Subject: Re: Re: Question Regarding GPL
From: Ben Tilly <btilly@gmail.com>
Date: Fri, 20 Jan 2006 20:50:09 -0800

 Fri, 20 Jan 2006 20:50:09 -0800
On 1/20/06, Rick Moen <rick@linuxmafia.com> wrote:
> Quoting Ben Tilly (btilly@gmail.com):
>
> > I'm not sure whether you're agreeing or disagreeing with me.
>
> Welcome to the law.  ;->

Is this where I break out my lawyer jokes?

[...]
> > Many people have released drivers that depend on that opinion being
> > correct.
>
> Note:  Torvalds's opinions on the matter have been known to change
> dramatically, without advance notice.  Compare his 1995-12-17 and
> 2002-10-17 proclamations on LKML, as archived here:  "Proprietary Kernel
> Modules" on http://linuxmafia.com/kb/Kernel/

On the surface that's that looks like a pretty big jump.  Though when
you squint at it sideways, he's at least somewhat consistent.  The
following statements appeared in some form in both opinions.

- It should be possible to port a non-Linux driver to work with Linux
without making it GPLed.

- He doesn't think people should write Linux-specific drivers and not GPL them.

- He doesn't like non-GPLed drivers.

- Linux-specific drivers are arguably derivative works of Linux.

- He sees the original limited kernel export table as providing a
plausible copyright barrier.

- Maintaining a binary-only Linux driver is very inconvenient.

When I look closely, the only actual inconsistency is that in the
first one he talks about how the structure of the kernel module
interface makes it possible to avoid copyright infringement, while in
the second he says that the kernel module interface was never
documented or meant for that purpose.  You can actually make him
completely consistent if you add to his later statement the
understanding that by "module interface" he means "current module
interface".  This is not entirely implausible since he does draw a
distinction in the 2002 opinion between the original and the current
interfaces.

Perhaps I'm just on a roll because I can vindicate a personal hero
:-), but a vague memory  that there was a significant factual change
there - his 1995 description of the kernel module interface would not
be an accurate description of the 2002 kernel module interface.  As
the facts change, his description of the consequences of the facts
should change.

However I did have to stretch a bit to make him consistent.  But when
I compare my understanding on key issues between now and 7 years ago,
I could forgive him a lot bigger inconsistencies than I had then.

But while I see the inconsistency, I'd have to disagree with your
characterization that Linus' opinions changed drastically.  The
emphasis was different in both cases, but that was a pretty long list
of very specific points that he made the same way both times.  And ot
wasn't an accident that he made those points.  For instance I just
googled and found the 1998 interview at
http://linuxgazette.net/issue32/rubini.html.  He again quickly hits
all of those points except the last two technical ones.  (Given the
forum, omitting technical details makes sense.)  I'd therefore call
him very consistent in all of the points that I listed.

> > What you've said is that the correct test to use is the one described
> > in that decision.  I just read that decision, and I am left no more
> > able to answer the fundamental question than I was before.
>
> My old landlord and colleague Richard Couture had a saying -- harsh,
> but relevant:  "Sorry to hear about  your  problem."  ;->  The relevant
> test is the one that will be applied.  I'm just telling the truth.

That's like answering a kid who said, "When will I see the house" with
a long treatise on optics.  Everything that you said is true,
relevant, and useless. :-)

> [Micro Star decision:]
>
> > I don't see the relevance.  U
>
> Relevance is that non-literal copying can infringe, and that
> copyright-encumbered content isn't necessarily limited to code.

Ah.  Which makes it much harder to write loadable modules without infringing.

Cheers,
Ben