I'm using Google to geocode addresses, which breaks down an address into geographic units like country, state, city, etc. Depending on the address being geocoded, some geographic units may or may not be returned (e.g. maybe city is not returned, but suburb will be).
I'm tempted to store the data of each broken down address in a single row of a table, like:
USA, California, San Francisco USA, California, San Jose USA, California, Los Angeles
Obviously, this doesn't seem normalized as data is being repeated. I still could potentially query this effectively, though (e.g. list all states for USA).
Another option would be to save the data in parent-child relationships, a tree. That would be more normalized, but potentially harder to maintain and query (I can use a library to make this a lot easier, though, so it's not much of a concern for me).
My biggest concern using a parent-child model is that the branches may not all be the same length (e.g. city is not returned, but instead suburb is after the state). If I were to query for all the nodes below the state level, I could get a mix of cities and suburbs -- which might not be ideal. I could avoid this problem if I saved the data in a single row, with each geographic unit in their appropriate columns.
What's the proper way to model address parts?