Node Types
Bringing node types into the new resolver system had me stumped for most of the morning, but I eventually started to get a picture of it.

For a start, I will be implementing it as a new resolver, and registering it as "internal". It will then get found based on model type rather than protocol. As before, it will build its data from the string pool, only now it will be using more than one string pool, so it will be appending data.

The trick is to make sure that the node type resolver uses the same string pool, on the same session, as all the other resolvers. I was concerned abotu how to get this, but SR was able to reassure me that I can get it easily.

The other important requirement is that constraints to be resolved against a node type model will occur last. This is so all other resolvers will have already populated the string pools with data returned from their query. This is a little harder to guarantee.

At the moment, the order of resolution is based on the approximate size of the data that the constraint expects to return. One way to be executed last would be to return a size of Long.MAX_VALUE. Unfortunately, several other resolvers try to bias themselves to go last by doing this. In this case the resolver absolutely must go last, so it can't necessarily rely on this trick.

In the interim, SR has suggested that I try returning Long.MAX_VALUE as a size. If another resolver tries to get in the way then we can deal with it then. Since most resolvers play well, this should not be a real problem, at least not in the general case.

Armed with this design I've started coding up the new resolver. It will probably take me a day or two.