RSpec/LetSetup false positive with nested groups

[dgmora]

I realize this is similar to #830, but this one here is happening without shared examples:

# frozen_string_literal: true

describe User, type: :model do
  let!(:user) { create(:user) }

  it 'is valid' do
    expect(user).to be_valid
  end

  context 'with no direct usage' do
    let!(:user) { create(:user, :special) } # <-- offense, but shouldn't!

    it 'creates something' do
      expect(SomeModel.count).to eq(2)
    end
  end
end

The second let declaration is constrained by a higher level let!. The let! is part of the setup, so changing the second let to before { ... } may not work, since the user will already be created and you are now creating a second user. You can't also change the first one, because it is used in another test.

I see that this could be refactored to let(:user) plus before. But it still looks like a false positive to me because is not actually possible to change that second let! to a before block. The first usage of let! is legit. I don't think you should refactor the whole test just to comply with RSpec/LetSetup

---

Expected behavior

If the same let! is declared in a higher level, those on the lower level shouldn't be marked as an offense because it's not possible to determine if you can change it to a before or not.

Actual behavior

The second let reports RSpec/LetSetup

Steps to reproduce the problem

See above

RuboCop RSpec version

$ bundle exec rubocop -V
1.74.0 (using Parser 3.3.7.2, rubocop-ast 1.38.1, analyzing as Ruby 3.3, running on ruby 3.3.6) [arm64-darwin23]
  - rubocop-performance 1.24.0
  - rubocop-factory_bot 2.27.1
  - rubocop-rails 2.30.3
  - rubocop-rspec 3.5.0
  - rubocop-rspec_rails 2.31.0