The Open Source Definition as Copyleft Regulationan in-depth review of commonly cited criteria
This is the second in a series inspired by MongoDB’s Server Side Public License and ensuing debate. It builds on the first, which sets out the vocabulary of copyleft design and regulation used here.
Changes since initial publication:
How does the Open Source Definition regulate maintenance of strong copyleft terms in open source licenses? Sporadically, incompletely, and unhelpfully for current controversies. A few Definition criteria set clear limits, but most don’t say anything about copyleft at all. A few, read out of context, beg readings that prohibit copyleft generally, or point to utterly unhelpful distinctions, clogging up debate.
In reading the Definition as a copyleft regulation, we can ask two practical questions:
Does the Definition prohibit specific copyleft design choices for trigger, reach, licensing, or distribution?
For example: Does the Definition prohibit triggering on running software without changes to provide network services?
Does the Definition prohibit copyleft design approaches more generally?
For example: Does the Definition prohibit designs that trigger on network use and reach beyond derivative works of the licensed software? Designs that seem too strong, in some other way?
In the end, the Open Source Definition provides neither holistic design-approach regulation, nor regulation for each copyleft design choice. That makes the Definition an incomplete copyleft regulation, at best.
That is altogether understandable, considering the way it was written, and why. The Definition wasn’t meant to answer current questions about the limits of open source copyleft design. Those questions hadn’t even been practically posed yet. But we’re asking them now. In order to establish meaningful regulation for open source copyleft, we’ll have to look beyond the Definition. That will be far easier to do without pretending that we’re still debating the Definition.
In another post in this series, I’ll begin to offer a usable framework. This post, however, will cover only what we can glean from the text of the Open Source Definition, its history, and the licenses the Open Source Initiative has approved as conforming.
Review the few Definition criteria that clearly regulate copyleft.
Outline a few recurring interpretive traps.
Consider the redistribution criterion, as a warm-up.
Dive into the discrimination criteria, first field-of-endeavor, then person-or-group.
Round out with contamination.
Clear Copyleft Regulations
Three Open Source Definition criteria—3 (Derived Works), 8 (License Must Not Be Specific to a Product), and 10 (License Must be Technology-Neutral)—clearly regulate copyleft design choices. Starting with those rules will give us a sense of the clarity missing elsewhere.
Allow the Same Terms for Derived Works
Criterion 3 of the Definition was brought over from the Debian Free Software Guidelines without change:
3. Derived Works
The license must allow modifications and derived works, and must allow them to be distributed under the same terms as the license of the original software.
Criterion 3 is a clear regulation of licensing design choice. To be open source, a copyleft rule must at least allow the same license terms for work within its reach. It can’t require some other set of terms, such as permissive terms or terms that aren’t open source.
It’s not clear from the text of the Definition whether the same terms must be the only terms on which “derived works” can be licensed. That was true of GPLv2, which the Definition, like the Debian Free Software guidelines from which it derived, was written to describe. However, there are three pieces of compelling textual and historical evidence for allowing the same terms among other choices, rather than having to require just the same terms.
First, criterion 3 says that the license must “allow” the same terms, not that the license must “require” them.
Second, permissive licenses like BSD, another license listed in the “Example Licenses” section of the Debian Free Software Guidelines, permit distributing “derived works” under any terms. There’s no reason evident from the text of the Definition why permissive licenses should be able to allow permissive “derived works”, but copyleft licenses should not.
Third, the Open Source Initiative has approved licenses that permit different terms for at least parts of “derived works”. For example, section 13 of GPLv3 and section 13 AGPLv3 create a compatibility bridge between those licenses, by making exceptions to their licensing rules for keeping GPLv3 and AGPLv3 contributions under their respective terms.
As a final note, the term “derived works” is close to, but not exactly, a relevant term of US copyright law. The legal term is “derivative work”. Of all the non-legal usages in the Definition, which does not read like the work of an American copyright lawyer, “derived work” feels the most like a legal allusion. But the difference shores up an interpretation that “derived work” goes beyond copyright derivative works, when copyleft reach does. We could read criterion 3 to regulate licensing rules for all work within copyleft reach.
