$get_used() is relative to $current_vars(), not names(.data)

[romainfrancois]

This does not yet deals with #5582 but there still was a little problem that attr(cols, "used") aka mask$get_used() is relative to the accumulated variables, not just names(.data).

Then, is this correct:

library(dplyr, warn.conflicts = FALSE)

iris %>%
  group_by(Species) %>%
  mutate(meanPL = mean(Petal.Length), .keep = "all") %>% 
  group_vars()
#> [1] "Species"

iris %>%
  group_by(Species) %>%
  mutate(meanPL = mean(Petal.Length), .keep = "none") %>% 
  group_vars()
#> [1] "Species"

iris %>%
  group_by(Species) %>%
  mutate(meanPL = mean(Petal.Length), .keep = "used") %>% 
  group_vars()
#> character(0)

iris %>%
  group_by(Species) %>%
  mutate(meanPL = mean(Petal.Length), .keep = "unused") %>% 
  group_vars()
#> [1] "Species"

^{Created on 2020-11-02 by the reprex package (v0.3.0.9001)}

[romainfrancois]

In particular "none" re-injects the grouping variables ...

[hadley]

Can you add a unit test please?