PHP True Async

• Version: 1.0
• Date: 2025-04-16
• Author: Edmond [HT], [email protected]
• Status: Under discussion
• First Published at: http://wiki.php.net/rfc/true_async
• Git: https://github.com/EdmondDantes/php-true-async-rfc

For several years, PHP has been attempting to carve out a niche in the development of long-running applications, where concurrent code execution becomes particularly useful. Production-ready solutions such as Swoole, AMPHP, ReactPHP, and others have emerged.

However, PHP still does not provide a comprehensive implementation for writing concurrent code. PHP extensions have no way to support non-blocking execution, even if they are capable of doing so. Swoole is forced to copy thousands of lines of code just for a few modifications, while AMPHP developers have to build drivers for MySQL, PostgreSQL, Redis, and other systems from scratch in user-land.

The goal of this RFC is to establish a standard for writing concurrent code in PHP, as well as a C-API interface that would allow PHP to be extended at a low level using C, Rust, C++, and other languages. This would enable extensions to support non-blocking I/O without the need to override PHP functions or duplicate code.

The True Async project pursues the following goals and values:

• From a PHP developer's perspective, the main value of this implementation is that they DO NOT NEED to change existing code (or if changes are required, they should be minimal) to enable concurrency. Unlike explicit async models, this approach lets developers reuse existing synchronous code inside fibers without modification.

• Code that was originally written and intended to run outside of a Coroutine must work EXACTLY THE SAME inside a Coroutine without modifications.

• A PHP developer should not have to think about how Coroutine switch and should not need to manage their switching—except in special cases where they consciously choose to intervene in this logic.

• If there is existing code or a familiar style, such as AMPHP interfaces, Go coroutines, Swoole API, and others, it is best to use what is most recognizable to a broad range of developers.

• The goal is to find a balance between flexibility and simplicity. On one hand, the implementation should allow leveraging existing solutions without requiring external libraries. On the other hand, it should avoid unnecessary complexity. Many design choices in this implementation are driven by the desire to free developers from concerns about compatibility with “external libraries” in favor of a standardized approach.

This RFC is quite complex due to the number of logical connections. Please use the diagrams from the table to simplify understanding.

This RFC describes the API and new syntax for writing concurrent code in PHP, which includes:

A lightweight execution thread that can be suspended (suspend) and resumed. Example:

spawn {
    echo "Start";
    suspend;  // Suspend the coroutine
    echo "Resumed";
};

A container that manages coroutine lifetimes. Example:

$scope = new Async\Scope();
spawn with $scope {
    // Coroutine bound to $scope
    spawn {
        // Coroutine bound to $scope
    };
};
// Dispose of the scope after 5 seconds
sleep(5);
$scope->disposeSafely();

Explicit group of coroutines with centralized result/error handling. Example:

$taskGroup = new Async\TaskGroup(captureResults: true);
spawn with $taskGroup task1();
spawn with $taskGroup task2();
[$result1, $result2] = await $taskGroup;

A special exception that implements cooperative cancellation: Example:

$coroutine = spawn {
    try {
        Async\delay(1000);
    } catch (Async\CancellationException $e) {
        echo "Coroutine cancelled";
    }
};
 
suspend;
 
$coroutine->cancel();

Coroutine/Scope-associated data storage. Example:

currentContext()->set('user_id', 123);
spawn {
    $userId = currentContext()->get("user_id");
    echo "User ID: $userId"; // 123
};

Group awaitable objects like any(), all(), ignoreError(). Example:

$results = await Async\all([spawn task1(), spawn task2()]);

spawn {
    $result = file_get_contents("file.txt");
 
    if($result === false) {
        echo "Error reading file.txt\n";
    }
 
    echo "File content: $result\n";
};
 
echo "Next line\n";

Output:

Next line
File content: ...

function fetchData(string $file): string
{
    $result = file_get_contents($file);
 
    if($result === false) {
        throw new Exception("Error reading $file");
    }
 
    return $result;
}
echo await spawn fetchData("file.txt");

echo await spawn fetchData("https://php.net/") until Async\timeout(2000);
echo await spawn fetchData("https://php.net/") until spawn sleep(2);

Transferring control from the coroutine to the Scheduler:

function myFunction(): void {
    echo "Hello, World!\n";
    suspend;
    echo "Goodbye, World!\n";
}
 
spawn myFunction();
echo "Next line\n";

Output:

Hello, World
Next line
Goodbye, World

function mergeFiles(string ...$files): string
{
    $taskGroup = new Async\TaskGroup(captureResults: true);
 
    foreach ($files as $file) {
       spawn with $taskGroup file_get_contents($file);
    }
 
    return array_merge("\n", await $taskGroup);
}

use Async\Scope;
use Async\TaskGroup;
use Async\AwaitCancelledException;
 
/**
 * Retrieves user profile data from different sources
 */
function getUserProfile(int $userId): array
{
    // Main task group with result capturing
    $profileTasks = new TaskGroup(captureResults: true);
 
    try {
        // Start fetching basic user information
        spawn with $profileTasks use ($userId):array {
            $userData = await spawn fetchUserData($userId);
            return ['basic' => $userData];
        };
 
        // Start fetching extended details in parallel
        spawn with $profileTasks use ($userId) {
            // Subtask group for supplementary information
            $orderTasks = new TaskGroup(Scope::inherit(), captureResults: true);
 
            // Request user orders with 2-second timeout
            spawn with $orderTasks use ($userId): array {
                try {
                    return ['orders' => await spawn fetchUserOrders($userId) until timeout(2000)];
                } catch (AwaitCancelledException) {
                    return ['orders' => ['status' => 'timeout']];
                }
            };
 
            // Request API2 user orders with 2-second timeout
            spawn with $orderTasks use ($userId): array {
                try {
                    return ['orders2' => await spawn fetchUserOrdersAPI2($userId) until timeout(2000)];
                } catch (AwaitCancelledException) {
                    return ['orders2' => ['status' => 'timeout']];
                }
            };
 
            // Wait for all order tasks to complete
            return array_merge(...await $orderTasks);
        };
 
        // Request user settings
        spawn with $profileTasks use ($userId): array {
          return ['settings' => await spawn fetchUserSettings($userId)];
        };
 
        // Merge all results into a single profile
        return array_merge(...await $profileTasks);
 
    } catch (Exception $e) {
        // Error handling
        error_log("Error fetching user profile: " . $e->getMessage());
        return ['error' => $e->getMessage()];
    }
}
 
// Usage
$profile = await spawn getUserProfile(123);
var_dump($profile);

function processBackgroundJobs(string ...$jobs): array
{
    $scope = new Scope();
 
    foreach ($jobs as $job) {
        spawn with $scope processJob($job);
    }
 
    // Waiting for all child tasks throughout the entire depth of the hierarchy.
    $scope->awaitCompletion(Async\timeout(300 * 1000));
}
 
function processJob(mixed $job): void {
    $scope = \Async\Scope::inherit();
    spawn $scope task1($job);
    spawn $scope task2($job);
 
    // Waiting for all child tasks in the current scope.
    $scope->awaitCompletion(Async\timeout(300 * 1000));
}

class HttpClient
{
    private Scope $scope;
 
    public function __construct()
    {
        $this->scope = new Scope();
    }
 
    public function request(array $data): \Async\Awaitable
    {
        return spawn with $this->scope use($data) {
            // This coroutine is bound to the MyClass instance
        };
    }
}
 
$service = new HttpClient;
$service->request(['login' => 'admin', 'password' => '1234']);
// HttpClient instance will stop all coroutines bound to it.
unset($service);

use Async\TaskGroup;
 
function fetchFirstSuccessful(string ...$apiHosts): string
{
    $taskGroup = new Async\TaskGroup(captureResults: false);
 
    foreach ($apiHosts as $host) {
        spawn with $taskGroup file_get_contents($host);
    }
 
    // Get the first successful result
    return await $taskGroup->race(ignoreErrors: true);
}

The implementation of this RFC should be carried out in a way that minimizes changes to the PHP core.

The proposed changes include: * syntax modifications to the language, * interfaces without implementations for Reactor and Scheduler, * an internal interface for the Scope class.

All other classes and functions from this RFC will be moved to a separate module, which will become part of the standard library. However, this module can be replaced with a different one if necessary.

The Coroutine, Scope and TaskGroup classes are not part of the PHP core.

The behavior of expressions and statements such as spawn, await, and suspend is not defined in the core, and can be overridden by an extension, which must adhere to the logic defined in this RFC.

Scheduler and Reactor must be implemented as PHP extensions that implement low-level interfaces.

The Scheduler and Reactor interfaces are part of the implementation of this RFC.

The behavior of Scheduler and Reactor must not contradict the logic of the RFC. Components cannot override the logic of expressions such as spawn, async, suspend, and so on. However, this RFC does not impose any restrictions on extending functionality.

It is allowed to use the Async namespace for new functions or objects in Scheduler and Reactor.

⚠️ Warning: Users should not make assumptions about the execution order of coroutines unless
this is a specific goal of a particular Scheduler implementation.

The Reactor is a component that implements the Event Loop. It may be exposed as a separate API in PHP-land, but its behavior is not defined within this RFC.

PHP allows for the implementation of forced coroutine suspension, which can be used in a preemptive multitasking algorithm.

This capability is particularly implemented in Swoole. However, the current RFC rejects preemptive multitasking due to the unpredictable behavior of code during context switches.

A coroutine can lose control literally at any PHP opcode, which can significantly affect the outcome and contradict the programmer's expectations. Writing code that can lose control at any moment is a complex domain where PHP does not seem like an adequate tool.

This RFC considers a scenario where a coroutine is abruptly stopped only in one case: if the Scheduler implements a runtime control mechanism similar to max_execution_time.

Please see [Maximum activity interval](#maximum-activity-interval) for more information.

This RFC is based on the principle of “Cancellable by design”, which can be described as follows:

By default, coroutines should be designed in such a way that their
cancellation at any moment does not compromise data integrity.

Coroutines launched without a defined Scope or lifetime must adhere to the “Cancellable by design” principle.

If a coroutine’s lifetime needs to be controlled — it MUST be done EXPLICITLY!

In practice, this means that if a coroutine is created using the expression spawn <callable>, the developer treats it as non-critical in terms of data integrity. If the developer needs to manage the coroutine’s lifetime, they will use the expression spawn with. In other words, the developer must take extra steps to explicitly extend the coroutine's lifetime.

All functions, classes, and constants defined in this RFC are located in the Async namespace. Extensions for Scheduler/Reactor are allowed to extend this namespace with functions and classes, provided that they are directly related to concurrency functionality.

A Coroutine is an execution container, transparent to the code,
that can be suspended on demand and resumed at any time.

Isolated execution contexts make it possible to switch between coroutines and execute tasks concurrently.

Any function can be executed as a coroutine without any changes to the code.

A coroutine can stop itself bypassing control to the Scheduler. However, it cannot be stopped externally.

⚠️ Warning:
It is permissible to stop a coroutine’s execution externally for two reasons:
* To implement multitasking.
* To enforce an active execution time limit.
Please see [Maximum activity interval](#maximum-activity-interval) for more information.

A suspended coroutine can be resumed at any time. The Scheduler component is responsible for the coroutine resumption algorithm.

A coroutine can be resumed with an exception, in which case an exception will be thrown from the suspension point.

This state diagram illustrates the lifecycle of a coroutine, showing how it transitions through various states during its execution:

States:

1. Created – The coroutine has been defined but not yet started.
2. Queued – The coroutine is queued
3. Running – The coroutine is actively executing.
4. Suspended – Execution is paused, usually waiting for a result or I/O.
5. Completed – The coroutine has finished successfully (via return).
6. Pending Cancellation – A cancellation was requested; the coroutine is cleaning up.

Key Transitions:

1. spawn moves a coroutine from Created to Running.
2. suspend and resume move it between Running and Suspended.
3. return/exit ends it in Completed.
4. cancel() initiates cancellation from Running or Suspended, leading to Pending Cancellation, and finally Cancelled.

