Update to lintr needed?

[joelnitta]

Before filing a bug

• I have installed the latest dev version of {precommit} with remotes::install_github('lorenzwalthert/precommit')
• I have installed the latest hook revisions (update with precommit::autoupdate())
• I have installed the latest release of the upstream Python framework pre-comit as described under the update instructions.

Describe the bug

Internal (non-exported) functions that are defined in a separate file from an exported function (e.g., functions in a utils.R file) trigger the lintr "[object_usage_linter] no visible global function definition" warning. I do not encounter this behavior when running lintr myself with lintr v3.0.2. It seems {precommit} is running v3.0.1 so I suspect that may be causing the problem:

precommit/renv.lock

Line 274 in 428e0f9

"Version": "3.0.1",

To Reproduce
Steps to reproduce the behavior:

1. Create utils.R with a non-exported function
2. Call that function from another file (e.g., exported.R)
3. Try to commit exported.R

Expected behavior
No warning from the lintr

Additional context

• My operating system is: Mac OSX
• My project does not use renv

[lorenzwalthert]

Thanks @joelnitta. The difference is probably due to the way {lintr} is called. We don't your whole source package prior to calling {lintr}, but we only lint the files staged for commit. This is arguably different from what you do when you call {lintr} interactively. Please see #440.

[joelnitta]

@lorenzwalthert it certainly looks like you are right - I think I've encountered the same thing as #440. There's one thing I don't understand though: It looks like {precommit} calls lintr::lint(path), but when I run lintr::lint("path_to_R_script") I don't get any [object_usage_linter] warnings like I do from {precommit}.

[lorenzwalthert]

Ok. Make sure to not devtools::load_all() before the {lintr} call. But have you tried also outside RStudio like

Rscript -e "lintr::lint(path)" 

With the same result? I think maybe {lintr} runs devtools::load_all() in RStudio or something like this. cc: @russHyde.

[joelnitta]

Ah, that does make the difference. I'm not using RStudio (I use vs-code), but if I run lintr::lint("path") in a fresh R session I get the object_usage_linter warnings; if I run devtools::load_all() first, no warnings. So yes this is the same thing. Closing now since it's a duplicate of #440.