The types of data
tel.github.io
tel.github.io
For example, one can imagine a relational type system which describes values (which are finite sets of tuples having named, unordered fields) by their cardinality (0, 1, >1, 0 or 1, >0, or any) the field names they contain, and the types of the values of those fields.
You can extend this system in several ways as well; the type can include keys (sets of fields for which each tuple in the set contain distinct values) and dependent field specifications (denoting fields which are present/have a given type only if some other given field has a certain value).
The reason I chose to call these "the" types of data is not because there exist no other reasonable mechanisms for defining things like types and data but instead because there's such a strong relationship between this presentation of types and the foundations of intuitionistic logic. You can also find these types in category theory (given a sufficiently nice category) which further cements their conceptual power.
Further, you can refine this presentation to make a dependent type theory and end up with a really wonderful place to express something like your relational type system.
One of my issues with algebraic types is the awkwardness of describing enumerable arrays and sets. Both end up being represented as some kind of binary tree. (I say "enumerable" because obviously functions can easily represent non-enumerable arrays and sets.) Relational types can describe both trivially.
I'm sure relational types have weaknesses likewise (though I can't think of any off the top of my head). A non-parametric relational type system – which doesn't permit field values to themselves be relations, such as in SQL – has obvious difficulties specifying recurrent structures. (You can of course flatten these using pointers, but without dependent constraints you can't catch bugs.)
Effectively, products are the quintessential "grouping" notion, sums the quintessential "choice" notion, and implications the quintessential "hypothetical" notion. So those are pretty important for describing any computation.
All that said, I really like relational types and wish they were more commonly used. I do think it takes some effort to model them in ADTs and that a system which provided them more directly would be fun.
Not only do the words not really make sense to me in the order they're presented; and not only do the assertions seem alien and foreign in nearly every respect; but I can't fathom a pragmatic path by which even if I accepted everything here as given, it would give me an intuition I could use to do anything different or better than I do now.
And that's neat, because there's apparently (given the footnotes) a bunch of people for whom that is not true, so I have to be missing something somewhere, and that's a learning opportunity. And certainly I've enjoyed your other blog posts, tel, for their clarity and brevity and informational content. Nevertheless this is problematic, because I'm probably your target audience, and I couldn't be any more baffled or concerned by the idea that you'd be blithely mapping "+" (sum, addition) to "\/" (logical-or, disjunction) across fields and asserting conceptual equivalence like it's a given, much less a foundational building block of an argument.
Now off to bounce harmlessly off that blog post a few more times.