To create coroutines, the spawn <callable> expression is used. It launches the <callable> in a separate execution context and returns an instance of the Async\Coroutine class as a result.

Let's look at two examples:

$result = file_get_contents('https://php.net');
echo "next line".__LINE__."\n";

This code:

1. first returns the contents of the PHP website,
2. then executes the ''echo'' statement.

$coroutine = spawn file_get_contents('https://php.net');
echo "next line".__LINE__."\n";

This code:

1. starts a coroutine with the ''file_get_contents'' function.
2. The next line is executed without waiting for the result of ''file_get_contents''.
3. The coroutine is executed after the ''echo'' statement.

The spawn construct is available in two variations:

• spawn function_call - creates a coroutine from a callable expression
• spawn closure_block - creates a coroutine and defines a closure

// Executing a known function
spawn [with <scope>] <function_call>;
 
// Closure form
spawn [with <scope>] [static] [use(<parameters>)][: <returnType>] {
    <codeBlock>
};

*where:*

function_call - a valid function call expression:

- call a standard PHP function:

spawn file_get_contents('file1.txt');

- call a user-defined function:

function example(string $name): void {
    echo "Hello, $name!";
}
 
spawn example('World');

- call a static method:

spawn Mailer::send($message);

- call a method of an object:

$object = new Mailer();
spawn $object->send($message);

- self, static or parent keyword:

spawn self::send($message);
spawn static::send($message);
spawn parent::send($message);

- call $class method:

$className = 'Mailer';
spawn $className::send($message);

- expression:

// Use result of foo()
spawn (foo())();
// Use foo as a closure
spawn (foo(...))();
// Use ternary operator
spawn ($option ? foo() : bar())();
// Scary example
spawn (((foo())))();

- call array dereference:

$array = [fn() => sleep(1)];
spawn $array[0]();

- new dereference:

class Test {
    public function wait(): void {
        sleep(1);
    }
}
 
spawn new Test->wait();

- call dereferenceable scalar:

spawn "sleep"(5);

- call short closure:

spawn (fn() => sleep(1))();

The spawn or with keywords does not affect the ability to define functions, constants, or classes with the same name:

function spawn(): void {} // <- Allowed
 
const SPAWN = 1; // <- Allowed
 
class Spawn {} // <- Allowed

The spawn expression cannot appear before namespace or use declarations:

spawn myFunction(); // <- Not allowed
namespace MyNamespace;
use MyNamespace\MyClass;

Allows creating a coroutine from a closure directly when using spawn:

spawn [with <scope>] [static] [use(<parameters>)[: <returnType>]] {
    <codeBlock>
};

- full form:

$file = 'main.log';
 
spawn use($file): string {
    $result = file_get_contents($file);
 
    if($result === false) {
        throw new Exception("Error reading $file");
    }
 
    return $result;
};

- short form:

spawn {
    return file_get_contents('main.log');
};

- with return type:

spawn use():string {
    return file_get_contents('main.log');
};

- with static keyword:

Define closure as static:

class Test {
    private $property = 'main.log';
    public function method(): void {
        spawn static {
            //$this->property <- not available
        };
    }
}

The with keyword allows specifying the scope in which the coroutine.

The operand for spawn with $scope can be either an Async\Scope object or a class that implements the Async\ScopeProvider interface. For example, such a class is TaskGroup.

$scope = new Async\Scope();
$taskGroup = new Async\TaskGroup($scope);
 
$coroutine = spawn with $scope use():string {
    return gethostbyname('php.net');
};
 
$coroutine = spawn with $taskGroup use():string {
    return gethostbyname('php.net');
};
 
function defineTargetIpV4(string $host): string {
    return gethostbyname($host);
}
 
spawn with $scope defineTargetIpV4($host);
spawn with $taskGroup defineTargetIpV4($host);

The scope expression can be: - A variable of the Async\ScopeProvider type:

spawn with $scope use():void {
    echo gethostbyname('php.net').PHP_EOL;
};

- The result of a method or function call:

spawn with $this->scope $this->method();
spawn with $this->getScope() $this->method();

The ScopeProvider interface allows objects to provide an Async\Scope instance that can be used in a spawn with expression.

This is useful when you want to abstract the scope management logic, letting higher-level structures (like a task group or a custom container) expose a scope without directly exposing internal details.

<?php
 
declare(strict_types=1);
 
namespace Async;
 
interface ScopeProvider
{
    /**
     * Returns the associated Scope instance.
     *
     * This scope will be used when spawning a coroutine via ''spawn with $provider''.
     *
     * @return Scope|null
     */
    public function provideScope(): ?Scope;
}

The provideScope method may return NULL; in this case, the current Scope will be used.

Example Use Case:

A task group can implement this interface to automatically provide its internal scope to spawn with:

class CustomTaskGroup implements ScopeProvider
{
    private Scope $scope;
 
    public function __construct()
    {
        $this->scope = new Scope();
    }
 
    public function provideScope(): ?Scope
    {
        return $this->scope;
    }
}

This allows you to spawn coroutines into the task group using:

spawn with $taskGroup {
    // This coroutine is bound to the TaskGroup's scope
};

The SpawnStrategy interface allows attaching a newly spawned coroutine to a custom user-defined context immediately after the spawn with expression is evaluated.

This is useful for scenarios where the coroutine should be registered, tracked, or logically grouped within a context (e.g., a TaskGroup or a custom task manager).

interface SpawnStrategy extends ScopeProvider
{
    /**
     * Called before a coroutine is spawned, before it is enqueued.
     *
     * @param   Coroutine   $coroutine  The coroutine to be spawned.
     * @param   Scope       $scope      The Scope instance.
     *
     */
    public function beforeCoroutineEnqueue(Coroutine $coroutine, Scope $scope): array;
 
    /**
     * Called after a coroutine is spawned, enqueued.
     *
     * @param Coroutine $coroutine
     * @param Scope     $scope
     */
    public function afterCoroutineEnqueue(Coroutine $coroutine, Scope $scope): void;
}

If the $scope object in a spawn with expression implements the SpawnStrategy interface, then the acceptCoroutine method will be called immediately after the coroutine is created.

Example:

A class like CustomTaskGroup might implement this interface to automatically collect all spawned coroutines under its management:

class CustomTaskGroup implements Async\SpawnStrategy
{
    private array $coroutines = [];
 
    public function afterCoroutineEnqueue(Coroutine $coroutine, Scope $scope): void
    {
        $this->coroutines[] = $coroutine;
        echo "Coroutine added to the group as ".$coroutine->getSpawnLocation()."\n";
    }
 
    // Additional methods for managing the group...
}
 
$customTaskGroup = new CustomTaskGroup();
 
spawn with $customTaskGroup {
    // This coroutine will be automatically added to the custom task group
};

The beforeCoroutineEnqueue() method is called after the coroutine has been created, but before it is added to the queue. It allows for additional operations to be performed with the coroutine and its context, and it returns an optional list of options for the Scheduler.

The list of options for the Scheduler is not part of this RFC
and is defined by the Scheduler implementation.

class HiPriorityStrategy implements Async\SpawnStrategy
{
    public function beforeCoroutineEnqueue(Coroutine $coroutine, Scope $scope): array
    {
        // Mark the coroutine as high priority before it is enqueued
        $coroutine->asHiPriority();
 
        return [];
    }
 
    // Additional methods ...
}
 
spawn with new HiPriorityStrategy() {
    // This coroutine will be marked as high priority
};

The Async\hiPriority(?Scope $scope = null) function allows launching a coroutine with high priority:

use function Async\hiPriority;
 
spawn {
    echo "normal priority\n";
};
 
spawn with hiPriority() {
    echo "high priority\n";
};

Expected output:

high priority
normal priority

If the $scope parameter is not specified, the current Scope will be used for launching.

The hiPriority strategy marks the coroutine as high-priority using the Coroutine::asHiPriority() method.

This action serves as a recommendation for the Scheduler, suggesting that the coroutine should be placed as close to the front of the queue as possible. However, the programmer MUST NOT rely on this outcome.

hiPriority can be useful in situations where resources need to be released as quickly as possible or when a critical section of code must be executed promptly. The programmer should not overuse it, as this may negatively affect the application's performance.

A coroutine can suspend itself at any time using the suspend keyword:

function example(string $name): void {
    echo "Hello, $name!";
    suspend;
    echo "Goodbye, $name!";
}
 
spawn example('World');
spawn example('Universe');

Expected output:

Hello, World!
Hello, Universe!
Goodbye, World!
Goodbye, Universe!

Basic syntax:

    suspend;

Wrong use:

    // not part of the expression
    suspend + $any;
    // not parameter
    my_function(suspend);

The suspend statement is considered executable code. Therefore, it MUST NOT appear before any use or namespace declarations at the top-level scope:

suspend; // <- Not allowed
namespace MyNamespace;
suspend; // <- Not allowed
use MyNamespace\MyClass;

The suspend keyword must not conflict with a function name, class name, or constant:

function suspend(): void { // <- Allowed
 
}
 
const SUSPEND = 1; // <- Allowed
 
class Suspend { // <- Allowed
}

The suspend keyword can be used only for the current coroutine.

The suspend keyword has no parameters and does not return any values, unlike the yield keyword.

The suspend keyword can be used in any function including from the main execution flow:

function example(string $name): void {
    echo "Hello, $name!";
    suspend;
    echo "Goodbye, $name!";
}
 
$coroutine = spawn example('World');
 
// suspend the main flow
suspend;
 
echo "Back to the main flow";

Expected output:

Hello, World!
Back to the main flow
Goodbye, World!

The suspend keyword can be a throw point if someone resumes the coroutine externally with an exception.

function example(string $name): void {
    echo "Hello, $name!";
 
    try {
        suspend;
    } catch (Exception $e) {
        echo "Caught exception: ", $e->getMessage();
    }
 
    echo "Goodbye, $name!";
}
 
$coroutine = spawn example('World');
 
// pass control to the coroutine
suspend;
 
$couroutine->cancel();

Expected output:

Hello, World!
Caught exception: cancelled at ...
Goodbye, World!

I/O operations invoked within the coroutine's context transfer control implicitly:

spawn function:void {
    echo "Start reading file1.txt\n";
    file_get_contents('file1.txt');
    echo "End reading file1.txt\n";
}
spawn function:void {
    echo "Start reading file2.txt\n";
    file_get_contents('file2.txt');
    echo "End readingfile2.txt\n";
}
 
echo "Main flow";

Expected output:

Start reading file1.txt
Start reading file2.txt
Main flow
End reading file1.txt
End reading file2.txt

Inside each coroutine, there is an illusion that all actions are executed sequentially, while in reality, operations occur asynchronously.

This RFC proposes support for core PHP functions that require non-blocking input/output, as well as support for cURL, Socket, and other extensions based on the PHP Stream API. Please see Unaffected PHP Functionality.

The Awaitable interface is a contract that allows objects to be used in the await expression.

The interface does not have any methods on the user-land side and is intended for objects implemented as PHP extensions, such as:

- Future - Cancellation

The following classes from this RFC also implement this interface:

- Coroutine - TaskGroup

Unlike Future, the Awaitable interface does not impose limitations on the number of state changes, which is why the Future contract is considered a special case of the Awaitable contract.

In the general case, objects implementing the Awaitable interface can act as triggers — that is, they can change their state an unlimited number of times. This means that multiple calls to await <Awaitable> may produce different results.

In contrast, Coroutine, Future and Cancellation objects change their state only once, so using them multiple times in an await expression will always yield the same result.

