Scatter 3d - Geometric position not respected i.e. a dataset added later appears on top even if it is behind spatially.

[igi-chris]

Example: https://codepen.io/chrisprosser/pen/MWqygKw

An example interaction is shown in the short video clip attached. The green cluster (trace 2) has higher values on the x-axis than the orange (trace 1). If I rotate so that they overlap (with the camera at the low x-axis value side) I would expect orange to appear in front as it is closer spatially, but the green appears on top. I then go on to show the same problem with the blue cluster (trace 0).

clusters.mp4

[igi-chris]

I investigated whether we still get the same issue if we put all of the data as a single trace, and it seems that we do still get the same issue: https://codepen.io/Chris-Prosser/pen/ExGOLaJ.

After more investigation, I found that if I remove the alpha from the line (from the above example) replacing: color: "rgba(16, 16, 16, 0.5)" with color: "rgb(16, 16, 16)" in the trace.line spec it resolves the issue. Here is a version that works as expected: https://codepen.io/Chris-Prosser/pen/PoXxQoB

[igi-chris]

I can also correct the position on the original example (with a trace per cluster) by removing the opacity from the markers, here is a multi-trace example working: https://codepen.io/Chris-Prosser/pen/RwEqyaG

It seems like the answer for now is don't use opacity on the 3d plots!

[alexcjohnson]

Thanks @igi-chris - apologies for not noticing this issue earlier, but yes, it's a known issue that we don't handle depth correctly with opacity in 3D #1267 (and other trace types too, eg #3243) - closing since the problem is covered in other issues.

To the best of our knowledge there's no order-independent transparency algorithm with anywhere near the performance we have for opaque objects, so this would require a very different approach.