Slashdot Mirror
eXtreme Programming (XP) in OSS projects?
ivi asks: "I first bumped into the XP approach to system development, some years ago, on Dr Dobbs' now-defunct Seminar-On-Demand site TechNetCast. XP has some short, simple rules for growing software from Users' Stories, eg, with programmers working in pairs,
showing prototypes to Users "early & often" et al. Download this XP site (under 400KB, zipped) for more.
So, who's used or using XP, does it work for OSS projects & what have your experiences been with it, so far?"
We tried XP approaches in my college programming classes. More often than not, it caused my partners and I to claw each others' eyes out, and generally despise your partner for the remainder of the project.
What worked MUCH better was designing the framework from the beginning, outlining how the different functions/processes/data structures etc. would have to communicate with each other, then splitting the programming tasks up.
After every one made progress, we'd recombine, work out bugs, etc.
Much more organic and realistic than the XP approach, which always seemed very articifical and limited.
I remember reading an article on O'Reilly about XP and Open Source .. go dig it up..
:-).
I don't have any experience with XP in writing open source projects, but I sure see a lot of projects that can benefit from it!!
I would love to see more software with comprehensive tests. It's very satisfying when you hack a minor change into a program and all the tests still pass. Then you can send the patch off to the author knowing it won't break anything (well, at least it won't break any of the tests
In this vein, I believe test-driven development (i.e., write tests, write any code that passes the test, then refactor to clean up the mess) is the most amazing gift from whoever the programming gods are. I'm glad XP has (re-)discovered it.
It forces much simpler, cleaner interfaces, and complete unit tests. There have been several times when I've wished that a particular open source program was split into many small objects that I could mix and match (i.e., I love this package BUT I need to replace the code that saves files to disk with code that saves files to a database..etc).
With TDD your design splits into many small simple classes. I think this would be great for OSS projects. A good example of a well-factored library (off the top of my head) is Mime::Tools for Perl.
Too many OSS developers think "I used Python, I must be doing object-oriented programming" or something along those lines.
Some other aspects of XP such as continuous integration seem useful.
Another thing about XP techniques (especially test-driven development again), is you can code in small bite-size chunks and not leave the code in a broken state. I think this is great for projects you might have "on the side". You can work an hour or two and your code will always be working and passing tests, none of these "egg-juggling" 10-hour coding marathons. You know what I mean, when you change something and it ripples to 20 files and you keep all the things you need to do in your head and you go half-insane running from file to file.
Pair programming, user stories, "Customers", etc., seems less useful.
But yeah test-driven development is worth trying, at least.
My initial impression of XP (I haven't tried it but had it explained to me) is that it takes a lot of the fun and programmer-as-artist aspect out of software development. Also it is good for making a larger number of so-so programmers as productive as a few really good programmers.
However, most OSS projects seem to be labors of love by groups of individuals working together in their spare time. I would bet most of the largest contributors to OSS projects are very good programmers who don't need to be "managed" to the extent that XP does. I don't see how many of the XP ideas could be applied when none of the developers are in the same room on a regular basis (or even available to meet on-line at one time), their time commitments are unpredictable, and in many cases have big egos and work on what they feel like, not what the customers ask for... After all, programmers have day jobs that are like that!
Something like... M'aud'ib! -ib! -IB! -IB! -*-KABOOM-*- ! (-:
Got time? Spend some of it coding or testing
Having participated in a number of OSS projects, and even led/maintained a few, I can't see how Extreme Programming can help - the model is clearly better suited for in-house programming than distributed programming.
Many of the same techniques are used in OSS projects though. "Release early, release often" is almost biblical to people from what can soon be called a Linux-background, and another rule is "No functionality is added early.", which fits my personal design philosophy pretty well, and a lot of OSS projects: I can't remember the last time I added functionality to a program.
What happens is a company has a bad experience with programming -- say they fire some old guy to hire a cheaper new guy and discover the new guy can't make head or tails of what is going on. Of course the new guy says this is due to lack of documentation. Or a department realizes that some software doesn't do what they needed, because they never told anyone what they needed and no one asked. Or the realization that a key piece of software that is keeping the whole company running is a mystery to everyone who works there and the source code is long lost.
Well, managers can't really get paid the big bucks for saying "Hey guys, next time let's try to be smart" which is the real solution. So instead they invent a plan, procedure, or rules to prevent it from happening again. Everything you do must be documented. Tens of thousands of dollars on verson trackers. A flow chart describing a series of meetings to produce a requirements document and another flow chart for modifying the requirements document.
Well, pretty soon people realize you have a bunch of people filling out TPS reports and one productive guy who is evading all the documentation and requirements procedures and just getting work done, maybe protected and covered for by a few co-workers and a boss. And someone thinks, jeez, if only everyone was like him.
But you can't just have the managers go out and say, "Alright, bad idea, toss it all out the window, let's just focus on common sense, communication, and being smart." That kind of implies you could hire fewer managers and just spend money on trying to get smart people.
So you have to package the "common sense" proposal in a bunch of buzzwords that make it sound like it's a new 6 sigma or total quality or some other plan for extracting the most out of dumb people by telling them to do smart things. This way you can slip it in past the Organization Men.
One of the most pathetic things you can ever see is a stagnant ossified department, which long ago chased away anyone with any ambition, attempting to implement pair programming and other feature of common sense. If you program in pairs, it's harder to say that person X did this and person Y did that, which managers see as a threat. Talking with the final consumer also undermines certain people's positions, so they usually interpret that "show prototypes early and often" as instructions to go through a formal process of beta release, evaluation, discussion meeting, plan for addressing issues, FASTER. Not fewer levels of BS and more communication, just do it harder and more often.
