Collective code ownership is limiting software quality

Reading the title of this posting you might think I was out of my mind. But in fact I really mean what I say: After lately thinking quite a bit about the software development process I´m now pretty sure, collective code ownership (CCO) as it is promoted by eXtreme Programming (XP) [1,2] is at least not just plain right.

Why´s that? Well, it´s because also CCO is just a coping technique. It´s an answer to certain problems and relies on a couple of premises. Here´s my guess what at least some of these XP premises are:

  1. Software development teams are highly volatile: For XP software teams seem to be ever changing. People constantly coming on board, leaving the team, being sick or run over by a truck ("truck factor"). You simply can´t rely on a person to be there for a longer period of time to assign him responsibility for any part of a project.
  2. Changes to software need to be applied immediately: If a bug is detected or if a customer requests some new functionality, the necessary changes to the software need to be done right away. Since the team cannot rely on any one developer to be there who might know best how to do this, every team member should be able and allowed to jump in. Waiting for a responsible person to be available again in reasonable time (depending on the issue) is no option.
  3. Specialization of developers is contraproductive: Developers who specialize themselves in certain technologies (e.g. GUI, database, cryptography) or disciplines (e.g. testing, integration, design) are of limited use to a software project. They just know who to do best one or two things and necessarily are mediocre in doing other stuff. That makes them hard to be assigned to arbitrary tasks on a software project. This makes it hard for them to jump in for other´s who are not available when some issue needs to be taken care of right away. Any kind of specialization thus would limit a team´s flexibility.
  4. Software quality is primarily determined by functional correctness: As long as changes to a software don´t break the phalanx of primarily functional unit tests all´s well. How the structure of some part of the code is changed by some developer or how a technology is used to quickly apply a change, is not relevant.
  5. There is hardly any difference between developing a new software and maintaining and existing software: XP does not distinguish between software lifecycle phases like "development" and "maintenance". Rather it´s continuous analysis, design, coding, testing. Thus throughout the whole life of a software the same software development skills are needed.
  6. Knowledge of implementation details should be maximized: The more details every developer on a team knows about as many parts of the software the easier it is for him to jump in and apply some change to any part. The better his understanding of the intricacies of the implementation, the better his judgement about the impact of a change.

As you can imagine, since I´m not in favor of CCO, I probably have a problem with one or more of the premises behind it. And you´re right: My thinking is, the premises are just trying to cure symptoms of some fundamental ailment. This ailment is a mixture of cultural ideosyncrasies and global "customs" of the software development industrie. Let me sketch the ailment to which CCO or XP tries to be a remedy by listing some common thought´s of someone suffering from it:

  1. I better not rely on any developer, because he might be gone tomorrow. (see premises: 1,6)
  2. I better be in full control of everything! (see premises: 1,2,3,6)
  3. It is possible for me to know and understand every detail of the software as well as the problem domain. (see premises: 2,3,6)
  4. It is possible for me to know all relevant technologies for the software sufficiently to choose and apply them for optimal quality. (see premises: 1,3,4)
  5. The customer is king! His wish is my immediate command. (see premises: 2,5)
  6. Software is so different from other stuff, I can´t rely on practises from other industries. (see premises: 3,4,5,6)

Let me put it very bluntly: Looking at these thoughts and the premises my impression is, CCO (and maybe XP) is for people suffering from a lot of fear (a, e), who need to be in control all the time (b, c, d), and are convinced they can be in control all the time (b, c, d, f).

Thoughts a. and e. for me seem to be "very inspired" by the American way of business. Thoughts b. and d. are prevalent among software developers around the world. The reason probably lying in the nature of software development which suggests maximum control over an infinitely flexible material (software) and thus attracts people in search of control. Thought f. is common not only for software developers but also laymen. The abstractness of the subject suggests it is so different from anything else mankind has dealt with so far, it is hard to apply best practices from other industry to this new one.

