I'm wondering about some edge cases that shouldn't be allowed but might pass through with this change. Please try out this code (I can't right now) and check that it still gives an error.

class A
  @a : Int32
  def initialize
    @a = 5
  end
end

abstract class B < A
  @b : Int32
  def initialize(@b)
  end
end

class C < B
end

Or maybe this is simpler but still checks the same thing (should give an error)

abstract class A
  @b : Int32
end

class B < A
end

(Feel free to use this code, I release it as public domain)

Thank you!

Two notes. First, with this change this compiles fine but shouldn't:

abstract class Base
  @a : Int32
end

class Sub < Base
end

p Sub.new

Second, if an abstract class defines non-nilable instance variables, in my opinion it should provide a constructor for those variables and make subclasses call super. That makes it more clear that this variable needs to be initialized. So I'm not sure this should be fixed (I don't consider this a bug, but a feature).

Thanks, @asterite and @BlaXpirit! It seems like the intended behavior may still be developing, but if this PR is indeed wanted eventually, I will gladly make the changes needed to make sure the edge cases are covered.

Closing because the solution is incorrect.