rfc:access_scope_from_magic_accessors

PHP RFC: Access Scope from Magic Accessors

Introduction

When using magic methods to access actual properties, respecting their declared visibility is often desired. Yet, accessing the calling scope to emulate the visibility restrictions is unreasonably difficult at the moment. This RFC proposes to pass the calling scope to magic accessors to make it trivial to get it.

Proposal

This RFC proposes to pass their calling scope to magic accessors: __get(), __set(), __isset(), __unset(), __call() and __callStatic().

The calling scope is defined as the name of the class that is calling a magic method. It's a string. It can also be null when a magic method is called outside of any class context, as is the case when calling them from the global scope or from the body of a named function. For closures, their scope is the class name they were created in (if any), or their rebound scope when using Closure::bind(). The calling scope is also the piece of information that is found under key “class” of frames returned by debug_backtrace().

Here are some examples:

class Foo {
    private $abc;
 
    public function __construct()
    {
        unset($this->abc); // enables magic methods when accessing the property
    }
 
    public function __get($name) {
        var_dump(debug_backtrace()[1]['class'] ?? null);
    }
}
 
$foo = new Foo();
 
$foo->abc; // NULL
 
class Bar {
    public function __construct($foo) {
        $foo->abc;
    }
}
 
$bar = new Bar($foo); // string(3) "Bar"
 
$f = fn ($foo) => $foo->abc;
 
$f($foo); // NULL
 
$g = \Closure::bind($f, null, Bar::class);
 
$g($foo); // string(3) "Bar"

Passing the calling scope to magic accessors would help to properly implement visibility-related logic. Right now, we have to call debug_backtrace() to write scope-sensitive logic, but it's very difficult to get it right: accounting for inheritance is hard (calling e.g. parent::__get() changes the outcome of debug_backtrace() inside this parent::__get() method.) Passing the calling scope would make this concern more evident and would solve the issue with inheritance. It would also fix the edge case of calling a magic method via ReflectionProperty::get/setValue(), where looking up at debug_backtrace() needs special care to extract the calling scope. Last but not least, removing calls to debug_backtrace() would improve the performance of such magic accessors.

Right now, the logic to access the calling scope is quite involving. This example doesn't handle chained inheritance calls:

class Foo extends Bar
{
	public function __get($name): mixed
	{
		$frame = debug_backtrace(\DEBUG_BACKTRACE_PROVIDE_OBJECT | \DEBUG_BACKTRACE_IGNORE_ARGS, 2)[1];
 
		$callingScope = $frame['class'] ?? null;
 
		if (\ReflectionProperty::class === $callingScope) {
			$callingScope = $frame['object']->class;
		}
 
		// Now we can implement visibility-related logic for accessing $name
 
		// Let's say we want to call parent::__get() and it uses the same logic to extract the calling
		// scope, it's now broken because instead of getting frame `[1]`, it should use frame `[2]`.
		// But knowing this requires more non trivial logic. 
	}
}

If this RFC is accepted, we'll be able to do this instead:

class Foo extends Bar
{
	public function __get($name, ?string $callingScope = null): mixed
	{
		// $callingScope is populated with the value know by the engine
 
		// Calling parent::__get($name, $callingScope) is done easily
		// without resorting to any specific logic
	}
}

The PHP engine will pass the calling scope to magic accessors even if they don't declare the corresponding argument. The corresponding value will always be accessible via func_get_arg().

This ensures that userland will be able to write code that works on both PHP <= 8.2 and PHP >= 8.3. (As a reminder, the engine forbids adding any extra arguments to the signature of magic methods: it's not allowed to declare a __get($name, $scope) on PHP <= 8.2. Note also that it is allowed to call this method with extra arguments, at least for userland classes.)

For consistency with the checks in place for the currently accepted arguments, the engine will throw a fatal error when the new argument is declared with a type that is not compatible with string|null.

Proposed PHP Version

PHP 8.3

RFC Impact

Unaffected PHP Functionality

Existing magic accessors won't be required to declare the new argument.

Backward Incompatible Changes

None

To Existing Extensions

Because the engine forbids calling internal methods with extra arguments, extensions that declare magic accessors (e.g. soap) will need to declare the new argument.

Proposed Voting Choices

Pass calling scope to magic accessors: yes/no?

Implementation

After the project is implemented, this section should contain

  1. the version(s) it was merged into
  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)
rfc/access_scope_from_magic_accessors.txt · Last modified: 2023/11/22 18:16 by nicolasgrekas