are unavailable.

In particular, using the C preprocessor it is possible through advanced techniques to perform an equality check like you have above but I don't know that it would be possible in GLSL.

[chrisj]

Having float in the conditional errors for me, it seems to restricted to integer values.

  #define FOO 6

  #if FOO == 6
    setColor(vec4(1.0, 0.0, 0.0, 1.0));
  #endif

[LDeakin]

Thanks both for catching that! I should have tested that basic use case. How about the proposed solution in c93afc3?

E.g. shader

#uicontrol invlerp normalized
void main() {
  DATA_VALUE_TYPE value = getDataValue();
  #if DATA_VALUE_TYPE_IS_UINT8_T
  if (value.value > 80u) {
    emitRGBA(vec4(1.0f, 0.0, 0.0f, 1.0f));
    return;         
  }
  #endif
  emitGrayscale(normalized());
}

[jbms]

I'm curious what your actual use case is for this.

[LDeakin]

The main use case for my lab is to check for a data-type dependent sentinel value for masking. Doing this at runtime in neuroglancer is much simpler than what we do currently, and I'd guess other people may find some utility in this.