[FFmpeg-devel] [PATCH] fate: Add ffhash as an md5 alternative

[James Almer]

On 07/06/13 6:19 PM, Reimar Döffinger wrote:
> On 07.06.2013, at 22:22, James Almer <jamrial at gmail.com> wrote:
>> On 07/06/13 4:27 PM, Michael Niedermayer wrote:
>>> On Fri, Jun 07, 2013 at 05:11:47AM -0300, James Almer wrote:
>>>> libavutil's MD5 implementation is still slower than openssl's, so
>>>> give the latter priority.
>>>
>>> doesnt work
>>>
>>> make distclean ; ./configure ; ...
>>> ffhash isnt build before use
>>
>> That's why there's a check to see if it's available, same as the other programs.
>> This is meant as a last resort in the unlikely case no other system md5 program
>> is available, but the user compiled ffhash with make alltools or similar.
> 
> But wouldn't it make sense to add it to the dependencies in the same way as ffmpeg for example?
> That would ensure it exists, though in case of cross-compilation it still won't actually work though if run on the host (I actually have no idea where it will be run).

Now that you mention cross-compilation, it seems that whichever program is run from 
md5.sh is run using $(TARGET_EXEC) if it was set.
Is that intended, considering it right now looks for system tools only?

Nonetheless, adding ffhash as a dependency for tests that use do_md5sum() would in 
99.99% of the cases build but not use it, since it would be the last program in a list 
of five that are checked for availability.

Michael submitted a patch to make use of (back then called) ffadler in md5.sh.
We could update it to use ffhash (with either adler or murmur3), which afaik was the 
whole point for writing ffadler in the first place.

Regards.