[int_19h]

It is far easier to store such structures as object graphs in-memory (i.e. your "foreign key" is a pointer/reference to the actual object). The navigation patterns would mostly be looking up properties on the item referenced by inventory, so it's not like you need to do joins etc (but even if you did, a join on in-memory object graph is still pretty easy and blazing fast).

For C++ especially, I would recommend looking at Boost multi_index library. This gives you the ability to do fast lookups on a variety of keys across the same data.

Pretty much the only benefit I can see from SQLite in those small dataset scenarios is when you need persistence and the ability to change subset of data in an atomic way (if you only need to save the entire in-memory dataset atomically, you can always just do the rename trick to ensure atomicity with far less overhead). Well, and, I guess, optimization of complicated queries - but I'm somewhat skeptical about the ability of their optimizer to use indices in a query that's really complicated; and simple ones are trivial to do explicitly.