I've been studying tidyeval semantics from a number of sources, but I'm getting a result I can't explain.
I'm using mutate_at and case_when to transform some variables by (1) retrieving their names using quotation, (2) modifying their names using gsub, and (3) referencing the data associated with the modified names.
In my minimal example, I'm creating foo$c as a transformation of foo$b which is meant to simply take on the value from foo$a. Steps (1) and (2) seem to be straightforward:
library(tidyverse)
library(rlang)
foo <- data.frame(a = 1, b = 2)
foo %>%
mutate_at(vars(c = b),
funs(case_when(
TRUE ~ gsub("b", "a", as_name(quo(.)))
)))
#> a b c
#> 1 1 2 a
foo$c contains the correct name of the variable we want to look at. I understand I need to transform the string into a symbol using sym() and then evaluate it. If I was using a simple mutate(), !! and sym() work fine:
foo %>%
mutate(c := !!sym(gsub("b", "a", as_name(quo(b)))))
#> a b c
#> 1 1 2 1
But when I do this inside the mutate_at(case_when()) I do not get the correct result:
foo %>%
mutate_at(vars(c = b),
funs(case_when(
TRUE ~ !!sym(gsub("b", "a", as_name(quo(.))))
)))
#> a b c
#> 1 1 2 2
To see what's going on I made a simple printing function. Without !! it looks from the printout as though gsub() and sym() are both producing the intended results:
look <- function(x) {
print(x)
print(typeof(x))
return(x)
}
foo %>%
mutate_at(vars(c = b),
funs(case_when(
TRUE ~ look(sym(look(gsub("b", "a", as_name(quo(.))))))
)))
#> [1] "a"
#> [1] "character"
#> a
#> [1] "symbol"
#> Error in mutate_impl(.data, dots): Evaluation error: object of type 'symbol' is not subsettable.
Once I put !! in front, the printout seems to show that we're getting a different result for gsub() and sym():
foo %>%
mutate_at(vars(c = b),
funs(case_when(
TRUE ~ !!(look(sym(look(gsub("b", "a", as_name(quo(.)))))))
)))
#> [1] "."
#> [1] "character"
#> .
#> [1] "symbol"
#> [1] "."
#> [1] "character"
#> .
#> [1] "symbol"
#> a b c
#> 1 1 2 2
I don't understand how adding !! can change the result from the nested sym(gsub()). Adding a new operation to the end shouldn't change the prior/interior result. I've read that !! is "not a function call, but a syntactic operation" but I don't fully appreciate that distinction or how that could change the result.
Using eval_tidy instead of !! seems to work fine, though I can't explain why:
foo %>%
mutate_at(vars(c = b),
funs(case_when(
TRUE ~ eval_tidy(look(sym(look(gsub("b", "a", as_name(quo(.)))))))
)))
#> [1] "a"
#> [1] "character"
#> a
#> [1] "symbol"
#> a b c
#> 1 1 2 1
