geom_boxplot() with data rows >1 and stat = "identity" fails

[paleolimbot]

In the revdep checks (#3303) at least one plot in an example failed to build that used geom_boxplot(stat = "identity") with more than one data row.

library(ggplot2)
ggplot(
  data.frame(x = "one value", y = 3, value = 4:6),
  aes(x, ymin = 0, lower = 1, middle = y, upper = value, ymax = 10)
) + 
  geom_boxplot(stat = "identity", position = "identity")
#> Error: Elements must equal the number of rows or 1

In 3.1.1, this works as expected (position dodge would make more sense here, but I wanted to make sure this wasn't the issue).

library(ggplot2)
ggplot(
  data.frame(x = "one value", y = 3, value = 4:6),
  aes(x, ymin = 0, lower = 1, middle = y, upper = value, ymax = 10)
) + 
  geom_boxplot(stat = "identity", position = "identity")

[clauswilke]

This is a problem with the group aesthetic. We infer grouping based on the categorical variables, and x has only one value, so all data gets thrown into one group. However, we need three groups here for proper drawing. Not sure why this worked in the past.

library(ggplot2)

# without group aesthetic
ggplot(
  data.frame(x = "one value", y = 3, value = 4:6),
  aes(x, ymin = 0, lower = 1, middle = y, upper = value, ymax = 10)
) + 
  geom_boxplot(stat = "identity")
#> Error: Elements must equal the number of rows or 1

# with group aesthetic
ggplot(
  data.frame(x = "one value", y = 3, value = 4:6),
  aes(x, group = 1:3, ymin = 0, lower = 1, middle = y, upper = value, ymax = 10)
) + 
  geom_boxplot(stat = "identity")

^{Created on 2019-05-08 by the reprex package (v0.2.1)}

The solution is that the package authors need to supply a group aesthetic. I don't think there's anything to be done on our side.

[paleolimbot]

I will contact the developer...there are some other things about that particular plot that could be improved.

On our end, GeomBoxplot$draw_group() implicitly assumes that nrow(data) == 1 without checking it.

ggplot2/R/geom-boxplot.r

Lines 211 to 220 in 92d2777

	whiskers <- new_data_frame(c(
	list(
	x = c(data$x, data$x),
	xend = c(data$x, data$x),
	y = c(data$upper, data$lower),
	yend = c(data$ymax, data$ymin),
	alpha = c(NA_real_, NA_real_)
	),
	common
	), n = 2)

If the previous version of new_data_frame() supported recycling, this probably would have worked with any length data. We could

• Do nothing and hope nobody ever does this again
• Support multiple rows
• Keep only the first row and drop the rest with a warning about setting a group aesthetic
• Error and tell the user to set a group aesthetic

I'd be happy to PR any of those if it's worth doing.

[clauswilke]

A couple of comments, in no particular order:

• This problem will only occur when people don't use geom_boxplot() with its own stat, so it's likely rather rare.
• The correct way to support multiple rows is via groups.
• If you add a test, it needs to have very low overhead, because this test will be run for every single boxplot that is drawn.
• I don't see a good case for taking just the first row and moving on. geom_boxplot() is encountering data in a format that it doesn't understand and that doesn't make sense, so it should error out.

[paleolimbot]

I've contacted the developer of HistDAWass with a suggestion of to fix this problem in their package code.

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/