On Sat, Dec 12, 2015 at 10:39 AM, Andreas Cadhalpun <andreas.cadhal...@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 _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
