Open Source and the Organization

Through the consulting work I've done in the years since the first edition of this book was published, it's become clear to me that there are special concerns that apply to organizations launching or participating in open source projects. Organizations contain formal management structures and informal social structures: both are affected by engagement with open source projects, and both sometimes need to be changed to better support open source activity by the individuals within the organization. In particular, government agencies have special pitfalls to watch out for when working with open source projects.

This section therefore examines organizational issues generally, and some issues specific to government agencies, and offers some advice about how to make organizational engagement with open source more likely to succeed. Many of these recommendations will be brief and somewhat generalized, not because there isn't more depth to go into, but because the specifics can vary so much from organization to organization that exploring all the possibilities here would require too much space. Please treat these bits of advice as starting points, not as complete recipes in themselves.

Dispel Myths Within Your Organization

In organizations that have been producing or using proprietary software for a long time, certain myths about open source software sometimes circulate. One traditional source of such myths is, of course, sales representatives from vendors of proprietary systems. But one can't attribute it all to them. It's just as often the case that someone had some bad experiences in an open source project, or used open source in the past without ensuring proper support channels, and since that was their first experience in an unfamiliar territory, the entire territory is now tainted.

Below are some of the myths I've encountered most frequently. First, the negative myths:

If it's open, that means anyone can change our code.

Believe it or not, you need to be prepared to respond to this. Sometimes people — particularly senior decision-makers who have limited technical experience — don't understand the difference between an upstream codebase allowing anyone to copy the code and modify the resultant copies, and someone modifying the particular instance that you deploy. The former is just the definition of open source, of course. The latter would be a security vulnerability, if it happened, but it has nothing to do with the license on the code. I mention this myth merely to prepare you for encountering it, because otherwise you might not expect that anyone could hold this particular misunderstanding. Trust me, they can, and you need to be ready to answer it.

Open source software is insecure, because anyone can see the code / change the code.

These are so easy to answer that I won't give a detailed refutation here; again, I merely note it so you can be prepared for it. If you find yourself having to explain why open source software is at least as secure as any other kind of software, if not more secure, you may wish to use the excellent resources provided by Dr. David A. Wheeler at http://www.dwheeler.com/#oss, and the arguments at http://www.mil-oss.org/learn-more/security-model-misconceptions.

Open source comes with no support.

There are plenty of companies that sell support for open source software, and they're not hard to find. There are also wonderfully helpful unofficial support communities on the Internet for different open source packages, of course, but often what organizations are looking for is a phone number to call, or a support chat room, that has a guaranteed response time. These are available, it's just that the source from which you procure the software may be unrelated to the source from which you procure the support. The best way to respond to this myth is to ask specifically for what packages support is desired, and then show some sources of support available for them.

If we open source this project, we'll have to spend a lot of time interacting with outside developers.

You open source your code, not your time and attention. You are never under any obligation to respond at all to outside parties, let alone engage substantively with them. You should only do so when engaging will benefit you — which it often will; after all, one of the key strengths of open source is that it enlarges the collective brain of your development team in direct proportion to how much they interact with other developers who become interested in the code. But that engagement is always under your control and at your discretion. If you don't want your team's attention going to bug reports or development questions from the outside, that's fine. Just be up front about that in project announcements and in the documentation, so that others can take that into account before they put a lot of energy into trying to communicate with your developers, and so they can decide whether forking to create a more open community would make sense for them (indeed, sometimes it might even be to your advantage for them to do that).

If we open source this project, then we'll have to release all our other stuff as open source too.

This myth usually results from a misunderstanding of copyleft licenses and the GNU General Public License (GPL) in particular. I won't go into detail here; see Chapter 9, Legal Matters: Licenses, Copyrights, Trademarks and Patents for a discussion of what the GPL actually stipulates. After reading that chapter, especially the section called “The Copyright Holder Is Special, Even In Copyleft Licenses”, you will be able to explain why this belief is incorrect.

Next, the positive myths:

Open source is cheaper.

Licensing costs are often not the largest cost with proprietary software; they are often outweight by training costs, installation and configuration costs, and other factors that make up the "total cost of ownership". But all of those other costs are, on average, the same for open source software. Don't make the mistake of pitching your organization on open source software on the grounds that it is cheaper. At least in terms of the most easily quantified costs, it is not. It is often cheaper in the long run, because it frees your organization from proprietary vendor lock-in (see the section called “Open Source and Freedom from Vendor Lock-In”), reduces training costs for new employees (because they arrive already familiar with the software), gives you greater ability to customize software to your needs, etc. But these are long-term benefits, and they may not show up directly on a balance sheet unless you take steps to make your accounting reveal them. In the short term, open source generally isn't cheaper than proprietary software, and shouldn't be pitched that way.

