Allow stat_bin() to compute over single-unique-value data

[yutannihilation]

Fixes #3043.

stat_bin() uses the width of the range of x to calculate the binwidth. So, it fails when x contains only one unique value, because the range is 0-width. Similaly to resolution(), we need to treat zero-ranges specially.

ggplot2/R/utilities-resolution.r

Lines 20 to 22 in 7f13dfa

	resolution <- function(x, zero = TRUE) {
	if (is.integer(x) || zero_range(range(x, na.rm = TRUE)))
	return(1)

As suggested in #3047 (comment), the width of the range should be 0.1, for consistency with the width of the expansion expand_default() gives for the 0-width range.

library(ggplot2)

d <- data.frame(x = rep(1, 100))
ggplot(d, aes(x = x)) +
  geom_histogram()
#> `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

^{Created on 2019-01-03 by the reprex package (v0.2.1)}

[clauswilke]

The default range for scales without data range is only 0.1 in the latest version of ggplot2. Maybe use the same value for default bin width?

library(ggplot2)

ggplot(data.frame(), aes(x = 1, y = 0)) + geom_point()

^{Created on 2018-12-30 by the reprex package (v0.2.1)}

[yutannihilation]

Thanks, I didn't notice that.

I still don't understand the details yet, but that 0.1 seems to come from this expansion, which is under control of Coord, not Scale. So, it's not enough to use 0.1 as the binwidth to plot a single stat_bin()ed point in the same x range as stat_identity() because the expansion will be added to 0.1. Let me think a bit more about what's the right behaviour...

ggplot2/R/coord-.r

Lines 153 to 155 in e9d4e5d

	expand_default <- function(scale, discrete = c(0, 0.6, 0, 0.6), continuous = c(0.05, 0, 0.05, 0)) {
	scale$expand %|W|% if (scale$is_discrete()) discrete else continuous
	}

library(ggplot2)

ggplot(data.frame(), aes(x = 1, y = 0)) +
  geom_point() +
  coord_cartesian(expand = FALSE)

^{Created on 2018-12-31 by the reprex package (v0.2.1)}

Here is the result with width <- 0.1 and the comparison with stat_identity():

library(ggplot2)
library(patchwork)

p1 <- ggplot(data.frame(), aes(x = 1, y = 2)) +
  geom_point() +
  ggtitle("stat_identity()")

d <- data.frame(x = c(1, 1))
p2 <- ggplot(d, aes(x = x, y = stat(count))) +
  geom_point(stat = "bin") +
  ggtitle("stat_bin()")


p1 / p2
#> `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

^{Created on 2018-12-31 by the reprex package (v0.2.1)}

[clauswilke]

I realize that the expansion will be added once more for stat_bin() relative to stat_identity(). I think that's fine.

[yutannihilation]

Now I'm convinced that the binwidth for 0-width range should be the same value as the width of the expansion for 0-width range, 0.1. I added the commits and updated the description.

[yutannihilation]

@clauswilke Could you review this again?

[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/