This is a slightly different Datalog variant implemented as an efficient, FFI-friendly Rust library. The major deviation from standard Datalog is the existence of a simple type system for rows as well as a choice of tuple- or record-structured data.

Deviation 1: Record-Structured Data

Any place where Datalog uses tuple-structured data (i.e. where you have comma-separated expressions in between parentheses) you can also use record-structured data, which allow fields to appear in any order. So the classic example

- parent ( terry, austin ).

can be expressed nearly equivalently using record-structured data as

- parent { parent: terry, child: austin }.

Similarly, a query to find all the children of terry could be written as

- parent { parent: terry, child: X }?

Deviation 2: Static Types

Datalog facts are usually dynamically typed, but this implementation requires (in part to support the syntax above) that facts and predicates have a consistent statically known type. For facts, this often involves a "table declaration" which dictates the structure of all future facts in that table. Some table declarations might look like

- @table edge ( integer, integer ).
- @table parent { parent: atom, child: atom }.

Queries also have static types, but these can generally be inferred from context in Datalog, and so do not need to be defined. If necessary, though, a query definition can be used.

- @query ancestor { parent: atom, child: atom }.
- ancestor { parent: X, child: Y}
    :- parent { parent: X, child: Y }.
- ancestor { parent: X, child: Y}
    :- parent { parent: X, child: Z},
       ancestor { parent: X, child: Y }.

Deviation 3: Implicit Variables

When defining a table's type signature, we can also include a few extra annotations about the meaning of that variable. For example, we might want a particular field of a table to be an auto-incrementing integer, so that every row would include a fresh value without us having to discover it first. If we define a table with the @autoincrement keyword, like so

- @table person { id: integer @autoincrement, name: string }.

then we can use the placeholder @auto value, which mean Datalog will examine the table to find the necessary "next" value.

- person { id: @auto, name: "alice" }.
- person { id: @auto, name: "bob" }.
- person { id: @auto, name: "carol" }.