@marten-seemann this is kinda essential for the test to actually make sense.

I don't understand this fix. What's the problem, and how is this PR fixing it?

The problem was that inheriting from TestCaseAddressRebinding requires seeing multiple client addresses for a pass, which is incorrect (too strict). It just needs to be multiple paths (i.e., multiple pairs of IP addresses and ports).

How is "path" defined here? Isn't it the 4-tuple? How is it possible to have multiple client addresses on the same path?

TestCaseAddressRebinding requires the runner to see the client use at least two distinct IP addresses to pass the test. When CM inherits from it, that extends to the CM test, which is too strict. It is enough that two paths are used by the client, i.e., two different IP address/port pairs.

Thank you for the explanation!