HeatmapLayer for v9

[donmccurdy]

So far, this PR resolves a number of runtime and compile-time errors in HeatmapLayer and its dependencies. The layer still does not render, but also doesn't log errors that appear serious. Feels like a good state to share WIP here and discuss.

Dependencies:

• chore(luma): Upgrade to luma.gl v9 alpha-50 #8379
• feat(engine): Transform → BufferTransform + TextureTransform luma.gl#1896

[donmccurdy]

Current dependencies on/for this PR:

• master
  • PR chore(luma): Upgrade to luma.gl v9 alpha-50 #8379
    • PR chore(luma): Upgrade to luma.gl v9 alpha-50 #8382
    • PR HeatmapLayer for v9 #8380 👈

This stack of pull requests is managed by Graphite.

[donmccurdy]

@felixpalmer I requested a review here, but as you'll see I have a lot of TODOs and questions scattered through the code, it's not a working PR yet. I'm hoping it might be helpful to talk through some of this code tomorrow – I'll schedule some time if that's ok!

[ibgreen]

OK I have done a first readthrough of this heatmap PR, there is a lot going on and hard to spot issues in a review. I’ll come back to it later, but initial comments

• My instinct would be be to divide and conquer, i.e. break out the transforms into separate functions or even better into separate files and try to unit test them separately.
• Your debug tools (buffer reads, framebuffer dumps etc) are good we should make sure they are available from luma.gl, though we could create a debug folder in deck.gl/core for now. And we should figure out how to make the luma spector support work out of the box.
• Ideally we should clean up all the debug enabling code from this PR, there is a lot of commented out code etc.

[donmccurdy]

@ibgreen thanks for your review! In the current state of the PR, nothing is being drawn to framebuffers — so all code here is intended to flush out errors rather than to be mergeable code. That's also why I've hidden <View id="basemap">, as having a second WebGL canvas on the page makes Spector.js more tedious to use. Once there are "signs of life" from the heatmap I'll remove most of this code. My main concerns are whether uniforms and attributes are being constructed with the new APIs correctly — note that the TextureTransform and BufferTransform take a superset of ModelProps as input.

---

Some further updates, thanks to your recent debugger tooling additions! No action or response required on any of this, just sharing my debugging notes —

I've noticed that the heatmap-layer-weights-transform model (the first transform step) does not have vertex attribute values the first time it draws. The second model and transform step, heatmap-layer-max-weights-transform, prints nothing to the console. Screenshot:

Stepping into the debugger, the missing lines for the second model are indeed empty objects, {}, which might indicate something to investigate here. The first step doesn't produce output though, so that could also be the cause.

The second time the gpu grid aggregator draws (e.g. on a zoom event), the heatmap-layer-weights-transform model does have vertex attribute values:

It still doesn't produce output, but this is probably helpful to know while debugging. It's also interesting that positions and positions64Low display the same array. Probably because they're interleaved?

[felixpalmer]

In v8, these are also LINEAR. This PR introduced the regression

[donmccurdy]

Huge thanks to @felixpalmer for debugging and getting this rendering again!