data masking doesn't work in anonymous functions used as mutate(data, y = map(x, function(x){})

[archavan]

I am not sure whether this issue falls under the purview of rlang or dplyr. Happy to post to another repo as appropriate.

Data masking doesn’t seem to be working when I use {{ }} in an anonymous function within a purrr::map() or lapply() call inside of a dplyr::mutate() call. It works when I use a predefined function or in a bare purrr::map() or lapply() call outside of dplyr::mutate(), though. Using the .data pronoun method also shows the same behaviour. Please see the reprex below.

library(tidyverse)
iris_nested <- nest(iris, data = -Species)

This doesn’t work

mutate(iris_nested, newcol = map(data,
                                 function(x) {
                                   select_col <- names(x)[2]
                                   select(x, {{ select_col }})
                                 }))
#> Error in (function (arg) : object 'select_col' not found

This works

select_2 <- function(x) {
  select_col <- names(x)[2]
  select(x, {{ select_col }})
}

mutate(iris_nested, newcol = map(data, select_2))
#> # A tibble: 3 × 3
#>   Species    data              newcol           
#>   <fct>      <list>            <list>           
#> 1 setosa     <tibble [50 × 4]> <tibble [50 × 1]>
#> 2 versicolor <tibble [50 × 4]> <tibble [50 × 1]>
#> 3 virginica  <tibble [50 × 4]> <tibble [50 × 1]>

This works too

map(iris_nested$data,  
    function(x) {
      select_col <- names(x)[2]
      select(x, {{ select_col }})
    }) %>% 
  map(head, 1)
#> [[1]]
#> # A tibble: 1 × 1
#>   Sepal.Width
#>         <dbl>
#> 1         3.5
#> 
#> [[2]]
#> # A tibble: 1 × 1
#>   Sepal.Width
#>         <dbl>
#> 1         3.2
#> 
#> [[3]]
#> # A tibble: 1 × 1
#>   Sepal.Width
#>         <dbl>
#> 1         3.3

^{Created on 2021-12-08 by the reprex package (v2.0.1)}

[lionel-]

This is an injection timing issue. I have plans to fix this for rlang 1.1 (so not the next release). However this fix will be a behaviour change that might cause too much backward incompatibility, so there is no guarantee that this will ever happen. You can track this at the original issue #845.

In the meantime, you can work around by storing your lambda function as a named function outside of the mutate, as in your select_2() example.