Developers will devote attention to this code just because we released it.

People with little experience in open source sometimes assume that the mere act of releasing code to the public will result in a flurry of attention from other developers — questions, patches, high-quality code review, bug reports, etc. But what actually happens, in most cases, is silence. Most good developers are busy people, and they're not going to pay attention to your project until they have some reason to. If your code is good and solves a real problem, you can expect word to travel to the right places eventually, and of course you can help that word along with tactically smart announcements and posts (see the section called “Publicity”). But until your code has had time to naturally accumulate credibility and mindshare, most people won't pay any attention, so you shouldn't expect that first release to be a big deal for anyone but you.

There is a situation in which this myth is not a myth. A large organization with a reputation and a dedicated public relations team can create buzz around an initial open source release. If you do this, then make sure not to squander that buzz: be ready to constructively engage the developer attention you attract right away.

Other companies / cities / whoever will pick up this software and start using it right away.

Adopting any software involves costs. Indeed, merely evaluating software involves costs. So when you release a new open source project that you and your team are excited about, that doesn't necessarily mean other entities are going to pick it up right away and run with it. Many of them may notice it, if you've done your announcement process well, but that just means they'll put it on their list of things to investigate based on organization-wide priorities — in other words, they'll take a closer look based on their schedule, not yours. So don't expect a flood of early adopters. You may get a few, and they should definitely be cultivated because they will provide the word-of-mouth that gets you more adopters. But in general you're more likely to see a trickle of early adopters over the first year or so after your initial release, than to see a flood of them immediately when the release is made.

We can casually copy open source code into our own code.

Open source licenses are still licenses, and they come with a few conditions. Virtually all of them require attribution at the source level and inclusion of the license together with the covered code. Some licenses, especially the copyleft licenses discussed in the section called “Aspects of Licenses”, cause the entire derivative work to be under the same open source license thus implying redistribution obligations that you may not want. Some have patent clauses that can affect your company in complex ways.[58] For all these reasons, incorporating open source code into software that will be distributed under a different license — whether open source or proprietary — cannot be done casually. Organizations that incorporate open source code into their products usually need a formal process for doing so, one that involves review by someone who understands the legal issues and the possible interactions between licenses.

Foster Pools of Expertise in Multiple Places

Sometimes organizations that are accustomed to procuring proprietary software treat open source software as if it were proprietary, in the sense that they assume there is exactly one authoritative provider of expert support, and that therefore it is necessary to have a commercial relationship with that provider.

That's not how open source works. One of the great strengths of open source is the availability of support from multiple, competing providers. It's perfectly fine, and often advisable, to have a commercial relationship with just one of those sources, but you must remember that support in open source is fundamentally a marketplace, not an add-on feature that just happens to come with the software license, as is often the case with proprietary software. Actually, even proprietary software sometimes has a competitive support marketplace — think for example of the third-party support providers for Oracle databases and Microsoft operating systems — but in open source these marketplaces tend to be more fluid and not as dominated by single, easily-recognizeable giants, because there isn't necessarily one commercial outfit that automatically assumes a place at the top of the hierarchy to sell gold-label support (e.g., as Oracle or Microsoft themselves would be, in the example just given).

The goal of fostering independent pools of expertise should even affect how you structure contracts to develop the software in the first place. If you hire a firm to develop new open source software, have a few of your own programmers working alongside them if possible, so that you accumulate some in-house expertise. This is not necessarily because you won't want to use the same firm for future maintenance — they might be a great choice — but just so that you'll have a better bargaining position and not be locked in. Essentially, the more people in different organizations who know the code, the better position that code is in, and the better position you are in. This is also one of the side benefits of holding hackathons, as discussed in the section called “Sponsoring Conferences, Hackathons, and other Developer Meetings”.

If your organization does not have enough in-house technical ability to participate in the development process directly alongside your contractor, or at least to perform knowledgeable review, then I strongly recommend finding a third-party to provide independent deployability and maintainability review while the project is under way, as described in the section called “Open Source IV&V”.

Establish Contact Early With Relevant Communities

Another way to foster independent sources of expertise is to establish contact with potentially interested technical communities early and often during development. They're almost always out there. For example, if you're developing software with geospatial functionality, there is an open source geospatial community that probably wants to hear about it; if you're developing software to process financial data, or medical data, there are open source fintech and medical data communities.