(Ok, ok, I don´t think every developer should get an appointment with a shrink :-) let me say: The "will to control" is very basic in all of us. It´s like a drive - and software makes it easy to satisfy. So whoever has an inclination to abstract thinking or is more on the rational side of life is prone to fall for software development. In addition, seeking control stands in the line of the successes of the 19th/20th century in several areas. For many it´s plain obvious almost anything can be controlled - so why not strive for control?)

So, what do you think? Is this the case? And if so, is it the way it should be? My guess is, the software industry should overhaul their thinking. Of course that´s not easy. The ailment is hard to cure, because it´s deeply ingrained in our "being as software developers". We grew up at with thoughts c. and f. at least and liked b. at lot. And early successes in software development might have fueled thought c.

But that´s plain wrong.

It´s contraproductive and stands in the way of higher quality software.

My two main premises are:

Software quality requires specialization: Number and complexity of software technologies and tools will continue to increase. I think it´s obvious, that it becomes ever harder to stay on top of current tools and technologies. 20 years ago database programming was plain simple and required knowledge of one API. Today it´s several APIs, languages, programming models (e.g. ADO.NET, O/R Mapping, SQL XML, SQL), plus the intricacies of database products like SQL Server 2005 (e.g. SQL CLR, T-SQL, Web service interface, SQL Service Broker, SQL Reporting Services). And what about (G)UI programming? There is at least WinForms, WebForms, AJAX, Flash, and WPF around the corner. Add to that a couple of new Smart Client frontend options VSTO and IBF and I guess you see what I mean. It´s impossible for a single developer to know all those and many, many other technologies well enough as to be able to apply them with profound understanding or even choose the best one for a certain problem scenario. Programming no longer is about knowing a language, an editor, and a debugger. Programming is about masterly weaving together a host of complex technologies. That´s also one reason, why software architecture is on the rise. There needs to be someone with an overview of all this and coordinating the weaving: the architect.

Also the number and complexity of software to develop will continue to increase. Customers are demanding and will continue to be so. They want faster, more scaleable software with slicker and more usable UIs on more devices than today. Fulfilling customer requirements will only be possible by deeply understanding and exploiting the new technologies and tools. Whether a customer requirement really, really makes sense is an altogether different question. Even if some cool feature is not really necessary for some user, it might be a distinguishing feature for your software or your company on the market. Employing more technologies, covering more device options, implementing more features, providing higher flexibility, and integrating with existing systems all require specialization along several dimensions (e.g. technology, problem domain, form factor). Resistance to this trend is futile.

Coping with complexity requires "not knowing": Information hiding is a well established notion in computer science. The whole object orientation paradigm rests on it. Then components as black boxes are also well established (even if not yet used to their full potential - but that´s a different topic). Services (as in service orientation (SO)) are the next black box generation knocking on the door. So I´d say  the history of software development among others from subroutines to services is a history of raising the level of abstraction when dealing with code, which means, it´s about "not wanting to know everything all the time": When you "glue together" components or classes you simply don´t want to be concerned with their implementation details. Why is that? Because your capacity for details/system parameters and dependencies is limited. Black box thinking and system decomposition into a hierarchy of subsystems is a way to deal with complexity. The more complex a system is, the deeper the hierarchy of black box "units". Each black box then is described by a specification which hides the box´s intricacies. The specification is a promise and thus a invites trust. It say "Trust me, trust the black box to do the job as specified. Don´t worry, you need not be in control here, you can focus on more important stuff." Trust thus is at the heart of dealing with complexity. Trying to know all details of a complex system is futile.

Based on these premises I´m arguing:

  1. Software becomes more complex; technologies and tools become more complex. To provide highest quality specialization is needed. Specialists should not interfere with the work of other specialists. 
  2. Specialists require clear units of code to apply their knowledge to. Clear units of composition on several levels of abstraction also make a complex system easier to understand. To benefit most from those units they need to follow the low coupling/high cohesion principles. Low coupling requires limiting the knowledge of unit internals.

Since CCO is favoring generalization instead of specialization, and since CCO is favoring spreading as much knowledge about implementation details as possible, well, I think CCO stands in the way of higher quality software.

