object_usage_linter ignores library functions

[renkun-ken]

From https://github.com/REditorSupport/vscode-r-lsp/issues/36.

Suppose the following code is stored in test.R:

library(jsonlite)

read <- function(file, ...) {
  read_json(file, ...)
}

Then object_usage_linter seems to ignore library(jsonlite) and mark read_json as unknown which is annoying.

> lintr::lint("test.R")
/Users/ken/Workspaces/test/vscode-r/test/test.R:4:3: warning: no visible global function definition for 'read_json'
  read_json(file, ...)
  ^~~~~~~~~

[tzakharko]

I was about to open an identical issue, so instead I am commenting here. In addition, maybe one should reconsider enabling object_usage_linter as default? If I understand correctly, it executes code during checks and is therefore a security hazard.

[assignUser]

This also happens if it is just in another .R file, it does not happen if the call is directly preceeded by an if without curly braces.
It also only happens within another function, on the top level of an .R file it does not trigger the linter:

[naupaka]

Adding my +1 to this. It looks like the environment in which the code from outside the function definition is evaluated is not preserved when evaluating code that makes up the function definition. If the file could be parsed and the package variables added to the environment used for testing here:
https://github.com/jimhester/lintr/blob/b93765553d3c1194e37be6efa047085de6d06d5a/R/object_usage_linter.R#L22

that might be one way to resolve this? I tried fiddling with it, but I'm not comfortable enough with the structure of the package to know precisely where the leverage point would be.

[AshesITR]

Note that a somewhat hacky work-around is to source the file before linting / manually attach all needed packages.
This of course isn't ideal and doesn't work reliably in non-interactive sessions.

It works because the default environment in non-packages is globalenv(), where the manual changes can be seen by lintr.

[AshesITR]

This has been fixed.