#datalog #ascent

ascent-byods-rels

data structures for Ascent relations, made possible by Ascent's BYODS feature

6 releases

0.7.0 Oct 6, 2024
0.6.0 Mar 31, 2024
0.6.0-alpha.1 Mar 30, 2024
0.6.0-alpha Mar 29, 2024

#668 in Data structures

Download history 13/week @ 2024-07-22 14/week @ 2024-09-23 137/week @ 2024-09-30 148/week @ 2024-10-07 27/week @ 2024-10-14

326 downloads per month

MIT license

340KB
7K SLoC

Logic programming in Rust

Rust Crates.io

Ascent is a logic programming language (similar to Datalog) embedded in Rust via macros.

For more information, check out the CC paper on Ascent.

In addition, this OOPSLA paper describes the "Bring Your Own Data Structures to Datalog" aspect of Ascent.

Examples

Computing all the connected nodes in a graph

ascent! {
   relation edge(i32, i32);
   relation path(i32, i32);
   
   path(x, y) <-- edge(x, y);
   path(x, z) <-- edge(x, y), path(y, z);
}

Using Ascent

  1. Install Rust.
  2. Make a new Rust project:
    cargo new my-ascent-project
    cd my-ascent-project
    
  3. Add ascent as a dependency in Cargo.toml:
    [dependencies]
    ascent = "*"
    
  4. Write some Ascent code in main.rs. Here is a complete example:
    use ascent::ascent;
    ascent! {
       relation edge(i32, i32);
       relation path(i32, i32);
       
       path(x, y) <-- edge(x, y);
       path(x, z) <-- edge(x, y), path(y, z);
    }
    
    fn main() {
       let mut prog = AscentProgram::default();
       prog.edge = vec![(1, 2), (2, 3)];
       prog.run();
       println!("path: {:?}", prog.path);
    }
    
  5. Run the program:
    cargo run
    

Examples

See the ascent/examples directory for a more complete set of examples.

Features

Lattices

Ascent supports computing fixed points of user-defined lattices. The lattice keyword defines a lattice in Ascent. The type of the final column of a lattice must implement the Lattice trait. A lattice is like a relation, except that when a new lattice fact (v1, v2, ..., v(n-1), vn) is discovered, and a fact (v1, v2, ..., v(n-1), v'n) is already present in the database, vn and v'n are joined together to produce a single fact.

This feature enables writing programs not expressible in Datalog. For example we can use this feature to compute the lengths of shortest paths between nodes in a graph.

ascent! {
   lattice shortest_path(i32, i32, Dual<u32>);
   relation edge(i32, i32, u32);

   shortest_path(x, y, Dual(*w)) <-- edge(x, y, w);

   shortest_path(x, z, Dual(w + l)) <-- 
      edge(x, y, w), 
      shortest_path(y, z, ?Dual(l));
}

In this example, Dual<T> is the dual of the lattice T. We use Dual<T> because we are interested in shortest paths, given two path lengths l1 and l2 for any given pair of nodes, we only store min(l1, l2).

Parallel Ascent

Ascent is capable of producing parallel code. The macros ascent_par! and ascent_run_par! produce parallelized code. Naturally, column types must be Send + Sync to work in parallel Ascent.

Parallel Ascent utilizes rayon, so the parallelism level can be controlled either via rayon's ThreadPoolBuilder or using the RAYON_NUM_THREADS environment variable (see here for more info).

BYODS

ascent-byods-rels on crates.io

BYODS (short for Bring Your Own Data Structures to Datalog) is a feature of Ascent that enables relations to be backed by custom data structures. This feature allows improving the algorithmic complexity of Ascent programs by optimizing the data structures used to back relations. For example, a program that requires transitive closure computation of a large graph could improve its performance by choosing a union-find based data structure trrel_uf (defined in ascent-byods-rels) for the transitive closure relation:

in Cargo.toml:

[dependencies]
ascent-byods-rels = "*"
ascent = "*"
use ascent_byods_rels::trrel_uf;
ascent! {
   relation edge(Node, Node);
   
   #[ds(trrel_uf)] // Makes the relation transitive
   relation path(Node, Node);
   
   path(x, y) <-- edge(x, y);
}

The #[ds(trrel_uf)] attribute directs the Ascent compiler to use the data structure provider defined in the module trrel_uf for the path relation. You can find a complete example of BYODS dramatically speeding up Ascent computations here.

