for NaN, y1 and y2 with sort is different from y

[aboveyunhai]

I was wondering why NaN or invalid value was treat as maximum for y1 and y2, while it's the minimum in y,
I could do something like

y2: (d) => isNaN(d.population) ? -Infinity : d.population

to fake the minimum or maximum, but feel like abusing the property, or maybe I misunderstood how y1 | y2 works?

timeseries = [
  {year: 2014, population: 7295.290765},
  {year: 2015, population: 7379.797139},
  {year: 2016, population: 7464.022049},
  {year: 2017, population: 7547.858925},
  {year: 2018, population: NaN},
  {year: 2019, population: 7713.468100},
  {year: 2020, population: 7794.798739}
];

Plot.barY(timeseries, {
  x: "year", 
  y: "population",
  sort: { x: "y" }
}).plot()

Plot.barY(timeseries, {
  x: "year", 
  y1: 0,
  y2: (d) => d.population,
  sort: { 
     x: "y2" // "y" 
   }
}).plot()

[Fil]

The difference is that if you specify y you get implicit stacking, and the invalid datum is stacked as a rectangle of height equal to 0. Whereas if you specify y2, there is no stacking, thus y2 is NaN and the invalid datum is discarded. The difference is visible if you set stroke: "red", inset: -1,.

(Note that it is not "minimum" nor "maximum": Plot and D3 put invalid values at the end when they sort. For instance if you say sort: { x: "-y2" } to sort by descending y2, the invalid datum will still be to the right.)

[aboveyunhai]

I see, then what would be the proper solution to address this issue?
Just a custom function inside the sort to replicate the sorting and make invalid to be in the front?

[Fil]

The easiest way is to indicate the domain explicitly in the definition of the scale, like so:

Plot.plot({
  x: {domain: [2014, 2015…]}
  ...
})

or with a function that orders the values as you see fit:

Plot.plot({
  x: {domain: d3.sort(timeseries, d => isNaN(d), d => d.population).map(d => d.year)}
  ...
})