`vars_pull()` and `eval_select()` work differently with `all_of()`

[DavisVaughan]

See also tidyverse/dplyr#7720

Alternative solution would be to implement eval_pull() and deprecate vars_pull() #189

b <- "a"

data <- data.frame(a = 1, b = 2) 

tidyselect::vars_pull(
  names(data),
  tidyselect::all_of(b)
)
#> [1] "b"

tidyselect::eval_select(
  rlang::quo(tidyselect::all_of(b)),
  data
)
#> a 
#> 1

This shows up in dplyr's pull() and select() in weird ways

b <- "a"
dplyr::tibble(a = 1, b = 2) |> 
  dplyr::pull(dplyr::all_of(b))
#> [1] 2
dplyr::tibble(a = 1, b = 2) |> 
  dplyr::select(dplyr::all_of(b)) |>
  dplyr::pull()
#> [1] 1

I'm having trouble wrapping my mind around what is happening in the vars_pull() case

It's possible vars_pull() shouldn't allow all_of() or any_of() at all? That function is really for returning exactly 1 name, right? And it's likely that with all_of() and any_of() you can get >1? And it doesn't seem to be working right anyways (i.e. not pulling from the surrounding env)