Support nested \donttest \dontrun

[lorenzwalthert]

style_text("
#' Remove the local CITES database
#'
#' @examples
#' \\donttest{
#' \\dontrun{
#' cites_db_delete()
#' }
#' }
cites_db_delete <- function() {
  for (t in dbListTables(cites_db())) {
    dbRemoveTable(cites_db(), t)
  }
  update_cites_pane()
}"
)
#> Error in if (from > to) {: missing value where TRUE/FALSE needed

^{Created on 2019-04-26 by the reprex package (v0.2.1)}

Taken from ropensci-archive/citesdb#3. Also note that these don't need to be completely nested to pass R CMD check:

#' test
#' @examples
#' \dontrun{\donttest{
#' 2
#' }
#' 1
#' }
test <- function()
  NULL

Approaches (simple to hard):

• catch error in roxygen and still style rest of input file.
• Fail more informatively on whole file.
• support nested but only where all R expressions are in most inner loop.
• support general case where donts and r expressions can occur anywhere.

[lorenzwalthert]

Seems to be seen as valid code by roygen ans R CMD CHECK, hence we should support it. Reference: https://stackoverflow.com/questions/1454211/what-does-not-run-mean-in-r-help-pages

[ycphs]

I had the same problem. The return was meaningless:

When processing R/style_wrappers.R: <text>:15:1: unexpected input
14: 
15: \

If an error occurs it shouldn't be generic return.

I would prefer to get the following information:

• location of the error
• is it part of the actual code
• is it part of the roxygen structure
• prompt continue to the end of the file with skipping the relevant part of the code (y/n)

[lorenzwalthert]

I don't think we are going to support styling this soon. It's pretty complicated to solve this in an elegant way and cover all cases I think and also might slow styler for roxygen2 sections that don't have nested `\dontrun' and friends. I agree we should fail more informatively. Before parsing roxygen, we could check if there are nested statements and if so, print the code (because exact line information will be harder) in a warning and proceed, so people can search their code for the pattern.

[IndrajeetPatil]

Since CRAN implemented a new check for excessive imports, I have been seeing a new trend among developers where they conditionally use the needed packages and add them to Suggests, and not Imports.
This means there are increasing number of instances where people are using either nested \donttest \dontrun or something along those lines.

So, yes, the need for this feature is increasing, and I am looking forward to this feature eventually being supported by styler :)

[lorenzwalthert]

@IndrajeetPatil any interest in spending some time to quantify how frequent this is? I have no feeling for this. E.g. what share of packages do use nested donts? Of which form are they (seee above), e.g. all R expression nested with same depth or not? E.g. could draw random sample from CRAN and investigate... Because the most general solution looks quite hard, but maybe it's not even necessary.

Also, I updated approaches in the initial post.

[IndrajeetPatil]

@lorenzwalthert Sure! Will do some digging on a random sample and report back the results.

[yjunechoe]

Hi! Just wanted to bump this with a similar issue for nesting \dontshow{} in \donttest{}. This is common when examples have setup code you want to hide (API auth, establishing connection to a background python/julia/etc. session) that also take too long (beyond time allotted for examples by CRAN).

Examples in a package of mine are along the lines of:

#' @examples
#' \donttest{
#' \dontshow{
#' # Setup code (ex: establishing connection)
#' }
#' # Actual example
#' \dontshow{
#' # Cleanup code (ex: closing connection)
#' }
#' }

I understand this is non trivial to handle, but if there are any workarounds that I can use in the meantime I'd really appreciate it!

P.S. - Github code search is now a thing, so here's a quick search of nested \dont*{}s (not too common but not extremely rare either): https://github.com/search?q=language%3Ar+%2F%5C%5Cdonttest%5C%7B%5Cs*%23%27%5Cs*%5C%5Cdont%28show%7Crun%29%2F+path%3AR%2F&type=code