[Samba] NT_STATUS_OBJECT_NAME_NOT_FOUND after installing the last git version

[Jelmer Vernooij]

On Tue, Feb 18, 2014 at 10:49:28AM +0100, Kai Blin wrote:
> On 2014-02-18 10:18, Volker Lendecke wrote:
> >On Tue, Feb 18, 2014 at 09:13:35PM +1300, Andrew Bartlett wrote:
> >>On Mon, 2014-02-17 at 23:21 +1100, Alain Foucher wrote:
> >>>Did i miss some issues ?
> >>
> >>Thank you very much for bringing this to our attention.
> >
> >Attached find a patchset that reverts the offending code. As
> >I don't have the time today to deal with this in an
> >appropriate manner, please review & push this until we have
> >a resolution for the problem this code causes.
> 
> I'd -1 these patches, but Andreas beat me to the review and +1ed them.
> 
> The main questions raised by this issue for me are:
> * Auth code is tricky, but it does have a lot of tests. Why didn't
> they catch this
> * The last time I changed this code to support unixid-mappings
> instead of plain uid/gid-specific mappings, I made fairly similar
> changes. I still didn't catch the problem with the patches. Can the
> auth code be simplified so the implications of changing it are
> better understood?
These seem like good questions to ask, regardless of whether the code landed.

> And last but not least a more personal question about code reviews
> and reverts like this: How can we resolve things like that without
> leaving the reviewer feeling like he wasted a lot of time on a
> review? I feel silly enough not having caught the problem, but now I
> also feel like I could have just spend the time I spent on the
> review drawing fluffy pink unicorns and be more productive that way.
> 
> I for one don't feel like spending more time trying to follow up on
> this issue now that the code is on the way out of master anyway. I
> also feel like I'll rather practice my unicorn drawing skills than
> reviewing code I can't care about enough to drop everything I'm
> doing and debug potential issues before the 10-minutes-to-revert
> window has passed.
What is this magical 10-minutes-to-revert window? I think you/we should be
free to revert any commits that introduce regressions (provided there are
no other side-effects of doing so), regardless of when they were
originally landed.

Cheers,

Jelmer