Defaults for bin counts

[yurivish]

Filing this issue at Mike's suggestion in the d3 slack. I ran into some surprising behavior using Plot.binX without specifying any arguments (https://observablehq.com/d/cf5ba27c74172e3e) and Mike suggested adding a default to handle this case.

Right now this produces the expected plot:

Plot.rectY(data, Plot.binX({ y: 'count' })).plot()

While this produces an unexpected (not empty, but also not correct) plot:

Plot.rectY(data, Plot.binX()).plot()

One question would be what channels to default to count – in this case I wanted the y channel, but I'm not sure if that's always the right default.

[Fil]

[REMOVED FLAWED ANALYSIS]
[At some point something] calls something like Float64Array.from([[0, 1], [2]]), which results in [NaN, 2].
In other words when a bin contains just 1 value [d0], the value is taken as d0. If it contains more values [d0, d1…], it is NaN. So the vertical bars you see are the bins that contain exactly one data point.

[Fil]

Going through this again; the bug is provoked by the following sequence:

1. The grouping operation has (correctly) transformed the data into 5 buckets, say [[0.01, 0.12, 0.19…], [0.21, 0.22, 0.39…], [0.41, 0.52, 0.59…], [0.61, 0.72, 0.79…], [0.84]]. Note how only the last of these buckets has 1 element.

2. The next transform is the implicit Plot.stackY transform called by Plot.rectY. It receives our list of buckets as data (an array of arrays), and applies:

Y = valueof(data, y, Float64Array)

with y being identity (the default for Plot.stackY).

This by the magic virtue of JavaScript's typecasting returns a Float64Array(NaN, NaN, NaN, NaN, NaN, 0.84). Et voilà, we get 1 bar of height 0.84.

The fix is simply (as you suggested) to add correct defaults in binX: binX(outputs = {y: "count"}, …

I'll add a test for binX, binY, and probably bin also needs a default output ({r: "count"} ?)