Error for dirname receiving a non-character input on reperx that nearly crashes R

[billdenney]

When trying to replicate a base R bug (https://bugs.r-project.org/show_bug.cgi?id=16441), I wanted to give a reprex for it. When trying to generate the reprex, there was an error in reprex.

reprex::reprex({
e <- new.env()
attr(e, "e") <- e
object.size(e)
})

That resulted in the following output on the R console.

i Rendering reprex...
Error in dirname(path[!is_missing]) : 
  a character vector argument expected

This is with R 4.1.0 on Windows 10.

[billdenney]

The error appears to be that out is NULL here, and that appears to be a unique output that is crashing R.

reprex/R/reprex_render.R

Lines 94 to 111 in 945d63d

	withr::with_dir(
	path_dir(input),
	out <- tryCatch(
	callr::r(
	function(input) {
	rmarkdown::render(
	input,
	quiet = TRUE, envir = globalenv(), encoding = "UTF-8"
	)
	},
	args = list(input = path_file(input)),
	spinner = is_interactive(),
	stdout = if (is.null(std_file)) NULL else path_file(std_file),
	stderr = if (is.null(std_file)) NULL else path_file(std_file)
	),
	error = function(e) e
	)
	)

[jennybc]

The root problem is that this error (C stack usage ... is too close to the limit) causes the rendering to stop.

For some reason, the error caught in reprex is NULL, which I was not expecting, which lead to us trying to proceed with a non-existent path.

I think the only thing I can do in reprex is to throw a marginally more informative error.

It is interesting to ask what's special about this error and whether some other package (evaluate? knitr?) or R itself could do something differently to allow it to be knit.

If I surround the tricky code with tryCatch(), it can be reprexed:

e <- new.env()
attr(e, "e") <- e
blah <- tryCatch(
  object.size(e),
  error = function(e) e
)
blah
#> <simpleError: C stack usage  7969228 is too close to the limit>
str(blah)
#> List of 2
#>  $ message: chr "C stack usage  7969228 is too close to the limit"
#>  $ call   : NULL
#>  - attr(*, "class")= chr [1:3] "simpleError" "error" "condition"

^{Created on 2021-11-01 by the reprex package (v2.0.1.9000)}

cc @cderv

[cderv]

Thanks Jenny for the ping. It seems this error (type of error) is not catched the same as other by evaluate

> evaluate::evaluate("stop('a')")
[[1]]
$src
[1] "stop('a')"

attr(,"class")
[1] "source"

[[2]]
<simpleError in eval(expr, envir, enclos): a>

> evaluate::evaluate('e <- new.env()
+                    attr(e, "e") <- e
+                    object.size(e)')
Error: C stack usage  15922884 is too close to the limit

So we should indeed look there.