Effect Composition: Pure Core, Imperative Shell

The Philosophy

Effect helps you structure applications with:

Pure core: Business logic with no side effects (easy to test)
Imperative shell: I/O operations at the boundaries (controlled)

This separation makes code more testable, maintainable, and composable.

Zero-Cost by Default

Stillwater’s Effect system follows the futures crate pattern: zero-cost by default, explicit boxing when needed.

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

// Zero heap allocations - compiler can inline everything
let effect = pure::<_, String, ()>(42)
    .map(|x| x + 1)           // Returns Map<Pure<...>, ...>
    .and_then(|x| pure(x * 2)) // Returns AndThen<Map<...>, ...>
    .map(|x| x.to_string());   // Returns Map<AndThen<...>, ...>

// Type: Map<AndThen<Map<Pure<i32, String, ()>, ...>, ...>, ...>
// NO heap allocation!
}

Each combinator returns a concrete type. The compiler knows the exact type at compile time and can fully optimize the effect chain.

The Problem

How do you test this code?

#![allow(unused)]
fn main() {
async fn create_user(email: String, age: u8) -> Result<User, Error> {
    // Validation mixed with I/O
    if !email.contains('@') {
        return Err(Error::InvalidEmail);
    }

    // Database call (requires real/mock DB)
    let existing = database.find_by_email(&email).await?;
    if existing.is_some() {
        return Err(Error::EmailExists);
    }

    // More I/O
    let user = User { email, age };
    database.save(&user).await?;
    Ok(user)
}
}

Problems:

Can’t test without database
Business logic mixed with I/O
Hard to reason about what’s pure vs effectful

The Solution: Effect

Effect separates pure logic from I/O:

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

#[derive(Clone)]
struct AppEnv {
    db: Database,
}

fn create_user(email: String, age: u8) -> impl Effect<Output = User, Error = AppError, Env = AppEnv> {
    // Pure validation first
    from_validation(validate_user(&email, age).map_err(AppError::Validation))
        // Then I/O
        .and_then(move |_| {
            from_fn(move |env: &AppEnv| env.db.find_by_email(&email))
        })
        // Pure logic
        .and_then(move |existing| {
            if existing.is_some() {
                fail(AppError::EmailExists)
            } else {
                pure(User { email, age })
            }
        })
        // More I/O
        .and_then(|user| {
            from_fn(move |env: &AppEnv| env.db.save(&user))
                .map(move |_| user)
        })
}

// Run at application boundary
let env = AppEnv { db };
let user = create_user(email, age).run(&env).await?;
}

Benefits:

Pure functions need no mocks
I/O is explicit via from_fn, from_async
Easy to test with mock environments
Zero heap allocations in the effect chain

Core API

Creating Effects

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

// Pure value (no I/O)
let effect = pure::<_, String, ()>(42);

// Failed effect
let effect = fail::<i32, _, ()>("error".to_string());

// From Result
let effect = from_result::<_, String, ()>(Ok(42));

// From Validation
let validation = Validation::success(42);
let effect = from_validation(validation);

// From sync function
let effect = from_fn(|env: &Env| {
    Ok::<_, String>(env.config.value)
});

// From async function
let effect = from_async(|env: &Env| async {
    env.db.fetch_user(123).await
});

// From Option
let effect = from_option::<_, _, ()>(Some(42), || "value missing");
}

Transforming Effects

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

// Map success value
let effect = pure::<_, String, ()>(21).map(|x| x * 2);
let result = effect.run(&()).await; // Ok(42)

// Map error value
let effect = fail::<i32, _, ()>("oops").map_err(|e| format!("Error: {}", e));

// Chain dependent effects
let effect = pure::<_, String, ()>(5)
    .and_then(|x| pure(x * 2))
    .and_then(|x| pure(x + 10));
let result = effect.run(&()).await; // Ok(20)
}

Validation Combinators

Stillwater provides declarative validation combinators that eliminate verbose and_then boilerplate when validating effect outputs:

