# swift-concurrency

Adopt this skill: https://proskills.md/skills/other/avdlee-swift-concurrency-agent-skill-skills-swift-concurrency

Page: https://proskills.md/skills/other/avdlee-swift-concurrency-agent-skill-skills-swift-concurrency
This document is the public skill bundle for that page. It contains SKILL.md and every helper file the catalog carries. No login is required.

## File: skills/swift-concurrency/SKILL.md
size: 12852
sha: 1dd24dff6546529bc516290002ded5debe002487
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/SKILL.md

````text
---
name: swift-concurrency
description: Diagnose Swift Concurrency issues, refactor callback-based code to async/await, and guide Swift 6 migration when working with tasks, actors, @MainActor, Sendable, data races, thread safety, or concurrency-related compiler and linter warnings.
---
# Swift Concurrency

## Fast Path

Before proposing a fix:

1. Analyze `Package.swift` or `.pbxproj` to determine Swift language mode, strict concurrency level, default isolation, and upcoming features. Do this always, not only for migration work.
2. Capture the exact diagnostic and offending symbol.
3. Determine the isolation boundary: `@MainActor`, custom actor, actor instance isolation, or `nonisolated`.
4. Confirm whether the code is UI-bound or intended to run off the main actor. When spawning unstructured tasks, inspect the synchronous prefix (everything before the first `await`): start on `@MainActor` only when that prefix truly needs main-actor access; otherwise use `Task { @concurrent in ... }` and hop back with `MainActor.run` only after the suspension. A trivial non-main line (for example, `print`) followed by main-actor work in the same prefix is not a reason to use `@concurrent`. For delayed retries, timers, and backoff tasks, separate the waiting from the UI mutation. The sleep often belongs off the main actor even when the final state update belongs on it.

Project settings that change concurrency behavior:

| Setting | SwiftPM (`Package.swift`) | Xcode (`.pbxproj`) |
|---|---|---|
| Language mode | `swiftLanguageVersions` or `-swift-version` (`// swift-tools-version:` is not a reliable proxy) | Swift Language Version |
| Strict concurrency | `.enableExperimentalFeature("StrictConcurrency=targeted")` | `SWIFT_STRICT_CONCURRENCY` |
| Default isolation | `.defaultIsolation(MainActor.self)` | `SWIFT_DEFAULT_ACTOR_ISOLATION` |
| Upcoming features | `.enableUpcomingFeature("NonisolatedNonsendingByDefault")` | `SWIFT_UPCOMING_FEATURE_*` |
| Approachable Concurrency | N/A (use individual upcoming features) | `SWIFT_APPROACHABLE_CONCURRENCY` |

> **Xcode 26 note**: New projects created in Xcode 26 will often start with `SWIFT_DEFAULT_ACTOR_ISOLATION = MainActor` and `SWIFT_APPROACHABLE_CONCURRENCY = YES` enabled by default. Treat these as likely defaults for newly created projects, not as confirmed settings.

If any of these are unknown, ask the developer to confirm them before giving migration-sensitive guidance. Do not guess, even for new Xcode 26 projects.

Guardrails:

- Do not recommend `@MainActor` as a blanket fix. Justify why the code is truly UI-bound.
- Prefer structured concurrency over unstructured tasks. Use `Task.detached` only with a clear reason.
- If recommending `@preconcurrency`, `@unchecked Sendable`, or `nonisolated(unsafe)`, require a documented safety invariant and a follow-up removal plan.
- Optimize for the smallest safe change. Do not refactor unrelated architecture during migration.
- Course references are for deeper learning only. Use them sparingly and only when they clearly help answer the developer's question.

## Quick Fix Mode

Use Quick Fix Mode when all of these are true:

- The issue is localized to one file or one type.
- The isolation boundary is clear.
- The fix can be explained in 1-2 behavior-preserving steps.

Skip Quick Fix Mode when any of these are true:

- Build settings or default isolation are unknown.
- The issue crosses module boundaries or changes public API behavior.
- The likely fix depends on unsafe escape hatches.

## Common Diagnostics

| Diagnostic | First check | Smallest safe fix | Escalate to |
|---|---|---|---|
| `Main actor-isolated ... cannot be used from a nonisolated context` | Is this truly UI-bound? | Isolate the caller to `@MainActor` or use `await MainActor.run { ... }` only when main-actor ownership is correct. | `references/actors.md`, `references/threading.md` |
| `Actor-isolated type does not conform to protocol` | Must the requirement run on the actor? | Prefer isolated conformance (e.g., `extension Foo: @MainActor SomeProtocol`); use `nonisolated` only for truly nonisolated requirements. | `references/actors.md` |
| `Sending value of non-Sendable type ... risks causing data races` | What isolation boundary is being crossed? | Keep access inside one actor, or convert the transferred value to an immutable/value type. | `references/sendable.md`, `references/threading.md` |
| `SwiftLint async_without_await` | Is `async` actually required by protocol, override, or `@concurrent`? | Remove `async`, or use a narrow suppression with rationale. Never add fake awaits. | `references/linting.md` |
| `wait(...) is unavailable from asynchronous contexts` | Is this legacy XCTest async waiting? | Replace with `await fulfillment(of:)` or Swift Testing equivalents. | `references/testing.md` |
| Core Data concurrency warnings | Are `NSManagedObject` instances crossing contexts or actors? | Pass `NSManagedObjectID` or map to a Sendable value type. | `references/core-data.md` |
| `@Observable` isolation or Sendable errors | Is the `@Observable` class annotated with the correct actor? | Add `@MainActor` for UI state; pass Sendable snapshots across boundaries. | `references/observation.md` |
| `Thread.current` unavailable from asynchronous contexts | Are you debugging by thread instead of isolation? | Reason in terms of isolation and use Instruments/debugger instead. | `references/threading.md` |
| SwiftLint concurrency-related warnings | Which specific lint rule triggered? | Use `references/linting.md` for rule intent and preferred fixes; avoid dummy awaits. | `references/linting.md` |
| `... cannot satisfy conformance requirement for a 'Sendable' type parameter` (`SendableMetatype`) | Does the conformance carry global-actor isolation? | Remove actor isolation from the conformance, or avoid passing the metatype across isolation boundaries. See `SendableMetatype` section in `references/actors.md`. | `references/actors.md` |

## When Quick Fixes Fail

1. Gather project settings if not already confirmed.
2. Re-evaluate which isolation boundaries the type crosses.
3. Route to the matching reference file for a deeper fix.
4. If the fix may change behavior, document the invariant and add verification steps.

## Smallest Safe Fixes

Prefer changes that preserve behavior while satisfying data-race safety:

- **UI-bound state**: isolate the type or member to `@MainActor`.
- **Shared mutable state**: move it behind an `actor`, or use `@MainActor` only if the state is UI-owned.
- **Background work**: when work must hop off caller isolation, use an `async` API marked `@concurrent`; when work can safely inherit caller isolation, use `nonisolated` without `@concurrent`. When spawning a `Task`, match entry isolation to its synchronous prefix. If nothing before the first `await` needs the main actor, use `Task { @concurrent in ... }` and hop back via `await MainActor.run { ... }` for the UI update. If the prefix mixes a trivial non-main statement with main-actor work, keep the inherited `@MainActor` start—splitting the cheap line off-main is not worth an extra hop.
- **Sendability issues**: prefer immutable values and explicit boundaries over `@unchecked Sendable`.

## Concurrency Tool Selection

| Need | Tool | Key Guidance |
|---|---|---|
| Single async operation | `async/await` | Default choice for sequential async work |
| Fixed parallel operations | `async let` | Known count at compile time; auto-cancelled on throw |
| Dynamic parallel operations | `withTaskGroup` | Unknown count; structured — cancels children on scope exit |
| Sync → async bridge | `Task { }` | Inherits actor context; use `Task.detached` only with documented reason |
| Shared mutable state | `actor` | Prefer over locks/queues; keep isolated sections small |
| UI-bound state | `@MainActor` | Only for truly UI-related code; justify isolation |

### Common Scenarios

**Network request with UI update**
```swift
Task { @concurrent in
    let data = try await fetchData()
    await MainActor.run { self.updateUI(with: data) }
}
```

**Processing array items in parallel**
```swift
await withTaskGroup(of: ProcessedItem.self) { group in
    for item in items {
        group.addTask { await process(item) }
    }
    for await result in group {
        results.append(result)
    }
}
```

## Task entry isolation

Match a `Task`'s entry isolation to its synchronous prefix (everything from `{` to the first `await`).

- If anything in that prefix needs `@MainActor`, keep the inherited `@MainActor` start.
- If nothing in that prefix needs `@MainActor`, prefer `Task { @concurrent in ... }` and hop back only for UI-owned mutation.

```swift
// ❌ Synchronous prefix is empty; first work hops away
Task {
    await hopToOtherIsolationDomain()
}

// ❌ Synchronous prefix is only `print` (trivial, non-main); first await hops away
Task {
    print("Also not main-thread-bound")
    await hopToOtherIsolationDomain()
}

// ✅ Start off the main actor, hop back only for UI work
Task { @concurrent in
    await hopToOtherIsolationDomain()
    await MainActor.run { updateUI() }
}

// ✅ Synchronous prefix DOES contain main-actor work — keep inheritance
Task {
    print("debug")              // trivial, non-main — rides along
    self.isLoading = true       // needs @MainActor, before any await
    await fetchData()
}
```

## Swift 6 Migration Quick Guide

Key changes in Swift 6:
- **Strict concurrency checking** enabled by default
- **Complete data-race safety** at compile time
- **Sendable requirements** enforced on boundaries
- **Isolation checking** for all async boundaries

### Migration Validation Loop

Apply this cycle for each migration change:

1. **Build** — Run `swift build` or Xcode build to surface new diagnostics
2. **Fix** — Address one category of error at a time (e.g., all Sendable issues first)
3. **Rebuild** — Confirm the fix compiles cleanly before moving on
4. **Test** — Run the test suite to catch regressions (`swift test` or Cmd+U)
5. **Only proceed** to the next file/module when all diagnostics are resolved

If a fix introduces new warnings, resolve them before continuing. Never batch multiple unrelated fixes — keep commits small and reviewable.

For detailed migration steps, see `references/migration.md`.

## Reference Router

Open the smallest reference that matches the question:

- Foundations
  - `references/async-await-basics.md` — async/await syntax, execution order, async let, URLSession patterns
  - `references/tasks.md` — Task lifecycle, cancellation, priorities, task groups, structured vs unstructured
  - `references/actors.md` — Actor isolation, @MainActor, global actors, reentrancy, custom executors, Mutex
  - `references/sendable.md` — Sendable conformance, value/reference types, @unchecked, region isolation
  - `references/threading.md` — Execution model, suspension points, Swift 6.2 isolation behavior
- Streams
  - `references/async-sequences.md` — AsyncSequence, AsyncStream, when to use vs regular async methods
  - `references/async-algorithms.md` — Debounce, throttle, merge, combineLatest, channels, timers
- Applied topics
  - `references/testing.md` — Swift Testing first, XCTest fallback, leak checks
  - `references/performance.md` — Profiling with Instruments, reducing suspension points, execution strategies
  - `references/memory-management.md` — Retain cycles in tasks, memory safety patterns
  - `references/core-data.md` — NSManagedObject sendability, custom executors, isolation conflicts
  - `references/observation.md` — @Observable with @MainActor, cross-isolation access, Sendable constraints
- Migration and tooling
  - `references/migration.md` — Swift 6 migration strategy, closure-to-async conversion, @preconcurrency, FRP migration
  - `references/linting.md` — Concurrency-focused lint rules and SwiftLint `async_without_await`
- Glossary
  - `references/glossary.md` — Quick definitions of core concurrency terms

## Verification Checklist

When changing concurrency code:

1. Re-check build settings before interpreting diagnostics.
2. Build and clear one category of errors before moving on. Do not batch unrelated fixes into the same change.
3. Run tests, especially actor-, lifetime-, and cancellation-sensitive tests.
4. Use Instruments for performance claims instead of guessing.
5. Verify deallocation and cancellation behavior for long-lived tasks.
6. Check `Task.isCancelled` in long-running operations.
7. Never use semaphores or ad hoc locking in async contexts when actor isolation or `Mutex` would express ownership more safely.

---

**Note**: This skill is based on the comprehensive [Swift Concurrency Course](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=skill-footer) by Antoine van der Lee.
````

## File: skills/swift-concurrency/references/_index.md
size: 2150
sha: 0a03426511360d60f65ff7d11638151cd2c996f1
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/_index.md

```text
# Reference Index

Quick navigation for the Swift Concurrency skill.

## Foundations

| File | Use it for |
|---|---|
| `async-await-basics.md` | closure-to-async bridges and foundational async/await usage |
| `tasks.md` | `Task`, cancellation, task groups, structured vs unstructured work |
| `actors.md` | actor isolation, `@MainActor`, reentrancy, isolated conformances |
| `sendable.md` | `Sendable`, `@Sendable`, region isolation, escape hatches |
| `threading.md` | execution model, suspension points, Swift 6.2 isolation behavior |

## Streams

| File | Use it for |
|---|---|
| `async-sequences.md` | deciding between `AsyncSequence`, `AsyncStream`, and one-shot async APIs |
| `async-algorithms.md` | debounce, throttle, merge, `combineLatest`, channels, timers |

## Applied Topics

| File | Use it for |
|---|---|
| `testing.md` | Swift Testing first, XCTest fallback, leak checks |
| `performance.md` | Instruments workflow, actor hops, suspension cost |
| `memory-management.md` | retain cycles, long-lived tasks, cleanup |
| `core-data.md` | `NSManagedObjectID`, `perform`, default isolation conflicts |
| `observation.md` | `@Observable` with `@MainActor`, cross-isolation access, Sendable constraints |

## Migration and Tooling

| File | Use it for |
|---|---|
| `migration.md` | rollout order, build settings, migration guardrails |
| `linting.md` | concurrency-focused lint rules |
| `glossary.md` | quick definitions |

## Problem Router

- "I need to fix a compiler error quickly" → `../SKILL.md`
- "I need to replace a callback with async/await" → `async-await-basics.md`
- "I need to protect shared mutable state" → `actors.md`
- "I need to pass data safely across boundaries" → `sendable.md`
- "I need stream operators" → `async-algorithms.md`
- "I need to understand why code runs where it runs" → `threading.md`
- "I need to stop a leak or lifetime issue" → `memory-management.md`
- "I need to migrate to Swift 6" → `migration.md`
- "I need to test async code" → `testing.md`
- "I need to optimize slow async code" → `performance.md`
- "I need to use @Observable safely with actors" → `observation.md`
```

## File: skills/swift-concurrency/references/actors.md
size: 17545
sha: 0efaa4c3723c9ad56cef6c679dd6b406194e394d
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/actors.md

````text
# Actors

Use this when:

- You need to protect class-based mutable state from concurrent access.
- You are choosing between `actor`, `@MainActor`, `nonisolated`, or `Mutex`.
- You are resolving protocol conformance issues on actor-isolated types.

Skip this file if:

- You mainly need to make a value safe to transfer across boundaries. Use `sendable.md`.
- You are debugging execution threads or suspension behavior. Use `threading.md`.

Jump to:

- Actor Isolation
- Global Actors / @MainActor
- Isolated vs Nonisolated
- Actor Reentrancy
- Isolated Deinit / Isolated Conformances (Swift 6.2+)
- `#isolation` Macro
- Mutex: Alternative to Actors
- Decision Tree

## What is an Actor?

Actors protect mutable state by ensuring only one task accesses it at a time. They're reference types with automatic synchronization.

