Does RuboCop "check the AST" on the changes it is doing?

[stdedos]

Inspired by #13499 (comment), and what I know by https://black.readthedocs.io/en/stable/*:

Could RuboCop compare the AST, to avoid creating issues such as when operator precedence is involved, or (implict) block starting which does not require parentheses, but the rewrite does require parentheses to keep semantics^?

I feel like, if that were the case, issues like #13299 would not have happened.

---

*: Ik that black is only a formatter, therefore "maybe" easier than RuboCop. But the { |x| } vs do ... end is also one of style, I'd say in this context.
^: or whichever is the correct Ruby terminology 😅

[dvandersluis]

I was actually looking into doing that after I made that comment but there are reasons that the AST could change without changing operator precedence.

For instance, this test:

rubocop/spec/rubocop/cop/style/block_delimiters_spec.rb

Lines 407 to 426 in 1b237c6

	it 'autocorrects do-end with `ensure` to {} if it is a functional block' do
	expect_offense(<<~RUBY)
	x = map do |a|
	^^ Prefer `{...}` over `do...end` for functional blocks.
	do_something
	ensure
	puts 'oh no'
	end
	RUBY
	expect_correction(<<~RUBY)
	x = map { |a|
	begin
	do_something
	ensure
	puts 'oh no'
	end
	}
	RUBY
	end

The AST changes (a begin node is inserted), but precedence is not affected and the code behaviour doesn't change.

Adding parentheses around a node also inserts a begin node, BTW.

[stdedos]

On very wrong terms, replacing do with { "removes an implicit begin node", which the begin then re-adds explicitly.

In that prism, it looks like a net-0 change.

[dvandersluis]

Sure, but that still means that the AST will have changed.