Tested and it looks good. However it's a big behavior change, despite it's bug. On that note it might have more sense this patch to go into master only.

This patch introduces a BC break as and such can not be merged into 5.4 (and I'm not sure if at all). Test code:

class A {
  function foo() {
    echo "I am A\n";
  }
}

class B extends A {
  function foo() {
    echo "I am B\n";
  }
  function who() {
        var_dump(get_class($this));
        call_user_func_array(array($this, "foo"), array());
  }
}

class C extends B {
  function foo() {
    echo "I am C\n";
  }

  function who() {
        var_dump(get_class($this));
        call_user_func_array(array($this, "foo"), array());
        parent::who();
  }
}

$c = new C;
$c->who();

Without the patch:

string(1) "C"
I am C
string(1) "C"
I am C

with the patch:

string(1) "C"
I am C
string(1) "C"
I am B

BC break and I'm not sure it's even right to call B's function when given C object. Note there's no "parent" in this example for call_user_func_array.

Also note that this would change a scope on the call even if we used not $this but any other object from the same class hierarchy, which also sounds wrong.

Oh, of course, overseen that ... while expanding the limitation of 'self' it breaks $this.

Comment on behalf of stas at php.net:

Closing pull for now as it produces BC break, please reopen or submit new pull with fixed code.

Guys, thank you for taking time and viewing my code.
Smalyshev, thank you for the test. I'll try to fix it.