There is a solution to this problem of content addressable structures. I was (paper, head) designing the same general approach 10 years ago, borrowing a sort of

There is a solution to this problem of content addressable structures. I was (paper, head) designing the same general approach 10 years ago, borrowing a sort of ~homoiconicity from Lisp (code and data are just tuples of structured hashes) as well. The problem, explicitly and concisely is this: Referential elements of structures introduce cascading re-hashing. There is however a solution to this by adding a layer of (invariant) semantic indirection for referential (as opposed to inlined embedded ‘value’ tuples) types. With this approach, at this point, we obtain 2 benefits: ‘object’ re-evaluation (rehash) is only necessary when object structure changes (not the referents); and in an object graph, cascades will only impact the immediate referents’ ‘value’ mappings. caveat that I felt extremely queasy seeing this post on hn — a session of self criticism ensued regarding the topic of “laziness” :D — so haven’t read beyond the upsetting intro to Unison.