[PATCH 4/5] xmllite/tests: Add basic test structure for IXmlReader

[Paul Vriens]

On 01/21/2010 06:54 PM, Reece Dunn wrote:
> 2010/1/21 Nikolay Sivov<bunglehead at gmail.com>:
>> On 1/21/2010 20:41, Reece Dunn wrote:
>>>
>>> 2010/1/21 Nikolay Sivov<bunglehead at gmail.com>:
>>>
>>>>
>>>> On 1/21/2010 19:49, Paul Vriens wrote:
>>>>
>>>>>
>>>>> On 21 jan 2010, at 17:26, Nikolay Sivov<bunglehead at gmail.com>    wrote:
>>>>>
>>>>>
>>>>>>
>>>>>> On 1/21/2010 19:19, Paul Vriens wrote:
>>>>>>
>>>>>>>
>>>>>>> On 01/18/2010 10:28 PM, Nikolay Sivov wrote:
>>>>>>>
>>>>>>>>
>>>>>>>> Add basic test structure for IXmlReader
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>> Hi Nikolay,
>>>>>>>
>>>>>>> This one crashes on Vista without a servicepack at:
>>>>>>>
>>>>>>> +    IXmlReader_Release(reader);
>>>>>>> +}
>>>>>>>
>>>>>>> All the calls before that one look ok (they produce the expected
>>>>>>> output).
>>>>>>>
>>>>>>> I guess "vista without an SP" can be considered broken(), but how do
>>>>>>> deal with that?
>>>>>>>
>>>>>>>
>>>>>>
>>>>>> That's a bit strange. Could you test for reader pointer around this:
>>>>>> ---
>>>>>>    hr = pCreateXmlReader(&IID_IXmlReader, (LPVOID*)&reader, NULL);
>>>>>>    ok(hr == S_OK, "Expected S_OK, got %08x\n", hr);
>>>>>> ---
>>>>>> Add reader = 0xdeadbeef; and test what is reader after a call.
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>
>>>>> See https://winetestbot.geldorp.nl/JobDetails.pl?Key=386
>>>>>
>>>>
>>>> Hm. So it's created ok, at least looks so.
>>>>
>>>> I don't see how to fix this. We could remove this _Release of course, but
>>>> will leak.
>>>>
>>>
>>> Can't you do something like:
>>>
>>>     ULONG count;
>>>     count = IXmlReader_AddRef(reader);
>>>     ok (count == 2, "Expected AddRef to return 2, got: %d\n", count)
>>>     while (count>    0)
>>>        count = IXmlReader_Release(reader);
>>>
>>
>> Maybe. It's ugly of course, cause it's not clear at all why refcount could
>> be zero just after creation
>> with success code and not zero pointer.
>> Btw, could you try it?
>
> I don't have access to a Windows test machine to try it out, and don't
> have access to my home Linux dev box, so can't create a patch for the
> test bot at the moment.
>
> - Reece
>
>
Just checked and it doesn't help. The first release is fine but the 
second crashes again.

It's really broken on Vista with no SP.

-- 
Cheers,

Paul.