```swift
actor Counter {
    var value = 0
    
    func increment() {
        value += 1
    }
}
```

**Key guarantee**: Only one task can access mutable state at a time (serialized access).

> **Course Deep Dive**: This topic is covered in detail in [Lesson 5.1: Understanding actors in Swift Concurrency](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## Actor Isolation

### Enforced by compiler

```swift
actor BankAccount {
    var balance: Int = 0
    
    func deposit(_ amount: Int) {
        balance += amount
    }
}

let account = BankAccount()
account.balance += 1 // ❌ Error: can't mutate from outside
await account.deposit(1) // ✅ Must use actor's methods
```

### Reading properties

```swift
let account = BankAccount()
await account.deposit(100)
print(await account.balance) // Must await reads too
```

Always use `await` when accessing actor properties/methods—you don't know if another task is inside.

## Actors vs Classes

### Similarities

- Reference types (copies share same instance)
- Can have properties, methods, initializers
- Can conform to protocols

### Differences

- **No inheritance** (except `NSObject` for Objective-C interop)
- **Automatic isolation** (no manual locks needed)
- **Implicit Sendable** conformance

```swift
// ❌ Can't inherit from actors
actor Base {}
actor Child: Base {} // Error

// ✅ NSObject exception
actor Example: NSObject {} // OK for Objective-C
```

## Global Actors

Shared isolation domain across types, functions, and properties.

### @MainActor

Ensures execution on main thread:

```swift
@MainActor
final class ViewModel {
    var items: [Item] = []
}

@MainActor
func updateUI() {
    // Always runs on main thread
}

@MainActor
var title: String = ""
```

### Custom global actors

```swift
@globalActor
actor ImageProcessing {
    static let shared = ImageProcessing()
    private init() {} // Prevent duplicate instances
}

@ImageProcessing
final class ImageCache {
    var images: [URL: Data] = [:]
}

@ImageProcessing
func applyFilter(_ image: UIImage) -> UIImage {
    // All image processing serialized
}
```

**Use private init** to prevent creating multiple executors.

> **Course Deep Dive**: This topic is covered in detail in [Lesson 5.2: An introduction to Global Actors](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## @MainActor Best Practices

### When to use

UI-related code that must run on main thread:

```swift
@MainActor
final class ContentViewModel: ObservableObject {
    @Published var items: [Item] = []
}
```

### Replacing DispatchQueue.main

```swift
// Old way
DispatchQueue.main.async {
    // Update UI
}

// Modern way
await MainActor.run {
    // Update UI
}

// Better: Use attribute
@MainActor
func updateUI() {
    // Automatically on main thread
}
```

### MainActor.assumeIsolated

**Use sparingly** - assumes you're on main thread, crashes if not:

```swift
func methodB() {
    assert(Thread.isMainThread) // Validate assumption
    
    MainActor.assumeIsolated {
        someMainActorMethod()
    }
}
```

**Prefer**: Explicit `@MainActor` or `await MainActor.run` over `assumeIsolated`.

> **Course Deep Dive**: This topic is covered in detail in [Lesson 5.3: When and how to use @MainActor](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## Isolated vs Nonisolated

### Default: Isolated

Actor methods are isolated by default:

```swift
actor BankAccount {
    var balance: Double
    
    // Implicitly isolated
    func deposit(_ amount: Double) {
        balance += amount
    }
}
```

### Isolated parameters

Reduce suspension points by inheriting caller's isolation:

```swift
struct Charger {
    static func charge(
        amount: Double,
        from account: isolated BankAccount
    ) async throws -> Double {
        // No await needed - we're isolated to account
        try account.withdraw(amount: amount)
        return account.balance
    }
}
```

### Isolated closures

```swift
actor Database {
    func transaction<T>(
        _ operation: @Sendable (_ db: isolated Database) throws -> T
    ) throws -> T {
        beginTransaction()
        let result = try operation(self)
        commitTransaction()
        return result
    }
}

// Usage: Multiple operations, one await
try await database.transaction { db in
    db.insert(item1)
    db.insert(item2)
    db.insert(item3)
}
```

### Generic isolated extension

```swift
extension Actor {
    func performInIsolation<T: Sendable>(
        _ block: @Sendable (_ actor: isolated Self) throws -> T
    ) async rethrows -> T {
        try block(self)
    }
}

// Usage
try await bankAccount.performInIsolation { account in
    try account.withdraw(amount: 20)
    print("Balance: \(account.balance)")
}
```

### Nonisolated

Opt out of isolation for immutable data:

```swift
actor BankAccount {
    let accountHolder: String
    
    nonisolated var details: String {
        "Account: \(accountHolder)"
    }
}

// No await needed
print(account.details)
```

### Protocol conformance

```swift
extension BankAccount: CustomStringConvertible {
    nonisolated var description: String {
        "Account: \(accountHolder)"
    }
}
```

> **Course Deep Dive**: This topic is covered in detail in [Lesson 5.4: Isolated vs. non-isolated access in actors](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## Isolated Deinit (Swift 6.2+)

Clean up actor state on deallocation:

```swift
actor FileDownloader {
    var downloadTask: Task<Void, Error>?
    
    isolated deinit {
        downloadTask?.cancel() // Can call isolated methods
    }
}
```

**Requires**: iOS 18.4+, macOS 15.4+

> **Course Deep Dive**: This topic is covered in detail in [Lesson 5.5: Using Isolated synchronous deinit](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## Global Actor Isolated Conformance (Swift 6.2+)

Protocol conformance respecting actor isolation:

```swift
@MainActor
final class PersonViewModel {
    let id: UUID
    var name: String
}

extension PersonViewModel: @MainActor Equatable {
    static func == (lhs: PersonViewModel, rhs: PersonViewModel) -> Bool {
        lhs.id == rhs.id && lhs.name == rhs.name
    }
}
```

**Enable**: `InferIsolatedConformances` upcoming feature.

> **Course Deep Dive**: This topic is covered in detail in [Lesson 5.6: Adding isolated conformance to protocols](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

### `SendableMetatype` Error with Isolated Conformances

Isolated conformances **cannot** satisfy a `SendableMetatype` requirement. This surfaces when you pass `MyClass.self` to a generic function whose type parameter requires `Sendable`.

```swift
protocol P {
    static func doSomething()
}

func doSomethingStatic<T: P & SendableMetatype>(_ type: T.Type) { }  // explicitly requires a Sendable type/metatype

@MainActor
class C { }

extension C: @MainActor P {
    static func doSomething() { }
}

@MainActor
func test(c: C) {
    doSomethingStatic(C.self)
    // ❌ main actor-isolated conformance of 'C' to 'P' cannot satisfy
    //    conformance requirement for a 'Sendable' type parameter
}
```

**Fix options**:

1. Remove actor isolation from the original conformance if the protocol requirements don't access actor state:

```swift
@MainActor
class C: P {
    nonisolated static func doSomething() { }  // ✅ Non-isolated requirement on a non-isolated conformance
}
```

2. Avoid passing the metatype across isolation boundaries — call the static method directly rather than routing through the generic function.

3. Make the generic function actor-aware so it accepts an isolated conformance (requires changing the callee's signature).

## Actor Reentrancy

**Critical**: State can change between suspension points.

```swift
actor BankAccount {
    var balance: Double
    
    func deposit(amount: Double) async {
        balance += amount
        
        // ⚠️ Actor unlocked during await
        await logActivity("Deposited \(amount)")
        
        // ⚠️ Balance may have changed!
        print("Balance: \(balance)")
    }
}
```

### Problem

```swift
async let _ = account.deposit(50)
async let _ = account.deposit(50)
async let _ = account.deposit(50)

// May print same balance three times:
// Balance: 150
// Balance: 150
// Balance: 150
```

### Solution

Complete actor work before suspending:

```swift
func deposit(amount: Double) async {
    balance += amount
    print("Balance: \(balance)") // Before suspension
    
    await logActivity("Deposited \(amount)")
}
```

**Rule**: Don't assume state is unchanged after `await`.

> **Course Deep Dive**: This topic is covered in detail in [Lesson 5.7: Understanding actor reentrancy](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## #isolation Macro

Inherit caller's isolation for generic code:

```swift
extension Collection where Element: Sendable {
    func sequentialMap<Result: Sendable>(
        isolation: isolated (any Actor)? = #isolation,
        transform: (Element) async -> Result
    ) async -> [Result] {
        var results: [Result] = []
        for element in self {
            results.append(await transform(element))
        }
        return results
    }
}

// Usage from @MainActor context
Task { @MainActor in
    let names = ["Alice", "Bob"]
    let results = await names.sequentialMap { name in
        await process(name) // Inherits @MainActor
    }
}
```

**Benefits**: Avoids unnecessary suspensions, allows non-Sendable data.

### Task Closures and Isolation Inheritance

When spawning unstructured `Task` closures that need to work with `non-Sendable` types, you must capture the isolation parameter to inherit the caller's isolation context.

**Problem**: `Task` closures are `@Sendable`, which prevents capturing `non-Sendable` types:

```swift
func process(delegate: NonSendableDelegate) {
  Task {
    delegate.doWork() // ❌ Error: capturing non-Sendable type
  }
}
```

**Solution**: Use `#isolation` parameter and capture it inside the `Task`:

```swift
func process(
  delegate: NonSendableDelegate,
  isolation: isolated (any Actor)? = #isolation
) {
  Task {
    _ = isolation  // Forces capture, Task inherits caller's isolation
    delegate.doWork()  // ✅ Safe - running on caller's actor
  }
}
```

**Why `_ = isolation` is required**: Per [SE-0420](https://github.com/swiftlang/swift-evolution/blob/main/proposals/0420-inheritance-of-actor-isolation.md), `Task` closures only inherit isolation when "a non-optional binding of an isolated parameter is captured by the closure." The `_ = isolation` statement forces this capture. The capture list syntax `[isolation]` should work but currently does not.

**When to use this pattern**:
- Spawning `Task`s that work with `non-Sendable` delegate objects
- Fire-and-forget async work that needs access to caller's state
- Bridging callback-based APIs to async streams while keeping delegates alive

**Note**: This pattern keeps the `non-Sendable` value alive and accessible within the `Task`. The `Task` runs on the caller's isolation domain, so no cross-isolation "sending" occurs.

> **Course Deep Dive**: This topic is covered in detail in [Lesson 5.8: Inheritance of actor isolation using the #isolation macro](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## Custom Actor Executors

**Advanced**: Control how actor schedules work.

### Serial executor

```swift
final class DispatchQueueExecutor: SerialExecutor {
    private let queue: DispatchQueue
    
    init(queue: DispatchQueue) {
        self.queue = queue
    }
    
    func enqueue(_ job: consuming ExecutorJob) {
        let unownedJob = UnownedJob(job)
        let executor = asUnownedSerialExecutor()
        
        queue.async {
            unownedJob.runSynchronously(on: executor)
        }
    }
}

actor LoggingActor {
    private let executor: DispatchQueueExecutor
    
    nonisolated var unownedExecutor: UnownedSerialExecutor {
        executor.asUnownedSerialExecutor()
    }
    
    init(queue: DispatchQueue) {
        executor = DispatchQueueExecutor(queue: queue)
    }
}
```

### When to use

- Integration with legacy DispatchQueue-based code
- Specific thread requirements (e.g., C++ interop)
- Custom scheduling logic

**Default executor is usually sufficient.**

> **Course Deep Dive**: This topic is covered in detail in [Lesson 5.9: Using a custom actor executor](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## Mutex: Alternative to Actors

Synchronous locking without async/await overhead (iOS 18+, macOS 15+).

### Basic usage

```swift
import Synchronization

final class Counter {
    private let count = Mutex<Int>(0)
    
    var currentCount: Int {
        count.withLock { $0 }
    }
    
    func increment() {
        count.withLock { $0 += 1 }
    }
}
```

### Sendable access to non-Sendable types

```swift
final class TouchesCapturer: Sendable {
    let path = Mutex<NSBezierPath>(NSBezierPath())
    
    func storeTouch(_ point: NSPoint) {
        path.withLock { path in
            path.move(to: point)
        }
    }
}
```

### Error handling

```swift
func decrement() throws {
    try count.withLock { count in
        guard count > 0 else {
            throw Error.reachedZero
        }
        count -= 1
    }
}
```

### Mutex vs Actor

| Feature | Mutex | Actor |
|---------|-------|-------|
| Synchronous | ✅ | ❌ (requires await) |
| Async support | ❌ | ✅ |
| Thread blocking | ✅ | ❌ (cooperative) |
| Fine-grained locking | ✅ | ❌ (whole actor) |
| Legacy code integration | ✅ | ❌ |

**Use Mutex when**:
- Need synchronous access
- Working with legacy non-async APIs
- Fine-grained locking required
- Low contention, short critical sections

**Use Actor when**:
- Can adopt async/await
- Need logical isolation
- Working in async context

> **Course Deep Dive**: This topic is covered in detail in [Lesson 5.10: Using a Mutex as an alternative to actors](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## Common Patterns

### View model with @MainActor

```swift
@MainActor
final class ContentViewModel: ObservableObject {
    @Published var items: [Item] = []
    
    func loadItems() async {
        items = try await api.fetchItems()
    }
}
```

### Background processing with custom actor

```swift
@ImageProcessing
final class ImageProcessor {
    func process(_ images: [UIImage]) async -> [UIImage] {
        images.map { applyFilters($0) }
    }
}
```

### Mixed isolation

```swift
actor DataStore {
    private var items: [Item] = []
    
    func add(_ item: Item) {
        items.append(item)
    }
    
    nonisolated func itemCount() -> Int {
        // ❌ Can't access items
        return 0
    }
}
```

### Transaction pattern

```swift
actor Database {
    func transaction<T>(
        _ operation: @Sendable (_ db: isolated Database) throws -> T
    ) throws -> T {
        beginTransaction()
        defer { commitTransaction() }
        return try operation(self)
    }
}
```

## Best Practices

1. **Prefer actors over manual locks** for async code
2. **Use @MainActor for UI** - all view models, UI updates
3. **Minimize work in actors** - keep critical sections short
4. **Watch for reentrancy** - don't assume state unchanged after await
5. **Use nonisolated sparingly** - only for truly immutable data
6. **Avoid assumeIsolated** - prefer explicit isolation
7. **Custom executors are rare** - default is usually best
8. **Consider Mutex for sync code** - when async overhead not needed
9. **Complete actor work before suspending** - prevent reentrancy bugs
10. **Use isolated parameters** - reduce suspension points

## Decision Tree

```
Need thread-safe mutable state?
├─ Async context?
│  ├─ Single instance? → Actor
│  ├─ Global/shared? → Global Actor (@MainActor, custom)
│  └─ UI-related? → @MainActor
│
└─ Synchronous context?
   ├─ Can refactor to async? → Actor
   ├─ Legacy code integration? → Mutex
   └─ Fine-grained locking? → Mutex
```

## Further Learning

For migration strategies, advanced patterns, and real-world examples, see [Swift Concurrency Course](https://www.swiftconcurrencycourse.com).

