Legends duplicated when collected

[robjhyndman]

This was working in R3.6.3, but not in R4.0.

library(ggplot2)
library(patchwork)
p1 <- ggplot(mtcars) + geom_point(aes(mpg, disp, col=gear))
p2 <- ggplot(mtcars) + geom_boxplot(aes(gear, disp, group = gear, col=gear))
(p1 | p2 ) +
  plot_layout(guides = "collect") & theme(legend.position = 'bottom')

^{Created on 2020-04-27 by the reprex package (v0.3.0)}

[bling1000]

The same happened to me ! I updated my R version to 4.0.0 recently.

[mikecuoco]

Same happened to me after I updated to R 4.0.0

[teng-gao]

+1 My R is 3.6.1

[hexuebao]

Same happened to me after I updated to R 4.0.0

[MMJansen]

Anyone know if a solution is to be expected soon? If I can help in anyway, please let me know.

[llrs]

No idea, but it would help if someone could find the source of the problem. At least then it would be easier to fix it (via PR or Thomas himself).

[thomasp85]

I will look into this within a week or two

[mcanouil]

So I took a look into that issue and the behaviour of all.equal is not the same in R 4.0.0 on grobs (https://github.com/thomasp85/patchwork/blob/master/R/guides.R#L31). Although, I don't know (yet) how to fix this.

library(patchwork)
library(ggplot2)
library(gtable)

p <- ggplot(iris, aes(Sepal.Length, Petal.Length, colour = Species)) +
  geom_point()

guides <- list(
  gtable_filter(ggplotGrob(p), "guide-box"), 
  gtable_filter(ggplotGrob(p), "guide-box")
)

unnamed <- lapply(guides, patchwork:::unname_grob)

i <- 2
j <- 1
res <- all.equal(unnamed[[i]], unnamed[[j]], check.names = FALSE, check.attributes = FALSE)
head(res)
#> [1] "Component \"grobs\": Component 1: Component 1: Component 1: Component 1: Component 2: Component 3: Component 3: Component 1: Component 2: Component 1: Component 2: Component 2: Component 1: Component 2: Component 9: 1 string mismatch"                                        
#> [2] "Component \"grobs\": Component 1: Component 1: Component 1: Component 1: Component 2: Component 3: Component 3: Component 1: Component 2: Component 2: Component 2: Component 2: Component 1: Component 2: Component 9: 1 string mismatch"                                        
#> [3] "Component \"grobs\": Component 1: Component 1: Component 1: Component 1: Component 2: Component 3: Component 3: Component 1: Component 2: Component 3: 1 string mismatch"                                                                                                         
#> [4] "Component \"grobs\": Component 1: Component 1: Component 1: Component 1: Component 2: Component 3: Component 3: Component 1: Component 2: Component 5: Component 1: Component 11: Component 3: Component 2: Component 2: Component 1: Component 2: Component 9: 1 string mismatch"
#> [5] "Component \"grobs\": Component 1: Component 1: Component 1: Component 1: Component 2: Component 3: Component 3: Component 1: Component 2: Component 5: Component 1: Component 11: Component 4: Component 2: Component 2: Component 1: Component 2: Component 9: 1 string mismatch"
#> [6] "Component \"grobs\": Component 1: Component 1: Component 1: Component 1: Component 2: Component 3: Component 3: Component 1: Component 2: Component 5: Component 1: Component 17: 1 string mismatch"

[thomasp85]

thanks - yeah, I suspected some change in all.equal() was the root cause

[mcanouil]

For the ones more expert in grid than me, the challenge to solve the issue is to return TRUE with the following code (in R 4.0.0):

library("grid")
x <- grob()
y <- grob()
all.equal(x, y)
#> [1] "Component \"name\": 1 string mismatch"

EDIT:

library("grid")
x <- patchwork:::unname_grob(grob())
y <- patchwork:::unname_grob(grob())
all.equal(x, y)
#> [1] TRUE

Maybe unname_grob needs to be called recursively on the guides.

EDIT2: Following my previous reprex, I think the issue comes from the new unit class used in grid.

# R 4.0.0
x <- unnamed[[1]]$grobs[[1]]$grobs[[1]]$grobs[[2]]$children[[1]]$widths[[2]]
y <- unnamed[[2]]$grobs[[1]]$grobs[[1]]$grobs[[2]]$children[[1]]$widths[[2]]
all.equal(x, y, check.names = FALSE, check.attributes = FALSE)
#> [1] "Component 1: Component 2: Component 1: Component 2: Component 9: 1 string mismatch" 
# this is the location of the component '$name         : chr "GRID.text.509"',  in the sub element of classes "text", "grob" and "gDesc"
class(x)
#> [1] "unit"    "unit_v2"
class(y)
#> [1] "unit"    "unit_v2"

# R 3.6.3
x <- unnamed[[1]]$grobs[[1]]$grobs[[1]]$grobs[[2]]$children[[1]]$widths[[2]]
y <- unnamed[[2]]$grobs[[1]]$grobs[[1]]$grobs[[2]]$children[[1]]$widths[[2]]
all.equal(x, y, check.names = FALSE, check.attributes = FALSE)
#> [1] TRUE
class(x)
#> [1] "unit.arithmetic" "unit"
class(y)
#> [1] "unit.arithmetic" "unit"

[RobinM92]

Is there a workaround for now that we can use?

[mcanouil]

Yes, to manually remove the duplicated legends.

library(patchwork)
library(ggplot2)
p <- ggplot(iris, aes(Sepal.Length, Petal.Length, colour = Species)) +
  geom_point()

p + p + plot_layout(guides = "collect")

(p + theme(legend.position = "none")) + p + plot_layout(guides = "collect")

[RobinM92]

Thanks, logical! To combine this with the legend positions new problems arise of course. E.g.

library(patchwork)
library(ggplot2)
p <- ggplot(iris, aes(Sepal.Length, Petal.Length, colour = Species)) +
  geom_point()

(p + theme(legend.position = "none")) + p + plot_layout(guides = "collect") & theme(legend.position = "bottom")

vs

library(patchwork)
library(ggplot2)
p <- ggplot(iris, aes(Sepal.Length, Petal.Length, colour = Species)) +
  geom_point()

(p + theme(legend.position = "none")) + p + plot_layout(guides = "collect") & theme(legend.position = "bottom")

Both are not ideal, but if you define the position for the first legend and then remove the later ones it works as we want to:

library(patchwork)
library(ggplot2)
p <- ggplot(iris, aes(Sepal.Length, Petal.Length, colour = Species)) +
  geom_point()

(p + plot_layout(guides = "collect") & theme(legend.position = "bottom")) + (p + theme(legend.position = "none")) 

[mcanouil]

Use guides to remove the legend then

library(patchwork)
library(ggplot2)

p <- ggplot(iris, aes(Sepal.Length, Petal.Length, colour = Species)) +
  geom_point()

p_theme <- p + theme(legend.position = "none")
p_guide <- p + guides(colour = "none")

p_theme + p + 
  plot_layout(guides = "collect") & theme(legend.position = "bottom")

p_guide + p + 
  plot_layout(guides = "collect") & theme(legend.position = "bottom")