Multiple issues with transforms

[dmt0]

I’ve been trying to add transforms to multiple trace types with varying degrees of success. Documentation doesn’t make it clear which trace types support transforms. All I know from discussion with @etpinard, is that anything that takes 2d arrays as input doesn’t support them.
react-chart-editor as of now does not prepopulate target/targetsrc:

      "transforms": [
        {
          "type": "filter",
          "target": [],
          "targetsrc": null
        }
      ]

Results:

scatter3d:

• filter: many warnings: [.Offscreen-For-WebGL-0x7fef4b0d6800]RENDER WARNING: Render count or primcount is 0.

-> https://codepen.io/etpinard/pen/dgWJZg

scatter3d + marker:’line’

• filter: Uncaught (in promise) TypeError: Cannot read property 'slice' of undefined at LinePlot.proto.update

-> https://codepen.io/etpinard/pen/Edmobo

cone, streamtube, scattergeo, table, heatmap (with 1d array input), contour (with 1d array input)

• any transforms: as I add them to data, they never get added to fullData

choropleth, scatterternary, scatterpolar, pie:

• aggregate: Error while plotting: TypeError: Cannot set property 'type' of undefined at Object.plots.supplyTransformDefaults

pie:

• filter: Error: <path> attribute d: Expected number, "MNaN,NaNa189.5,18…".

-> https://codepen.io/etpinard/pen/wYdppB

2d contour histogram:

• filter: Error while plotting: TypeError: Cannot read property 'length' of undefined at makeCrossings (plotly-with-meta.js?ab3a:146042)

Edit:
Add this one to the list:

• When defining a scatter with just Y defined, and than add an aggregate:
  Uncaught TypeError: Cannot set property 'type' of undefined at Object.plots.supplyTransformDefaults
  https://codepen.io/dmt0/pen/xJoWoY

This one got fixed in: #3093

[etpinard]

Ok. Thanks for the report!

What does targetsrc: null which isn't a plotly.js attribute) do?. I suspect it injects target: [] in the user data, but I would like to confirm

[etpinard]

But yeah, we could do better than declaring transforms in the schema for all trace types (including the ones that don't support them yet):

plotly.js/src/plots/attributes.js

Lines 160 to 167 in 7bba266

	transforms: {
	_isLinkedToArray: 'transform',
	editType: 'calc',
	description: [
	'An array of operations that manipulate the trace data,',
	'for example filtering or sorting the data arrays.'
	].join(' ')
	}

[dmt0]

yes, in fullData I see target: Array[0]

[etpinard]

Most of these problems occur because transforms[i].target: [] turns data arrays into [] during the calc step (as filter transforms are applied using _module.calcTransform) and the trace type module plot method expect empty data array to be associated with visible: false (from the supplyDefault logic) traces.

Note that

Plotly.newPlot(gd, [{
  y: [1, 2, 1, 3],
  transforms: [{
    type: 'filter',
    target: []
  }]
}], {
  showlegend: true
})

has a legend item, whereas

Plotly.newPlot(gd, [{
  y: [],
}], {
  showlegend: true
})

does not, which I believe is correct.

---

Perhaps, the most robust to handle this would be to make transform items with target: [] (or any other falsy value?) set enable to false similar to how traces with empty data array get visible:false during supplyDefaults.

[etpinard]

scatterpolar now looks like:

https://codepen.io/etpinard/pen/YJVYYd

[etpinard]

As far as I can tell, @antoinerg 's #3158 removed the transforms attribute container from all traces that do not support transforms.