Of course, no every software will employ every new technology or feature. But in general nobody can escape the trend towards more technologies. And to apply those technologies to produce high quality software, a lot of expertise and experience is necessary. To believe, any good programmer just needs to sit down a couple of days or maybe weeks to learn a bunch of technologies to be able to use them like an expert is plain wrong. Sure, a good programmer will learn a lot about those technologies - but it will take much longer to become an expert. Also, simply to choose which technology to use at all for a given problem (e.g. use async delegates, MSMQ, Queued Components, SQL Service Broker, or Virtual Shared Memory for async processing) requires already expert knowledge, since the implications of choosing one technology might me far reaching.

No other industry has this ideal of omniescient "engineers". Building a computer, car, house, airplane, or producing a movie all requires the interplay of a large number of specialists - who are all perfectly happy to be "limited" to a certain area of the product planning or building process.

However, there is one area, where generalists are still needed in other industries: maintenance! One a house has been built, once the car has been manufactured generalists are taking over from the specialists. A car repair guy or a janitor are examples of generalists. They know quite a bit about the respective products, they can repair them to a certain extent, they can assess damage. But they also know their limitations. If the damage is too large, they call in a specialist again to fix it (or the product needs to be replaced).

So I guess, also for software we need to distinguish between the production and the maintenance phase of the lifecycle when talking about developer responsibilities. During initial development I strongly believe we need to move to more specialization and clear responsibilities. Then, after release, the software is handed over to generalists doing further bug fixing and also adding new features within reason. If features require deeper knowledge of certain technologies, then specialists have be called in again.

You say, this won´t work? Well, I´d like to ask: Has anybody tried? I don´t think so. Nobody has tried to do it like this, because so far only a very limited notion of specialization has established itself in the minds of the developer community. And hence projects are not planned and organized accordingly. There probably is not even a capable enough "ecosystem" for projects and developers to develop software in this manner.

Although my position is somewhat a priori and just resting on experience in other industries (which probably cannot be carried over 100%), I´d say, any flat rejection also is a priori, because the whole situation our industry is in is so new, nobody can really say "We´ve tried it, and it does not work." So what we need are empirical studies. We need trials which compare several approaches. But already now I´m pretty sure, CCO is not the future. It cannot be, because technological advancements and rising complexity cannot be delt with by keeping up the illusion, omnicience and full control for everyone on a team was possible.

I think, I understand where XP and its CCO is coming from. I really sympathize with them. I´d love the world to be like CCO sees it - but it is very different (or at least becoming different). I too love to have the impression, I know every detail about a software and be ready to solve problems in every part. I love the feeling to be on top of all sorts of technologies. But more and more I have to admit: I cannot know "it" all anymore. I´m under the impression I used to in the 1980s or even maybe until the mid 1990s. But today... I have to focus more and more. I have to not (!) read a lot of interesting articles and books simply to be able to become (or stay) an expert in at least two or three technological areas.

I would feel really, really bad, if I´d tell a customer, I was capable of deliveringy a high quality software product all by myself (assuming a moderately sized project). I could not. I could not for not being an expert in so many technologies which potentially could be used in that software product.

Maybe you feel different. Maybe you think you can handle it all. Maybe you can stay on top of dozens of technologies and also an entire problem domain to be the "jack of all trades". Maybe you´re even working in a team, where all your colleagues are of your caliber and you´re all smoothly sailing along delivering highest quality software.

Well, then I congratulate you! No, I even envy you and your colleagues. And I´d like to learn how you do it. CCO for you sure is the best way to handle your code base.

But what with the rest of us mere mortal software developers? I don´t think we ever can aspire to such lucidity. I think we´ve to develop a coping strategy for fighting technological diversity and rising software complexity. And my guess is, CCO does not help. Taking a long term perspective, I´d even say, CCO for us mere mortals stands in the way of higher software quality.

