Compatability with ggplot2 3.6.0

[teunbrand]

Hi there,

Apologies for not posting an issue first.
The ggplot2 package is planning an update for around May 2025 and a reverse dependency test identified a problem with the feasts package.
The long and short of it is that ggplot2 handles labels differently now than feast's test expect.
This PR removes a test that no longer holds and adapts a small label addition.
You can test the changes yourself with the development version of ggplot2 (pak::pak("tidyverse/ggplot2"))

Best,
Teun

[teunbrand]

The expected label is derived from the global mapping of the plot, while ggplot2 now prioritises layers for default mappings (so the path geoms is the one to provide x/y).

[mitchelloharawild]

No problem, in this case the code is better updated to return + ggplot2::labs(x = "Re(1/root)", y = "Im(1/root)").

Am I correct in understanding that ggplot(aes(x = mpg, y = hp)) will now return default axis labels of x = "x" and y = "y" rather than x = "mpg" and y = "hp"? Since geom_*() in this case is inheriting from the default ggplot aes(), it feels like a regression that the path geom wouldn't produce the same x/y labels.

[teunbrand]

will now return default axis labels

It will be mpg/hp in that case. Essentially it goes over the layers one-by-one and picks up the labels, so the order in which aesthetics are evaluated matters. It ignores the global aesthetics because they only matter when they are inherited by a layer. The relevant issue is tidyverse/ggplot2#5894 if you want to read more.

[mitchelloharawild]

Thanks for the reference! I'm not against the change, and I think the relevant issue is well motivated.

It seems there could be an issue with using annotate() and the new automatic layer-based labels (or I misunderstand their intended interaction -- I haven't read the relevant issue closely yet).

Here's an adapted version of gg_arma() that removes the model and the package NSE handling junk. The default aes(x = Re(1/root), y = Im(1/root), colour = UnitCircle) results in labels labs(x = "x", y = "y", colour = "UnitCircle") for these layers:

... +
  ggplot2::annotate(
    "path", x = cos(seq(0, 2 * pi, length.out = 100)),
    y = sin(seq(0, 2 * pi, length.out = 100))
  ) +
  ggplot2::geom_vline(xintercept = 0) +
  ggplot2::geom_hline(yintercept = 0) +
  geom_point()

Removing annotate() produces the expected labels labs(x = "Re(1/root)", y = "Im(1/root)", colour = "UnitCircle").

Full (not so minimal) MRE ({reprex} isn't working for me at the moment, so apologies for the botched output):

plot_data <- structure(list(
  type = structure(c(1L, 1L, 1L, 1L, 1L, 1L, 1L, 
                     1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 
                     2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L), levels = c("AR roots", "MA roots"), class = "factor"), 
  root = c(0.743346174645867+0.743346174645867i, 
           -0.743346174645867+0.743346174645867i, -0.743346174645867-0.743346174645867i, 
           1.01542975837224-0.272083583726372i, 0.272083583726371+1.01542975837224i, 
           -1.01542975837224+0.272083583726371i, -0.272083583726371-1.01542975837224i, 
           1.01542975837224+0.272083583726372i, -0.272083583726372+1.01542975837224i, 
           -1.01542975837224-0.272083583726372i, 0.272083583726372-1.01542975837224i, 
           0.743346174645867-0.743346174645868i, 3.89255799705759+6.88853197873318e-16i, 
           0.7382028071715+0.7382028071715i, -0.877092143351749+0.506389385068237i, 
           -0.506389385068237-0.877092143351749i, 0.877092143351748-0.506389385068237i, 
           0.506389385068238+0.877092143351749i, -1.00840378774125+0.270200980569754i, 
           -0.27020098056976-1.00840378774126i, 1.01277877013646+2.37053503325252e-14i, 
           -6.48524393002789e-15+1.01277877013648i, -1.01277877013648+1.03769254626384e-15i, 
           0.270200980569773-1.00840378774126i, 1.00840378774127-0.270200980569749i, 
           0.270200980569755+1.00840378774126i, -1.00840378774126-0.270200980569754i, 
           3.1285544130455e-15-1.01277877013649i, 0.877092143351752+0.506389385068235i, 
           -0.50638938506824+0.877092143351747i, -0.877092143351755-0.506389385068242i, 
           0.506389385068258-0.877092143351732i, 1.00840378774125+0.270200980569741i, 
           -0.7382028071715+0.738202807171498i, -0.738202807171499-0.738202807171506i, 
           0.738202807171505-0.738202807171487i, -0.270200980569757+1.00840378774125i, 
           1.71020671745964+2.77960735720925e-15i), 
  UnitCircle = structure(c(2L, 
                           2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 
                           2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 
                           2L, 2L, 2L, 2L, 2L), levels = c("Outside", "Within"), class = "factor")),
  row.names = c(NA, -38L), class = c("tbl_df", "tbl", "data.frame"))


plot_data

library(ggplot2)
ggplot(plot_data, aes(x = Re(1/root), y = Im(1/root),
                      colour = UnitCircle)) +
  ggplot2::annotate(
    "path", x = cos(seq(0, 2 * pi, length.out = 100)),
    y = sin(seq(0, 2 * pi, length.out = 100))
  ) +
  ggplot2::geom_vline(xintercept = 0) +
  ggplot2::geom_hline(yintercept = 0) +
  geom_point() +
  ggplot2::coord_fixed(ratio = 1) +
  facet_grid(cols = vars(type))

[mitchelloharawild]

Looking into annotate() a bit more, since it is essentially setting up a geom_path() I can understand why it is producing x/y labels. However the resulting labels were a bit surprising to me before I looked into the code since annotate() doesn't have a traditional data+mapping syntax like other layer functions. It is probably difficult to implement, but perhaps annotate() layers should be ignored for layer-based labels.

Perhaps it is best if I just add the desired labels into the plot explicitly.

[teunbrand]

I like the suggestion and I've added it to tidyverse/ggplot2#6290

[mitchelloharawild]

Thanks @teunbrand.

Both of these changes I think are better resolved differently, which I've discussed in response to each changed line comment.

[mitchelloharawild]

No problem, in this case the code is better updated to return + ggplot2::labs(x = "Re(1/root)", y = "Im(1/root)").

Am I correct in understanding that ggplot(aes(x = mpg, y = hp)) will now return default axis labels of x = "x" and y = "y" rather than x = "mpg" and y = "hp"? Since geom_*() in this case is inheriting from the default ggplot aes(), it feels like a regression that the path geom wouldn't produce the same x/y labels.

[teunbrand]

Thanks for the swift review! If these changes are suboptimal, you can see this PR as just pointing out some tests that are incompatible with the upcoming ggplot2 release and a request for these to be updated. Feel free to close this PR whenever you feel fit.

[mitchelloharawild]

No problem, we can work on fixing the changes together in this PR. I've fixed the first issue with the S3 dispatch conflict (I'm not sure why this is an issue only now though, it seems unrelated to the ggplot2 changes that I know of... but I haven't tested this in isolation).

The other issue could be an upstream problem with the labeling change, more details in the code comment thread.