CDDL+GPL and stuff

[Konstantin Svist]

Arnoud Engelfriet wrote:
> Konstantin Svist wrote:
>   
>> So you're saying the question is not whether someone distributes a 
>> binary that includes both module and kernel already linked, but the mere 
>> fact that the module is written so as to link to the kernel?
>>     
>
> No, I'm saying that you have to look at whether the module
> qualifies as a "derivative work" under copyright law. The mechanism
> used for interaction with the pre-existing work (the kernel) is
> not the criterion for that, although it can be an indication.
>
>   
>> In other words, "linking" does not refer to the actual process by which 
>> a binary program is built, but rather the process of writing code that 
>> will be able to "link".
>>     
>
> The question is, did you write the code independently or did
> you base yourself on kernel code? In the latter case, the code
> is a derivative work of the kernel code *regardless* of how the
> two interact.
>
> For example, if I copy&paste some code from the kernel in my own
> program which I sell separately, it's a derivative work even though
> no linking is going on.
>
> If I take an old UNIX device driver that surprisingly happens to work 
> without modification under Linux, it's not a derivative work
> even though it links. 
>   

Well, it's pretty easy to argue about corner cases - but what about ZFS?
It was not written for Linux specifically, but it doesn't "link" to it, 
at least not as-is. Does altering ZFS code to make it link with Linux 
kernel violate the GPL?
Or is it just a concern that it's a gray area?
What about using a layer - a small piece of GPL/LGPL code that will be 
the only thing that links to the Kernel directly - and on its other 
side, to existing ZFS code? (I think I heard that that's what ATI/nVidia 
do with their proprietary drivers, by the way)



Thanks,
Konstantin