Literature

[1] Explanations of CCO, http://www.xpexchange.net/english/intro/collectiveCodeOwnership.html, http://www.extremeprogramming.org/rules/collective.html
[2] Interview with Ward Cunningham on CCO, http://www.artima.com/intv/ownership.html

Comments

# re: Collective code ownership is limiting software quality

Sunday, April 2, 2006 1:50 PM by Jeremy

I've found it much easier to do architecture within a team that practices collective code ownership. More team members have a greater grasp of the system as a whole because they're rotating through different areas of the system. That makes it much easier to communicate architectural ideas to your team members because they better understand the context of the decisions. It also seems to enable more people in the team to positively contribute to the architecture discussions.



And I'll add one more reason to your alphabet list:

g.) Can the team continue to function when a key team member is out for two weeks.



# re: Collective code ownership is limiting software quality

Sunday, April 2, 2006 4:18 PM by Ralf

@Jeremy: I´m not saying, an architect should share his thoughts with the team. To involve the team during designing the whole system, modelling its components is good. And since I´m in favor for clear responsibilities, I think whenever contracts between components need to be "negotiated", then the "owners" of those components should be the ones to do it unter the supervision of the architect.

So the more responsibilties a developer has and the more the architect invites the team to "review" the architecture, the more everyone knows about the whole system.

But this positive knowledge about the whole is different from collective code ownership which means extensive knowledge of implementation details. This extensive knowledge runs contrary to any decoupling efforts during modelling.

Again: I´m all for communication and collaboration, asking co-developers for help, opening one´s thoughts to others. But that´s different from inviting everybody to work on any part of a system.

As for your thought g.: 1. How do other industries deal with this situation? 2. Why not plan for such situations and assign a backup developer? 3. As I suggested: Maybe even other developers can touch code in another responsibility domain - but then they are guest and behave like that. The never own it. And the owner - once he returns from his two weeks vacations - is responsible for reviewing whatever others have done to his code.

Maybe the software industry has to ackownledge it´s trying to do the impossible: build very, very complex products with a wrong view on the ressources needed.

You can´t have the cake and eat it: use evermore complex technologies to build evermore complex stuff - and use a project organization like 30 years ago. Sure, Agile Methods are to the rescue everywhere... but can they defy pretty simple logic and experience in other industries?

Software development is not even on par with other industries when it comes to using "components". Loose coupling, contract first design, IoC - it´s not widely in use. So I´d say: Since XP is not even about component orientation how can it make a statement about a component oriented production process? Components are about boundaries - in any industry. Collective code ownership does not recognize boundaries, because it favors free diseminattion of implementation details.

So CCO seems to be in contradiction with component orientation.

Please get me right: I´m all for flexibility and openness to contributions from any team member. But does that necessarily entail the absence of clear responsibilties?

But I´m getting carried away here... ;-) Also, the issue of non-CCO needs to be seen as just a part of a bigger picture. Software development needs to change in more regards - but the reason is all the same:

Software development needs to change in order to cope with higher complexity, higher diversity, more options, faster evolution, higher demands. Just advancing some tools and trying to keep everybody as informed about everything is futile.

How we organize projects, how we model software, how the market pays for software, how developers organize themselves - this all has to change - and I´m sure it will change within the next 10 years.

# re: Collective code ownership is limiting software quality

Sunday, April 2, 2006 11:09 PM by Jeremy Miller

All I meant to say Ralf is that it is actually easier in many ways to be an architect on a team doing CCO. I wasn't trying to imply you were against the communication, just that I think communicating an architecture is easier inside a team that practices CCO.

Being able to build a system with loose coupling is mostly a matter of the technical skill and knowledge of the team more than any particular process or practice anyway. I would contend though that the practice of TDD within XP enforces/promotes the usage of loose coupling. XP certainly goes much easier when you take advantage of IoC/DI and other strategies for loose coupling.

I think the assignment of people to a specific part of the code versus CCO to be orthogonal to defining a solid separation of concerns. One thing CCO definitely contributes to is the elimination of duplication. I think knowing more about the system as a whole makes it easier to put related functionality in a consistent place.

As far as software trying to borrow from other industries, at some point I think we need to realize that software is its own thing. Trying to apply practices from manufacturing or construction too literally doesn't work. I was an engineer before I went into software development. The behavior of steel is easily predictable, software isn't. We need to create our own canon of software wisdom.

