Re: [tlug] Proprietary code linked with GPL code

[Edward Middleton <edwardmiddleton@example.com>]

On Fri, 2005-12-16 at 14:57 +0900, Stephen J. Turnbull wrote:
> >>>>> "Edward" == Edward Middleton <edwardmiddleton@example.com> writes:
> 
>     Edward> My understanding is that if the GPL kernel used a non-GPL
>     Edward> standard interface
> 
> Interfaces cannot be copyrighted, because they are not expressive
> works.  You can prevent linking to those interfaces by your license
> for the work that exposes them, but you can't license the interface
> itself, anymore than you can copyright and license the English
> language.

OK put it a more long winded way. If it is an interface that is specific
to a GPL kernel then it would be undeniable that it was written
explicitly to interface with that kernel and as such is a derivative
work.  If it uses a generic interface that is supported by a non GPL
kernel then it is arguable that the driver was not derived from the GPL
kernel.

>     Edward> then a module implementing that interface would not be
>     Edward> derivative of that kernel.  If the driver had to be
>     Edward> written for the specific kernel rather then for a standard
>     Edward> interface then it would be a derivative.  This is one of
>     Edward> two reasons why a number of companies want a standard
>     Edward> binary interface for Linux.
> 
> This only matters if the ABI has non-GPL implementations.
> 
> However, if the only kernels it works with are GPL-licensed, then the
> FSF's position is that standardizing the ABI is irrelevant.  It runs
> in the same memory space, it's linked, so it's a derivative.

Which is a good general rule, but it is possible to run windows drivers
under Linux but that does not make them a derivative work of the Linux
kernel.

>     Edward> I would contend that this doesn't apply to the Linux
>     Edward> kernel either, the issue is whether it is a derivative
>     Edward> work, linking is just a test for determining that a work
>     Edward> is derivative.
> 
> Which is exactly what _I_ said!  It's a little bit slippery whether
> loadable modules are ipso facto part of a derivative work, but my
> position is that until they are shown to be linkable both logically
> and legally to some other product, they are a derivative of the
> kernel.
> 
> My claim is that the license Linus intended for the kernel is like the
> LGPL; it excludes certain kinds of linkage from the GPL obligations.
> That's why I asked whether the documents exist or not, as existence
> would prove that the GPL is modified in the case of Linux licensing.

Well I guess if distributors of proprietary binary drivers can prove
this in court then you are correct.  But Cisco couldn't

-- 
Edward Middleton <edwardmiddleton@example.com>