No Product-Specific Licensing
Criterion 8 of the Debian Free Software Guidelines had to be generalized for the Open Source Definition:
8. License Must Not Be Specific to
The rights attached to the program must not depend on the program’s being part of a
Debian system.If the program is extracted from Debianand used or distributed without Debian but otherwisewithin the terms of the program’s license, all parties to whom the program is redistributed should have the same rights as those that are granted in conjunction with the Debian system.
If we read “software distribution” and “product” to cover not just distributions of multiple programs, like Linux distributions, but also particular distributions of the licensed program itself, like Firefox and Iceweasel, criterion 8 becomes a clear regulation of copyleft licensing design. An open source copyleft rule cannot vary the terms required for parts taken out of the program and parts left with or as a part of it.
Criterion 8 could also function as a regulation of copyleft reach, insofar as a copyleft rule does not set rules for licensing code that does or doesn’t remain part of the original distribution.
Open Source Definition criterion 10 was not part of the Debian Free Software Guidelines, which sports an “Example Licenses” section as number 10, instead. Here is the new text, peculiar to the Open Source Definition:
10. License Must Be Technology-Neutral
No provision of the license may be predicated on any individual technology or style of interface.
Criterion 10 is a clear regulation of distribution and licensing design choices. An open source copyleft license cannot require licensees to use a specific technology for licensing or distributing code within its reach. For example, an open source copyleft license cannot require licensees to distribute work within its reach via FTP, or to require that licensees for derived works accept terms for that work by electronic signature.
The first Artistic License 1.0, also among the “Example Licenses” listed in the Debian Free Software Guidelines, includes a condition that can be satisfied by making changes “Freely Available” as defined, with Usenet and “an archive site like ftp.uu.net” offered as example methods. Subsequently approved copyleft licenses specifying classes or kinds of methods for distributing source, without mandating any particular one.
Having reviewed some clear copyleft regulations in the Open Source Definition, we move on to criteria that tempt us to misinterpret them as copyleft regulations. As we’ll see, a few common problems plague more than one of these criteria. As such, it’s worth reviewing them first, before we dive into criteria one by one.
Excluding Copyleft Generally
Ignoring history, it’s possible to read several criteria of the Open Source Definition, straight off the page, to prohibit copyleft altogether. These readings crop up each time a new, stronger copyleft license appears, forcing large parts of the conversation to start from before square one, rather than from where the debate left off the last time. The generality of the Definition invites these readings.
Generality was intentional. In fact, generalization was the point. The Open Source Definition was adapted from the Debian Free Software Guidelines, which were made part of the Debian Social Contract in 1997. The Debian Project drafted and adopted the Guidelines to describe the kinds of licenses project participants recognized and accepted at that time, as a category of license terms.
That category included GPLv2, then considered a very strong activist copyleft license, and a very popular choice. A section of the Guidelines entitled “Example Licenses”, removed in the Definition but still a part of Guidelines, lists “GPL” first. The Open Source Initiative approved GPLv2 under its adaptation of the Guidelines. Thus, as a threshold matter, any reading of any criterion of the Open Source Definition that excludes GPLv2 can’t be right.
The Open Source Initiative has since approved other, stronger copyleft licenses. Readings of the Definition criteria that exclude those licenses can’t reflect how the Definition has been understood since.
Oddly Specific Readings of Very General Language
Starting from a reading of some Definition criterion that limits copyleft and paring back to avoid GPLv2 or subsequently approved licenses, it’s possible to end up with a very specific reading that limits open source copyleft in a very specific way.
It is only the history that leads to that specificity. The text of the Definition remains as general as it was the day it was published. The path from general text to highly specific meaning is not predetermined by the text. It could not be accepted or rejected by those who reviewed or accepted the Debian Free Software Guidelines or the Open Source Definition when proposed. It cannot be awarded retroactive, undocumented consensus status twenty years in arrears.
This error often manifests as the position that the outer limit of open source copyleft is whatever the last approved stronger-copyleft license happened to be. Starting debate from what’s new is better than starting from whether copyleft can ever conform to the Definition. But looking back at a general Definition, plus a history of progressively stronger approved copyleft licenses, it never follows that the last approved stronger-copyleft license has to be the last ever. That debate has to reach beyond the Definition and even Open Source Initiative history, to the industry, community, and history of the movement. It has to take into account that the Open Source Initiative only definitively accepts licenses, declines to approve many licenses that conform to the Definition, and never affirmatively and finally rejects any terms. It is up for debate, not precluded by the Definition.
