I have database rows containing serialized objects.
I want to deserialize these, but the class has changed, some properties went private so the deserialization no longer works.
Is there a way to force deserialization to an array or a stdClass? (or anything that won't cause an error upon deserialization)
I want to avoid migrating data with a script. I'd rather have backward compatibility with the objects serialized in the old format.
Not really, or at least i would be pretty afraid to use something like this in production.
However, unserialize will use the autoload system or the function name specified in the unserialize_callback_func ini setting. So with a little hacking you can make this work:
// this a serialized object with the class "SomeMissingClass"
$str = 'O:16:"SomeMissingClass":1:{s:1:"a";s:1:"b";}';
ini_set('unserialize_callback_func', 'define_me'); // set your callback_function
// unserialize will pass in the desired class name
function define_me($classname) {
// just create a class that has some nice accessors to it
eval("class $classname extends ArrayObject {}");
}
$object = unserialize($str);
print $object['a']; // should print 'b'
You can use something like this to migrate your data to a little more handy format.
UPDATE:
I've consulted with my repressed memories on this (i've faced something like this once) and remembered an other solution:
So you have your SomeClass with a private property named $a
class SomeClass {
private $a;
public function getA(){
return $this->a;
}
}
And you have the serialized version of it:
$str = 'O:9:"SomeClass":1:{s:1:"a";s:1:"b";}';
When you unserialize it, and dump the result it will look like this, which is no good:
$a = unserialize($str);
var_dump($a->getA()); // prints 'null'
var_dump($a);
/*
prints:
object(SomeClass)#1 (2) {
["a":"SomeClass":private]=>
NULL
["a"]=>
string(1) "b"
}
*/
Now, when an object get's unserialized, php will call it's __wakeup magic method. The data you need is there in the object, but not under the private property but a similarly named public one. You can't reach that with the $this->a since it will look for the wrong one,
however the method get_object_vars() will return these properties and you can reassign them inside a __wakeup():
class SomeClass {
private $a;
public function getA(){
return $this->a;
}
public function __wakeup(){
foreach (get_object_vars($this) as $k => $v) {
$this->{$k} = $v;
}
}
}
$str = 'O:9:"SomeClass":1:{s:1:"a";s:1:"b";}';
$a = unserialize($str);
print $a->getA();
I think it's needless to say that you should save yourself the further headache and convert your data to some dedicated data exchange format.
Aside from manually munging the data in the database itself, which is always a risky proposition, I think your only option is to roll back the class code to an older version, extract the data, then re-store it in a more sensible way that can be more easily dealt with in future code revisions.
You do have the old classes in your SVN/GIT/CVS, right?
Related
The I18n class in CakePHP provides this method to create instances:
public static function getInstance() {
static $instance = array();
if (!$instance) {
$instance[0] = new I18n();
}
return $instance[0];
}
Among other considerations (please correct me if I'm wrong), I understand it helps to use class instances from the convenience functions:
/**
* Returns a translated string if one is found; Otherwise, the submitted message.
*/
function __($singular, $args = null) {
// ...
$translated = I18n::translate($singular);
// ...
}
echo __('Hello, World!');
This looks cleaner than having to pass the instance around as argument (or, even worse, using a randomly named global variable). But I can't imagine a reason why $instance is an array rather than a plain object.
What can be the purpose of using a one-item array to store class instances?
I would suspect this to be leftovers from older PHP4/CakePHP versions where the instances were assigned by reference.
https://github.com/cakephp/cakephp/blob/1.2.0/cake/libs/i18n.php
function &getInstance() {
static $instance = array();
if (!$instance) {
$instance[0] =& new I18n();
$instance[0]->l10n =& new L10n();
}
return $instance[0];
}
$_this =& I18n::getInstance();
Assigning by reference doesn't work with static, the reference is not being remembered, but it works when assigned to an array entry.
So this was most probably just a workaround for a PHP limitation.
One possible reason for this is to keep all singleton class instances in one global - (static is a synonym of global in this case) array variable for monitoring or not messing the global/local namespace with individual variables for each singleton. If each of the static variables were with random names e.g $translated it would be more easier to overwrite and mess its value. - bug again for me, this is extremely rear possibility.
For example the I18Nn instance would be with [0] key, other class would have other key. You should check outher singleton classes how manage the static $instance array values.
