guide_legend's effectiveness depends on order of aesthetic mappings

[william-l-barcelona-fr]

I found a possible problem with guide_legend()s effectiveness within guides() depending on the order of aesthetic mappings in the ggplot() call.

I expected that nrow = 2 in the guide_legend() call for shape in guides would have an effect, but it doesn't seem to be working. It does work if I switch shape to be mapped before color though.

Here is the code to reproduce the bug:

library(ggplot2)
  
ggplot(
  mtcars,
  aes(x = wt, y = mpg, color = as.factor(cyl), shape = as.factor(cyl))
) +
  geom_point() +
  guides(shape = guide_legend(nrow = 2))

ggplot(
  mtcars,
  aes(x = wt, y = mpg, shape = as.factor(cyl), color = as.factor(cyl))
) +
  geom_point() +
  guides(shape = guide_legend(nrow = 2))

Should the order of the aesthetic mappings matter here?

[teunbrand]

The colour and shape legends need to be merged, and which one of the legends ends up assimilating the other is somewhat arbitrary. I'm not convinced the arbitrariness is a bug per se. I think the correct way to approach this problem currently is to give the same legend to both aesthetics, so using guides(shape = guide_legend(nrow = 2), colour = guide_legend(nrow = 2)). We could also make guide merging more strict and only allow merges when the guide dimensions match, but that'd also result in the same approach.

[william-l-barcelona-fr]

Thank you. Targeting the earliest-specified aesthetic for any guide_legend customization that we want to be shared seems to be working in the cases I've tried, but I take your point that this isn't guaranteed to work in all cases or forever and the safest idea is to specify any desired settings for all relevant legends.