Legend keys from geom_sf() change upon repeat printing of the same plot

[francisko-rezende]

I'm having a super weird issue with ggsave and, after sharing it at RStudio community, Mara Averick suggested I open an issue here. So here I am 😅

Basically, the legends of plots I generate using ggsave do not match what is displayed in RStudio's "Plots" window nor the legends of plots I generate using RStudio's "Export" option. This is happening for maps created using geom_sf.

Here's an example:

library(rnaturalearth)
library(ggplot2)

world <-
  rnaturalearth::ne_countries(scale = "medium", returnclass = "sf") # creates a r object  called "world" with spatial data

ggplot2::ggplot(data = world) + #starts plot with "world" as its data
  geom_sf(aes(color = continent), show.legend = "point") # draws map with colored continent outlines

^{Created on 2020-04-10 by the reprex package (v0.3.0)}

As you can see, the map above has points as its legend key. That is the output I wanted to get, however, when I use ggsave("ggsave_test.pdf") to save the map as a pdf file this is what I get:

The weird thing is that this problem seems to be exclusive to ggsave since saving the plot using RStudio's "Export" functionality or knitting a Rmarkdown document result in the plot having points as its legend key.

This happens for .ps, .pdf, .jpeg, .tiff, .png, .bmp, and .svg files both when leaving ggsave()'s plot argument blank (e.g., when it uses last_plot() like in the reprex above), and when using the saved map as the argument for plot. There's an exception though I used purrr::walk to test the different file formats and the .eps file that was generated using the named map as ggsave()'s "plot" argument has the points as its key ! The plots generated using ggsave(filename = here::here("ggsave_no_loop.eps"), plot = map), and purrr::walk without the named map as plot's argument have the squares as their keys though.

I think this might be a macOS-specific issue (or even something that is happening only in my computer) because this problem doesn't happen in rstudio.cloud.

Cheers and thanks for your amazing work!

[clauswilke]

It looks like the problem is the use of the last_plot() function, not ggsave(). I suspect the show.legend parameter is dropped when the plot is cloned.

library(rnaturalearth)
library(ggplot2)

world <-
  rnaturalearth::ne_countries(scale = "medium", returnclass = "sf") # creates a r object  called "world" with spatial data

ggplot(data = world) + 
  geom_sf(aes(color = continent), show.legend = "point")

last_plot()

^{Created on 2020-04-10 by the reprex package (v0.3.0)}

[clauswilke]

Simpler reprex. Just printing the same plot twice triggers the problem.

library(ggplot2)

nc <- sf::st_read(system.file("shape/nc.shp", package = "sf"), quiet = TRUE)
p <- ggplot(nc[1:3, ]) +
  geom_sf(aes(color = NAME), show.legend = "point")

p

p

^{Created on 2020-04-10 by the reprex package (v0.3.0)}

[clauswilke]

The problem is in this line:

ggplot2/R/layer-sf.R

Line 53 in b41e1b9

self$show.legend <- TRUE

When we added it, we got confused about when it is executed. It is only executed when the plot is built, not when it is originally constructed. As a consequence, the legend type changes to the default after one printing.

library(ggplot2)

nc <- sf::st_read(system.file("shape/nc.shp", package = "sf"), quiet = TRUE)
p <- ggplot(nc[1:3, ]) +
  geom_sf(aes(color = NAME), show.legend = "point")
p$layers[[1]]$show.legend
#> [1] "point"
invisible(ggplot_build(p))
p$layers[[1]]$show.legend
#> [1] TRUE

^{Created on 2020-04-10 by the reprex package (v0.3.0)}

[FriGruner]

Hi Claus, I am experiencing exactly the same issue with geom_ribbon_pattern. Is there a chance that you could fix this, too? Thanks a lot!

[clauswilke]

@FriGruner Please open a new issue and provide a complete, reproducible example. Refer to this issue here for reference.

[clauswilke]

Also, I'm not aware that geom_ribbon_pattern() even exists, so you may want to check that this is even a ggplot2 problem.