Tests in pillar fail with the dev version of ggplot2

[krlmlr]

Is this us? Should ggplot2 do things differently? Should we investigate further?

This is with the dev version of ggplot2.

library(pillar)

data <- tibble::tibble(
  x = num((1:10) / 100, fixed_exponent = -3, notation = "eng"),
  y = num((1:10) / 100, scale = 100, label = "%"),
  z = num(10^(-5:4), notation = "si")
)

basic <-
  ggplot2::ggplot(data, ggplot2::aes(x, y)) +
  ggplot2::geom_point() +
  ggplot2::theme_minimal(36)

basic
#> Error in make_title(..., self = self): unused arguments (list(), "x")

log_scale <-
  ggplot2::ggplot(data, ggplot2::aes(x, z)) +
  ggplot2::geom_point() +
  scale_y_num(transform = "log10") +
  ggplot2::theme_minimal(36)

log_scale
#> Error in make_title(..., self = self): unused arguments (list(), "x")

^{Created on 2025-03-13 with reprex v2.1.1}

[teunbrand]

I think this might be a regression due to #6200 and a custom make_title() method in pillar (https://github.com/r-lib/pillar/blob/6a3c8ee066495dfd5543bf5a3ffd61c76897b9dd/R/ggplot2.R#L68).
We might need to try to be backwards compatible in ggplot2.

[krlmlr]

Is there a way for pillar to be forward-compatible instead?

[teunbrand]

In the upcoming version, the name argument will accept functions that receive the default title and return formatted titles. From eyeballing the pillar method, it might be possible to capture the additional formatting in a separate function that can be given to the name argument. If that is feasible, then it might be possible to avoid having to implement a make_title() method. It might be a bit of a pain to bridge different ggplot2 versions though.

EDIT: nevermind, found a simpler solution which I've put in r-lib/pillar#729.