Issues with label control when combining `scale_fill_discrete` with `guide_colorsteps`

[Breeze-Hu]

Hi! Happy Merry Christmas in Advance!

I defined a function to alternate the labels in the legend.
However, whether using single or double intervals, the last label (as shown as 0.4 in the figure) always appears. When running the function independently, it produces puzzling results. The generated labels do not match those in the figure.

library(ggplot2)

v <- ggplot(faithfuld, aes(waiting, eruptions, z = density))

v +
  geom_contour_filled(breaks = seq(0, 0.04, length.out = 8)) +
  scale_fill_discrete(
    palette = scales::pal_brewer(palette = "Blues"),
    labels = function(brks) {
      brks[seq_along(brks) %% 2 != 0] <- ""
      return(brks)
    } 
  ) +
  guides(fill = guide_colorsteps(show.limits = TRUE, position = 'bottom')) +
  theme(legend.key.width = unit(40, 'pt'))

a  <- function(brks) {
      brks[seq_along(brks) %% 2 != 0] <- ""
      return(brks)
    }

a(seq(0, 0.04, length.out = 8))

#[1] ""                    "0.00571428571428571" ""                   
#[4] "0.0171428571428571"  ""                    "0.0285714285714286" 
#[7] ""                    "0.04"           

Best wishes,
Hu

[Breeze-Hu]

Oh! guide_colorsteps(show.limits = TRUE) should have affected the results. But why didn't the 0 on the leftmost side appear?

[Breeze-Hu]

When using show.limits = FALSE, the function label processes the vector after removing the two boundary data points. This is reasonable.

library(ggplot2)

v <- ggplot(faithfuld, aes(waiting, eruptions, z = density))

v +
  geom_contour_filled(breaks = seq(0, 0.04, length.out = 7)) +
  scale_fill_discrete(
    palette = scales::pal_brewer(palette = "Blues"),
    labels = function(brks) {
      print(brks)
      brks[seq_along(brks) %% 2 != 0] <- ""
      print('----')
      print(brks)
      return(brks)
    } 
  ) +
  guides(fill = guide_colorsteps(show.limits = FALSE, # show.limits = FALSE
                                 position = 'bottom')) +
  theme(legend.key.width = unit(40, 'pt'))

#> [1] 0.00667 0.01333 0.02000 0.02667 0.03333
#> [1] "----"
#> [1] ""        "0.01333" ""        "0.02667" ""

^{Created on 2025-12-25 with reprex v2.1.1}

---

However, after setting show.limits = TRUE, the boundary data is fed into the function for processing. This results in the permanent loss of left boundary data (for this particular function in the case study).

library(ggplot2)

v <- ggplot(faithfuld, aes(waiting, eruptions, z = density))

v +
  geom_contour_filled(breaks = seq(0, 0.04, length.out = 7)) +
  scale_fill_discrete(
    palette = scales::pal_brewer(palette = "Blues"),
    labels = function(brks) {
      print(brks)
      brks[seq_along(brks) %% 2 != 0] <- ""
      print('----')
      print(brks)
      return(brks)
    } 
  ) +
  guides(fill = guide_colorsteps(show.limits = TRUE, position = 'bottom')) +
  theme(legend.key.width = unit(40, 'pt'))

#> [1] 0.00667 0.01333 0.02000 0.02667 0.03333
#> [1] "----"
#> [1] ""        "0.01333" ""        "0.02667" ""       
#> [1] 0.00 0.04
#> [1] "----"
#> [1] ""     "0.04"

^{Created on 2025-12-25 with reprex v2.1.1}

Add conditional logic to the function to prevent issues:

library(ggplot2)

v <- ggplot(faithfuld, aes(waiting, eruptions, z = density))

v +
  geom_contour_filled(breaks = seq(0, 0.04, length.out = 7)) +
  scale_fill_discrete(
    palette = scales::pal_brewer(palette = "Blues"),
    labels = function(brks) {
      print(brks)
      ifelse(length(brks)==2, brks, brks[seq_along(brks) %% 2 != 0] <- "")  # ADD ifeles 
      print('----')
      print(brks)
      return(brks)
    } 
  ) +
  guides(fill = guide_colorsteps(show.limits = TRUE, # show.limits = FALSE
                                 position = 'bottom')) +
  theme(legend.key.width = unit(40, 'pt'))

#> [1] 0.00667 0.01333 0.02000 0.02667 0.03333
#> [1] "----"
#> [1] ""        "0.01333" ""        "0.02667" ""       
#> [1] 0.00 0.04
#> [1] "----"
#> [1] 0.00 0.04

^{Created on 2025-12-25 with reprex v2.1.1}

[Breeze-Hu]

Feel free to close this issue, unless further clarification or discussion is needed. 😀

Happy holidays!

[Breeze-Hu]

AND I also have another confusion: Manually set breaks cannot inherently control the behavior of limits, and the following two outcomes are different. Is it possible to adjust this?

library(ggplot2)
library(rcolors)

set.seed(2026)
data = data.frame(
  y = seq(1:10),
  a = runif(n = 10, min = 0, max = 10),
  b = runif(n = 10, min = -2, max = 0)
) |>
  tidyr::pivot_longer(c(a, b), names_to = 'class', values_to = 'value')

ggplot(data = data, aes(x = class, y = y, fill = value)) +
  geom_tile() +
  scale_fill_stepsn(
    # limits = c(-10, 10),
    breaks = c(-10, -8, -6, -2, 0, 2, 6, 8, 10),
    colours = rcolors::get_color(rcolors$MPL_coolwarm, n = 8) |> rev(),
  )

# Explicitly annotate the limits
ggplot(data = data, aes(x = class, y = y, fill = value)) +
  geom_tile() +
  scale_fill_stepsn(
    limits = c(-10, 10),
    breaks = c(-10, -8, -6, -2, 0, 2, 6, 8, 10),
    colours = rcolors::get_color(rcolors$MPL_coolwarm, n = 8) |> rev(),
  )

^{Created on 2026-01-05 with reprex v2.1.1}

[Breeze-Hu]

Okay, it seems it's not about the relationship between breaks and limits. Rather, it's about how to handle manually set breaks?

[teunbrand]

IIRC there are two separate calls to the labels function, once when formatting the inner breaks and once when formatting the limits. I'm not sure we're committed to formatting the breaks in order, so the approach to take every 2nd break doesn't hold up with how breaks/limits are presented.

The whole computation around breaks for binned scales is gnarly, so I'm hesitant to change anything for fear of introducing a bug elsewhere.

[Breeze-Hu]

Happy New Year!

Yes, I realize this is a highly complex control issue. That's why I've hesitated many times. I believe it's acceptable to maintain the current behavior without making changes, to avoid disrupting too many things. 😀

Feel free to close this issue

[Breeze-Hu]

Would you be able to take a look at this issue? eliocamp/ggnewscale#77
I'm not sure if it's related to the timing of running after_scale() in ggplot2?

[teunbrand]

Feel free to close this issue

Thanks!