The error to do with highly specific readings of general Definition text based on the Definition’s history or past decisions isn’t that they’re necessarily wrong or indefensible. The error is in claiming that the wording of the Definition, or its acceptance as the frame for debate, compels them. There may be very good reasons for such specific rules, but they’re not in the Definition, and don’t derive from its authority.
The concept of precedent proper remains highly controversial among license-review participants. Old decisions are frequently criticized. On the other hand, there doesn’t seem to be any willingness to remove or deprecate licenses approved in the past. It’s clear that past decisions, and especially use of approved licenses in apparent reliance on Open Source Initiative approval, always seems to matter for new submissions. It’s just not clear how.
Restriction versus Condition
Some open source license specialists distinguish “restrictions”, which hold back permission to take actions governed by licenses, from “conditions”, which add requirements to meet after doing what’s permitted. The word “restrict” appears six times in the Definition, but “restriction” doesn’t mean “not a condition”.
Specialists speak of prohibitions on copying, distributing, or using software for commercial purposes as “restrictions”. A noncommercial license has a commercial use restriction. Conversely, they speak of notice-retention rules, which require keeping copies of copyright notices and license terms with copies of software, as “conditions”. Copyleft rules, which require sharing source and giving a license, are almost always classed conditions, too.
It’s not at all clear that the Definition means “restrict” in that oddly particular way, or what that particular way is. The word “condition” never appears in the Definition. Moreover, both the Debian Free Software Guidelines and the Open Source Definition predate today’s larger community of open source licensing specialists, and may predate this jargon usage, too. The restriction-bad, condition-good association may in fact stem from the Definition, leading those trying to decide what “restriction” means in the Definition around in a circle.
But perhaps the distinction in jargon reflects a distinction in the law that affects software licenses? After all, many of the licensing specialists who use the jargon distinction come from the legal camp.
To enforce a copyright license for software, copyright holders bring lawsuits for, among other things, copyright infringement: claims that others took action controlled by copyright without a license from the copyright holder. For an infringement lawsuit based on copyleft, the claim is that the licensee went beyond the permission they received under the public license by failing to distribute source code reached by the license, and licensing it as specified, after triggering the requirement to do so. Whether a defendant lacked permission because the license restricted distribution of closed source—sounds like a “restriction”—or because they had permission to distribute source broadly, but failed to meet an additional requirement of making it open source—sounds like a “condition”—the legal claim is copyright infringement.
In the end, a copyright license creates a list of things that a licensee can do with a work of copyrighted software. Every action covered by copyright but not permitted by the license or a legal exception, like fair use, remains infringement. Either the license gave permission to do what the defendant did, or the defendant committed infringement. Whether we call a license rule a restriction or a condition, it reduces the amount of permission granted by a license.
If the difference between “restriction” and “condition” isn’t in what counts as infringement, perhaps it’s in the consequences. I’ve heard some lawyers explain it in terms of legal “remedies”: what you can get from a court when you win a lawsuit.
Successful copyright infringement lawsuits lead to court orders for defendants to stop taking action with the software. Thus, courts can issue orders to stop making and distributing proprietary software in a way that breaks a copyleft rule. However, at least under US law, it’s not at all clear that courts will order defendants to fulfill conditions, say by forcing defendants to distribute their source code and license it as required to satisfy copyleft. Likewise if terms get enforced as a contract, instead of or in addition to being enforced as a copyright license. Specific performance—orders to do what you promise to do, instead of paying for damages cause by your failure to do so—is exceedingly rare in contract law, and typically only available for promises of specific and unique services, where money couldn’t adequately compensate for broken promises. It’s far more likely that courts will order defendants to stop working with the copyrighted software and award money damages as compensation for the infringement.
In other words, under US law, we expect courts to award money, not code. Legal confusion on this point contributed to the fear, uncertainty, and doubt about copyleft licenses as “viral” licenses that could force large companies to share their closed software.
At this point, we are deep in the law. All of this speaks from legal background that those involved in creating the Definition did not share at the time. Even if there were a clear legal distinction between restrictions and conditions, based on legal remedies or otherwise, it’s not at all clear that the Definition means “restrict” in that oddly particular way.
The Definition describes licenses, which are legal tools. But the Definition doesn’t speak in the terms that US lawyers would choose. It speaks of the practical effects licenses have, as a client might describe the goals for a license to be drafted, not of the particular ways a license should be written, as a lawyer might set out a plan for the drafting job.
