[mmporong]

Numeric#<=> returns 0 when the argument is the receiver itself and nil for anything else, as the RDoc comment above the signature says ("zero or nil"). The return type was Integer, which leaves out the nil case:

n = Numeric.new
n <=> n            # => 0
n <=> Numeric.new  # => nil
n <=> 1            # => nil

This changes the return type to Integer? and adds NumericInstanceTest#test_op_cmp for both results. With the current signature the new test fails on the nil case ("Call trace does not match one of method definitions"); with Integer? the Numeric tests pass, and so do the Integer, Float, Rational, Complex and Comparable tests (Ruby 4.0.9). rbs validate and rubocop pass.

The only values this implementation returns are 0 and nil, so 0? would also be accurate, like Kernel#<=> since #1945. This uses Integer? because the subclasses in core already do: Float#<=> and Complex#<=> return Integer?, and Integer#<=> / Rational#<=> return Integer? for an untyped argument. It also stays compatible with #3210, which discusses whether <=> that subclasses override should be typed as Integer?. It can be changed to 0? if that is preferred.

[ksss]

I support your idea.
I think it is also correct from the perspective of LSP.
The "contract type" should be expressed rather than the "implementation type".

I also support #3210, but I did not mention it because it is in a draft state.
@apiology What do you think?