Yes. 100% yes. It is very explicit in what must be included and you are explicitly ignoring what it says about scripts:
" For an executable work, complete source code means all the source code for all modules it contains, plus any associated interface definition files, plus the scripts used to control compilation and installation of the executable. "
Scripts used to control compilation and installation != signing keys. It means scripts used to control compilation and installation.
You may want it to include something more, but it doesn't. That's why GPLv3 was written.
This is a pretty big assumption, and in fact, is not actually true in a lot of cases :)
Let's assume it is here.
"Why isn't that data file also part of "the scripts used to control installation of the executable"?"
Because the data file literally isn't a script used to control installation of the executable? It's a data file.
At best, btw, you'll be able to argue that it would have to include the signing key, but not any password for it. So you are still pretty screwed :)
Your argument also swallows the definition. Where does it end? Do i have to include the interpreter for the script, and provide source? Why not, since it's required for the script to function. (Note, this would not be covered by the next exception in a lot of cases)
If i give you a bash script The installation script also almost certainly refers to the executable, but that's called out explicitly as something i have to give you source to. Why bother if it's included in the "scripts to control compilation/installation" part? (In general, courts are going to read definitions into things that make parts of enumerated lists differentiable from each other ) In general, i would say you can go down this path, but you won't get a court to come with you.
Scripts means scripts. Data files means data files.
If the key is literally part of the script, yes, you'd have to ship that. Not the password, mind you.
If they wanted something else, they should have written something else. Which they did, in gplv3 ;)
You're referring to "scripts used to control compilation", but how are signing keys not part of the source code? GPLv2 uses a specific definition of source code, "the preferred form of the work for making modifications to it." Don't you have to be arguing that a form not allowing modifications to be used or tested is the "preferred form of the work for making modifications"?
You are trying to take parts out and remove context from them.
I don't have to argue at all about the preferred form, because it very explicitly says what that is: "The source code for a work means the preferred form of the work for making modifications to it. For an executable work, complete source code means all the source code for all modules it contains, plus any associated interface definition files, plus the scripts used to control compilation and installation of the executable. However, as a special exception, the source code distributed need not include anything that is normally distributed (in either source or binary form) with the major components (compiler, kernel, and so on) of the operating system on which the executable runs, unless that component itself accompanies the executable."
The first two sentences go together. It defines, explicitly, what the preferred form of complete source code for an executable work includes.
So that isn't going to include signing keys unless you want to argue that one of the explicitly enumerated parts includes signing keys.
For years, people have argued back and forth about whether the scripts include data files. The general consensus is "no", because otherwise, it has no differentiable meaning from the other parts.
I use my GitHub private keys way more often than final code signing keys - be those keys used for as "nefarious" a use as tivoization lock-in, or as benign a use as proving authorship so Windows can tell my executables apart from some asshole's malware better.
During standard development I'm either using temporary, machine-local certificates, or using unsigned code. If that means I need a devkit, or an emulator, or a dongle, or whatever else might be required to iterate that doesn't require the final "publishing" steps, in a well designed system it shouldn't require final publishing certs. Because for security's sake, you'd consider going to such lengths as keeping those code signing certs on a tightly controlled, air gapped machine, to reduce the chance of getting pwned and dealing with cert invalidation. Or worse.
Which makes it a very far cry from being part of your compilation phase in any capacity, nevermind a script used to control it.
But the "preferred form of the work for making modifications" presumably involves each person using their own github keys (and/or repositories) etc. Anybody can create an account and use github. That is not true in the tivoization case because in that case the only keys that can be used are the original ones.
Are you sure? Based on what? I've managed to quote you without the use of copy and paste in that previous post - I'm not sure how I accomplished that if I didn't read what you wrote - so please be more specific.
This also begs the question: Did you read what I wrote?
> But the "preferred form of the work for making modifications" presumably involves each person using their own github keys (and/or repositories) etc.
And the "preferred form of the work for making modifications" also presumably involves each person using their own signing keys (and/or hardware that may not require them) etc.
> Anybody can create an account and use github.
Anyone can buy their own hardware, and compile for it, and install on it. Just don't pick tivoized hardware.
> That is not true in the tivoization case because in that case the only keys that can be used are the original ones.
The source code is not tivoized - hardware is. So don't pick tivoized hardware.
You're still talking about "scripts used to control compliation" when my point was that it doesn't need to be that if it falls under the definition of source code proper.
> And the "preferred form of the work for making modifications" also presumably involves each person using their own signing keys (and/or hardware that may not require them) etc.
Which works perfectly well up to exactly the point where the person distributing the software delivers it on hardware that does require signing keys and doesn't allow the user/owner to use their own keys. Then they need yours because you prevented them from using their own.
