Stacked area charts render with gaps when the X coordinates are not continuous

[wylieconlon]

Describe the bug
Gaps appear in the rendering when you skip an X coordinate in your stacked area chart. It only appears in stacked areas, since stacked bars render correctly.

To Reproduce
This is an example of discontinuous data:

      <AreaSeries
        splitSeriesAccessors={['split']}
        stackAccessors={['x']}
        id={getSpecId('y')}
        xAccessor={'x'}
        yAccessors={['y']}
        data={[
          {
            x: 5,
            y: 10,
            split: 'a',
          },
          {
            x: 5,
            y: 30,
            split: 'b',
          },
          {
            x: 10,
            y: 2,
            split: 'a',
          },
          // {
          //   x: 10,
          //   y: 3,
          //   split: 'b',
          // },
          {
            x: 20,
            y: 30,
            split: 'a',
          },
          {
            x: 20,
            y: 20,
            split: 'b',
          },
        ]}
        xScaleType={ScaleType.Linear}
        yScaleType={ScaleType.Linear}
      />

Expected behavior
With the data above, I would expect the chart library to insert an extra vertex at the missing X coordinates.

[nickofthyme]

@markov00 would this be a non-standard behaviour through a prop or should it always ignore discontinuous data?

[wylieconlon]

As a consumer of elastic-charts I would expect stacking to work regardless of the input data. I would expect that the stacking is always cumulative, so that every vertex in the bottom stack will create a vertex in stacks above.

[nickofthyme]

Ok cool, that makes sense! I'll make it happen. Thanks @wylieconlon

[markov00]

@wylieconlon what is the expected output for the above example?

Do you expect that, by default, the missing value is configured as a zero as in the following screenshot

Or something like this

The second one shows better the fact that there is no value at x = 10, where the first imply that there is a value and it's 0.

I think we could go with something configurable by the consumer/user

@nickofthyme I think we can partially resume this: #55

[wylieconlon]

@markov00 I see cases where both of those charts are correct. I just tried recreating this in google sheets and found that they do the first behavior, treating missing values as 0 and connecting the lines. Elasticsearch aggregations support both behaviors, they are null by default but can be configured to have a value: https://www.elastic.co/guide/en/elasticsearch/reference/current/search-aggregations-metrics-avg-aggregation.html#_missing_value

I actually think I was expecting a rendering that is in between the two, because the "gaps" rendering would look very strange in the example I gave at the top, or with any other stacks.

Imagine the shape I've drawn here, but with a fill underneath the dotted line:

[markov00]

@wylieconlon I like the idea of having a dashed line that describe the missing part of the chart.
I think we can use a fitting function #55 to define the value of the missing point: null, 0, last available, the next available etc.

I think the segmented/dashed line can take a bit more time to implement because we need to separate that line from the rest of the chart but we can work on adding the fitting function to that.

[nickofthyme]

Based on slack/zoom conversations with @wylieconlon and @markov00

LineSeriesSpec and AreaSeriesSpec should allow the user to specify a fitting function to fill null values in the dataset. Similar to elastic/kibana#17717.

For the meantime, the style of the "fitted" line should be dashed and area fill, if applicable, will be the same as the "non-fitted" fill color. Later improvement could pass a style to be applied.

Edge cases:

• Stacked areas where void is in lower series such that the upper series would potentially have to fill the void in the lower series.

Code

import React from 'react';
import {
  Axis,
  Chart,
  getAxisId,
  getSpecId,
  Position,
  ScaleType,
  Settings,
  LineSeries,
  AreaSeries,
  DataGenerator,
} from '../src';

const dg = new DataGenerator();
const data = dg.generateGroupedSeries(10);

export class Playground extends React.Component {
  render() {
    return (
      <>
        <div className="chart">
          <Chart className="story-chart">
            <Settings
              theme={{
                areaSeriesStyle: {
                  point: {
                    visible: true,
                  },
                },
              }}
            />
            <Axis
              id={getAxisId('bottom')}
              position={Position.Bottom}
              title={'Bottom axis'}
              showOverlappingTicks={true}
            />
            <Axis id={getAxisId('left')} title={'Left axis'} position={Position.Left} />
            <AreaSeries
              id={getSpecId('test')}
              xScaleType={ScaleType.Linear}
              yScaleType={ScaleType.Linear}
              xAccessor={'x'}
              yAccessors={['y']}
              splitSeriesAccessors={['g']}
              stackAccessors={['x']}
              data={data.map((d, i) => {
                if (i === 5) {
                  return {
                    ...d,
                    y: null,
                  };
                }
                return d;
              })}
            />
          </Chart>
        </div>
      </>
    );
  }
}

[wylieconlon]

Looking at existing work in Kibana, I think this area chart rendering issue can also create single dots:

[nickofthyme]

Yeah that's a good point, I think would be the case for fit of None.

On a related note...I think the should not be points for the fitted data points. This is misleading because it could be assumed as actual data points but they are not.

I was thinking to draw a line with not points for all fit functions. Then the line would be dashed and area filled below, depending on the fit type.

[markov00]

@wylieconlon the example shown already works in elastic-chart (infra is using elastic charts). The problem here is more on how to handle stacked area chart more than single area chart that work as expected (the main difference is that the style of the area chart is configured to shoiw the area points)

[wylieconlon]

@markov00 Sorry if I was unclear, but I showed that screenshot not because it's a problem in isolation but because I'm not sure what the correct stacking behavior would be there. Imagine that we're tracking total network traffic and want to split it by machine, and the traffic looks like above. How would it stack?