Point cluster dev

[dy]

supersedes #2493

[etpinard]

looks like you brought in some non-ES5 tokens making our image tests 100% not functional.

[dy]

@etpinard is the test-image build fail caused by scattergl update? Looks like something other causes that.

[etpinard]

you mean this:

see -> #2493 (comment)

[etpinard]

Looks like

is also having issues in nw.js

---

@dfcreative Can you try setting up https://gist.github.com/etpinard/d27a44bd5dbee5490f20 ?

You don't anything docker related. Just make sure to use nw.js version 0.12 -> https://dl.nwjs.io/v0.12.0/

[etpinard]

this should probably be options.marker && count >= TOO_MANY_POINTS like in scattergl/convert.js

[etpinard]

@dfcreative nice! You got the tests to ✅

---

I noticed a few things by looking at the diff:

• scatterpolargl tree logic isn't the same as scattergl https://github.com/plotly/plotly.js/pull/2499/files#r177424508

• colors in gl2d_scatter-color-clustering are a little different. Is this on purpose?

• there's a lot more @flaky scattergl tests. Can you think of any reason what that's the case? This added flakiness might be hidding real problems - as @alexcjohnson noticed before in multiple selections on parcoords axes #2415

• can you please check that this new regl-scatter2d version is at least as good as the previous in terms of memory consumption?

[etpinard]

@dfcreative I'm noticing that drag performance slows down after toggling layout.dragmode from 'drag' to 'select' back to 'drag' . Is this expected?

Moreover, with

var gd = document.getElementById('graph') 
var x = []
var y = []
var ms = []
var N = 1e6
var i = 0

for(; i < N; i++) {
  x.push(Math.random())
  y.push(Math.random())
  ms.push(i % 2 ? 20 : 10)
}

console.time('plot')
Plotly.newPlot(gd, [{
  type: 'scattergl',
  mode: 'markers',
  x: x,
  y: y,
  marker: {size: ms}
}], {
  dragmode: 'closest',
  //xaxis: {range: [-0.1, 2.2]},
  //yaxis: {range: [-0.1, 1.1]}
}, {scrollZoom: true})

selection is slower than on master (see https://codepen.io/etpinard/pen/QQBzyV). Is this expected for 1e6 random points?

[etpinard]

... Oh and we should make sure the new regl-scatter2d works in IE11.

[dy]

scatterpolargl tree logic isn't the same as scattergl

Fixed. Although that affects only data layering order, for polar plots that is a bit less actual.

colors in gl2d_scatter-color-clustering are a little different. Is this on purpose?

That is not clustering, that is adjusted shader behavior in case of no-border gl-vis/regl-scatter2d@023c421. It used to add thin black pixel outline instead of keeping the point color, now that "antialias" has the point color.

flakiness

test-jasmine2 passed or failed sporadically, regardless of changes. According to errors, fullLayout is undefined.

IE11

Thanks, fixed. In fact it supports multiple colors now in IE and works even better.

Memory
Selection performance

Working on that.

[dy]

Ok, with the latest commit selection should work almost as fast as possible, init time is ok, no tree overcalculation and observable memory leaks.

[alexcjohnson]

did this get used?

[dy]

Whoops.

[etpinard]

Well this is amazing! Performance off the latest commit is nothing short of spectacular. Great job @dfcreative 🎉 All my concerns have been addressed.

Now, would you mind publishing a major version of regl-scatter2d (that's v3.0.0 I think) and point-cluster and update the package.json and package-lock.json accordingly? Once that's done, we'll merge this thing right up and move on the sploms 🚀

[etpinard]

Ok. Well-earned 💃 Let's merge this thing.