We use option 3 sometimes…

Option 3 i.e “Getting a stub just before the method call” is essentially “setter” injection in a factory.

We have different grades of dependency. Mandatory references will always be injected via ctor. Optional references will be usually ” setter” injected – e.g. diagnostics. Also, an arguable advantage is of using this approach is the code is simple to read!

He briefly mentions that TOOD conflicts with OOD – from my experience thats a huge understatement. Usually, using IOC + SRP etc means code is easier to test – but this makes the code much more complex – e.g. at the high level of the domain you may sometimes end up with classes that have a ridiculous number of dependencies. 100% TOOD + 100% OOD != solved problem. In my opinion fully abiding by OOD means TOOD is broken or vice versa – you cant have the best of both worlds, you must instead trike a happy medium. In my experience finding the happy medium can be challenging!

Additional notes:
Wouldnt it be so much easier if c# had virtual methods by default!
Really like InternalsInvisibleTo, may adopt this