Informative error message when "primary pipe input" is missing

[jankowtf]

Just a really minor thing I noticed (and possibly this is rather a magrittr issue?):

While refactoring a function, I accidentally dropped the initial df %>% part of a pipe that used :=.

But before I realized this, I was puzzled for a couple minutes by the error below as it didn't suggest the reason was the missing df %>% but not being able to find function :=.

Maybe one could come up with error messages along the lines of primary pipe input 'x' missing?

library(magrittr)

df <- tibble::tibble(x = 1:3)
col <- rlang::sym("x")

df %>% 
  dplyr::mutate(!!col := !!col * 10)
#> # A tibble: 3 x 1
#>       x
#>   <dbl>
#> 1    10
#> 2    20
#> 3    30

# When accidentally dropping `df %>%`:
  dplyr::mutate(!!col := !!col * 10)
#> Error in `:=`(!!col, !!col * 10): could not find function ":="

[romainfrancois]

Unfortunately, I don't think there is much we can do here, as the error happens before S3 dispatch, i.e. the first argument has to be evaluated before the mutate generic knows which method to call.

[krlmlr]

We can do a tryCatch() in the mutate() generic:

mutate <- function(.data, ...) {
  quo <- enquo(.data)

  tryCatch(
    force(.data),
    error = function(e) {
      abort(paste0("`.data` must be a data source. ", conditionMessage(e)))
    }
  )

  UseMethod("mutate")
}

This doesn't look very helpful right now, perhaps if we had a tryCatch() implementation with better call traces:

dplyr::mutate(!!col := !!col * 10)
#> Error: `.data` must be a data source. could not find function ":="
#> Backtrace:
#>     █
#>  1. └─dplyr::mutate(`:=`(!!col, !!col * 10))
#>  2.   └─base::tryCatch(...) /home/kirill/git/R/dplyr/R/manip.r:442:2
#>  3.     └─base:::tryCatchList(expr, classes, parentenv, handlers)
#>  4.       └─base:::tryCatchOne(expr, names, parentenv, handlers[[1L]])
#>  5.         └─value[[3L]](cond)

^{Created on 2019-01-29 by the reprex package (v0.2.1.9000)}

[romainfrancois]

We'd have to do that for every generic then. Sounds painful.

[hadley]

The cost-benefit tradeoff does not favour adding a tryCatch() here.

[krlmlr]

We do this in ggplot(), and it really helps, especially for beginners. The tryCatch() could be extracted into a helper function, so we only need to remember to call that function (using the new whiskers syntax in rlang) at the beginning of each generic.

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/