````

## File: skills/swift-concurrency/references/async-algorithms.md
size: 20705
sha: 0dd545af772946da2679a5b663c336606ace3a64
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/async-algorithms.md

````text
# AsyncAlgorithms Package

Use this when:

- You need time-based operators (debounce, throttle, timers).
- You need to combine multiple async sequences (merge, combineLatest, zip).
- You are migrating from Combine or RxSwift operators to Swift Concurrency equivalents.

Skip this file if:

- You need basic `AsyncStream` bridging for callbacks or delegates. Use `async-sequences.md`.
- You are choosing between `Task`, `async let`, or task groups. Use `tasks.md`.

Jump to:

- Quick Start
- Time-Based Operators
- Combining Operators
- Multi-Consumer Scenarios
- Combine Migration Guide
- Best Practices

---

## Quick Start

Top 5 most common operators:

```swift
import AsyncAlgorithms

// 1. Debounce rapid inputs
for await query in searchQueryStream.debounce(for: .milliseconds(500)) {
    await performSearch(query)
}

// 2. Throttle repeated actions
for await _ in buttonClicks.throttle(for: .seconds(1)) {
    await performAction()
}

// 3. Merge multiple independent streams
for await message in chat1Messages.merge(chat2Messages) {
    display(message)
}

// 4. Combine dependent values
for await (username, email) in usernameStream.combineLatest(emailStream) {
    validateForm(username: username, email: email)
}

// 5. Zip paired operations
for await (image, metadata) in imageStream.zip(metadataStream) {
    await cache(image: image, metadata: metadata)
}
```

> **See**: [AsyncAlgorithms on GitHub](https://github.com/apple/swift-async-algorithms)

---

## Overview & Installation

### What is AsyncAlgorithms?

Extends Swift's AsyncSequence with time-based operators, stream combination tools, and multi-consumer primitives.

**Use for**:
- Time-based operations: debounce, throttle, timers
- Combining streams: merge, combineLatest, zip, chain
- Multi-consumer scenarios: AsyncChannel for backpressure
- Specific operators: removeDuplicates, chunks, adjacentPairs, compacted

**Use standard library for**:
- Bridging callbacks: AsyncStream
- Simple iteration: for await in sequence
- Single-value operations: async/await

### Installation

```swift
dependencies: [
    .package(url: "https://github.com/apple/swift-async-algorithms", from: "1.0.0")
]

targets: [
    .target(
        name: "MyTarget",
        dependencies: [
            .product(name: "AsyncAlgorithms", package: "swift-async-algorithms")
        ]
    )
]
```

Import:

```swift
import AsyncAlgorithms
```

---

## Time-Based Operators

### debounce(for:tolerance:clock:)

Wait for inactivity before emitting. Use for rapid inputs like search fields.

#### Example: ArticleSearcher

```swift
import AsyncAlgorithms

@Observable
final class ArticleSearcher {
    @MainActor private(set) var results: [Article] = []
    private var searchQueryContinuation: AsyncStream<String>.Continuation?

    private lazy var searchQueryStream: AsyncStream<String> = {
        AsyncStream { continuation in
            searchQueryContinuation = continuation
        }
    }()

    func search(_ query: String) {
        searchQueryContinuation?.yield(query)
    }

    func startDebouncedSearch() {
        Task { @MainActor in
            for await query in searchQueryStream.debounce(for: .milliseconds(500)) {
                self.results = []
                self.results = await APIClient.searchArticles(query)
            }
        }
    }
}
```

**Benefits**: Automatic cancellation, backpressure, cleaner than manual Task.sleep.

#### ❌ Anti-Pattern

```swift
// Bad: Every keystroke spawns new task
func search(_ query: String) {
    Task {
        try? await Task.sleep(for: .milliseconds(500))
        await performSearch(query)
    }
}
```

**Problem**: Multiple tasks execute simultaneously, causing out-of-order results.

**Solution**: Use `debounce()` for automatic backpressure.

---

### throttle(for:clock:reducing:)

Emit at most one value per interval. Use for repeated actions like button taps.

#### Example: Like Button

```swift
import AsyncAlgorithms

struct LikeButton: View {
    @State private var tapStream = AsyncStream<Void> { continuation in
        // Continuation stored externally
    }
    @State private var isLiked = false

    var body: some View {
        Button(action: {
            tapStream.continuation?.yield()
        }) {
            Image(systemName: isLiked ? "heart.fill" : "heart")
        }
        .task {
            await handleThrottledTaps()
        }
    }

    private func handleThrottledTaps() async {
        for await _ in tapStream.throttle(for: .seconds(1)) {
            await toggleLike()
        }
    }

    private func toggleLike() async {
        isLiked.toggle()
        await APIClient.updateLikeStatus(isLiked: isLiked)
    }
}
```

#### Understanding reducing Parameter

```swift
// .latest (default): Keep most recent value
for await value in events.throttle(for: .seconds(1)) {
    process(value)
}

// .oldest: Keep first value
for await value in events.throttle(for: .seconds(1), reducing: .oldest) {
    process(value)
}

// Custom: Sum all values
for await value in events.throttle(for: .seconds(1)) { $0 + $1 } {
    process(value)
}
```

---

### AsyncTimerSequence

Emit values at regular intervals. Use for periodic refresh or countdown timers.

#### Example: Feed Refresh

```swift
import AsyncAlgorithms

@MainActor @Observable
final class FeedViewModel {
    private(set) var articles: [Article] = []
    private var refreshTask: Task<Void, Never>?

    func startAutoRefresh() {
        refreshTask = Task {
            for await _ in AsyncTimerSequence(interval: .seconds(30)) {
                await refreshFeed()
            }
        }
    }

    private func refreshFeed() async {
        articles = await APIClient.fetchLatestArticles()
    }
}
```

#### ❌ Anti-Pattern

```swift
// Bad: Manual timer implementation
func startTimer() {
    Task {
        while !Task.isCancelled {
            performAction()
            try? await Task.sleep(for: .seconds(1))
        }
    }
}
```

**Solution**: Use `AsyncTimerSequence`.

---

## Combining Operators

### merge(_:...)

Combine sequences into one, emitting as they arrive. **Stable operator ✅**

Use for independent data sources that don't depend on each other.

#### Example: Multi-Room Chat

```swift
import AsyncAlgorithms

actor ChatManager {
    private var messageContinuations: [String: AsyncStream<ChatMessage>.Continuation] = [:]

    func getMessagesStream(roomID: String) -> AsyncStream<ChatMessage> {
        AsyncStream { continuation in
            messageContinuations[roomID] = continuation
        }
    }

    func receiveMessage(_ message: ChatMessage) {
        messageContinuations[message.roomID]?.yield(message)
    }

    func startMonitoring(rooms: [String]) -> AsyncStream<ChatMessage> {
        let streams = rooms.map { getMessagesStream(roomID: $0) }
        return streams.merge()
    }
}

// Usage
let manager = ChatManager()
let mergedMessages = await manager.startMonitoring(rooms: ["general", "random"])

for await message in mergedMessages {
    print("[\(message.roomID)] \(message.text)")
}
```

**Behavior**: Values emit as they arrive from any source. Order interleaved by timing. Cancellation propagates to all sources.

---

### combineLatest(_:...)

Combine sequences, emitting tuple when any source emits. Always uses latest values. **Stable operator ✅**

Use for dependent values that need synchronization.

#### Example: Form Validation

```swift
import AsyncAlgorithms

struct SignupForm: View {
    @State private var usernameStream = AsyncStream<String> { /* ... */ }
    @State private var emailStream = AsyncStream<String> { /* ... */ }
    @State private var passwordStream = AsyncStream<String> { /* ... */ }
    @State private var formState = FormState.incomplete

    var body: some View {
        Form {
            TextField("Username", text: $username)
            TextField("Email", text: $email)
            SecureField("Password", text: $password)
        }
        .task {
            await validateForm()
        }
    }

    private func validateForm() async {
        for await (username, email, password) in
                usernameStream.combineLatest(emailStream, passwordStream)
        {
            formState = await validate(
                username: username,
                email: email,
                password: password
            )
        }
    }
}
```

#### ❌ Anti-Pattern

```swift
// Bad: Manual value combining
actor FormValidator {
    private var currentUsername: String = ""
    private var currentEmail: String = ""

    func updateUsername(_ username: String) {
        currentUsername = username
        checkForm()
    }
}
```

**Solution**: Use `combineLatest()`.

---

### zip(_:...)

Combine sequences by pairing elements in order. **Stable operator ✅**

#### Example: Image + Metadata

```swift
import AsyncAlgorithms

struct ImageLoader {
    func loadImagesWithMetadata(urls: [URL]) async throws -> [LoadedImage] {
        let imageStream = AsyncThrowingStream<UIImage, Error> { continuation in
            Task {
                for url in urls {
                    let image = try await downloadImage(from: url)
                    continuation.yield(image)
                }
                continuation.finish()
            }
        }

        let metadataStream = AsyncThrowingStream<ImageMetadata, Error> { continuation in
            Task {
                for url in urls {
                    let metadata = try await fetchMetadata(for: url)
                    continuation.yield(metadata)
                }
                continuation.finish()
            }
        }

        var results: [LoadedImage] = []
        for try await (image, metadata) in imageStream.zip(metadataStream) {
            results.append(LoadedImage(image: image, metadata: metadata))
        }
        return results
    }
}
```

**Behavior**: Emits tuple when all sequences emit. Maintains order. Finishes when shortest sequence finishes.

---

### chain(_:...)

Concatenate sequences sequentially. **Stable operator ✅**

#### Example: Paginated Loading

```swift
import AsyncAlgorithms

struct ArticlePaginator {
    func loadAllArticles() -> AsyncStream<[Article]> {
        AsyncStream { continuation in
            Task {
                var page = 1
                var hasMore = true
                while hasMore {
                    let articles = try await fetchPage(page: page)
                    continuation.yield(articles)
                    hasMore = articles.count == 20
                    page += 1
                }
                continuation.finish()
            }
        }
    }
}

// Usage: Chain cache + network
for await articles in loadFromCacheStream().chain(loadFromNetworkStream()) {
    display(articles)
}
```

**Behavior**: Emits all values from first sequence before starting second.

---

## Utility Operators

### removeDuplicates()

Remove adjacent duplicates. **Stable operator ✅**

```swift
import AsyncAlgorithms

actor ChatHistory {
    private var messageStream = AsyncStream<ChatMessage> { /* ... */ }

    func getUniqueMessages() -> AsyncStream<ChatMessage> {
        messageStream.removeDuplicates()
    }
}
```

---

### chunks() and chunked()

Collect values into batches. **Stable operator ✅**

```swift
import AsyncAlgorithms

struct BatchProcessor {
    func processLargeDataset(dataStream: AsyncStream<DataItem>) async {
        for await batch in dataStream.chunks(count: 100) {
            await processBatch(batch)
        }
    }

    func chunkedByTime(dataStream: AsyncStream<DataItem>) async {
        for await batch in dataStream.chunked(by: .seconds(5)) {
            await processBatch(batch)
        }
    }
}
```

---

### compacted() and adjacentPairs()

```swift
import AsyncAlgorithms

// Remove nil values
for await value in optionalValuesStream.compacted() {
    process(value)
}

// Pair adjacent elements
for await (previous, current) in valuesStream.adjacentPairs() {
    let difference = current - previous
}
```

---

## Multi-Consumer Scenarios

### AsyncChannel

AsyncSequence with backpressure. **Stable operator ✅**

Use for producer-consumer patterns with flow control.

#### Example: Message Queue

```swift
import AsyncAlgorithms

actor MessageQueue {
    private let channel = AsyncChannel<Message>()

    func getMessages() -> AsyncStream<Message> {
        channel
    }

    func enqueue(_ message: Message) async {
        await channel.send(message)
    }

    func startProcessing() {
        Task {
            for await message in channel {
                await process(message)
            }
        }
    }
}

// Multiple producers
let queue = MessageQueue()
Task { await queue.enqueue(Message(type: .userAction, content: "tap")) }
Task { await queue.enqueue(Message(type: .network, content: "data")) }
queue.startProcessing()
```

#### ❌ Anti-Pattern

```swift
// Bad: Values split unpredictably
let stream = AsyncStream<Int> { continuation in
    for i in 1...10 {
        continuation.yield(i)
    }
    continuation.finish()
}

Task { for await value in stream { print("Consumer 1: \(value)") } }
Task { for await value in stream { print("Consumer 2: \(value)") } }
```

**Problem**: Each value goes to only one consumer.

**Solution**: Use `AsyncChannel` for multi-consumer scenarios.

---

### AsyncThrowingChannel

Like AsyncChannel but can emit errors. **Stable operator ✅**

#### Example: WebSocket

```swift
import AsyncAlgorithms

actor WebSocketConnection {
    private let channel = AsyncThrowingChannel<WebSocketMessage, Error>()

    func getMessages() -> AsyncThrowingStream<WebSocketMessage, Error> {
        channel
    }

    func receiveMessage(_ message: WebSocketMessage) async {
        await channel.send(message)
    }

    func reportError(_ error: Error) async {
        await channel.finish(throwing: error)
    }
}

// Usage
do {
    for await message in connection.getMessages() {
        handle(message)
    }
} catch {
    print("WebSocket error: \(error)")
}
```

---

## Combine Migration Guide

### Operator Mapping Table

| Combine | AsyncAlgorithms | Status | Alternative |
|---------|-----------------|---------|-------------|
| `.debounce()` | `debounce()` | ✅ Stable | - |
| `.throttle()` | `throttle()` | ✅ Stable | - |
| `.merge()` | `merge()` | ✅ Stable | - |
| `.combineLatest()` | `combineLatest()` | ✅ Stable | - |
| `.zip()` | `zip()` | ✅ Stable | - |
| `.concat()` | `chain()` | ✅ Stable | - |
| `.removeDuplicates()` | `removeDuplicates()` | ✅ Stable | - |
| `.timer()` | `AsyncTimerSequence` | ✅ Stable | - |
| `.share()` | - | - | `AsyncChannel` |
| `.flatMap()` | - | - | `TaskGroup` |
| `.receive(on:)` | - | - | `Task` / `@MainActor` |
| `.eraseToAnyPublisher()` | - | - | `any AsyncSequence` |

---

### Migration Examples

#### Example 1: ArticleSearcher

**Before: Combine**

