Check for dropped aesthetics

[clauswilke]

Check for dropped aesthetics and issue a warning if aesthetics are dropped that shouldn't be. Fixes #3250.

As a consequence of this change, stats now need to advertise which aesthetics they are dropping, via the new dropped_aes field. If they don't, unit tests may fail because code now generates warnings that previously ran quietly.

Example of the new feature:

library(tidyverse)

ggplot(mtcars, aes(mpg, fill = am)) + geom_density(alpha = 0.4)
#> Warning: The following aesthetics were dropped during statistical transformation: fill
#> ℹ This can happen when ggplot fails to infer the correct grouping structure in
#>   the data.
#> ℹ Did you forget to specify a `group` aesthetic or to convert a numerical
#>   variable into a factor?

ggplot(mtcars, aes(mpg, fill = am, group = am)) + geom_density(alpha = 0.4)

^{Created on 2022-06-09 by the reprex package (v2.0.0)}

[thomasp85]

Generally looks good - should we update the docs for the affected stats to say that these aesthetics are dropped? Seems like a nice thing to do

[clauswilke]

Let me think about documenting the dropped aesthetics. In some cases it's just a temporary variable that gets dropped, so documenting that is not so useful. In other cases (e.g. stat_density_2d()) one of the input variables gets dropped, so documenting that makes sense.

[thomasp85]

yeah - "thankfully" the stats aes documentation is not automatic so it is just a matter of cherry picking where to add it to the docs

[thomasp85]

@clauswilke is it possible to get the docs updated soon? I plan to start the release process soonish to give the revdep maintainers the summer to fix potential issues

[clauswilke]

@thomasp85 I think I can do it today. I was waiting for you to merge the vctrs PR, but that has happened, correct?

[thomasp85]

Ah, sorry -- I missed that. Yes, vctrs has been merged in

[clauswilke]

@thomasp85 I think I've addressed everything. I documented the stats where it seemed appropriate. In particular, I didn't document stat_boxplot() because I think it'd be more confusing if I did than if I didn't.

[thomasp85]

lgtm