Looking closer at the GPL, it seems like most requirements only kick in once you "convey" GPL-covered code. If you make your users get the GPL component themselves from a 3rd party (e.g. PyPI or other package repository), then you might be okay. I'd be curious for input from others, but it seems like the following flow avoids GPL virality by avoiding "conveying" the GPL-covered code to the end-user:

1. You give your user a non-GPL python package with requirements.txt file (no bundled dependencies)

2. Your user pip-installs the dependencies (including some GPL-licensed ones)

3. Your user runs the application

As long as your country doesn't consider use of an API prohibited under the copyright of the implementing code, I think steps 1-3 would be fine (though not very practical for a product).

I'd be curious for others input, though, as this has bugged be for a while in the R community where several core libraries (like the Matrix package) are GPL licensed but many packages that depend on GPL packages claim to be licensed under MIT or some other license.

The end result will most probably violate the GPL. However, this is only realized by each user; you could probably argue that you never test your software and only look at API documentation.

And then it becomes a problem of proving your users violating the GPL. So you'd have to go after each one of them, which will be incredibly difficult, and proving damages would be even more difficult.

It's an asshole way of exploiting "Wo kein Kläger, da kein Richter" (where's no plaintiff, there's no judge) since actually proving that the developers violated the GPL will be difficult, unless they have a CI system that readily documents this.

As a pure copyright license the GPL concerns itself with distribution only. You cannot violate the GPL as an end user without distributing something.

Yes, all of the above is on the condition that some distribution happened (and you can prove that).

However, distribution also happens in places you might not expect. As a business, I'd stray far away from such constructs even if I only use this construct internally.

However, this is purely based on the wording of the GPL. For example, the EUPL explicitly covers the creation of derivative works - and I'd argue that the proposed circumvention would create a derivative work.

Yeah, the crux of the issue would definitely be whether use of an API is prohibited by default under copyright law for a country (i.e. does using a library make something a derivative work of the library). In the US, at least, the Google v Oracle case makes me think this is worst case fair use (for many contexts) and best case too functional to be covered by copyright in the first place.

Though I can certainly imagine that a multinational company might not be confident of the copyright status of API usage in all countries they operate in.

I'd further argue that it would be important if your program does anything useful without the GPL parts; if so, you can probably argue that it is not a derivative work. If you however only build an extension to some GPL'ed software that fundamentally needs the GPL'ed code to properly function and cannot (easily) be used by any other software, then you probably will create a derivative work.

To be pedantic, you can't actually violate the GPL. You infringe on the owner's copyrights if your actions don't fall in the GPL's allowed actions.

Yeah for interpreted languages this may be OK. For compiled languages it becomes harder to not "convey" anything derived from GPL code you linked against. The GPL even explicitly exempts system components for this reason.

Even for compiled languages, you can get around this with a properly architected plugin architecture. Your core project (non-GPL) exposes a runtime plugin interface. It has no clue what plugins will be used, but provides all the operations needed for the plugin to do what it needs. Create a plugin which links to the GPL code. GPL the plugin itself. The user can then be directed to install the GPL plugin. I can’t see how this would be a violation (by the core developer) of the GPL, as the core doesn’t have the first clue what license its plugins are using, and indeed, multiple plugins could be used that have contradictory licenses.

I’m trivializing here but for sake of brevity:

LGPL allows dynamic linkage without open sourcing your components.

GPL requires any linkage to open source your parts, but allows for process boundary to not be viral.

AGPL is the nuclear don’t touch because it is vital to all possible usage types.

Again, that’s trivializing because there’s many subtle details that in many cases are intentionally not clarified in the licenses themselves

There is no difference whatsoever between the GPL and the AGPL for software you distribute.

The only difference between the AGPL and the GPL relates to software you don't distribute, but that you make available over a network, such as hosting an AGPL web server. The AGPL says that as long as public users interact with this web server, they have the right to get the code and deployment instructions and AGPL rights to be able to run the exact same service themselves, just as if they had downloaded the software.

The FSF's (very reasonable) position is that linking to a piece of software makes your software a derivative work, but that launching it as a separate process or accessing it over a network generally doesn't. However, if you exchange very complex internal data structures with this GPL software running in a separate process, whether you do it over a network or through pipes or command line arguments or whatever, then your program is still a derivative work, in their opinion.

