March 7, 2020

Copyleft Has No Possewar is over, want it or not

Even the guy who isn’t on Twitter has seen that WhiteSource report showing permissive licenses outpacing copyleft. And I caught wind of some commentary, like this piece from Matt Assay.

I’m not here to argue for copyleft over permissive. I’m not here to join the permissive gloat squad, either. I’m here to relate what I see in practice, in the field.

The idea of permissive v. copyleft as be-all, end-all “open” is inane, played out, and pragmatically vacuous. I wish we’d get over it and have better conversations.

The WhiteSource article, and the reader appetite that puts the bait to its click, perpetuate the mentality of a long closed marketing experiment. The social notion that copyleft is one tribe, permissive another, and new licensing choices represent some kind of wonky perpetual Ragnarök between them, descends to us by memes and muscle memories old enough to drink. They are blowback of guerrilla actions meant to win a ruinous war of nerdy words long since lost on both sides. With the benefit of hindsight, permissive v. copyleft makes about as much sense, and promises as much genuinely interesting drama, as pen v. pencil or screw v. nail. Form follows function. Beware all contrary configurations.

Exactly as WhiteSource presumably expected, most of the commentary I’ve seen takes the form of permissive-license people chortling self-congratulation. Theirs, surely, is the side of history, so the vaguely labeled axes of the WhiteSource infographics say. Permissive shall rise. GPL shall fall. Slowly, perhaps even the niggling attribution and change-notice rules of popular permissive licenses will slough away, leaving something like Blue Oak. We might even abandon licensing outright, warranty disclaimers and liability limits be damned, for the pure glory of the true “public domain”.

When Matt amplifies Chris DiBona’s point that license restrictions discourage adoption, he makes a good point, but well short of a complete case. Single-minded focus on adoption to the exclusion of everything else makes things very simple. It’s easy to run around accumulating evidence for that ideology selectively. What’s more, it’s clear and easy to prescribe what others ought to do, thus producing more selective evidence and feeding the feedback loop. But monomania of adoption reflects reality no better than single-minded quality, ideology, aesthetics, or marketing obsession. Companies pay for software all the time, available under no free open license at all, because it is good, sold very well, or for other countervailing reasons. The magic of free, permissive licensing is simply overwhelmed by other factors.

It’s complicated. Reality always is.

That being said, assuming for the sake of argument that popular copyleft licenses were adequately maintained—they’re not—I would still expect to see numbers favoring permissive licenses on a per-artifact basis. That’s not to say counting packages sums up licensing reality, any better than KLOC bespeaks a coder’s craft. To wit, I currently boast 359 public npm packages to my name, most of them permissively licensed, and I can assure you that my numeric contribution to permissive-license supremacy reflects no underlying merit in quality, innovation, maintenance, or design.

I wonder how they count the kernel. I wonder why they count at all.

When considering copyleft and permissive licensing, never consider those options alone. They are never exclusive choices. Whether the developer or developers fly solo or work for a firm, open licensing as we know it represents but one short segment of a much longer line. The spectrum of licensing and distribution choices runs from highly protected trade secrecy to public abandonment of all right to exclude, and that’s still flattening it out onto one dimension.

The role of copyleft on that spectrum is now and has been marginal in the broader economics sense, liminal in the broader anthropological sense. Copyleft does work, and thrives, when and where there is both a valid case for permissive open release and a valid case for withholding on non-open terms, such as noncommercial terms, or no public terms at all. In other words, copyleft tends to make strategic sense precisely when other factors—innovation, quality, market position, and so on—can and do overwhelm price and permission alone. That was true of GNU core and the kernel in their early days, with first-mover advantage away from restricted distribution and paid licensing. It’s true today of company projects riding strong network effects or technical leads, like MongoDB and x265.

A certain and very natural fate befalls successful copyleft projects whose tenure at the cutting edge, or whose other competitive advantages, fade with time: they’re cloned or otherwise supplanted by permissively licensed alternatives, or succumb to de facto permissive licensing by exempting their primary use cases from copyleft. The canonical example remains GCC, but the same process plays out for oodles of libraries, frameworks, and other projects each year, just as it plays out for closed software reimplemented in the open. The mark of a successful software project, open, closed, or anything between, is a new, upstart substitute somewhere on the spectrum closer to completely available, effectively ownerless, and unrestricted, free as the words of the English language. Underpricing as sincerest form of flattery. So far, so good.

