When you do "alias this" and have an invariant, any methods that are forwarded to the aliased member do not invoke your invariant methods. This prevents me from writing a really sleek 10-liner to the tune of: struct ValueRestrictedInteger(int lowerBound, int upperBound) { int value; alias value this; this (int rhs) { value = rhs; } invariant() { assert (value >= lowerBound && value <= upperBound); } void forDemonstrationOnly() {} } unittest { ValueRestrictedInteger!(0, 100) x = 0; x += 10; x -= 100; //This works, but I don't think it should x.forDemonstrationOnly(); //This causes the assertion to fire ValueRestrictedInteger!(0, 100) y = -100; //This also would hit the assertion } Forum post on same topic: http://forum.dlang.org/thread/ijoqlxqlacccwmqnwpdn@forum.dlang.org
I think from angel's comment that discussion, this bug is invalid. There is no method to call, so it cannot hook the invariant, x -= 100 translates directly to machine instructions. What I would suggest is instead of using alias this, use opDispatch and opBinary to implement inlined functions that forward to the member. Then your invariant will be called. Instead of using alias this for inheritance, use composition.
I think there is still a problem. Does it actually make sense to have a class invariant, when you have an alias this? It seems as though having an 'alias this' is like exposing a public member variable: once you do this, the class invariant cannot be trusted to be true at all times. I feel uneasy about this; it suggests that 'alias this' is bad practice. It may be worth adding to the documentation at http://dlang.org/contracts. That states: "Invariants are used to specify characteristics of a class or struct that always must be true (except while executing a member function)." I think that section should explicitly state that direct modification of members in a way which violates the invariant, will not be detected until a member function is called.
It's not just with alias this. Any modification of a member variable outside using the methods can result in breaking the invariant. I think it's not a bad idea to mention the pitfalls, but I don't think alias this is special in that regard.