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

[David Chisnall via llvm-dev]

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.  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++).

>> 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.  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.

David