d3-axis

d3-axis copied to clipboard

demo: https://observablehq.com/@d3/multiline-tick-labels-609

How would you encode a leading question mark in your tick format, if you wanted it, for example?

I agree it's better not to introduce a new DSL, but… hmm, why would you ever want that? And you can do it by adding a space before the question mark, or using two question marks if you don't want repeats.

I suspect we could get away with a single boolean option that defaults to true, and no DSL (other than newlines).

Agree, this would probably work well, and it's very simple.

I can see two cases: one is the Month\nYear thing, and the second is someone who returns a "block of text" under each tick, like 

1            2
pie          pies
in           in 
the          the
sky.         sky. 

I don't think someone would ever want to mix the two?

Another alternative with no "\n" DSL would be to allow the tickformat to return an array of lines. But this seems much simpler and "intuitive".

I think the \n is fine—that’s not really a DSL (at least not a new one). And in the other case you mentioned, you turn off the repetition filter and just do it yourself. So I think we’re good with just the option to turn off the tick repetition filter).

Should the first line be wrapped in a tspan? (if we're in a multiline context). Might make it easier to manipulate.

We should also, I guess, trim the trailing \s — this way, people who have a hand-crafted tickFormat which returns \n at the end wouldn't suddenly get loads of empty tspans?

To further the discussion:

Added two options:

• axis.tickMultiline, defaults to "auto"
• axis.tickNoRepeat, defaults to true

The names of these options is certainly bad. tickNoRepeat does a double negation, so we may want to write as tickSuppressRepeats or something.

One choice tickMultiline offers is between "all in tspans" or "first line in text, next lines in tspan". In my tests having the first line in a tspan shifts it 1px to the right, I don't understand why. I suppose that for simplicity we could always do <text>first line<tspan>second line… and that it would be enough? In that case, tickMultiline is probably not be useful at all (users who might be hit by the change can switch back with an explicit .replace("\n", " ")…), so I guess we can probably remove it?

Last remark: to make empty lines really count when they're followed by a non-empty line (see the axisBottomScaleLinearQuarters example), I used a non-breaking space character—it feels really fragile, there must certainly be a better way.

TODO:

• [ ] add a textAnchor option
• [ ] add a dx, dy option

Maybe no extra API surface area is required for textAnchor, dx, and dy as they can be set using post-selection.