rfc:fiber

PHP RFC: Fiber

Introduction

Fibers are primitives for implementing light weight cooperative concurrency in PHP. Basically they are a means of creating Closure that can be paused and resumed. The scheduling of fiber must be done by the programmer and not the VM.

As opposed to Generator (a stackless light weight concurrency implementation), each fiber comes with a stack. This enables the fiber to be paused from deeply nested function calls within the fiber block.

When a fiber is created it will not run automatically. Rather it must be explicitly asked to run using the Fiber::resume method. The code running inside the fiber can give up control by calling Fiber::yield in which case it yields control back to caller (the caller of the Fiber::resume).

Proposal

Why not make it as a Extension?

Fiber is a major language feature, that allows significant benefits for asynchronous frameworks. Providing it as an optional extension, just doesn't make sense.

Implementation

Proposed API

final class Fiber {
  public const STATUS_SUSPENDED = 1;
  public const STATUS_RUNNING   = 2;
  public const STATUS_FINISHED  = 3;
  public const STATUS_DEAD      = 4;
 
  /**
   * @param callable $callable any php callable to be paused
   * @param int $stack_size fiber stack init size
   */
  public function __construct(callable $callable = null, int stack_size = null) {}
 
  /**
   * pause the current fiber and ~return~ the $arg1
   * as the Fiber::resume's return value.
   */
  public static function yield($arg1) {}
 
  /**
   * Start or resume a fiber.
   * 
   * If the fiber is not started, call resume will init
   * the $callable with all args.
   *
   * If the fiber is paused, call resume will send the first arg
   * as the last Fiber::yield's return value.
   */
  public function resume($arg1...) {}
 
  /**
   * Throw an exception into the fiber.
   * 
   * You code can use try/catch to process error in the
   * top level function call. Some framework make heavy
   * usage of this feature.
   */
  public function throw(Throwable $e) {}
}

Usage Demo

function sub1()
{
	return Fiber::yield(1);
}
$fiber = new Fiber(function ($a, $b) {
	$c = Fiber::yield($a + $b);
 
	$d = sub1();
	return $d.$c;
});
 
echo $fiber->resume(1, 2);     // echo 3
echo $fiber->resume("world");  // echo 1
echo $fiber->resume("hello "); // echo "hello world"

Implementation Detail

In our simple implementation, we only backup/restore the zend stack. We cannot pause a Fiber during internal function call like `array_map`.

Backward Incompatible Changes

“Fiber” are now globally defined classes, which might collide with user defined classes of the same name in the global namespace. However, the risk of the introduction of them is considered to be very low, since the global namespace should not be used by PHP users.

Proposed PHP Version(s)

7.3

RFC Impact

To SAPIs

None

To Existing Extensions

None

To Opcache

None

New Constants

None

php.ini Defaults

- fiber.stack_size default stack size for A fiber

Open Issues

Why not support backup/restore the c stack?

Martin Schröder is working on this at https://github.com/fiberphp/fiber-ext/pull/30.

And here is the comparison.

PropertyStackless FiberNative Fiber
1 Minimum Memory UsageVM stack only (4 KB)VM & C stack (4 KB + 4 KB)
2 Supported Architectursany platformx86 at this time
3 Yield in Internal Functionunsupportedsupported
4 Yield in Iteratorunsupportedsupported

Stackless fiber use less memory and are not platform-dependend (1 & 2) which makes them very portable and efficient. They do however lack support for anything that involves internal function calls (3) including opcode handlers (4, e.g. foreach loop).

Native fibers are very platform-dependend (2) and use more memory because they do need to allocate a C call stack (1). While memory allocation will be done using mmap() it will still reserve virtual memory (can be problematic for a large number of fibers on 32 bit systems due to limited virtual memory addressing). The big advantage is that all kinds of internal function call (3 & 4) are supported without any changes to the existing codebase.

Why not introduce helper like Fiber::alive(),Fiber::running()?

And as a language feature, Fiber should only offer the essential API. User can implement these methods in user land code easily.

Why not introduce a dedicate method other than Fiber::resume() for Fiber initialization?

Both Ruby's Fiber and Lua's coroutine using the same resume() API to init and resume their coroutine.

Why not introduce keyword like async/await?

Introducing new keywords will cause BC impact. Both Ruby's Fiber and Lua's coroutine does not have such keyword as well.

Unaffected PHP Functionality

None

Future Scope

This sections details areas where the feature might be improved in future, but that are not currently proposed in this RFC.

Proposed Voting Choices

2/3+1 voting majority

Patches and Tests

Implementation

After the project is implemented, this section should contain

  1. the version(s) it was merged to
  2. a link to the git commit(s)
  3. a link to the PHP manual entry for the feature
  4. a link to the language specification section (if any)

References

Rejected Features

rfc/fiber.txt · Last modified: 2018/04/18 22:55 by lvht