Having covered some common interpretive traps, let’s have a look at the first Open Source Definition criterion that tends to snare readers.
Open Source Definition criterion 1 reads more or less as criterion 1 of the Debian Free Software Guidelines did:
1. Free Redistribution
of a Debian component maynot restrict any party from selling or giving away the software as a component of an aggregate software distribution containing programs from several different sources. The license maynot require a royalty or other fee for such sale.
We could strain to read criterion 1 as a rule against copyleft licenses whose reach and licensing terms create license-compatibility issues. In other words, copyleft licensing rules that create conflicts with other copyleft licensing rules, preventing developers who combine work under both licenses from sharing the combination together.
We’d have to read “component” and “aggregate software distribution” to cover changed versions and programs based on the licensed work. Considering that “derived works” gets used for this meaning in another, dedicated criterion, that seems a stretch. And I don’t think developers in practice would go out of their way to say “component of an aggregate software distribution” when they meant “part of a new program” or “basis of a new version”.
But even if we accepted that interpretation, we’d end up with an oddly specific reading that we’d have to pare back to save GPLv2. GPLv2 has created plenty of compatibility problems. Arguably, GPLv2 invented modern license incompatibility, along with modern copyleft. But there’s no basis in the text for paring the broad reading back to save it. Criterion 1 says “restrict”, and GPLv2 is full of “conditions”. But as described above, there’s no good reason to read the specialist distinction between “restrictions” and “conditions” into the Definition generally, or condition 1 specifically.
Overall, “redistribution” here seems to mean something far more like sharing collections of software than changing or building it. We’ll see that again in the contamination criterion, later on.
Two Open Source Definition criteria, 5 and 6, prohibit discrimination, the first against “any person or group of persons”, the second against “specific fields of endeavor”. Yet those criteria don’t prohibit copyleft, even though copyleft discriminates by design.
Quoting Richard Stallman, from Copyleft: pragmatic idealism:
I make my code available for use in free software, and not for use in proprietary software, in order to encourage other people who write software to make it free as well. I figure that since proprietary software developers use copyright to stop us from sharing, we cooperators can use copyright to give other cooperators an advantage of their own: they can use our code.
That is discrimination. We might also call it reciprocity, fair play, or the golden rule in action. But it is discrimination.
Evidently not the kind of discrimination the Definition prohibits. The Open Source Initiative has approved all of the major licenses drafted by the Free Software Foundation, which Stallman founded. A Debian’s FAQ points out specifically that the antidiscrimination criteria don’t prohibit copyleft.
Field of Endeavor Discrimination
Criterion 6 of the Open Source Definition reads:
6. No Discrimination Against Fields of Endeavor
The license must not restrict anyone from making use of the program in a specific field of endeavor. For example, it may not restrict the program from being used in a business, or from being used for genetic research.
This language was copied from the Debian Free Software Guidelines verbatim.
Ignoring open source licensing history, it’s possible to read criterion 6 to prohibit all copyleft licenses whatsoever. After all, as a practical matter, copyleft terms prevent use of copyleft software in the creation of closed software. That tracks the general meaning of “discriminate”. Per the Oxford English Dictionary:
intransitive. Economics (orig. U.S.). To treat goods, trading partners, etc., more or less favourably according to circumstances. Frequently with against (also in favour of). Also in prepositional passive with against.
How can we square that reading with a long list of approved copyleft licenses? Different readers answer that question differently. Some focus on “restriction”. Some focus on “making use”. Some focus on “specific field of endeavor”.
Overall, natural readings of “restrict” and “making use” are very broad. So broad, in fact, that they describe GPLv2. But “specific field of endeavor” doesn’t have any everyday meaning. If we read it broadly, it works with the other key terms to exclude GPLv2, which can’t be right. But if we track the examples, focusing on endeavors where open or closed is a choice, we end up with a reading that works.
The license must not restrict anyone from making use of the program in a specific field of endeavor.
In ordinary English, the meaning of “restrict” is very broad. Per the Oxford English Dictionary:
To limit (a person or thing); to confine to or within certain limits.
To prohibit or prevent from.
Applying this ordinary meaning, “restrict anyone from making use of the program in a specific field of endeavor” would cover rules under which “making use” in a “specific field of endeavor” is infringement, however that happens. Considering GPLv2, the question would then become how we read “making use” and “specific field of endeavor”.
