relocate.data.frame() shouldn't rely on vec_unique() retaining names

[DavisVaughan]

Currently, relocate() relies on the fact that vec_unique() keeps names

dplyr/R/relocate.R

Lines 83 to 85 in 8abb54b

	pos <- vec_unique(c(lhs, to_move, rhs))
	out <- .data[pos]
	new_names <- names(pos)

We think we should actually drop names from the result of vec_unique(), since the returned names are reliant on the fact that only the first encountered duplicate value is returned (this is also consistent with unique()). See r-lib/vctrs#1442 and r-lib/vctrs#1532 where we tried to implement this, but found that dplyr relies on this behavior currently.

---

One easy way to fix this is to use vec_unique_loc() to get the locations, and then vec_slice(), which will retain names.

That should work in the immediate term but it may be worth looking at the code to also see why we currently get names off the unique values, and if there is something else we should be doing instead, since that feels unreliable.

[eutwt]

The lhs and rhs in c(lhs, to_move, rhs) are already unique, and neither contain any elements in to_move.

dplyr/R/relocate.R

Lines 80 to 85 in 8abb54b

	lhs <- setdiff(seq2(1, where - 1), to_move)
	rhs <- setdiff(seq2(where + 1, ncol(.data)), to_move)
	pos <- vec_unique(c(lhs, to_move, rhs))
	out <- .data[pos]
	new_names <- names(pos)

This means the only effect of vec_unique is to remove duplicates in to_move. And to_move only contains duplicates when the user supplies the same column twice with different names (if they supply the same column twice with no name or the same name, eval_select would give a unique output).

So, the only thing vec_unique does here is silently prevent users from creating copies of columns with relocate. Maybe there should be a check for this up front with an error that copying columns using relocate is not supported?

df <- tibble(a = 1, b = 2)

#### current behavior
df %>% 
  relocate(new = b, new2 = b)
# # A tibble: 1 × 2
#     new     a
#   <dbl> <dbl>
# 1     2     1

#### no `vec_unique()` i.e. `pos` defined as `pos <- c(lhs, to_move, rhs)`
df %>% 
  relocate(new = b, new2 = b)
# # A tibble: 1 × 3
#     new  new2     a
#   <dbl> <dbl> <dbl>
# 1     2     2     1