First, take a look at this PHP 5.5.8 code which implements lazy initialization of class properties with using a Trait:
trait Lazy
{
private $__lazilyLoaded = [];
protected function lazy($property, $initializer)
{
echo "Initializer in lazy() parameters has HASH = "
. spl_object_hash($initializer) . "\n";
if (!property_exists($this, $property)
|| !array_key_exists($property, $this->__lazilyLoaded))
{
echo "Initialization of property " . $property . "\n";
$this->__lazilyLoaded[$property] = true;
$this->$property = $initializer();
}
return $this->$property;
}
}
class Test
{
use Lazy;
private $x = 'uninitialized';
public function x()
{
return $this->lazy('x', function(){
return 'abc';
});
}
}
echo "<pre>";
$t = new Test;
echo $t->x() . "\n";
echo $t->x() . "\n";
echo "</pre>";
The output is as follow:
uninitialized
Initializer in lazy() parameters has HASH = 000000001945aafc000000006251ed62
Initialization of property x
abc
Initializer in lazy() parameters has HASH = 000000001945aafc000000006251ed62
abc
Here are my questions and things I'd like to discuss and improve, but I don't know how.
Based on the HASH values reported, it may appear that the initializer function is created only once.
But actually uniqueness is not guaranteed between objects that did not reside in memory simultaneously. So the question remains unanswered - whether the initializer gets created only once, and it matters for performance I think, but I'm not sure.
The way it's implemented now is not very safe in that if I refactor the code and change property $x to something else, I might forget to change the 'x' value as a first parameter to lazy() method. I'd be happy to use & $this->x instead as a first parameter, but then inside lazy() function I don't have a key to use for $__lazilyLoaded array to keep track of what has been initialized and what has not. How could I solve this problem? Using hash as a key isn't safe, nor it can be generated for callbacks like array($object, 'methodName')
If $this->x is a private property, it's safe for outer world to call the x() method, but for the class' methods it's still unsafe to access the raw $this->x property as it can be still uninitialized. So I wonder is there a better way - maybe I should save all the values in some Trait's field?
The global aim is to make it:
a) Fast - acceptable enough for small and medium software applications
b) Concise in syntax - as much as possible, to be used widely in the methods of the classes which utilize the Lazy trait.
c) Modular - it would be nice if objects still held their own properties; I don't like the idea of one super-global storage of lazily-initialized values.
Thank you for your help, ideas and hints!
So the question remains unanswered - whether the
initializer gets created only once, and it matters for performance I
think, but I'm not sure.
Well, closure instance is created only once. But anyway, performance will depend not on closure instance creation time (since it is insignificant), but closure execution time.
I'd be happy to use & $this->x instead as a first parameter, but then
inside lazy() function I don't have a key to use for $__lazilyLoaded
array to keep track of what has been initialized and what has not. How
could I solve this problem? Using hash as a key isn't safe, nor it can
be generated for callbacks like array($object, 'methodName')
I can propose the following solution:
<?php
trait Lazy
{
private $_lazyProperties = [];
private function getPropertyValue($propertyName) {
if(isset($this->_lazyProperties[$propertyName])) {
return $this->_lazyProperties[$propertyName];
}
if(!isset($this->_propertyLoaders[$propertyName])) {
throw new Exception("Property $propertyName does not have loader!");
}
$propertyValue = $this->_propertyLoaders[$propertyName]();
$this->_lazyProperties[$propertyName] = $propertyValue;
return $propertyValue;
}
public function __call($methodName, $arguments) {
if(strpos($methodName, 'get') !== 0) {
throw new Exception("Method $methodName is not implemented!");
}
$propertyName = substr($methodName, 3);
if(isset($this->_lazyProperties[$propertyName])) {
return $this->_lazyProperties[$propertyName];
}
$propertyInializerName = 'lazy' . $propertyName;
$propertyValue = $this->$propertyInializerName();
$this->_lazyProperties[$propertyName] = $propertyValue;
return $propertyValue;
}
}
/**
* #method getX()
**/
class Test
{
use Lazy;
protected function lazyX() {
echo("Initalizer called.\r\n");
return "X THE METHOD";
}
}
echo "<pre>";
$t = new Test;
echo $t->getX() . "\n";
echo $t->getX() . "\n";
echo "</pre>";
Result:
c:\Temp>php test.php
<pre>X THE METHOD
X THE METHOD
</pre>
c:\Temp>php test.php
<pre>Initalizer called.
X THE METHOD
X THE METHOD
</pre>
c:\Temp>
You cannot always be protected from forgetting something, but it is easier to remember when all things are close to each other. So, I propose to implement lazy loaders as methods on corresponding classes with specific names. To provide autocomplete #method annotation can be used. In a good IDE refactoring method name in annotation will allow to rename method across all project. Lazy loading function will be declared in the same class so renaming it also is not a problem.
By declaring a function with a name, starting with "lazy", in my example you both declare a corresponding accessor function, with name starting with "get" and it's lazy loader.
If $this->x is a private property, it's safe for outer world to call the x() method, but for the class' methods it's still unsafe to
access the raw $this->x property as it can be still uninitialized. So
I wonder is there a better way - maybe I should save all the values in
some Trait's field?
Trait fields are available in the class, that uses specific trait. Even private fields. Remember, this is composition, not inheritance. I think it's better to create private trait array field and store your lazy properties there. No need to create a new field for every property.
