the Plot.stack offset option can be a function

[Fil]

use case: diverging semantic bars following a Likert scale
see https://talk.observablehq.com/t/diverging-stacked-bar-chart-in-plot/6028

[mbostock]

I’d like this bulleted list to fit on one line. Can you use a short description here, and move the longer description to the paragraph below that talks about offsets? E.g., “If offset is a function, blah blah blah.” We use this style elsewhere in the README if you want to find precedent.

[mbostock]

I like this.

I also wonder if we should have a diverging offset, where the z channel is defined as -2, -1, 0, 1, 2, etc. values, and then you can set the order to z, too, and not need any custom code to do this.

[Fil]

Yes this could be another possibility, but we'd have to pass z independently of fill (which would still be the labels for the color scale), and it might be hard to pass more than one neutral z (since there's only one zero—or maybe two if we check the sign of -0 ;-) ).

[Fil]

I wonder if allowing the user to mutate values in-place is the correct API here. Maybe when it's a function we want them to return [X1, X2]?

[mbostock]

I wonder if allowing the user to mutate values in-place is the correct API here.

I had the same thought. For precedent, you could take a look at the map transform: when the map is implemented a function, it returns an array of values:

plot/src/transforms/map.js

Line 60 in 66142e2

const M = f(take(S, I));

However if the map is implemented a map object with a map method, we use mutation by assigning to the passed T array:

plot/src/transforms/map.js

Lines 67 to 72 in 66142e2

	const mapCumsum = {
	map(I, S, T) {
	let sum = 0;
	for (const i of I) T[i] = sum += S[i];
	}
	};

Also d3-shape’s stack.offset uses mutation:

https://github.com/d3/d3-shape/blob/main/README.md#stack_offset

I think mutation is fine here, mostly because it feels like a pretty low-level API already given the arguments.

[mbostock]

I refactored the example to make it easier to reuse this technique. (I expect in the future we could make these even easier with a new option for the stack transform, but this is fine for now.)

I also switched the dataset to CSV and added attribution.