Now, whether or not a team practicing CCO needs to have a clearly defined architect or technical lead role, that's a different subject I'll leave to everyone else.

All in all a thought provoking post.

# re: Collective code ownership is limiting software quality

Monday, April 3, 2006 4:03 AM by Ralf

@Jeremy: I fully understand, it "feels easier" ;-) to be an architect on a CCO team. But I´d argue this so only at the beginning. The reason: Once the team starts and you do not have even have a microkernel framework and to strictly separated development of components, then the coupling of what you´ve design will start to increase. Slowly, but steadily as the developers carry over their knowledge of implementation details of one part of the software into all sorts of other parts.

So what I mean is: if software development agrees on components and loose coupling as having large value... then it´s only very consequential to support this by clear responsibilities as well as clearly separated development of components (in their own testbeds).

I like TDD quite a bit - as a "prototyping technique". TDD for me is useful to shape component contracts once you know which components a software system consits of. But TDD itself does not enforce loose coupling, I´d say.

Hm... why is "assignment of people to a specific part" orthogonal to CCO?

Avoiding duplicate functionality should not be something that somehow happens, but be the result of a planning process. But of course, duplication might creep in over time, so refactoring is necessary. But CCO is not - rather, it stands in the way of planning, I´d say.

Sure, not every practice from other industries can be applied to software development. And software development is different. But my feeling is, software development is so happy to be different, that it´s trying to resist the transfer of experiences from other industries more than is good for it.

Software quality cannot be calculated like the behaviour of steel or concrete. Software quality can only assessed empirically by testing, testing, testing. And of course XP thus is right by emphazising testing and refactoring. But CCO does not help here.

CCO and an architect role are orthogonal, from my point of view. Nevertheless I´d say: a software architect is a necessary role.

# re: Collective code ownership is limiting software quality

Thursday, April 6, 2006 8:10 PM by Maruis Marais

M8, I think you are missing the point of CCO completely. You are mainly talking about technology and frankly, no one developer will ever be an expert in every technology out there today. The field of software development is too big and wide for any one developer to have an expert understanding of the whole discipline. I personally believe you should have a good base to work from (Good OO understanding) and a general understanding of technologies available. Now, at the same time the team should spread the technology research between them.

In our company, we might have four or five projects running at the same time. We have SQL experts, but they also have general programming capabilities. Thus giving our team the ability the share the unique knowledge between project teams as needed and at the same time spreading the SQL knowledge through pairing with non-SQL experts when implementing specific SQL functionality. Thus, the project does not have a bus factor at any point. We have a concept of assigning specific technologies the company thinks it needs to invest in, to technology groups. These might be three or four individuals in a group to research and become experts in a specific technology.

On another point, when using TDD to drive your implementation and code, at some point if you have done it incorrectly, you will hurt. Why? Because you will have HIGH coupling between your object and you will start to feel the pain of fragile, breaking test. So what does this mean? Developers today still write procedural code. Go look at some of your code that you’ve written maybe yesterday, run a metrics tool like SourceMonitor over it, and if you find any code above 10 in Cyclomatic Complexity, I can guarantee you’ve written procedural code that is breaking some sort of OO rule.

The way TDD forces you to write loosely coupled code is through having to replace external dependencies in the class under test with Stubs or Mocks using Inversion of Control and Dependency injection. Moreover, because Mock frameworks normally can only mock interfaces, this will force you to communicate to the external component through an interface, which means you have low coupling between these two components. Knowing your Design Patterns will also help you to practice TDD more efficient.

On another note, I think you are also missing the relationship between the XP practices and the roles these practices play in supporting each other. If you look at CCO in isolation, you are missing crucial benefits you are getting when you look at CCO in context to the other practices. What do I mean? Well, all the practices of XP support each other and as a whole make developing software easier and more productive, reducing the risk of failure. Maybe you should have a read of this essay explaining this in more detail.

