[FFmpeg-devel] [PATCH 4/6 v2] avutil/mastering_display_metadata: add a new allocator function that returns a size

[James Almer]

On 3/27/2024 9:40 AM, Anton Khirnov wrote:
> Quoting James Almer (2024-03-27 13:35:35)
>> On 3/27/2024 4:41 AM, Anton Khirnov wrote:
>>> Quoting James Almer (2024-03-25 22:13:25)
>>>> On 3/25/2024 6:02 PM, Andreas Rheinhardt wrote:
>>>>> James Almer:
>>>>>> I don't mind a function like that being added to simplify future
>>>>>> additions, but this API is orthogonal to the frame side data one. It's
>>>>>> also used in packets, for example, and right now lavf is using
>>>>>> sizeof(AVMasteringDisplayMetadata) because
>>>>>> av_mastering_display_metadata_alloc() is not useful.
>>>>>>
>>>>>
>>>>> The API proposed by me is supposed to make API like
>>>>> av_mastering_display_metadata_alloc_size() redundant and therefore these
>>>>> two additions are not orthogonal.
>>>>
>>>> Just because there's a frame side data type for MDM does not make the
>>>> new alloc function redundant.
>>>
>>> The commit message says:
>>>
>>>> av_mastering_display_metadata_alloc() is not useful in scenarios where you need to
>>>> know the runtime size of AVMasteringDisplayMetadata.
>>>
>>> In what scenarios besides side data do you need to know the size of this
>>> struct?
>>
>> None within our libraries that i can think of, but library users can
>> have scenarios we need to provide for. MDM is a standalone API, so lets
>> not try to make its usability depend on a separate one.
>> I'm replacing a helper with a better one, it should not be so controversial.
> 
> Breaking API to benefit hypothetical use cases IS a controversial change
> in my view.

Then I can leave the old on in place. But don't insist on telling people 
to use a frame side data specific alloc function if they want a MDM 
struct, because said function will not be usable for all side data types 
anyway (see video enc params, iamf params), so it's just making things 
more complex for the user for no reason.