Comparison of Different Awaitable Classes:

The await keyword is used to wait for the completion of another coroutine or any object that implements the Awaitable interface.:

function readFile(string $fileName):string
{
    $result = file_get_contents($fileName);
 
    if($result === false) {
        throw new Exception("Error reading file1.txt");
    }
 
    return $result;
}
 
$coroutine = spawn readFile('file1.txt');
 
echo await $coroutine;
// or
echo await spawn readFile('file2.txt');

await suspends the execution of the current coroutine until the awaited one returns a final result or completes with an exception.

function testException(): void {
    throw new Exception("Error");
}
 
try {
    await spawn testException();
} catch (Exception $e) {
    echo "Caught exception: ", $e->getMessage();
}

Await basic syntax:

    await <awaitExp> [until <cancellationExp>];

where:

- awaitExp - An expression whose result must be an object with the Async\Awaitable interface. - cancellationExp - An expression that limits the waiting time.

Must be an object with the ''Async\Awaitable'' interface.

Await expression:

- A variable of the Awaitable interface

    $readFileJob = spawn file_get_contents('file1.txt');
 
    $result = await $readFileJob;

- A function that returns an Async\Awaitable object:

    function getContentsJobStarter(string $fileName): \Async\Coroutine {
        return spawn file_get_contents($fileName);
    }
 
    $result = await getContentsJobStarter('file1.txt');

- A new coroutine:

    $result = await spawn file_get_contents('file1.txt');

- A new Awaitable object:

    $result = await new Async\Future();

- A static method:

    $result = await SomeClass::create();

- A method of an object:

    $service = new Mailer();
    $result = await $service->sendMail("[email protected]", "Hello!");

- A method of a class:

    $serviceClass = 'Mailer';
    $result = await $serviceClass::sendAll();

- A valid expression:

    $result = await ($bool ? foo() : bar());

The await expression can be used just like any other expression together with other statements:

    $coroutine = spawn {...};
 
    if(await $coroutine === true) {
        echo "Success";
    } else {
        echo "Failure";
    }
 
    foreach (await spawn $coroutine as $value) {
        echo $value;
    }

The await or until keywords does not affect the ability to define functions, constants, or classes with the same name:

function await(): void {} // <- Allowed
 
const AWAIT = 1; // <- Allowed
$x = AWAIT + 2;  // <- Allowed
 
class Await {} // <- Allowed

The await expression cannot appear before namespace or use declarations:

await file_get_contents('file1.txt'); // <- Not allowed
namespace MyNamespace;
use MyNamespace\MyClass;

The expression await + until <cancellationExp> allows limiting the wait time to the moment the cancellationExp trigger fires.

When cancellationExp is triggered, an exception Async\AwaitCancelledException is thrown.

The wait operation is often combined with a cancellation token. In modern programming languages, the cancellation token is typically passed as an additional parameter to functions, which makes the semantics somewhat unclear.

For example:

await all([...], $cancellation);

Clearer semantics would allow us to logically and visually separate the wait operation into two conditions: 1. What we're waiting for 2. How long we're willing to wait

For example:

await all([...]) until $cancellation;
// or if timeout() returns a awaitable object
await all([...]) until timeout(5);

basic syntax:

    await <awaitExp> [until <cancellationExp>];

where cancellationExp:

- A variable of the Async\Awaitable interface

    $cancellation = Async\timeout(5000);
    $result = await $coroutine until $cancellation;

- A function that returns an Awaitable object

    function getCancellation(): \Async\Awaitable {
        return spawn sleep(5);
    }
 
    $result = await $coroutine until getCancellation();

- A new coroutine

    $result = await $coroutine until spawn sleep(5);

The until keyword allows using coroutines as a CancellationToken. If an exception occurs in a coroutine that participates in until, that exception will be thrown at the point where the await expression is called.

Example:

function cancellationToken(): void {
    throw new Exception("Error");
}
 
try {
    await spawn sleep(5) until spawn cancellationToken();
} catch (Exception $exception) {
    echo "Caught exception: ", $exception->getMessage();
}

Expected output:

Caught exception: Error

⚠️ Warning: Note that completing the coroutine's await
does not affect the lifetime of the coroutine used with until.

The use of spawn/await/suspend is allowed in almost any part of a PHP program. This is possible because the PHP script entry point forms the main execution thread, which is also considered a coroutine.

As a result, keyword like suspend and currentCoroutine() will behave the same way as in other cases.

If only one coroutine exists in the system, calling suspend will immediately return control.

The register_shutdown_function handler operates in synchronous mode, after asynchronous handlers have already been destroyed. Therefore, the register_shutdown_function code should not use the concurrency API. The suspend keyword will have no effect, and the spawn operation will not be executed at all.

Coroutine Scope — the space associated with coroutines created using the spawn expression.

Sometimes it is necessary to gain control not only over a currently running coroutine, but also over all coroutines that will be launched within a new one — without having direct access to them. This could be the case for web server code that handles requests in separate coroutines and does not know how many additional coroutines will be launched, or a JobExecutor that wants to manage the lifecycle of running jobs.

Without such control, the application code loses the ability to resist runtime errors, which increases the risk of a complete service failure.

This is why the Coroutine Scope pattern is of critical importance in the context of ensuring reliability.

The main use cases for Scope are:

1. Controlling the lifetime of coroutines created within a single scope (Point of responsibility) 2. Handling errors from all coroutines within the scope 3. Binding the lifetime of the scope's coroutines to the lifetime of a PHP object 4. Creating a hierarchy of scopes to manage coroutines in a structured way

Binding Scope to objects is a good practice that has proven effective in Kotlin. By allowing coroutines to be tied to an object (this could be a Screen or a ViewModel), it is possible to avoid the error where coroutines outlive the object that manages them.

For frameworks, it can be useful to be able to control all coroutines created within a Scope, to apply context-dependent constraints to them.

By default, all coroutines are associated with the Global Coroutine Scope:

spawn file_get_contents('file1.txt'); // <- global scope
 
function readFile(string $file): void {
    return file_get_contents($file); // <- global scope
}
 
function mainTask(): void { // <- global scope
    spawn readFile('file1.txt'); // <- global scope
}
 
spawn mainTask(); // <- global scope

If an application never creates custom Scopes, its behavior is similar to coroutines in Go:

• Coroutines are not explicitly linked to each other.
• The lifetime of coroutines is not limited.

The expression spawn with $scope creates a new coroutine bound to the specified scope. Scope is propagated between coroutines. If a coroutine is launched within a specific Scope, that Scope is considered the current one. Any expression like spawn <callable> will create a coroutine within the current Scope.

Coroutines created during the execution of this new coroutine will become sibling tasks:

use Async\Scope;
 
$scope = new Scope();
 
spawn with $scope { // <- new scope
 
    echo "Sibling task 1\n";
 
    spawn { // <- $scope is current scope
        echo "Sibling task 2\n";
 
        spawn { // <- $scope is current scope
            echo "Sibling task 3\n";
        };
    };
};
 
$scope->awaitCompletion(Async\signal(SIGTERM));

Expected output:

Sibling task 1
Sibling task 2
Sibling task 3

Structure:

main()                          ← defines a $scope
└── $scope = new Scope()
    ├── task1()                 ← runs in the $scope
    ├── task2()                 ← runs in the $scope
    ├── task3()                 ← runs in the $scope

Thus, the expression spawn with $scope creates a new branch of sibling coroutines, where the new coroutine exists at the same level as all subsequent ones.

The code that owns a Scope object becomes the Point of responsibility for all coroutines executed within that Scope.

A good practice is to ensure that a Scope object has only ONE owner.
Passing $scope as a parameter to other functions or assigning it to multiple objects
is a potentially dangerous operation that can lead to complex bugs.
If you need to interact with the Scope in different parts of the program,
use Async\ScopeProvider containers or other appropriate mechanisms.

Warning: In general, it is strongly discouraged to wait on a Scope; instead, prefer using a TaskGroup.

The Scope class does not implement the Awaitable interface, and therefore cannot be used in an await expression. Awaiting a Scope is a potentially dangerous operation that should be performed consciously, not accidentally.

There are several Use-Cases where waiting for a Scope might be necessary: * Structured concurrency: when a parent awaits the completion of all child coroutines. * Waiting for Scope tasks to complete the cancellation process.

The structured concurrency pattern with waiting for all child coroutines can be useful for applications whose lifetime is explicitly limited by external conditions. For example, the user might stop a console application.

To support a task awaiting in a controlled manner, Scope provides two specific methods:

- public function awaitCompletion(Awaitable $cancellation): void {} - public function awaitAfterCancellation(?callable $errorHandler = null, ?Awaitable $cancellation = null): void {}

The awaitCompletion method blocks the execution flow until all tasks within the scope are completed. The awaitAfterCancellation method does the same but is intended to be called only after the scope has been cancelled.

use Async\Scope;
 
$scope = new Scope();
 
spawn with $scope {
 
    echo "Sibling task 1\n";
 
    spawn {
        echo "Sibling task 2\n";
 
        spawn {
            echo "Sibling task 3\n";
        };
    };
};
 
$scope->awaitCompletion(Async\signal(SIGTERM));

Expected output:

Sibling task 1
Sibling task 2
Sibling task 3

The Scope awaiting methods do not capture any task results, so they cannot be used to await return values.

ℹ️ Note: If you need to retrieve the result of a group of tasks, use the Async\TaskGroup class.

The awaitCompletion method can only be used with an explicitly defined cancellation token. This requirement helps prevent indefinite waiting.

Awaiting the $scope object also allows handling exceptions from coroutines within the $scope:

use Async\Scope;
 
$scope = new Scope();
 
spawn with $scope {
    spawn {
        spawn {
            throw new Exception("Error occurred");
        };
    };
};
 
try {
    $scope->awaitCompletion(Async\signal(SIGTERM));
} catch (Exception $exception) {
    echo $exception->getMessage()."\n";
}

Expected output:

Error occurred

Calling the awaitCompletion() method after the Scope has been cancelled will immediately throw a cancellation exception.

$scope = new Scope();
 
try {
    spawn with $scope task1();
    spawn with $scope task2();
    $scope->cancel();
    // Wait all tasks
    $scope->awaitCompletion(Async\signal(SIGTERM));
} catch (Exception $exception) {
    echo "Caught exception: ",$exception->getMessage()."\n";
}

Expected output:

Caught exception: cancelled at ...

In this example, $scope->awaitCompletion(Async\signal(SIGTERM)); will immediately throw an exception.

If you need to wait for the Scope to complete after it has been cancelled, use the special method awaitAfterCancellation, which is designed for this case.

$scope = new Scope();
 
spawn {
    try {
        $scope->awaitCompletion(Async\signal(SIGTERM));
    } catch (\Async\CancellationException $exception) {
        $scope->awaitAfterCancellation();
        echo "Caught exception: ",$exception->getMessage()."\n";
    }
};
 
spawn with $scope use($scope) {
    $scope->cancel();
 
    try {
        Async\delay(1000);
    } finally {
        sleep(1);
        echo "Finally\n";
    }
};

Expected output:

Finally
Caught exception: cancelled at ...

In this example, the line Finally will be printed first because $scope->awaitAfterCancellation() waits for all coroutines inside the Scope to complete.

The awaitAfterCancellation method is used in scenarios where final resource cleanup is required after all child tasks are guaranteed to have finished execution.

There is also a risk of indefinite waiting, so it is recommended to explicitly specify a timeout.

A hierarchy can be a convenient way to describe an application as a set of dependent tasks:

• Parent tasks are connected to child tasks and are responsible for their execution time.
• Tasks on the same hierarchy level are independent of each other.
• Parent tasks should control their child's tasks.
• Child tasks MUST NOT control or wait for their parent tasks.
• It is correct if tasks at the same hierarchy level are only connected to tasks of the immediate child level.

WebServer
├── Request Worker
│   ├── request1 task
│   │   ├── request1 subtask A
│   │   └── request1 subtask B
│   └── request2 task
│       ├── request2 subtask A
│       └── request2 subtask B

The work of a web server can be represented as a hierarchy of task groups that are interconnected. The Request Worker is a task responsible for handling incoming requests. There can be multiple requests. Each request may spawn subtasks. On the same level, all requests form a group of request-tasks.

