Ian Murdock's recent piece on the whole Debian/Nexenta controversy contained what is an almost-universal misunderstanding (so common, in fact, that I held the same misunderstanding myself until fairly recently) about what the GPL actually says about linking in other software with a GPL-licenced work. Ian writes:
It seems to me the argument that linking a GPL application to a CDDL library and asserting that that somehow makes the library a derivative work of the application is, to say the least, a stretch&emdash;not to mention the fact that we're talking about libc here, a library with a highly standard interface that's been implemented any number of times and, heck, that's even older than the GPL itself. It's interpretations like this, folks, that give the GPL its reputation of being viral, and I know how much Richard Stallman hates that word.
The core misunderstanding there is that the GPL somehow claims that libraries linked into a GPL-licenced binary are derived works. Nothing in the text of the GPL states that, and a quick scan of the GPL FAQ doesn't provide any opinions of that nature, either.
What the GPL does state are a few conditions which you have to abide by in order to be given permission by the copyright holder to exercise rights ordinarily held exclusively by said copyright holder. One of those is that, in order to be given permission to distribute a compiled binary, you must be able to provide source code for the binary itself, as well as all interface definition files, build scripts, and modules contained in the binary, and that all of this material must be provided under terms no more restrictive than those of the GPL.
The GPL isn't trying to lay a "derived work" claim on any of those other parts. It's just saying to you, as someone who wants to distribute a binary, "here are some other things that must be OK if you want to do this". A similar clause exists with regards to patents and court judgments (section 7, for those of you playing along at home). This doesn't mean that the GPL is attempting to claim that the patents or court judgments are derived works. A licence could also say "you must delete all copies of any Microsoft product from your hard drive before duplicating this software", but that wouldn't mean that the licence is attempting to claim that all Microsoft products are derived works of the licenced software!
As anyone who's looked at a typical proprietary software EULA probably already knows, a software licence can claim pretty much anything (and, apparently, thanks to the DMCA, in the US they can make it stick). The licence doesn't make the claim on the basis of derivation, it makes it because that's just what the licence claims. The GPL is no different in that respect. If you can't fulfill the licence conditions (either because you don't want to, or because -- as in the case of the clause under discussion -- the module you are including in your binary can't be licenced under terms equivalent to the GPL) then you can't do whatever it is that the licence would otherwise allow you to do.
To be fair to Ian, some of the comments he made to his blog post indicate that he does understand the distinction between a claim due to derivation and one due to a simple licence term, but I thought it worth making the point crystal clear to everyone out there in the blogsphere, because it's a very common misunderstanding, and I'd hate for it to spread any further than it has already.