```swift
import Combine

final class ArticleSearcher: ObservableObject {
    @Published private(set) var results: [Article] = []
    @Published var searchQuery = ""

    init() {
        $searchQuery
            .debounce(for: .milliseconds(500), scheduler: DispatchQueue.main)
            .removeDuplicates()
            .flatMap { query in
                APIClient.searchArticles(query)
                    .catch { _ in Just([]) }
            }
            .receive(on: DispatchQueue.main)
            .assign(to: &$results)
    }
}
```

**After: AsyncAlgorithms**

```swift
import AsyncAlgorithms

@Observable
final class ArticleSearcher {
    @MainActor private(set) var results: [Article] = []
    private var searchQueryContinuation: AsyncStream<String>.Continuation?

    private lazy var searchQueryStream: AsyncStream<String> = {
        AsyncStream { continuation in
            searchQueryContinuation = continuation
        }
    }()

    func search(_ query: String) {
        searchQueryContinuation?.yield(query)
    }

    func startDebouncedSearch() {
        Task { @MainActor in
            for await query in searchQueryStream
                .debounce(for: .milliseconds(500))
                .removeDuplicates()
            {
                do {
                    self.results = try await APIClient.searchArticles(query)
                } catch {
                    self.results = []
                }
            }
        }
    }
}
```

**Benefits**: Simpler error handling, no cancellables, automatic cancellation.

---

#### Example 2: Multi-Source Loading

**Before: Combine Merge**

```swift
import Combine

final class ArticleLoader: ObservableObject {
    @Published private(set) var items: [Item] = []

    func loadAllSources() {
        let source1 = APIClient.fetchItems(from: .source1)
        let source2 = APIClient.fetchItems(from: .source2)

        Publishers.Merge(source1, source2)
            .scan([]) { accumulated, new in
                accumulated + new
            }
            .receive(on: DispatchQueue.main)
            .assign(to: &$items)
    }
}
```

**After: TaskGroup**

```swift
import AsyncAlgorithms

@Observable
final class ArticleLoader {
    @MainActor private(set) var items: [Item] = []

    func loadAllSourcesParallel() async {
        await withTaskGroup(of: [Item].self) { group in
            group.addTask {
                await APIClient.fetchItems(from: .source1)
            }
            group.addTask {
                await APIClient.fetchItems(from: .source2)
            }

            for await newItems in group {
                items.append(contentsOf: newItems)
            }
        }
    }
}
```

**Key difference**: For parallel execution, use `TaskGroup` instead of `flatMap`.

---

#### Example 3: Form Validation

**Before: Combine**

```swift
import Combine

final class FormValidator: ObservableObject {
    @Published var username = ""
    @Published var email = ""

    @Published private(set) var formState: FormState = .incomplete

    init() {
        Publishers.CombineLatest2($username, $email)
            .map { username, email in
                validate(username: username, email: email)
            }
            .assign(to: &$formState)
    }
}
```

**After: AsyncAlgorithms or async let**

```swift
import AsyncAlgorithms

@Observable
final class FormValidator {
    var username = ""
    var email = ""

    @MainActor private(set) var formState: FormState = .incomplete

    // Option 1: combineLatest for stream-based validation
    func startStreamValidation() {
        Task { @MainActor in
            for await (username, email) in
                    usernameStream.combineLatest(emailStream)
            {
                self.formState = validate(
                    username: username,
                    email: email
                )
            }
        }
    }

    // Option 2: async let for simple validation
    func validateForm() async {
        let (username, email) = await (username, email)
        formState = validate(
            username: username,
            email: email
        )
    }
}
```

**Choose**:
- `combineLatest()`: Continuous validation as fields change
- `async let`: One-time validation when all values available

---

## Common Mistakes Agents Make

- **Manual debounce with `Task.sleep`**: This creates multiple concurrent tasks and risks out-of-order results. Use the stream-based `debounce(for:)` operator from AsyncAlgorithms instead.
- **Sharing `AsyncStream` across multiple consumers**: Values split unpredictably between consumers. Use `AsyncChannel` for multi-consumer scenarios with backpressure. Note: `AsyncChannel` is point-to-point, not broadcast like Combine's `.share()`.
- **Looking for a `.flatMap` equivalent**: Use `TaskGroup` for fan-out; the semantics differ from Combine/Rx `flatMap`.
- **Looking for `.receive(on:)` equivalent**: Use `@MainActor` or `Task` context for isolation instead.

## Best Practices

1. **Use time-based operators** for rapid inputs: debounce() for search, throttle() for buttons
2. **Combine streams** with merge/combineLatest instead of manual state management
3. **Use AsyncChannel** for multi-consumer scenarios with backpressure
4. **Ensure Sendable conformance** when using operators across isolation boundaries
5. **Leverage cancellation** - Task cancellation propagates through all operators
6. **Choose right tool**: AsyncAlgorithms for complex streams, AsyncStream for bridging callbacks
7. **Avoid manual sleep loops** - use AsyncTimerSequence instead

---

## Further Learning

- [AsyncAlgorithms Documentation](https://github.com/apple/swift-async-algorithms)
- [Combine Migration Guide](migration.md)
- [Async Sequences](async-sequences.md)
- [Tasks](tasks.md) - Task groups and structured concurrency
````

## File: skills/swift-concurrency/references/async-await-basics.md
size: 7686
sha: cd9c928286048e15804fc5e32e5db52f328f3249
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/async-await-basics.md

````text
# Async/Await Basics

Use this when:

- You are starting fresh with async/await and need foundational patterns.
- You are converting callback-based code to async/await.
- You need to understand execution order and the sync-to-async bridge.

Skip this file if:

- You need parallel execution with task groups or `async let`. Use `tasks.md`.
- You need stream-based async iteration. Use `async-sequences.md`.

Jump to:

- Function Declaration
- Execution Order
- Parallel Execution with async let
- URLSession with Async/Await
- Migration Strategy

## Function Declaration

Mark functions with `async` to indicate asynchronous work:

```swift
func fetchData() async -> Data {
    // async work
}

func fetchData() async throws -> Data {
    // async work that can fail
}
```

**Key benefit over closures**: The compiler enforces return values. No forgotten completion handlers.

> **Course Deep Dive**: This topic is covered in detail in [Lesson 2.1: Introduction to async/await syntax](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## Calling Async Functions

### From synchronous context

Use `Task` to bridge from sync to async:

```swift
Task {
    let data = try await fetchData()
}
```

### From async context

Use `await` directly:

```swift
func processData() async throws {
    let data = try await fetchData()
    // process data
}
```

## Execution Order

Structured concurrency executes top-to-bottom in the order you expect:

```swift
let first = try await fetchData(1)   // Waits for completion
let second = try await fetchData(2)  // Starts after first completes
let third = try await fetchData(3)   // Starts after second completes
```

Code after `await` only executes once the awaited function returns.

> **Course Deep Dive**: This topic is covered in detail in [Lesson 2.2: Understanding the order of execution](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## Parallel Execution with async let

Use `async let` to run multiple operations concurrently:

```swift
async let data1 = fetchData(1)
async let data2 = fetchData(2)
async let data3 = fetchData(3)

let results = try await [data1, data2, data3]
```

### How async let works

- **Starts immediately**: The function executes right away, even before `await`
- **Structured concurrency**: Automatically canceled when leaving scope
- **Error handling**: If one fails, others are implicitly canceled when awaiting grouped results
- **No redundant keywords**: Don't use `try await` in the `async let` line itself

```swift
// Redundant - avoid this
async let data = try await fetchData()

// Correct - errors handled at await point
async let data = fetchData()
let result = try await data
```

### When to use async let

**Use when:**
- Tasks don't depend on each other
- Number of tasks known at compile-time
- Want automatic cancellation on scope exit

**Avoid when:**
- Tasks must run sequentially
- Need dynamic task spawning (use `TaskGroup`)
- Need manual cancellation control

### Limitations

- Cannot use at top-level declarations (only within function bodies)
- Tasks not explicitly awaited may be canceled implicitly

> **Course Deep Dive**: This topic is covered in detail in [Lesson 2.3: Calling async functions in parallel using async let](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## URLSession with Async/Await

URLSession provides async alternatives to closure-based APIs:

```swift
// Closure-based (old)
URLSession.shared.dataTask(with: request) { data, response, error in
    guard let data = data, error == nil else { return }
    // handle response
}.resume()

// Async/await (modern)
let (data, response) = try await URLSession.shared.data(for: request)
```

### Benefits over closures

- No optional `data` or `response` to unwrap
- Automatic error throwing
- Compiler enforces return values
- Simpler error handling with do-catch

### Complete network request pattern

```swift
func fetchUser(id: Int) async throws -> User {
    let url = URL(string: "https://api.example.com/users/\(id)")!
    var request = URLRequest(url: url)
    request.httpMethod = "GET"
    
    let (data, response) = try await URLSession.shared.data(for: request)
    
    guard let httpResponse = response as? HTTPURLResponse,
          (200...299).contains(httpResponse.statusCode) else {
        throw NetworkError.invalidResponse
    }
    
    return try JSONDecoder().decode(User.self, from: data)
}
```

### POST requests with JSON

```swift
func createUser(_ user: User) async throws -> User {
    let url = URL(string: "https://api.example.com/users")!
    var request = URLRequest(url: url)
    request.httpMethod = "POST"
    request.setValue("application/json", forHTTPHeaderField: "Content-Type")
    request.httpBody = try JSONEncoder().encode(user)
    
    let (data, response) = try await URLSession.shared.data(for: request)
    
    guard let httpResponse = response as? HTTPURLResponse,
          (200...299).contains(httpResponse.statusCode) else {
        throw NetworkError.invalidResponse
    }
    
    return try JSONDecoder().decode(User.self, from: data)
}
```

> **Course Deep Dive**: This topic is covered in detail in [Lesson 2.4: Performing network requests using URLSession and async/await](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## Typed Errors (Swift 6)

Specify exact error types for better API contracts:

```swift
enum NetworkError: Error {
    case invalidResponse
    case decodingFailed(DecodingError)
    case requestFailed(URLError)
}

func fetchData() async throws(NetworkError) -> Data {
    do {
        let (data, _) = try await URLSession.shared.data(from: url)
        return data
    } catch let error as URLError {
        throw .requestFailed(error)
    } catch {
        throw .invalidResponse
    }
}
```

Callers know exactly which errors to handle.

## Migration Strategy

When converting closure-based code:

1. **Add new async method alongside old one** - keeps code compiling
2. **Update method signature** - add `async`, remove completion parameter
3. **Replace closure calls with await** - use URLSession async APIs
4. **Remove optional unwrapping** - async APIs return non-optional values
5. **Simplify error handling** - use do-catch instead of nested closures
6. **Return directly** - compiler enforces return values

## Common Patterns

### Sequential execution (when order matters)

```swift
let user = try await fetchUser(id: 1)
let posts = try await fetchPosts(userId: user.id)
let comments = try await fetchComments(postIds: posts.map(\.id))
```

### Parallel execution (when independent)

```swift
async let user = fetchUser(id: 1)
async let settings = fetchSettings()
async let notifications = fetchNotifications()

let (userData, settingsData, notificationsData) = try await (user, settings, notifications)
```

### Mixed execution

```swift
// Fetch user first (required for next step)
let user = try await fetchUser(id: 1)

// Then fetch related data in parallel
async let posts = fetchPosts(userId: user.id)
async let followers = fetchFollowers(userId: user.id)
async let following = fetchFollowing(userId: user.id)

let profile = Profile(
    user: user,
    posts: try await posts,
    followers: try await followers,
    following: try await following
)
```

## Further Learning

For in-depth coverage of async/await patterns, error handling strategies, and real-world migration scenarios, see [Swift Concurrency Course](https://www.swiftconcurrencycourse.com).

````

## File: skills/swift-concurrency/references/async-sequences.md
size: 17136
sha: 24fdfdae26f14e670b6778fb9ca4a834fe521324
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/async-sequences.md

````text
# Async Sequences and Streams

Use this when:

- You need to iterate over values that arrive over time.
- You are bridging callback-based or delegate-based APIs to async/await.
- You need to choose between `AsyncSequence`, `AsyncStream`, or a regular async method.

Skip this file if:

- You need time-based operators like debounce, throttle, or merge. Use `async-algorithms.md`.
- You are choosing between `Task`, `async let`, or task groups. Use `tasks.md`.

Jump to:

- AsyncSequence Protocol
- AsyncStream / AsyncThrowingStream
- Bridging Callbacks and Delegates
- Stream Lifecycle and Cleanup
- Buffer Policies
- Standard Library Integration
- Limitations
- When to Use AsyncAlgorithms

## AsyncSequence

Protocol for asynchronous iteration over values that become available over time.

### Basic usage

```swift
for await value in someAsyncSequence {
    print(value)
}
```

**Key difference from Sequence**: Values may not all be available immediately.

### Custom implementation

```swift
struct Counter: AsyncSequence, AsyncIteratorProtocol {
    typealias Element = Int
    
    let limit: Int
    var current = 1
    
    mutating func next() async -> Int? {
        guard !Task.isCancelled else { return nil }
        guard current <= limit else { return nil }
        
        let result = current
        current += 1
        return result
    }
    
    func makeAsyncIterator() -> Counter {
        self
    }
}

// Usage
for await count in Counter(limit: 5) {
    print(count) // 1, 2, 3, 4, 5
}
```

### Standard operators

Same functional operators as regular sequences:

```swift
// Filter
for await even in Counter(limit: 5).filter({ $0 % 2 == 0 }) {
    print(even) // 2, 4
}

// Map
let mapped = Counter(limit: 5).map { $0 % 2 == 0 ? "Even" : "Odd" }
for await label in mapped {
    print(label)
}

// Contains (awaits until found or sequence ends)
let contains = await Counter(limit: 5).contains(3) // true
```

### Termination

Return `nil` from `next()` to end iteration:

```swift
mutating func next() async -> Int? {
    guard !Task.isCancelled else {
        return nil // Stop on cancellation
    }
    
    guard current <= limit else {
        return nil // Stop at limit
    }
    
    return current
}
```

> **Course Deep Dive**: This topic is covered in detail in [Lesson 6.1: Working with asynchronous sequences](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## AsyncStream

Convenient way to create async sequences without implementing protocols.

### Basic creation

```swift
let stream = AsyncStream<Int> { continuation in
    for i in 1...5 {
        continuation.yield(i)
    }
    continuation.finish()
}

for await value in stream {
    print(value)
}
```

### AsyncThrowingStream

For streams that can fail:

```swift
let throwingStream = AsyncThrowingStream<Int, Error> { continuation in
    continuation.yield(1)
    continuation.yield(2)
    continuation.finish(throwing: SomeError())
}

do {
    for try await value in throwingStream {
        print(value)
    }
} catch {
    print("Error: \(error)")
}
```

> **Course Deep Dive**: This topic is covered in detail in [Lesson 6.2: Using AsyncStream and AsyncThrowingStream in your code](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## Bridging Closures to Streams

### Progress + completion handlers

