Take ownership of the `mutate()` step in `group_by()`

[lionel-]

To simplify the error message:

mtcars %>%
  group_by(cyl = 1 + "") %>%
  mutate(new = 1 + "")
#> Error in `group_by()` at dplyr/R/mutate.R:156:2:
#> ! Problem adding computed columns.
#> Caused by error in `mutate()`:
#> ! Problem while computing `cyl = 1 + ""`.
#> Caused by error in `1 + ""`:
#> ! non-numeric argument to binary operator

[hadley]

Is there some simple way to do this? Because if it's easy count() should also take responsibility for group_by(), as in #6182.

[lionel-]

To keep the dplyr API user-oriented, we didn't add call arguments to the generics or data frame methods, instead we used a sentinel approach. It works by setting dplyr_local_error_call() before calling other dplyr verbs, then they should borrow the call automatically.

This is still experimental but is used in some places, e.g. in the slice() derivatives like slice_head().

[hadley]

Hmmm, adding dplyr_local_error_call() at the obvious place in count.data.frame() doesn't work, and add_computed_columns() already has what looks like an existing attempt to solve this problem, so I think probably someone else will need to take a look at this.

[lionel-]

Sounds good I'll take a look after the workshop.