[llvm-dev] RFC #2: Improving license & patent issues in the LLVM community

[David Chisnall via llvm-dev]

On 3 Nov 2016, at 17:11, Daniel Berlin <dberlin at dberlin.org> wrote:
> On Thu, Nov 3, 2016 at 9:13 AM, David Chisnall <David.Chisnall at cl.cam.ac.uk> wrote:
> On 3 Nov 2016, at 15:55, Daniel Berlin <dberlin at dberlin.org> wrote:
> >
> >> I and the other libc++ contributors were all happy to have our code relicensed under the MIT license (or contribute it under those terms originally), so there is some pretty clear evidence that we explicitly did *not* require this attribution.
> >
> > That license does, in fact, require attribution:
> > "The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. “
> 
> Again, unless I’m misreading this, there is no requirement that code that merely uses libc++ headers include a libc++ attribution. 
> 
> You are misreading this :)
> 
> I can tell you it's not even a consensus, i have not found a single lawyer who believes otherwise.
> 
> I can forward you the discussion that occurred around this if you like (i have to acquire permission from the folks involved first). There was 100% universal agreement on this.
>  
> That was the explicit requirement for the switch to the MIT license.  The proposed new license would mean that this is only true for code compiled with LLVM.
> 
> We must include the copyright and attribution when we distribute FreeBSD, because we include the whole of libc++.  We do not need to separately include it in programs that simply link to libc++ (irrespective of how they were compiled) in packages, because they do not contain ‘substantial portions’ of the code (probably - it’s a bit of a grey area with respect to template-heavy C++).
> 
> This is not correct ;)
>  
> 
> >> Indeed, we relicensed under the MIT license (after long discussion) specifically to avoid this requirement.
> >
> > If we did, we screwed it up then. We probably meant the zlib license, which does not require attribution  ;)
> 
> Assuming that your interpretation of this clause - which is the first time I have ever heard it interpreted to include either linking or including headers - is correct. 
> 
> This is interesting, as i've said, i've literally never heard it interpreted otherwise.
>  
> I’d be quite surprised if it were, as the lawyers that reviewed it at the time were happy that it did not impose these constraints, but it’s possible.
> 
> Like I said, there was discussion of it on the open source counsel mailing lists i belong to, and there was universal agreement that it required attribution.
> 

If your interpretation is correct, then almost every graphical X11 application is infringing the license with respect to libX, the authors of the license were incorrect as to what it does, every project that I’ve been involved with that has used it has adopted it assuming that it means something different, and every lawyer that I’ve spoken to at companies adopting such code has been wrong.  This may be the case, but you’ll forgive me if I take your opinion with a grain of salt here.

The LLVM project's intention on moving libc++ and compiler-rt to the MIT license was that no code that simply linked to unmodified versions of these libraries would be required to contain attribution.  A quick look on opensource.org points to TL;DRLegal, which concurs with this: you must include the copyright notice in a copy of the library, but you are not required to include any attribution.  In fact, a quick skim of similar sites includes the same consensus, so I’m not sure who is on this mailing list that you allude to, but they seem to have some quite different opinions to the normal consensus.

David