```swift
// Old closure-based API
struct FileDownloader {
    enum Status {
        case downloading(Float)
        case finished(Data)
    }
    
    func download(
        _ url: URL,
        progressHandler: @escaping (Float) -> Void,
        completion: @escaping (Result<Data, Error>) -> Void
    ) throws {
        // Implementation
    }
}

// Modern stream-based API
extension FileDownloader {
    func download(_ url: URL) -> AsyncThrowingStream<Status, Error> {
        AsyncThrowingStream { continuation in
            do {
                try self.download(url, progressHandler: { progress in
                    continuation.yield(.downloading(progress))
                }, completion: { result in
                    switch result {
                    case .success(let data):
                        continuation.yield(.finished(data))
                        continuation.finish()
                    case .failure(let error):
                        continuation.finish(throwing: error)
                    }
                })
            } catch {
                continuation.finish(throwing: error)
            }
        }
    }
}

// Usage
for try await status in downloader.download(url) {
    switch status {
    case .downloading(let progress):
        print("Progress: \(progress)")
    case .finished(let data):
        print("Done: \(data.count) bytes")
    }
}
```

### Simplified with Result

```swift
AsyncThrowingStream { continuation in
    try self.download(url, progressHandler: { progress in
        continuation.yield(.downloading(progress))
    }, completion: { result in
        continuation.yield(with: result.map { .finished($0) })
        continuation.finish()
    })
}
```

## Bridging Delegates

### Location updates example

```swift
final class LocationMonitor: NSObject {
    private var continuation: AsyncThrowingStream<CLLocation, Error>.Continuation?
    let stream: AsyncThrowingStream<CLLocation, Error>
    
    override init() {
        var capturedContinuation: AsyncThrowingStream<CLLocation, Error>.Continuation?
        stream = AsyncThrowingStream { continuation in
            capturedContinuation = continuation
        }
        super.init()
        self.continuation = capturedContinuation
        
        locationManager.delegate = self
        locationManager.startUpdatingLocation()
    }
}

extension LocationMonitor: CLLocationManagerDelegate {
    func locationManager(_ manager: CLLocationManager, didUpdateLocations locations: [CLLocation]) {
        for location in locations {
            continuation?.yield(location)
        }
    }
    
    func locationManager(_ manager: CLLocationManager, didFailWithError error: Error) {
        continuation?.finish(throwing: error)
    }
}

// Usage
let monitor = LocationMonitor()
for try await location in monitor.stream {
    print("Location: \(location.coordinate)")
}
```

## Stream Lifecycle

### Termination callback

```swift
AsyncThrowingStream<Int, Error> { continuation in
    continuation.onTermination = { @Sendable reason in
        print("Terminated: \(reason)")
        // Cleanup: remove observers, cancel work, etc.
    }
    
    continuation.yield(1)
    continuation.finish()
}
```

**Termination reasons**:
- `.finished` - Normal completion
- `.finished(Error?)` - Completed with error (throwing stream)
- `.cancelled` - Task canceled

### Cancellation

Streams cancel when:
- Enclosing task cancels
- Stream goes out of scope

```swift
let task = Task {
    for try await status in download(url) {
        print(status)
    }
}

task.cancel() // Triggers onTermination with .cancelled
```

**No explicit cancel method** - rely on task cancellation.

## Buffer Policies

Control what happens to values when no one is awaiting:

### .unbounded (default)

Buffers all values until consumed:

```swift
let stream = AsyncStream<Int> { continuation in
    (0...5).forEach { continuation.yield($0) }
    continuation.finish()
}

try await Task.sleep(for: .seconds(1))

for await value in stream {
    print(value) // Prints all: 0, 1, 2, 3, 4, 5
}
```

### .bufferingNewest(n)

Keeps only the newest N values:

```swift
let stream = AsyncStream(bufferingPolicy: .bufferingNewest(1)) { continuation in
    (0...5).forEach { continuation.yield($0) }
    continuation.finish()
}

try await Task.sleep(for: .seconds(1))

for await value in stream {
    print(value) // Prints only: 5
}
```

### .bufferingOldest(n)

Keeps only the oldest N values:

```swift
let stream = AsyncStream(bufferingPolicy: .bufferingOldest(1)) { continuation in
    (0...5).forEach { continuation.yield($0) }
    continuation.finish()
}

try await Task.sleep(for: .seconds(1))

for await value in stream {
    print(value) // Prints only: 0
}
```

### .bufferingNewest(0)

Only receives values emitted after iteration starts:

```swift
let stream = AsyncStream(bufferingPolicy: .bufferingNewest(0)) { continuation in
    continuation.yield(1) // Discarded
    
    Task {
        try await Task.sleep(for: .seconds(2))
        continuation.yield(2) // Received
        continuation.finish()
    }
}

try await Task.sleep(for: .seconds(1))

for await value in stream {
    print(value) // Prints only: 2
}
```

**Use case**: Location updates, file system changes - only care about latest.

## Repeated Async Calls

Use `init(unfolding:onCancel:)` for polling:

```swift
struct PingService {
    func startPinging() -> AsyncStream<Bool> {
        AsyncStream {
            try? await Task.sleep(for: .seconds(5))
            return await ping()
        } onCancel: {
            print("Pinging cancelled")
        }
    }
    
    func ping() async -> Bool {
        // Network request
        return true
    }
}

// Usage
for await result in pingService.startPinging() {
    print("Ping: \(result)")
}
```

## Standard Library Integration

### NotificationCenter

```swift
let stream = NotificationCenter.default.notifications(
    named: .NSSystemTimeZoneDidChange
)

for await notification in stream {
    print("Time zone changed")
}
```

### Combine publishers

```swift
let numbers = [1, 2, 3, 4, 5]
let filtered = numbers.publisher.filter { $0 % 2 == 0 }

for await number in filtered.values {
    print(number) // 2, 4
}
```

### Task groups

```swift
await withTaskGroup(of: Image.self) { group in
    for url in urls {
        group.addTask { await download(url) }
    }
    
    for await image in group {
        display(image)
    }
}
```

## Limitations

### Single consumer only

Unlike Combine, streams support one consumer at a time:

```swift
let stream = AsyncStream { continuation in
    (0...5).forEach { continuation.yield($0) }
    continuation.finish()
}

Task {
    for await value in stream {
        print("Consumer 1: \(value)")
    }
}

Task {
    for await value in stream {
        print("Consumer 2: \(value)")
    }
}

// Unpredictable output - values split between consumers
// Consumer 1: 0
// Consumer 2: 1
// Consumer 1: 2
// Consumer 2: 3
```

**Solution**: Create separate streams or use third-party libraries (AsyncExtensions).

### No values after termination

Once finished, stream won't emit new values:

```swift
let stream = AsyncStream<Int> { continuation in
    continuation.finish() // Terminate immediately
    continuation.yield(1) // Never received
}

for await value in stream {
    print(value) // Loop exits immediately
}
```

## Decision Guide

### Use AsyncSequence when:

- Implementing standard library-style protocols
- Need fine-grained control over iteration
- Building reusable sequence types
- Working with existing sequence protocols

**Reality**: Rarely needed in application code.

### Use AsyncStream when:

- Bridging delegates to async/await
- Converting closure-based APIs
- Emitting events manually
- Polling or repeated async operations
- Most common use case

---

## When to Use AsyncAlgorithms vs Standard Library

### Use AsyncAlgorithms when:

- **Time-based operations** need debounce/throttle/timer
- **Combining multiple async sequences** (merge, combineLatest, zip)
- **Multi-consumer scenarios** require backpressure (AsyncChannel)
- **Complex operator chains** that Combine would handle naturally
- **Need specific operators** not in standard library

### Use Standard Library when:

- **Bridging callback APIs** → AsyncStream
- **Simple iteration** → for await in sequence
- **Single-value operations** → async/await
- **Basic transformations** → map/filter/contains

### Quick Decision Table

| Need | Solution |
|------|----------|
| Debounce search input | ✅ AsyncAlgorithms.debounce() |
| Throttle button clicks | ✅ AsyncAlgorithms.throttle() |
| Merge independent streams | ✅ AsyncAlgorithms.merge() |
| Combine dependent values | ✅ AsyncAlgorithms.combineLatest() or async let |
| Pair values from two sources | ✅ AsyncAlgorithms.zip() |
| Bridge callback API | AsyncStream |
| Multi-consumer with backpressure | ✅ AsyncChannel |
| Periodic timer | ✅ AsyncTimerSequence |
| Simple async iteration | for await in... |

> **See**: [async-algorithms.md](async-algorithms.md) for detailed usage examples with real-world patterns.

### Use regular async methods when:

- Single value returned
- No progress updates needed
- Simple request/response pattern

```swift
// Use this
func fetchData() async throws -> Data

// Not this
func fetchData() -> AsyncThrowingStream<Data, Error>

> **Course Deep Dive**: This topic is covered in detail in [Lesson 6.3: Deciding between AsyncSequence, AsyncStream, or regular asynchronous methods](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)
```

## Common Patterns

### Progress reporting

```swift
func download(_ url: URL) -> AsyncThrowingStream<DownloadEvent, Error> {
    AsyncThrowingStream { continuation in
        Task {
            do {
                var progress: Double = 0
                while progress < 1.0 {
                    progress += 0.1
                    continuation.yield(.progress(progress))
                    try await Task.sleep(for: .milliseconds(100))
                }
                
                let data = try await URLSession.shared.data(from: url).0
                continuation.yield(.completed(data))
                continuation.finish()
            } catch {
                continuation.finish(throwing: error)
            }
        }
    }
}
```

### Monitoring file system

```swift
func watchDirectory(_ path: String) -> AsyncStream<FileEvent> {
    AsyncStream(bufferingPolicy: .bufferingNewest(1)) { continuation in
        let source = DispatchSource.makeFileSystemObjectSource(
            fileDescriptor: fd,
            eventMask: .write,
            queue: .main
        )
        
        source.setEventHandler {
            continuation.yield(.fileChanged(path))
        }
        
        continuation.onTermination = { _ in
            source.cancel()
        }
        
        source.resume()
    }
}
```

### Timer/polling

```swift
func timer(interval: Duration) -> AsyncStream<Date> {
    AsyncStream { continuation in
        Task {
            while !Task.isCancelled {
                continuation.yield(Date())
                try? await Task.sleep(for: interval)
            }
            continuation.finish()
        }
    }
}

// Usage
for await date in timer(interval: .seconds(1)) {
    print("Tick: \(date)")
}
```

## Best Practices

1. **Always call finish()** - Streams stay alive until terminated
2. **Use buffer policies wisely** - Match your use case (latest value vs all values)
3. **Handle cancellation** - Set `onTermination` for cleanup
4. **Single consumer** - Don't share streams across multiple consumers
5. **Prefer streams over closures** - More composable and cancellable
6. **Check Task.isCancelled** - Respect cancellation in custom sequences
7. **Use throwing variant** - When operations can fail
8. **Consider regular async** - If only returning single value

## Debugging

### Add termination logging

```swift
continuation.onTermination = { reason in
    print("Stream ended: \(reason)")
}
```

### Validate finish() calls

```swift
// ❌ Forgot to finish
AsyncStream { continuation in
    continuation.yield(1)
    // Stream never ends!
}

// ✅ Always finish
AsyncStream { continuation in
    continuation.yield(1)
    continuation.finish()
}
```

### Check for dropped values

```swift
let stream = AsyncStream(bufferingPolicy: .bufferingNewest(1)) { continuation in
    for i in 1...100 {
        continuation.yield(i)
        print("Yielded: \(i)")
    }
    continuation.finish()
}

// If consumer is slow, many values dropped
for await value in stream {
    print("Received: \(value)")
    try? await Task.sleep(for: .seconds(1))
}
```

## Common Mistakes Agents Make

```swift
// ❌ Values after finish() are silently dropped
continuation.finish()
continuation.yield(1) // Never received

// ❌ Stream never terminates (forgot finish)
AsyncStream { continuation in
    continuation.yield(1)
    // Missing: continuation.finish()
}

// ❌ Wrapping a single-value API in a stream — use a regular async function instead
func fetchUser() -> AsyncStream<User> { ... } // Overkill for one result
```

- **Sharing a single `AsyncStream` between multiple consumers**: Values split unpredictably. There is no built-in broadcast; use `AsyncChannel` for point-to-point multi-consumer patterns.
- **Forgetting `onTermination`** when bridging delegate or observer APIs, causing resources to leak.

## Further Learning

For real-world migration examples, performance patterns, and advanced stream techniques, see [Swift Concurrency Course](https://www.swiftconcurrencycourse.com).

