`pseudo_log_trans()` doesn't work when mapped to color or fill

[ccsarapas]

Using scales 1.4.0. Reprex:

library(ggplot2)
library(scales)

p <- ggplot(diamonds) +
  geom_point(aes(x, y, color = price))

p_identity <- p +
  scale_color_viridis_b(
    "identity",
    n.breaks = 6, limits = c(1, NA)
  )

p_log10 <- p +
  scale_color_viridis_b(
    "log10",
    transform = log_trans(base = 10), n.breaks = 6, limits = c(1, NA)
  )

p_pseudo_log10 <- p +
  scale_color_viridis_b(
    "pseudo log10",
    transform = pseudo_log_trans(base = 10), n.breaks = 6, limits = c(1, NA)
  )

cowplot::plot_grid(p_identity, p_log10, p_pseudo_log10, ncol = 3)

The colors scale is transformed correctly for log_trans(). I would expect pseudo_log_trans() to look the same as log_trans(), but instead it matches the untransformed scale.

[teunbrand]

The transformation should be mostly irrelevant for binned scales, the only thing that would matter are the breaks.
The log10 breaks have a different default break algorithm than the other two transformations, so that is why you observe this result.
You can use transform = pseudo_log_trans(base = 10), breaks = log_breaks(n = 6) to mimic the log10 scales.

[ccsarapas]

But log_breaks() fails when there are zeroes in the data or limits -- ie, the typical use case for pseudo_log_trans().

p +
  scale_color_viridis_b(
    transform = pseudo_log_trans(base = 10), 
    breaks = log_breaks(n = 6),
    limits = c(0, NA)
  )
# Error in seq.default(min, max, by = by) : 'from' must be a finite number

[teunbrand]

Maybe the algorithm needs to be made robust, or at least throw a more informative error