Scope is fit for implementing this concept:

WebServer
├── Request Worker
│   ├── request1 Scope
│   │   ├── request1 subtask A
│   │   │   └── subtask A Scope
│   │   │       ├── sub-subtask A1
│   │   │       └── sub-subtask A2
│   │   └── request1 subtask B
│   └── request2 Scope
│       ├── request2 subtask A
│       └── request2 subtask B
│           └── subtask B Scope
│               └── sub-subtask B1

A new child Scope can be created using a special constructor: Scope::inherit(). It returns a new Scope object that acts as a child. A coroutine created within the child Scope can also be considered a child relative to the coroutines in the parent Scope.

An example:

use Async\Scope;
use Async\CancellationException;
 
function connectionChecker($socket, callable $cancelToken): void
{
    while (true) {
        if(feof($socket)) {
            $cancelToken("The connection was closed by user");
            return;
        }
 
        Async\delay(1000); // throw CancellationException if canceled
    }
}
 
function connectionLimiter(callable $cancelToken): void
{
   Async\delay(10000);
   $cancelToken("The request processing limit has been reached.");
}
 
function connectionHandler($socket): void
{
    // Note that a parent Scope can stop the execution of all coroutines
    // belonging to a child Scope at any moment.
    $scope = Scope::inherit();
 
    //
    // Passing ''$scope'' via ''use'' into a single coroutine is equivalent to the logic:
    // the lifetime of ''$scope'' equals the lifetime of the coroutine.
    // In this way, we create a coroutine-closure that acts as a Point of Responsibility.
    // This code is one example of how to implement Points of Responsibility.
    //
    spawn with $scope use($socket, $scope) {
 
        $limiterScope = Scope::inherit(); // child scope for connectionLimiter and connectionChecker
 
        // We do not provide direct access to the Scope object in other functions, because this is an antipattern!
        $cancelToken = fn(string $message) => $scope->cancel(new CancellationException($message));
 
        // Limiter coroutine
        spawn with $limiterScope connectionLimiter($cancelToken);
 
        // A separate coroutine checks that the socket is still active.
        spawn with $limiterScope connectionChecker($socket, $cancelToken);
 
        try {
            sendResponse($socket, dispatchRequest(parseRequest($socket)));
        } catch (\Exception $exception) {
            fwrite($socket, "HTTP/1.1 500 Internal Server Error\r\n\r\n");
        } finally {
            fclose($socket);
            // Explicitly cancel all coroutines in the child scope
            $scope->cancel();
        }
    };
}
 
function socketServer(): void
{
    // Main Server $scope
    $scope = new Scope();
 
    // Child coroutine that listens for a shutdown signal
    //
    // Note that we are passing ''$scope'' to another function!
    // This is acceptable here because the code is within a single visual block,
    // and the risk of an error due to oversight is minimal.
    spawn with $scope use($scope) {
        try {
            // Note: The ''signal'' function is not part of this RFC,
            // but it may be implemented in the standard library in the future.
            // This example shows how such a function could be used.
            // The ''signal'' function returns a trigger ''Awaitable''.
            await Async\signal(SIGINT);
        } finally {
            $scope->cancel(new CancellationException('Server shutdown'));
        }
    }
 
    try {
       // The main coroutine that listens for incoming connections
       // The server runs as long as this coroutine is running.
       await spawn with $scope {
           while ($socket = stream_socket_accept($serverSocket, 0)) {
               connectionHandler($socket);
           }
       };
    } catch (\Throwable $exception) {
        echo "Server error: ", $exception->getMessage(), "\n";
    } finally {
        echo "Server should be stopped...\n";
 
        // Graceful exit
        try {
            $scope->cancel();
            // Await for all coroutines to finish but not more than 5 seconds
            $scope->awaitAfterCancellation(cancellation: Async\timeout(5000));
            echo "Server stopped\n";
        } catch (\Throwable $exception) {
            // Force exit
            echo "Server error: ", $exception->getMessage(), "\n";
            throw $exception;
        }
    }
}

Let's examine how this example works.

1. socketServer creates a new Scope for coroutines that will handle all connections. 2. Each new connection is processed using connectionHandler() in a separate Scope,

 which is inherited from the main one.

3. connectionHandler creates a new Scope for the connectionLimiter and connectionChecker coroutines. 4. connectionHandler creates coroutine: connectionLimiter() to limit the processing time of the request. 5. connectionHandler creates coroutine, connectionChecker(), to monitor the connection's activity.

 As soon as the client disconnects, ''connectionChecker'' will cancel all coroutines related to the request.

6. If the main Scope is closed, all coroutines handling requests will also be canceled.

GLOBAL <- globalScope
│
├── socketListen (Scope) <- rootScope
│   │
│   ├── connectionHandler (Scope) <- request scope1
│   │   └── connectionLimiter (Coroutine) <- $limiterScope
│   │   └── connectionChecker (Coroutine) <- $limiterScope
│   │
│   ├── connectionHandler (Scope) <- request scope2
│   │   └── connectionLimiter (Coroutine) <- $limiterScope
│   │   └── connectionChecker (Coroutine) <- $limiterScope
│   │

The connectionHandler doesn't worry if the lifetimes of the connectionLimiter or connectionChecker coroutines exceed the lifetime of the main coroutine handling the request, because it is guaranteed to call $scope->cancel() when the main coroutine finishes.

$limiterScope is used to explicitly define a child-group of coroutines that should be cancelled when the request is completed. This approach minimizes errors.

On the other hand, if the server receives a shutdown signal, all child Scopes will be cancelled because the main Scope will be cancelled as well.

Note that the coroutine waiting on await Async\signal(SIGINT) will not remain hanging in memory if the server shuts down in another way, because $scope will be explicitly closed in the finally block.

The cancel method cancels all child coroutines and all child Scopes of the current Scope.:

use function Async\Scope\delay;
 
$scope = new Scope();
 
spawn with $scope {
    spawn {
        delay(1000);
        echo "Task 1\n";
    };
 
    spawn {
        delay(2000);
        echo "Task 2\n";
    };
};
 
$scope->cancel();

Expected output:

Coroutine Scope has several resource cleanup strategies that can be triggered either explicitly, on demand, or implicitly when the Scope object loses its last reference.

There are three available strategies for Scope termination:

The main goal of all three methods is to terminate the execution of coroutines that belong to the Scope or its child Scopes. However, each method approaches this task slightly differently.

The disposeSafely method is used by default in the destructor of the Async\Scope class. Its key feature is transitioning coroutines into a zombie coroutine state. A zombie coroutine continues execution but is tracked by the system differently than regular coroutines. (See section: [Zombie coroutine policy](#zombie-coroutine-policy)).

A warning is issued when a zombie coroutine is detected.

use function Async\Scope\delay;
 
$scope = new Scope();
 
await spawn in $scope {
    spawn {
        delay(1000);
        echo "Task 1\n";
    };
 
    spawn {
        delay(2000);
        echo "Task 2\n";
    };
 
    echo "Root task\n";
};
 
$scope->disposeSafely();

Expected output:

Root task
Warning: Coroutine is zombie at ... in Scope disposed at ...
Warning: Coroutine is zombie at ... in Scope disposed at ...
Task 1
Task 2

The $scope variable is released immediately after the coroutine Root task completes execution, so the child coroutine Task 1 does not have time to execute before the disposeSafely method is called.

disposeSafely detects this and signals it with a warning but allows the coroutine to complete.

The Scope::dispose method differs from Scope::disposeSafely in that it does not leave zombie coroutines. It cancels all coroutines. When coroutines are detected as unfinished, a warning is issued.

Example:

use function Async\Scope\delay;
 
$scope = new Scope();
 
await spawn in $scope {
    spawn {
        delay(1000);
        echo "Task 1\n";
    };
 
    spawn {
        delay(2000);
        echo "Task 2\n";
    };
 
    echo "Root task\n";
};
 
$scope->dispose();

Expected output:

Warning: Coroutine is zombie at ... in Scope disposed at ...
Warning: Coroutine is zombie at ... in Scope disposed at ...
Warning: Coroutine is zombie at ... in Scope disposed at ...

The disposeAfterTimeout method is a delayed version of the disposeSafely method. The $timeout parameter must be greater than zero but less than 10 minutes.

use Async\Scope;
 
class Service
{
    private Scope $scope;
 
    public function __construct() {
        $this->scope = new Scope();
    }
 
    public function __destruct() {
        $this->scope->disposeAfterTimeout(5000);
    }
 
    public function run(): void {
        spawn with $this->scope {
 
            spawn {
                delay(1000);
                echo "Task 2\n";
                delay(5000);
                echo "Task 2 next line never executed\n";
            };
 
            echo "Task 1\n";
        };
    }
}
 
$service = new Service();
$service->run();
 
sleep(1);
unset($service);

Expected output:

Task 1
Warning: Coroutine is zombie at ... in Scope disposed at ...
Task 2

The dispose*() methods can be called multiple times, which is not considered an error.

If the Scope::cancel() method is called with a parameter after the Scope has already been cancelled, PHP will emit a warning indicating that the call will be ignored.

If a Scope has child Scopes, the coroutines in the child Scopes will be canceled or disposed first, followed by those in the parent — from the bottom up in the hierarchy. This approach increases the likelihood that resources will be released correctly. However, it does not guarantee this, since the exact order of coroutines in the execution queue cannot be determined with 100% certainty.

During the release of child Scopes, the same cleanup strategy is used that was applied to the parent Scope.

If the disposeSafely method is called, the child Scopes will also be released using the disposeSafely strategy. If the dispose method is used, the child Scopes will use the same method for cleanup.

The disposeAfterTimeout method will delay the execution of disposeSafely for the specified time.

When the cancel() or dispose() method is called, the Scope is marked as closed. Attempting to launch a coroutine with this Scope will result in a fatal exception.

$scope = new Scope();
 
spawn with $scope {
    echo "Task 1\n";
};
 
$scope->cancel();
 
spawn with $scope { // <- Fatal error
    echo "Task 2\n";
};

Detecting erroneous situations when using coroutines is an important part of analyzing an application's reliability.

The following scenarios are considered potentially erroneous:

1. A coroutine belongs to a global scope and is not awaited by anyone (a **zombie coroutine**).
2. The root scope has been destroyed (its destructor was called), but no one awaited
 it or ensured that its resources were explicitly cleaned up (e.g., by calling ''$scope->cancel()'' or ''$scope->dispose()'').
3. Tasks were not cancelled using the ''cancel()'' method, but through a call to ''dispose()''.
 This indicates that the programmer did not intend to cancel the execution of the coroutine,
 yet it happened because the scope was destroyed.
4. An attempt to await a coroutine from within itself.
5. Awaiting ''$scope'' from within itself or from one of its child scopes.
6. Stuck tasks in the cancellation state.
7. Using ''TaskGroup'' with a result capturing without an ''await'' expression.
8. Deadlocks caused by circular dependencies between coroutines.

PHP will respond to such situations by issuing warnings, including debug information about the involved coroutines. Developers are expected to write code in a way that avoids triggering these warnings.

An attempt to use the expression await $coroutine from within the same coroutine throws an exception.

$coroutine = null;
$coroutine = spawn use(&$coroutine) {
    await $coroutine; // <- Fatal error: A coroutine cannot await itself. Coroutine spawned at ...
};

Using the Scope::awaitCompletion() from a coroutine that belongs to the same $scope or to one of its child scopes will throw a fatal exception. This condition makes it impossible to perform the $globalScope->awaitCompletion call.

$scope = new Scope();
 
spawn with $scope use($scope) {
    $scope->awaitCompletion(Async\timeout(1000)); // <- Fatal error: Awaiting a scope from within itself or
                                                  // its child scope would cause a deadlock. Scope created at ...
};

The only way to create zombie coroutines is by using the spawn expression in the globalScope. However, if the initial code explicitly creates a scope and treats it as the application's entry point, the initializing code gains full control — because spawn <callable> will no longer be able to create a coroutine in globalScope, thus preventing the application from hanging beyond the entry point.

