[PATCH v2 1/6] ntoskrnl.exe/tests: Silent a todo_wine failing test after a while.

[Rémi Bernon]

On 6/3/21 5:44 PM, Zebediah Figura (she/her) wrote:
> On 6/3/21 10:42 AM, Rémi Bernon wrote:
>> On 6/3/21 5:40 PM, Zebediah Figura (she/her) wrote:
>>> On 6/3/21 7:55 AM, Rémi Bernon wrote:
>>>> Signed-off-by: Rémi Bernon <rbernon at codeweavers.com>
>>>> ---
>>>>
>>>> v2: * Fix spurious failures first.
>>>>
>>>>       * Don't try NULL preparsed data, it crashes old windows versions.
>>>>
>>>>       * Squash new HidP_Get*Caps tests together, add a few more with
>>>>         specific caps lookup.
>>>>
>>>>    dlls/ntoskrnl.exe/tests/driver_hid.c | 3 ++-
>>>>    1 file changed, 2 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/dlls/ntoskrnl.exe/tests/driver_hid.c
>>>> b/dlls/ntoskrnl.exe/tests/driver_hid.c
>>>> index 44ec1b09526..79b3e708752 100644
>>>> --- a/dlls/ntoskrnl.exe/tests/driver_hid.c
>>>> +++ b/dlls/ntoskrnl.exe/tests/driver_hid.c
>>>> @@ -103,6 +103,7 @@ static const unsigned char report_descriptor[] =
>>>>    static NTSTATUS WINAPI driver_internal_ioctl(DEVICE_OBJECT *device,
>>>> IRP *irp)
>>>>    {
>>>> +    static int test_failed = 0;
>>>>        IO_STACK_LOCATION *stack = IoGetCurrentIrpStackLocation(irp);
>>>>        const ULONG in_size =
>>>> stack->Parameters.DeviceIoControl.InputBufferLength;
>>>>        const ULONG out_size =
>>>> stack->Parameters.DeviceIoControl.OutputBufferLength;
>>>> @@ -176,7 +177,7 @@ static NTSTATUS WINAPI
>>>> driver_internal_ioctl(DEVICE_OBJECT *device, IRP *irp)
>>>>            case IOCTL_HID_READ_REPORT:
>>>>                ok(!in_size, "got input size %u\n", in_size);
>>>> -            todo_wine ok(out_size == 2, "got output size %u\n",
>>>> out_size);
>>>> +            if (test_failed++ < 10) todo_wine ok(out_size == 2, "got
>>>> output size %u\n", out_size);
>>>>                ret = STATUS_NOT_IMPLEMENTED;
>>>>                break;
>>>>
>>>
>>> Does this test fail on Windows? Can you provide a link to the test 
>>> results?
>>>
>>
>> It's not failing on windows, but the todo_wine are spamming the test
>> results, making the test be reported as failing because too much data is
>> printed.
>>
> 
> Shame on me for introducing that test and never fixing it, but maybe we 
> could just fix it the right way instead?
> 

Yes but it's a little bit more complicated than it seems, and I'm a bit 
afraid of the possible unforeseen consequences.

There's apparently several ways to request reports (in polled and 
non-polled modes for instance). and I couldn't test the 
IOCTL_HID_GET_INPUT_REPORT way yet for instance, I don't know if it 
passes a packet with the full report length, including 0 report id or if 
it uses a different length depending on when report ids are used or not.

Also in winebus.sys, both ways are handled by deliver_last_report, which 
expects a full report.

Then there's the question whether reading hidraw devices give us full 
reports or not, which I haven't investigated yet.

I'm perfectly fine with leaving the test unfixed for now, and have the 
spurious failures, it just looked bad for my additional tests ;)
-- 
Rémi Bernon <rbernon at codeweavers.com>