just a quick test with your patch, got a memleak:

<?php

class b {
  public function dummy() {
      echo self::class;
    }
}

lead to a memleak:

[Sun Sep  9 22:19:52 2012]  Script:  '/tmp/1.php'
Zend/zend_language_scanner.l(1907) :  Freeing 0x2B1498A80288 (5 bytes), script=/tmp/1.php
=== Total 1 memory leaks detected ===

I've located the leak and will work the fix out, thanks- (I forgot to compile with --enable-debug).

I am working on a few more tests and some remaining issues (parent::class in type hints and constants e.g.)

@ralphschindler would love to get this merged, as the RFC ended positively. Could you drop me a message to discuss the remaining work?

I need to rework the patch as per the conversations that were had on the internals list. My plan is to do it during the holidays this week (American Thanksgiving), so I should be able to get this updated by end of week.

In its current state, it's not good for merge.

@ralphschindler, cool, thanks! If you need help with anything, drop me a note.

@ralphschindler any news?

@ralphschindler ping

@ralphschindler ping again.

@ralphschindler By the way, if you want this in 5.5 you need to hurry. We're close to feature freeze ;)

@nikic I have it mostly working, in two different forms, any chance you could help me bring it to completion?

@ralphschindler Depends on how much help you need, but in principle yes :) What are you still missing / what issues are you having with the implementation?

@nikic I will push what I have to my github account, and give you a summary of the decisions I need to make.

Mostly, I need to decide how to handle situations like this:

class Foo {
    public function bar(static::class $foo, parent::class $bar, self::class $baz) {}
}

I believe I had self::class working in that scenario, but the other two were not with the current implementation of the FETCH_CLASS_CONST.

@ralphschindler Assuming you mean $foo = static::class etc.

Parameter initializers (and all other initializers) in PHP only accept compile-time scalars. As static::class and parent::class are not resolvable at compile time they should just throw an error there. parent::class is theoretically resolvable at compile-time (because it only needs the name and not the actual CE) and you can obtain it by checking the op array for the FETCH_CLASS op of the class declaration. But I don't think that it's worth adding such a hack for something nobody will do anyway (what's supposed to be the use for initializing a parameter to the name of the parent class?)

I agree with @nikic here. If parent and static throw an error, that’s more than fine.

Since I decided to go in a different direction with the implementation and wanted to keep a clean commit history, I re-branched from master and applied my fixes there and will open up a new PR. see #256

@ralphschindler: You don't need a new pull request, just force-push to this branch, the history will update here.