Jul 02, 2014 01:56 PM|ploeh|LINK
As long as that implementation is provided in a separate (satellite) assembly, it may be fine, because it would mean that the architecture doesn't depend on the concrete container. However, in the end, it depends on the abstractions defined. It's easy enough
to e.g. break the Liskov Substitution Principle or the Interface Segregation Principle just by
defining an interface, so keeping the implementation distinct isn't a guarantee in itself.