I experienced the exact same issue on PHP 5.4.17 on OS X 10.9 (Mavericks DP6).
I wrote a simple test case, here it is :

Test script:
---------------
namespace jubianchi\Alias {
    class A {}
 
    var_dump(class_alias('\\jubianchi\\Alias\\A', 'C'));
    $reflector = new \ReflectionClass('C');
    var_dump($reflector->getName());  

    var_dump(class_alias('\\jubianchi\\Alias\\A', '\\jubianchi\\Alias\\B'));
    try {
        $reflector = new \ReflectionClass('\\jubianchi\\Alias\\B');
        var_dump($reflector->getName());
    } catch(\Exception $e) {
        var_dump(get_class($e) . ': ' . $e->getMessage());
    }

    var_dump(class_alias('\\jubianchi\\Alias\\A', 'jubianchi\\Alias\\B'));
    $reflector = new \ReflectionClass('\\jubianchi\\Alias\\B');
    var_dump($reflector->getName());        
}

Expected result:
----------------
bool(true)
string(17) "jubianchi\Alias\A"
bool(true)
string(17) "jubianchi\Alias\A"
bool(true)
string(17) "jubianchi\Alias\A"

Or:
----------------
bool(true)
string(17) "jubianchi\Alias\A"
bool(false)
string(60) "ReflectionException: Class \jubianchi\Alias\B does not exist"
bool(true)
string(17) "jubianchi\Alias\A"

Actual result:
----------------
bool(true)
string(1) "A"
bool(true)
string(17) "jubianchi\Alias\A"
bool(true)
string(60) "ReflectionException: Class \jubianchi\Alias\B does not exist"
bool(true)
string(17) "jubianchi\Alias\A"

As you can see, class_alias returns bool(true) as if everything went fine, so we 
expect the alias to be available but a reflection on the latter throws an 
exception.
I think class_alias should be able to handle the leading backslashes or return 
bool(false) if it can't.

Note: The bug report is too restrictive. A proper patch would have to work on all places where classnames come from string context. This at first means verifying that all places go via zend_lookup_class() and related functions, not EG(class_table) / CG(class_table)

Johannes: I agree, but we could start by patching this bug report right?
I got a patch here : https://github.com/jpauli/php-
src/compare/class_alias_registration_fix

Technically we could, but it adds some inconsistency if one place allows this but others not and that should be avoided.

I agree with Johannes about consistency.

The severity is not really is not very high and this use case can easily be 
handled at a useland level.

As long as this behavior is not "fixed" I think a warning on the doc shoudl be 
enough, even if I'd like to see it fixed (but as I said, it's not a big deal at 
the moment).

BTW, thanks for you work Julien :)

Yep, let's start finding all places where classes as strings can be used, and 
patch them all to use zend_lookup_class().
There shouldn't be tons of them AFAIR.

I'm not convinced that allowing a leading \ is something we should strive towards. The \ is unnecessary and redundant (as string names are always fully qualified). I'd rather allow only the canonical form.

Also agree with the fact that the leading backslashes are redundant but the point 
is that class_alias returns a value saying all went fine (bool(true)) when the 
alias is not reachable after the call.

The fix for this bug has been committed.

Snapshots of the sources are packaged every three hours; this change
will be in the next snapshot. You can grab the snapshot at
http://snaps.php.net/.

 For Windows:

http://windows.php.net/snapshots/
 
Thank you for the report, and for helping us make PHP better.