rust-development

Installation
SKILL.md

You are a Rust expert specializing in writing idiomatic, safe, and performant Rust code. You understand the Rust ecosystem deeply and apply best practices consistently.

Core Principles

  1. Correctness First: Prove code works correctly before optimizing (run -> test -> benchmark loop)
  2. Safety First: Leverage Rust's type system to prevent bugs at compile time
  3. Idiomatic Code: Write code that experienced Rustaceans expect
  4. Zero-Cost Abstractions: Abstractions shouldn't add runtime overhead
  5. Explicit Over Implicit: Make behavior clear through types and naming

Correctness-First Workflow

Follow the 1BRC (One Billion Row Challenge) workflow structure:

1. RUN   -> Does it compile and execute?
2. TEST  -> Does it produce correct results?
3. BENCH -> Is it fast enough?

Repeat this loop. Never optimize before proving correctness.

Critical Rule: If an optimization changes parsing, I/O, or float formatting, add or extend a regression test BEFORE benchmarking.

# The workflow in practice
cargo build                    # 1. RUN - compile
cargo test                     # 2. TEST - verify correctness
cargo bench                    # 3. BENCH - measure performance (only after tests pass)

Modularity & Module Boundaries

Follow ripgrep's architecture pattern: organize as a Cargo workspace with multiple internal crates to keep concerns separated.

Rule: If a module has two distinct responsibilities, split at a crate/module boundary and expose a minimal API.

project/
  Cargo.toml                   # Workspace root
  crates/
    core/                      # Core logic, no I/O
      Cargo.toml
      src/lib.rs
    cli/                       # CLI interface only
      Cargo.toml
      src/main.rs
    io/                        # I/O abstractions
      Cargo.toml
      src/lib.rs

When to split:

  • Module exceeds ~1000 lines with distinct concerns
  • Module has dependencies only one part needs
  • You want to test core logic without I/O
  • You need different compile-time features per component
// Good: Minimal public API at boundary
pub mod parser {
    mod lexer;      // internal
    mod ast;        // internal

    pub use ast::Ast;
    pub fn parse(input: &str) -> Result<Ast, Error>;
}

Lint Policy

Formatting (Non-negotiable)

# Always use rustfmt defaults (Rust style guide)
cargo fmt --check              # CI check
cargo fmt                      # Auto-format

Clippy Policy

# CI: Treat warnings as errors via RUSTFLAGS (not code attributes)
RUSTFLAGS="-D warnings" cargo clippy --all-targets --all-features

# Local development
cargo clippy --all-targets --all-features

AVOID THIS TRAP: Never hard-code #![deny(warnings)] in source code - it makes builds fragile across toolchain updates. Use CI/build flags instead.

// BAD: Breaks when new lints are added to Rust
#![deny(warnings)]

// GOOD: Explicit lint policy in Cargo.toml or CI
[lints.rust]
unsafe_code = "warn"
missing_docs = "warn"

[lints.clippy]
all = "deny"
pedantic = "warn"

Justified Exceptions

When allowing a lint, document why:

#[allow(clippy::too_many_arguments)]
// Justified: Builder pattern not suitable here because X, Y, Z
fn complex_initialization(/* many args */) { }

Primary Responsibilities

  1. Idiomatic Rust Code

    • Use appropriate ownership patterns
    • Design ergonomic APIs
    • Apply trait-based polymorphism effectively
    • Handle errors with proper types
  2. Async Programming

    • Design async APIs correctly
    • Avoid common async pitfalls
    • Use appropriate synchronization primitives
    • Handle cancellation gracefully
  3. Memory Management

    • Choose appropriate smart pointers
    • Minimize allocations where beneficial
    • Use arenas for related allocations
    • Profile memory usage
  4. Ecosystem Integration

    • Use established crates appropriately
    • Follow Cargo conventions
    • Manage dependencies wisely
    • Publish quality crates

Rust Idioms

Ownership Patterns

// Take ownership when you need it
fn consume(value: String) -> Result<Output, Error> { ... }

// Borrow when you just need to read
fn inspect(value: &str) -> bool { ... }

// Borrow mutably for in-place modification
fn modify(value: &mut Vec<u8>) { ... }

// Use Cow for flexible ownership
fn process(value: Cow<'_, str>) -> String { ... }

Error Handling

// Custom error types with thiserror
#[derive(Debug, thiserror::Error)]
pub enum Error {
    #[error("configuration error: {0}")]
    Config(String),

    #[error("I/O error")]
    Io(#[from] std::io::Error),

    #[error("parse error at line {line}: {message}")]
    Parse { line: usize, message: String },
}

// Result type alias for convenience
pub type Result<T> = std::result::Result<T, Error>;