Using `ensure()` with Closures

The ensure() method validates an effect’s success value and fails if the predicate returns false:

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

#[derive(Debug, PartialEq)]
enum Error {
    Negative,
    TooLarge,
}

// Before: verbose and_then pattern
let effect = pure::<_, Error, ()>(5)
    .and_then(|x| {
        if x > 0 {
            pure(x)
        } else {
            fail(Error::Negative)
        }
    });

// After: declarative ensure
let effect = pure::<_, Error, ()>(5)
    .ensure(|x| *x > 0, Error::Negative);

let result = effect.run(&()).await;
assert_eq!(result, Ok(5));
}

Using `ensure_with()` for Lazy Errors

When you need the value to construct the error:

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

#[derive(Debug, PartialEq)]
struct RangeError {
    value: i32,
    min: i32,
}

let effect = pure::<_, RangeError, ()>(-5)
    .ensure_with(
        |x| *x >= 0,
        |x| RangeError { value: *x, min: 0 }
    );

let result = effect.run(&()).await;
assert_eq!(result, Err(RangeError { value: -5, min: 0 }));
}

Using `ensure_pred()` with Composable Predicates

For reusable validation logic, use predicates from the predicate module:

#![allow(unused)]
fn main() {
use stillwater::prelude::*;
use stillwater::predicate::*;

#[derive(Debug, PartialEq)]
enum Error {
    InvalidAge,
}

let valid_age = between(18, 120);

let effect = pure::<_, Error, ()>(25)
    .ensure_pred(valid_age, Error::InvalidAge);

let result = effect.run(&()).await;
assert_eq!(result, Ok(25));

// Fails for invalid ages
let effect = pure::<_, Error, ()>(15)
    .ensure_pred(valid_age, Error::InvalidAge);

let result = effect.run(&()).await;
assert_eq!(result, Err(Error::InvalidAge));
}

Using `unless()` for Inverse Validation

The unless() method fails when the predicate is TRUE (inverse of ensure):

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

#[derive(Debug, PartialEq)]
enum Error {
    UserBanned,
}

struct User {
    id: u32,
    is_banned: bool,
}

let effect = from_fn(|_: &()| User { id: 1, is_banned: false })
    .unless(|u| u.is_banned, Error::UserBanned);

let result = effect.run(&()).await;
assert!(result.is_ok());

// Fails when user is banned
let effect = from_fn(|_: &()| User { id: 2, is_banned: true })
    .unless(|u| u.is_banned, Error::UserBanned);

let result = effect.run(&()).await;
assert_eq!(result, Err(Error::UserBanned));
}

Using `filter_or()` Alias

filter_or() is an alias for ensure() following functional programming conventions:

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

let effect = pure::<_, &str, ()>(5)
    .filter_or(|x| *x > 0, "must be positive");

let result = effect.run(&()).await;
assert_eq!(result, Ok(5));
}

Chaining Multiple Validations

Combine multiple validation checks for comprehensive validation:

#![allow(unused)]
fn main() {
use stillwater::prelude::*;
use stillwater::predicate::*;

#[derive(Debug, PartialEq)]
enum Error {
    TooShort,
    TooLong,
    NotAlpha,
}

let effect = pure::<_, Error, ()>(String::from("hello"))
    .ensure_pred(len_min(3), Error::TooShort)
    .ensure_pred(len_max(10), Error::TooLong)
    .ensure_pred(is_alphabetic(), Error::NotAlpha);

let result = effect.run(&()).await;
assert_eq!(result, Ok(String::from("hello")));

// Fails at first violation (fail-fast)
let effect = pure::<_, Error, ()>(String::from("hi"))
    .ensure_pred(len_min(3), Error::TooShort)  // fails here
    .ensure_pred(len_max(10), Error::TooLong)
    .ensure_pred(is_alphabetic(), Error::NotAlpha);

let result = effect.run(&()).await;
assert_eq!(result, Err(Error::TooShort));
}

Real-World Example

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

#[derive(Clone)]
struct Database;

impl Database {
    async fn fetch_user(&self, id: u32) -> Result<User, DbError> {
        // ... database logic
    }
}

