Please reverse the 'expected' and 'actual' results, I swapped them around.

To confirm what the expect result should be:

string(48) "(input)  O:15:"Test_TestClassA":1:{s:1:"x";i:4;}"
string(8) "__wakeup"
string(23) "(output) true (passing)"
string(48) "(input)  O:15:"Test_TestClassB":1:{s:1:"x";i:4;}"
string(8) "__wakeup"
string(23) "(output) true (passing)"
string(49) "(input)  O:16:"Test_TestClassAA":1:{s:1:"x";i:4;}"
string(8) "__wakeup"
string(23) "(output) true (passing)"
string(49) "(input)  O:16:"Test_TestClassBB":1:{s:1:"x";i:4;}"
string(8) "__wakeup"
string(23) "(output) true (passing)"

https://3v4l.org/rWFHv

To re-iterate, failing on all normal PHP versions yet working as expected on a third party PHP implementation (HHVM)

The behavioral change has been introduced with commit 5328d42[1]
(which fixed bug #67072).  Restoring the documented behavior (if
that even will be possible) doesn't appear to make sense after
such a long time, so the documentation needs to be fixed.

[1] <https://github.com/php/php-src/commit/5328d4289946e260232f3195ba2e0f0eb173d5ef>

I would not call this a documentation issue.

With the exception of say, creating an alternative serialisation parser either in PHP or as a module, how would one go about using the features of Serializable whilst keeping backwards compatibility with older serialised data?

I don't seem to recall when implementing JsonSerializable being unable to use serialize()...

I, personally, would advise against storing long-lasting data in
serialized format, anyway.

Anyhow, Anatol, what do you think?  Would it be possible and
sensible to restore the old behavior?

Thanks for digging that deep, Christoph. The original issue was local but turned out to reveal a common vulnerability with handling of the internal classe unserialisation. There was fixes on top of that as well. In the end, the decision of the 5.4, 5.5 and 5.6 RMs was to keep the essential part of the fix. Please check the links from Ferenc in the ticket you've linked. In the light of the issues revealed, the sentence "Classes that implement this interface no longer support __sleep() and __wakeup()." has still the precedence.  

@westie sure in some grades it might be painful to migrate an app from 5.4 to 7.x. There has been much more change than just this one in the meanwhile. The most viable way would be to recreate a clean serialized data for your app. And, manipulated serialize strings was never promised to work. In the end, it is a documentation issue today.

Thanks.

Thank you @cmb and @ab for your interest and insights into this issue.

Firstly, despite a warning or two in the comments in the documentation for serialize (https://secure.php.net/serialize) the documentation itself does not imply that output of serialize is mutable; to quote, it says that it "[...] generates a *storable* representation of a value" (emphasis mine)

I hope you don't find the tone of what I am going to suggest too abrasive but since this has now been classed a documentation issue as opposed to an undocumented breaking change, perhaps the documentation for serialize() also needs to be edited, to specifically warn against the use of it for any planned long term storage of object data, as well as a warning against relying on data being able to be unserialised between different versions of PHP?

Anyhow, I'm investigating alternative ways around this.

Thanks for your support!

Sorry, all my mistake!  I committed the respective note to the
manual[1] *after* the behavioral change had been introduced.

> […] to specifically warn against the use of it for any planned
> long term storage of object data […]

That's just my personal opinion, and it's not so much regarding
potential breakages in the serialization format, but rather to
prevent issues with changes to userland classes.

> […] as well as a warning against relying on data being able to
> be unserialised between different versions of PHP?

Well, exchanging serialized data between different versions of PHP
should be fine, if the class definitions are identical.

[1] <http://svn.php.net/viewvc/?view=revision&amp;revision=339383>

This bug has been fixed in the documentation's XML sources. Since the
online and downloadable versions of the documentation need some time
to get updated, we would like to ask you to be a bit patient.

Thank you for the report, and for helping us make our documentation better.

For further reference, I have (for migration and curiosity purposes) created a library that "fixes" this behaviour.

Any objects that were serialised before the class definition was changed to implement the `Serializable` interface will become unserialisable once again, using oddly enough, the unserialize method.

(The inverse is also true, __wakeup would be called when unserialising objects that no longer support unserialisation via `Serializable`)

At the time of writing the support is limited to the types I can easily replicate, but I hope to change that.

https://github.com/OUTRAGElib/unserialize