Date and number formatting for ordinal axis ticks.

[mbostock]

It’s possible to have an ordinal axis where the domain value is dates, for example with a bar chart. This isn’t a temporal axis (which is continuous), but it’d still be nice to support d3-time-format specifiers, e.g. tickFormat: "%b". Probably the axis should look at the domain values and check if they are temporal. Similarly if the domain values are numbers, d3-format specifiers should be supported.

[mbostock]

A related technique that could be useful here is D3’s multi-scale time format, but configured to a specific time interval:

https://github.com/d3/d3-scale/blob/8afe6bd4af7a1f0be6b4d231126bdc8a83659ab5/src/time.js#L59-L67

For example, maybe scale.tickFormat(count) on a UTC scale could return a tick format function that returns the empty string if the given time isn’t the expected interval. Then could say something like:

d3.scaleUtc().domain(d3.extent(data, d => d.date)).tickFormat(10)

Or perhaps more directly:

d3.utcTickFormat(...d3.extent(data, d => d.date), 10)

And maybe even the 10 could be optional.

Related d3/d3-time#8.

[mbostock]

I also find this pattern useful:

tickFormat: ((i, f) => d => i(d) < d ? null : f(d))(d3.timeDay, d3.timeFormat("%A %m/%d"))

Maybe that could be better expressed as:

tickFormat: d3.timeFormat("%A %m/%d", d3.timeDay)

Though I would actually prefer to use date.toLocaleString for this instead of d3-time-format (so it can be properly localized), and only use d3-time for the filtering. Maybe you could say something like:

tickFormat: Plot.formatTime("en", {month: "numeric", day: "numeric"})

And then Plot would be smart enough to return the empty string for times that are not UTC midnight? Though that feels like it would be easy to misuse with an ordinal date axis if the dates are at irregular intervals.