On 15/04/14 15:57, Roman Savchenko wrote:
> Hi lu,

Please don't top post, if possible reply inline as I'm doing here.

> for my opinion it easier to provide deep copy for all data it need, with
> serialization/deserialization for huge structs/arrays, it will not break
> any api and will not need to make changes in other code.

You basically have a scaled down frame repeated twice.

> But in elegant way. in my opinion, its to changes pair (uint8_t *buff, int
> size) to AVBufferRef*. And provide api, that allow to chose: create new
> buf_ref (with ref counting 1) or "ref" buf_ref. But in this way need some
> guarantees, such as user can not change internal buffer of AVBufferRef,
>  live time of AVBufferRef will be correct, and no race conditions. And its
> required a lot of changes in code.

av_buffer_* functions are available.

Probably we could provide this information as secondary frame (see
https://wiki.libav.org/Blueprint/MultiAVFrame)

lu
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to