add support for Not with multiple positional indices

[bkamins]

No description provided.

[bkamins]

@nalimilan - when doing a final check before merging I realized that:

select(df, Not(1, 1])

will error, while

select(df, Not(:a, :a])

will work. The reason is that former is translated to Not([1, 1]) which is disallowed (duplicate column selected), while the latter to Not(Cols(:a, :a)) which is OK, as Cols allows duplicates (and deduplicates them).

I have documented the difference, but I do not fully like that we have such a corner case when we have the difference (however, it seems that it is best what we can have so I have not proposed a change as I do not see any improvement here). What do you think?

[nalimilan]

Ah. Can't we avoid using Cols implicitly?

[bkamins]

I think it is the opposite. We would want to use Cols, but when integers are passed we cannot. However, now I see that what we can (and should) do is to run unique in case of integers. The reason is that Not-negation drops duplicates. I will push an update to the PR soon.

The current behavior is:

julia> using DataFrames

julia> df = DataFrame(x=1:3, y=2:4, z=3:5)
3×3 DataFrame
 Row │ x      y      z
     │ Int64  Int64  Int64
─────┼─────────────────────
   1 │     1      2      3
   2 │     2      3      4
   3 │     3      4      5

julia> df[Not(2, 2), :]
2×3 DataFrame
 Row │ x      y      z
     │ Int64  Int64  Int64
─────┼─────────────────────
   1 │     1      2      3
   2 │     3      4      5

julia> df[:, Not(2, 2)]
ERROR: ArgumentError: Elements of [2, 2] must be unique

and it is inconsistent. df[:, Not(2, 2)] should just drop second column and all is good then.

[bkamins]

Thank you!