Description:
------------
When __sleep includes uninitialized typed properties, serialize() will succeed (behaves as if the value was null), but unserialize() will cause a TypeError to be thrown if the value wasn't nullable.

I didn't see any mention of how uninitialized typed properties should behave for __sleep in the RFC or RFC/PR discussion.

Is the current behavior in 7.4 and 8.0-dev for __sleep on uninitialized properties by design, or is it an edge case that was overlooked? It seems surprising that data serializes successfully, but throws when it gets unserialized for the same class declaration.

I checked https://wiki.php.net/rfc/typed_properties_v2 , https://externals.io/message/102333 , and https://externals.io/message/103148 , and didn't see any references to __sleep (just "I wouldn't lose sleep
if we wanted to keep the GA date where it is.")
I also didn't see any mention of __sleep in php/php-src#3734 , or in the changes of the PR. (or in bugs.php.net)

Noticed in https://github.com/igbinary/igbinary/pull/250#discussion_r355812953

Test script:
---------------
<?php
class HasProp {
    public int $x;
    public function __sleep() { return ['x']; }
}
$x = new HasProp();
// string(28) "O:7:"HasProp":1:{s:1:"x";N;}"
var_dump($s = serialize($x));
// Fatal error: Uncaught TypeError: Typed property HasProp::$x must be int, null used in ...
unserialize($s);


Expected result:
----------------
Unserialize() should not throw, and should generate an instance of HasProp, with an uninitialized HasProp.

Ideally, serialize() should skip over the property $x. This would make it match the original data's representation for both typed nullable and non-nullable properties.



Actual result:
--------------
This throws a TypeError when unserializing the data it just serialized