Fix bug in discrete position scales with continuous limits (fixes #3918)

[paleolimbot]

This PR fixes the case where only continuous values were provided to a discrete scale, which was also provided continuous limits. I introduced this bug in #3380, which changed the way that discrete scales were expanded into continuous limits (old logic, new logic)

library(ggplot2)

df <- data.frame(x = 1:14, y = -2:-15)

ggplot(df, aes(x, y)) +
  scale_y_discrete(limits = -1:-16, labels = LETTERS[1:16])

^{Created on 2020-03-31 by the reprex package (v0.3.0)}

Note that is.discrete() in this PR is linked to is.discrete() in ScaleDiscretePosition$train()...I don't think that the discrete scale training logic is likely to change, but the fix in this PR depends on that behaviour. The other approach would be to check is.null(scale$range$range), which is inconvenient the way the expansion code is currently set up (but totally possible if anyone thinks it matters).

[clauswilke]

In the example you provided, I don't understand why there's a label "P" underneath the plot panel. Is this really supposed to be there?

[paleolimbot]

I've modified the reprex a bit to be more realisitic...the drawing of out-of-bounds limits is a separate guides issue for another day.

With the new warning:

library(ggplot2)

df <- data.frame(x = 1:14, y = -2:-15)
ggplot(df, aes(x, y)) +
  scale_y_discrete(limits = -2:-15, labels = LETTERS[1:14])
#> Warning: Continuous limits supplied to discrete scale.
#> Did you mean `limits = factor(...)` or `scale_*_continuous()`?

^{Created on 2020-04-02 by the reprex package (v0.3.0)}

Also, flagging issue #3918 because it didn't pick up the reference in the title.

[clauswilke]

I wonder whether this warning is in the wrong place. It's making statements about scale functions and their arguments. Shouldn't it be located inside the appropriate scale functions? (I.e., check the limit argument to scale_*_discrete() rather than checking here.)

[paleolimbot]

I was wondering about that too. scale_x|y_discrete() don't have a limits arg (they pass it via ..., so what do you think about putting it in discrete_scale()?

[clauswilke]

Yes, discrete_scale() seems the right place.

[thomasp85]

LGTM, @clauswilke do you have further comments?

[clauswilke]

No further comments from me.