Investigate possibility to infer "real" type of substitute to improve non-virtual member calls detection

[tpodolak]

As reported in here nsubstitute/NSubstitute#499 (comment) NSubstitute.Analyzers fails to find non-virtual member calls when substituting for class but assigning it to interface e.g

IUnitOfWorkCoreCqrs cqrsUow = Substitute.ForPartsOf<CqrsUow>();
cqrsUow.Received(1).RegisterAfterCommitCallBackItem(Arg.Any<Func<Task>>());

In order to correctly report non-virtual member calls, we have to find actual substitute assignment

IUnitOfWorkCoreCqrs cqrsUow = Substitute.ForPartsOf<CqrsUow>();

or

IUnitOfWorkCoreCqrs cqrsUow = null;
cqrsUow  = Substitute.ForPartsOf<CqrsUow>();

based on the type infered from expression, we should be able to enhance CanBeSetuped method.

Possible roslyn API to use

var dataFlowAnalysis = syntaxNodeContext.SemanticModel.AnalyzeDataFlow(accessedMember);

Possible problems

• performace degradation, we might decide not to implement it

[dtchepak]

Thanks for tracking this issue.

This gets even more fun if we have a IUnitOfWorkCoreCqrs field that gets modified by multiple threads and is sometimes assigned instances of different types. 😂

There may be some simple cases we can identify (local variable, single assignment?), but my initial feeling is that it may not be worth the cost (finding a reliable implementation, plus the performance overhead). Happy to be proven wrong though! :)