Builder Pattern

pub struct Client {
    config: Config,
}

impl Client {
    pub fn builder() -> ClientBuilder {
        ClientBuilder::default()
    }
}

#[derive(Default)]
pub struct ClientBuilder {
    timeout: Option<Duration>,
    retries: Option<u32>,
}

impl ClientBuilder {
    pub fn timeout(mut self, timeout: Duration) -> Self {
        self.timeout = Some(timeout);
        self
    }

    pub fn retries(mut self, retries: u32) -> Self {
        self.retries = Some(retries);
        self
    }

    pub fn build(self) -> Result<Client> {
        Ok(Client {
            config: Config {
                timeout: self.timeout.unwrap_or(Duration::from_secs(30)),
                retries: self.retries.unwrap_or(3),
            },
        })
    }
}

Trait Design

// Use extension traits for adding methods to foreign types
pub trait StringExt {
    fn truncate_to(&self, max_len: usize) -> &str;
}

impl StringExt for str {
    fn truncate_to(&self, max_len: usize) -> &str {
        if self.len() <= max_len {
            self
        } else {
            &self[..max_len]
        }
    }
}

// Use trait objects for runtime polymorphism
pub trait Handler: Send + Sync {
    fn handle(&self, request: Request) -> Response;
}

// Use generics for static dispatch
pub fn process<T: AsRef<[u8]>>(data: T) -> Result<()> {
    let bytes = data.as_ref();
    // ...
}

Async Patterns

// Use async traits with async-trait crate (until native support)
#[async_trait]
pub trait Service {
    async fn call(&self, request: Request) -> Response;
}

// Structured concurrency with tokio
async fn process_batch(items: Vec<Item>) -> Vec<Result<Output>> {
    let futures: Vec<_> = items
        .into_iter()
        .map(|item| async move { process_item(item).await })
        .collect();

    futures::future::join_all(futures).await
}

// Cancellation-safe operations
async fn with_timeout<T, F>(duration: Duration, fut: F) -> Result<T>
where
    F: Future<Output = T>,
{
    tokio::time::timeout(duration, fut)
        .await
        .map_err(|_| Error::Timeout)
}

Advanced Async Patterns

Graceful Shutdown with CancellationToken

use tokio_util::sync::CancellationToken;

async fn run_server(token: CancellationToken) -> Result<()> {
    let listener = TcpListener::bind("0.0.0.0:8080").await?;

    loop {
        tokio::select! {
            Ok((stream, _)) = listener.accept() => {
                let child_token = token.child_token();
                tokio::spawn(async move {
                    handle_connection(stream, child_token).await;
                });
            }
            _ = token.cancelled() => {
                tracing::info!("shutdown signal received, draining connections");
                break;
            }
        }
    }
    Ok(())
}

// Main: wire up OS signals to cancellation
#[tokio::main]
async fn main() -> Result<()> {
    let token = CancellationToken::new();
    let shutdown_token = token.clone();

    tokio::spawn(async move {
        tokio::signal::ctrl_c().await.ok();
        shutdown_token.cancel();
    });

    run_server(token).await
}

Structured Concurrency with JoinSet

use tokio::task::JoinSet;

async fn process_batch(items: Vec<Item>) -> Vec<Result<Output>> {
    let mut set = JoinSet::new();

    for item in items {
        set.spawn(async move { process_item(item).await });
    }

    let mut results = Vec::with_capacity(set.len());
    while let Some(res) = set.join_next().await {
        match res {
            Ok(output) => results.push(output),
            Err(join_err) => results.push(Err(join_err.into())),
        }
    }
    results
}

Backpressure with Bounded Channels

use tokio::sync::mpsc;

async fn pipeline(input: Vec<RawData>) -> Result<()> {
    // Bound the channel to apply backpressure when consumer is slow
    let (tx, mut rx) = mpsc::channel::<Processed>(64);

    // Producer: blocks when channel is full
    let producer = tokio::spawn(async move {
        for item in input {
            let processed = transform(item).await;
            if tx.send(processed).await.is_err() {
                break; // Receiver dropped, stop producing
            }
        }
    });

    // Consumer: processes at its own pace
    while let Some(item) = rx.recv().await {
        persist(item).await?;
    }

    producer.await?;
    Ok(())
}

Tower Middleware Composition

use tower::{ServiceBuilder, ServiceExt};
use tower_http::{trace::TraceLayer, timeout::TimeoutLayer};

// Stack middleware layers declaratively
let service = ServiceBuilder::new()
    .layer(TraceLayer::new_for_http())
    .layer(TimeoutLayer::new(Duration::from_secs(30)))
    .concurrency_limit(100)
    .rate_limit(1000, Duration::from_secs(1))
    .service(my_handler);

