The interval option for line and area should round down

[Fil]

closes #1931

[mbostock]

Hmm, the intersection with #2390 is interesting… Do we like this? My initial assumption is that the dense interval transform should produce zero instead of undefined… not sure what is happening here.

[mbostock]

Can you explain what you’ve changed here? How does this change the behavior? I see some refactoring that seems to try to reduce code duplication but I can’t follow the logical change that is actually causing the interval transform to round down…

[mbostock]

This is the change (outputting x: "x1") that matters. 👀

[mbostock]

Hmm, this seems wrong to me. Here is my test case:

test(async function aaplInterval() {
  const aapl = (await d3.csv<any>("data/aapl.csv", d3.autoType)).slice(0, 81);
  return Plot.plot({
    x: {grid: true, ticks: "week"},
    marks: [
      Plot.dotY(aapl, {x: "Date", y: "High", interval: "week"}),
      Plot.lineY(aapl, {x: "Date", y: "High", interval: "week", reduce: "max"}),
      Plot.lineY(aapl, {x: "Date", y: "High", interval: "week", reduce: "min"})
    ]
  });
});

In main, the dot and line align because both round to the middle of the interval:

But in this branch, the dot and line don’t align because the line rounds down:

Maybe #1931 is just ill conceived? Or should we get rid of maybeIntervalMidX and maybeIntervalMidY, and have the dot and text marks round down, too? That would look like this:

[mbostock]

And if we change the dot and text marks to round down to the start of the interval, they don’t work as well to label rects (here representing the extent of a week along the x-axis) as in the seattlePrecipationSum test:

Hmmm! 🤔

[mbostock]

I think probably #1931 is ill conceived and should be closed; if you want the dot to align with the line, the dot should have the interval option applied, too.