Infer type annotations for instance variables

[jez]

Input

→ View on sorbet.run

# typed: strict

class A
  extend T::Sig

  X = 0

  sig {void}
  def initialize
    @y = 0
  end
end

Observed output

editor.rb:10: Use of undeclared variable @y https://srb.help/6002
    10 |    @y = 0
            ^^
Errors: 1

Expected behavior

We should infer the annotation for @x because it is a literal type. We should use the same type defaulting rules that we use for constants (as depicted) to also handle instance variables.

---

• This should probably only work for instance variable assignments that happen inside initialize or the top level of a class.
• Ideally we would consolidate the logic that handles constants to handle instance variables.
  To this end, we can probably move the logic to be done in a rewriter pass.
  The rewriter pass would insert a T.let inside assignments to either constants or instance variables when the RHS is a literal.
  (As opposed to now, where it is done in the internals of namer/resolver.)

[amomchilov]

With the advent of #6599 (which infers that X = A.new means X = T.let(A.new, A), would there be any desire to apply the same logic to ivars?

E.g.

#typed: strict

class C
  extend T::Sig

  sig { void }
  def initialize
    @foo = Foo.new # inferred as `T.let(Foo.new, Foo)`
  end  
    
  # Stretch goal: A further refinement of https://github.com/sorbet/sorbet/issues/995
  def memoed_bar
    @bar ||= Bar.new # inferred as `T.let(Bar.new, T.nilable(Bar))`
  end
end

Related: #995