You may even have already announced your project to those people when you began, as discussed in the section called “Announcing”. But there's more you can do to create external reservoirs of knowledge. When your project runs across a design issue that you suspect others may have encountered before, it's fine to ask them how they handled it, as long as you do your homework by first finding out what you can from their code and documentation and then asking any remaining questions. You can also arrange small-scale contracts with developers who are active in related projects, to serve two goals at once: improving your project's quality while establishing mindshare in places that may be strategically useful later.

Don't Let Publicity Events Drive Project Schedule

Although open source projects are amenable to software project management techniques, in general if you have an active developer community you do lose some control over the exact timing of events in the life of the project, especially the scheduling of releases. Or rather, you can still have as much control as you want, but now there are other things you can lose if you exercise that control in the wrong way. For example, if the release manager (see the section called “Release Manager”) is someone from outside your organization, and she's doing a good job, then if you try to force the release to be on a certain precise date, you may cause her and many of the developers participating in release-specific work to give up and devote their attention to something else. You'd gain fine-grained control of the release schedule, but at the cost of lower quality releases and the possible loss of some of your development community.

This is just one example illustrating the general principle that if you have publicity needs related to an open source project, you generally shouldn't let those needs drive the project's schedule. If you arrange a press conference for the project reaching 1.0 and being deployed live, but then the developers decide on an extra two weeks of testing because of some last-minute bugs, you'll have some improvising to do. (This example is drawn from real life, by the way.)

There are two ways to achieve this independence, and they are not mutually exclusive. One way is to just let project events drive publicity instead of the other way around, such as by preparing release announcements ahead of time but being ready to publish them based on when the release is actually done. The other way is to create publicity events that are not bound to development milestones, but are rather associated with project-related things that are able to be scheduled, such as conference appearances, hackathons, major deployments, etc.

You might be tempted to try a third way: to bring the development community into the scheduling process, so that through consensus you are able schedule certain milestones accurately enough to tie timed publicity to them. While that may sound like a good idea, in practice it rarely works. An exception to this is if the whole project is on board with doing time-based releases, as described in Time-Based Releases vs Feature-Based Releases. If the development community as a whole shares that goal, then they will make the sacrifices necessary to keep to the time-based cycle — but your organization must also be willing to abide by that schedule, even if it doesn't always align with business needs.

An open source development community's first priority is the software itself, and making sure it meets the needs its participants are working toward. Of course the community wants releases and other deadlines to be met with reasonable regularity, and every development community makes tradeoffs for that. But even with the best of intentions among all parties, you can never guarantee how that tradeoff will be decided in a particular case, when things get down to the wire. The outcome of a community's decision-making process cannot be anticipated with perfect accuracy, by definition — if it could, there would be no need for a decision-making process. So while it's fine to try to influence the community's priorities in ways that work to your advantage, you should avoid relying on that for scheduling purposes, because you won't succeed at it every time.

The Key Role of Middle Management

If you intend to have long-term organizational engagement with open source software projects, the people in your middle layer of management will play a key role in determining whether you succeed or fail.

Supervising programmers who spend part or all of their time on open source projects is more complex than supervising programmers on purely internal projects. Many aspects of the developers' work and schedule will be strongly influenced by external factors not under the control of management, and in any case the developers' own desires may not always perfectly line up with the employer's. After all, each developer now has two unrelated audiences to satisfy: her employer, as embodied by her direct manager, and her colleagues in the open source project, many of whom may work for other employers.

If a manager is not sufficiently sensitive to this dynamic, then developers can start to feel as though they serve two conflicting masters. Sometimes this conflict is the result of poor planning, and other times it is real and unavoidable. Good management can prevent the former situation from happening in the first place, and in the latter situation, good management is essential for recognizing it and addressing it in some way that gives the developer clarity and a way to handle the conflicted situation.

Middle managers also have not only the usual upward and lateral internal reporting responsibilities, but are to some degree responsible for the image — the open source brand identity — of the organization in the projects where its developers participate. This is a lot like having an entire extra constituency to satisfy, and managers who have no experience with open source participation themselves are unlikely to have a solid understanding of how to position the organization and its developers within the project.

Furthemore, the middle layer of management is often best positioned to serve as a communications conduit and information filter between the project (that is, the whole project including all its other participants) and the company. The wealth of information available from the activity in an open source project is most useful to the organization if there is a filtered channel by which the most interesting activities can be communicated to the relevant stakeholders within the organization — stakeholders who might include other technical staff, executives, and sales team members. But both by their position and their temperament, the programmers themselves are often not best suited to serve as this conduit. They may have a very deep understanding of the particular projects they work on, but they often have a less complete view of the organization's interests — for example, in a commercial environment, the programmers often do not have a clear idea of how the project fits into the company's various lines of business or into its sales processes. Middle managers are better positioned to maintain the requisite bidirectional sensitivity: aware enough of the project to ask the programmers for more information when necessary, and aware enough of the organization to have a sense of what in the project might be unexpectedly relevant to the organization.