That is, if you run gcc in a separate process but then mmap() it's memory and search for its entry points and jump to the memory locations of GCC functions in order to generate optimized code for a piece of GCC AST you provide, then your program is a derivative of GCC (according to the FSF, at least) even though GCC is running in a separate process.

A few articles I've read suggest that the AGPL terms trigger if you MODIFY AGPL covered code. So if you (1) setup an AGPL service over the network "as-is" with no changes, (2) use the AGPL-covered service from your proprietary code over the network (ie, across a process boundary), then you are in the clear, and your proprietary code can remain closed.

If you instead, modify/customize the source of the AGPL-covered service, then you are required to make available your changes to AGPL-covered product under a AGPL license.

Is this a correct understanding of the situation? If the product implements a publicly documented protocol to communicate over the network (ie, not "complex internal structures", and/or replaceable with another product), is that permitted?

I am not a lawyer, but my understanding is the same as yours - if you host a proprietary service that has some AGPL components (say, an AGPL authentication service that users interact directly with to get a token, that they then pass to the proprietary parts), then you only have to make sure that the source code for the AGPL part is made available (or not even that if you are using it completely unmodified).

> LGPL allows dynamic linkage without open sourcing your components.

Also static linkage provided you provide users with the object files so that they can re-link with a modified library.

> GPL requires any linkage to open source your parts

With the exception of system components. Maybe that is what is causing the confusion because it's not intuitive why you can ship proprietary software that uses glibc but can't ship proprietary plugins for a GPL program.

> AGPL is the nuclear don’t touch because it is vital to all possible usage types.

That is an absurd viewpoint bordering on FUD. Affero licenses are only a problem if you are trying to work around the license terms. The license does fix loopholes for networked use that allowes SAAS providers to skirt the GPL but it is NOT any more viral than the GPL. Commercial use of AGPL software is absolutely possible - see e.g. cloud providers. Most GPL software probably should be AGPL.

> The [AGPL] license does fix loopholes for networked use that [allows] SAAS providers to skirt the GPL but it is NOT any more viral than the GPL.

One person’s skirting is another person’s complying. Most of what AGPL proponents argue is skirting GPL by SaaS companies seems like straightforward compliance/conformance to me.

Linux dual licenses glibc so that it can be linked without falling under the GPL

And I’m not saying that AGPL can’t be used, in the same way I’m not saying other GPL can’t be used. You’re misunderstanding the context of the comment.

The comment was in regards to: what are you allowed to do without having to fall under the purview of the license.

In that sense, the AGPL specifically exists to close off many areas that the GPL didn’t touch.

Pipe/exec into GPL generally considered okay (not a derivative work)

What you described only okay with LGPL

And pipe/exec not because it's a magic boundary, but because that kind of interface is often so generic and simple it's not constituting a derivate work to use that boundary. But one could easily come up with an example where it is.

It is a "magic" boundary in that it is an interface to another process, i.e. effectively another program altogether.

Another process is not necessarily "another program altogether". What presents as "a program" to the user may be a collection of several processes using a variety of IPC methods to interface with each other. "Process" has a fairly concrete technical meaning but "program" is more loosely goosey. A shell script could be called "a program" even though it's executed by a shell process and invokes dozens of other processes, without which it is nothing.

That's a nitpick.

Within this discussion if you take a GPL'd piece of software, a "program", and interface to it via pipe/exec you're just using an external program, you're not linking to it and are 'immune' from GPL 'contamination'. If not it would effectively not be possible, or very difficult, to run commercial license programs on Linux, for instance.

It would not, however, Linux has:

"GPL-2.0, with an explicit syscall exception"

Deleted/Edited (sorry...):

That's for the kernel, right?

My comment was general and about all the interactions betweem programs.

How else do you run programs on Linux?

I think that's probably an interesting case, and whether it's a valid defence is going to depend on what you're doing.

If you are (for example) shipping a product which is effectively a thin wrapper around a GPL 'd component, but instead of linking to a GPL'd library you've written and dutifully GPL'd a small shim program that turns the library into an executable and reads function arguments over a pipe, then you call that from your 'non-GPL' code... I don't think you'll find the process boundary is some sort of absolute defence in law.