There’s still a way to use global variables and new Scope to launch a coroutine that runs unchecked:

$GLOBALS['my'] = new Scope();
spawn with $GLOBALS['my'] { ... };

But such code can't be considered an accidental mistake.

To avoid accidentally hanging coroutines whose lifetimes were not correctly limited, follow these rules:

• Use separate Scopes for different coroutines. This is the best practice,

as it allows explicitly defining lifetime dependencies between Scopes.

• Use Scope::dispose(). The dispose() method cancels coroutine execution and logs an error.
• Don’t mix semantically different coroutines within the same Scope.
• Avoid building hierarchies between Scopes with complex interdependencies.
• Do not use cyclic dependencies between Scopes.
• The principle of single point of responsibility and Scope ownership.

Do not pass the Scope object to different coroutine functions (unless the action happens in a closure).

Do not store ''Scope'' objects in different places.
Violating this rule can lead to manipulations with ''Scope'',
which may cause a deadlock or disrupt the application's logic.
* Child coroutines should not wait for their parents.
Child Scopes should not wait for their parents.

namespace ProcessPool;
 
use Async\Scope;
use Async\TaskGroup;
 
final class ProcessPool
{
    private Scope $watcherScope;
    private Scope $jobsScope;
    private TaskGroup $pool;
 
    /**
     * List of pipes for each process.
     * @var array
     */
    private array $pipes = [];
    /**
     * Map of process descriptors: pid => bool
     * If the value is true, the process is free.
     * @var array
     */
    private array $descriptors = [];
 
    public function __construct(readonly public string $entryPoint, readonly public int $max, readonly public int $min)
    {
        // Define the coroutine scopes for the pool, watcher, and jobs
        $this->watcherScope = new Scope();
        $this->jobsScope = new Scope();
        $this->pool = new TaskGroup(captureResults: false);
    }
 
    public function __destruct()
    {
        $this->watcherScope->dispose();
        $this->pool->dispose();
        $this->jobsScope->dispose();
    }
 
    public function start(): void
    {
        spawn with $this->watcherScope $this->processWatcher();
 
        for ($i = 0; $i < $this->min; $i++) {
            $taskGroup->add(spawn with $this->poolScope $this->startProcess());
        }
    }
 
    public function stop(): void
    {
        $this->watcherScope->cancel();
        $this->pool->cancel();
        $this->jobsScope->cancel();
    }
 
    private function processWatcher(): void
    {
        while (true) {
            try {
                await $this->pool;
            } catch (StopProcessException $exception)  {
                echo "Process was stopped with message: {$exception->getMessage()}\n";
 
                if($exception->getCode() !== 0 || count($this->descriptors) < $this->min) {
                    spawn with $this->pool $this->startProcess();
                }
            }
        }
    }
}

The example above demonstrates how splitting coroutines into Scopes helps manage their interaction and reduces the likelihood of errors.

Here, watcherScope monitors tasks in poolScope. When a process finishes, the watcher detects this event and, if necessary, starts a new process or not. The monitoring logic is completely separated from the process startup logic.

The lifetime of watcherScope matches that of poolScope, but not longer than the lifetime of the watcher itself.

The overall lifetime of all coroutines in the ProcessPool is determined by the lifetime of the ProcessPool object or by the moment the stop() method is explicitly called.

Coroutines whose lifetime extends beyond the boundaries of their parent Scope are handled according to a separate policy.

This policy aims to strike a balance between uncontrolled resource leaks and the need to abruptly terminate coroutines, which could lead to data integrity violations.

If there are no active coroutines left in the execution queue and no events to wait for, the application is considered complete.

Zombie coroutines differ from regular ones in that they are not counted as active. Once the application is considered finished, zombie coroutines are given a time limit within which they must complete execution. If this limit is exceeded, all zombie coroutines are canceled.

The delay time for handling zombie coroutines can be configured using a constant in the php.ini file: async.zombie_coroutine_timeout, which is set to two seconds by default.

If a coroutine is created within a user-defined Scope, the programmer can set a custom timeout for that specific Scope using the Scope::disposeAfterTimeout(int $ms) method.

Async\TaskGroup is a container for controlling a group of coroutines.

function mergeFiles(string ...$files): string
{
    $taskGroup = new Async\TaskGroup(captureResults: true);
 
    foreach ($files as $file) {
       spawn with $taskGroup file_get_contents($file);
    }
 
    return array_merge("\n", await $taskGroup);
}
 
echo await spawn mergeFiles(['file1.txt', 'file2.txt', 'file3.txt']);

TaskGroup is an explicit and safe way to manage a group of tasks.

Using the Scope class and the spawn expression, you can create groups of coroutines. However, the code that creates $scope and/or awaits it might not be aware of which coroutines will be added. Moreover, the wait strategy of $scope can lead to resource leaks if a programmer mistakenly uses the spawn <callable> expression and adds a coroutine to the Scope that lives indefinitely.

The TaskGroup class is an explicit pattern for managing a group of coroutines. Unlike Scope, tasks cannot be added to it accidentally.

In a TaskGroup, a task can only be added explicitly, using the spawn with expression. A TaskGroup is not propagated through the execution context by child coroutines.

And unlike Scope, TaskGroup can capture the results of tasks, which makes it convenient for awaiting results.

If Scope is used to create a shared space for coroutines, then TaskGroup is intended for explicit control over child tasks. In this role, TaskGroup serves as a complement to the logic of Scope.

The TaskGroup constructor accepts several parameters:

1. $scope – the Scope in which the tasks will be executed.

  If this parameter is not provided, a new, separate ''Scope'' will be created.

2. $captureResults – an option to capture the results of the tasks. 3. $bounded – an option to Scope::dispose all Scope tasks when the TaskGroup is disposed.

Once a $taskGroup is created, it can be used in a spawn with $taskGroup expression, which has an additional effect for task groups: a coroutine is created within the TaskGroup's $scope, and the coroutine is added to the task group.

A TaskGroup holds a reference to the Scope in which the tasks will be executed. If this is the only reference to the Scope, the TaskGroup will automatically call Scope::dispose() as soon as the TaskGroup::dispose or TaskGroup::cancel method is invoked.

The expression spawn with $taskGroup creates a coroutine in the $scope specified in the TaskGroup, and additionally adds the task to the group. If child coroutines create other coroutines using the spawn expression, they will be added to the TaskGroup's scope, but not to the task group itself.

use Async\Scope;
use Async\TaskGroup;
 
function task1() {
    spawn subtask();
}
 
$scope = new Scope();
$taskGroup1 = new TaskGroup($scope);
$taskGroup2 = new TaskGroup($scope);
 
spawn with $taskGroup1 task1();
spawn with $taskGroup2 task2();
 
$scope->awaitCompletion(Async\signal(SIGTERM));

Structure:

main()                             ← defines a $scope
└── $scope = new Scope()
    ├── task1()                    ← runs in the $scope
    ├── subtask()                  ← runs in the $scope
    ├── task2()                    ← runs in the $scope
└── $taskGroup1 = new TaskGroup($scope)
    ├── task1()                    ← runs in the $scope
└── $taskGroup2 = new TaskGroup($scope)
    ├── task2()                    ← runs in the $scope

The tasks task1() and task2() belong to different groups but are in the same Scope. The coroutine subtask(), which was launched from task1(), does not belong to any group.

If $scope is disposed, all task groups will be cancelled. However, cancelling a task group will not cancel tasks in the Scope if the reference count to $scope is greater than one.

The TaskGroup class implements the Awaitable interface, so it can be used with the await expression. The await $taskGroup expression captures both the results of execution and any exceptions that occur in the tasks.

If the constructor option captureResults: true is specified, then the await $taskGroup expression will return the results of all tasks that were added to the group. If the results are no longer needed, the TaskGroup::disposeResults() method should be used to discard them.

function processInBatches(array $files, int $limit): array
{
    $allResults = [];
    $taskGroup  = new Async\TaskGroup(captureResults: true);
    $count      = 0;
 
    foreach ($files as $file) {
        spawn with $taskGroup file_get_contents($file);
 
        if (++$count >= $limit) {
            $allResults = [...$allResults, ...await $taskGroup];
            $taskGroup->disposeResults();
            $count = 0;
        }
    }
 
    $allResults = [...$allResults, ...await $taskGroup];
    return $allResults;
}
 
$results = await spawn processInBatches(['file1.txt', 'file2.txt', 'file3.txt', 'file4.txt'], limit: 2);
echo implode("\n", $results);

The $taskGroup object can be used in an await expression multiple times. If the captureResults mode is not enabled, the await expression will always return NULL.

Be careful when capturing coroutine results, as this may cause memory leaks or keep large amounts of data in memory.
Plan the waiting process wisely, and use the TaskGroup::disposeResults() method.

When a TaskGroup is disposed, all tasks belonging to it will be cancelled using cancel, without issuing any warnings. No tasks will turn into zombie coroutines.

This behavior is consistent with Scope::cancel().

The reason for this behavior lies in the fact that TaskGroup only keeps track of explicitly added tasks. If a task group is being disposed, it means the user clearly understands that all coroutines launched within it should also be terminated.

TaskGroup is designed to complement the behavior of Scope where needed. Although a single Scope can have multiple TaskGroups, in most cases it is reasonable to create a TaskGroup along with a unique Scope that belongs only to it. This leads to clear and memorable behavior: the lifetime of a TaskGroup equals the lifetime of its Scope.

If this rule is followed, then an exception in a coroutine will lead to the cancellation of the Scope, which is required before the cancellation operation, and will trigger TaskGroup::dispose. On the other hand, releasing a TaskGroup object automatically leads to the disposal of its Scope. This architecture helps reduce the likelihood of resource leakage errors.

The TaskGroup class allows for implementing a pattern in which tasks are divided into two groups: * explicit (or target) tasks, whose results are needed * implicit (or secondary) tasks, which are created within explicit tasks and whose results are not needed.

Explicit tasks belong directly to the TaskGroup and are created using the spawn with $taskGroup expression. All other tasks are considered as secondary.

This separation helps produce code that manages resources more efficiently than code that waits for the completion of all child coroutines within a Scope.

The following code demonstrates this idea:

use Async\Scope;
use Async\TaskGroup;
 
function targetTask(int $i): void
{
    spawn {
        // subtask should be added to the same scope
    };
}
 
$taskGroup = new TaskGroup(scope: new Scope(), captureResults: true);
 
for($i = 0; $i < 10; $i++) {
    spawn with $taskGroup targetTask($i);
}
 
// wait for only the tasks that were added to the TaskGroup
$results = await $taskGroup;

The expression await $taskGroup will wait only for the completion of the target tasks that were explicitly added to the TaskGroup.

The result of await $taskGroup will include the outcomes of all coroutines from the TaskGroup, but not of other implicit coroutines that were created during the execution of targetTask().

Once $taskGroup is destroyed, the Scope it references will also be disposed of, which means all other implicit coroutines will be cancelled using one of the three strategies (see the corresponding section).

The reverse is also true: if the Scope is disposed, the associated TaskGroups will be disposed as well.

$scope = new Async\Scope();
$taskGroup = new Async\TaskGroup(scope: $scope, captureResults: false);
spawn with $taskGroup {
    // this task will be added to the task group
    Async\delay(1000);
    echo "This line will be executed\n";
};
 
sleep(1);
$scope->dispose();

Expected output:

There are no warnings about zombie coroutines in the output because the task was canceled using $taskGroup->dispose().

However, if the Scope contains other coroutines that were created outside the TaskGroup, they will follow the general rules. In the case of the Scope::disposeSafely() strategy, a warning will be issued if unfinished tasks are detected, as they would become zombie coroutines.