#[derive(Clone)]
struct AppEnv {
    db: Database,
}

#[derive(Debug)]
enum AppError {
    Db(DbError),
    UserBanned,
    Underage,
}

struct User {
    id: u32,
    age: u8,
    is_banned: bool,
}

fn fetch_valid_user(id: u32) -> impl Effect<Output = User, Error = AppError, Env = AppEnv> {
    from_fn(move |env: &AppEnv| env.db.fetch_user(id))
        .map_err(AppError::Db)
        .unless(|u| u.is_banned, AppError::UserBanned)
        .ensure(|u| u.age >= 18, AppError::Underage)
}

// Usage
let env = AppEnv { db: Database };
let user = fetch_valid_user(123).run(&env).await?;
}

Why Use Effect Validation Combinators?

Before (12 lines):

#![allow(unused)]
fn main() {
from_fn(|env: &AppEnv| fetch_data(env))
    .and_then(|data| {
        if data.value > 0 {
            pure(data)
        } else {
            fail(Error::InvalidValue)
        }
    })
    .and_then(|data| {
        if data.count < 100 {
            pure(data)
        } else {
            fail(Error::TooMany)
        }
    })
}

After (3 lines):

#![allow(unused)]
fn main() {
from_fn(|env: &AppEnv| fetch_data(env))
    .ensure(|data| data.value > 0, Error::InvalidValue)
    .ensure(|data| data.count < 100, Error::TooMany)
}

Running Effects

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

// With environment
let env = AppEnv { /* ... */ };
let result = effect.run(&env).await;

// With unit environment (when Env = ())
use stillwater::RunStandalone;
let result = effect.run_standalone().await;
}

When to Use `.boxed()`

Boxing is needed in exactly three situations:

1. Storing in Collections

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

// Different effect types can't be stored in the same Vec
// Boxing gives them a uniform type
let effects: Vec<BoxedEffect<i32, String, ()>> = vec![
    pure(1).boxed(),
    pure(2).map(|x| x * 2).boxed(),
    pure(3).and_then(|x| pure(x * 3)).boxed(),
];

// Process them
for effect in effects {
    let result = effect.run(&()).await?;
    println!("Result: {}", result);
}
}

2. Recursive Effects

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

// Recursive function needs concrete return type
fn countdown(n: i32) -> BoxedEffect<i32, String, ()> {
    if n <= 0 {
        pure(0).boxed()
    } else {
        pure(n)
            .and_then(move |x| countdown(x - 1).map(move |sum| x + sum))
            .boxed()
    }
}

let sum = countdown(5).run(&()).await?; // 15
}

3. Match Arms with Different Effect Types

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

enum DataSource {
    Cache,
    Database,
    Remote,
}

fn fetch_data(source: DataSource) -> BoxedEffect<String, String, ()> {
    match source {
        DataSource::Cache => {
            // Just pure value
            pure("cached data".to_string()).boxed()
        }
        DataSource::Database => {
            // Effect with map
            pure("db")
                .map(|s| format!("{} data", s))
                .boxed()
        }
        DataSource::Remote => {
            // Effect with and_then
            pure("remote")
                .and_then(|s| pure(format!("{} data", s)))
                .boxed()
        }
    }
}
}

Reader Pattern

The Reader pattern provides functional dependency injection. Stillwater includes three helpers:

`ask()` - Access the Environment

Returns the entire environment:

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

#[derive(Clone)]
struct Config {
    api_key: String,
    timeout: u64,
}

// Get the whole environment
let effect = ask::<String, Config>();

let config = Config {
    api_key: "secret".into(),
    timeout: 30,
};

let result = effect.run(&config).await.unwrap();
assert_eq!(result.api_key, "secret");
}

`asks()` - Query Environment

Extract a specific value:

