drop_na() incompatible with list-cols

[tiernanmartin]

It seems that drop_na()'s dependency on stats::complete.cases is tripping it up when any list-cols are present:

library(tidyverse) 
library(purrr)

set.seed(1)

# Give mtcars a list-col and some NAs
mtcars_nas <- 
  mtcars %>% 
  rownames_to_column() %>% 
  mutate(LIST_COL = map(mpg, ~sample(LETTERS, nchar(.x)))) %>% 
  mutate(mpg = sample(c(mpg, rep(NA, 32)), n(), replace = TRUE)) %>% 
  select(rowname, mpg, LIST_COL) %>% 
  as_tibble

mtcars_nas
#> # A tibble: 32 × 3
#>              rowname   mpg  LIST_COL
#>                <chr> <dbl>    <list>
#> 1          Mazda RX4    NA <chr [2]>
#> 2      Mazda RX4 Wag  21.4 <chr [2]>
#> 3         Datsun 710  15.0 <chr [4]>
#> 4     Hornet 4 Drive  16.4 <chr [4]>
#> 5  Hornet Sportabout    NA <chr [4]>
#> 6            Valiant  19.7 <chr [4]>
#> 7         Duster 360    NA <chr [4]>
#> 8          Merc 240D  15.2 <chr [4]>
#> 9           Merc 230  10.4 <chr [4]>
#> 10          Merc 280    NA <chr [4]>
#> # ... with 22 more rows

# Try to drop the incomplete rows - fails
mtcars_nas %>% tidyr::drop_na()
#> Error in stats::complete.cases(data): invalid 'type' (list) of argument

Can you add a check to drop_na() to override this behavior?

[jameshowison]

Unsurprisingly I also have this problem when I have NAs in a list column, so I'm commenting to add that possibility to the example above (where the NAs are in other, non-list columns).

[hadley]

NB: I'm pretty sure I worked around this problem in ggplot2, so the code should already exist.