> The source code is not tivoized - hardware is. So don't pick tivoized hardware.
The hardware is being chosen by the person distributing it with GPL software.
Ahh. Expanding the definition of source code enough to include code signing keys would have also put them into the (sub)category of "scripts used to control compilation and installation" in my mind, making the distinction moot.
It's not moot in the context of the specific point you intended, of course, and which you've now provided, which I thank you for.
> Which works perfectly well up to exactly the point where the person distributing the software delivers it on hardware that does require signing keys and doesn't allow the user/owner to use their own keys.
Which works perfectly well up to exactly the point where the person distributing the software delivers it on hardware (my VCS server) that does require VCS keys and doesn't allow the user/owner to use their own keys.
> The hardware is being chosen by the person distributing it with GPL software.
...and by the person purchasing the product that has the GPL software in it, by choosing to buy that product.
You distribute your software by conveying the VCS server hardware to the customer?
> ...and by the person purchasing the product that has the GPL software in it, by choosing to buy that product.
All GPL licensees have the rights the license grants them, not just the ones who buy or don't buy specific hardware. What good is the license at all if distributors can just say "I'm not doing that part, if you don't like it buy from somebody else"? And what if they all do that, or nobody else makes equivalent hardware?
So even a reaching definition of "preferred form for modifications" doesn't hit signing keys, just like "preferred form" of a Rails app doesn't include your EC2 credentials.
Picture the scenario at the manufacturer where one developer is being tasked with creating a new feature using a consumer unit fresh off the line, for a business demo. What resources do they utilize in the course of doing this? This is the company's "preferred form" of the source tree for modifying an as-shipped device, and is what needs to be made public.
The company could supply a firmware update to overwrite the bootloader and stub out signature checking, if that is what makes sense. Or a well-specified description to use JTAG to do the same. Or perhaps they don't actually care about security and 'make dist' creates an installable image.
The thing is, all of these arguments ("preferred form for modification", the key being part of the installation script, code+signature being the executable work) seem like uphill battles when trying to show the manufacturer is violating them in an informal forum like HN.
But the intent and spirit of the license is straightforward and even spelled out in the preamble! "Our General Public Licenses are designed to make sure ... you can change the software". Fulling these goals is the entire point of the contract and only consideration for receiving a license. So the manufacturer is actually the party that must work to show how what they're releasing actually fulfills the specific terms.
What is the relevance of that supposed to be? You can put any part of the development process on any number of different computers or distribute the work between different developers however you like. And the relevant "preferred form for modifications" is presumably the form the user uses to make modifications to the device delivered to the user in any event.
> So even a reaching definition of "preferred form for modifications" doesn't hit signing keys, just like "preferred form" of a Rails app doesn't include your EC2 credentials.
Why wouldn't it include your EC2 credentials (or other root access to the instance) if conveying an EC2 instance to the customer is how the customer receives the software from you? But now you're wading into the SaaS vs. software distribution ambiguity which has nothing to do with the case where you've bought a physical device with the software on it.
I'm having a hard time digging it up, but I swore I have seen a video where RMS lays out the exact same theory I'm saying here.
It's not a terribly outlandish theory. I'm just getting a lot of flak for bucking the status quo from people who grew up with DRM kool aid and see Free software in terms of the corporations that lock it down.
That everyone includes RMS, by the way. Here's a transcript[0] from 2006 where he's talking about it and says:
>For instance, the Tivo itself is the prototype of tivoisation. The Tivo contains a small GNU/Linux operating system, thus, several programs under the GNU GPL. And, as far as I know, the Tivo company does obey GPL version 2. They provide the users with source code and the users can then modify it and compile it and then install it in the Tivo. That's where the trouble begins because the Tivo will not run modified versions, the Tivo contains hardware designed to detect that the software has been changed and shuts down. So, regardless of the details of your modification, your modified version will not run in your Tivo.
So while you may be correct that it's never been litigated, it would seem that if the strongest software freedom advocates like RMS and the FSF thought that it did violate GPLv2, surely rather than telling people they should upgrade to V3 because it blocks tivoization [1], they would have just litigated against tivo and set the precedent.
[0] http://fsfe.org/campaigns/gplv3/tokyo-rms-transcript#tivoisa...
RMS was likely wrong about at least one thing in that clip - I would think signature checking, especially an early incarnation of it, would be done by the bootloader (ie software/firmware) rather than a specific piece of hardware.
This could be minor (thinking of embedded firmware as hardware isn't wrong in the freedom sense), or that framework could have misguided his thinking. He didn't use a specific term for the compiled source code, only "it". The GPL2 speaks in terms of "executable", but an unsigned or missigned binary is not actually executable. The "executable" would consist of the machine code and header, which would necessarily include a correct signature to be "executable".
