heatmapgl TODO list

[etpinard]

Before getting merged into the main plotly.js dist bundle, we need to make sure following problems are addressed

• heatmapgl test-images are not reproducible. Maybe this is just a matter of bumping the gl-shader requirement.
• top-most and right-most points aren't being catch by the hover picking routine

• double-check that we are handling NaNs correctly. One good test would be to Plotly.restyle(gd, 'type', 'heatmapgl') all our heatmap test mocks.
• include only attributes that are relevant to the gl2d version (instead of simply copying the heatmap attributes). For example, heatmapgl doesn't support zsmooth, xgap and ygap.

See the heatmapgl-primetime branch for more some WIP work.

[etpinard]

Maybe @dfcreative would be interested in looking some of the items ⏫ ?

cc @monfera @jackparmer

[dy]

Sure!

[jackparmer]

That would be awesome @dfcreative ! Want to come by the office for lunch this Friday?

[dy]

@jackparmer sure!

[dy]

Ok, as for

top-most and right-most points aren't being catch

that seems to be wrongly rendered pickBuffer:

only 9 levels out of 10 are here. It does not contain ids >= 90 and whole decimals 10, 20, 30, 40, 50, 60, 70, 80, which refer to the last row/col

[dy]

test-images are not reproducible

All the difference I see across all the heatmap tests is in fonts rendering and antialiasing

original:

result:

Can it be the reason of failing tests @etpinard?

[etpinard]

@dfcreative It's similar to https://github.com/plotly/streambed/issues/6307#issuecomment-231818023

Generating one heatmapgl mock at a time works fine. But inside a loop, image generation is off.

[dy]

Ok, the problem of top-right points is in this line, which ignores edge vertices. It is easily fixable for the x axis by adding dx, but difficult for the y axis, because pick shader interpolates value.
A fact that pickBuffer in some magic way transforms float array of ids into vec4s of ints adds to the complexity of issue.

[dy]

Ok, top-most and right-most picking problem should be fixed by gl-vis/gl-heatmap2d#2

[jackparmer]

Hey @dfcreative - Do you have an idea for this one by chance?

heatmapgl test-images are not reproducible. Maybe this is just a matter of bumping the gl-shader requirement.

What do you think is your timeline for the other 2?

[dy]

@jackparmer all heatmap related tests are reproduced fine on my machine, I guess I should discuss that with @etpinard this friday, to see the issue in action to understand what's wrong.

[dy]

@etpinard ok, last gl-vis/gl-heatmap2d#2 should fix the images/picking issues

[etpinard]

@dfcreative nice job. Your patch even works on CI:

The first two items are ✅

[etpinard]

Oh. I spoke too soon. Picking is still broken for a few other mocks.

For example, Earth_heatmap when replace data[0].type from 'heatmap' to 'heatmapgl':

[etpinard]

Same story for the 4 mock:

[dy]

@etpinard gl-vis/gl-heatmap2d@04cae76 should fix picking, tested on various mocks. My render of 4.json is different though, I guess because of custom layout options
. Should it be fixed?

[etpinard]

@dfcreative thanks! I'll give it a test 🚗 on Monday.

[etpinard]

@dfcreative with your last commit, this is looking really good. I'll spend a few more minutes double-checking, but looks like we're really to something mergeable. 🍻