Think carefully about who occupies the middle management positions that serve as the interface between the organizations's priorities and the open source project's development direction, and provide them with extra training if necessary. It is best if the managers themselves have had direct experience as participants in some open source project. This doesn't have to be the same project as the one for which they are now managing developers; the situations and tensions that arise in open source projects tend to be similar, so experience from one project will generally translate well to other projects. But a manager who has never dealt with open source projects first-hand at all will start with limited ability to understand the various pressures faced by the organization's developers operating in open source environments, and limited ability to play the role of communications conduit between the organization and the project.

InnerSourcing

InnerSource or innersourcing means using standard open source development practices only within the boundaries of an organization. For example, a company might move all of its projects to GitHub (albeit in private, not public, repositories), and declare that, inside the company, any engineer can report bugs and contribute pull requests to any project anywhere in the company. Innersourcing also often includes sincere efforts at internal cultural change: managers encouraging developers to speak their mind on both technical and process issues, developers being given more latitude to choose which projects and teams they work with, etc.

In early 2016 I conducted interviews[59] with open source specialists at a number of medium- and large-sized technology companies, many of whom had observed innersourcing efforts and were willing to talk about the results. What they reported was pretty consistent from company to company, and consistent with my own experience as a consultant: innersourcing really can make a positive difference, in several ways, but it's also definitely not the same as true open source.

For companies that already participate in open source projects, innersourcing can reduce the difference between internal development practices and external ones. If some of your engineers participate in upstream open source projects anyway, where they must use typical open source collaboration tools and adhere to open source standards for submitting and reviewing code and documentation, then moving the company's internal engineering infrastructure and conventions in that direction means less context-switching overhead for existing staff, an easier onboarding process for new hires, and sometimes improved technical compatibility between internal and external projects. (For these reasons, innersourcing is also often used as the first "baby steps" toward genuine corporate participation in open source projects.)

But the benefits of innersourcing go beyond that. When accompanied by a real commitment to reduce managerial and organizational barriers to engineers participating in projects across the company, innersourcing can improve morale, help spread expertise around the company more effectively (because expertise will come closer to flowing along its natural paths, rather than being directed by management), and make software development more efficient.[60]

Nevertheless, innersource is not the same as open source, nor is it even "open source lite". The managers we talked to reported that innersourced projects don't have the provocative, uncontrolled energy of truly open source projects, because all the actors in innersourcing are, ultimately, embedded in the same hierarchical authority structure. Fundamentally, open source dynamics require at least the potential for totally permissionless modification (i.e., you don't have to worry what someone else might think of a fork). When software only circulates within a given management hierarchy, then that potential for permissionless collaboration vanishes — and with it, the potential for true open source behavior vanishes too. The permission structure that governs one's behavior with respect to the code is not just a matter of the code's license: it's also about whom you report to, what others in the hierarchy might think about your changes, etc.

In the long run, the dynamics of open source collaboration require an external supply of freedom. There must always be people who could, in principle, fork or do whatever they want without worrying about consequences to the original authors' organization. When that external freedom is removed, everything changes.

Innersourcing also fails the "portable résumé" test — an employee can't take the code with her, and her work will not be publicly visible (see the section called “Hiring Open Source Developers”). She can be alienated from her work, if she leaves the company, which means that her motivation to personally invest is reduced.

None of this means that innersourcing isn't worth it. It can be very beneficial on its own terms, and is also sometimes useful as an intermediate step for a traditionally closed company that's still figuring out how to do open source participation. Just don't imagine that innersourcing is somehow "just like open source, but inside our company". They're two different things, and shouldn't be confused.



[58] I am strongly opposed to software patents of any kind, for the reasons given in the section called “Patents”, but if you are a patent holder I would still like you to at least be aware of the possible patent consequences of incorporating open code into your programs.

[59] Actually, my friend and business partner James Vasile and I both conducted these interviews, and we were much aided by O'Reilly Media providing introductions to open source staff at a few companies where we did not have personal contacts.

[60] If you're interested in learning more, see http://innersourcecommons.org/, where Danese Cooper, Head of Open Source at PayPal, has been organizing a number of resources about InnerSource.