See BYODS.MD for more information on BYODS.

ascent_run!

In addition to ascent!, we provide the ascent_run! macro. Unlike ascent!, this macro evaluates the ascent program when invoked. The main advantage of ascent_run! is that local variables are in scope inside the Ascent program. For example, we can define a function for discovering the (optionally reflexive) transitive closure of a relation like so:

fn tc(r: Vec<(i32, i32)>, reflexive: bool) -> Vec<(i32, i32)> {
   ascent_run! {
      relation r(i32, i32) = r;
      relation tc(i32, i32);
      tc(x, y) <-- r(x, y);
      tc(x, z) <-- r(x, y), tc(y, z);
      tc(x, x), tc(y, y) <-- if reflexive, r(x, y);
   }.tc
}

In the above example, we initialize the relation r directly to shorten the program.

We also provide ascent_run_par!, the parallel version of ascent_run!.

Conditions and Generative Clauses

The syntax is designed to be familiar to Rust users. In this example, edge is populated with non-reflexive edges from node. Note that any type that implements Clone + Eq + Hash can be used as a relation column.

ascent! {
   relation node(i32, Rc<Vec<i32>>);
   relation edge(i32, i32);
   
   edge(x, y) <--
      node(x, neighbors),
      for &y in neighbors.iter(),
      if x != y;
}

Negation and Aggregation

Ascent supports stratified negation and aggregation. Aggregators are defined in ascent::aggregators. You can find sum, min, max, count, and mean there.

In the following example, the average grade of students is stored in avg_grade:

use ascent::aggregators::mean;
type Student = u32;
type Course = u32;
type Grade = u16;
ascent! {
   relation student(Student);
   relation course_grade(Student, Course, Grade);
   relation avg_grade(Student, Grade);

   avg_grade(s, avg as Grade) <--
      student(s),
      agg avg = mean(g) in course_grade(s, _, g);
}

You can define your own aggregators if the provided aggregators are not sufficient. For example, an aggregator for getting the 2nd highest value of a column can have the following signature:

fn second_highest<'a, N: 'a>(inp: impl Iterator<Item = (&'a N,)>) -> impl Iterator<Item = N>
where N: Ord + Clone

Aggregators can even be parameterized! For an example of a parameterized aggregator, lookup the definition of percentile in ascent::aggregators.

Macro definitions

It may be useful to define macros that expand to either body items or head items. Ascent allows you to do this.

You can find more about macros in Ascent macros here.

Misc

  • #![measure_rule_times] causes execution times of individual rules to be measured. Example:

    ascent! {
       #![measure_rule_times]
       // ...
    }
    let mut prog = AscentProgram::default();
    prog.run();
    println!("{}", prog.scc_times_summary());
    

    Note that scc_times_summary() is generated for all Ascent programs. With #![measure_rule_times] it reports execution times of individual rules too.

  • With #![generate_run_timeout], a run_timeout function is generated that stops after the given amount of time.

  • The feature wasm-bindgen allows Ascent programs to run in WASM environments.

  • struct declarations can be added to the top of ascent! definitions. This allows changing the name and visibility of the generated type and introduction of type/lifetime parameters and constraints.

    ascent! {
       struct GenericTC<N: Clone + Eq + Hash>;
       relation edge(N, N);
       // ...
    }
    

    Hint: If you get a "private type ... in public interface (error E0446)" warning, you can fix it by making the generated Ascent type private, as done in the above example.

Dependencies

~5–10MB
~99K SLoC