Using NULL and NA in pl$DataFrame() shouldn't be identical

[etiennebacher]

I don't know if pl$DataFrame(x = NULL) should generate a DataFrame of height 1 to begin with.

length(NULL)
#> [1] 0

arrow::arrow_table(x = NULL)
#> Table
#> 0 rows x 1 columns
#> $x <null>

^{Created on 2024-09-02 with reprex v2.1.1}

In my opinion, since NULL is zero length, it makes more sense to generate a DataFrame with zero height like arrow::arrow_table() does.

(On next branch, this pattern generates a DataFrame with height 0.)

neopolars::pl$DataFrame(x = NULL)
#> shape: (0, 1)
#> ┌──────┐
#> │ x    │
#> │ ---  │
#> │ null │
#> ╞══════╡
#> └──────┘

^{Created on 2024-09-02 with reprex v2.1.1}

Originally posted by @eitsupi in #1216 (comment)

[eitsupi]

In looking into this, I found that arrow and nanoarrow map the null type of Apache Arrow to vctrs::unspecified.
This seems reasonable to me, and I have achieved the same conversion in the next branch.

arrow::arrow_table(x = vctrs::unspecified(10))
#> Table
#> 10 rows x 1 columns
#> $x <null>
#>
#> See $metadata for additional Schema metadata

arrow::arrow_table(x = vctrs::unspecified(10)) |> tibble::as_tibble()
#> # A tibble: 10 × 1
#>        x
#>    <???>
#>  1     .
#>  2     .
#>  3     .
#>  4     .
#>  5     .
#>  6     .
#>  7     .
#>  8     .
#>  9     .
#> 10     .

neopolars::pl$DataFrame(x = vctrs::unspecified(10))
#> shape: (10, 1)
#> ┌──────┐
#> │ x    │
#> │ ---  │
#> │ null │
#> ╞══════╡
#> │ null │
#> │ null │
#> │ null │
#> │ null │
#> │ null │
#> │ null │
#> │ null │
#> │ null │
#> │ null │
#> │ null │
#> └──────┘

neopolars::pl$DataFrame(x = vctrs::unspecified(10)) |> tibble::as_tibble()
#> # A tibble: 10 × 1
#>        x
#>    <???>
#>  1     .
#>  2     .
#>  3     .
#>  4     .
#>  5     .
#>  6     .
#>  7     .
#>  8     .
#>  9     .
#> 10     .

^{Created on 2024-09-03 with reprex v2.1.1}