`unnest_longer()` uses a mix of `keep_empty = TRUE` and `keep_empty = FALSE`

[DavisVaughan]

Currently it keeps NULL elements, but drops empty vectors like integer().

It should work like unchop(keep_empty = FALSE) which drops both NULL and integer().

This is heavily related to #1339. I imagine that to be able to add keep_empty to unnest_longer() we will have to resolve this. I think it requires changes to tidyr:::elt_to_long()'s handling of NULL. When keep_empty = FALSE (the default) it should probably use unspecified(0L) and an empty vector for the index if needed. The current behavior looks to be what should happen for keep_empty = TRUE.

Fixing this is probably a small breaking change. But I think it is definitely worth it for package consistency.

library(tidyr)

df <- tibble(
  x = list(NULL, 1L, integer())
)

df
#> # A tibble: 3 × 1
#>   x        
#>   <list>   
#> 1 <NULL>   
#> 2 <int [1]>
#> 3 <int [0]>

# Inconsistent:
# Keeps the `NULL` as `NA`.
# Drops the `integer()`.
unnest_longer(df, x)
#> # A tibble: 2 × 1
#>       x
#>   <int>
#> 1    NA
#> 2     1

# Consistent with our definition of "empty".
# - Drops `NULL`
# - Drops `integer()` (size 0 vectors)
unchop(df, x)
#> # A tibble: 1 × 1
#>       x
#>   <int>
#> 1     1

# - Keeps `NULL` as NA
# - Keeps `integer()` as NA
unchop(df, x, keep_empty = TRUE)
#> # A tibble: 3 × 1
#>       x
#>   <int>
#> 1    NA
#> 2     1
#> 3    NA

^{Created on 2022-06-02 by the reprex package (v2.0.1)}