Closed
Description
When using ggnewscale, legends can disappear if they have the same title and breaks:
library(ggplot2)
library(ggnewscale)
data <- expand.grid(y = 1:4, x = 1:4)
data$z <- c("a", "b")
ggplot(data, aes(x, y)) +
geom_tile(data = ~.x[.x$x == 1, ], aes(fill = z)) +
scale_fill_brewer(palette = 1) +
new_scale_fill() +
geom_tile(data = ~.x[.x$x == 2, ], aes(fill = z)) +
scale_fill_brewer(palette = 2) 
What's going on is that ggplot2 is merging both guides because they have the same title and labels. Changing the name of each scale makes them distinct.
ggplot(data, aes(x, y)) +
geom_tile(data = ~.x[.x$x == 1, ], aes(fill = z)) +
scale_fill_brewer(name = "1", palette = 1) +
new_scale_fill() +
geom_tile(data = ~.x[.x$x == 2, ], aes(fill = z)) +
scale_fill_brewer(name = "2", palette = 2) 
The problem is that guide_train.legend computes the hash using the title, labels, direction and name:
Line 229 in 12e61df
In the first example, all of those characteristics are the same, and the guides are merged. Ideally, the hash should include information on the colours included in key[[aes_column_name]]
This small hack is a small proof of concept of how it would work:
my_guide <- function(...) {
guide <- guide_legend(...)
class(guide) <- c("my_guide", class(guide))
guide
}
guide_train.my_guide <- function (guide, scale, aesthetic = NULL) {
guide <- NextMethod("guide_train")
guide$hash <- with(guide,
digest::digest(list(title,
key$.label, direction, name, key[, 1] # should be `key[[aes_column_name]]`
)))
guide
}
ggplot(data, aes(x, y)) +
geom_tile(data = ~.x[.x$x == 1, ], aes(fill = z)) +
scale_fill_brewer(palette = 1, guide = my_guide()) +
new_scale_fill() +
geom_tile(data = ~.x[.x$x == 2, ], aes(fill = z)) +
scale_fill_brewer(palette = 2, guide = my_guide()) 