ntdll/tests: Dump the handle table on failure.

[Sebastian Lackner]

On 18.08.2016 10:26, Huw Davies wrote:
> On 17 Aug 2016, at 15:07, Sebastian Lackner <sebastian at fds-team.de> wrote:
>>
>> On 17.08.2016 16:01, Huw Davies wrote:
>>>
>>>> On 17 Aug 2016, at 14:54, Sebastian Lackner <sebastian at fds-team.de> wrote:
>>>>
>>>> I assume this is for bug 40106, right? I also investigated it back then,
>>>> and it looked like a kernel bug to me. If I remember correctly it was
>>>> somehow triggered by a large number of allocated handles.
>>>
>>> Yes, I was wondering whether it was a simple as needing to compare
>>> the LOWORD of the handle.
>> I think I tried that back then, and it was not sufficient. Afaik the whole
>> structure was somehow messed up (including other fields not included in your
>> trace yet) and I suspected that it was returning random kernel memory.
>> I am not sure about all details anymore, and maybe I also made just a stupid
>> mistake while testing. Nevertheless, thanks for your help, maybe you can find
>> out more. ;)
> 
> It turns out that both Vista and 2008 only copy the first handle entry into
> the buffer, so if the required handle isn’t first in the list the tests would fail.
> Patch sent to detect and win_skip this behaviour.
> 
> Huw.
> 

I thought I had tested this aswell, so I'm a bit surprised that the solution is
that easy. Nevertheless, thanks for working on this. ;)

Regards,
Sebastian