Confusing error when forgetting to pipe input (i.e., first arg isn't a dataframe)

[kcarnold]

Students often forget pipes. I wonder if we could make clearer errors in that case. The error for arrange is particularly inscrutable:

library(tidyverse)
diamonds %>% 
  group_by(cut) %>% 
  summarize(n = n())
  arrange(desc(n))
#> Warning in is.na(x): is.na() applied to non-(list or vector) of type 'closure'
#> Error in x[!nas]: object of type 'closure' is not subsettable

^{Created on 2021-09-23 by the reprex package (v2.0.1)}

The other possible errors in this pipeline are also confusing (no applicable method for 'group_by' applied to an object of class "function" and Error: n() must only be used inside dplyr verbs.)

The errors in this particular pipeline could be improved somewhat by adding an informational group_by.default, tweaking the logic around the existing n() error, and some additional type-checking in desc. ggplot has some nice messaging like "Did you use %>% instead of +?".

Vaguely related: #4127

[romainfrancois]

Maybe something like:

arrange <- function(.data, ..., .by_group = FALSE) {
  .data <- impose(.data, "arrange")
  UseMethod("arrange")
}

impose <- function(.data, .fn) {
  withCallingHandlers(.data, error = function(e) {
    abort(c(
      glue("Error when evaluating `{.fn}(.data=)`."),
      i = "This might to be a missing pipe."
    ), parent = e)
  })
}

This'd give something like this:

[master*] 226.48 MB > arrange(desc(n))
Error: 
  Error when evaluating `arrange(.data=)`.
  ℹ This might to be a missing pipe.
Caused by error in `[`: 
  object of type 'closure' is not subsettable
Run `rlang::last_error()` to see where the error occurred.
Warning message:
In is.na(x) : is.na() applied to non-(list or vector) of type 'closure'

but we did rule it out in #4127. Or perhaps this is a variant of UseMethod() that enforces object to be greedy ?

[hadley]

The specific error here is coming from xtfrm:

library(dplyr, warn.conflicts = FALSE)
xtfrm(n)
#> Warning in is.na(x): is.na() applied to non-(list or vector) of type 'closure'
#> Error in x[!nas]: object of type 'closure' is not subsettable

^{Created on 2022-04-16 by the reprex package (v2.0.1)}

We could fix this either by adding some additional type checks to desc() or by maybe switching from xtrfm() to vctrs::vec_proxy_order(), but it doesn't have quite the behaviour we're looking for either:

vctrs::vec_proxy_order(mean)
#> function (x, ...) 
#> UseMethod("mean")
#> <bytecode: 0x7fabf3671100>
#> <environment: namespace:base>

^{Created on 2022-04-16 by the reprex package (v2.0.1)}