Description:
------------
I am seeing weird behavior that I believe was introduced by this patch:

https://bugs.php.net/bug.php?id=69210

It seems that if you call serialize(debug_backtrace()) where the backtrace contains an object with a __sleep method that returns a non-string (in this case, an array containing a member variable of that object), the member variable is overridden with the __toString representation of that object. 

That's a bit confusing to understand but I hope my example code will illustrate it better.

Test script:
---------------
<?php

class Obj {
    public function __construct($mix) {
        foreach ($mix as $k => $v) {
            $this->$k = $v;
        }
    }

    public function __toString() {
        $str = '';
        foreach ($this as $k => $v) {
            $str = "$k => $v\n";
        }

        return $str;
    }
}

class A {
    public $obj;

    public function setObj($obj) {
        $this->obj = $obj;
    }

    public function getObj() {
        return $this->obj;
    }

    function __sleep() {
        return ['test' => $this->obj];
    }
}

function printA(A $a) {
    var_dump($a);

    serialize(debug_backtrace(0));
}

$obj = new Obj(['hello'=>'there']);
$a = new A();
$a->setObj($obj);

var_dump($a->getObj());

printA($a);

var_dump($a->getObj());

Expected result:
----------------
object(Obj)#1 (1) {
  ["hello"]=>
  string(5) "there"
}
object(A)#2 (1) {
  ["obj"]=>
  object(Obj)#1 (1) {
    ["hello"]=>
    string(5) "there"
  }
}
object(Obj)#1 (1) {
  ["hello"]=>
  string(5) "there"
}

Actual result:
--------------
object(Obj)#1 (1) {
  ["hello"]=>
  string(5) "there"
}
object(A)#2 (1) {
  ["obj"]=>
  object(Obj)#1 (1) {
    ["hello"]=>
    string(5) "there"
  }
}
string(15) "hello => there"