As described earlier, it’s hard to argue that the general distinction between “restrictions” and “conditions” among licensing specialists applies in criterion 6, instead of the ordinary meaning.
The license must not restrict anyone from making use of the program in a specific field of endeavor.
Ordinary Meaning of Making Use
The everyday meaning of “use” is extremely broad, covering anything licensees might do with software. Following that meaning, “restrict anyone from making use of the program” would describe copyleft conditions that prevent licensees from distributing copyleft software with closed changes, as part of closed programs, or simply under different, closed terms. The question becomes whether distributing closed software is a “specific field of endeavor”.
Actions Governed by the License
Some read “making use” in criterion 6 to mean specifically taking any action controlled by the license, under intellectual property law. Under copyright law, that means, broadly speaking, making copies, publicly distributing copies, and making changes or new work based on the old.
Trigger language for all kinds of conditions in old open source licenses usually speaks in terms of publicly distributing copies, making changes or new software based on the licensed software, or both. Both of these actions are controlled by copyright, and governed by copyright licenses for software. Triggers in the earliest copyleft conditions, like GPLv2’s, also speak in terms of these actions. This reading of “making use” would therefore catch GPL.
Running the Software
Some read “use” even more specifically, to mean one specific kind of use: “running” or “executing” software. This reading misses early, GPLv2-style copyleft licenses, because those licenses triggered copyleft only on distribution or making new code based on the licensed code, not running the licensed software. But it would catch stronger, network copyleft licenses that OSI has approved since.
Under copyright law, owners of copyright do not have an explicit right to control and license running of their software, as do patent owners. However, copying software is an unavoidable step in running it, for installation, loading into RAM for execution, and so on. Since copyright holders can control copying of their software, they have indirect power to control running of their software. A specific section of US copyright law makes an exception to this rule, for copies of software that users own, especially when purchased as a part of a package, with a computing device, to keep it running as delivered. But well known federal case law indicates that having a license, especially one imposing notable use restrictions, is not the same as owning a copy.
Copyleft licenses approved after GPLv2, in particular “network copyleft” licenses that trigger on provision of the software as a service to others over a network, go beyond distributing, making changes, and building new software, explicitly or implicitly. They cover running it. Some, like OSL, actually say “use” in their trigger language. Others, like AGPLv3, seem to take great pains to avoid saying “use”, “run”, or “execute” in their triggers.
Regardless of wording, a copyright license can only control the actions covered by copyright. Providing services to others over a network is not distributing copies publicly or making changes or new work based on the software. In fact, copyleft rules based on providing network services ensure that licensees distribute, when they otherwise haven’t. The only action governed by copyright that remains as a firm basis for those rules, then, is copying. When a copyright license makes rules about running a program, it makes rules about making copies, as one must to run the program.
I’ve heard from a few commentators that the language of AGPL, an early network copyleft license, was meant to create a kind of “public performance” right for software. Public performance is another action controlled by copyright, which covers creative work far beyond software. Licenses for other kinds of copyrighted work, like film, theatrical, and musical works, control public performance of those works in various ways. Technically, software work falls into a legal category where public performance controls apply, but what public performance means for software remains unclear. The authors of AGPL would have liked public performance to clearly cover running software for others to interact with over a network, but it did not, and it still doesn’t. So they wrote a license condition to the same effect.
Copyright licenses conditions attempting to control rights other than those controlled by copyright law have to fall back on control of the actions copyright does control. In the case of “synthetic” public performance rights—lawyers would say “constructive”—that means falling back on control of running the software, via control of copying and modifying the software.
Either way, a reading this specific strays a long way from the general text of the Definition. Criterion 6 says “making use”, which is not a legal term. For more specificity, it could have easily said “running” or “executing”, and all involved would have known what that meant.
Taking “making use” to mean specifically running the program, but not publicly performing the software, or not providing the software to others over a network, is even more specific, particular to FSF’s network copyleft, which isn’t the only approved approach, and also anachronistic. Neither the need for such a fine-grained distinction, nor any particular implementation of it, were pressing issues at the relevant times for the Guidelines or the Definition.
Specific Field of Endeavor
The license must not restrict anyone from making use of the program in a specific field of endeavor.
