QBICE

Query-Based Incremental Computation Engine

QBICE is a high-performance, asynchronous incremental computation framework for Rust. Define your computation as a graph of queries, and QBICE automatically determines what needs to be recomputed when inputs change—minimizing redundant work through intelligent caching and dependency tracking.

Features

• 🚀 Incremental Computation — Only recomputes what's necessary when inputs change
• ⚡ Async-First Design — Built on Tokio for efficient concurrent execution
• 🔄 Cycle Detection — Automatically detects and handles cyclic dependencies
• 🔒 Type-Safe — Strongly-typed queries with associated value types
• 🧵 Thread-Safe — Safely share the engine across multiple threads
• 📊 Visualization — Generate interactive HTML dependency graphs

Quick Start

use std::sync::Arc;
use qbice::{
    Identifiable, StableHash,
    config::DefaultConfig,
    engine::{Engine, TrackedEngine},
    executor::{CyclicError, Executor},
    query::Query,
};

// Define an input query
#[derive(Debug, Clone, PartialEq, Eq, Hash, StableHash, Identifiable)]
struct Variable(u64);

impl Query for Variable {
    type Value = i64;
}

// Define a computation query
#[derive(Debug, Clone, PartialEq, Eq, Hash, StableHash, Identifiable)]
struct Sum {
    a: Variable,
    b: Variable,
}

impl Query for Sum {
    type Value = i64;
}

// Define the executor
struct SumExecutor;

impl<C: qbice::config::Config> Executor<Sum, C> for SumExecutor {
    async fn execute(
        &self,
        query: &Sum,
        engine: &TrackedEngine<C>,
    ) -> Result<i64, CyclicError> {
        let a = engine.query(&query.a).await?;
        let b = engine.query(&query.b).await?;
        Ok(a + b)
    }
}

#[tokio::main]
async fn main() {
    // Create and configure the engine
    let mut engine = Engine::<DefaultConfig>::new();
    engine.register_executor::<Sum, _>(Arc::new(SumExecutor));

    // Set input values
    {
        let mut session = engine.input_session();
        session.set_input(Variable(0), 10);
        session.set_input(Variable(1), 20);
    }

    // Query the engine
    let engine = Arc::new(engine);
    let tracked = engine.clone().tracked();
    let result = tracked.query(&Sum {
        a: Variable(0),
        b: Variable(1),
    }).await;

    assert_eq!(result, Ok(30));
}

Core Concepts

Queries

A query represents a unit of computation with an input key and an output value. Queries are identified by their type and content hash, ensuring stable identification across program runs.

#[derive(Debug, Clone, PartialEq, Eq, Hash, StableHash, Identifiable)]
struct FileContents {
    path: Arc<str>,  // Use Arc for cheap cloning
}

impl Query for FileContents {
    type Value = Arc<[u8]>;  // Use Arc for cheap cloning
}

Performance Tip: Both query types and their values are cloned frequently internally. Use Arc<T>, Arc<str>, or Arc<[T]> for heap-allocated data to ensure O(1) cloning.

Executors

Executors define how to compute values for queries. They must behave as pure functions—given the same inputs, they must always produce the same output.

struct FileContentsExecutor;

impl<C: Config> Executor<FileContents, C> for FileContentsExecutor {
    async fn execute(
        &self,
        query: &FileContents,
        engine: &TrackedEngine<C>,
    ) -> Result<Arc<[u8]>, CyclicError> {
        // Query dependencies through the engine
        let base_path = engine.query(&BasePath).await?;
        let full_path = format!("{}/{}", base_path, query.path);

        // Pure computation based on inputs
        let contents = std::fs::read(&full_path).unwrap_or_default();
        Ok(contents.into())
    }
}

Important: Executors should not read from global mutable state, system time, or external sources without modeling them as query dependencies. This ensures correct incremental behavior.

Engine Lifecycle

// 1. Create and configure
let mut engine = Engine::<DefaultConfig>::new();
engine.register_executor::<MyQuery, _>(Arc::new(MyExecutor));

// 2. Set inputs
{
    let mut session = engine.input_session();
    session.set_input(InputQuery(0), value);
}

// 3. Wrap in Arc and query
let engine = Arc::new(engine);
let tracked = engine.clone().tracked();
let result = tracked.query(&MyQuery { ... }).await?;

// 4. Update inputs (requires dropping TrackedEngine first)
drop(tracked);
{
    let engine_mut = Arc::get_mut(&mut engine).unwrap();
    let mut session = engine_mut.input_session();
    session.set_input(InputQuery(0), new_value);
}

// 5. Query again - only affected computations rerun
let tracked = engine.tracked();
let result = tracked.query(&MyQuery { ... }).await?;

Incremental Updates

QBICE tracks dependencies automatically. When you update inputs, only queries that depend on changed values are recomputed:

// Initial computation
let tracked = engine.clone().tracked();
assert_eq!(tracked.query(&Sum { a: Variable(0), b: Variable(1) }).await, Ok(300));
drop(tracked);

// Update one input
{
    let engine_mut = Arc::get_mut(&mut engine).unwrap();
    let mut session = engine_mut.input_session();
    session.set_input(Variable(0), 150);  // Changed!
}

// Only Sum is recomputed, not unrelated queries
let tracked = engine.tracked();
assert_eq!(tracked.query(&Sum { a: Variable(0), b: Variable(1) }).await, Ok(350));

Handling Cycles

QBICE detects cyclic dependencies and returns CyclicError. For intentional cycles (e.g., fixed-point computations), implement scc_value:

impl<C: Config> Executor<Reachable, C> for ReachableExecutor {
    async fn execute(&self, query: &Reachable, engine: &TrackedEngine<C>) -> Result<bool, CyclicError> {
        if query.from == query.to {
            return Ok(true);
        }
        // Check reachability through neighbors...
        Ok(false)
    }

    fn scc_value() -> bool {
        false  // Default value when cycle detected
    }
}

Visualization

Generate interactive HTML visualizations of your dependency graph:

engine.visualize_html(&my_query, "dependency_graph.html")?;

This creates an interactive page with:

• Pan and zoom navigation
• Click-to-inspect node details
• Search and filtering
• Color-coded dependency edges

Execution Styles

For advanced optimization, queries can have different execution styles:

• Normal — Standard dependency tracking (default)
• Projection — Fast extractors for parts of other queries
• Firewall — Boundaries that limit dirty propagation

impl<C: Config> Executor<MyQuery, C> for MyExecutor {
    async fn execute(&self, ...) -> Result<T, CyclicError> { ... }

    fn execution_style() -> ExecutionStyle {
        ExecutionStyle::Firewall  // Limits dirty propagation
    }
}

Requirements

• Rust 1.88.0 or later (Edition 2024)
• Tokio runtime

License

This project is licensed under MIT License.

Contributing

Contributions are welcome! Please feel free to submit issues and pull requests.