Granular .drop argument?

[lionel-]

My first expectation would have been that .drop only applies for specific variables, i.e. in the following a and b would still keep their groups:

data %>% group_by(a, b, .drop = FALSE) %>% group_by(c, .drop = TRUE)

Since creating empty groups is expensive, a more granular drop attribute might be useful.

Originally posted by @lionel- in #4091

[hadley]

I think we'll solve this by allowing the user to supply their own grouping data frame, which we're tracking in #4722