[Cellar] Re: [PATCH] ffv1: Support CRC with initial -1 and final -1 values

[Steve Lhomme <[e-mail]>]

On 26 Sep 2024, at 21:37, Michael Niedermayer <[e-mail]> wrote:
> 
> Hi Jerome
> 
> On Thu, Sep 26, 2024 at 11:12:58AM +0200, Jerome Martinez wrote:
>> On 26/09/2024 02:39, Michael Niedermayer wrote:
>>> Does someone want an option like a SHA-256 or something "stronger" than a CRC ?
>> 
>> In my opinion a CRC is enough because the need is to catch a corruption and
>> not to authenticate, and 1/1^32 per slice (Matroska has also CRC but only
>> per slice) seems a good balance to me (overhead vs usefulness)
> 
> yes, i agree
> 
> 
>> 
>>> or some kind of error correction code, but iam not sure this is the right place for
>>> that
>> 
>> I personally have a a PoC for error correction code at the container level
>> (Matroska). I think that it is a better place for ECC because the whole
>> content, including the container, would have to be repaired.
> 
> yes, container or some packaging level further outside
> like what a zip or tar file with mkv inside would be
> 
> I also wanted to work on some error correction but like so many things
> i had not found the time for that since the last time i said i wanted ;(
> But this wont conflict with you because my plans are more toward a container
> indepedannt solution

For the record, we did consider having some error correction in Matroska. But eventually it was ruled out. Some of the cons is that it makes the files much bigger and it depends on what kind of errors you accept. So it’s more tied to the physical layer (or transfer) of the data, than the data themselves. IMO it should be handled separately from files.