For the best part of two decades, through wrapt, I have been dealing with the mechanics of monkey patching in Python. Anyone who has followed wrapt will know I am quite pedantic about correctness, to the point of caring whether a wrapper preserves every last introspectable detail of the thing it wraps. For much of that time I have wanted the same standard from the tools I use when testing code, and unittest.mock, which does what it does well enough, was never designed to give it. A fabricated Mock answers every method call and verifies nothing. A patched call records a flat list of calls, with no return values and no sense of what was called from what. The calls an object makes to itself are invisible, because the substitute never runs the real code at all. What I want a test to be sure of, that the right calls happened, in the right order, with the real code actually running, sits just outside what substitution can express.
Seeing the real calls as they happen, with the real code still doing the work, was also something I had already spent years on for a quite different purpose. I was the original author of the New Relic Python agent, written while I worked there, and that left me with a lasting interest in instrumenting Python programs. Attaching observation to code you do not control, recording what flows through it, and doing so without disturbing the program being watched, is a problem I have never really stopped thinking about.
