Broken support for `#' @importFrom` with backticks not documented

[krlmlr]

roxygen2 7.3.0 fails for several of my packages. The only vaguely related NEWS entry I found is

• @importFrom throws a friendlier error if you try and import a non-existing functions (@MichaelChirico, Check imports when generating the NAMESPACE file #1409).

Is the new behavior intended, is it worth waiting for a fix?

roxygen2 7.3.0

library(roxygen2)

roc_proc_text(namespace_roclet(), "
  #' @importFrom purrr `%>%`
  NULL
  ")
#> ✖ <text>:2: @importFrom Excluding unknown export in from purrr: `` `%>%` ``.
#> [1] "importFrom(purrr,)"

roc_proc_text(namespace_roclet(), "
  #' @importFrom purrr %>%
  NULL
  ")
#> [1] "importFrom(purrr,\"%>%\")"

^{Created on 2024-01-14 with reprex v2.0.2}

roxygen2 7.2.3

library(roxygen2)

roc_proc_text(namespace_roclet(), "
  #' @importFrom purrr `%>%`
  NULL
  ")
#> [1] "importFrom(purrr,`%>%`)"

roc_proc_text(namespace_roclet(), "
  #' @importFrom purrr %>%
  NULL
  ")
#> [1] "importFrom(purrr,\"%>%\")"

^{Created on 2024-01-14 with reprex v2.0.2}

[MichaelChirico]

I confirm from git bisect 38166f0 is the culprit

[dchiu911]

I am also getting similar errors when replacing backticks with quotations

#' @importFrom magrittr "%>%"
#' @importFrom purrr "%||%"

@importFrom Excluding unknown export in from magrittr: `"%>%"`.
@importFrom Excluding unknown export in from purrr: `"%||%"`.

[jdblischak]

Try #' @importFrom magrittr %>%. This should work with roxygen2 7.2.3 and 7.3.0

[klmr]

I think this is also responsible for the ‘box’ build failures: https://github.com/klmr/box/actions/runs/7535202942

(Edit: actually on second look this looks unrelated, but it’s also new in ‘roxygen2’ 7.3.0; I’ll investigate further.)

[MichaelChirico]

@klmr I believe yours is related to this NEWS bullet:

roxygen2/NEWS.md

Lines 16 to 20 in 767ec1b

	* The NAMESPACE roclet now reports if you have S3 methods that are missing
	an `@export` tag. All S3 methods need to be `@export`ed (which confusingly
	really registers the method) even if the generic is not. This avoids rare,
	but hard to debug, problems (#1175). You can suppress the warning with
	`@exportS3Method NULL` (#1550).

[klmr]

@MichaelChirico Thanks, but the warning isn’t the issue I am concerned about, although it is a bit confusing, and in my case incorrect (you do not need to declare methods in NAMESPACE, even in R >3.6, for an internally defined generic as long as it is only used internally). Unfortunately I also can’t silence the warning in a backwards-compatible way: adding #' @exportS3Method NULL causes a different warning on previous versions of ‘roxygen2’. :-(

I am talking about the bug described in #1575.