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

#[derive(Clone)]
struct AppEnv {
    database: String,
    cache: String,
}

// Query just the database field
let effect = asks(|env: &AppEnv| env.database.clone());

let env = AppEnv {
    database: "postgres".into(),
    cache: "redis".into(),
};

let result = effect.run(&env).await.unwrap();
assert_eq!(result, "postgres");
}

`local()` - Modify Environment

Run an effect with a temporarily modified environment:

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

#[derive(Clone)]
struct Config {
    debug: bool,
    timeout: u64,
}

fn fetch_data() -> impl Effect<Output = String, Error = String, Env = Config> {
    asks(|cfg: &Config| format!("fetched with timeout {}", cfg.timeout))
}

let config = Config {
    debug: false,
    timeout: 30,
};

// Run with modified timeout
let effect = local(
    |cfg: &Config| Config { timeout: 60, ..*cfg },
    fetch_data()
);

let result = effect.run(&config).await.unwrap();
assert_eq!(result, "fetched with timeout 60");
// Original config still has timeout=30
}

Parallel Effects

Heterogeneous Parallel (Zero-Cost)

For 2-4 effects of different types, use par2, par3, par4:

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

let (num, text) = par2(
    pure::<_, String, ()>(42),
    pure::<_, String, ()>("hello".to_string()),
    &(),
).await;
let num = num?;
let text = text?;
}

Homogeneous Parallel (Requires Boxing)

For collections of effects, use par_all, race, par_all_limit:

#![allow(unused)]
fn main() {
use stillwater::prelude::*;

// par_all - run all, collect all results
let effects: Vec<BoxedEffect<i32, String, ()>> = vec![
    pure(1).boxed(),
    pure(2).boxed(),
    pure(3).boxed(),
];
let results = par_all(effects, &()).await?; // [1, 2, 3]

// race - return first completed result
let effects: Vec<BoxedEffect<String, String, ()>> = vec![
    pure("first completed".to_string()).boxed(),
    pure("second completed".to_string()).boxed(),
];
let result = race(effects, &()).await?; // one completed value, or RaceError::Empty

// par_all_limit - run with concurrency limit
let effects: Vec<BoxedEffect<i32, String, ()>> = /* many effects */;
let results = par_all_limit(effects, 10, &()).await?; // max 10 concurrent
}

Real-World Example: User Registration

use stillwater::prelude::*;

// Environment with dependencies
#[derive(Clone)]
struct AppEnv {
    db: Database,
    email_service: EmailService,
}

// Error type
#[derive(Debug)]
enum AppError {
    ValidationError(Vec<String>),
    EmailExists,
    DatabaseError(String),
    EmailError(String),
}

// Pure validation (no I/O, easy to test)
fn validate_user(email: &str, age: u8) -> Validation<(), Vec<String>> {
    Validation::all((
        validate_email(email),
        validate_age(age),
    ))
    .map(|_| ())
}

// Effect composition (I/O at boundaries)
fn register_user(
    email: String,
    age: u8,
) -> impl Effect<Output = User, Error = AppError, Env = AppEnv> {
    // 1. Validate input (pure)
    from_validation(
        validate_user(&email, age)
            .map_err(AppError::ValidationError)
    )
    // 2. Check if email exists (I/O)
    .and_then(move |_| {
        from_fn(move |env: &AppEnv| {
            env.db.find_by_email(&email)
                .map_err(|e| AppError::DatabaseError(e.to_string()))
        })
    })
    // 3. Check uniqueness (pure logic)
    .and_then(move |existing| {
        if existing.is_some() {
            fail(AppError::EmailExists)
        } else {
            pure(())
        }
    })
    // 4. Create user (pure)
    .map(move |_| User { email: email.clone(), age })
    // 5. Save to database (I/O)
    .and_then(|user| {
        from_fn(move |env: &AppEnv| {
            env.db.save_user(&user)
                .map_err(|e| AppError::DatabaseError(e.to_string()))
        })
        .map(move |_| user)
    })
    // 6. Send welcome email (I/O)
    .and_then(|user| {
        from_fn(move |env: &AppEnv| {
            env.email_service.send_welcome(&user.email)
                .map_err(|e| AppError::EmailError(e.to_string()))
        })
        .map(move |_| user)
    })
}

// Usage at application boundary
#[tokio::main]
async fn main() -> Result<(), AppError> {
    let env = AppEnv {
        db: Database::connect("postgres://...").await?,
        email_service: EmailService::new(),
    };

    let user = register_user(
        "user@example.com".to_string(),
        25
    ).run(&env).await?;

    println!("Registered: {:?}", user);
    Ok(())
}

Testing Effects

The key benefit: pure functions need no mocks!