If we count copyleft rules as “restrictions”, and accept that open source copyleft licenses trigger on “use”, we could still decide that proprietary software development doesn’t count as a “specific field of endeavor”. That would explain how approved copyleft licenses from GPLv2 on conform to criterion 6.
Any Activity at All
If we read “specific field of endeavor” to mean any specified activity at all, at any level of generality, it comes to include development and distribution of closed software. Unless we adopt one of the improbable meanings of “restrict” or “making use”, such a broad meaning of “specific field of endeavor” would exclude GPLv2, the license that both the Guidelines and the Definition clearly meant to include.
The term “field of endeavor” has no specific meaning in intellectual property law. The term does appear in US immigration law, but again, there’s no reason to believe the Definition means to invoke that usage specifically, or any legal meaning, to begin with.
Fields with Choice of Open or Closed
If we read “restrict” and “make use” in the ordinary, natural ways, “specific field of endeavor” can’t mean closed development. At a minimum, GPLv2 restricts companies from using GPLv2 code in closed sourced that they distribute to others. Later, stronger copyleft licenses restrict companies from using copyleft code in closed source in other ways, like providing network services.
The Open Source Definition and software licensing history provide a number of examples of specific fields of endeavor that fit the history of copyleft approvals. Criterion 6 mentions business and genetic research. Software licenses prohibiting military use have been widely considered non-conforming, as have licenses prohibiting use in violation of human rights.
OSI’s annotated Definition reemphasizes commercial use. As all major open advocacy groups, including OSI and FSF, have enthusiastically urged, “commercial” and “closed” are not synonymous, any more than “noncommercial” and “open” are. Many people and firms, in many industries, produce, distribute, and develop open source software for pay or other commercial advantage. Business firms, genetic researchers, and military personnel can employ and make open as well as closed software. Software is but a means to those endeavors. The choice of open or closed, as user or developer, stands independent of the choice to endeavor in those specific fields.
Some firms do business in software development primarily. Many of those develop or consume closed software exclusively, for a variety of industries or other use cases. But those firms face the same choice of whether to take or make work on open or closed terms. Activist copyleft was designed to create incentives to choose open, by excluding firms that choose closed. The Definition was written in large part to describe one of those licenses, GPLv2.
Excluding closed software development doesn’t mean locking “specific field of endeavor” down to just business or genetic research. We could extrapolate to other fields that offer a choice of open or closed, such as hyrdraulic fracturing, environmental protection, alcoholic beverages, addition recovery, nuclear weapons, disarmament activism, the practice of law, and so on.
Person and Group Discrimination
Open Source Definition criterion 5 reads:
5. No Discrimination Against Persons or Groups
The license must not discriminate against any person or group of persons.
This language was brought over from the Debian Free Software Guidelines without any changes.
Criterion 5 has typically been used to exclude licenses that blacklist particular people or entities by name, such as lists of firms or particular national governments. Sometimes developers provided reasons for these lists, and sometimes they expressed the reasons as categories, excluding all people or entities described.
For any activity, we could define a group: the group of people performing that activity. For example, the group of people developing closed software. Discriminating against those people in those groups, by requiring them to develop software when they’d choose not to, or cannot, would disqualify approved copyleft licenses, including GPLv2. That can’t be right.
Approved licenses of all kinds—permissive and copyleft—discriminate against other people and groups on the basis of their choices. Perhaps most notably, the defensive termination provisions of many licenses take back permissions granted by licenses to those who sue developers or users under patents. Those rules are highly discriminatory against firms that actively enforce their patents. They’re especially discriminatory against firms that primarily acquire and sue on patents. Those are evidently not the kinds of people or groups of people criterion 5 is meant to protect.
Discrimination in Copyleft Terms
This section was added on November 7, 2018.
Even if copyleft itself does not inherently discriminate in ways criteria 5 and 6 prohibit, particular copyleft designs could fail one or both of those criteria with trigger, reach, licensing, or distribution terms that discriminate in prohibited ways. In that sense criteria 5 and 6 regulate all copyleft design choices, as a consequence of regulating all open source license terms.
