Implicitly stack bars

[mbostock]

If a barY is specified without y1 or y2, implicitly apply the stackY transform. Same with barX, x1, and x2.

This is stacked:

Plot.barY(crimea, {x: "date", y: "deaths", fill: "cause"})

This overlaps:

Plot.barY(crimea, {x: "date", y2: "deaths", fill: "cause"})

The idea is that this reduces the possibility of unintentionally overlapping opaque bars, but you can still opt-out of stacking easily if that’s what you intend.

[mbostock]

I’d love help getting the unit tests working again…

[mbostock]

I get the impression that this should also apply to the area mark.

[Fil]

Overlapping areas are more common than overlapping bars, but yes it would be consistent if bars and areas worked similarly.
But then there is this pesky issue of #325 and you might want to also include the "seriesX" transform (probably a poor choice of a name) explored in https://observablehq.com/d/f6a7975f2ad4519a

[mbostock]

Added areas. I don’t know how to address your other comment.

[mbostock]

Oops, need to stack rect too. Works well though! It’s fun that you can turn on or off stacking by switching between y and y2.

[Fil]

A foolproof solution for stacked areas should include the seriesX transform (and similarly seriesY transform for areaX).

So I suppose 7594f0e#diff-8f62f80a02a317ce3253c0795d19093acea8456b90e720732b62455ef9d96203R94 might become something like maybeStackYSeriesX(…)

I thought it would be a plugin, but I can research it now as a PR to this PR or to main.

The risk with seriesX is that it might create a combinatory explosion if you make a mistake in your choice of x, transforming n data points into n^2.

[mbostock]

Do you think this is a blocker? It doesn’t feel like one for me.

[Fil]

No, not a blocker; it's just a case where the default doesn't work out of the box; let's say that, if the series are incomplete on x and you want to represent them with a stacked areaY, you need the plugin; in the meantime we recommend using stacked rects (or bars, or rules).

[mbostock]

Yep. I feel like the seriesX transform should probably instead be an option to the stack transform that fills any missing values across stacks with zero.