I want to filter out components of a function's argument by class, motivated by this question (which looks for a function f
so that f(c(mpg, gear))
is equivalent to select(mtcars, mpg)
).
One way to do so is:
f0 <- \(x){
names_in <- x |>
enquo() |>
quo_get_expr()
mtcars |> select(names_in[[2]]) ## !
}
... but I need to pick the second component of c(mpg, gear)
, because the first is c
. However, a user might want to call f0(mpg)
only, so the first quosure component has to be used instead of the second.
My idea was to Filter
the list of components returned by quo_get_expr
and drop what evaluates to class function
(here: in environment mtcars
). While Map
ping does return the expected list of booleans, Filter
ing actually Reduce
s the list of components to a nested function (treating mpg
etc. as calls):
f1 <- function(x){
names_in <- enquo(x) |> quo_get_expr()
is_no_function <- \(nm) !(nm |> eval(envir = mtcars) |> inherits('function'))
list(map_result = Map(names_in, f = is_no_function),
filter_result = Filter(names_in, f = is_no_function)
)
}
output:
## > f1(c(mpg, gear))
## $map_result
## $map_result[[1]]
## [1] FALSE
##
## $map_result[[2]]
## [1] TRUE
##
## $map_result[[3]]
## [1] TRUE
##
##
## $filter_result
## mpg(gear)
expected output:
## ...
## $filter_result
## [1] mpg gear
While the original question already has an elegant solution with tidyselect
, I'd be grateful for a hint why Filter
ing behaves like it does in this example, and what would be the proper way to proceed with the enquo
output.
I'm not sure the eval
is really needed here. Do you intend to allow other functions like f1(sum(mpg, gear))
? I think you can just more explictly work with the call and strip out the function.
f2 <- function(x){
names_in <- enquo(x) |> quo_get_expr()
if (is.symbol(names_in)) {return(list(names_in))}
stopifnot(is.call(names_in))
names_in <- as.list(names_in)
stopifnot(names_in[[1]]==quote(`c`))
return(names_in[-1])
}
Which returns
f2(c(mpg, gear))
# [[1]]
# mpg
# [[2]]
# gear
f2(mpg)
# [[1]]
# mpg
You get the unexpected result of mpg(gear)
because what's being returned is a call
object. See
x <- f1(c(mpg,gear))
class(x$filter_result)
All call objects begin with a symbol for the function, then a list of their parameters. So if you have a list like
as.call(list(quote(a), quote(b), quote(c)))
# a(b, c)
You can see it interprets the list as a call to the function a
with parameters b
and c