Elastic promises "open"—delivers proprietary [LWN.net]

Elastic promises "open"—delivers proprietary

Posted Feb 1, 2021 18:52 UTC (Mon) by jejb (subscriber, #6654)
In reply to: Elastic promises "open"—delivers proprietary by kemitchell
Parent article: Elastic promises "open"—delivers proprietary

> I asked about another source for this argument, because I suspect it was either FSF gloss or kernel lore. Both of those institutions have developed a "folk law" of GPLv2 interpretation that doesn't necessarily follow from its text.

I already told you I was present for the discussion: this isn't some received wisdom, it's something I remember happening. Likely we may have email archives from the time as well (although I'm not certain about how long ago it was, so finding them is going to be a pain) and I can bet intel is going to have preserved the records to make absolutely sure we have no claim requiring their open sourcing of icc.

This isn't just the kernel being different, it has been the FSF practice since the early days as well: the gnu tools were created and released under GPLv2 long before we had a usable gcc. The only way to build them in those days was with the proprietary tools on whatever unix platform you had. Most of those projects still have autoconf fragments that check for the proprietary compilers if you need written evidence.

> I don't think estoppel would apply so cleanly as you think. Especially if there's no written statement out there telling people they can share some parts of complete source code / Corresponding Source without licensing under GPL.

My understanding based on having a lawyer take me through the various estoppel doctrines for another matter is that if it becomes standard practice it may be relied on without a written promise. icc isn't the only proprietary toolchain that has been used to compile Linux: the practice is rife throughout the embedded space as well.

Plus, as I said before, the clear exclusion of identifiable sections of the corresponding source that aren't derived works of the program in section 2 means that to require the scripts and the compiler to be under GPLv2 you'll have a way to prove in court that they're derived works of the Program the licence was protecting.

to post comments

Elastic promises "open"—delivers proprietary

Posted Feb 1, 2021 19:22 UTC (Mon) by kemitchell (subscriber, #124442) [Link]

I don't doubt this has come up before, or that folks involved at the time came to an understanding. But I don't think that understanding, whatever it was, flows inexorably from the plain text of GPLv2.

Nothing strange about that idea. Both FSF and the kernel have developed, and often shared, lots of understandings and expectations about how the license applies to their work. Some of that ends up documented, like the kernel user-space exception or the FSF GPL FAQs. Some of it's scattered around in mailing list archives and between devs' ears.

I'm interested in the written record of the conversation around this because that would give me a sense of the arguments deployed, as well as who, and which projects, might be held to that interpretation. If it's an FSF thing, it may not apply to the kernel, and vice versa. Or it may be the accepted norm for both FSF projects and the kernel, but not necessarily for other GPLv2-licensed projects.

It's a tangent, but on your note about build tools: Especially GPLv3 spilled a bit of ink to catch scripts, but not tools they invoke, like compilers. Which, as you point out, weren't free early on in free software history. Ref:

However, it [Corresponding Source] does not include the work’s System Libraries, or general-purpose tools or generally available free programs which are used unmodified in performing those activities but which are not part of the work.

One of the complaints against SSPLv1 is that it threatens to reach the kernel and other system-level utilities, because it doesn't have this kind of limitation in section 13. Mongo sent a proposed revision of SSPL to the OSI mailing list that partially addressed those concerns. I suspect they initially left it out because they'd seen it was possible to argue that the definition of "Corresponding Source" in AGPLv3 "insulates" containerized services from the reach of copyleft. It's analogous to the (somewhat dubious) idea that you can "wrap" GPL code, and then invoke the wrapper in a way that avoids triggering copyleft.