Rectanglers over-eager to git rid of emptiness

[jennybc]

Adapting the Toothless & Dory example, but where I've removed the films metadata for Toothless. When I hoist() or unnest_wider(), films is destined to become a list-column. I think the entry for Dory should be NULL, not an unspecified vector of length 1. Or maybe even an unspecified vector of length 0? Or a character vector of length 0?

library(tidyverse)

df <- tibble(
  character = c("Toothless", "Dory"),
  metadata = list(
    list(
      species = "dragon",
      color = "black"
    ),
    list(
      species = "clownfish",
      color = "blue",
      films = c("Finding Nemo", "Finding Dory")
    )
  )
)
df
#> # A tibble: 2 x 2
#>   character metadata        
#>   <chr>     <list>          
#> 1 Toothless <named list [2]>
#> 2 Dory      <named list [3]>

df %>% 
  hoist(metadata, films = "films")
#> # A tibble: 2 x 3
#>   character films     metadata        
#>   <chr>     <list>    <list>          
#> 1 Toothless <???>     <named list [2]>
#> 2 Dory      <chr [2]> <named list [2]>

(df2 <- df %>% 
  unnest_wider(metadata))
#> # A tibble: 2 x 4
#>   character species   color films    
#>   <chr>     <chr>     <chr> <list>   
#> 1 Toothless dragon    black <???>    
#> 2 Dory      clownfish blue  <chr [2]>
df2$films
#> [[1]]
#> <unspecified> [1]
#> 
#> [[2]]
#> [1] "Finding Nemo" "Finding Dory"
lengths(df2$films)
#> [1] 1 2

^{Created on 2019-11-20 by the reprex package (v0.3.0.9001)}

[DavisVaughan]

Should hoist() and unnest_wider() (and maybe unnest_longer() and unnest_auto()) have a keep_empty argument that they pass on to their corresponding usages of simplify_col? Right now it is hard coded as TRUE. That is where it changes from NULL to unspecified(1).

At least for this one example, flipping that to FALSE retains NULL for the Toothless films element

tidyr/R/rectangle.R

Line 127 in cb51247

keep_empty = TRUE

tidyr/R/rectangle.R

Line 219 in cb51247

keep_empty = TRUE,

Additional note: unnest() also already has a keep_empty argument, and it is defaulted to FALSE.

[jennybc]

I think setting keep_empty = FALSE fixes it for the wrong reason. It happens to bypass the call to init_col(). Plus keep_empty = FALSE is an odd way to express that we want to retain emptiness.

[hadley]

I think there's something wrong with the logic in init_col() — we only want to use unspecified() if the result is going to be a vector. Maybe we need a can_simplify() that we can use earlier in the function?