The TaskGroup class allows you to wait for the first task to complete using the race() method.

use Async\TaskGroup;
 
function fetchFirstSuccessful(string ...$apiHosts): string
{
    $taskGroup = new Async\TaskGroup(captureResults: false);
 
    foreach ($apiHosts as $host) {
        spawn with $taskGroup use ($host) {
            $response = file_get_contents($host);
 
            if($response === false) {
                throw new Exception("Failed to fetch data from $host");
            }
 
            return $response;
        });
    }
 
    return await $taskGroup->race(ignoreErrors: true);
}

The TaskGroup::race() method returns an Awaitable trigger that can be used multiple times to obtain the first completed task. The race() trigger clears the internal result storage after completion, so you won't be able to retrieve the same result twice.

If you need to get the first available result, use the firstResult() method. The TaskGroup::firstResult() trigger returns the first available result. Even if it is called repeatedly, the result will remain the same until the TaskGroup::disposeResults() method cancels the previous values.

The ignoreErrors parameter specifies the error ignoring mode. If it is set to true, exceptions from tasks will be ignored, and the race()/firstResult() triggers will return the first successful task.

The TaskGroup::getErrors() method will return an array of exceptions.

You can combine TaskGroup with Scope::inherit() to create a task group within a child Scope, thereby forming a hierarchy between groups:

use Async\TaskGroup;
use Async\Scope;
 
$taskGroupParent = new TaskGroup(captureResults: false);
 
spawn with $taskGroupParent {
    $taskGroupChild = new TaskGroup(Scope::inherit(), captureResults: false);
 
    spawn with $taskGroupChild {
        // this task will be added to the child task group
    };
 
    // wait for the child task group to finish
    await $taskGroupChild;
};
 
await $taskGroupParent;

Structure:

main()
└── $taskGroupParent = new TaskGroup()                          <- parent task group scope
    ├── $taskGroupChild = new TaskGroup(Scope::inherit())       <- child task group scope

Since each TaskGroup is associated with its own Scope, and Scope instances are connected through parent-child relationships, cancelling a parent TaskGroup will automatically cancel the entire hierarchy.

The TaskGroup class allows you to cancel all tasks in the group using the TaskGroup::cancel() method.

This method behaves the same way as TaskGroup::dispose, with the only difference being that it allows you to pass a specific exception.

use Async\TaskGroup;
 
$taskGroup = new Async\TaskGroup(captureResults: false);
spawn with $taskGroup {
    try {
        suspend;
    } catch (Throwable $throwable) {
        echo "Task was cancelled: ", $throwable->getMessage(), "\n";
    }
};
 
// pass control to the task
suspend;
 
$taskGroup->cancel(new \Async\CancellationException('Custom cancellation message'));

Expected output:

Task was cancelled: Custom cancellation message

TaskGroup does not introduce additional logic for handling coroutine exceptions. When a developer uses the expression await $taskGroup, they are capturing the exceptions of all tasks contained within $taskGroup.

In other words, await $taskGroup is equivalent to simultaneously using await $coroutine for each task. If no one awaits $taskGroup, the exception handling follows the general Flow, and the error will propagate to the Scope.

An additional method TaskGroup::all(bool $ignoreErrors = false, $nullOnFail = false): Awaitable {} provides a trigger that fires when all tasks in the group have completed.

At the same time, it captures any errors, which can be retrieved using TaskGroup::getErrors().

// Returns an array of all tasks with their results ignoring errors
return $taskGroup->all(ignoreErrors: true);

The trigger TaskGroup::all() returns an array of results with numeric indices, where each index corresponds to the ordinal number of the task.

If a task completed with an error, its numeric index is missing from the array.

Using the option $nullOnFail, you can specify that the results of failed tasks should be filled with NULL instead.

$taskGroup = new Async\TaskGroup(captureResults: true);
spawn with $taskGroup {return 'result 1';}
spawn with $taskGroup {throw new Exception('Error');}
 
var_dump(await $taskGroup->all(ignoreErrors: true, nullOnFail: true));

Expected output:

array(2) {
  [0]=>
  string(8) "result 1"
  [1]=>
  NULL
}

The method TaskGroup::getErrors() returns an array with numeric indices and exceptions, where each index corresponds to the ordinal number of the task.

Note: The method TaskGroup::disposeResults clears all results and errors at the moment it is called.
Coroutines then reset their ordinal indices starting from zero.

What happens when a coroutine that belongs to a Scope but does not belong to a TaskGroup throws an exception?

1. If the exception is not handled, it will propagate to the Scope. 2. If a Scope has no exception handler, it invokes the dispose() strategy, which cancels all coroutines, including any TaskGroup associated with the Scope. In this case, the await point of the TaskGroup will receive a CancellationException.

Example:

use Async\TaskGroup;
 
$taskGroup = new Async\TaskGroup(captureResults: false);
 
spawn with $taskGroup {
    spawn { // <- subcoroutine in the same scope
        throw new Exception('Error in coroutine');
    };
 
    sleep(1);
};
 
try {
    await $taskGroup;
} catch (Async\CancellationException $exception) {
    echo "Caught exception: ", $exception->getMessage(), "\n";
}

Expected output:

Caught exception: TaskGroup was cancelled at ...

If you need to handle this type of exception, use the Scope::setExceptionHandler method before calling await $taskGroup:

use Async\TaskGroup;
 
$scope = new Async\Scope();
$taskGroup = new Async\TaskGroup(scope: $scope, captureResults: false);
$scope->setExceptionHandler(function (Async\Scope $scope, Async\Coroutine $coroutine, Throwable $e) {
    echo "Caught exception: {$e->getMessage()}\n in coroutine: {$coroutine->getSpawnLocation()}\n";
});
 
spawn with $taskGroup {
    spawn { // <- subcoroutine in the same scope
        throw new Exception('Error in coroutine');
    };
 
    sleep(1);
};
 
await $taskGroup;

Please see [Error Handling](#error-handling) for more details.

Structured concurrency allows organizing coroutines into a group or hierarchy to manage their lifetime or exception handling.

The parent task is required to take responsibility for its child tasks and must not complete before the children have finished their execution.

To implement structured concurrency, it is recommended to use the TaskGroup class.

The following code implements this idea:

use Async\Scope;
 
function copyFile(string $sourceFile, string $targetFile): void
{
    $source = fopen($sourceFile, 'r');
    $target = fopen($targetFile, 'w');
 
    $buffer = null;
 
    try {
        // Child scope
        $tasks = new \Async\TaskGroup(Scope::inherit());
 
        // Read data from the source file
        spawn with $tasks use(&$buffer, $source) {
            while (!feof($source)) {
                if ($buffer === null) {
                    $chunk = fread($source, 1024);
                    $buffer = $chunk !== false && $chunk !== '' ? $chunk : null;
                }
 
                suspend;
            }
 
            $buffer = '';
        };
 
        // Write data to the target file
        spawn with $tasks use(&$buffer, $target) {
            while (true) {
                if (is_string($buffer)) {
                    if ($buffer === '') {
                        break; // End of a file
                    }
 
                    fwrite($target, $buffer);
                    $buffer = null;
                }
 
                suspend;
            }
 
            echo "Copy complete.\n";
        };
 
        await $tasks;
    } finally {
        fclose($source);
        fclose($target);
    }
}
 
$copyTasks = new \Async\TaskGroup;
 
spawn with $copyTasks copyFile('source.txt', 'target.txt');
spawn with $copyTasks copyFile('source2.txt', 'target2.txt');
 
await $copyTasks;

The example creates two task groups: a parent and a child. The parent task group handles the copy operations directly, while the child tasks perform file reading and writing. File descriptors will not be closed until all child copy tasks have completed. The main code will not finish until all copy operations are completed.

Libraries and frameworks often use variables that are shared within a request to store common data. These variables are not Global in the general sense, but they essentially reflect a shared state related to the request or execution scope.

For example, the TokenStorage class (https://github.com/symfony/symfony/blob/7.3/src/Symfony/Component/Security/Core/Authentication/Token/Storage/TokenStorage.php) from Symfony allows retrieving the user token multiple times, as it is stored in a variable.

Or /src/Illuminate/Auth/TokenGuard.php from Laravel:

    /**
     * Get the currently authenticated user.
     */
    public function user()
    {
        // If we've already retrieved the user for the current request we can just
        // return it back immediately. We do not want to fetch the user data on
        // every call to this method because that would be tremendously slow.
        if (! is_null($this->user)) {
            return $this->user; // <-- Shared state
        }
 
        $user = null;
 
        $token = $this->getTokenForRequest();
 
        // some code skipped
 
        return $this->user = $user;
    }

This code assumes that a single process/thread always handles only one request at a time. However, in a concurrent web server environment, shared states can no longer be used because the execution context may switch unexpectedly.

You can use Coroutine ID and Map to associate a unique coroutine ID with specific data. However, in this case, you must ensure that the data is properly released when the coroutine ceases to exist.

In addition to storing request-specific data, concurrent code must also ensure the proper handling of input/output descriptors. For example, when implementing a protocol, data must be sent in a specific sequence. If a socket is used by two coroutines simultaneously for reading/writing, the order of operations may be disrupted.

Another example is database transactions. Code that starts a transaction cannot release the database connection socket until the transaction is completed.

The Async\Context class is designed to help solve these issues.

The Async\Context class defines three groups of methods: * Methods for retrieving values from the Map, considering parent contexts * Methods for retrieving values only from the current context * Methods for modifying or removing keys in the current context

Context Slots are an efficient mechanism for managing memory associated with Scope or coroutine lifetimes. Once all coroutines owning the Scope complete, or the Scope itself is terminated, all data in the slots will be released.

This helps the programmer associate data with coroutines without writing explicit cleanup code.

To ensure data encapsulation between different components, Coroutine Scope Slots provide the ability to associate data using key objects. An object instance is unique across the entire application, so code that does not have access to the object cannot read the data associated with it.

This pattern is used in many programming languages and is represented in JavaScript by a special class, Symbol.

$key = 'pdo connection';
 
if(currentContext()->has($key)) {
    $pdo = currentContext()->get($key);
} else {
    $pdo = new PDO('sqlite::memory:');
    currentContext()->set($key, new PDO('sqlite::memory:'));
}

Coroutine Scope Slots can automatically dereference WeakReference. If you assign a WeakReference to a slot and then call find(), you will receive the original object or NULL.

function task(): void
{
    // Should return the original object
    $pdo = currentContext()->find('pdo');
}
 
$pdo = new PDO('sqlite::memory:');
currentContext()->set('pdo', new WeakReference($pdo));
 
spawn task();

The context belongs to the Scope and is created along with it. If a Scope is inherited from a parent, the new context also inherits the parent. Thus, the hierarchy of Scope objects forms exactly the same hierarchy of contexts.

use Async\Scope;
use function \Async\currentContext;
use function \Async\rootContext;
 
function handleRequest($socket): void
{
    echo currentContext()->get('request_id')."\n"; // <-- From request context
    echo currentContext()->get('server_id')."\n"; // <-- From server context
    echo rootContext()->get('request_id')."\n"; // <-- Should be NULL
}
 
function startRequestHandler($socket): void
{
    $requestScope = Scope::inherit(); // <-- Inherit server context
    $requestScope->context->set('request_id', uniqid()); // <-- Override server context slot
 
    // Handle request in separate coroutine and scope
    spawn with $requestScope handleRequest($socket);
}
 
function startServer(): void
{
    $serverScope = new Scope();
    $serverScope->context->set('server_id', uniqid());
    $serverScope->context->set('request_id', null);
 
    while (true) {
        $socket = stream_socket_accept($serverSocket, 0);
        startRequestHandler($socket);
    }
}

The special functions Async\currentContext() and Async\rootContext() help quickly access the current context from any function.

Async\rootContext() returns the context at the very root of the hierarchy, if it exists, or the global application context if it does not.

While a Scope can serve as a shared context in the coroutine hierarchy, a coroutine's local context is a personal data store strictly tied to the coroutine's lifetime. The local context allows associating data slots that are automatically freed once the coroutine completes.

