[PATCH] winegstreamer: Fixup raw audio caps to be compatible with IMFMediaType.

[Zebediah Figura]

On 11/6/20 2:51 PM, Derek Lesho wrote:
> 
> On 11/6/20 2:32 PM, Zebediah Figura wrote:
>> On 11/6/20 2:20 PM, Derek Lesho wrote:
>>> On 11/6/20 2:13 PM, Zebediah Figura wrote:
>>>
>>>> This is done in a rather inconsistent way relative to how video streams
>>>> are handled.
>>> Yes, because the goals are different for each of the paths.  The video
>>> path is just an enhancement to report video formats in a defined order
>>> as if they were coming from a decoder, since right now we're skipping
>>> the decoder MFT step.  The step for fixing up the audio caps is meant to
>>> be a generic solution for any caps which are un-representable as a
>>> IMFMediaType object.  This same path is used for compressed h.264 video
>>> on my local branch for example.
>>>
>> In both cases you're doing conversion from a type which may not be
>> representable into a type which is.
> No, in the case of the uncompressed video streams, the type is almost
> definitely re-presentable.  Think of it as a necessary hack for the
> bypassing of the decoder MFT we are doing.  On the other hand, there are
> plenty of cases where uncompressed audio may be read from a container,
> and the fixup path would still be necessary in those cases.

You can't assume that the type is representable. In fact, you should
make no assumptions about the type whatsoever. This is not only true in
theory, but in practice—I've seen decoders output GST_VIDEO_FORMAT_RGB.

>>   The reasons for doing this
>> conversion may be different, but there is no reason for the mechanism
>> to be.
> I would say there is, the conversion we're doing for the video streams
> is unconditional, entirely specific to the media source output, and
> doesn't output 1 fixed up caps structure per input caps structure.  On
> the other hand, the audio and compressed type format would be necessary
> any time we want to feed gstreamer buffers with those caps to a media
> foundation component, and is a 1 to 1 conversion in every case. 

I don't see any of those as reasons for the code structure to be
different. Note thought that even if they were, you probably want to
make it possible to convert even from some representable formats. Not
all systems can play back 64-bit float PCM, for example.

> Examples of other areas where this would be necessary, off the top of my
> head, would be a separate uncompressed-audio-emitting media source from,
> say, a microphone, or any MFT which outputs compressed video or raw
> audio, such an encoder MFT.
>>
>> Moreover, the goals are not entirely orthogonal; not all video will be
>> output in the four types you have listed.
> All video streams that take the videoconvert enumeration path
> (uncompressed video) won't need any transformation to align with an
> IMFMediaType object.  The only potential incompatibility would be the
> layout, but that problem would never surface with the current media
> source we are pretending that our output types have gone through a stand
> media foundation decoder.  An instance where we would want to put this
> type of code in the make_mf_compatible_caps path would be a media source
> that provides uncompressed video on windows, such as from a webcam or
> screen capture.  In the code for this media source, we'd unconditionally
> put the video caps through the make_mf_compatible_caps path, and add
> code there to replace any unsupported layout with its closest equivalent
> defined in media foundation.

"won't need any transformation" is only true because you're *already*
applying a transformation. This code path *is* the transformation. If
you did a similar thing with the audio stream, it "won't need any
transformation" either.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: OpenPGP digital signature
URL: <http://www.winehq.org/pipermail/wine-devel/attachments/20201106/fff57f08/attachment.sig>