after_scale fails to find some aesthetic variables

[atusy]

I am highlighting some points by a combination of stage() and after_scale().
I found that after_scale can find aesthetic variables, but not variables in the original data.
It is fine, if that's a design.
As a workaround, I let after_scale find the variable in the data by mapping it to an aesthetic variable.
It works in some case, but not always.
As far as I tried, it seems the error is related to discrete variables.

library(ggplot2)
g <- ggplot(mtcars) +
    aes(wt, mpg, highlight = am == 1) +
    geom_point(
        aes(colour = stage(cyl, after_scale = ifelse(highlight, colour, 'gray'))),
        shape = 21
    )

# Okay
g

# Okay
g + aes(fill = am)

# Error
g + aes(fill = as.factor(am))
#> Error in ifelse(highlight, colour, "gray"): object 'highlight' not found

# Error
ggplot(iris) +
    aes(Sepal.Length, Sepal.Width, highlight = Species == "setosa") +
    geom_point(
        aes(
            colour = stage(
                Species, after_scale = ifelse(highlight, colour, 'gray')
            )
        )
    )
#> Error in ifelse(highlight, colour, "gray"): object 'highlight' not found

^{Created on 2020-01-13 by the reprex package (v0.3.0.9001)}

[yutannihilation]

The error seems to occur on drawing legends. I don't get what's the expected behaviour here...

library(ggplot2)

g <- ggplot(mtcars) +
  aes(wt, mpg, highlight = am == 1) +
  geom_point(
    aes(colour = stage(cyl, after_scale = ifelse(highlight, colour, 'gray'))),
    shape = 21
  )

g + aes(fill = as.factor(am)) + theme(legend.position = "none")

^{Created on 2020-01-13 by the reprex package (v0.3.0)}

[thomasp85]

Defining a random aesthetic (highlight) will invariantly lead to undefined behaviour. There is no way to know how the stat and geom will treat it or carry it over. This has always been true and was also the case for the old stat() and ..var.. notations

[yutannihilation]

I agree there's no garantee that such random aesthetics work, but this is a different problem. The problem is that limited variables are used to draw legends. Probably this can happen with more standard aesthetics.

[thomasp85]

I do t think I follow you @yutannihilation — what is the problem with the legend?

[yutannihilation]

I'll try creating some example tomorrow...

[thomasp85]

Ah - I see from your comment now. That is a separate issue all together. Can you open a new one?

as for this: after_stat() has only access to columns returned by the stat. after_scale() has only access to the final aesthetics. There is no way columns from the original dataset can be carried over as the data might get changed considerably during preparations

[yutannihilation]

Sure. I don't feel it's a completely different issue, but let's discuss on a new one.

[yutannihilation]

@thomasp85 Here it is: #3729