[PATCH][PYTHON] Fix replPropertyMetaData sort order for schema extension attributes

[Stefan Metzmacher]

Am 24.08.2015 um 06:03 schrieb Andrew Bartlett:
> On Wed, 2015-08-12 at 11:45 +1200, Andrew Bartlett wrote:
>> On Wed, 2015-08-05 at 20:19 +1200, Andrew Bartlett wrote:
>>> On Wed, 2015-08-05 at 09:23 +0200, Stefan Metzmacher wrote:
>>>> Hi Andrew,
>>>>
>>>>>>>> - Do we have a test that checks the objectClass sorting 
>>>>>>>> against the 
>>>>>>>> LDAP
>>>>>>>> server
>>>>>>>>   (and passed against windows)?
>>>>>>>
>>>>>>> No, not currently.  We should have it check that the 
>>>>>>> correct
>>>>>>> attributeID is added, that it is sorted, and that when 
>>>>>>> entries are
>>>>>>> modified, replaced or deleted, that the version is 
>>>>>>> incremented. 
>>>>>>
>>>>>> That an answer to the wrong question...
>>>>>
>>>>> Ah. :-)
>>>>>
>>>>>> My question was about the sorting of objectClass values:
>>>>>> https://git.samba.org/?p=abartlet/samba.git/.git;a=commitdiff
>>>>>> ;h
>>>>>> =c74751d5e89e9b36a0987589f4dd0e481137561d
>>>>>
>>>>> I have no intention of matching Windows behaviour there.  The 
>>>>> docs (and
>>>>> I should reference those) say the order is undefined for AUX 
>>>>> classes.
>>>>
>>>> I'm sorry but we really need to match Windows, everything else 
>>>> will
>>>> cause problems
>>>> with poorly written admin scripts.
>>>
>>>> And we need tests to prove it our behavior matches a Windows 
>>>> server.
>>>
>>> Then please NAK the patch until you can come up with a behaviour 
>>> you 
>>> can
>>> live with. 
>>>
>>> To be clear, each time the objectClass list is modified, the 
>>> portion 
>>> of
>>> the list with AUX classes reverses. 
>>
>> I've dropped these patches from the replmetadata-sort-minimal branch,
>> but I still think this is the best solution for having the dbcheck 
>> code
>> testable. 
>>
>>>>>>
>>>>>>>> - Please add BUG: https:... tags to the commit messages
>>>>>>>>   We want to backport this at least to v4-3-test and v4-2
>>>>>>>> -test, but
>>>>>>>> maybe also to v4-1-test...
>>>>>>>
>>>>>>> The key is to backport 
>>>>>>> 61b978872fe86906611f64430b2608f5e7ea7ad8, then
>>>>>>> these to fix up broken databases. 
>>>>>>>
>>>>>>> The big remaining issue is that our msDS-IntID handling is 
>>>>>>> totally
>>>>>>> broken.  We use that only when pushing the values over the 
>>>>>>> network with
>>>>>>> GetNCChanges, but not during storage in 
>>>>>>> replPropertyMetaData.
>>>>>>>   This
>>>>>>> means for the an object with the same attribute modified on 
>>>>>>>
>>>>>>> two servers
>>>>>>> we can end up with metadata for an entry twice, once for 
>>>>>>> each 
>>>>>>> style of
>>>>>>> attributeID.
>>>>>>
>>>>>> Ok, do we have bug reports for all this problems?
>>>>>
>>>>> Not yet.
>>>>
>>>> Please add them and cc me.
>>>
>>> Certainly.
>>
>> See bug https://bugzilla.samba.org/show_bug.cgi?id=11443
>>
>> Now that this has been done, and with the sorting issues put aside 
>> for
>> the moment, can you please review the remaining patches?
>>
>> git://git.samba.org/abartlet/samba.git replmetadata-sort-minimal
>> https://git.samba.org/?p=abartlet/samba.git/.git;a=shortlog;h=refs/he
>> ad
>> s/replmetadata-sort-minimal
>>
>> (On a positive note, while dbcheck is left flapping, it is at least
>> tested in the dbcheck-oldrelease tests, so the changes are still 
>> being
>> tested). 
> 
> Can I please get a review of these changes?
> 
> As users move from Samba 4.0 and 4.1 due to moving out of security /
> maintenance support, fixing this will only become more critical.
> 
> Remember that without this patch and running dbcheck on the existing
> databases, it is not possible to replicate older versions to a modern
> Samba version.

Pushed a bit reordered and with more BUG: markers

metze

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 198 bytes
Desc: OpenPGP digital signature
URL: <http://lists.samba.org/pipermail/samba-technical/attachments/20150824/ffa36225/signature.sig>