In fact you may be looked upon unkindly in court for having knowingly attempted a license circumvention mechanism.

The law is not necessarily sensitive to the exact inner workings of a turing machine, and the definition of a derivative work may or may not be affected by such shenanigans.

If the whole product you're selling is just a thin wrapper around GPL code, why not simply sell the GPL code or GPL the wrapper? Surely the wrapper would be trivial for a competitor to replicate so the reason you're able to make money from it must be something else like other services you provide with it, or just the value of you having chosen and distributed that particular GPL program.

It was more of a thought experiment to show that a process boundary isn’t likely to be some sort of perfect defence, than a real example of how someone might operate.

GPL v2.0 is pretty vague on that:

If identifiable sections of that work are not derived from the Program, and can be reasonably considered independent and separate works in themselves, then this License, and its terms, do not apply to those sections when you distribute them as separate works.

I use the pipe/exec loophole to include a GPL program in a proprietary one. I personally decided that the proprietary code can be considered an independent and separate work because it's still does something useful without the GPL component. Also, you could swap the GPL component for a 3rd party proprietary one, which actually exists and uses the same interface for communicating via files. So to me, it's just a generic interface and I'm distributing the GPL program together as "aggregation" of separate works. For the sake of human relations, I also got approval from the GPL program's author to do this but he admitted he wasn't sure, and neither am I.

That quote is part of section 2, I believe. That section states "You may modify your copy or copies of the Program or any portion of it, thus forming a work based on the Program...", so if my program uses a GPL'd program via pipe/exec I am not modifying the GPL's program at all, just using it, and so my interpretation is that section 2 (and in fact the whole GPL license), and therefore your quote, does not apply at all.

An example of what I have in mind: Let's say that you write a program and for some functionality it needs to grep files. Now, instead of programming this from scratch you just exec/pipe grep and grab its output. That's does not make your program GPL'd although grep is.

Packaging grep with your program to make distribution easier (if needed) does not, either.

I assume it never matters if you modify it or not since you're fully allowed to fork and redistribute GPL code. It doesn't even prohibit modifications that enable compatibility with proprietary software.

The ambiguity is in whether the two programs (proprietary and GPL) together count as a "work based on the program". Your grep example seems to support the idea that they don't, and it wouldn't fall under the operating system components exemption because it "accompanies the executable". But then where is the boundary? Others have said you can't make a GPL shim to convert a library to an executable then use it in this way.

Well, in the beginning of section 2, which I quoted, a "work based on the program" is defined as a modification of the program. If you use the program as is then you do not modify it and therefore your own program is not a "work based on the program".

A GPL library is a trickier case because it's all about linking to it and potentially how that is done.

My guess is that if the shim converts the library into an executable that is usable in itself generically then you're fine (e.g. you turn some 'libgrep' into grep then uses grep). But if it only implements something like RPC through pipes specifically for use with your own program then someone might decide that this is just convoluted 'linking' and falls within the licence.

I see. It almost seems to be starting to make sense. That even brings clarity to:

"But when you distribute the same sections as part of a whole which is a work based on the Program..."

No matter how you distribute them, they never formed a "work based on the Program" anyway so the GPL doesn't spread.

If all that's correct, it seems to leave a loophole that I mentioned where you can fork the original program to make it specially compatible with your proprietary program and redistribute it under the GPL. Now when you combine them you're not modifying it because you (or your friend if need be) did that before you obtained it and the modification is none of your business.

It's extremely entertaining to see people here arguing that programming and software interfaces should be copyrightable and therefore protected by the GPL.

Pretty sure the prevalent opinion was that it was going to be the end of the world if the US Supreme Court ruled that APIs were copyrightable in the Oracle v Google case.

Just say'n.

I think there is a way to make APIs copyrightable:

1.) Hire a bonafide poet (e.g., someone with an MFA degree in Creative Writing).

2.) Have them write Haikus on salary for your company ("work made for hire" under copyright law).

3.) Use the Haikus in your API names.

e.g. for a sqrt() function in Java:

public double sqrt___an_ancient_pond_$_a_frog_jumps_in_$_the_splash_of_water(double number);