[PATCH v4 3/3] mfplat: Add MFCreateFile stub.

[Zebediah Figura]

On 11/2/18 5:34 PM, Sven Baars wrote:
> On 02-11-18 21:55, Zebediah Figura wrote:
>> On 11/2/18 3:27 PM, Sven Baars wrote:
>>> Signed-off-by: Sven Baars <sven.wine at gmail.com>
>>> ---
>>> v3: Actually read the file and handle all flags.
>>> v4: Rebased on the previous patch.
>>>
>>>    dlls/mfplat/main.c            |  78 ++++++++++++++++++++++-
>>>    dlls/mfplat/mfplat.spec       |   2 +-
>>>    dlls/mfplat/tests/Makefile.in |   2 +
>>>    dlls/mfplat/tests/mfplat.c    | 114 ++++++++++++++++++++++++++++++++++
>>>    dlls/mfplat/tests/resource.rc |  24 +++++++
>>>    dlls/mfplat/tests/test.mp4    | Bin 0 -> 1554 bytes
>>>    include/mfapi.h               |   2 +
>>>    7 files changed, 220 insertions(+), 2 deletions(-)
>>>    create mode 100644 dlls/mfplat/tests/resource.rc
>>>    create mode 100644 dlls/mfplat/tests/test.mp4
>>>
>>> diff --git a/dlls/mfplat/main.c b/dlls/mfplat/main.c
>>> index 7865a8935f..130597e8e7 100644
>>> --- a/dlls/mfplat/main.c
>>> +++ b/dlls/mfplat/main.c
>>> @@ -850,6 +850,8 @@ typedef struct _mfbytestream
>>>        mfattributes attributes;
>>>        IMFByteStream IMFByteStream_iface;
>>>        IMFAttributes IMFAttributes_iface;
>>> +
>>> +    HANDLE file;
>>>    } mfbytestream;
>>>      static inline mfbytestream *impl_from_IMFByteStream(IMFByteStream
>>> *iface)
>>
>> This strikes me as a little janky. Arguably it could make sense, but
>> I'd like to get a sense of how MFCreateMFByteStreamOnStream would fit
>> in with this (plus any other APIs that create byte stream objects).
>> Taking a less cursory look at the API of IMFByteStream, my intuition
>> leads me to think it would make more sense just to create separate
>> objects; there doesn't seem to be a lot of room for shared code.
>>
>>
> Hi Zebediah,
> 
> I thought the same, but a lot may be shared (at least the IMFAttributes
> interface that I added), and for now, not doing this would mean I have
> to duplicate another 500 lines of code. Maybe it would be better if the
> person who actually implements MFCreateMFByteStreamOnStream duplicates
> this code? I'm happy to copy-paste everything if you think that's better.
> 
> Sven
> 
> 

Even if there's a lot of shared code, it may be better just to use 
common helper functions. Anyway, besides possibly 
IMFAsyncCallback/IMFAsyncResult handling, it looks like most of 
IMFByteStream is a thin veneer over specific I/O methods.

Besides, with regard to IMFAttributes, as I think over it a bit 
more—since here the interface in question doesn't actually derive from 
IMFAttributes, I think we could simplify the code considerably by using 
existing mfattributes methods in the vtbl directly, with the exception 
of IUnknown methods.