Unintended consequences from implicitly implemented interfaces

In 2.3.0 we resolved #477, which means that class fakes now implement the interfaces of the base class. It was supposed to be a non-breaking enhancement, but it turned out to be a breaking change:

FakeItEasy 2.2.0:

List<int> list = A.Fake<List<int>>();
list.Add(1);
list.Count.Dump(); // 1
((ICollection<int>)list).Count.Dump(); // 1

FakeItEasy 2.3.0:

List<int> list = A.Fake<List<int>>();
list.Add(1);
list.Count.Dump(); // 1
((ICollection<int>)list).Count.Dump(); // 0, oops!

Basically, what happened is this:

• List<int> implicitly implements ICollection<int>; we used to inherit that implementation
• the changes made for #477 mean that we now create the fake as if o => o.Implements<ICollection<int>>() was passed in the creation options
• by doing this, we introduce a new, separate implementation of ICollection<int>, which is not configured and returns default values

So, it’s clearly a breaking change; whether it’s a good or a bad change is debatable (Moq does the same as FIE 2.3.0, NSub and RhinoMocks do the same as FIE 2.2.0), but anyway, it wasn’t our intention when we addressed #477.

We used to be able to chose between 2 behaviors:

• one where the interface members were implemented by the base class. list.Count was always the same, whether it was called via List<int> or via ICollection<int>. It was less flexible (Count couldn’t be configured), but it was simple to understand. It was the default behavior.
• one where the members of interfaces of the faked class are now independent of the base class implementation. It’s more flexible, but also more complex and difficult to understand. This behavior was enabled by doing o => o.Implements<ICollection<int>>(), but is now the default behavior (for all interfaces of the faked class)

Now we can only have the second behavior.

So, should we revert, or keep the new behavior?

(in case it’s not obvious, I’m in favor of reverting)

[thomaslevesque]

Thanks @adamralph

Since we all agree, let’s revert.

[adamralph]

Neither path is particularly attractive but I think the safer option is to revert, release 2.3.1, with the removal of #477 in the release notes, re-open #477 with no milestone and re-assess it.