Free and open source licences: what the argument was about

1980s onwardsWhat the argument was not about

Price. Nothing in any of the licences concerned requires software to be given away, and several of the organisations most associated with them sell it.

The argument was about what a recipient may do with what they received, and specifically about whether a licence should compel them to extend the same permissions to whoever they pass it on to.

1989The mechanism, which surprises people

The licence that compels sharing does so by using copyright, not by rejecting it. The author holds the exclusive right to permit copying, and grants permission on the condition that derivative works carry the same terms.

Which means the strategy depends entirely on copyright being strong and enforceable. Weaken it and the condition becomes unenforceable, and the arrangement collapses into the permissive one.

The two positions, stated fairly

One holds that a recipient's freedom is not secure unless it is passed on, so the condition is the point. The other holds that attaching conditions is itself a restriction, and that the widest possible use, including by people who close their derivatives, is the greater good.

Both are coherent. They disagree about whose freedom is being protected: the next recipient's, or the immediate one's.

1998The rebranding

The term now in general use was coined deliberately in the late nineties by people who thought the existing vocabulary was frightening commercial adopters, and who wanted a phrase emphasising practical benefit rather than a moral claim.

It worked, comprehensively, and it is also the origin of a permanent confusion: the two terms describe nearly the same set of licences and very different positions about why they matter.

2000s onwardsWhat actually happened

Permissive licensing won by volume. The great majority of components in a modern application carry terms allowing closed derivatives, and large companies both consume and publish under them heavily.

Copyleft won specific and enormous battles, most obviously an operating system kernel that runs a large share of the world's infrastructure and remains under terms requiring changes to be shared.

2007 onwardsWhere the argument moved

The conditions attach to distribution, and running a service is not distributing anything. A company can therefore build on shared code, modify it, and offer it as a service without passing anything on.

One licence extended the condition to network use. More recently, several companies have adopted terms that restrict competing services specifically, which are not open source under the accepted definition and are usually described as source available. That is where the current disagreement is, and it is the old one about who benefits from the sharing.

What a licence does not do

Enforce itself. Conditions are meaningful only if somebody is willing and able to pursue a violation, which is expensive, slow, and undertaken rarely and by a small number of organisations.

Compliance in practice therefore rests mostly on reputation and on the difficulty of hiding, which is a social mechanism dressed as a legal one, and it is why the vocabulary of these arguments is moral as much as legal.

present dayThe obligations that actually bite

For most organisations the practical question is narrower than the philosophical one: which of the components in a product carry conditions, and what those conditions require when the product is shipped.

Answering it requires knowing what is in the product, which is harder than it sounds, since a modern application pulls in hundreds of components each pulling in more. The tooling for producing that inventory exists and is a recent development, driven as much by security concerns as by licensing ones.

Why the choice is usually made by default

Almost nobody chooses a licence by reasoning about the positions above. They copy whatever the ecosystem they are working in uses, which is why licence choice correlates so strongly with language community rather than with the author's views.

Which means the aggregate outcome of the argument was decided by defaults in project templates and by what the largest projects in each community happened to pick, rather than by the argument.

present dayWhat the term open source excludes

The definition maintained since the rebranding has ten clauses, and the ones that matter in current disputes are that a licence may not discriminate against a field of endeavour or against particular persons or groups.

Which is why terms forbidding competing services, or forbidding particular industries, fall outside it however reasonable their authors' motives. The definition is a boundary rather than a judgement, and treating a licence as bad because it falls outside it confuses the two.

present dayThe unpaid maintainer problem

None of the licences say anything about who does the work, and the outcome is a large body of infrastructure maintained by very few people, often unpaid, on which enormous commercial activity depends.

That was not a consequence anybody argued for. It follows from permission to use without obligation to contribute, which is the permissive position implemented, and it is the strongest practical argument the other side has acquired since the original disagreement.

What we cannot verify

The licences, their versions and the definitions are published and can be read. Accounts of intent behind particular licence changes come from the organisations that made them. Figures for the prevalence of each licence family depend entirely on what is counted, and vary substantially between the surveys that publish them.