Alas, a broader, freely conversable vocabulary of software building blocks does not mean a freer profession, industry, or society. A certain increase in absolute capability simply raises table stakes for competition. And just as the competitive advantage of projects under copyleft terms fades in time, the advantage and simple relevance of permissively licensed software fades, too. In some cases, the gravity well of zero price, broad permission, and wide adoption actually turns around and swallows quality, innovation, and design. The Internet we have today is in many ways a chewing gum hack job of crafty patches keeping decades-olds protocols afloat and making steam. Similar attempts to artificially prolong performance of x86 begat Meltdown, Spectre, and whatever new perf-reverting vuln comes next.

Neither does an inevitability of progressively permissive substitutes obviate the need for the copyleft waystation. It’s tempting to look back at the history of banner projects, such as Linux or the Internet, and proclaim that every project should skip the intermediate steps of their journeys directly to the de facto permissive promised land. That is a bit like prescribing Utopia by all-consuming, immediate revolution. Whether the trap is snuffing out developer enthusiasm, starving the project for cash, surrendering institutional independence, or involving too many heads in design too soon, incrementalism in licensing, disclosure, and the development process more generally protects precious, fragile resources precisely when they are precious and fragile.

It’s tempting to smooth out paths to glory—forget about Lions’ commentary, Minix, the first Linux license, the user-space exception—especially when big companies churn out well publicized, permissively licensed projects with big marketing budgets, immune from all intermediate worldly constraint. But it is also misleading and dangerous. Software production is a process, not a specification. Waterfall licensing makes no more sense than waterfall development.

Turning the marginal nature of copyleft around, we should also expect a surfeit of permissive licensing as a dominant strategy for offsetting lack of differentiating quality, network effects, and other advantages. In other words, if you want to win the Next Hot Web Framework lottery or the like, and you do not have a true step change in functionality, efficiency, or other value to differentiate yourself from the throng, you will almost certainly choose a permissive license to stay in the running. Even though the odds remain overwhelmingly against your runaway success. Especially when adoption—literally any adoption—represents a new and exciting development of your coding career.

The odds those fledgling projects play should translate to an expectation about permissive-license population statistics. We can expect a glut of permissively licensed alternatives on public GitHub, with relatively little technical or strategic relevance, in well trodden genres, just as we can expect a glut of highly derivative pop songs grasping for our ear time on SoundCloud. Some of those dice rolls will turn up winning numbers, take the spotlight for n minutes of fame, and perhaps advance the state of the art or prevailing fashion by some small step. But apart from the occasional monumental act of enterprise-incumbent loss-leading, we shouldn’t expect revolutions or even the state of the art on permissive terms, just more of the same. More copyleft at the relatively thin margin. More permissive once commodified, or to commodify by act of targeted corporate dumping.

I write all of this with some confidence, because I see the kinds of results WhiteSource’s report projects through my law practice. The majority of clients who seek my help on licensing strategy end up choosing permissive licenses. This despite many clients seeking me out precisely for my work and writing on copyleft.

Quite independent of what terms other developers have chosen for other projects, this merely reflects the lay of the strategic land that we choose our licenses on, broadly speaking. Nobody sees all of open source, from any angle. But I’m reassured to think my own periscope view reveals something like a representative picture.

It also strikes me how few of my clients come to me preloaded with abiding allegiance toward permissive or copyleft as such. When devs are on the clock, money is on the line, or foundational ideals are at stake, developers, business people, and all creatures between feel and understand that neither orthodoxy nor tribalism nor Twitter plaudits will save them from a less than pragmatic licensing choice. The instincts and identities they’ve seen or developed online mostly fall away. They choose a license as they’d choose a language or library, not religion.

In surveying the byproducts of those decision processes, we shouldn’t impute any partisanship, ideology, or affiliation to others in confirmation of our own. There are more nails than screws in a well built home. That doesn’t make hammers—or screwdrivers—the one true hand tool.

more posts about: Open Source Licensing Copyleft

Your thoughts and feedback are always welcome by e-mail.

more articlesback to topedit on GitHubrevision history