expose scales

[Fil]

exposing scales, extracted from #484

build at https://observablehq.com/@observablehq/expose-plot-scales-538

With many tests!

[mbostock]

Instead of materializing a family field on the internal scale object, I wonder if we can have a function that takes a specific scale type and returns either a corresponding family, or a boolean. For example:

function isTemporalType(type) {
  return type === "time" || type === "utc";
}

Or the other approach:

const quantitative = Symbol("quantitative");
const temporal = Symbol("temporal");
const ordinal = Symbol("ordinal");

function family(type, key) {
  switch (type) {
    case "diverging":
    case "diverging-sqrt":
    case "diverging-pow":
    case "diverging-log":
    case "diverging-symlog":
    case "cyclical":
    case "sequential":
    case "linear":
    case "sqrt":
    case "threshold":
    case "quantile":
    case "pow":
    case "log":
    case "symlog":
      return quantitative;
    case "identity":
      return registry.get(key) === position ? quantitative : ordinal;
    case "utc":
    case "time":
      return temporal;
    case "ordinal":
    case "categorical":
    case "point":
    case "band":
      return ordinal;
  }
}

This would be useful when coercing channel values based on the scale type, for example. (I haven’t reviewed the suggested code above closely…)

[mbostock]

I’m going to start working on this. I rebased against main and I got a few test failures. I skipped one and regenerated the polylinear test. Is this the expected output?

[Fil]

All tests are fixed now. For reference the correct polylinear test is:

(The error was in maybePiecewiseRange which was checking the old type==quantitative.)

[mbostock]

Thanks for the help. I’m going to continue rewriting the tests today (schedule permitting).

[mbostock]

Okay, I think this is 99% ready. The last things I want to think through are:

1. The semantics of the isDescendingDomain test versus the reverse options. Thank you for replying to my notebook! Probably either semantics are fine, I just want to feel comfortable with it before committing.
2. The internal confusion around scale options vs. axis options for label and labelAnchor, and specifically whether we should also materialize axis-related options in the returned scale object.

The latter one embodies the difference between Plot’s internal and external representations. Externally, scales and axes options are on the same object, but internally, we split them up. Probably we want to favor the external representation when we materialize scale objects and return them from plot.scale. Although I suppose alternatively we could add a plot.axis method in the future, and you’d be expected to splat them together if needed. But at any rate since we are already exposing the label axis option on the returned scale object, I’m guessing that we do in fact want to include all the axis options in the exposed scale and use a unified representation.

Okay, maybe that’s more like 92% ready. 😉