Changes in jitter.width of geom_point() since v3.5.1.9000

[jbengler]

Part 1

In this example the jitter is wider since v3.5.1.9000

ggplot(mtcars, aes(x = factor(cyl), y = disp, color = factor(cyl))) +
  geom_point(
    position = position_jitterdodge(
      jitter.width = 0.4, jitter.height = 0, dodge.width = 0, seed = 42
      )
    )

Output of ggplot2 v3.5.1

Output of ggplot2 v3.5.1.9000

Part 2

Now, it gets even more interesting. If the grouping viable is discrete, but not a factor, the jitter is more narrow since v3.5.1.9000

mtcars %>%
  mutate(new_var = c(rep("A", 11), rep("B", 11), rep("C", 10))) %>%
  ggplot(aes(x = new_var, y = disp, color = new_var)) +
  geom_point(
    position = position_jitterdodge(
      jitter.width = 0.4, jitter.height = 0, dodge.width = 0, seed = 42
    )
  )

Output of ggplot2 v3.5.1

Output of ggplot2 v3.5.1.9000

Conclusion

I am not sure what to make of this. I guess this might probably not hurt too much. However, in tidyplots it triggers a lot of vdiffr failures.

[teunbrand]

Thanks for the report Jan!

The relevant PR that affected this behaviour is #5922.
In the dev version, the jitter is equally wide regardless of whether the grouping is a factor or character.
The behaviour in 3.5.1, where you get unequal jitter widths based on factor/character, is undesirable.

it triggers a lot of vdiffr failures

I'm sure there will be other changes that will trigger small changes as well. vdiffr snapshots aren't checked on CRAN, so it won't be an immediate problem.