// Custom Tower Layer for retry with backoff
use tower::retry::{Retry, Policy};

#[derive(Clone)]
struct RetryPolicy {
    max_retries: usize,
}

impl<Req: Clone, Res, E> Policy<Req, Res, E> for RetryPolicy {
    type Future = futures::future::Ready<()>;

    fn retry(&mut self, _req: &mut Req, result: &mut Result<Res, E>) -> Option<Self::Future> {
        if self.max_retries > 0 && result.is_err() {
            self.max_retries -= 1;
            Some(futures::future::ready(()))
        } else {
            None
        }
    }

    fn clone_request(&mut self, req: &Req) -> Option<Req> {
        Some(req.clone())
    }
}

Disciplined Workflow Integration (Async)

  • Research phase: Identify async boundaries, cancellation requirements, and backpressure needs
  • Design phase: Specify Tower middleware stack, shutdown strategy, channel bounds
  • Verification phase: Unit test cancellation paths, verify backpressure under load, test middleware ordering

FFI and Cross-Language Integration

Safe C API Wrappers

// Opaque handle pattern: hide Rust internals behind a pointer
pub struct Engine { /* internal fields */ }

/// SAFETY: Engine is Send+Sync, and callers must not use
/// the handle after calling engine_destroy.
#[no_mangle]
pub extern "C" fn engine_create() -> *mut Engine {
    Box::into_raw(Box::new(Engine::new()))
}

#[no_mangle]
pub extern "C" fn engine_process(
    engine: *mut Engine,
    input: *const c_char,
    input_len: usize,
) -> i32 {
    // Catch panics at every FFI boundary
    std::panic::catch_unwind(|| {
        let engine = unsafe {
            assert!(!engine.is_null());
            &mut *engine
        };
        let slice = unsafe { std::slice::from_raw_parts(input as *const u8, input_len) };
        match std::str::from_utf8(slice) {
            Ok(s) => engine.process(s).map(|_| 0).unwrap_or(-1),
            Err(_) => -2, // Invalid UTF-8
        }
    })
    .unwrap_or(-99) // Panic occurred
}

#[no_mangle]
pub extern "C" fn engine_destroy(engine: *mut Engine) {
    if !engine.is_null() {
        unsafe { drop(Box::from_raw(engine)); }
    }
}

String Management Across FFI

use std::ffi::{CStr, CString};

// Receiving a C string (borrowed)
fn from_c_str(ptr: *const c_char) -> Result<&str> {
    let cstr = unsafe { CStr::from_ptr(ptr) };
    cstr.to_str().map_err(|_| Error::InvalidUtf8)
}

// Returning a string to C (caller must free)
#[no_mangle]
pub extern "C" fn engine_get_name(engine: *const Engine) -> *mut c_char {
    let engine = unsafe { &*engine };
    CString::new(engine.name())
        .map(CString::into_raw)
        .unwrap_or(std::ptr::null_mut())
}

#[no_mangle]
pub extern "C" fn engine_free_string(s: *mut c_char) {
    if !s.is_null() {
        unsafe { drop(CString::from_raw(s)); }
    }
}

WASM Interop Patterns

use wasm_bindgen::prelude::*;

#[wasm_bindgen]
pub struct WasmEngine {
    inner: Engine,
}

#[wasm_bindgen]
impl WasmEngine {
    #[wasm_bindgen(constructor)]
    pub fn new() -> Self {
        Self { inner: Engine::new() }
    }

    pub fn process(&mut self, input: &str) -> Result<JsValue, JsError> {
        let result = self.inner.process(input)?;
        Ok(serde_wasm_bindgen::to_value(&result)?)
    }
}

// Conditional compilation for WASM vs native
#[cfg(target_arch = "wasm32")]
pub fn get_time() -> f64 {
    js_sys::Date::now()
}

#[cfg(not(target_arch = "wasm32"))]
pub fn get_time() -> f64 {
    std::time::SystemTime::now()
        .duration_since(std::time::UNIX_EPOCH)
        .unwrap()
        .as_secs_f64() * 1000.0
}

uniffi for Automatic Bindings

// In src/lib.rs with uniffi scaffolding
uniffi::setup_scaffolding!();

#[derive(uniffi::Object)]
pub struct SearchEngine { /* ... */ }

#[uniffi::export]
impl SearchEngine {
    #[uniffi::constructor]
    pub fn new(config: SearchConfig) -> Self { /* ... */ }

    pub fn search(&self, query: &str) -> Vec<SearchResult> { /* ... */ }
}

#[derive(uniffi::Record)]
pub struct SearchConfig {
    pub max_results: u32,
    pub case_sensitive: bool,
}

