return function argument as string

[queermath]

Hi all! I'm writing a function that takes some data x and a function aggr_func and I would like to return a list with the aggregated data & the function passed in aggr_func. rlang::expr_text() and rlang::expr(), but while this works outside of the function context it does not work with an argument! Here's a reprex:

library(rlang)
#> Warning: package 'rlang' was built under R version 3.5.3

# expected behavior
expr_text(expr(mean))
#> [1] "mean"
test_func <- function(x, aggr_func) {
    list(
        x = aggr_func(x),
        aggr_func = expr_text(expr(aggr_func))
    )
}

# returns name of argument ("aggr_func") instead of "mean"
test_func(1:10, mean)
#> $x
#> [1] 5.5
#> 
#> $aggr_func
#> [1] "aggr_func"

^{Created on 2019-05-16 by the reprex package (v0.2.1)}

Using rlang::!! doesn't help entirely--it returns what's getting applied to the data and not the text from the argument:

library(rlang)
#> Warning: package 'rlang' was built under R version 3.5.3
test_func <- function(x, aggr_func) {
    list(
        x = aggr_func(x),
        aggr_func = expr_text(expr(!!aggr_func))
    )
}
test_func(1:10, mean)
#> $x
#> [1] 5.5
#> 
#> $aggr_func
#> [1] "function (x, ...) \nUseMethod(\"mean\")"

^{Created on 2019-05-16 by the reprex package (v0.2.1)}

[lionel]

Inside the function, use enexpr() instead of expr(). Whereas expr() captures your code, enexpr() captures the code of the user of your function.

Also I would use rlang::as_label() instead of expr_text() since the latter is for multi-lines strings. The former always returns a single-line string, so is appropriate when you need to turn something into a descriptive label that can be used, for example, for plot axes.

[queermath]

Solved, c/o @hadley:

library(rlang)
#> Warning: package 'rlang' was built under R version 3.5.3
test_func <- function(x, aggr_func) {
    func <- enexpr(aggr_func)
    list(
        x = aggr_func(x),
        aggr_func = func
    )
}
test_func(1:10, mean)
#> $x
#> [1] 5.5
#> 
#> $aggr_func
#> mean

^{Created on 2019-05-16 by the reprex package (v0.2.1)}

[queermath]

thanks! hadley beat you to it on twitter but the other advice is also helpful! new to rlang and it comes up when i least expect it...

[chrisengelhardt]

I think the response by @lionel better answers your OP, as his suggestion returns the bare function name as a string in the list.

[chrisengelhardt]

Something like this:

test_func <- function(x, aggr_func) {
  
  capture_exp <- enexpr(aggr_func)
  
  list(
    x = aggr_func(x),
    aggr_func = as_label(capture_exp)
  )
}

test_func(1:10, mean)

[queermath]

thanks for pointing out the difference! now that i see i could do either, i'm not sure which is preferred...presumably there are memory savings to just storing the string? i don't know that i would ever use the function by pulling it from this data but it might be worthwhile to have...

[chrisengelhardt]

Surprisingly, storing the string seems to be more costly than storing the expression. However, in practice the difference is probably trivial (112 B vs 56 B).

pryr::object_size(expr(mean)) #56 B
pryr::object_size("mean") #112 B

[hadley]

You might also ask yourself what you would want this to return:

test_fun(function(x) sd(x) / mean(x))

Sometimes generalising the problem can make it easy to see what you really want.