````

## File: skills/swift-concurrency/references/core-data.md
size: 14888
sha: 72840c6d9c53e1c66d916bf7208853ab1517e0b3
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/core-data.md

````text
# Core Data and Swift Concurrency

Use this when:

- You need to use Core Data with async/await or actors.
- `NSManagedObject` instances are crossing context or actor boundaries.
- You are resolving default `@MainActor` isolation conflicts with generated NSManagedObject subclasses.

Skip this file if:

- The issue is general actor isolation, not Core Data specific. Use `actors.md`.
- You need general Sendable guidance. Use `sendable.md`.

Jump to:

- Core Principles
- Data Access Objects (DAO) Pattern
- Working Without DAOs (NSManagedObjectID)
- Bridging Closures to Async
- Custom Actor Executor (Advanced)
- Default MainActor Isolation
- SwiftUI Integration
- Common Mistakes

## Core Principles

### Thread safety still matters

Core Data's thread safety rules don't change with Swift Concurrency:
- Can't pass `NSManagedObject` between threads
- Must access objects on their context's thread
- `NSManagedObjectID` is thread-safe (can pass around)

### NSManagedObject cannot be Sendable

```swift
@objc(Article)
public class Article: NSManagedObject {
    @NSManaged public var title: String // ❌ Mutable, can't be Sendable
}
```

**Don't use `@unchecked Sendable`** - hides warnings without fixing safety.

> **Course Deep Dive**: This topic is covered in detail in [Lesson 9.1: An introduction to Swift Concurrency and Core Data](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## Available Async APIs

### Context perform

```swift
extension NSManagedObjectContext {
    func perform<T>(
        schedule: ScheduledTaskType = .immediate,
        _ block: @escaping () throws -> T
    ) async rethrows -> T
}
```

### What's missing

No async alternative for:
```swift
func loadPersistentStores(
    completionHandler: @escaping (NSPersistentStoreDescription, Error?) -> Void
)
```

Must bridge manually (see below).

## Data Access Objects (DAO)

Thread-safe value types representing managed objects.

### Pattern

```swift
// Managed object (not Sendable)
@objc(Article)
public class Article: NSManagedObject {
    @NSManaged public var title: String?
    @NSManaged public var timestamp: Date?
}

// DAO (Sendable)
struct ArticleDAO: Sendable, Identifiable {
    let id: NSManagedObjectID
    let title: String
    let timestamp: Date
    
    init?(managedObject: Article) {
        guard let title = managedObject.title,
              let timestamp = managedObject.timestamp else {
            return nil
        }
        self.id = managedObject.objectID
        self.title = title
        self.timestamp = timestamp
    }
}
```

### Benefits

- **Sendable**: Safe to pass across isolation domains
- **Immutable**: No accidental mutations
- **Clear API**: Explicit data transfer

### Drawbacks

- **Requires rewrite**: All fetch/mutation logic
- **Boilerplate**: DAO for each entity
- **Complexity**: Additional layer of abstraction

> **Course Deep Dive**: This topic is covered in detail in [Lesson 9.2: Sendable and NSManageObjects](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## Working Without DAOs

Pass only `NSManagedObjectID` between contexts.

### Basic pattern

```swift
@MainActor
func fetchArticle(id: NSManagedObjectID) -> Article? {
    viewContext.object(with: id) as? Article
}

func processInBackground(articleID: NSManagedObjectID) async throws {
    let backgroundContext = container.newBackgroundContext()
    try await backgroundContext.perform {
        guard let article = backgroundContext.object(with: articleID) as? Article else {
            return
        }
        // Process article
        try backgroundContext.save()
    }
}
```

### NSManagedObjectID is Sendable

```swift
// Safe to pass between tasks
let articleID = article.objectID

Task {
    await processInBackground(articleID: articleID)
}
```

## Bridging Closures to Async

### Load persistent stores

```swift
extension NSPersistentContainer {
    func loadPersistentStores() async throws {
        try await withCheckedThrowingContinuation { continuation in
            self.loadPersistentStores { description, error in
                if let error {
                    continuation.resume(throwing: error)
                } else {
                    continuation.resume(returning: ())
                }
            }
        }
    }
}

// Usage
try await container.loadPersistentStores()
```

## Simple CoreDataStore Pattern

Enforce isolation at API level:

```swift
nonisolated struct CoreDataStore {
    static let shared = CoreDataStore()
    
    let persistentContainer: NSPersistentContainer
    private var viewContext: NSManagedObjectContext {
        persistentContainer.viewContext
    }
    
    private init() {
        persistentContainer = NSPersistentContainer(name: "MyApp")
        persistentContainer.viewContext.automaticallyMergesChangesFromParent = true
        
        Task { [persistentContainer] in
            try? await persistentContainer.loadPersistentStores()
        }
    }
    
    // View context operations (main thread)
    @MainActor
    func perform(_ block: (NSManagedObjectContext) throws -> Void) rethrows {
        try block(viewContext)
    }
    
    // Background operations
    @concurrent
    func performInBackground<T>(
        _ block: @escaping (NSManagedObjectContext) throws -> T
    ) async rethrows -> T {
        let context = persistentContainer.newBackgroundContext()
        return try await context.perform {
            try block(context)
        }
    }
}
```

### Usage

```swift
// Main thread operations
@MainActor
func loadArticles() throws -> [Article] {
    try CoreDataStore.shared.perform { context in
        let request = Article.fetchRequest()
        return try context.fetch(request)
    }
}

// Background operations
func deleteAll() async throws {
    try await CoreDataStore.shared.performInBackground { context in
        let request = Article.fetchRequest()
        let articles = try context.fetch(request)
        articles.forEach { context.delete($0) }
        try context.save()
    }
}
```

### Why this pattern works

- **@MainActor**: Enforces view context on main thread
- **@concurrent**: Forces background execution
- **Compile-time safety**: Wrong isolation = error
- **Simple**: No custom executors needed

## Custom Actor Executor (Advanced)

**Note**: Usually not needed. Consider simple pattern first.

> **Course Deep Dive**: This topic is covered in detail in [Lesson 9.3: Using a custom Actor executor for Core Data (advanced)](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

### Implementation

```swift
final class NSManagedObjectContextExecutor: @unchecked Sendable, SerialExecutor {
    private let context: NSManagedObjectContext
    
    init(context: NSManagedObjectContext) {
        self.context = context
    }
    
    func enqueue(_ job: consuming ExecutorJob) {
        let unownedJob = UnownedJob(job)
        let executor = asUnownedSerialExecutor()
        
        context.perform {
            unownedJob.runSynchronously(on: executor)
        }
    }
    
    func asUnownedSerialExecutor() -> UnownedSerialExecutor {
        UnownedSerialExecutor(ordinary: self)
    }
}
```

### Actor usage

```swift
actor CoreDataStore {
    let persistentContainer: NSPersistentContainer
    nonisolated let modelExecutor: NSManagedObjectContextExecutor
    
    nonisolated var unownedExecutor: UnownedSerialExecutor {
        modelExecutor.asUnownedSerialExecutor()
    }
    
    private init() {
        persistentContainer = NSPersistentContainer(name: "MyApp")
        let context = persistentContainer.newBackgroundContext()
        modelExecutor = NSManagedObjectContextExecutor(context: context)
    }
    
    func deleteAll<T: NSManagedObject>(
        using request: NSFetchRequest<T>
    ) throws {
        let objects = try context.fetch(request)
        objects.forEach { context.delete($0) }
        try context.save()
    }
}
```

### Drawbacks

- **Hidden complexity**: Executor details obscure Core Data
- **Forces concurrency**: Even for main thread operations
- **Not simpler**: More code than `perform { }`
- **Error prone**: Easy to use wrong context

**Recommendation**: Use simple pattern instead.

## Default MainActor Isolation

### Problem with auto-generated code

When default isolation set to `@MainActor`, auto-generated managed objects conflict:

```swift
// Auto-generated (can't modify)
class Article: NSManagedObject {
    // Inherits @MainActor, conflicts with NSManagedObject
}
```

**Error**: `Main actor-isolated initializer has different actor isolation from nonisolated overridden declaration`

### Solution: Manual code generation

1. Set entity to "Manual/None" code generation
2. Generate class definitions
3. Mark as `nonisolated`:

```swift
nonisolated class Article: NSManagedObject {
    @NSManaged public var title: String?
    @NSManaged public var timestamp: Date?
}

> **Course Deep Dive**: This topic is covered in detail in [Lesson 9.4: Autogenerated Core Data Objects and Default MainActor Isolation Conflicts](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)
```

**Benefit**: Full control over isolation.

## Common Patterns

### Fetch on main thread

```swift
@MainActor
func fetchArticles() throws -> [Article] {
    let request = Article.fetchRequest()
    return try viewContext.fetch(request)
}
```

### Background save

```swift
func saveInBackground() async throws {
    let context = container.newBackgroundContext()
    try await context.perform {
        let article = Article(context: context)
        article.title = "New Article"
        try context.save()
    }
}
```

### Pass ID, fetch in context

```swift
@MainActor
func displayArticle(id: NSManagedObjectID) {
    guard let article = viewContext.object(with: id) as? Article else {
        return
    }
    // Use article
}

func processArticle(id: NSManagedObjectID) async throws {
    try await CoreDataStore.shared.performInBackground { context in
        guard let article = context.object(with: id) as? Article else {
            return
        }
        // Process article
        try context.save()
    }
}
```

### Batch operations

```swift
@concurrent
func deleteAllArticles() async throws {
    try await CoreDataStore.shared.performInBackground { context in
        let request = NSFetchRequest<NSFetchRequestResult>(entityName: "Article")
        let deleteRequest = NSBatchDeleteRequest(fetchRequest: request)
        try context.execute(deleteRequest)
    }
}
```

## SwiftUI Integration

### Environment injection

```swift
@main
struct MyApp: App {
    let persistentContainer = NSPersistentContainer(name: "MyApp")
    
    var body: some Scene {
        WindowGroup {
            ContentView()
                .environment(\.managedObjectContext, persistentContainer.viewContext)
        }
    }
}
```

### View usage

```swift
struct ContentView: View {
    @Environment(\.managedObjectContext) private var viewContext
    @FetchRequest(
        sortDescriptors: [NSSortDescriptor(keyPath: \Article.timestamp, ascending: true)]
    ) private var articles: FetchedResults<Article>
    
    var body: some View {
        List(articles) { article in
            Text(article.title ?? "")
        }
    }
}
```

## Best Practices

1. **Pass NSManagedObjectID only** - never managed objects
2. **Use perform { }** - don't access context directly
3. **@MainActor for view context** - enforce main thread
4. **@concurrent for background** - force background execution
5. **Manual code generation** - control isolation
6. **Keep it simple** - avoid custom executors unless needed
7. **Enable Core Data debugging** - catch thread violations
8. **Merge changes automatically** - `automaticallyMergesChangesFromParent = true`
9. **Use background contexts** - for heavy operations
10. **Test with Thread Sanitizer** - catch violations early

## Debugging

### Enable Core Data concurrency debugging

```swift
// Launch argument
-com.apple.CoreData.ConcurrencyDebug 1
```

Crashes immediately on thread violations.

### Thread Sanitizer

Enable in scheme settings to catch data races.

### Assertions

```swift
@MainActor
func fetchArticles() -> [Article] {
    assert(Thread.isMainThread)
    // Fetch from viewContext
}
```

## Decision Tree

```
Need to access Core Data?
├─ UI/View context?
│  └─ Use @MainActor + viewContext
│
├─ Background operation?
│  ├─ Quick operation? → perform { } on background context
│  └─ Batch operation? → NSBatchDeleteRequest/NSBatchUpdateRequest
│
├─ Pass between contexts?
│  └─ Use NSManagedObjectID only
│
└─ Need Sendable type?
   ├─ Can refactor? → Use DAO pattern
   └─ Can't refactor? → Pass NSManagedObjectID
```

## Migration Strategy

### For existing projects

1. **Enable manual code generation** for all entities
2. **Mark entities as nonisolated** if using default @MainActor
3. **Wrap Core Data access** in CoreDataStore
4. **Use @MainActor** for view context operations
5. **Use @concurrent** for background operations
6. **Pass NSManagedObjectID** between contexts
7. **Test with debugging enabled**

### For new projects

1. **Start with simple pattern** (CoreDataStore)
2. **Manual code generation** from the start
3. **Consider DAOs** if heavy cross-context usage
4. **Enable strict concurrency** early

## Common Mistakes

### ❌ Passing managed objects

```swift
func process(article: Article) async {
    // ❌ Article not Sendable
}
```

### ❌ Accessing context from wrong thread

```swift
func background() async {
    let articles = viewContext.fetch(request) // ❌ Not on main thread
}
```

### ❌ Using @unchecked Sendable

```swift
extension Article: @unchecked Sendable {} // ❌ Doesn't make it safe
```

### ❌ Not using perform

```swift
func save() async {
    backgroundContext.save() // ❌ Not on context's thread
}
```

## Common Mistakes Agents Make

- **Passing `NSManagedObject` instances across actors**: Always transfer `NSManagedObjectID` or a Sendable value snapshot instead.
- **Using `@unchecked Sendable` on `NSManagedObject`**: This does not make it thread-safe. The object is still bound to its context's queue.
- **Skipping `perform { }`**: All background context access must go through `perform` or `performAndWait`.
- **Accessing `viewContext` from a background task**: The view context belongs to the main actor; access it only from `@MainActor`-isolated code.

## Further Learning

For Core Data best practices, migration strategies, and advanced patterns:
- [Core Data Best Practices](https://github.com/avanderlee/CoreDataBestPractices)
- [Swift Concurrency Course](https://www.swiftconcurrencycourse.com)

````

## File: skills/swift-concurrency/references/glossary.md
size: 7081
sha: f38ed0d2d03b6a23d7d494b3ea15d808bddaaf2c
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/glossary.md

```text
# Glossary

Use this when:

- You need a quick definition of a Swift Concurrency term.
- You encounter unfamiliar terminology in other reference files.

Skip this file if:

- You need implementation patterns, not definitions. Use the relevant reference file instead.

## Actor isolation

A rule enforced by the compiler: actor-isolated state can only be accessed from the actor's executor. Cross-actor access requires `await`.

## Global actor

A shared isolation domain applied via attributes like `@MainActor` or a custom `@globalActor`. Types/functions isolated to the same global actor can interact without crossing isolation.

## Default actor isolation

A module/target-level setting that changes the default isolation of declarations. App targets often choose `@MainActor` as the default to reduce migration noise, but it changes behavior and diagnostics.

## Strict concurrency checking

Compiler enforcement levels for Sendable and isolation diagnostics (minimal/targeted/complete). Raising the level typically reveals more issues and can trigger the “concurrency rabbit hole” unless migrated incrementally.

## Sendable

A marker protocol that indicates a type is safe to transfer across isolation boundaries. The compiler verifies stored properties and captured values for thread-safety.

## @Sendable

An annotation for function types/closures that can be executed concurrently. It tightens capture rules (captured values must be Sendable or safely transferred).

## Suspension point

An `await` site where a task may suspend and later resume. After a suspension point, you must assume other work may have run and (for actors) state may have changed (reentrancy).

## Reentrancy (actors)

While an actor is suspended at an `await`, other tasks can enter the actor and mutate state. Code after `await` must not assume actor state is unchanged.

## nonisolated

Marks a declaration as not isolated to the surrounding actor/global actor. Use only when it truly does not touch isolated mutable state (typically immutable Sendable data).

## nonisolated(nonsending) (Swift 6.2+ behavior)

An opt-out to prevent “sending” non-Sendable values across isolation while still allowing an async function to run in the caller’s isolation. Used to reduce Sendable friction when you do not need to hop executors.

## @concurrent (Swift 6.2+ behavior)

An attribute used to explicitly opt a nonisolated async function into concurrent execution (i.e., not inheriting the caller’s actor). It is used during migration when enabling `NonisolatedNonsendingByDefault`.
Also valid on `Task { @concurrent in ... }` to opt the task body out of the enclosing actor's isolation; pick this when the task's synchronous prefix (everything before the first `await`) does not need the main actor.

## @preconcurrency

An annotation used to suppress Sendable-related diagnostics from a module that predates concurrency annotations. It reduces noise but shifts safety responsibility to you.

## Region-based isolation / sending

Mechanisms that model ownership transfer so certain non-Sendable values can be moved between regions safely. The `sending` keyword enforces that a value is no longer used after transfer.

## AsyncSequence

A protocol for types that provide asynchronous, sequential iteration over elements. Conforms to the `for await` loop pattern. Use for streaming data where elements arrive over time.

## AsyncStream

A concrete implementation of `AsyncSequence` that bridges callback-based or delegate-based APIs to async/await. Provides `yield()` to emit values and `finish()` to complete the stream.

## Continuation

A mechanism to bridge callback-based APIs to async/await. `withCheckedContinuation` and `withCheckedThrowingContinuation` provide safe bridging with runtime checks. `withUnsafeContinuation` variants skip checks for performance-critical code.

## Task Local

Task-scoped storage that propagates values through the task hierarchy automatically. Declared with `@TaskLocal` and accessed via the wrapper's static property. Child tasks inherit parent task locals.

## Cooperative thread pool

Swift's threading model where tasks run on a limited pool of threads managed by the runtime. Tasks yield cooperatively at suspension points, allowing other tasks to run. Avoid blocking operations that would starve the pool.

## Executor

The scheduling mechanism that determines where and when actor code runs. `MainActor` uses the main thread executor. Custom actors use the default executor unless a custom executor is specified.

## Structured concurrency

A pattern where child tasks have a well-defined relationship to parent tasks. Child tasks must complete before the parent scope exits. Provides automatic cancellation propagation and prevents orphaned tasks. Implemented via `async let` and `TaskGroup`.

## Isolation domain

A boundary that protects mutable state from concurrent access. Each actor instance defines its own isolation domain. The `@MainActor` global actor defines a shared isolation domain for UI work. Code must cross isolation boundaries explicitly via `await`.

## Task priority

A hint to the runtime about the relative importance of a task. Priorities include `.high`, `.medium`, `.low`, `.userInitiated`, `.utility`, and `.background`. Higher priority tasks are scheduled before lower priority ones. Priority can escalate when a high-priority task awaits a low-priority one.

## Cancellation

A cooperative mechanism to signal that a task should stop. Check `Task.isCancelled` or call `Task.checkCancellation()` (throws) in long-running work. Cancellation propagates to child tasks in structured concurrency.

## Debounce

Wait for a period of inactivity before emitting a value. Used to reduce API calls for rapid inputs like search fields. Implemented as `debounce(for:tolerance:clock:)` in AsyncAlgorithms.

## Throttle

Emit at most one value per time interval, discarding intermediate values. Used to prevent excessive calls from repeated actions like button taps. Implemented as `throttle(for:clock:reducing:)` in AsyncAlgorithms.

## Merge (AsyncAlgorithms)

Combine multiple asynchronous sequences into one, emitting values as they arrive from any source. Order is interleaved based on emission timing. Stable operator.

## CombineLatest (AsyncAlgorithms)

Combine multiple asynchronous sequences, emitting a tuple whenever any source emits a new value. Always uses the latest value from each sequence. Stable operator.

## Zip (AsyncAlgorithms)

Combine multiple asynchronous sequences by pairing elements in order. Waits for all sequences to emit before producing a tuple. Stable operator.

## AsyncChannel

An AsyncSequence with backpressure sending semantics. Allows multiple producers to send values safely to multiple consumers with flow control. Stable operator.

## AsyncThrowingChannel

Like AsyncChannel but can emit errors through the stream. Stable operator.

## AsyncTimerSequence

An AsyncSequence that emits a value at regular intervals. Replaces timer-based publishers and manual sleep loops. Stable operator.

