stat_function() doesn't work without dummy data

[clauswilke]

I would have expected the second and third example below to work, but it doesn't.

library(ggplot2)

# works
ggplot(data.frame(x = c(-5, 5)), aes(x)) + 
  stat_function(fun = dnorm)

# doesn't work
ggplot() + xlim(-5, 5) + 
  stat_function(fun = dnorm)

# doesn't work
ggplot() + 
  stat_function(fun = dnorm, xlim = c(-5, 5))

^{Created on 2020-05-01 by the reprex package (v0.3.0)}

The problem seems to be that stat_function() doesn't return any data in the second and third case, or the data gets somehow deleted. It would be nice if we could solve this before the 3.3.1 release, I assume (hope!) it's a simple fix.

library(ggplot2)

p <- ggplot(data.frame(x = c(-5, 5)), aes(x)) + 
  stat_function(fun = dnorm)
head(layer_data(p))
#>      x            y PANEL group colour size linetype alpha
#> 1 -5.0 1.486720e-06     1    -1  black  0.5        1    NA
#> 2 -4.9 2.438961e-06     1    -1  black  0.5        1    NA
#> 3 -4.8 3.961299e-06     1    -1  black  0.5        1    NA
#> 4 -4.7 6.369825e-06     1    -1  black  0.5        1    NA
#> 5 -4.6 1.014085e-05     1    -1  black  0.5        1    NA
#> 6 -4.5 1.598374e-05     1    -1  black  0.5        1    NA

p <- ggplot() + xlim(-5, 5) + 
  stat_function(fun = dnorm)
layer_data(p)
#> data frame with 0 columns and 0 rows

p <- ggplot() + 
  stat_function(fun = dnorm, xlim = c(-5, 5))
layer_data(p)
#> data frame with 0 columns and 0 rows

^{Created on 2020-05-01 by the reprex package (v0.3.0)}

[clauswilke]

We can work around this issue by adding at least one group explicitly.

library(ggplot2)

# doesn't work
ggplot() + xlim(-5, 5) +
  stat_function(fun = dnorm)

# works
ggplot() + xlim(-5, 5) +
  stat_function(aes(group = 1), fun = dnorm)

^{Created on 2020-05-01 by the reprex package (v0.3.0)}

@thomasp85 Can you see a way to simulate this via setup_data() or default_aes or required_aes? I couldn't get any of these approaches to work. I could reimplement compute_panel() and ignore the check for an empty data frame, but I'd rather not do that.

[clauswilke]

The reason why setup_data() doesn't work is it isn't even called if the input data is empty:

ggplot2/R/layer.r

Lines 271 to 278 in 49aca36

	compute_statistic = function(self, data, layout) {
	if (empty(data))
	return(new_data_frame())
	params <- self$stat$setup_params(data, self$stat_params)
	data <- self$stat$setup_data(data, params)
	self$stat$compute_layer(data, params, layout)
	},

For the same reason, overriding compute_panel() won't do anything.

[clauswilke]

Never mind, I figured it out. PR #3984.

library(ggplot2)

ggplot() + xlim(-5, 5) + 
  geom_function(fun = dnorm)

ggplot() + xlim(-5, 5) + 
  geom_function(fun = dnorm) + 
  scale_y_log10()

ggplot() + 
  geom_function(fun = dnorm, xlim = c(-5, 5))

ggplot() + 
  geom_function(fun = dnorm)

^{Created on 2020-05-02 by the reprex package (v0.3.0)}

And everything else I've fixed previously still works.

library(tidyverse)

data <- tibble(
  x = 1:50,
  y = x*x - 2*x + 500 + 200*rnorm(50),
  a = sample(letters[1:3], 50, replace = TRUE)
)

ggplot(data, aes(x, y)) +
  geom_point(aes(color = "data")) +
  geom_function(aes(color = "function"), fun = ~.x*.x - 2*.x + 500) +
  scale_y_log10() + scale_x_log10()

ggplot(data, aes(x, y, color = a)) +
  geom_point() +
  geom_function(fun = ~.x*.x - 2*.x + 500) +
  scale_y_log10() + facet_wrap(~a)

^{Created on 2020-05-02 by the reprex package (v0.3.0)}