New warning `missing values or values outside the scale range` with `geom_area(position = position_fill())`

[jbengler]

Dear @teunbrand

I am struggling to wrap my head around this new warning message that occurs with ggplot2 v3.5.1.9000 but not v3.5.1.

This might be connected to #6211 and #6269, which report the same warning message but in different contexts.

Best
Jan

mtcars |>
  ggplot(aes(x = am, y = gear, fill = cyl, group = cyl)) +
  geom_area()

# works

mtcars |>
  ggplot(aes(x = am, y = gear, fill = cyl, group = cyl)) +
  geom_area(position = position_fill())

# Warning message:
# Removed 6 rows containing missing values or values outside the scale range (`geom_area()`).

[teunbrand]

Hi Jan!

I don't think the similarity of the error message is sufficient to implicate #6211 or #6269. I'm reasonably sure that the change you're seeing is because of #6244. The data does contain missing values (note the first y), so these should be warned about but that didn't happen earlier.

library(ggplot2)

p <- mtcars |>
  ggplot(aes(x = am, y = gear, fill = cyl, group = cyl)) +
  geom_area(position = position_fill())

layer_data(p) |> head()
#>      fill      x  y group PANEL align_padding flipped_aes      ymin ymax   xmin
#> 1 #132B43 -0.002 NA     1     1          TRUE       FALSE        NA   NA -0.002
#> 2 #132B43 -0.001  1     1     1         FALSE       FALSE 0.6000000    1 -0.001
#> 3 #132B43  0.000  1     1     1         FALSE       FALSE 0.7000000    1  0.000
#> 4 #132B43  0.001  1     1     1         FALSE       FALSE 0.6999900    1  0.001
#> 5 #132B43  0.999  1     1     1         FALSE       FALSE 0.6923077    1  0.999
#> 6 #132B43  1.000  1     1     1         FALSE       FALSE 0.7142857    1  1.000
#>     xmax colour linewidth linetype alpha
#> 1 -0.002     NA       0.5        1    NA
#> 2 -0.001     NA       0.5        1    NA
#> 3  0.000     NA       0.5        1    NA
#> 4  0.001     NA       0.5        1    NA
#> 5  0.999     NA       0.5        1    NA
#> 6  1.000     NA       0.5        1    NA

^{Created on 2025-02-19 with reprex v2.1.1}

[jbengler]

Hey Teun!

Thank you very much for your prompt reply!

The option na.rm = TRUE solves my problem.

However, I am still not sure where the missing values are actually coming from. Consider the following simplified example.

df <-
  tibble::tribble(
    ~season,   ~family, ~count,
    "Summer",    "Bird",    10,
    "Summer",    "Fish",    12,
    "Winter",    "Bird",     4,
    "Winter",    "Fish",     1
  )

df |>
  ggplot(aes(x = season, y = count, fill = family, group = family)) +
  geom_area()

df |>
  ggplot(aes(x = season, y = count, fill = family, group = family)) +
  geom_area(position = position_fill())

[teunbrand]

So stat = "align" sets the first observation to x = 0.999, y = 0 for both groups. Then position_fill() 'stacks' these zeroes in the lines below:

ggplot2/R/position-stack.R

Lines 224 to 228 in d835cfe

	heights <- c(0, cumsum(y))
	if (fill) {
	heights <- heights / abs(heights[length(heights)])
	}

From the code above we can anticipate a 'divide by zero' error, and 0 / 0 == NaN, so that is where the missing values are coming from. Setting stat = "identity" would fix the problem.

However, it might make sense to avoid the divide by zero error too. I'll take that to be the action prompted by this issue then.

[jbengler]

Sounds great. Thanks again!