Constant time ObjectClass handling

[Matthias Dieter Wallnöfer]

Okay, the patches which let "dbcheck" detect and fix wrong "objectClass" 
attributes are ready for review in my own "master" branch.

This should be enough to close [Bug 8635] Samba 4 denies Linux/Samba 
single sign on.

Thanks,
Matthias

Andrew Bartlett schrieb:
> On Tue, 2012-04-10 at 15:40 +0200, Matthias Dieter Wallnöfer wrote:
>> Hi Andrew,
>>
>> Andrew Bartlett schrieb:
>>> Thanks.  I'll push these at some point over easter, except for:
>>>
>>> commit 0dac7b21d97d504653375b8cfddec8a5f35bfc23
>>> Author: Matthias Dieter Wallnöfer<mdw at samba.org>
>>> Date:   Fri Apr 6 12:40:25 2012 +0200
>>>
>>>       LDB/s4:dsdb - don't allocate objects on reallocable structures
>>>
>>>       The "msg->elements" or "struct ldb_message_element" objects (array)
>>> are subject
>>>       to changes when adding/removing message elements. As far as I
>>> understand
>>>       "talloc" cannot handle this correctly.
>>>       "msg" however remains always the same for a lifetime of a message.
>>>
>>>
>>> Do you have any evidence that talloc doesn't do this properly?  It
>>> certainly is meant to, so if this isn't working we should investigate
>>> what is really going on.
>> Okay, I have not been aware of this circumstance. So I wrote this patch
>> just in case talloc could not handle this - I was simply not sure about,
>> let us drop it.
> Thanks,
>
>>> One questions however: On the key patch:
>>>
>>> commit 27078fac8555ef48b8da98650fe944180a177887
>>> Author: Matthias Dieter Wallnöfer<mdw at samba.org>
>>> Date:   Wed Apr 4 18:40:00 2012 +0200
>>>
>>>       s4:dsdb - move "objectclass_sort()" out from the objectclass LDB
>>> module into the schema code
>>>
>>>       This allows it to be useful for the dbchecker utility in respect to
>>>       object class problems.
>>>       Fix up the API to only work with standardised LDB
>>> "ldb_message_element"
>>>       structures which do allow much easier interoperations. As a
>>> consequence this
>>>       leads to some changes in the objectclass module as well.
>>>
>>> How did you solve the test failures?  When I tried to have it sort
>>> before doing the get_last_structural_class() and
>>> check_unrelated_objectclasses() I couldn't make it pass the ldap.py
>>> python tests.
>> I am a bit puzzled. But the patches should now be passing ldap.py - as
>> far as I have tested. Or are you still getting failures?
> I first worked on a very similar patch, but wasn't able to make it pass
> the tests.  That's all.
>
> I'll get these into the tree after I do the alpha19 release.
>
> Andrew Bartlett
>