Zig Async I/O Skill (0.16.0+)

Overview

Zig 0.16.0 introduces a redesigned async I/O system based on the std.Io interface. Unlike the old async/await (removed in 0.11), the new design decouples concurrency expression from execution models, allowing code to work optimally across synchronous, multi-threaded, and event-driven contexts.

Critical Concept: Asynchrony ≠ Concurrency

• async: Operations can proceed out-of-order (sequential awaiting is valid)
• concurrent: Operations must proceed simultaneously (requires parallelism)

Table of Contents

• [Bundled Resources](#bundled-resources)
• [Core Concepts](#core-concepts)
• [Workflows](#workflows)
• [Common Patterns](#common-patterns)

Bundled Resources

References

Fundamentals:

• references/async-overview.md - New async I/O design philosophy
• references/io-interface.md - std.Io interface and primitives
• references/async-vs-concurrent.md - Critical distinction explained

Patterns:

• references/future-handling.md - Future.await() and Future.cancel()
• references/resource-management.md - Defer patterns for async resources
• references/io-implementations.md - Blocking, ThreadPool, EventLoop, Stackless

Advanced:

• references/message-passing.md - Io.Queue for synchronization
• references/vectorized-io.md - sendFile() and drain() operations
• references/migration-guide.md - From old async or sync code

Examples

Complete async I/O demonstrations:

• examples/basic_async.zig - Simple async operations with io.async()
• examples/concurrent_tasks.zig - True concurrency with io.concurrent()
• examples/file_operations.zig - Async file I/O patterns
• examples/http_server.zig - Async HTTP server with event loop
• examples/producer_consumer.zig - Message passing with Io.Queue
• examples/cancellation.zig - Proper task cancellation patterns

Templates

Starting points for async code:

• assets/templates/async-function.zig - Async function template
• assets/templates/async-server.zig - Async server template
• assets/templates/async-client.zig - Async client template
• assets/templates/threaded-io.zig - Thread pool I/O setup

Core Concepts

The std.Io Interface

const Io = struct {
    /// Spawn async work (may execute immediately or be scheduled)
    fn async(self: *Io, func: anytype, args: anytype) Future

    /// Spawn concurrent work (fails if parallelism unavailable)
    fn concurrent(self: *Io, func: anytype, args: anytype) !Future

    /// Message passing primitive
    fn Queue(comptime T: type) type
};

Future Operations

const Future = struct {
    /// Wait for result (idempotent)
    fn await(self: *Future, io: *Io) !T

    /// Cancel and retrieve result (idempotent)
    fn cancel(self: *Future, io: *Io) !T
};

I/O Implementations

1. Blocking I/O - Standard syscalls, zero overhead
2. Thread Pool - std.Io.Threaded multiplexes across OS threads
3. Event Loop - io_uring/kqueue with green threads (future)
4. Stackless Coroutines - State machine rewriting (future)

Workflows

Writing Async Code

1. Accept Io parameter - Functions take io: *std.Io like allocators
2. Spawn async work - Use io.async() for potentially concurrent operations
3. Await results - Call future.await(io) when needed
4. Handle cancellation - Use defer future.cancel(io) for cleanup
5. Choose concurrent carefully - Only use io.concurrent() when simultaneous execution required

Async vs Concurrent Decision

Use io.async() when:

• Operations can proceed in any order
• Sequential awaiting is acceptable
• Want to work with blocking I/O implementations

Use io.concurrent() when:

• Operations MUST run simultaneously
• Solving producer-consumer deadlocks
• Require true parallelism

Resource Management Pattern

var future = io.async(operation, .{io, args});
defer if (future.cancel(io)) |result| {
    cleanup(result);
} else |_| {};

// Use future...
try future.await(io);

This single pattern handles both success and failure cases.

Common Patterns

Pattern 1: Parallel File Operations

fn saveFiles(io: *std.Io, data: []const u8) !void {
    var fut_a = io.async(saveFile, .{io, data, "a.txt"});
    var fut_b = io.async(saveFile, .{io, data, "b.txt"});

    try fut_a.await(io);
    try fut_b.await(io);
}

Pattern 2: Producer-Consumer with Queue

var queue = io.Queue(Task).init();

// Producer
var producer = try io.concurrent(produce, .{io, &queue});

// Consumer
var consumer = try io.concurrent(consume, .{io, &queue});

try producer.await(io);
try consumer.await(io);

Pattern 3: Cancellation on Timeout

var work = io.async(longOperation, .{io});
var timeout = io.async(sleep, .{io, 5000});

const result = io.race(&.{work, timeout});
if (result == 1) { // timeout won
    _ = work.cancel(io) catch {};
    return error.Timeout;
}

Pattern 4: Passing Io Like Allocator

pub fn processData(io: *std.Io, allocator: Allocator, data: []const u8) !void {
    // Use io just like allocator
    var future = io.async(helper, .{io, allocator, data});
    try future.await(io);
}

Key Differences from Old Async

Old Model (0.10.x and earlier):

• async and await keywords
• Stackless coroutines baked into language
• Execution model virality
• Removed in 0.11 for redesign

New Model (0.16.0+):

• io.async() and future.await(io) methods
• Execution model as runtime parameter
• No function coloring
• Works with synchronous code

Migration Guide

From synchronous code:

// Before (sync)
try saveFile(data, "a.txt");
try saveFile(data, "b.txt");

// After (async)
var fut_a = io.async(saveFile, .{io, data, "a.txt"});
var fut_b = io.async(saveFile, .{io, data, "b.txt"});
try fut_a.await(io);
try fut_b.await(io);

From old async/await:

// Before (old async - removed)
var frame_a = async saveFile(data, "a.txt");
var frame_b = async saveFile(data, "b.txt");
try await frame_a;
try await frame_b;

// After (new async)
var fut_a = io.async(saveFile, .{io, data, "a.txt"});
var fut_b = io.async(saveFile, .{io, data, "b.txt"});
try fut_a.await(io);
try fut_b.await(io);

Best Practices

1. Pass Io explicitly - Like allocators, pass as parameter not global
2. Default to async - Use io.async() unless concurrent truly needed
3. Always handle cancellation - Use defer pattern for resource cleanup
4. Understand async ≠ concurrent - Async allows out-of-order, concurrent requires simultaneous
5. Test with multiple implementations - Verify code works with Blocking and ThreadPool
6. Use Queue for synchronization - Io.Queue handles producer-consumer patterns
7. Avoid blocking in async - Don't call blocking syscalls directly in async functions
8. Leverage idempotency - await and cancel can be called multiple times safely

Version Information

Minimum Zig Version: 0.16.0 (unreleased as of writing)

This skill targets the new async I/O design planned for Zig 0.16.0 based on:

• Loris Cro's blog post: "Zig's New Async I/O"
• Andrew Kelley's post: "Zig's New Async I/O (Text Version)"

Status: Implementation in progress, API subject to change

Additional Resources

Load references for deep dives:

• references/async-overview.md - Complete design philosophy
• references/io-implementations.md - Implementation strategies
• references/message-passing.md - Advanced synchronization