A rowwise summary problem where we don't want to use `rowwise()`

miles.md

library(tidyverse)

(df <- tibble(
  id = 1:3,
  wut = letters[3:1],
  x1 = 1:3,
  x2 = 4:6,
  x3 = 7:9
))
#> # A tibble: 3 x 5
#>      id wut      x1    x2    x3
#>   <int> <chr> <int> <int> <int>
#> 1     1 c         1     4     7
#> 2     2 b         2     5     8
#> 3     3 a         3     6     9

col_prod <- function(...) {
  args <- list(...)
  Reduce(`*`, args)
}

df %>%
  mutate(y = pmap_dbl(select(., starts_with("x")), col_prod))
#> # A tibble: 3 x 6
#>      id wut      x1    x2    x3     y
#>   <int> <chr> <int> <int> <int> <dbl>
#> 1     1 c         1     4     7    28
#> 2     2 b         2     5     8    80
#> 3     3 a         3     6     9   162

df %>%
  mutate(y = pmap_dbl(across(starts_with("x")), col_prod))
#> # A tibble: 3 x 6
#>      id wut      x1    x2    x3     y
#>   <int> <chr> <int> <int> <int> <dbl>
#> 1     1 c         1     4     7    28
#> 2     2 b         2     5     8    80
#> 3     3 a         3     6     9   162

^{Created on 2021-02-24 by the reprex package (v1.0.0.9002)}

Wouldn't it be safer to use across(...) instead if select (.,...)?

Yeah that's cleaner (added above). I'm not sure I understand the safety concern.

Also not sure that this is meaningfully different than a rowwise() solution (but I think we've only had partial information about the context).

It's safer because it "respects" previous changes in mutate.

For example, I've added a transformation on x1, but only the across solution "caught" this in the computation of y:

library(tidyverse)

(df <- tibble(
  id = 1:3,
  wut = letters[3:1],
  x1 = 1:3,
  x2 = 4:6,
  x3 = 7:9
))
#> # A tibble: 3 x 5
#>      id wut      x1    x2    x3
#>   <int> <chr> <int> <int> <int>
#> 1     1 c         1     4     7
#> 2     2 b         2     5     8
#> 3     3 a         3     6     9

col_prod <- function(...) {
  args <- list(...)
  Reduce(`*`, args)
}

df %>%
  mutate(x1 = exp(x1),
         y = pmap_dbl(select(., starts_with("x")), col_prod))
#> # A tibble: 3 x 6
#>      id wut      x1    x2    x3     y
#>   <int> <chr> <dbl> <int> <int> <dbl>
#> 1     1 c      2.72     4     7    28
#> 2     2 b      7.39     5     8    80
#> 3     3 a     20.1      6     9   162

df %>%
  mutate(x1 = exp(x1),
         y = pmap_dbl(across(starts_with("x")), col_prod))
#> # A tibble: 3 x 6
#>      id wut      x1    x2    x3      y
#>   <int> <chr> <dbl> <int> <int>  <dbl>
#> 1     1 c      2.72     4     7   76.1
#> 2     2 b      7.39     5     8  296. 
#> 3     3 a     20.1      6     9 1085.

^{Created on 2021-02-25 by the reprex package (v1.0.0)}

It's safer because it "respects" previous changes in mutate

Yeah, true.