We should never double-type anything, so it would be a good idea to add a test that checks that. One way to do that is by having a macro call that would execute twice.

It would also be good to add a test that's a bit more straightforward.

I think this case is quite complicated and we'll have to do something similar to what type_assign_op does. In those cases, we have a write = read op rhs situation, while here it's if (read) write = rhs.

I've started implementing this based on the current tests. We'll have to go through all these AK-variants to see what else can come up here, and test/implement it accordingly.

This approach is going to lead to some code duplication with type_assign_op, but we can look into that later.

Of note: testing the correct code generation isn't very trivial because the difference between null ? set(newval) : val and set(null ? newval : val) isn't very observable unless we can track the set-operation. This means that we should add an optimization test for all cases to make sure we're generating the correct code.