In the rest of the documentation for this site we try to lay out all of the pieces of the Neo4j.rb gems to explain them one at a time. Sometimes, though, it can be instructive to see examples. The following are examples of code where somebody had a question and the resulting code after fixes / refactoring. This section will expand over time as new examples are found.
Goal: Find all contacts for a user two hops away, but don’t include contacts which are only one hop away¶
user.contacts(:contact, :knows, rel_length: 2).where_not( uuid: user.contacts.pluck(:uuid) )
This works, though it makes two queries. The first to get the
uuid s for the
where_not and the second for the full query. For the first query,
user.contacts.pluck(:id) could be also used instead, though associations already have a pre-defined method to get IDs, so this could instead be
This doesn’t take care of the problem of having two queries, though. If we keep the
rel_length: 2, however, we won’t be able to reference the nodes which are one hop away in order. This seems like it would be a straightforward solution:
user.contacts(:contact1).contacts(:contact2).where_not('contact1 = contact2')
And it is straightforward, but it won’t work. Because Cypher matches one subgraph at a time (in this case roughly
contact one is always just going to be the node which is in between the user in question and
contact2. It doesn’t represent “all users which are one step away”. So if we want to do this as one query, we do need to first get all of the first-level nodes together so that we can then check if the second level nodes are in that list. This can be done as:
user.as(:user).contacts .query_as(:contact).with(:user, first_level_ids: 'collect(ID(contact))') .proxy_as(User, :user) .contacts(:other_contact, nil, rel_length: 2) .where_not('ID(other_contact) IN first_level_ids')
And there we have a query which is much more verbose than the original code, but accomplishes the goal in a single query. Having two queries isn’t neccessarily bad, so the code’s complexity should be weighed against how both versions perform on real datasets.