#![allow(unused)]
fn main() {
#[cfg(test)]
mod tests {
    use super::*;

    // Test pure validation (no mocks needed!)
    #[test]
    fn test_validate_user() {
        let result = validate_user("user@example.com", 25);
        assert!(result.is_success());

        let result = validate_user("invalid", 15);
        assert!(result.is_failure());
    }

    // Test effectful code with mock environment
    #[derive(Clone)]
    struct MockEnv {
        users: Vec<User>,
    }

    impl MockEnv {
        fn find_by_email(&self, email: &str) -> Result<Option<User>, String> {
            Ok(self.users.iter().find(|u| u.email == email).cloned())
        }
    }

    #[tokio::test]
    async fn test_effect_with_mock_env() {
        let env = MockEnv { users: vec![] };

        let effect = from_fn(|env: &MockEnv| env.find_by_email("test@example.com"))
            .and_then(|existing| {
                if existing.is_some() {
                    fail("Email exists")
                } else {
                    pure(User {
                        email: "test@example.com".to_string(),
                        age: 25,
                    })
                }
            });

        let result = effect.run(&env).await;
        assert!(result.is_ok());
    }
}
}

Performance Considerations

The Effect trait is zero-cost by default:

No heap allocations for effect chains
Compiler can fully inline combinators
Same performance as hand-written async code

Boxing happens only when you call .boxed():

Collections of effects
Recursive effects
Match arms with different types

For I/O-bound work (API calls, database queries), boxing overhead is negligible compared to actual work.

Common Patterns

Pattern 1: Validate Then Execute

#![allow(unused)]
fn main() {
from_validation(validate_input(input))
    .and_then(|valid| execute_with_db(valid))
}

Pattern 2: Read, Decide, Write

#![allow(unused)]
fn main() {
from_fn(|env: &Env| env.db.fetch(id))
    .and_then(|data| {
        let result = pure_business_logic(data);
        from_fn(move |env: &Env| env.db.save(result))
    })
}

Pattern 3: Error Context

#![allow(unused)]
fn main() {
create_user(email, age)
    .context("Creating user account")
    .and_then(|user| {
        send_welcome_email(&user)
            .context("Sending welcome email")
    })
}

Pattern 4: Conditional Effect

#![allow(unused)]
fn main() {
fn conditional_fetch(use_cache: bool) -> BoxedEffect<String, String, AppEnv> {
    if use_cache {
        from_fn(|env: &AppEnv| Ok(env.cache.get("data"))).boxed()
    } else {
        from_async(|env: &AppEnv| async { env.db.fetch().await }).boxed()
    }
}
}

When to Use Effect

Use Effect when:

Separating I/O from business logic
Testing effectful code
Composing async operations
Dependency injection needed

Use plain async fn when:

Simple CRUD operations
No complex composition
Testing not critical
Maximum simplicity needed

Summary

Effect trait: Zero-cost effect composition following futures pattern
Pure core: Business logic is easy to test (no mocks)
Imperative shell: I/O at boundaries via from_fn, from_async
Environment: Provides dependency injection
Boxing: Use .boxed() only when type erasure is needed
Composition: Via map, and_then, or_else, etc.
Reader pattern: ask(), asks(), local() for environment access

Next Steps

Learn about Error Context
Explore the Reader Pattern in depth
See the Migration Guide if upgrading from 0.10.x
Check out testing_patterns example
Read the API docs

Keyboard shortcuts

Stillwater Documentation