How to reduce the legend symbol thickness for a bar chart in ggplot2

Question

Was asked to reduce the legend symbol thickness for a bar chart in ggplot2 (need them so thin that they look like narrow horizontal lines). Here is a simplification of my case:

library(ggplot2)

# Simple bar chart example
g <- ggplot(mpg, aes(class)) +
  geom_bar(aes(fill = drv))
g

# Failed attempt to reduce the thickness of the legend symbol using guides(). 
# I also tried negative values, but that gives errors. 
# However, increasing the size works well. I need the symbols very thin.
g2 <- g + guides(fill = guide_legend(override.aes = list(size = 0.1)))
g2

# Also adjusting with some theme options is not really working for me
# nor what I really need because is also reducing the distance between the labels.
g + theme(legend.key.height = unit(0.1, "mm"))

Perhaps there is no other way around than editing the legend grobs themselves with the functionality of the grid package or do this outside of R, like Inkscape (?).

^{Created on 2019-05-21 by the reprex package (v0.2.1)}

Answer 1

One solution is to change legend shape. This solution is not perfect (too complicated) as you have to add geom_point layer for which you can change shape.

library(ggplot2)
ggplot(mpg, aes(class, fill = drv)) +
  # Remove legend for bar
  geom_bar(show.legend = FALSE) +
  # Add points to a plot, but invisible as size is 0
  # Shape 95 is a thin horizontal line
  geom_point(aes(y = 0, color = drv), size = 0, shape = 95) +
  # Reset size for points from 0 to X
  guides(fill = guide_legend(override.aes = list(size = 10)))

---

Another solution is to add geom_line layer (ie, line is a thin bar):

library(ggplot2)
ggplot(mpg, aes(class, fill = drv)) +
  geom_bar(show.legend = FALSE) +
  geom_line(aes(y = NA, color = drv), size = 2)

Answer 2

Finally, I made use of the creative solution of @PoGibas to add a geom_line and then edit manually the guides. Since the color aesthetic was used by another geom, I had to use other available aesthetics, and linetype was a good candidate. I accepted @PoGibas 's answer, but hopefully the code below adds to the diversity of the solution:

library(ggplot2)
#> Warning: package 'ggplot2' was built under R version 3.5.3
library(scales)  # just for getting the default colors

ggplot(mpg, aes(x = class)) +
  geom_bar(aes(fill = drv),
           show.legend = FALSE)+
  geom_line(aes(y = 0, # y must be provided, so force it to 0 (if forced to NA, it appears in the OY axis)
                linetype = drv)) + # can be any unused aesthetic 
  guides(linetype = guide_legend(override.aes = list(linetype = "solid",
                                                     color = scales::hue_pal()(3),
                                                     size = 1)))

So, can apply the same principle if we 'force' use alpha aesthetic as well and then edit the guides as needed.

  ggplot(mpg, aes(x = class)) +
    geom_bar(aes(fill = drv),
             show.legend = FALSE)+
    geom_line(aes(y = 0,
                  alpha = drv)) +
    guides(alpha = guide_legend(override.aes = list(alpha = 1, # 1 is recycled 3 times here, as size is below as well
                                                    color = scales::hue_pal()(3),
                                                    size = 2)))
#> Warning: Using alpha for a discrete variable is not advised.

^{Created on 2019-05-21 by the reprex package (v0.2.1)}