Fix issue 1546, fill aesthetic does not show in the legend guide for stat_summary

[clauswilke]

I think the way to fix this issue is to give geom_smooth() a first-class se parameter that is considered when drawing the panel. This allows the geom and the legend drawing code to always be in synch. See resulting behavior in the examples below.

One note: Currently one regression test fails because the geom_smooth() and draw_group() functions have different default arguments and the test checks for that. I think the test is wrong in this particular case, but I wanted to bring this up for discussion. Relevant lines in the code, with explanation for why it should be this way:
https://github.com/clauswilke/ggplot2/blob/ed016144c97cf67424498453a2f4e06925fdba92/R/geom-smooth.r#L119-L126

Examples of geom_smooth() with stat_summary():

library(ggplot2)
ggplot(mtcars, aes(cyl, mpg, fill = factor(am))) +
  geom_smooth(stat = "summary")
#> No summary function supplied, defaulting to `mean_se()

ggplot(mtcars, aes(cyl, mpg, fill = factor(am))) +
  geom_smooth(stat = "summary", se = FALSE)
#> No summary function supplied, defaulting to `mean_se()

# stat_summary doesn't have an `se` parameter, hence
# default is `se = FALSE`
ggplot(mtcars, aes(cyl, mpg, fill = factor(am))) +
  stat_summary(geom = "smooth")
#> No summary function supplied, defaulting to `mean_se()

ggplot(mtcars, aes(cyl, mpg, fill = factor(am))) +
  stat_summary(geom = "smooth", se = TRUE)
#> No summary function supplied, defaulting to `mean_se()

And, for completeness, all combinations of geom_smooth() / stat_smooth() with se = TRUE / se = FALSE work as expected:

library(ggplot2)
ggplot(iris, aes(Sepal.Length, Sepal.Width, color = Species, fill = Species)) +
  geom_smooth()
#> `geom_smooth()` using method = 'loess' and formula 'y ~ x'

ggplot(iris, aes(Sepal.Length, Sepal.Width, color = Species, fill = Species)) +
  geom_smooth(se = FALSE)
#> `geom_smooth()` using method = 'loess' and formula 'y ~ x'

ggplot(iris, aes(Sepal.Length, Sepal.Width, color = Species, fill = Species)) +
  stat_smooth()
#> `geom_smooth()` using method = 'loess' and formula 'y ~ x'

ggplot(iris, aes(Sepal.Length, Sepal.Width, color = Species, fill = Species)) +
  stat_smooth(se = FALSE)
#> `geom_smooth()` using method = 'loess' and formula 'y ~ x'

Created on 2018-05-10 by the reprex package (v0.2.0).

[hadley]

I think this approach is reasonable; the way parameters works is all a little hacky, so you should feel free to disable the test for this case (IIRC there are already a few other geoms where it's disabled)

[clauswilke]

I've disabled the consistency check for geom_smooth(). (You were correct, there was already a list of such geoms.) I have added two simple visual tests, since there weren't any for geom_smooth() or stat_smooth() yet.

[hadley]

Thanks!

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/