Axis guide for logarithmic ticks

[teunbrand]

This PR aims to fix #5325, in particular through #5325 (comment).

Briefly, it adds a new axis guide, guide_axis_logticks() that draws ticks at log10-spaced positions.

In comparison to annotation_logticks():

• Only base 10 ticks. None of these base 2 or base e ticks. Even ?annotation_logticks says: "These tick marks probably make sense only for base 10.", which I agree with.
• Proper axis, meaning label placement works out naturally and don't have to set clip = "off" when outside. Of course, can do equivalent of annotation_logticks(outside = FALSE) by setting negative tick lengths.
• Takes style from theme
• Has access to scale, so no need to guess whether user has transformed through scale or coord. User only needs to intervene when manually transformed.

In comparison with guide_axis():

• Exact same key is used for the labels. The ticks get their own 'shadow'-key.
• Ticks all use axis.ticks.{position}.{aes} theme setting, so no tick uses the axis.minor.ticks.{position}.{aes} setting.

Why WIP?

• Should we be able to control the style of the minor/5 ticks and smallest/1 ticks seperately from the major/10 ticks? If so, do we handle that in the guide or make new theme settings?
• Should we mark annotation_logticks() as superseded?
• Once/if POC: secondary axis functionality in guide_axis() #5410 gets merged, we may think about providing a custom breaks and/or labels argument.

A bunch of demonstrations; ticks are placed in non-transformed, original data ranges. The label placement shows that guide_axis_logticks() does not intervene with that part of the guide and takes these verbatim from the scale.

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

# Tiny wrapper for demonstrating scale/guide interactions
demo <- function(limits, trans = "identity", guide = "axis_logticks", ...) {
  scales::demo_continuous(limits, trans = trans, guide = guide, ...)
}

demo(c(0.1, 100))
#> scale_x_continuous(trans = trans, guide = guide)

Of course, these ticks start to make more sense with a log transformation:

demo(c(0.1, 100), trans = "log10")
#> scale_x_continuous(trans = trans, guide = guide)

Tick lengths are set in the guide. If not a <unit> they are interpreted as relative to the theme's tick length setting.

demo(
  c(0.1, 100), trans = "log10",
  guide = guide_axis_logticks(long = 5, mid = 2, short = 1)
)
#> scale_x_continuous(trans = trans, guide = guide)

By default, the ticks are placed 1 order of magnitude outside the limits. To clip these to the actual limits rather than the expanded limits, you can set expanded = FALSE.

demo(
  c(0.1, 100), trans = "log10", expand = expansion(add = 0.5),
  guide = guide_axis_logticks(expanded = FALSE)
)
#> scale_x_continuous(trans = trans, guide = guide, expand = expansion(add = 0.5))

When a user has manually transformed their data, the scale doesn't know about this so the ticks would be off. To intervene, one can set prescale_base to calculate appropriate ticks.

demo(log10(c(0.1, 100)), guide = guide_axis_logticks(prescale_base = 10))
#> scale_x_continuous(trans = trans, guide = guide)

Because ticks are placed in original data-space, this works naturally with compound transformations, like the reverse log10 transform.

demo(c(0.1, 100), trans = c("log10", "reverse"))
#> scale_x_continuous(trans = trans, guide = guide)

When the data include 0 or negative numbers, the ticks are mirrored around 0.
The reason for this is as follows:

1. It works out really naturally with pseudo-log (and alike) transformations.
2. A true log transform cannot include 0 or negative numbers, so we're not harming these transformations by mirroring.

demo(c(-100, 100), trans = "pseudo_log")
#> scale_x_continuous(trans = trans, guide = guide)

Of course, some decision then has to be made where near 0 one should stop drawing ticks, as otherwise one would end up with an infinite amount of ticks. The default is 0.1 but the user can control this too.

demo(
  c(-100, 100), trans = "pseudo_log", 
  guide = guide_axis_logticks(negative_small = 1)
)
#> scale_x_continuous(trans = trans, guide = guide)

Just to show that I'm not cheating this mirroring by seeing if the scale name is "pseudo-log", we can do the same with custom transformations:

demo(c(-100, 100), trans = trans_new("", asinh, sinh))
#> scale_x_continuous(trans = trans, guide = guide)

^{Created on 2023-10-28 with reprex v2.0.2}

[teunbrand]

Now has customisation options for short ticks. Long ticks inherit from axis.ticks, middle ticks inherit from axis.minor.ticks and smallest ticks are set in guide_axis_logticks(short_theme) and inherit from the minor ticks. We chose not to define a new theme element for the shortest ticks, as it is only relevant in this axis.

Still unsure if there wouldn't be a better argument name than short_theme.

In example below, you can see that the smallest ticks can be turned off on the x-axis, and the inheritance on the y-axis.

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

ggplot(msleep, aes(bodywt, brainwt)) +
  geom_point(na.rm = TRUE) +
  scale_x_log10(guide = guide_axis_logticks(
    short = 2,
    short_theme = element_blank()
  )) +
  scale_y_log10(guide = guide_axis_logticks(
    short_theme = element_line(linewidth = 2)
  )) +
  theme(
    axis.minor.ticks.y.left = element_line(colour = "red")
  )

^{Created on 2023-11-22 with reprex v2.0.2}

[thomasp85]

LGTM

[teunbrand]

Thanks for the review Thomas!