Circular Dependency should only be an exception if no scope defined.

[hpebley3]

Currently an activation exception is thrown whenever a circular dependency is found. It does not make sense to me, and I think it should be changed, that if a scope is introduced that keeps the dependency from sending the resolver into infinite recursion, then the exception should not be thrown.

In other words, currently the CircularDependenciesTests all bind the test classes as InSingletonScope and expect an exception. It seems like these tests should not expect an exception since the second instance of Foo (needed by Bar) would be the first instance already created. Another set of test cases should not bind as InSingletonScope and should expect an exception.

Is there a reasonable argument against this?

[hpebley3]

This is different from v1.5 where circular dependencies worked when Using<SingletonBehavior> was used. If this is a feature and not a bug, it should probably be included in the breaking behavior documentation.

[remogloor]

We must distinguish the possible circular references.

• If they are on the constructor its technically not possible to resolve. Its not possible to create a Foo that requires a Bar in the constructor which itsself requires a Foo in the constructor.
• If they are on properties Ninject can handle them.
• The mixed case constructor and property is tricky. It the one with property is created first Ninject can handle. If the constructor injected object is created first Ninject can not. The second case is technically possible but would require a major change in the implementation (Activation after everything is created).