object_usage_linter should respect NAMESPACE file

[grimmjulian]

Description

object_usage_linter() reports no visible global function definition warnings for functions that are imported via the package NAMESPACE (e.g. importFrom() / import()). Those functions are actually available at runtime, so these warnings are false positives for package code.

Reproducible example

In NAMESPACE

importFrom(dplyr,mutate)

In R/file.R

mutate(data, x = x + 1)

leads to:

file.R:1:1: warning: no visible global function definition for 'mutate'

No warning should be emitted because mutate() is imported via NAMESPACE and therefore available to the package code.

Proposed Solution

When running the object_usage_linter in a package project, it should parse the NAMESPACE file for imports and collect all imports from importFrom. This avoids noisy false positives for properly authored packages and makes lintr behave consistently with R CMD check.

Notes:

I am not sure how to handle general package import() in NAMESPACE as the whole package needs to be checked which functions are exported.

[MichaelChirico]

For now you are expected to use pkgload::load_all() before running this linter

[d-morrison]

also happens when internal functions call other internal functions

For now you are expected to use pkgload::load_all() before running this linter

@MichaelChirico is there a different long-term solution in the works?

[grimmjulian]

For now you are expected to use pkgload::load_all() before running this linter

My problem is more when using https://github.com/REditorSupport/languageserver when I get the warnings for each line. Then I cannot use pkgload::load_all() everytime.