Fix bug in guide_axis_logticks(negative.small)

[teunbrand]

This PR aims to fix #6121.

Briefly, is ensures that the limits it uses to calculate log-breaks are sorted.
Moreover, it picks a better default value for negative.small when the scale maximum is small.

Some examples based on the issue's regex; first showing that it now doesn't error and doesn't render absurd tick marks:

devtools::load_all("~/packages/ggplot2")
#> ℹ Loading ggplot2

p <- data.frame(
  "Y_val" = c(0, 1e-6, 1e-5, 1e-4),
  "X_val" = c(10, 20, 30, 40)
) |>
  ggplot(aes(x = X_val, y = Y_val)) + 
  geom_point() +
  scale_y_continuous(transform = pseudo_log_trans(sigma = 1e-7, base = 10))

p + guides(y = "axis_logticks")

Second, showing the plot with a reasonable negative.small argument for this plot:

p + guides(y = guide_axis_logticks(negative.small = 1e-7))

Lastly, showing the plot with an inappropriate negative.small argument (which still doesn't throw an error).

p + guides(y = guide_axis_logticks(negative.small = 1))

^{Created on 2024-09-30 with reprex v2.1.1}

[thomasp85]

There is something unnerving about defining a variable in one if block (small), and then using it in another, even though the blocks have the same condition.

To be clear, it is perfectly valid, but I think it is a brittle structure

[thomasp85]

Actually, I'd rather you define small <- NULL and then in the first has_negative block you set it to something. Then in this block, instead of testing has_negative you test !is.null(small)

[teunbrand]

This part of the code that we're commenting on is outdated. It has mostly been replaced by scales::minor_breaks_log.

[teunbrand]

Let's wait on {scales} to release a new version so we can use r-lib/scales#452 for this

[thomasp85]

Actually, I'd rather you define small <- NULL and then in the first has_negative block you set it to something. Then in this block, instead of testing has_negative you test !is.null(small)

[thomasp85]

LGTM