project_size that respects different x/y zoom

[Aystein]

What I want to achieve:
Subclass the scatterplot layer so it renders a rectangle (done) and applies the radiusUnits... part only to the x coordinates. The y coordinates offset should be a fixed offset in a unit of my choice (meter/pixel). This is required for a DNA track I want to implement. So basically I can zoom this track using the OrthographicController using a locked y zoom (to 1) and only the x part changes. The x domain would be base pairs and the y domain always lets say [-1, 1].

However this does seem to give weird results, since the project_size functions defined here https://deck.gl/docs/api-reference/core/project#project_size_to_pixel do not seem to respect the x/y part of the zoom

I tried it with something like this:

  float offsetXPixels = positions.x * outerRadiusPixels;

  float offsetYWorld = positions.y * project_size(10.0);

  vec3 offset = vec3(
    project_pixel_size(offsetXPixels), // X converted to world AFTER pixel scaling
    offsetYWorld,                      // Y stays world → respects x/y zoom
    0.0
  );

  DECKGL_FILTER_SIZE(offset, geometry);

  gl_Position = project_position_to_clipspace(
    instancePositions,
    instancePositions64Low,
    offset,
    geometry.position
  );

but it will only scale correctly if zoom is a single scalar (then it correctly has a fixed height but will scale along the x axis)

Is this a limitation of deck.gl or is there a simple workaround?

[dvmoritzschoefl]

I digged a bit into the shader code and found that for project_size, the x/y units per meter are correctly applied. Using this snippet instead

  vec2 size = project_size(positions.xy);

  float xPixelSize = clampAbsolute(project_size_to_pixel(size.x * 0.45, UNIT_COMMON), scatterplot.radiusMinPixels, scatterplot.radiusMaxPixels);

  vec3 offset = vec3(
      project_pixel_size(xPixelSize),
      size.y * 5.0,
      0.0
  );

works perfectly now. But I wonder if there should also be a project_pixel_size that preserves scaling (or converts to common first, then applies scaling) or am I missing something here?