```

## File: skills/swift-concurrency/references/linting.md
size: 5334
sha: e0129b2c1a817ff3f685e529a7c56b29912b74f9
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/linting.md

````text
# Linting & Concurrency

Use this when:

- SwiftLint flags `async_without_await` or other concurrency-related warnings.
- You need to decide whether to suppress, fix, or reconfigure a concurrency lint rule.

Skip this file if:

- The issue is a compiler diagnostic, not a lint rule. Use `actors.md`, `sendable.md`, or `threading.md`.

Jump to:

- SwiftLint Concurrency Rules Overview
- `async_without_await` Rule
- Suppression Strategies

## SwiftLint Concurrency Rules Overview

SwiftLint provides several rules targeting async/await and concurrency patterns. Understanding when to fix vs. suppress is critical.

| Rule | Default | Purpose |
|------|---------|---------|
| `async_without_await` | warning | Flags `async` functions that never await |
| `unowned_variable_capture` | warning | Warns about `unowned` in closures (risky in async) |
| `class_delegate_protocol` | warning | Ensures delegates are class-bound (AnyObject) |
| `weak_delegate` | warning | Delegates should be weak to avoid retain cycles |

## SwiftLint: `async_without_await`

- **Intent**: A declaration should not be `async` if it never awaits.
- **Never "fix"** by inserting fake suspension (e.g. `await Task.yield()`, `await Task { ... }.value`). Those mask the real issue and add meaningless suspension points.
- **Legit use of `Task.yield()`**: OK in tests or scheduling control when you truly need a yield; not as a lint workaround.

### Diagnose why the declaration is `async`
1) **Protocol requirement** — the protocol method/property is `async`.
2) **Override requirement** — base class API is `async`.
3) **`@concurrent` requirement** — stays `async` even without `await`.
4) **Accidental/legacy `async`** — no caller needs async semantics.

### Preferred fixes (order)
1) **Remove `async`** (and adjust call sites) when no async semantics are needed.
2) If `async` is required (protocol/override/@concurrent):
   - Re-evaluate the upstream API if you own it (can it be non-async?).
   - If you cannot change it, keep `async` and **narrowly suppress the rule** where appropriate (common for mocks/stubs/overrides).

### Suppression examples (keep scope tight)
```swift
// swiftlint:disable:next async_without_await
func fetch() async { perform() }

// For a block:
// swiftlint:disable async_without_await
func makeMock() async { perform() }
// swiftlint:enable async_without_await
```

### Quick checklist
- [ ] Confirm if `async` is truly required (protocol/override/@concurrent).
- [ ] If not required, remove `async` and update callers.
- [ ] If required, prefer localized suppression over dummy awaits.
- [ ] Avoid adding new suspension points without intent.

## Compiler Warnings: Sendable & Isolation

The Swift compiler generates concurrency-related warnings based on strict concurrency checking level.

### Common Warning Patterns

**"Capture of non-sendable type"**
```swift
// Warning: Capture of 'self' with non-sendable type 'MyClass' in a `@Sendable` closure
Task {
    self.doWork() // 'self' is non-Sendable
}
```

**Fixes (in order of preference):**
1. Make the type `Sendable` if it's truly thread-safe
2. Use `@MainActor` isolation if it's UI-related
3. Capture only Sendable values instead of `self`
4. Use `@unchecked Sendable` with documented safety invariant (last resort)

**"Non-sendable result returned"**
```swift
// Warning: Non-sendable type 'MyResult' returned by implicitly async call
let result = await actor.getData() // Returns non-Sendable type
```

**Fixes:**
1. Make the return type Sendable
2. Return Sendable projections (IDs, copies of data)
3. Keep processing within the actor's isolation

### Actor Isolation Warnings

**"Main actor-isolated property accessed from non-isolated context"**
```swift
// Warning: Main actor-isolated property 'title' cannot be referenced from a non-isolated context
func updateTitle() {
    viewModel.title = "New" // viewModel is @MainActor
}
```

**Fixes:**
1. Mark the calling function `@MainActor`
2. Use `await MainActor.run { }` for one-off access
3. Reconsider if the property truly needs @MainActor isolation

## Suppression Strategies

### When to Suppress vs. Fix

**Fix when:**
- The warning identifies a real data race risk
- The fix is straightforward (add Sendable, adjust isolation)
- The code is new or actively maintained

**Suppress when:**
- Protocol/inheritance requires the signature
- Third-party code forces the pattern
- Migration is in progress (with tracked ticket)

### Suppression Annotations

```swift
// Suppress Sendable warnings for legacy imports
@preconcurrency import LegacyFramework

// Suppress for a single declaration
nonisolated(unsafe) var legacyCallback: (() -> Void)?

// Type-level suppression (use sparingly)
struct LegacyWrapper: @unchecked Sendable {
    // Document why this is safe
    private let lock = NSLock()
    private var value: Int
}
```

### Documentation Requirements

When using suppression annotations, document:
1. **Why** the suppression is needed
2. **What** invariant makes it safe
3. **When** it can be removed (link to migration ticket)

```swift
/// Thread-safe: Internal lock protects all mutations.
/// TODO: Remove @unchecked when migrated to actor (JIRA-1234)
final class ThreadSafeCache: @unchecked Sendable {
    private let lock = NSLock()
    private var storage: [String: Data] = [:]
}
```
````

## File: skills/swift-concurrency/references/memory-management.md
size: 12736
sha: b33b99cd5390b6b5c157873e3f446de7143fb97d
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/memory-management.md

````text
# Memory Management

Use this when:

- A task or async sequence is keeping objects alive longer than expected.
- You suspect a retain cycle between a task and its owner.
- You need to verify deallocation behavior or use `isolated deinit`.

Skip this file if:

- You mainly need to protect mutable state from races. Use `actors.md`.
- You are debugging slow async code. Use `performance.md`.

Jump to:

- Core Concepts (Task Capture)
- Retain Cycles
- One-Way Retention
- Async Sequences and Retention
- Isolated Deinit (Swift 6.2+)
- Detection and Testing
- Common Patterns

## Core Concepts

### Tasks capture like closures

Tasks capture variables and references just like regular closures. Swift doesn't automatically prevent retain cycles in concurrent code.

```swift
Task {
    self.doWork() // ⚠️ Strong capture of self
}
```

### Why concurrency hides memory issues

- Tasks may live longer than expected
- Async operations delay execution
- Harder to track when memory should be released
- Long-running tasks can hold references indefinitely

> **Course Deep Dive**: This topic is covered in detail in [Lesson 8.1: Overview of memory management in Swift Concurrency](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

## Retain Cycles

### What is a retain cycle?

Two or more objects hold strong references to each other, preventing deallocation.

```swift
class A {
    var b: B?
}

class B {
    var a: A?
}

let a = A()
let b = B()
a.b = b
b.a = a // Retain cycle - neither can be deallocated
```

### Retain cycles with Tasks

When task captures `self` strongly and `self` owns the task:

```swift
@MainActor
final class ImageLoader {
    var task: Task<Void, Never>?
    
    func startPolling() {
        task = Task {
            while true {
                self.pollImages() // ⚠️ Strong capture
                try? await Task.sleep(for: .seconds(1))
            }
        }
    }
}

var loader: ImageLoader? = .init()
loader?.startPolling()
loader = nil // ⚠️ Loader never deallocated - retain cycle!
```

**Problem**: Task holds `self`, `self` holds task → neither released.

## Breaking Retain Cycles

### Use weak self

```swift
func startPolling() {
    task = Task { [weak self] in
        while let self = self {
            self.pollImages()
            try? await Task.sleep(for: .seconds(1))
        }
    }
}

var loader: ImageLoader? = .init()
loader?.startPolling()
loader = nil // ✅ Loader deallocated, task stops
```

### Pattern for long-running tasks

```swift
task = Task { [weak self] in
    while let self = self {
        await self.doWork()
        try? await Task.sleep(for: interval)
    }
}
```

> **Course Deep Dive**: This topic is covered in detail in [Lesson 8.2: Preventing retain cycles when using Tasks](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=lesson-reference)

Loop exits when `self` becomes `nil`.

## One-Way Retention

Task retains `self`, but `self` doesn't retain task. Object stays alive until task completes.

```swift
@MainActor
final class ViewModel {
    func fetchData() {
        Task {
            await performRequest()
            updateUI() // ⚠️ Strong capture
        }
    }
}

var viewModel: ViewModel? = .init()
viewModel?.fetchData()
viewModel = nil // ViewModel stays alive until task completes
```

**Execution order**:
1. Task starts
2. `viewModel = nil` (but object not deallocated)
3. Task completes
4. ViewModel finally deallocated

### When one-way retention is acceptable

Short-lived tasks that complete quickly:

```swift
func saveData() {
    Task {
        await database.save(self.data) // OK - completes quickly
    }
}
```

### When to use weak self

Long-running or indefinite tasks:

```swift
func startMonitoring() {
    Task { [weak self] in
        for await event in eventStream {
            self?.handle(event)
        }
    }
}
```

## Async Sequences and Retention

### Problem: Infinite sequences

```swift
@MainActor
final class AppLifecycleViewModel {
    private(set) var isActive = false
    private var task: Task<Void, Never>?
    
    func startObserving() {
        task = Task {
            for await _ in NotificationCenter.default.notifications(
                named: .didBecomeActive
            ) {
                isActive = true // ⚠️ Strong capture, never ends
            }
        }
    }
}

var viewModel: AppLifecycleViewModel? = .init()
viewModel?.startObserving()
viewModel = nil // ⚠️ Never deallocated - sequence continues
```

**Problem**: Async sequence never finishes, task holds `self` indefinitely.

### Solution 1: Manual cancellation

```swift
func startObserving() {
    task = Task {
        for await _ in NotificationCenter.default.notifications(
            named: .didBecomeActive
        ) {
            isActive = true
        }
    }
}

func stopObserving() {
    task?.cancel()
}

// Usage
viewModel?.startObserving()
viewModel?.stopObserving() // Must call before release
viewModel = nil
```

### Solution 2: Weak self with guard

```swift
func startObserving() {
    task = Task { [weak self] in
        for await _ in NotificationCenter.default.notifications(
            named: .didBecomeActive
        ) {
            guard let self = self else { return }
            self.isActive = true
        }
    }
}
```

Task exits when `self` deallocates.

## Isolated deinit (Swift 6.2+)

Clean up actor-isolated state in deinit:

```swift
@MainActor
final class ViewModel {
    private var task: Task<Void, Never>?
    
    isolated deinit {
        task?.cancel()
    }
}
```

**Limitation**: Won't break retain cycles (deinit never called if cycle exists).

**Use for**: Cleanup when object is being deallocated normally.

## Common Patterns

### Short-lived task (strong capture OK)

```swift
func saveData() {
    Task {
        await database.save(self.data)
        self.updateUI()
    }
}
```

**When safe**: Task completes quickly, acceptable for object to live until done.

### Long-running task (weak self required)

```swift
func startPolling() {
    task = Task { [weak self] in
        while let self = self {
            await self.fetchUpdates()
            try? await Task.sleep(for: .seconds(5))
        }
    }
}
```

### Async sequence monitoring (weak self + guard)

```swift
func startMonitoring() {
    task = Task { [weak self] in
        for await event in eventStream {
            guard let self = self else { return }
            self.handle(event)
        }
    }
}
```

### Cancellable work with cleanup

```swift
func startWork() {
    task = Task { [weak self] in
        defer { self?.cleanup() }
        
        while let self = self {
            await self.doWork()
            try? await Task.sleep(for: .seconds(1))
        }
    }
}
```

## Detection Strategies

### Add deinit logging

```swift
deinit {
    print("✅ \(type(of: self)) deallocated")
}
```

If deinit never prints → likely retain cycle.

### Memory graph debugger

1. Run app in Xcode
2. Debug → Debug Memory Graph
3. Look for cycles in object graph

### Instruments

Use Leaks instrument to detect retain cycles at runtime.

## Decision Tree

