Consolidate scale definition of position aesthetics

[teunbrand]

trafficstars

This PR aims to fix #3342 and fix #4966.

Briefly, all position scales now use the definitions of x- and y-aesthetics from the gglobal_global environment. This ensures that rarer position aesthetics, like xintercept, contribute to the scale limits and are transformed by the scale.

I have been forewarned that this might lead to unexpected issues, but I don't foresee any at the moment. It might lead to some visual changes, but I'd argue they might be intended.

Reprex from #4966:

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

set.seed(1234)
df = tibble(
  x = as.Date(0:19, origin = ymd("2022-01-01")), 
  y = rnorm(20)
)
df |>
  ggplot(aes(x, y)) + 
  geom_point() + 
  geom_hline(yintercept = 5) +                 
  geom_vline(xintercept = ymd("2022-01-10")) +
  geom_vline(xintercept = ymd("2022-03-01"))

^{Created on 2024-01-12 with reprex v2.0.2}

[michaelgrund]

Are there any plans to put the solution regarding this issue into the next release? It's really annoying that vertical lines plotted far outside the date limits are not considered.

[teunbrand]

It's really annoying that vertical lines plotted far outside the date limits are not considered.

I totally agree with you there.

We want the next release to be a low-risk hotfix where we mostly fix mistakes from 3.5.0 and not put in something disruptive. This PR has a non-zero risk of breaking the assumption of some packages that depend on ggplot2, so I think it'd be wise to postpone this a little bit further (or run revdepcheck for this PR). Any thoughts @thomasp85?

[thomasp85]

yes, this will be postponed

[thomasp85]

After a lot of digging I found that the addition of "intercept" to the list of positional aesthetics was the cause of the troubles I remembered. Since this is not touched here I think it's safe. Will run revdepcheck regardless

[thomasp85]

Bunch of revue-check failures. I'm pretty sure that most or all are unrelated to this but please sift through them

[teunbrand]

I can see the irony in some of the revdep failures. Basically, people have found workarounds for the problem that date scales don't accept {x/y}intercept aesthetics. However, this PR now breaks those workarounds by making it easier.

For example this now correctly throws the error that it will not accept any non-dates.

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

# Analogous to current workarounds
ggplot(economics, aes(date, unemploy)) +
  geom_line() +
  geom_vline(xintercept = as.numeric(as.Date("1983-01-31")))
#> Error in `transformation$transform()` at ggplot2/R/scale-.R:628:3:
#> ! `transform_date()` works with objects of class <Date> only

Whereas now the solution is to simply drop the as.numeric() workaround.

ggplot(economics, aes(date, unemploy)) +
  geom_line() +
  geom_vline(xintercept = as.Date("1983-01-31"))

^{Created on 2024-05-21 with reprex v2.1.0}

[teunbrand]

TODO: try to catch these situations, convert to numeric with warning

[teunbrand]

Numeric input now throws warnings instead of errors:

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

ggplot(economics, aes(date, unemploy)) +
  geom_line() +
  geom_vline(xintercept = as.numeric(as.Date("1983-01-31")))
#> Warning in scale_x_date(): A <numeric> value was passed to a Date scale.
#> ℹ The value was converted to a <Date> object.

ggplot(economics, aes(as.POSIXct(date), unemploy)) +
  geom_line() +
  geom_vline(xintercept = as.numeric(as.POSIXct("1983-01-31")))
#> Warning in scale_x_datetime(): A <numeric> value was passed to a Datetime scale.
#> ℹ The value was converted to a <POSIXt/POSIXct> object.

^{Created on 2024-06-04 with reprex v2.1.0}