where() with NA values

[MyKo101]

The where() function produces relatively uninformative errors when functions output NA values. It would be useful for the user if this was checked.

In this example, the function, ~any(str_detect(.x,",")) returns TRUE and FALSE values, but also returns NA values, and so the below throws an uninformative error:

gss_cat %>%
  select(where(~any(str_detect(.x,","))))
Error: `where()` must be used with functions that return `TRUE` or `FALSE`.

This can be rectified using na.rm=T in the any() function, as below:

gss_cat %>%
  select(where(~any(str_detect(.x,","),na.rm=T)))

I believe that having this throw an error is important, as it draw attention to the NA values, but since all the values are logical, the error should be more informative.

This is also in contrast to the select_if() version of this call:

gss_cat %>%
  select_if(~any(str_detect(.,",")))

which treats NA values the same as FALSE

[NeuronalMike]

I believe this is a limitation with the anonymous function you have used rather than an issue with the where() function .

fruit <- c("apple", "banana", "pear", "pinapple", NA); stringr::str_detect(fruit, "a"); [1] TRUE TRUE TRUE TRUE NA
Open to being wrong, though I do believe its a limit of the stringr::str_detect function. Maybe stringr or string has a better alternative?

[hadley]

Minimal reprex:

library(dplyr, warn.conflicts = FALSE)

df <- data.frame(
  a = c("x", NA), 
  b = c(NA, NA), 
  c = c("y", "y")
)

df |> select(where(~ any(stringr::str_detect(.x, "x"))))
#> Error in `FUN()`:
#> ! `where()` must be used with functions that return `TRUE` or `FALSE`.

^{Created on 2022-08-10 by the reprex package (v2.0.1)}

Interestingly the problem doesn't occur with grepl() because it appears to never match missing strings.