Need to handle cases where attribute may not be a symbol we have

[CJ-Wright]

For example

# code from xyz
# b = {'hi': 'world'}

from xyz import b

a = b.values()

We would observe the symbol as xyz.b.values but I don't think we'll know that b has a valid symbol values. In this case we should maybe state that once we get to the end of our chain of attributes, then we stop looking. The length of the chain depends on the type of symbol: modules can have as many as needed, functions zero, constants zero, classes one. We could handle this if we knew that b was a dict, but we'd have to have that metadata in hand. And that could be very difficult if we got b from a function where the return type could be anything.

[CJ-Wright]

That approach does produce a nasty blind spot, where we don't see changes in type (for instance b moves from dict to list) as API changes, although that would certainly break the example code and need to be pinned accordingly.