Difference between `scale_color_discrete()` and `scale_colour_discrete()`

[cbrnr]

I think I found a problem with the functions scale_color_discrete() and scale_colour_discrete(). I thought that one of these functions was just an alias for the other, so in theory they should behave exactly the same.

In one of my recent projects, I needed a global color scale, so I thought that I'd redefine one of these functions with custom colors (let me know if this is not a good idea and if there is a better approach). It turns out that this only works for scale_colour_discrete(), but not for scale_color_discrete()!

Redefining the scale_color_discrete() function does not have the desired effect (easy should be green, medium should be blue, and hard should be red):

library(ggplot2)

scale_color_discrete = function(...) {
    scale_color_manual(..., values=c(easy="green", medium="blue", hard="red"))
}

df = data.frame(
    x=1:3,
    y=1:3,
    z=c("easy", "medium", "hard")
)

ggplot(data=df, mapping=aes(x=x, y=y, color=z)) +
    geom_point(size=10)

This only works with the scale_colour_discrete() function (same code as before, the only difference is the spelling of the function):

library(ggplot2)

scale_colour_discrete = function(...) {
    scale_color_manual(..., values=c(easy="green", medium="blue", hard="red"))
}

df = data.frame(
    x=1:3,
    y=1:3,
    z=c("easy", "medium", "hard")
)

ggplot(data=df, mapping=aes(x=x, y=y, color=z)) +
    geom_point(size=10)

I know that @hadley will probably argue that "colour" is the correct spelling, but I'd still say that this qualifies as a bug 😄.

[teunbrand]

let me know if this is not a good idea and if there is a better approach

I'd like to say this approach is unsupported. I recognise that this is an approach people take, but I'm not quite sure (if and) where this is documented or advertised.

The currently favoured approach is by using options, so you can do:

my_scale <- function(...) {
  scale_color_manual(..., values=c(easy="green", medium="blue", hard="red"))
}
options("ggplot2.discrete.colour" = my_scale)

However, we got something cooking that'll supersede that approach in the sense that default palettes will be included in the theme.

"colour" is the correct spelling

We support both spellings in places we consider the user facing API. Internally ggplot2 renames everything to "colour", including populating default colour scales with the scale_colour_*() functions. Because overriding scales isn't part of the API, overriding scale_colour_discrete() works but scale_color_discrete() doesn't.

For these reasons, I think this is off-label usage and therefore it is not a bug.

[cbrnr]

Thanks for the explanation! I did not really find the favored approach you mention, could you point me to the relevant documentation page? I'll definitely make sure to use it from now on (and of course I'm looking forward to the new approach).

Regarding spelling, do you not think that doing options("ggplot2.discrete.colour" = my_scale) is part of the public API then? Because I assume that only colour is working (same with dplyr.summarise.inform where dplyr.summarize.inform does not work either).

[teunbrand]

could you point me to the relevant documentation page?

Sure, it is in the description section of ?scale_colour_discrete and the details section of the ?scale_colour_continuous docs.

Regarding spelling, do you not think that doing options("ggplot2.discrete.colour" = my_scale) is part of the public API then?

Yes and I'd accept an argument that this would be bug, but we're superseding the options() approach, so it doesn't really make sense to fix a bug on a part of the API that is discouraged. (It might even be mildly helpful to get people to use the newer approach)

[cbrnr]

Thanks! I guess the dplyr.summarise.inform issue is worth reporting then (unless they are also working on superseding the options() approach, but I'll check first).

Edit: I reported it here: tidyverse/dplyr#7631.