Internals of ggtintshade

ggtintshade uses a bit of a hack to let two aesthetics interact. The default ggplot2 API deliberately keeps aesthetics independent, and for the most part that makes sense. Here, though, we want color (a hue) and tintshade (a lightness) to interact: the tint of each point depends on both its color group and its tintshade value.

The hack has two parts:

  1. A per-layer cache. Each layer gets its own environment, created when the geom is built and carried along on the ggproto object, so there is no shared or global state.
  2. The (ab)use of use_defaults. This is a late step in ggplot_build() where the mapped colors and the mapped tintshade values are finally available side by side. There we compute each point’s tint, recolor the points, and record the final color in the cache keyed by its tintshade value. The guide later reads those colors back when it draws the legend.

This vignette rebuilds the following plot using a stripped-down geom_point_tintshadedemo to show the general mechanism.

library(ggplot2)
library(ggtintshade)

metab_data <- data.frame(
  metab = rep(c("Alanine", "Threonine", "Glycine",
                "Glycine betaine", "Proline betaine", "Carnitine",
                "DMSP", "DMS-Ac", "Isethionate"), 3),
  metab_group = rep(rep(c("Amino acid", "Betaine", "Sulfur"), each = 3), 3),
  tripl = rep(c("A", "B", "C"), each = 9),
  area  = runif(27)
)

ggplot(metab_data) +
  aes(x = tripl, y = area, color = metab_group, tintshade = metab) +
  geom_point_tintshade(size = 4)

The cache, created in the constructor

Initialise the cache in the geom constructor so each layer gets its own — no shared or global state to worry about. The environment is attached to a per-layer ggproto instance.

geom_point_tintshadedemo <- function(mapping = NULL, data = NULL, ..., size = NULL) {
  cache <- new.env(parent = emptyenv())
  cache$lookup <- character(0)

  geom <- ggproto(NULL, GeomPointTintshadeDemo, tintshade_cache = cache)
  layer(
    geom = geom, mapping = mapping, data = data,
    stat = "identity", position = "identity",
    params = list(size = size, ...)
  )
}

Tinting the points in use_defaults

use_defaults() is called (at least) twice during ggplot_build(): once for the panel data and once for the legend key. We only want to touch the panel data, so we skip the legend key. The legend has an .id column, which the panel data does not, so we use that to disambiguate them.

Inside, we rank each point’s tintshade value within its color group, spread those ranks across the tintshade range, and feed the result to colorspace::lighten() (positive lightens, negative darkens). We overwrite data$colour with the tint and record it in the cache under its numeric tintshade value, for the guide to recall later.

# Rank tintshade values within each color group and spread over the range.
local_tint <- function(color, tintshade) {
  ave(tintshade, color, FUN = function(v) {
    ranks <- match(v, sort(unique(v)))
    scales::rescale(ranks, to = range(tintshade))
  })
}

GeomPointTintshadeDemo <- ggproto("GeomPointTintshadeDemo", GeomPoint,
  tintshade_cache = NULL,   # supplied by the constructor
  use_defaults = function(self, data, params = list(), ...) {
    data <- ggproto_parent(GeomPoint, self)$use_defaults(data, params, ...)

    if (!is.null(self$tintshade_cache) && is.null(data$.id) && !is.null(data$tintshadedemo)) {
      t <- local_tint(data$colour, data$tintshadedemo)
      data$colour <- colorspace::lighten(data$colour, 2 * t - 1)   # 0.5 -> no change
      self$tintshade_cache$lookup[sprintf("%.10f", data$tintshadedemo)] <- data$colour
    }
    data
  }
)
GeomPointTintshadeDemo$default_aes$tintshadedemo <- NA   # register the new aesthetic

scale_tintshadedemo_discrete <- function(..., range = c(0.2, 0.8)) {
  discrete_scale("tintshadedemo", palette = function(n) seq(range[1], range[2], length.out = n))
}

With just this, the panel is correctly tinted but the legend is incorrect because we haven’t yet defined the guide so it defaults to the solid black entries.

ggplot(metab_data) +
  aes(x = tripl, y = area, color = metab_group, tintshadedemo = metab) +
  geom_point_tintshadedemo(size = 4)

Reading the cache back in the guide

The guide has two parts. In get_layer_key() we grab the layer’s cache and stash it on the guide’s params. In draw(), which runs after the build has filled the cache, we look up each key’s tintshade value and recolor it. We also redefine the default scale (constructed from the aesthetic name) to use the new guide.

GuideTintshadeDemo <- ggproto("GuideTintshadeDemo", GuideLegend,
  get_layer_key = function(self, params, layers, data, theme = NULL) {
    params <- ggproto_parent(GuideLegend, self)$get_layer_key(params, layers, data, theme)
    params$tintshade_cache <- layers[[1]]$geom$tintshade_cache
    params
  },
  draw = function(self, theme, position = NULL, direction = NULL, params = self$params) {
    keys <- params$decor[[1]]$data
    params$decor[[1]]$data$colour <-
      params$tintshade_cache$lookup[sprintf("%.10f", keys$tintshadedemo)]
    ggproto_parent(GuideLegend, self)$draw(theme, position, direction, params)
  }
)

guide_tintshadedemo <- function(...) {
  new_guide(..., available_aes = "tintshadedemo", super = GuideTintshadeDemo)
}

scale_tintshadedemo_discrete <- function(..., range = c(0.2, 0.8)) {
  discrete_scale("tintshadedemo", palette = function(n) seq(range[1], range[2], length.out = n),
                 guide = guide_tintshadedemo())
}

And we have our expected behavior.

ggplot(metab_data) +
  aes(x = tripl, y = area, color = metab_group, tintshadedemo = metab) +
  geom_point_tintshadedemo(size = 4)

The real package generalizes this in a few ways and adds some additional functionality but the two-part hack is the one shown here.

Vignette last built on 2026-07-15