The local coroutine context is accessible via the Async\coroutineContext() function, which returns an Async\Context object. The Async\Context class provides the same methods for working with slots as the Scope class:

function task(): void
{
    coroutineContext()->set('data', 'This local data');
 
    spawn {
         // No data will be found
         echo coroutineContext()->find('data')."\n";
    };
}

Using a coroutine's local context can be useful for associating objects with a coroutine that MUST be unique to each coroutine.

For example, a database connection:

<?php
 
namespace Async;
 
use PDO;
use RuntimeException;
 
class ConnectionProxy
{
    private PDO $connection;
 
    public function __construct(PDO $connection)
    {
        $this->connection = $connection;
    }
 
    public function __destruct()
    {
        ConnectionPool::default()->releaseConnection($this->connection);
    }
}
 
final class ConnectionPool
{
    static private $pool = null;
 
    public static function default(): ConnectionPool
    {
        if (self::$pool === null) {
            self::$pool = new ConnectionPool();
        }
 
        return self::$pool;
   }
 
    private array $pool = [];
    private int $maxConnections = 10;
 
    public function getConnection(): ConnectionProxy
    {
        if (!empty($this->pool)) {
            return new ConnectionProxy(array_pop($this->pool));
        }
 
        if (count($this->pool) < $this->maxConnections) {
            return new ConnectionProxy(PDO("mysql:host=localhost;dbname=test", "user", "password"));
        }
 
        throw new RuntimeException("No available database connections.");
    }
 
    public function releaseConnection(PDO $connection): void
    {
        $this->pool[] = $connection;
    }
}
 
function getDb(): ConnectionProxy
{
    static $key = new Key('db_connection');
 
    $context = Async\coroutineContext();
 
    if ($context->has($key)) {
        return $context->get($key);
    }
 
    $connection = ConnectionPool::default()->getConnection();
 
    $context->set($key, $connection);
 
    return $connection;
}
 
function printUser(int $id): void
{
    $db = getDb();
    $stmt = $db->query("SELECT * FROM users WHERE id = $id");
    $users = $stmt->fetchAll(PDO::FETCH_ASSOC);
    print_r($users);
}
 
spawn printUser(1);
spawn printUser(2);

This code relies on the fact that an instance of the ConnectionProxy class will be destroyed as soon as the coroutine completes. The destructor will be called, and the connection will automatically return to the pool.

The following functions allow combining Awaitable objects or capturing errors from Awaitable objects:

The combinators any, all, and anyOf propagate exceptions from their composite triggers. If you need to ignore errors, you should use the captureErrors or ignoreErrors combinator.

The Async\all method modifies the execution result and returns an array of values. The order of the values matches the order of $triggers.

use function Async\all;
 
$results = await all([
    spawn fetchUserData(),
    spawn fetchUserSettings()
]);
 
print_r($results);

Expected output:

text
Array
(
    [0] => ... // result of fetchUserData()
    [1] => ... // result of fetchUserSettings()
)

Any returns a trigger that fires as soon as at least one item from the list has completed:

use function Async\any;
 
// Returns when at least one API call
$results = await any([
    spawn fetchDataFromAPI1(),
    spawn fetchDataFromAPI2(),
    spawn fetchDataFromAPI3(),
]);

The any trigger can be invoked multiple times, and each time it will respond to the completion of the next item from the list. This allows you to organize a loop of calls.

The anyOf combinator is used to wait for at least N items from the list:

// Returns when at least 2 images are loaded
$results = await Async\anyOf(2, [
    spawn loadImage('preview.jpg'),
    spawn loadImage('medium.jpg'),
    spawn loadImage('full.jpg'),
]);

The captureErrors combinator ignores trigger errors and waits for the first successful execution. Use it together with any, all, or anyOf to wait for events while ignoring errors.

The captureErrors combinator modifies the trigger results. It returns an array with two elements: * the first element is the operation result * the second element is an array of errors

use function Async\captureErrors;
 
[$result, $errors] = await captureErrors(Async\all([spawn fetchUserData(), spawn fetchUserSettings()]));
 
if(empty($errors)) {
    // $result contains successful result
} else {
    // $result contains null
    // $errors contains an exception or empty array
}

The ignoreErrors combinator also allows you to ignore errors, but instead of returning errors, as a result, it sends them to a special handler function.

use function Async\any;
use function Async\ignoreErrors;
 
// Returns when at least one API call ignores errors
$results = await ignoreErrors(any([
    spawn fetchDataFromAPI1(),
    spawn fetchDataFromAPI2(),
    spawn fetchDataFromAPI3(),
]), fn(Throwable $throwable) => null);

Since Awaitable objects act as triggers rather than futures, their state can change multiple times. This can be useful in complex scenarios.

For example:

function getFirstAvailable(array $sources, int $errorTolerance = 0): mixed
{
    if($errorTolerance <= 0) {
        $errorTolerance = count($sources) / 2;
    }
 
    $errors     = 0;
    $trigger    = Async\any($sources);
 
    while($errors < $errorTolerance) {
        try {
            return await $trigger;
        } catch (Exception $e) {
            $errors++;
        }
    }
 
    throw new Exception("sources failed: {$errors} errors");
}

The function will return the first successful value with error tolerance, which by default is set to 50% of the total number of $sources.

The combinators any, all, and anyOf can accept an iterator as a source of triggers. In this case, the iterator will be executed in a concurrent environment. The iterator can be asynchronous. In this case, all will wait not only for all triggers but also for the iterator to finish.

The captureErrors and ignoreErrors combinators do not affect errors that occur inside the iterator providing the triggers.

If an exception is thrown inside the iterator, it will be passed through captureErrors or ignoreErrors further up the call stack.

use function Async\all;
use function Async\captureErrors;
 
function processAllUserTasks(int ...$userIds): iterable
{
    $taskGroup = new Async\TaskGroup(captureResults: true);
 
    foreach ($userIds as $userId) {
        spawn with $taskGroup getUser($userId);
    }
 
    while ($taskGroup->isFinished() === false) {
 
        $user = await $taskGroup->race();
 
        foreach ($user['tasks'] ?? [] as $task) {
            yield spawn processUserTask($task);
        }
    }
}
 
$results = await all(processAllUserTasks());

The standard async library includes two functions similar to usleep(): * Async\delay(int $ms): void * Async\timeout(int $ms): Awaitable

The delay function suspends the execution of a coroutine for the specified number of milliseconds. Unlike usleep, the delay function will throw a cancellation exception if the coroutine is cancelled.

The timeout function is similar to delay, but it returns an Awaitable object:

use function Async\timeout;
use function Async\delay;
 
try {
    delay(1000); // suspends the coroutine for 1 second
    // Try to fetch data from the URL within 1 second
    echo await spawn file_get_content('https://php.net/') until timeout(1000);
} catch (\Async\AwaitCancelledException) {
    echo "Operation was cancelled by timeout\n";
} catch (\Async\CancellationException) {
    echo "Operation was cancelled by user\n";
}

The supervisor allows organizing a group of tasks in which an error in one task stops all its child tasks but does not affect sibling tasks.

To create a Supervisor, you can use the following approach:

class Supervisor implements Async\SpawnStrategy
{
    private $supervisor;
 
    public function __construct() {
        $this->supervisor = new Async\Scope();
    }
 
    public function provideScope(): ?Scope
    {
        return Async\Scope::inherit($this->supervisor);
    }
 
    public function beforeCoroutineEnqueue(Coroutine $coroutine, Scope $scope): array
    {
        $coroutine->onFinally(fn() => $scope->dispose());
        return [];
    }
}
 
$supervisor = new Supervisor();
 
spawn with $supervisor {
 
    spawn {
        echo "This code is not been executed\n";
    }
 
    throw new Exception("Task 1");
};
 
spawn with $supervisor {
    // This task will not be canceled
    echo "Task 2\n";
};

To achieve the desired behavior, the Supervisor creates each task in a separate child Scope and links the lifetime of the Scope to the lifetime of the task. If an exception occurs in a task or in a child coroutine, only the task where the exception happened will be cancelled, not all tasks within the Supervisor.

The Supervisor class may be included in the standard library in future RFCs.

An uncaught exception in a coroutine follows this flow:

1. If the coroutine is awaited using the await keyword,

 the exception is propagated to the awaiting points.
 If multiple points are awaiting, each will receive the same exception
 (**Each await point will receive the exact same exception object, not cloned**).

2. The exception is passed to the Scope. 3. If the Scope has an exception handler defined, it will be invoked. 4. If the Scope does not have an exception handler, the cancel() method is called,

 canceling all coroutines in this scope, including all child scopes.

5. If the Scope has responsibility points, i.e., the construction Scope::awaitCompletion,

 all responsibility points receive the exception.

6. Otherwise, the exception is passed to the parent scope if it is defined. 7. If there is no parent scope, the exception falls into globalScope,

 where the same rules apply as for a regular scope.

Example:

use Async\Scope;
 
$scope = new Scope();
 
spawn with $scope {
    throw new Exception("Task 1");
};
 
$exception1 = null;
$exception2 = null;
 
$scope2 = new Scope();
 
spawn with $scope2 use($scope, &$exception1) {
    try {
        $scope->awaitCompletion(Async\signal(SIGTERM));
    } catch (Exception $e) {
        $exception1 = $e;
        echo "Caught exception1: {$e->getMessage()}\n";
    }
};
 
spawn with $scope2 use($scope, &$exception2) {
    try {
        $scope->awaitCompletion(Async\signal(SIGTERM));
    } catch (Exception $e) {
        $exception2 = $e;
        echo "Caught exception2: {$e->getMessage()}\n";
    }
};
 
$scope2->awaitCompletion(Async\signal(SIGTERM));
 
echo $exception1 === $exception2 ? "The same exception\n" : "Different exceptions\n";

If an exception reaches globalScope and is not handled in any way, it triggers Graceful Shutdown Mode, which will terminate the entire application.

The Scope class allows defining an exception handler that can prevent exception propagation.

For this purpose, two methods are used:

1. setExceptionHandler – triggers for any exceptions thrown within this Scope.
2. setChildScopeExceptionHandler – triggers for exceptions from child Scopes.

The methods setExceptionHandler and setChildScopeExceptionHandler cannot be used with the globalScope.
If an attempt is made to do so, an exception will be thrown.

Example:

$scope = new Scope();
$scope->setExceptionHandler(function (Async\Scope $scope, Async\Coroutine $coroutine, Throwable $e) {
    echo "Caught exception: {$e->getMessage()}\n in coroutine: {$coroutine->getSpawnLocation()}\n";
});
 
spawn with $scope {
    throw new Exception("Task 1");
};
 
$scope->awaitCompletion(Async\signal(SIGTERM));

Using these handlers, you can implement the Supervisor pattern, i.e., a Scope that will not be canceled when an exception occurs in coroutines.

If the setExceptionHandler or setChildScopeExceptionHandler handlers throw an exception,
it will be propagated to the parent Scope or the global Scope.

The setChildScopeExceptionHandler method allows handling exceptions only from child Scopes, which can be useful for implementing an algorithm where the main Scope runs core tasks, while child Scopes handle additional ones.

For example:

use Async\Scope;
use Async\Coroutine;
 
final class Service
{
    private Scope $scope;
 
    public function __construct()
    {
        $this->scope = new Scope();
 
        $this->scope->setChildScopeExceptionHandler(
        static function (Scope $scope, Coroutine $coroutine, \Throwable $exception): void {
            echo "Occurred an exception: {$exception->getMessage()} in Coroutine {$coroutine->getSpawnLocation()}\n";
        });
    }
 
    public function start(): void
    {
        spawn with $this->scope $this->run();
    }
 
    public function stop(): void
    {
        $this->scope->cancel();
    }
 
    private function run(): void
    {
        while (($socket = $this->service->receive()) !== null) {
 
            $scope = Scope::inherit($this->scope);
 
            // supervisor pattern
            (spawn with $scope $this->handleRequest($socket))->onFinally(
                static function () use ($scope) {
                    $scope->disposeSafely();
                }
            );
        }
    }
}

$this->scope listens for new connections on the server socket. Canceling $this->scope means shutting down the entire service.