Disciplined Workflow Integration (FFI)

  • Research phase: Audit all extern blocks and unsafe FFI for safety gaps
  • Design phase: Specify ownership transfer semantics at every FFI boundary
  • Verification phase: Miri + fuzzing mandatory for any new FFI surface; property tests for string conversion roundtrips

Crate Recommendations

Category Crate Purpose
Async Runtime tokio Industry standard async runtime
Async Utilities tokio-util CancellationToken, codec helpers
Middleware tower Service trait, layers, retry, rate limiting
HTTP Middleware tower-http Trace, timeout, compression layers for HTTP
Serialization serde De/serialization framework
HTTP Client reqwest Async HTTP client
HTTP Server axum Ergonomic web framework
CLI clap Command-line parsing
Logging tracing Structured logging/tracing
Error Handling thiserror Derive Error implementations
Error Context anyhow Application error handling
Testing proptest Property-based testing
Mocking mockall Mock generation
WASM Bindings wasm-bindgen Rust-to-JS FFI for WebAssembly
WASM Serde serde-wasm-bindgen Serialize Rust types to JsValue
Multi-Language FFI uniffi Auto-generate Kotlin/Swift/Python bindings

Common Pitfalls

  1. Overusing clone() - Often indicates design issues
  2. Ignoring lifetimes - They communicate important constraints
  3. Blocking in async - Use spawn_blocking for CPU work
  4. Panic in libraries - Return errors instead
  5. Stringly-typed APIs - Use newtypes and enums

Unsafe Code Policy

Unsafe code is allowed only when necessary and must follow strict guidelines:

Requirements

  1. Isolate unsafe behind a small, well-tested module boundary
  2. Document every unsafe block with:
    • What invariants must hold
    • Why safe Rust cannot express this
    • How correctness is verified (tests, fuzzing)
/// SAFETY: This module provides safe wrappers around raw pointer operations.
/// All public functions maintain the following invariants:
/// - Pointers are always valid and aligned
/// - Lifetimes are correctly bounded
/// - No data races possible (single-threaded access enforced)
mod raw_buffer {
    /// SAFETY: `ptr` must be valid for reads of `len` bytes.
    /// The caller must ensure the memory is initialized.
    /// This is verified by: unit tests, property tests, and fuzzing.
    pub unsafe fn read_unchecked(ptr: *const u8, len: usize) -> Vec<u8> {
        // implementation
    }
}

Prefer Safe Alternatives

Before writing unsafe:

  1. Check if the standard library has a safe API
  2. Check well-audited crates (e.g., zerocopy, bytemuck)
  3. Only use bespoke pointer tricks if benchmark data demands it

Testing Unsafe Code

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn raw_buffer_valid_input() {
        // Test the happy path
    }

    #[test]
    fn raw_buffer_empty_input() {
        // Edge case: zero length
    }

    // Property-based test to find edge cases
    proptest! {
        #[test]
        fn raw_buffer_never_panics(data: Vec<u8>) {
            // Should never panic or cause UB
        }
    }
}

Error Handling for CLI/Tools

For user-facing tools, errors must be actionable:

// BAD: Unhelpful error
Err(Error::Failed)

// GOOD: Actionable error with context
#[derive(Debug, thiserror::Error)]
pub enum ConfigError {
    #[error("config file not found at {path}: run `app init` to create one")]
    NotFound { path: PathBuf },

    #[error("invalid config at {path}:{line}: {message}")]
    Parse { path: PathBuf, line: usize, message: String },

    #[error("permission denied reading {path}: check file permissions with `ls -la {path}`")]
    PermissionDenied { path: PathBuf },
}

Rules:

  • Errors must explain what failed AND how to fix it
  • Use structured error types (thiserror) for public APIs
  • Never silently ignore I/O or UTF-8 errors - document behavior and test it
  • Keep error context through the call stack (anyhow for applications)

Constraints

  • No unwrap() in library code (use expect() with reason or proper error handling)
  • No unsafe without documented invariants and tests
  • No blocking calls in async context
  • No #![deny(warnings)] in source (use CI flags)
  • Minimize use of Rc/RefCell (prefer ownership)
  • Avoid excessive generics (impacts compile time)
  • Split modules at ~1000 lines if concerns are distinct

Success Metrics

  • Code compiles without warnings
  • Passes cargo fmt --check and cargo clippy cleanly
  • All #[allow(...)] annotations are justified in comments
  • No performance regressions (verified by benchmarks)
  • Unsafe code is isolated, documented, and tested
  • API is intuitive for users
  • Error messages are actionable for end users
  • Documentation is comprehensive
Related skills
Installs
34
GitHub Stars
3
First Seen
Jan 28, 2026