New warning in 'designer' from object_usage_linter

[MichaelChirico]

Haven't confirmed this is a regression yet, but we observe this warning on main that is not present on v3.0.2:

Possible bug in lintr: Couldn't parse usage message ‘<anonymous>: no visible global function definition for 'find_cache_dir'
’. Ignoring 1 usage warnings. Please report an issue at https://github.com/r-lib/lintr/issues.

[MichaelChirico]

Similar warnings also in other packages:

** latrend
Possible bug in lintr: Couldn't parse usage message ‘<anonymous>: no visible binding for global variable 'id'
’. Ignoring 1 usage warnings. Please report an issue at https://github.com/r-lib/lintr/issues.
** mlrCPO
Possible bug in lintr: Couldn't parse usage message ‘<anonymous>: local variable 't' assigned but may not be used
’. Ignoring 1 usage warnings. Please report an issue at https://github.com/r-lib/lintr/issues.
** rextendr
Possible bug in lintr: Couldn't parse usage message ‘<anonymous>: no visible binding for global variable 'wrap__hello_world'
’. Ignoring 1 usage warnings. Please report an issue at https://github.com/r-lib/lintr/issues.

[MichaelChirico]

Here's an MRE for the type of error that's unrecognized:

lintr::lint(linters = lintr::object_usage_linter(), "
foo <- function() bar()
")
Warning message:
In parse_check_usage(fun, known_used_symbols = known_used_symbols,  :
  Possible bug in lintr: Couldn't parse usage message ‘<anonymous>: no visible global function definition for 'bar'
’. Ignoring 1 usage warnings. Please report an issue at https://github.com/r-lib/lintr/issues.

[MichaelChirico]

c.f. the codetools::checkUsage() output:

codetools::checkUsage({
  foo <- function() bar()
})
# <anonymous>: no visible global function definition for ‘bar’

codetools::checkUsage({
  foo <- function() {
    bar()
  }
})
# <anonymous>: no visible global function definition for ‘bar’ (:3)

[AshesITR]

Not really a regression, the old version just kept silent about the problem.

We should try to fix this though. At this point I'm wondering whether it might be worthwhile to fork and fix the issues in codetools directly.

[MichaelChirico]

would you mind having a crack at a fix?

I thought it would be as simple as making this a maybe() but I think that's having an interaction with all the other maybe()s I couldn't figure out quickly:

https://github.com/r-lib/lintr/blob/ed91a2ff3dc31631e93eb98f65bb5779606b6d59/R/object_usage_linter.R#L288

[AshesITR]

I can take a look, do you have MWEs for all the "Couldn't parse usage message" warnings you found?

[AshesITR]

It was surprisingly easy to remedy - you were almost there. Just had to make the message part fit lazily.