But I cannot say I like the whole scheme. Can you explain the use of it for you? May be we can come with better solution.
_toString() is called when an object is used as string. How can I do something similar for numerical values, something like __toInt(), or __toArray(). Do such methods exist? Is there a work around? Is it a bad idea to use something like that even if there is a workaround for it?
There is no __toArray magic-method (just check the ones that exist here), but then, there shouldn't be, IMO.
Though people have asked for a magic toArray method, it doesn't look like such a method will be implemented any time soon.
Considering what objects are for, and how we use them, a toInt method wouldn't make much sense, and since all objects can be cast to an array, and can be iterated over, I see very little point in using __toArray anyway.
To "convert" on object to an array, you can use either one of the following methods:
$obj = new stdClass;
$obj->foo = 'bar';
var_dump((array) $obj);
//or
var_dump(json_decode(json_encode($obj), true));
This can be done with both custom objects, as stdClass instances alike.
As far as accessing them as an array, I can't see the point. Why write a slow magic method to be able to do something like:
$bar = 'foo';
$obj[$bar];
if you can do:
$obj->{$bar}
or if you can do:
foreach($obj as $property => $value){}
Or, if you need something a tad more specific, just implement any of the Traversable interfaces.
And for those rare cases, where you want an object to produce an array from specific properties in a very particular way, just write a method for that and call that method explicitly.
class ComplexObject
{
private $secret = null;
private $childObject = null;
public $foo = null;
//some methods, then:
public function toArray()
{//custom array representation of object
$data = array();
foreach($this->childObject as $property => $val)
{
if (!is_object($this->childObject->{$property}))
{
$data[$property] = $val;
}
}
$data['foo'] = $this->foo;
return $data;
}
//and even:
public function toJson()
{
return json_encode($this->toArray());
}
}
Ok, you have to call these methods yourself, explicitly, but that's not that hard, really... is it?
In perl I'm used to doing
my $foo = new WhatEver( bar => 'baz' );
and now I'm trying to figure out if PHP objects can ever be constructed this way. I only see this:
my $foo = new WhatEver();
$foo->{bar} = 'baz';
is it possible to do it in one step?
You can lay out your constructor as follows:
class MyClass {
public function __construct($obj=null) {
if ($obj && $obj instanceof Traversable || is_array($obj)) {
foreach ($obj as $k => $v) {
if (property_exists($this,$k)) {
$this->{$k} = $v;
}
}
}
}
}
This has a serie of drawbacks:
This is inefficient
The variables you create will not show up on any doc software you use
This is the open door to all forms of slackery
However, it also presents the following benefits:
This can be extended pretty safely
It allows you to lazy-implement variables
It also allows you to set private variables, provided that you know their names. It is pretty good in that respect if not abused.
The parameters passed in the parentheses (which can be omitted, by the way, if there aren't any) go to the constructor method where you can do whatever you please with them. If a class is defined, for example, like this:
class WhatEver
{
public $bar;
public function __construct($bar)
{
$this -> bar = $bar;
}
}
You can then give it whatever values you need.
$foo = new WhatEver('baz');
There are a few ways to accomplish this, but each has its own drawbacks.
If your setters return an instance of the object itself, you can chain your methods.
my $foo = new WhatEver();
$foo->setBar("value")->setBar2("value2");
class WhatEver
{
public $bar;
public $bar2;
public function setBar($bar)
{
$this->bar = $bar;
return $this;
}
public function setBar2($bar2)
{
$this->bar2 = $bar2;
return $this;
}
}
However, this doesn't reduce it to one step, merely condenses every step after instantiation.
See: PHP method chaining?
You could also declare your properties in your constructor, and just pass them to be set at creation.
my $foo = new WhatEver($bar1, $bar2, $bar3);
This however has the drawback of not being overtly extensible. After a handful of parameters, it becomes unmanageable.
A more concise but less efficient way would be to pass one argument that is an associative array, and iterate over it setting each property.
The implicit assumption here is that objects have meaningful, presumably public, properties which it is up to the calling code to provide values for. This is by no means a given - a key aspect of OOP is encapsulation, so that an object's primary access is via its methods.
The "correct" mechanism for initialising an object's state is its constructor, not a series of property assignments. What arguments that constructor takes is up to the class definition.
Now, a constructor might have a long series of named parameters, so that you could write $foo = new WhatEver(1, "hello", false, null) but if you want these to act like options, then it could take a single hash - in PHP terms, an Array - as its argument.
So, to answer the question, yes, if your constructor is of the form function __construct(Array $options) and then iterates over or checks into $options. But it's up to the constructor what to do with those options; for instance passing [ 'use_safe_options' => true ] might trigger a whole set of private variables to be set to documented "safe" values.
