enquo（）函数错误：参数已被求值

Question

I'm trying to write several embedded functions that end up filtering a set of columns ( id_type in filter_data() ) that is defined by a vector ( id_types in load_data() ). Based on the new dplyr programming vignette , I'm trying to use enquo to make my code run, but am getting an error that I don't know how to deal with.

The actual code is more complicated (and actually requires the different functions), but this is the simplest code I can use to replicate the error:

library(dplyr)
library(purrr)

data <- tibble(id_a = c(1,1,2,2,3),
             id_b = 991,
             id_c = c(45,45,45,1,80),
             units_sold = c(21,20,24,4,5))
id_types <- c("id_a", "id_b", "id_c")

load_data <- function(){

  bind_rows(pmap(list(list(data),
                  id_types),
             filter_data))
}

filter_data <- function(df, id_type) {
  quo_id_type <- enquo(id_type)

  filter(df, !!quo_id_type == 1)
}

load_data()

throws the following error:

Error in (function (x, strict = TRUE)  : 
  the argument has already been evaluated
Called from: (function (x, strict = TRUE) 
    {
    caller_env <- parent.frame()
    if (identical(caller_env, globalenv())) {
        stop("must be called in a function")
    }
    if (missing(x)) {
        stop("argument \"x\" is missing")
    }
    .Call(rlang_capturearg, NULL, NULL, pairlist(caller_env, 
        strict), get_env())
})(id_type)

My desired output: I want the code to iterate through all the id_types , filtering df where either id_a == 1 , id_b == 1 , or id_c == 1 (rows 1, 2, and 4 on the original df), and bind those together in a new data frame.

How can I get this code to work without removing the nested functions?

Thank you.

Answer 1

I still don't understand why the original code didn't work, but I did find another solution.

Replace this:

quo_id_type <- enquo(id_type)

filter(df, !!quo_id_type == 1)

With this:

filter_at(df, vars(id_type), all_vars(. == 1))

Answer 2

Quick fix:

If you need the filter_data function, you can switch to this, which is a bit easier and more obvious ( .data refers to your df you pass into filter , and [[ selects the id_type column when id_type is a string):

filter_data <- function(df, id_type) {
  filter(df, .data[[id_type]] == 1)
}

If you can drop the filter_data function altogether, you can go with:

filter_at(data, vars(id_types), any_vars(. == 1))

(Also dropping the bind_rows and pmap calls). Note that you'll need to change all_vars to any_vars , since you're evaluating all of the vars in id_types at once.

Why the original won't work:

In this example, the argument you pass to filter is a string (ie, "id_a"), but even this by itself won't work:

> filter(data, "id_a" == 1)
# A tibble: 0 x 4
# ... with 4 variables: id_a <dbl>, id_b <dbl>, id_c <dbl>,
#   units_sold <dbl>

The filter condition evaluates to FALSE , since the string "id_a" is not equal to the integer 1 , so no rows satisfy the condition, so it returns an empty tibble .

---

The argument you pass to the function is really id_types[[1]] , not "id_a", so that's what you're enquo() ing. You can see that by trying:

f <- function(id_type) enquo(id_type)  
f(id_types[[1]])  
f("id_a")

Which returns:

<quosure: global>
~id_types[[1]]
<quosure: empty>
~"id_a"

---

You can fix that (and print the other problem) by changing filter_data to this (but I wouldn't recommend actually doing this):

filter_data <- function(df, id_type) {
  id_type <- id_type
  quo_id_type <- enquo(id_type)
  print(quo(filter(df, !!quo_id_type == 1)))

  filter(df, !!quo_id_type == 1)
}