Been thinking a bit about "custom unification", or extensible unification. It seems it is a pretty common thing: • A library that implements unification / custom unification in JavaScript • Extending unification in clojure's core.logic • An unification library for Python • Also TIL SWI prolog supports maps/dictionaries Wondering how much it makes sense to use more complex data structures in prolog-like searches. Datalog seems to be an example that going "the other way", being more restrictive, may be more productive:

Datalog disallows complex terms as arguments of predicates, e.g., p (1, 2) is admissible but not p (f (1), 2)

Perhaps unification of complex/custom data structures is more productive for pattern matching than for general logic programming ... but not sure (SWI prolog's dict maybe contradicts this idea).

FWIW I came to be interested in Datalog precisely because it didn't have nested data structures. Nested data structures of arbitrary depth must be constructed and deconstructed using ordinary functional-style recursive algorithms, which goes against the whole philosophy of Datalog (in my opinion): performing recursive relational queries (select, join, union, etc.) on flat datasets.