groupingsets with autonamed j creates extra column for ungrouped set

[Henrik-P]

I stumbled over this when I was rolling up sample size (.N) for different sets of grouping variables, including no grouping.

If using groupingsets with an unnamed expression in j which is "autonamed" (e.g. .N) and including an "ungrouped" grouping set, then an additional variable is created for the ungrouped set.

Some data:

d <- data.table(x = c("a", "a", "b"))
d
#    x
# 1: a
# 2: a
# 3: b

With a "grouped" grouping set only, an unnamed expression with .N in j works: the result is autonamed N:

groupingsets(d, j = .N, by = "x", sets = list("x"))
#    x N
# 1: a 2
# 2: b 1

However, when also including an "ungrouped" grouping set (character()), an additional column is generated for the ungrouped result:

groupingsets(d, j = .N, by = "x", sets = list("x", character()))
#       x  N V1
# 1:    a  2 NA
# 2:    b  1 NA
# 3: <NA> NA  3

With an ungrouped grouping set only, the N column is included despite it was only used for the grouped result in the previous example:

groupingsets(d, j = .N, by = "x", sets = list(character()))
#       x  N V1
# 1: <NA> NA  3

Similar examples with named expressions in j, like j = .(N = .N), works.

Additional (admittedly contrived) attempts with .GRP and .I in j give similar results. Thus, it seems like this behaviour may be related to autonaming.