Stats use default_aes prior to computation

[teunbrand]

This PR aims to fix #3860.

Briefly, it makes a Stat$use_defaults() method that evaluates Stat$default_aes prior to Stat$setup_data().
This is convenient for filling in optional aesthetics, like weight.

[teunbrand]

Also came across this:

ggplot2/R/stat-density-2d.R

Line 124 in d31c051

default_aes = aes(colour = "#3366FF", size = 0.5),

Where it seems weird to me that the stat declares aesthetics for the geom.

[thomasp85]

I think the contract is that default_aes in a Stat means "the default aesthetics to pass on to the geom if not provided by the user. So it is not exactly the same as default_aes in Geom

[teunbrand]

I'm not sure that the contract as you describe it is honoured currently in the CRAN version. Note that the lines are blue, not red.

library(ggplot2)
packageVersion("ggplot2")
#> [1] '3.5.1'

update_stat_defaults(StatDensity2d, aes(colour = "red"))
StatDensity2d$default_aes
#> Aesthetic mapping: 
#> * `colour` -> "red"
#> * `size`   -> 0.5

ggplot(faithful, aes(eruptions, waiting)) +
  stat_density_2d()

With this PR, the updated aesthetic is enforced:

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

update_stat_defaults(StatDensity2d, aes(colour = "red"))

ggplot(faithful, aes(eruptions, waiting)) +
  stat_density_2d()

^{Created on 2024-12-03 with reprex v2.1.1}

[thomasp85]

My guess is that it doesn't work because the geom "wins". This is also as it should be I think. The stats should not define "graphical" defaults such as colour and the defaults aesthetics for stats are primarily there for aesthetics such as width etc

This is also why they can't be resolved prior to computations as the defaults can rely on values computed by the stat