facet_wrap() and facet_grid() with a date variable is broken

[paleolimbot]

The revdeps (#3303) identified that using facet_grid() or facet_wrap() with a date variable is broken. This probably affects other S3 vectors as well.

library(ggplot2)
df <- data.frame(date = as.Date("2019-01-01"), x = 1, y = 1)
p <- ggplot(df, aes(x, y)) + geom_point()
p + facet_grid(vars(date))
#> Error in scale_apply(layer_data, x_vars, "train", SCALE_X, x_scales):
p + facet_wrap(vars(date))
#> Error in scale_apply(layer_data, x_vars, "train", SCALE_X, x_scales):

The error is in Facet(Grid|Wrap)$map_data(), where there is a join between the layout and the panels. In one of these, date is a factor, and in the other, it is an S3 Date. This changed because of different behaviour between join_keys() and its previous plyr counterpart.

[clauswilke]

@yutannihilation Do you have time to look into this issue?

[yutannihilation]

Sure, will do.

[yutannihilation]

Quickly confirmed #3327 fixes this issue. Sorry, I'm afraid I won't be very responsive until the end of this month...

[paleolimbot]

I would be happy to PR a test for this if that is helpful.

[thomasp85]

I don’t think it makes sense to test for some arbitrary ordering as long as we don’t break something in revdeps... we will probably have to actively break this once we move to rbinding through vctrs at is is unlikely to follow the arbitrary rules of plyr, but it is best to break only once

[yutannihilation]

Though ordering is just an implementational detail, I think it makes sense to ensure facetting by Date is not broken.

[thomasp85]

I’m simply weary because the old ordering (which is now replicated) is logically wrong. Setting up a unit test may give it some undeserved credibility

[yutannihilation]

Yeah, I agree that we don't need a test about ordering, but the one about facetting by Date (probably it would not be a unit test).

[yutannihilation]

(But, since I don't have enough time to help right now and it seems not necessary for the release, please ignore my comments. Sorry...)