Open Source projects are usually XP in an extreme way, except for pair programming. Almost all code is written by lone wolfs and casually reviewed by others if at all. Other than that, the quick-release and review, constant prototyping, etc all just naturally happens. Even more so since the programmer is usually the user he is trying to satisfy, which is the tightest loop there is.
If your open source project is being developed by a team of people all sitting in the same building, directed by a customer who's paying for the software and who accepts XP, then yes, XP should work just fine.
If your open source project is being developed by volunteers all over the world in different time zones, then forget it. Getting XP to work in that situation is probably possible, but difficult and probably expensive.
But if you, like so many others, are using XP as an excuse for shoddy development practices (the "look, we don't need to do requirements analysis or design or documentation because we're using XP" syndrome), then by all means go ahead. It'll work just fine as an excuse.
So let's get to the details. I understand XP pretty darn well. We deployed it successfully at my shop, making all the classic mistakes (and some of our own) on the way. We finally got everything right, and once we did, it worked really well.
I'm going to assume that "open source project" means a project with volunteers all over the world (or all over the country), and no paying customer.
Problem 1: The customer role
The "customer" in XP is a person who represents the users and can make decisions regarding what to implement and when. It's not necessarily a paying customer or an outside person.
The customer is a key person in XP. If you don't have a customer, you can't do XP and if the developers don't do what the customer requires, you can't do XP.
Developers don't make good customers. They're too involved in the technical side of the project and rarely prioritize well. You really need an outsider who is focused only on the end result.
Developers have to listen to the customer. The customer decides what gets done. The developers have no choice in the matter. They can tell the customer how much it will cost to get something done, but in the end they have to do what the customer wants.
Do you have a customer on your project? Will your volunteer developers do what the customer wants? If either answer is no, forget about XP.
Problem 2: Pair programming
Pair programming gets left out of a lot of XP projects. That's unfortunate, because pair programming is a key ingredient of XP. Without it, the process hobbles along.
There are lots of reasons why people give up on pair programming. The poster mentions one: not leaving that big programmer ego at home (it's just programming, not personal). There are others.
What you need to understand is that pair programming is how knowledge and skills are communicated in an XP project. Pair programming compensates for lack of formal design and documentation. Pair programming cuts down on both trivial and serious bugs (the brain not writing the code is usually thinking about the big picture). We found that code written by pairs was consistently better by all subjective and objective criteria we used.
Can you do pair programming in your open source project? If your developers are all in one place and work at the same time, then it's easy. But if they're not, you'll find it very difficult. Instant messaging or even speakerphones don't really help much. And remember, pairs are supposed to be unstable. If you always work with the same person, you'll lose a lot of the benefits of pair programming.
Problem 3: Communication
XP developers tend to be pretty chatty, in my experience. At our shop, verbal communication was a very important component, and to ensure that it was easy, we all worked in the same room (no partitions at all).
How easy is it for your developers to communicate? Instant messaging helps, but isn't all that good since you can't count on instant feedback. A phone works pretty well, but it should be a speakerphone (a headset is OK, but less effective). Are your developers working at the same time? If they're not, how are they going to talk to each other while they work?
In conclusion . . .
T
Perhaps it was counterproductive for you, but I really like it and lots of other people do too.
Some of the other rules in the article are nice in theory, but not really practical in real life. For example: customers aren't always available, and often don't want to be either.
I would not want to work on a piece of software where the person paying for it refused to be around to help me figure out what it's supposed to do! If I just have to guess, then I'm going to be wrong and nobody's going to be happy.
Involving the customer too much usually ends up in them wanting something other than what they originally agreed to, but within the same time frame and for the same price
EXACTLY!! You've perfectly described the whole reason for XP. There's a reason XP's motto is "embrace change". Customers very often change their minds. You can say, "I know you're paying me, but too bad, you said you wanted X and that's what you're going to get." How does that help anyone? I like to say, "Sure I can change it."
(and then they get upset when they can't have it).
But they can have it! They are the ones who are paying for the software, and XP is all about letting them get what they want.
The first thing I did were to picture the whole system as a group of functional boxes with interfaces. The next were to define the interface mechanism. Then I grabbed my nearest bosses and made them customers showing them a new release every two weeks. I distributed the tasks of implementing the functional boxes among the programmers in the group who I placed in the biggest room awailable facing each other. In the group we used nightly builds to always have the CVS build clean. My bosses were forced by us to prioritize among tasks and give us input for the next release. One day over a year later we could experience a decreased pressure and a product that we felt confident about. At that point we started to refactor all bits and pieces that were just implemented the easiest way as well as the build system. Now another two years down the line we have catched up and we have got rooms, some still with two programmers and we are increasingly using a traditional way of programming, with a spec phase, an implementation phase and a QA phase etc. We only kept some featured of XP that we like, nightly builds and refactoring.
I think that XP has its cons and pros depending on the situation. We could never have delivered all these demos, prototypes and releases during the course of these three years without XP, simply because we had to attract customers and venture capital all along. We didn't use XP to the full extent either, no automatic testing and no true pair programming but that was only because we didn't have time to change ourselfs. If you sit on a fully financed military project for instance I doubt that XP will work for you. If you lack time or funding or simply needs to go extremelly fast to something that works XP might be the way to go, just ensure that everybody in the project agrees.
Good Luck!