Each new connection is handled in a separate Scope, which is inherited from $this->scope. If an exception occurs in a coroutine created within a child Scope, it will be passed to the setChildScopeExceptionHandler handler and will not affect the operation of the service as a whole.

A responsibility point is code that explicitly waits for the completion of a coroutine or a Scope:

$scope = new Scope();
 
spawn with $scope {
  throw new Exception("Task 1");
};
 
try {
    $scope->awaitCompletion(Async\signal(SIGTERM));
} catch (\Throwable $e) {
     echo "Caught exception: {$e->getMessage()}\n";
}

The Scope class provides a method for handling exceptions:

$scope = new Scope();
 
spawn with $scope {
    throw new Exception("Task 1");
};
 
$scope->setExceptionHandler(function (Exception $e) {
    echo "Caught exception: {$e->getMessage()}\n";
});
 
$scope->awaitCompletion(Async\signal(SIGTERM));

An exception handler has the right to suppress the exception. However, if the exception handler throws another exception, the exception propagation algorithm will continue.

The onFinally method allows defining a callback function that will be invoked when a coroutine or scope completes. This method can be considered a direct analog of defer in Go.

$scope = new Scope();
 
spawn with $scope {
    throw new Exception("Task 1");
};
 
$scope->onFinally(function () {
    echo "Task 1 completed\n";
});
 
$scope->awaitCompletion(Async\signal(SIGTERM));

Or for coroutines:

function task(): void
{
    throw new Exception("Task 1");
}
 
$coroutine = spawn task();
 
$coroutine->onFinally(function () {
    echo "Task completed\n";
});

The onFinally semantics are most commonly used to release resources, serving as a shorter alternative to try-finally blocks:

function task(): void
{
    $file = fopen('file.txt', 'r');
    onFinally(fn() => fclose($file));
 
    throw new Exception("Task 1");
}
 
spawn task();

The cancellation operation is available for coroutines and scopes using the cancel() method:

function task(): void {}
 
$coroutine = spawn task();
 
// cancel the coroutine
$coroutine->cancel(new Async\CancellationException('Task was cancelled'));

The cancellation operation is implemented as follows:

1. If a coroutine has not started, it will never start. 2. If a coroutine is suspended, its execution will resume with an exception. 3. If a coroutine has already completed, nothing happens.

The CancellationException, if unhandled within a coroutine, is automatically suppressed after the coroutine completes.

⚠️ Warning: You should not attempt to suppress CancellationException exception,
as it may cause application malfunctions.

$scope = new Scope();
 
spawn with $scope {
    sleep(1);
    echo "Task 1\n";
};
 
$scope->cancel(new Async\CancellationException('Task was cancelled'));

Canceling a Scope triggers the cancellation of all coroutines within that Scope and all child Scopes in hierarchical order.

Note: CancellationException can be extended by the user
to add metadata that can be used for debugging purposes.

In the context of coroutines, it is not recommended to use catch \Throwable or catch CancellationException.

Since CancellationException does not extend the \Exception class, using catch \Exception is a safe way to handle exceptions, and the finally block is the recommended way to execute finalizing code.

try {
    $coroutine = spawn {
        sleep(1);
        throw new \Exception("Task 1");
    };
 
    spawn use($coroutine) {
        $coroutine->cancel();
    };
 
    try {
        await $coroutine;
    } catch (\Exception $exception) {
        // recommended way to handle exceptions
        echo "Caught exception: {$exception->getMessage()}\n";
    }
} finally {
    echo "The end\n";
}

Expected output:

The end

try {
    $coroutine = spawn {
        sleep(1);
        throw new \Exception("Task 1");
    };
 
    spawn use($coroutine) {
        $coroutine->cancel();
    };
 
    try {
        await $coroutine;
    } catch (Async\CancellationException $exception) {
        // not recommended way to handle exceptions
        echo "Caught CancellationException\n";
        throw $exception;
    }
} finally {
    echo "The end\n";
}

Expected output:

Caught CancellationException
The end

The CancellationException affects PHP standard library functions differently. If it is thrown inside one of these functions that previously did not throw exceptions, the PHP function will terminate with an error.

In other words, the cancel() mechanism does not alter the existing function contract. PHP standard library functions behave as if the operation had failed.

Additionally, the CancellationException will not appear in get_last_error(), but it may trigger an E_WARNING to maintain compatibility with expected behavior for functions like fwrite (if such behavior is specified in the documentation).

Sometimes it's necessary to execute a critical section of code that must not be cancelled via CancellationException.

For example, this could be a sequence of write operations or a transaction.

For this purpose, the Async\protect function is used, which allows executing a closure in a non-cancellable (silent) mode.

function task(): void
{
    Async\protect(fn() => fwrite($file, "Critical data\n"));
}
 
spawn task();

If a CancellationException was sent to a coroutine during protect(), the exception will be thrown immediately after the execution of protect() completes.

The use of loops or unsafe operations inside a critical section can be checked by static analyzers.

This RFC intentionally does not define rules for tracking the execution time of cancelled coroutines. The reason is that cancellation operations may be long-running—for example, rollback strategies—and may require blocking the function being cancelled.

Intentionally stopping coroutines that are in the cancellation state is a dangerous operation that can lead to data loss. To avoid overcomplicating this RFC, it is proposed to delegate the responsibility for such logic to the Scheduler implementation.

The exit/die keywords called within a coroutine result in the immediate termination of the application. Unlike the cancel() operation, they do not allow for proper resource cleanup.

When an unhandled exception occurs in a Coroutine the Graceful Shutdown mode is initiated. Its goal is to safely terminate the application.

Graceful Shutdown cancels all coroutines in globalScope, then continues execution without restrictions, allowing the application to shut down naturally. Graceful Shutdown does not prevent the creation of new coroutines or close connection descriptors. However, if another unhandled exception is thrown during the Graceful Shutdown process, the second phase is triggered.

Second Phase of Graceful Shutdown - All Event Loop descriptors are closed. - All timers are destroyed. - Any remaining coroutines that were not yet canceled will be forcibly canceled.

The further shutdown logic may depend on the specific implementation of the Scheduler component, which can be an external system and is beyond the scope of this RFC.

The Graceful Shutdown mode can also be triggered using the function:

Async\gracefulShutdown(?CancellationException $cancellationException = null): void {}

from anywhere in the application.

A situation may arise where there are no active Coroutines in the execution queue and no active handlers in the event loop. This condition is called a Deadlock, and it represents a serious logical error.

When a Deadlock is detected, the application enters Graceful Shutdown mode and generates warnings containing information about which Coroutines are in a waiting state and the exact lines of code where they were suspended.

This RFC does not require the implementation of this tool but describes its potential use.

The Scheduler can implement a limit on the continuous execution time of a coroutine to regain control from tasks that may have “hung” due to a programmer's error.

The criterion is calculated as the maximum interval of active coroutine execution during which the coroutine does not yield control.

It is reasonable to set the maximum interval to short time periods: 3–5 seconds for applications that handle requests. If the maximum interval is exceeded, the Scheduler must generate a warning with precise information about which coroutine and on which line the situation occurred.

If this interval is exceeded, the Scheduler can interrupt the coroutine’s execution at any point, on any line, not just at suspension points or I/O function calls.

The Scheduler can cancel a coroutine using cancel(), which will throw an exception at the suspension point, or it can terminate it without the possibility of resumption (depending on the implementation).

The Coroutine class implements methods for inspecting the state of a coroutine.

The Coroutine::getAwaitingInfo() method returns an array with debugging information about what the coroutine is waiting for, if it is in a waiting state.

The format of this array depends on the implementation of the Scheduler and the Reactor.

The Async\Scope::getChildScopes() method returns an array of all child scopes of the current scope.

The method Async\Scope::getCoroutines() returns a list of coroutines that are registered within the specified Scope.

The Async\getCoroutines() method returns an array of all coroutines in the application.

• [Async functions](https://github.com/EdmondDantes/php-true-async-rfc/tree/main/examples/Async/Async.php)
• [Coroutine](https://github.com/EdmondDantes/php-true-async-rfc/tree/main/examples/Async/Coroutine.php)
• [Coroutine Context](https://github.com/EdmondDantes/php-true-async-rfc/tree/main/examples/Async/Context.php)
• [Coroutine Scope](https://github.com/EdmondDantes/php-true-async-rfc/tree/main/examples/Async/Scope.php)
• [Task Group](https://github.com/EdmondDantes/php-true-async-rfc/tree/main/examples/Async/TaskGroup.php)

Simultaneous use of the True Async API and the Fiber API is not possible.

1. If new Fiber() is called first, the Async\spawn function will fail with an error.
2. If Async\spawn is called first, any attempt to create a Fiber will result in an error.

PHP 8.5 / PHP 9.0

The True Async module activates the reactor within the context of php_request_startup\php_request_shutdown() request processing. Therefore, using concurrency is reasonable only for long-life scenarios implemented via CLI.

It is expected that True Async will enable the integration of built-in web servers into PHP, which will be embedded into the reactor’s event loop.

• PHP Socket Extension.
• Curl Extension.
• MySQL PDO Extension.
• Redis Extension.

Does not affect.

No new constants are introduced.

• async.zombie_coroutine_timeout - default 5 seconds

None.

• Fiber API.
• PHP Sockets.
• Proc Functions.
• Shell/Exec Functions.
• gethostbyname/gethostbyaddr/gethostname/gethostbynamel

This RFC assumes the subsequent development of additional RFCs:

* An RFC for a low-level API for C interfaces, which will describe how PHP will allow extensions and standard library code to interact with the concurrent environment. * An RFC for a set of standard primitives, such as Future, Channel, etc. * And possibly others...

This RFC provides for the subsequent expansion of functionality to achieve a complete toolkit for working with concurrent logic. It proposes development in areas:

• Support for Pipe
• Development of new and revision of existing extensions
• Refactoring of input-output code to improve performance and better integration with the Event Loop
• Functions for collecting metrics

The Future->map()->catch()->finally() call chain is rightly criticized for excessive complexity and difficulty of comprehension. Pipe (not UNIX-like-pipe) can solve this problem and create a more intuitive and understandable interface for describing sequences of asynchronous function calls.

Input-output modules such as PHP Streams can be redesigned with asynchronous capabilities in mind and better optimized for operation in this environment.

It would also be appropriate to add support for pipe in such a way that it can be used regardless of the operating system using fopen() functions. This would make the API more consistent.

Yes or no vote. 2/3 required to pass.

* Proof of concept: https://github.com/EdmondDantes/php-src/tree/async/ext/async

This code is an attempt to implement logic using Fiber. Most of it can be used to create the next version. I would be happy if someone would like to join me in this project!

Links to external references, discussions or RFCs

• First Discussion - https://externals.io/message/126402
• Second Discussion - https://externals.io/message/126537

Additional links:

• Comparison of concurrency models in programming languages
• Cooperative cancellation
• Understanding Concurrency Bugs in Real-World Programs with Kotlin Coroutines
• Understanding Real-World Concurrency Bugs in Go
• Static Analysis for Asynchronous JavaScript Programs

The following can be considered as competing solutions to the current implementation:

• Swoole (https://github.com/swoole/swoole-src) – a C++ library that implements a full feature set for concurrent programming. The advantage and disadvantage of Swoole is that it is a standalone solution that does not directly affect the language itself.

• The Swow (https://github.com/swow/) project is a C library that provides a good lightweight API while not affecting the language itself and not requiring changes to PHP.

• Examples of code: https://github.com/EdmondDantes/php-true-async-rfc/tree/main/examples

Keep this updated with features that were discussed on the mail lists.