For example, a copyleft license that triggers on use to provide a paid network service would fail criterion 6, not because it discriminates against development of closed network services, but because it discriminates against business. A copyleft license that reaches source code that is derivative work based on the copyleft code in the normal case, but also scripts for deployment, if the scripts are used to deploy for United States federal government use, would fail criterion 5, because it discriminates against the United States government and its service providers. Copyleft terms that allow individuals to license without an express patent grant, but require patent grants from corporations would fail criterion 5, because it discriminates against corporations. A copyleft design that requires distributing source code to users in the normal case, but allows sponsors of the project’s foundation to distribute by general publication would fail criterion 5, because it discriminates against non-sponsors.
However, the history of approved licenses indicates that open source copyleft designs can discriminate between means of creating new software from copyleft software in at least some terms. At a minimum, open source copyleft designs can trigger copyleft selectively, so that some methods of developing new software don’t trigger copyleft, while others do.
Approved “weak copyleft” licenses that only reach changes to the licensed software, like MPL 2.0, discriminate between development of closed programs that do and do not modify the original work’s files. As compared to AGPLv3, its sister license, GPLv3 discriminates against development of closed programs for distribution to end users, as opposed to for provision as services.
These approvals apply rules about selectivity in various copyleft terms that aren’t at all clear from the text of criteria 5 and 6.
Open Source Definition criterion 9 involved a slight heading change:
9. License Must Not
The license must not place restrictions on other software that is distributed along with the licensed software. For example, the license must not insist that all other programs distributed on the same medium must be
I’ve seen criterion 9 read to regulate copyleft reach and licensing. But for the reasons that follow, I don’t see those readings in the text and its history.
When discussing criterion 9, in contrast to the nondiscrimination criteria, I find it’s useful to bring all the interpretive boundaries up front:
However we read the first sentence of criterion 9, it needs to accord with the same-media example in its second sentence.
We can’t read criterion 9 to exclude GPLv2, since the Definition was written to describe, not exclude, that copyleft license. The last paragraph of GPLv2 section 2 mirrors the example in criterion 9:
In addition, mere aggregation of another work not based on the Program with the Program (or with a work based on the Program) on a volume of a storage or distribution medium does not bring the other work under the scope of this License.
Reaching now into history, criterion 9 can’t exclude network-copyleft licenses like AGPL and OSL.
More controversially, criterion 9 evidently didn’t exclude stronger-trigger copyleft or “reciprocal” licenses such as Plan 9, Open Watcom, and RPL, which were approved by the Open Source Initiative, rejected by the Free Software Foundation, and likely fail Debian’s Desert Island and Dissident tests.
The license must not place restrictions on other software that is distributed along with the licensed software.
As noted earlier on, it’s tempting to fall back on an informal or legal distinction between “restriction” and “condition”, but there’s no cause to read Open Source Definition language in either of those specific ways. The natural reading of “restrictions” is broad.
However, that broad reading doesn’t seem to include “restrictions” in the sense of terms required by copyleft licensing rules. That reading butts up against, and potentially conflicts with, criterion 3’s clear copyleft rule of permitting the same terms. Criterion 3 has the effect of ensuring that “derived works” can be open source, since the “same terms” must meet the Definition, and are thus open source terms. While an open source license cannot impose restrictions contrary to criterion 1 (Free Redistribution), criterion 6 (No Discrimination Against Fields of Endeavor), or criterion 9, it explicitly can restrict distribution of modified copies under criterion 4 (Integrity of The Author’s Source Code).
All of this, plus the provided example, make criterion 9 far easier to read as a rule against terms that take away permission to distribute the licensed software based on aspects of other software that don’t have anything to do with copyleft, and could appear in permissive licenses, as well. In other words, restating the Definition: “The license must not restrict what other software can be distributed along with the licensed software.”
This reading clearly covers the example: requiring other software on the same medium to be open source. We could extrapolate. For example, a permissive license could not prohibit distribution along with software by a competitor, or software for nonfree operating systems, or software under licenses without express patent permissions.
It also speaks to the concerns of a project like Debian, whose sprawling, heterogeneous distribution channels could never implement every non-conforming license’s restrictions in software, to ensure they’re followed. To prevent that explosion of complexity, and a dearth of distributors willing to take the resulting risk, it made sense for Debian to exclude software that creates such problems. A “contamination” metaphor also makes sense. Just as common carriers of mail or shipping containers enforce hazardous shipping regulations to keep one package from leaking over others, criterion 9 heads off license-based trouble arising “in transit” to users. Once users receive their copies, their rights to build new software come under criterion 3 (Derived Works), a clear copyleft licensing regulation.