As of PHP 5.4 (which introduced [ ... ] as an alternative to array( ... )), it only takes a few more character strokes than the Perl version:
$foo = new WhatEver( ['bar' => 'baz'] );
Assume this class code:
class Foo {
function method() {
echo 'works';
}
}
Is there any way to store a reference to the method method of a Foo instance?
I'm just experimenting and fiddling around, my goal is checking whether PHP allows to call $FooInstance->method() without writing $FooInstance-> every time. I know I could write a function wrapper for this, but I'm more interested in getting a reference to the instance method.
For example, this pseudo-code would theoretically store $foo->method in the $method variable:
$foo = new Foo();
$method = $foo->method; //Undefined property: Foo::$method
$method();
Apparently, as method is a method and I'm not calling it with () the interpreter thinks I'm looking for a property thus this doesn't work.
I've read through Returning References but the examples only show how to return references to variables, not methods.
Therefore, I've adapted my code to store an anonymous function in a variable and return it:
class Foo {
function &method() {
$fn = function() {
echo 'works';
};
return $fn;
}
}
$foo = new Foo();
$method = &$foo->method();
$method();
This works, but is rather ugly. Also, there's no neat way to call it a single time, as this seems to require storing the returned function in a variable prior to calling it: $foo->method()(); and ($foo->method())(); are syntax errors.
Also, I've tried returning the anonymous function directly without storing it in a variable, but then I get the following notice:
Notice: Only variable references should be returned by reference
Does this mean that returning/storing a reference to a class instance method is impossible/discouraged or am I overlooking something?
Update: I don't mind adding a getter if necessary, the goal is just getting a reference to the method. I've even tried:
class Foo {
var $fn = function() {
echo 'works';
};
function &method() {
return $this->fn;
}
}
But from the unexpected 'function' (T_FUNCTION) error I'd believe that PHP wisely doesn't allow properties to store functions.
I'm starting to believe that my goal isn't easily achievable without the use of ugly hacks as eval().
It is. You have to use an array, with two values: the class instance (or string of the class name if you are calling a static method) and the method name as a string. This is documented on the Callbacks Man page:
A method of an instantiated object is passed as an array containing an object at index 0 and the method name at index 1.
Demo (Codepad):
<?php
class Something {
public function abc() {
echo 'called';
}
}
$some = new Something;
$meth = array($some, 'abc');
$meth(); // 'called'
Note this is also works with the built-ins that require callbacks (Codepad):
class Filter {
public function doFilter($value) {
return $value !== 3;
}
}
$filter = new Filter;
$test = array(1,2,3,4,5);
var_dump(array_filter($test, array($filter, 'doFilter'))); // 'array(1,2,4,5)'
And for static methods -- note the 'Filter' instead of an instance of a class as the first element in the array (Codepad):
class Filter {
public static function doFilter($value) {
return $value !== 3;
}
}
$test = array(1,2,3,4,5);
var_dump(array_filter($test, array('Filter', 'doFilter'))); // 'array(1,2,4,5)'
// -------- or -----------
var_dump(array_filter($test, 'Filter::doFilter')); // As of PHP 5.2.3
Yes, you can. PHP has a "callable" pseudo-type, which is, in fact, either just a string or an array. Several functions (usort comes to mind) accept a parameter of the "callback" type: in fact, they just want a function name, or an object-method pair.
That's right, strings are callable:
$fn = "strlen";
$fn("string"); // returns 6
As mentioned, it's possible to use an array as a callback, too. In that case, the first element has to be an object, and the second argument must be a method name:
$obj = new Foo();
$fn = array($obj, "method");
$fn(); // calls $obj->method()
Previously, you had to use call_user_func to call them, but syntax sugar in recent versions make it possible to perform the call straight on variables.
You can read more on the "callable" documentation page.
No, as far as I know it's not possible to store a reference to a method in PHP. Storing object / class name and a method name in an array works, but it's just an array without any special meaning. You can play with the array as you please, for example:
$ref = [new My_Class(), "x"];
// all is fine here ...
$ref();
// but this also valid, now the 'reference' points to My_Other_Class::x()
// do you expect real reference to behave like this?
$ref[0] = new My_Other_Class();
$ref();
// this is also valid syntax, but it throws fatal error
$ref[0] = 1;
$ref();
// let's assume My_Class::y() is a protected method, this won't work outside My_Class
$ref = [new My_Class(), 'y'];
$ref();
this is prone to error as you loose syntax checking due to storing the method name as string.
you can't pass reliably a reference to a private or a protected method this way (unless you call the reference from a context that already has proper access to the method).
Personally I prefer to use lambdas:
$ref = function() use($my_object) { $my_object->x(); }
If you do this from inside $my_object it gets less clunky thanks to access to $this:
$ref = function() { $this->x(); }
this works with protected / private methods
syntax checking works in IDE (less bugs)
unfortunately it's less concise