What is monkey patching in Python?

Question presented to candidate: "What is monkey patching? Show me how you would do it, tell me where it is genuinely useful, and why many teams treat it as a code smell outside of tests."

What a strong answer should cover:

• Monkey patching is replacing or adding an attribute (a method, function or value) on a class, instance or module at run time, without editing its source; it works because classes and modules are mutable objects whose attributes are looked up when used.
• Scope: patching a class affects every instance, old and new; patching an instance needs types.MethodType to bind self; patching a module affects only code that looks the name up through the module.
• The from-import trap: from clock import now copies a reference, so patching clock.now does not affect it; with unittest.mock.patch you patch where the name is looked up.
• Main legitimate use is testing: unittest.mock.patch / patch.object (stdlib since 3.3) or pytest monkeypatch, which restore the original automatically; others are hotfixing a third-party bug and libraries like gevent that patch the stdlib.
• Risks: global, invisible side effects, order-dependent behaviour, breakage on library upgrades, and built-in types such as str cannot be patched at all.

Clarifying questions expected:

• Is this for tests or for production code? The acceptable answer differs.

Code / implementation expected: Yes — patch a class method and a module function, show the from-import trap, and the same done safely with unittest.mock.patch.