# re: Collective code ownership is limiting software quality

Sunday, April 9, 2006 4:11 PM by Ralf

@Maruis: I don´t understand your argument. You support mine by saying "no one developer will ever be an expert in every technology" and go on to describe how your company keeps expert groups - but you don´t follow my conclusion. Why? If there are experts, then they naturally cannot (!) be qualified to work on each and every detail of a project. And just because experts hand down some knowledge to others by pari programming, the receiving developers don´t become experts themselves.

I´m also in favor of decoupling components and using IoC/Microkernels and Mocks. All great stuff. But what does it have to do with CCO?

If you´re a fan of TDD, great! Go for it. It´s one way of finding the best signatures for the interfaces you arrived at during modelling. Unless, though, you don´t explicitly model components (and not classes), I doubt your software has a sound architecture. But that´s a different matter.

Why I´m missing any relationship between XP practices? I´m not concerned with pair programming, refactoring etc. I was just focussing on CCO - which does not become just right by being tied in with other XP practices. And others are not invalidated just because I don´t think CCO is of much value.

I like XP - but XP is (or at least should not be) written in stone.

# re: Collective code ownership is limiting software quality

Tuesday, April 11, 2006 12:55 PM by Scott Bellware

Ralf, rather than speculate at length about Agile and XP, why not actually earn some practical and empirical knowledge and experience?

You can’t guess at the nature of XP from a waterfall perspective – it’s like trying to figure out quantum theory from a perspective where the smallest unit of material decomposition is a molecule.

# re: Collective code ownership is limiting software quality

Tuesday, April 11, 2006 1:06 PM by Jeffrey Palermo

On an XP team, CCO is a must because an XP team goes fast! An XP team pairs, so knowledge is transferred quickly. An XP pair can change any code in the system. There is no "That's John's code". It's code, and it's in the system, and it needs to be changed - done.

I don't understand why you try to keep secrets about the code on your team. I can't imagine how your team benefits by confining a single developer to a small part of the codebase.

# re: Collective code ownership is limiting software quality

Wednesday, April 12, 2006 10:31 AM by Ralf

@Rajesh: 1. I understand that "the idea is that everyone is responsible for the project". And that´s of course true. But it´s true for any project. This is the essence of the term "team". But responsibility in general does entail any special division of work.

Although it sounds nice, truely responsible programmers communicate changes before they apply them, I doubt this is universal practice. Given the fast pace of software projects you´re describing, there simply is no time to wait for tomorrow´s stand up meeting. Whoever thinks he knows better how to do something will simply do it.

2. "Never be blocked" sounds very good. But "good feelings" don´t sound all too good to me. Limited knowledge (let alone naivity or even hypocricy) often generate "good feelings", but not good quality.

Also, please read my posting carefully: I´m not saying, nobody must touch other´s code. I´m saying: There should be clear responsibilities and non-owners should behave like guests and all changes by non-owners need to be reviewed by the owner. I urgency requires it, a guest might change another person´s code - but this should be kept to a minimum due to two main reasons: 1. The guest might not know enough about the technology he´s using. 2. The gues learns details purposedly kept hidden behind a carefully designed interface.

Also: Please explain to me, why other industries don´t have a problem with specialization and clear responsibilities? Why does it work in lean car manufacturing all over the world?

3. "Specialization of developers is contraproductive" is a claim not proven, not even by the text you´re citing. Likewise - I have to admit - my claim also hasn´t been proven. Why haven´t you´re and my claim not been proven: because empirical studies in software development are scarce.

By the way: Heinlein´s view of what a human being should be able to do seems quite naive. Also he does not recognize that being a human in itself already is a specialization. In addition Heinlein neglects how far specialization has carried insects compared to humans and what the scenarios for a post-human world are.

Maybe I should answer Heinlein like this: Generalization is for settlers conquering a new world.

4. I´m not saying "quality" is primarily determined by "functional correctness". Right to the contrary!

6. Please explain the connection between your "knowledge of implementation details should be maximized" and the common architectural recommendations of black box thinking and loose coupling and information hiding?

-Ralf

# re: Collective code ownership is limiting software quality

Sunday, September 3, 2006 6:52 AM by maruis@xtra.co.nz

There are some good comments on this post. Really interesting viewpoints that is worth the read.