Meta-programming data-mask column aliases

tidyverse
,
1

I'm playing around with some code for a data frame in which a particular "role" ("foo" below) has been defined to refer to a specific column (bar) at construction time:

set_foo_col <- \(x, col) `attr<-`(x, "foo", col)
get_foo_col <- \(x) attr(x, "foo")

data <- tibble::tibble(bar = 1:10) |>
  set_foo_col("bar")

I wanted to save myself from having to write a bunch of code like dplyr::mutate(x, baz = .data[[get_foo_col(x)]]), instead hoping I could do some meta-programming shenanigans to use .foo as an alias in data-mask contexts (i.e. dplyr::mutate(x, baz = .foo) instead).

This naïve solution fails, but expresses my intent, I hope:

my_mutate <- function(.data, ...) {
  data <- rlang::enquo(.data)
  dots <- rlang::enquos(...)
  expr <- rlang::expr(dplyr::mutate(!!data, !!!dots))
  env <- rlang::env(
    rlang::caller_env(),
    .foo = rlang::sym(get_foo_col(.data)) # This is wrong, but what to do instead?
  )
  rlang::eval_tidy(expr, env = env)
}

This fails with:

data |> my_mutate(baz = .foo)
#> Error in `dplyr::mutate()`:
#> ℹ In argument: `baz = .foo`.
#> Caused by error:
#> ! object '.foo' not found

Created on 2024-07-18 with reprex v2.1.1

Some things I have tried that didn't work.
  • I can create a data mask from scratch (with rlang::as_data_mask and friends), but I can only figure out how to use this new mask by reimplementing tons of dplyr internals. That is, I don't know how to just "pass" it to the existing dplyr::mutate implementation.
  • I could maybe modify the existing data mask in .data with dplyr:::peek_mask, but that is not publicly exposed, so I'd rather avoid that.
  • I tried delaying execution with rlang::env_bind_active, but cannot find a solution that works in general. In particular, the combination with dplyr::pick seemed attractive, but I cannot quite make it work.
  • Ultimately, I considered just walking the AST in dots to replace .foo leaf symbols with .data[[get_foo_col(x)]], but that feels hacky?

Any pointers on how to approach something like this?