[FFmpeg-devel] [PATCH v2] avcodec/nvenc: Include NVENC SDK header

[Hendrik Leppkes]

On Sat, Dec 12, 2015 at 10:39 AM, Andreas Cadhalpun
<andreas.cadhalpun at googlemail.com> wrote:
> On 12.12.2015 01:46, Philip Langdale wrote:
>> On 2015-12-12 00:03, Andreas Cadhalpun wrote:
>>> On 11.12.2015 09:41, Carl Eugen Hoyos wrote:
>>>> My point is that so far several people have said that if nvenc
>>>> is a system library there is no issue (and I fully agree). I
>>>> didn't see a mail (and even less a patch with a commit message
>>>> that says so) that claims nvenc is a system library (only that
>>>> it "should" be one).
>>>
>>> So let's ask: Is someone here who claims that nvenc is a system
>>> library and can explain why?
>>
>> I'm not going to claim it's a system library.
>
> Interesting...
>
>> I'm, instead, going to
>> ask why we're having this conversation about nvenc, when the qsx/mfx
>> situation is exactly the same.
>
> We have this conversation, because someone sent a patch to enable it
> by default, together with including the header and removing the
> 'die_license_disabled nonfree nvenc' line.
>
>> The functionality is provided by a
>> proprietary set of modules that are part of the intel driver on windows
>> and a separate (almost undiscoverable) download on linux (actually,
>> that's worse than nvenc where the functionality is shipped with the
>> driver in both cases). The only structural difference is that ffmpeg
>> links against a wrapper library for mfx and dlopens in the nvenc
>> case, but because of your following statement, that cannot make any
>> difference.
>
> Since this requires the mfx wrapper to link, it is not enabled by
> default. As the license situation seems similar, it might be a good idea
> to add a 'die_license_disabled nonfree libmfx' line. But these don't have
> any effect on the legal situation anyway, they are just a help
> for our users.
>
>>>> I am glad we agree that there is no difference (license-wise) if
>>>> a library is linked statically, dynamically or via dynamic
>>>> loading;-)
>>>
>>> There is that, at least. ;-)
>>
>> Oh, and do you know what's funny - I just realised that the primary ffmpeg
>> code base is LGPL and not GPL, so this whole conversation is slighlty
>> pointless.
>
> No, it's not, because the LGPL and GPL are very similar in terms of the
> requirements about distributing object code of (L)GPL-ed source code.
>
>> Combining ffmpeg with proprietary libraries is covered under section 6 and
>> section 7,
>
> These sections only cover "work that uses the Library" (defined in section 5),
> not the Library itself.
>
>> so even if building the nvenc codec is considered to combine
>> ffmpeg with nvenc in this sense, it would be acceptable. The key requirement
>> is that the LGPL covered parts can be rebuilt and modified as desired, and
>> that is certainly true.
>>
>> These sections are generally thought of as enabling a larger proprietary
>> program to pull in an LGPL library, but the language is symmetric.
>
> No, see section 4:
> "You may copy and distribute the Library (or a portion or derivative of it,
> under Section 2) in object code or executable form under the terms of Sections
> 1 and 2 above provided that you accompany it with the complete corresponding
> machine-readable source code"
>
>> Note that I actually don't believe with have a GPL problem here,
>
> Why?
>
>> but as a step forward, if we can all agree that the nvenc codec is a valid
>> part of an lgpl build of ffmpeg, that's a step forward.
>
> I don't agree with that interpretation, see above explanation.
>

We should just add an exception into the license to explicitly allow
using it with the NVIDIA CUDA library and be done with this debate for
ever.
You know that Open-Source has failed when the project itself is
arguing days and days for including a feature on license reasons that
any closed-source app would just write, enable and offer to its users
without a second thought.

- Hendrik