For comparison, say you had some GPL software which checked in with a dongle or license server, and refused to run if there was no license/dongle. Then you'd be within your rights to edit it to remove the license check.
But here, the device itself refuses to load the program because it isn't blessed. Consider another implementation: in the factory, the hash of the GPL firmware is burnt into a ROM, and the bootloader checks that before running the code. Clearly you can build a modified executable, and you still can't run it on the device. But crucially, there is no header that you're missing; the device just can't run that exe. Or suppose the modified executable is too big to fit in memory - are they obliged to come upgrade the device? The point is, the license covers the software, not the loading of that software.
IANAL, while this may appear against the spirit of the license, it's hard to see how the license prohibits it. The existence of GPLv3 implies the FSF thinks so too.
To put it bluntly: an ought does not imply an is.
Please find me any real-world device that actually has a symmetric checksum burnt into OTP or mask _ROM_ while the code itself resides in flash.
No. Not really. GPLv2 requires that the modifier of source code release the tools to build/compile that source code into an executable. Whether or not the source code can be run directly (ex. Ruby, Python) or requires a compile/assemble/link process (ex. C, Java) is irrelevant. The GPLv2 also does not entitle or give anyone rights to any method to execute that recontributed and rebuilt code, for example if the code was ported to a proprietary hardware or some closed sourced ISA. However, IANAL and could very well be wrong because the GPL is such an unwieldy beast of unknowns.
In short: yes.
Being able to access and modify and redistribute source code isn't the same as having the ability to deploy it to a particular piece of hardware.
Likewise, if I open source a .NET project that doesn't mean it has to include Visual Studio even though VS is the preferred way to build it.
But the preferred form for modification includes being able to test your changes. Source without installation scripts is most likely not the preferred form - ie what do engineers at the company use?
> if I open source a .NET project that doesn't mean it has to include Visual Studio, which is the preferred way to build it
Yes, because having been designed in the era of expensive compilers, GPL2 explicitly covers this:
> However, as a special exception, the source code distributed need not include anything that is normally distributed (in either source or binary form) with the major components (compiler, kernel, and so on) of the operating system on which the executable runs, unless that component itself accompanies the executable
You're trying to draw a box around "Windows" and define operating system as everything in that box. But minesweeper is not an operating system component, it's just something that comes with Windows. A hypervisor is an operating system component, even though not all versions of Windows have one.
So is a compiler.
I imagine there exist developer-oriented PC OEMs that offer PCs with Windows and Visual Studio already installed. So then it is included with the OS by the OEM. The fact that that would change the outcome in your interpretation implies that there is something wrong with it.
That in practice nobody except FSF and Debian cares is another thing.
This means that someone sending a PGP signed e-mail saying "I audited that the source code of GNU Foo with hash 0xdeadbeef... is free of backdoors" is violating GNU Foo's copyrights unless he releases his private keys.
This is obviously a very undesirable outcome, and so it's resonable to hope that hashing does not create derivative works, and thus signing doesn't either.
Including the scripts to install in the "preferred form" also seems very problematic because then you might also be forced to include a full copy of the hard drive of your development machine since that's an even more "preferred form" under that theory.
This is also a very undesirable outcome, so it's again reasonable to hope that "preferred form" does not necessarily include things that were absent in the original work and that aren't essential for reproducing the binary derived work itself.
Note that this means that I could sell a device that only runs signed code and could separately sell signed apps for that device that the user can buy and install later, and as long as the code that ships with the device is not GPLv3, I could use GPLv3 code in the signed apps and would NOT be required to release signing keys under GPLv3.
(The Apple App store incompatibility with GPLv3 has nothing to do with the app signing requirements. It is due to the terms of service that users are required to agree to in order to access the store. These impose what the GPL considers to be additional, incompatible conditions).
It seems odd, if they felt the way you think they do, that they would have written this so narrowly.
Does not explicitly. From GPL2:
> Accompany it [executable code] with the complete corresponding machine-readable source code ... The source code for a work means the preferred form of the work for making modifications to it
So, you're telling me that the preferred form for making modifications to the source code of a router does not include the scripts which install onto the router to test those modifications?
> Not by any legal definition of "bad faith"
The GPL2 and the FSF are very explicit on what the GPL2 is supposed to engender. Accepting the license and trying to circumvent it on a possible technicality (assertion that a signed blob isn't a derivative work, when the only reason anyone is interested in the source code is to run it) looks pretty bad-faith to me.
IANAMOAG (I am not a member of a guild, and you probably aren't either). Even if we were, we could still disagree until it was actually litigated. I see no reason to prematurely capitulate Linux devices being non-Free.