Update lintr hook and snippet_generate for lintr hook

[BjarkeHautop]

This changes the snippet_generate('additional-deps-lintr') to also include suggested dependencies, so {lintr} works on vignettes, README, etc. Modifed default hook to include args: [--load_package] for lintr.

These changes prevents the default hooks from automatically failing when calling use_precommit() on a package, since {lintr} would fail if the package had any dependencies (or simply when the package calls its own function from another file).

Also removed an old print about roxygenize that always appeared.

[lorenzwalthert]

Hi @BjarkeHautop, thanks for your contributions. Before you spend time on code changes, let's discuss them in an issue as per the CONTRIBUTING.md. The --load-package is not the default and I'd like to keep it that way. The reason for it is that it won't be compatible with pre-commit.ci since installing the whole dependency tree would most likely trigger a timeout there. Also, I am not sure I'd agree with your statement that --load-package is required for the hooks to work on packages. {styler} does not have it and things seem to work fine.

[BjarkeHautop]

I originally mentioned the motivation here: #648 (I realize that isn’t an issue).

For {styler}, things work because the config excludes all the locations where it would otherwise fail, so it doesn't actually run on the whole package:

exclude: >
          (?x)^(
          inst/.*|
          man/.*|
          tests/.*|
          touchstone/.*|
          vignettes/.*|
          )$

My concern is that a user sees the package, tries the default hooks on their own (non-trivial) package, and lintr-hook will fail with no clear way to address it. So, the user now needs to look at github issues and/or go through documentation to find out why (in particular that happend to me at #647), or they may simply decide not to use the package at all.

That said, the part of the PR in commit 1aa1ff5 (possibly with a small modification) still addresses the issue that adding only hard dependencies via snippet_generate("additional-deps-lintr") is not enough to make lintr-hook work out of the box.

[lorenzwalthert]

Sorry I don’t care if issue or pull request, but I meant that I never endorsed the idea of a change in the default before you started to implement it. I see what you are saying. Another way would be to only include R/ by default for the {lintr} hook in the config template and put a comment there saying for other directories, you need the load package option?

Also, I am happy to accept changes with regard to snippet generation anyways, as long as the default is not changed.

[BjarkeHautop]

No, that won't fix the issue. See e.g. the following two simple examples, which would fail lintr-hook currently (first example equivalent to #647):

#' @importFrom checkmate assert_count
plus_one <- function(n) {
  assert_count(n)
  n + 1
}

Fails lintr-hook because it uses @importFrom pkg without loading the pkg.

Another example of it failing is trying to use a function defined in a different R file. So have this in R/plus_one.R

plus_one <- function(n) {
  n + 1
}

And this in R/plus_two.R:

plus_two <- function(n) {
  plus_one(plus_one(n))
}

It will also fail the lintr-hook.

I guess a possible solution is to avoid doing the object-usage check from lintr if one doesn't want to load the package. So, add this to .lintr

linters: linters_with_defaults(
    object_usage_linter = NULL
  )