label_comma(scale_cut = cut_short_scale()) creates misleading rounding of labels

[mcaselli]

With data ranging to 1.25E6, we get labels of 400K 800K and 1M, which is no good: 1M is labelling 1.2E6.
Should be something like 400K 800K 1.2M, or 0.4M 0.8M 1.2M

library(ggplot2)
library(scales)
library(patchwork)

p <- data.frame(
  category = c("a", "b"),
  value = c(1.25E6, 0.82E6)
) |>
  ggplot(aes(x = category, y = value)) +
  geom_col()

p_1 <- p + labs(title = "default", subtitle = "OK")

p_2 <- p + 
  scale_y_continuous(labels = label_comma()) + 
  labs(title = "label_comma()", subtitle = "OK")

p_3 <- p + 
  scale_y_continuous(labels = label_comma(scale_cut = cut_short_scale())) +
  labs(title = "label_comma(scale_cut = cut_short_scale()", subtitle = "NOT OK")

p_1 + p_2 + p_3

^{Created on 2025-11-10 with reprex v2.1.1}

[joranE]

The behavior here is a result of scale_cut, not label_comma specifically. You'd see the same labelling with label_number.

The "obvious" solution here is to specify accuracy = 0.1, but I sympathize because while that ensures we get 1.2M it also forces the other labels to be 800.0K, which (subjectively) I find unsightly. I have worked around this frequently, but only with fairly laborious manual labelling approaches.