```
Task captures self?
├─ Task completes quickly?
│  └─ Strong capture OK
│
├─ Long-running or infinite?
│  ├─ Can use weak self? → Use [weak self]
│  ├─ Need manual control? → Store task, cancel explicitly
│  └─ Async sequence? → [weak self] + guard
│
└─ Self owns task?
   ├─ Yes → High risk of retain cycle
   └─ No → Lower risk, but check lifetime
```

## Best Practices

1. **Default to weak self** for long-running tasks
2. **Use guard let self** in async sequences
3. **Cancel tasks explicitly** when possible
4. **Add deinit logging** during development
5. **Test object deallocation** in unit tests
6. **Use Memory Graph** to verify no cycles
7. **Document lifetime expectations** in comments
8. **Prefer cancellation** over weak self when possible
9. **Avoid nested strong captures** in task closures
10. **Use isolated deinit** for cleanup (Swift 6.2+)

## Testing for Leaks

### Unit test pattern

```swift
func testViewModelDeallocates() async {
    var viewModel: ViewModel? = ViewModel()
    weak var weakViewModel = viewModel
    
    viewModel?.startWork()
    viewModel = nil
    
    // Give tasks time to complete
    try? await Task.sleep(for: .milliseconds(100))
    
    XCTAssertNil(weakViewModel, "ViewModel should be deallocated")
}
```

### SwiftUI view test

```swift
func testViewDeallocates() {
    var view: MyView? = MyView()
    weak var weakView = view
    
    view = nil
    
    XCTAssertNil(weakView)
}
```

## Common Mistakes

### ❌ Forgetting weak self in loops

```swift
Task {
    while true {
        self.poll() // Retain cycle
        try? await Task.sleep(for: .seconds(1))
    }
}
```

### ❌ Strong capture in async sequences

```swift
Task {
    for await item in stream {
        self.process(item) // May never release
    }
}
```

### ❌ Not canceling stored tasks

```swift
class Manager {
    var task: Task<Void, Never>?
    
    func start() {
        task = Task {
            await self.work() // Retain cycle
        }
    }
    
    // Missing: deinit { task?.cancel() }
}
```

### ❌ Assuming deinit breaks cycles

```swift
deinit {
    task?.cancel() // Never called if retain cycle exists
}
```

## Examples by Use Case

### Polling service

```swift
final class PollingService {
    private var task: Task<Void, Never>?
    
    func start() {
        task = Task { [weak self] in
            while let self = self {
                await self.poll()
                try? await Task.sleep(for: .seconds(5))
            }
        }
    }
    
    func stop() {
        task?.cancel()
    }
}
```

### Notification observer

```swift
@MainActor
final class NotificationObserver {
    private var task: Task<Void, Never>?
    
    func startObserving() {
        task = Task { [weak self] in
            for await notification in NotificationCenter.default.notifications(
                named: .someNotification
            ) {
                guard let self = self else { return }
                self.handle(notification)
            }
        }
    }
    
    isolated deinit {
        task?.cancel()
    }
}
```

### Download manager

```swift
final class DownloadManager {
    private var tasks: [URL: Task<Data, Error>] = [:]
    
    func download(_ url: URL) async throws -> Data {
        let task = Task { [weak self] in
            defer { self?.tasks.removeValue(forKey: url) }
            return try await URLSession.shared.data(from: url).0
        }
        
        tasks[url] = task
        return try await task.value
    }
    
    func cancelAll() {
        tasks.values.forEach { $0.cancel() }
        tasks.removeAll()
    }
}
```

### Timer

```swift
actor Timer {
    private var task: Task<Void, Never>?
    
    func start(interval: Duration, action: @Sendable () async -> Void) {
        task = Task {
            while !Task.isCancelled {
                await action()
                try? await Task.sleep(for: interval)
            }
        }
    }
    
    func stop() {
        task?.cancel()
    }
}
```

## Common Mistakes Agents Make

- **Forgetting `[weak self]` in stored tasks**: When `self` owns the task and the task captures `self`, a retain cycle prevents deallocation.
- **Strong capture in infinite `AsyncSequence` loops**: `for await` over an infinite sequence with a strong `self` capture keeps the object alive forever.
- **Not cancelling stored tasks on cleanup**: If the task outlives its owner, it retains captured objects indefinitely.
- **Assuming `isolated deinit` breaks retain cycles**: `isolated deinit` runs cleanup on the correct actor, but if a cycle prevents `deinit` from being called at all, the cleanup never executes.
- **Using `try?` in loops with `Task.sleep`**: `try?` can swallow `CancellationError`, causing the loop to continue running after cancellation. Always check `Task.isCancelled` explicitly.

## Debugging Checklist

When object won't deallocate:

- [ ] Check for strong self captures in tasks
- [ ] Verify tasks are canceled or complete
- [ ] Look for infinite loops or sequences
- [ ] Check if self owns the task
- [ ] Use Memory Graph to find cycles
- [ ] Add deinit logging to verify
- [ ] Test with weak references
- [ ] Review async sequence usage
- [ ] Check nested task captures
- [ ] Verify cleanup in deinit

## Further Learning

For migration strategies, real-world examples, and advanced memory patterns, see [Swift Concurrency Course](https://www.swiftconcurrencycourse.com).

````

## File: skills/swift-concurrency/references/migration.md
size: 33770
sha: 2416573cab9cfcfe49807a069f7435d1aaff5cc1
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/migration.md

Content is not inlined in the catalog. Read the upstream file: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/migration.md

## File: skills/swift-concurrency/references/observation.md
size: 14420
sha: 525403f70c002b28452a045d81053e2964db5e7c
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/observation.md

Content is not inlined in the catalog. Read the upstream file: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/observation.md

## File: skills/swift-concurrency/references/performance.md
size: 15490
sha: 76996336b8a2b4cc29e058a27e975766c409b85f
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/performance.md

Content is not inlined in the catalog. Read the upstream file: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/performance.md

## File: skills/swift-concurrency/references/sendable.md
size: 13603
sha: 5a4414fcbf124ea043cd7f798029fe95ba6e9259
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/sendable.md

Content is not inlined in the catalog. Read the upstream file: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/sendable.md

## File: skills/swift-concurrency/references/tasks.md
size: 17019
sha: 961d6182e02049029f535dc01bac7878b8d50510
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/tasks.md

Content is not inlined in the catalog. Read the upstream file: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/tasks.md

## File: skills/swift-concurrency/references/testing.md
size: 13757
sha: 65d4ffa70d4fe59bf36699be2bbb7bfe7c72e92f
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/testing.md

Content is not inlined in the catalog. Read the upstream file: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/testing.md

## File: skills/swift-concurrency/references/threading.md
size: 16440
sha: 06c520ff3a50575008c4364afa1bfb47c31539ed
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/threading.md

Content is not inlined in the catalog. Read the upstream file: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/references/threading.md

## File: skills/swift-concurrency/assets/logo-small.png
size: 4283
sha: ea10ab0c63fda7ce199f80b2f5a0dfb2e9d058d5
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/assets/logo-small.png

Content is not inlined in the catalog. Read the upstream file: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/assets/logo-small.png

## File: skills/swift-concurrency/assets/logo.png
size: 26017
sha: 62972993ff53469a9c176eee7102e09404a86f75
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/assets/logo.png

Content is not inlined in the catalog. Read the upstream file: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/assets/logo.png

## File: skills/swift-concurrency/agents/openai.yaml
size: 249
sha: cbae65abd4277d0ce6e08be2a32e8eeb176615c8
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/skills/swift-concurrency/agents/openai.yaml

```text
interface:
  display_name: "Swift Concurrency"
  short_description: "Diagnose data races, adopt async/await and actors, and guide Swift 6 migration."
  icon_small: "./assets/logo-small.png"
  icon_large: "./assets/logo.png"
  brand_color: "#F05138"
```

## File: README.md
upstream: https://github.com/avdlee/swift-concurrency-agent-skill/blob/main/README.md

````text
<p align="center">
    <a href="https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=swift-concurrency-skill">
        <img width="900px" src="assets/github_readme_banner.jpg" alt="Swift Concurrency Agent Skill banner">
    </a>
</p>

# Swift Concurrency Agent Skill

Expert guidance for any AI coding tool that supports the [Agent Skills open format](https://agentskills.io/home) — safe concurrency, performance, and Swift 6+ migration.

Based on the comprehensive [Swift Concurrency Course](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=swift-concurrency-skill), distilled into actionable, concise references for agents.

This repository is packaged as a portable [Agent Plugin](https://agent-plugins.org) (spec 1.0.0): compatible clients discover the skill automatically from the root `plugin.json` manifest and the `skills/` directory. Client-specific manifests for Claude Code, Cursor, Codex, and pi are included as well, and the plugin is published in the OpenAI Plugins Directory for ChatGPT and Codex.

## Who this is for
- Teams migrating to Swift 6 / strict concurrency who need safe defaults and quick triage.
- Developers debugging data races, isolation errors, or flaky async tests.
- Anyone wanting performance-minded concurrency patterns (actors, tasks, Sendable, async streams).

## See also my other skills:
- [SwiftUI Expert](https://github.com/AvdLee/SwiftUI-Agent-Skill)
- [Core Data Expert](https://github.com/AvdLee/Core-Data-Agent-Skill)
- [Swift Testing Expert](https://github.com/AvdLee/Swift-Testing-Agent-Skill)

## How to Use This Skill

### Option A: Using skills.sh (recommended)
Install this skill with a single command:
```bash
npx skills@latest add https://github.com/avdlee/swift-concurrency-agent-skill --skill swift-concurrency
```

For more information, visit the [skills.sh platform page](https://skills.sh/avdlee/swift-concurrency-agent-skill/swift-concurrency).

> Installed before the Agent Plugins migration? Run `npx skills@latest update` using skills CLI 1.5.24 or newer. It migrates the saved skill path without treating the skill as deleted.

Then use the skill in your AI agent, for example:  
> Use the swift concurrency skill and analyze the current project for Swift Concurrency improvements

### Option B: ChatGPT & Codex (OpenAI Plugins Directory)

This plugin is published in the universal OpenAI Plugins Directory shared by ChatGPT and Codex:

1. In the ChatGPT desktop app, select **Codex** (or ChatGPT with **Work** enabled), or use the plugin browser in Codex CLI.
2. Open **Plugins** and search for "Swift Concurrency".
3. Install the plugin. The skill becomes available in new chats and sessions.

### Option C: Claude Code Plugin

#### Personal Usage

To install this Skill for your personal use in Claude Code:

1. Add the marketplace:
   ```bash
   /plugin marketplace add AvdLee/Swift-Concurrency-Agent-Skill
   ```

2. Install the Skill:
   ```bash
   /plugin install swift-concurrency@swift-concurrency-agent-skill
   ```

#### Project Configuration

To automatically provide this Skill to everyone working in a repository, configure the repository's `.claude/settings.json`:

```json
{
  "enabledPlugins": {
    "swift-concurrency@swift-concurrency-agent-skill": true
  },
  "extraKnownMarketplaces": {
    "swift-concurrency-agent-skill": {
      "source": {
        "source": "github",
        "repo": "AvdLee/Swift-Concurrency-Agent-Skill"
      }
    }
  }
}
```

When team members open the project, Claude Code will prompt them to install the Skill.

### Option D: Cursor Plugin

Load the plugin from a local clone by placing it in `~/.cursor/plugins/local`, or install it from the Cursor Marketplace once listed. Cursor supports both the portable Agent Plugins manifest (`plugin.json`) and the Cursor Plugin manifest (`.cursor-plugin/plugin.json`) included in this repository.

### Using another tool?

See [INSTALLATION.md](INSTALLATION.md) for all other installation options: pi, Gemini CLI, Autohand Code, and manual installation.

**How to verify**: 

Your agent should reference the triage/playbook in `skills/swift-concurrency/SKILL.md` and jump into the relevant reference file for your error or task.

## What This Skill Offers

This skill gives your AI coding tool comprehensive Swift Concurrency guidance. It can:

### Guide Your Concurrency Decisions
- Choose the right tool for the job (async/await, actors, tasks, task groups)
- Understand when to use `@MainActor`, custom actors, or `nonisolated`
- Navigate isolation domains and prevent data races at compile time
- Apply `Sendable` conformance correctly for value and reference types

### Write Safe Concurrent Code
- Avoid common pitfalls like actor reentrancy and retain cycles
- Prevent data races with proper isolation
- Handle task cancellation and error propagation correctly
- Manage memory safely in concurrent contexts

### Optimize Performance
- Choose between serialized, asynchronous, and parallel execution
- Reduce actor contention and unnecessary suspension points
- Understand the tradeoffs of parallelism

### Migrate to Swift 6
- Step-by-step migration strategies for existing codebases
- Enable strict concurrency checking incrementally
- Rewrite closure-based code to async/await
- Migrate from Combine/RxSwift to Swift Concurrency
- Use migration tooling for upcoming Swift features

### Test Concurrent Code
- Write reliable tests using Swift Testing (recommended) or XCTest
- Handle `@MainActor` isolation in tests
- Use `withMainSerialExecutor` for deterministic testing
- Avoid flaky tests with proper async handling

### Integrate with Core Data
- Safely pass data between isolation domains using `NSManagedObjectID`
- Implement the Data Access Object (DAO) pattern
- Use custom actor executors when needed
- Avoid common Core Data concurrency pitfalls

## What Makes This Skill Different

**Expert Knowledge**: Based on real-world experience migrating large production codebases to Swift 6, distilled from the comprehensive [Swift Concurrency Course](https://www.swiftconcurrencycourse.com?utm_source=github&utm_medium=agent-skill&utm_campaign=swift-concurrency-skill).

**Non-Opinionated**: Focuses on industry-standard best practices and compile-time safety, not architectural preferences. Works with any Swift project, coding style, or architecture.

**Swift 6.2 Ready**: Covers the latest Swift Concurrency features including:
- Default Actor Isolation
- `isolated deinit`
- Global Actor Conformance for protocols
- `nonisolated(nonsending)` and `@concurrent`
- Approachable Concurrency build settings
- Concurrency-safe notifications (iOS 26+)

**Practical & Concise**: Assumes your AI agent is already smart. Focuses on what developers need to know, not what they already understand. Includes code examples for every pattern.

## See it in action
[![Youtube Video](assets/agent_skills_watch_on_youtube.png)](https://www.youtube.com/watch?v=khekVi1PK3o)

## Skill Structure

```
skills/swift-concurrency/
├── SKILL.md                # Main skill file with decision trees
└── references/
    ├── _index.md               # Quick navigation index for all reference files
    ├── actors.md               # Actor isolation, global actors, reentrancy
    ├── async-algorithms.md     # AsyncAlgorithms operators: debounce, merge, zip, Combine migration
    ├── async-await-basics.md   # Fundamentals of async/await syntax
    ├── async-sequences.md      # AsyncSequence and AsyncStream patterns
    ├── core-data.md            # Core Data integration patterns
    ├── glossary.md             # Terms & concepts for Swift Concurrency
    ├── linting.md              # Linting rules for strict concurrency
    ├── memory-management.md    # Retain cycles, weak self, isolated deinit
    ├── migration.md            # Step-by-step Swift 6 migration guide
    ├── observation.md          # @Observable isolation, observation tracking, Sendable snapshots
````
