Negated named variables used in across() can't be used with .by argument in mutate()/summarise() etc.

[heliconone]

When using negated selections in across() (e.g. !a or !(a:b)) the negated variable(s) can't subsequently be used with the .by argument in dplyr's verbs.

library(dplyr, warn.conflicts = FALSE)

dat <- tibble(
  a = rep(c("a", "b"), 2),
  b = 1:4,
  c = 5:8
)

dat |>
  summarise(across(!a, mean), .by = a)
#> Error in `summarise()`:
#> ℹ In argument: `across(!a, mean)`.
#> Caused by error in `across()`:
#> ! Can't select columns that don't exist.
#> ✖ Column `a` doesn't exist.

^{Created on 2024-03-29 with reprex v2.1.0}

There are many obvious workarounds but I would have imagined the above would work.

[DavisVaughan]

The docs for across() mention:

You can't select grouping columns because they are already automatically handled by the verb

This includes doing negation. You should just assume that across() can't see a because its a grouping variable, and do everything() here instead