The license must not place restrictions on other software that is distributed along with the licensed software.
With “distributed”, we have much the same interpretive problem as “use” in condition 6. We could read it informally, to mean sharing copies of software, or as a legal allusion to distribution to the public under copyright law, or as something in between.
But “distributed” in criterion 9 is more troublesome that “use” in condition 6, considering GPLv2. Reading “use” in oddly specific ways in condition 6 avoided GPLv2, since GPLv2 doesn’t trigger on those specific actions. Conversely, even reading “distributed” in criterion 9 in an very specific, legal way, still describes GPLv2’s trigger. Distribution-based triggers are a hallmark of GPLv2-style copyleft, and the specific design choice led to the ASP loophole and the need for network-copyleft licenses.
Whether we read “distributed” informally or legally, if we reach criterion 9 to address copyleft design, it catches GPLv2.
The license must not place restrictions on other software that is distributed along with the licensed software.
Straining to make a copyleft regulation out of criterion 9—despite “restriction”, criterion 3, GPLv2, “distributed”, and the given example—is a lot of work for no reward. The two-word phrase “other software” would become the limit of copyleft reach. The word “other” would tell us there’s a limit, but without giving any hint of where it lies, or why. Semantically, it’s a blank slate.
License approvals have written a bit on that slate. GPLv3 and AGPLv3 reach beyond the software itself in its definition of “Corresponding Source”. So does RPL 1.5, in its definition of “Required Components”. All reach “other software” as I think developers would mean and understand the phrase, talking shop. Which points back to questions about the meaning of “restriction”.
If this analysis feels unfair to the Open Source Definition, that’s because it is. We’re asking the Definition to do more and different work than it was ever written to do. The Definition wasn’t laid down to express consensus on how far open source copyleft could go. Software copyleft was less than ten years old. The Definition wasn’t written to be dissected this way.
Installed at the center of the Open Source Initiative process, reviewers look to the Definition as the total framework for license analysis, describing and debating in its terms. But when those debates reach copyleft design choices that open source theory implies, but the Open Source Definition does not express, the structure of the Definition boxes discussion in, rather than guiding or supporting it. We begin to force-fit our own broader thoughts and instincts about copyleft regulation into odd nooks and crannies, because they have to go somewhere, and the Definition is where. All of this makes writing about copyleft regulation, and reading what others have to say, even harder, especially for those who can’t cite Definition criteria from memory.
We should accept that the Open Source Definition was written to generalize a broad category of licenses with very few strong-copyleft examples, at a time of great GPL cohesion. We should accept that in its time, for its purpose, anticipating and resolving as yet theoretical problems in copyleft design would have been wasteful, foolish, and potentially dangerous to the effort then coalescing, which the Definition wasn’t. We should accept that as a consequence of its time and purpose, the Definition offers at best an incomplete approach to regulating copyleft, and no particularly good vocabulary for expressing a complete one.
I agree with colleagues, some of whom reject the strong-copyleft license I wrote specifically to test the limits of open source copyleft, that the Open Source Definition alone can’t suffice as the whole basis for assessing new strong-copyleft licenses. But I disagree strongly that an unwritten consensus on the limits of copyleft is latent in the Definition, for an elect who can see or speak it from authority. I disagree strongly that the Open Source Initiative or any other group should graft a new, complete or selective copyleft regulation onto the Definition, exclude those who disagree with it, and declare those remaining consensus.
The way forward is acknowledging that licenses written to the Definition—in 2000 or 2018—conform to the Definition. That includes RPL, AGPLv3, and now SSPL. That doesn’t mean the job of assessing them is done. It means it’s time to put down the Definition, as a tool, and to pick up a more specialized one, for the finer job of assessing new copyleft designs. If that tool doesn’t exist, perhaps we should make it. That would be a project and a product, necessarily, of 2018, not 1998. It could only be.
In posts to come, I’ll set out guideposts for comprehensive regulation of copyleft in open source licenses. I don’t anticipate unanimous acclaim for every point I propose. But I hope even those who disagree will come to appreciate how freedom from a ceremonial duty to torture every new copyleft regulating rule into the Open Source Definition, by exegesis fair, foul, or flummoxing, makes the task tractable, debatable, and honest in a way it rarely has been.
Your thoughts and feedback are always welcome by e-mail.
more articles — revision history — back to top