`...` of `across()` are evaluated too early?

[lionel-]

Not sure why this doesn't work:

library(dplyr)

tibble(x = tibble(foo = 1)) %>%
  mutate(across(
    everything(),
    mutate,
    foo = foo + 1
  ))
#> Error: Problem with `mutate()` input `..1`.
#> ✖ object 'foo' not found
#> ℹ Input `..1` is `(function (.cols = everything(), .fns = NULL, ..., .names = NULL) ...`.

Ideally across() would work the same way as map():

tibble(x = tibble(foo = 1)) %>%
  map(
    mutate,
    foo = foo + 1
  )
#> $x
#> # A tibble: 1 x 1
#>     foo
#>   <dbl>
#> 1     2

[romainfrancois]

😞 it's again because of the instrumented "top across", e.g. regular across works fine:

library(dplyr, warn.conflicts = FALSE)

tibble(x = tibble(foo = 1)) %>%
  mutate(identity(across(
    everything(),
    mutate,
    foo = foo + 1
  )))
#> # A tibble: 1 x 1
#>   x$foo
#>   <dbl>
#> 1     2

^{Created on 2021-03-11 by the reprex package (v0.3.0)}

[romainfrancois]

Also, should it say ?

Input `..1` is `dplyr:::top_across(...)`

[lionel-]

I think it's referring to the inner mutate that is passed directly to across. Maybe top_across() evaluates (parts of?) the expansion too early and should instead recreate the same calls? Then the error would refer to the symbol mutate instead of the result of evaluating that symbol.

[romainfrancois]

🤔 actually ... are meant to be evaluated early, this however works fine:

library(dplyr, warn.conflicts = FALSE)

tibble(x = tibble(foo = 1)) %>%
  mutate(across(
    everything(),
    ~mutate(.x, foo = foo + 1)
  ))
#> # A tibble: 1 x 1
#>   x$foo
#>   <dbl>
#> 1     2

^{Created on 2021-03-11 by the reprex package (v0.3.0)}

[romainfrancois]

nevermind, it's "just" that top_across() was eagerly evaluating ... with list(...) when it can be more lazy with enquos(...)

[lionel-]

Need to be careful about multiple evaluations though. If the mapped function is not a data masking function, the arguments inside ... must be evaluated only once. Maybe worth adding a unit test for that case.

The best thing is not to capture or collect ... in any way but just pass them on in the calls, and evaluate those calls in the proper environment where the dots are defined. It's also possible to "forward" dots by binding the ... symbol in an environment to the DOTSXP object of another environment. (I don't know if any of this applies here, just laying out the options in case it's useful).