Modifying state for `par()` and environmental variables using withr

I have a legacy function that relies on objects not passed in as input arguments. I cannot change that function.

To document them, I'd like to temporarily change the state in an isolated environment. I'd like to use the withr package for that.

However, I can't quite seem to get this to work the way I want to.

Calling ls(), the variables I'd like to assign temporarily (example_text and example_df) remain in the execution environment.

What do I need to do so they are removed too? I need to also use par() here. Is there a way to do this already or do I need to write a new withr function that combines the cleanup of setting graphical parameters and temporarily defined variables?

# Bad example function in a library that relies on parameters defined in
# global environemnt
bad_function <- function(){
  plot(example_df)
}
par_1 <- par()
# Execute
withr::with_par(
  new = list(oma = c(rep(2, 4)), cex = 0.55),
  code = {
    example_df <- mtcars
    example_text <- "whatever"
    bad_function()
    text(x = 1, y = 2,labels = example_text)
  }
)

par_2 <- par()
# parameters are cleaned up
identical(par_1, par_2)
#> [1] TRUE
# variables `example_df` and `example_text` that I do not want to show up still do
ls()
#> [1] "bad_function" "example_df"   "example_text" "par_1"        "par_2"

^{Created on 2025-07-08 with reprex v2.1.1}

1) Wrap the withr(...) in a local and set the environment of a copy of the bad_function to that of the environment of the local -- just 3 extra lines added to the code in the question and marked by ## plus we have change the text line to mtext to improve the example.

bad_function <- function(){
  plot(example_df)
}

local({ ##
  withr::with_par(
    new = list(oma = c(rep(2, 4)), cex = 0.55),
    code = {
      example_df <- mtcars
      example_text <- "whatever"
      environment(bad_function) <- environment() ##
      bad_function()
      mtext(example_text, line = 5) ##
    }
  )
}) ##

ls()
## [1] "bad_function"

2) proxy environment The code above works for the reproducible example in the question but if in the actual situation the body of bad_function references additional objects in environment(bad_function) as suggested in a comment then use this. It creates a proxy environment e which contains example_df whose parent is environment(bad_function) and resets environment(bad_function) to that.

good2 <- function(example_df, example_text) {
  e <- new.env(parent = environment(bad_function))
  e$example_df <- example_df
  environment(bad_function) <- e
  withr::with_par(
    new = list(oma = c(rep(2, 4)), cex = 0.55),
    code =  { bad_function(); mtext(example_text, line = 5) }
  )
}

good2(mtcars, "whatever")
ls()

3) proto A variation of (2) which is slightly more concise can be developed using the proto package. This creates a proto object (an environment with certain methods) whose parent is environment(bad_function). Inserting a function into it as shown will change the environment of that function to that of the proto object eliminating one step.

library(proto)

good3 <- function(example_df, example_text) {
  p <- proto(environment(bad_function), example_df = example_df,
    bad_function = bad_function)
  withr::with_par(
    new = list(oma = c(rep(2, 4)), cex = 0.55),
    code = { with(p, bad_function)(); mtext(example_text, line = 5) }
  )
}

good3(mtcars, "whatever")
ls()

4) trace A different approach is to to insert example_df into bad_function using trace.

bad_function <- function(){
  plot(example_df)
}
par_1 <- par()

good4 <- function(example_df, example_text) {
  withr::with_par(
    new = list(oma = c(rep(2, 4)), cex = 0.55),
    code = {
      trace(bad_function, bquote(example_df <- .(example_df)), print = FALSE) ##
      bad_function()
      untrace(bad_function) ## 
      mtext(example_text, line = 5)
  })
}

good4(mtcars, "whatever")
ls()