possible bug when using map_depth() and discard_at()

[gkuljanin]

My team is running into a possible bug in our purrr programming. We have a nested list with information at various depth levels. We manipulate the nested list to discard/keep certain information. We discovered list elements change in an undesired way when discarding list elements at certain list depths.

Consider the following nested list:

library(purrr)

nested_list <- list(
  a_1 = list(
    a_2 = list(
      a_3 = list(
        a_4_0 = matrix(c(2, 2, 3, 3), 2, 2), 
        a_4_1 = list(
          a_5 = 9
        )
      )
    )
  )
)

nested_list
#> $a_1
#> $a_1$a_2
#> $a_1$a_2$a_3
#> $a_1$a_2$a_3$a_4_0
#>      [,1] [,2]
#> [1,]    2    3
#> [2,]    2    3
#> 
#> $a_1$a_2$a_3$a_4_1
#> $a_1$a_2$a_3$a_4_1$a_5
#> [1] 9

Clearly, we see that the a_4_0 element is a matrix. Now, when we discard element a_5, we see an undesired change to a_4_0:

nested_list %>% 
  map_depth(
    .depth = 4, 
    \(ndx) {discard_at(ndx, at = "a_5")}
  )
#> $a_1
#> $a_1$a_2
#> $a_1$a_2$a_3
#> $a_1$a_2$a_3$a_4_0
#> [1] 2 2 3 3
#> 
#> $a_1$a_2$a_3$a_4_1
#> named list()

Somewhere in the discarding process, a_4_0 turns into a vector instead of remaining a matrix. We did not discover the issue at lower depths.

We can convert the vector back into a matrix, but it would be nice if that change did not happen in the first place.

Thank you for the great purrr package!

^{Created on 2025-01-16 with reprex v2.1.0}