Slashdot Mirror
Extreme Programming Refactored
The Previous Extreme - What XP Set out to Fix It had previously been accepted practice to spend months (years, even, on large-scale projects) gathering requirements, then another year or two on design, before a single line of production code had been written. The infamous "big bang delivery" occurred when this gigantic monolith of a software system was finally delivered to the customer, only for the customer to retort that this was nothing like what he wanted.
It was also accepted practice to divide the system into separate subsystems, and attempt to integrate them after several months. By this time, each subsystem would have taken on a life of its own, and integrating these disparate monoliths together gave a whole new meaning to "plug and pray."
How XP "Fixes" It XP takes the development process to the other extreme, by shortening the "waterfall" lifecycle to weeks, days, even minutes. In fact, Kent Beck describes XP as a "waterfall run through a blender." Iterations typically last for a week or two; there is a high emphasis on code quality via unit testing; and code is integrated "constantly" so that it never becomes out of synch with the rest of the project. Beck is often quoted for saying that the XP practices "turn the dial all the way up to 10" -- that is, if something is good (testing, integrating, pair programming etc), well then, let's do it all the time.
There's a lot to be gained from learning about XP, and agile practices in general. However, many feel that XP has taken things too far. By taking things to the opposite extreme, we're just introducing a fresh set of problems. The optimum solution, then, must lie somewhere between these two extremes. That is fundamentally what Extreme Programming Refactored (XPR) is about.
Optimizing the Process There's been a lot of controversy surrounding this book. It grew out of an equally controversial article that appeared on the author's website. XP advocates were arguing on Yahoo! Groups over XPR's good and bad points, miraculously, months before the book was even available. XP zealots were even posting messages telling others not to buy the book, before they'd even had a chance to read it for themselves and find out what it's all about.It's important to note that XPR isn't the anti-XP slam piece that some people had been expecting. It does rip into the XP practices in plenty of detail, but importantly it also describes alternatives, and talks about the good aspects of XP.
The authors make the argument that "turning the dial up to 10" mostly isn't such a good thing, and that to achieve our "holy grail" development process, we just need to turn the dial down a little bit -- let the milk simmer rather than boil over. We can do this by adding in some additional practices that take the burden off the overloaded, heavily interdependent XP practices.
For example, not everyone likes to pair program (with two people sitting at one computer). It just isn't for everyone. However, XP relies on everybody in the team pairing all the time. So if you don't like to "pair up," what choice do you have but to leave the project? XPR adjusts the other practices, placing a bigger emphasis on up-front design and documentation, so that pairing up doesn't have to be mandatory.
XPR also argues that it is possible to achieve a decent design before writing the code. The authors don't want a return to "BDUF" (Big Design Up Front), but instead to achieve an ideal middle-ground. The result is more akin to the monthly Sprints found in Scrum (www.controlchaos.com).
Similarly, XPR argues that the customer (and users) usually do have a pretty good idea what they want from a new system, and that they don't have to see a live system first before realizing that they wanted something entirely different. The authors argue in favour of interaction design as a way of achieving this goal.
XPR achieves all of this with more than a mild dose of satire. It's important to realize this -- the book is essentially "taking the piss" out of the more extreme XP practices, and the quasi-religious Extremo culture that has quickly grown up around XP. It has lots of serious things to say, but has a slight danger of that being lost "in the chuckles." There again, the danger is less to do with the book, and more to do with the reader.
XP sealots will never be swayed by such a book, naturally, but they are not the audience. It is for those undecided, or the cowed XP skeptics who know something is very wrong at the heart of the beast, but haven't have the words to say it. Even for zealots, I'd hope they'd put the hatred for long enough to at least temper their XP actions, to turn the dial down a little, to read the contents with the possibility in their minds that XP isn't the final perfect expression of all programming methodologies. Just for a while...
If you are scared of the contents, a favorite XP accusation, then of course you'll point out the 'needless humour,' blah blah, anything rther than address what the book says. Form will be far more addressable than content. It's the old "ignore this man, he wears a colourful tie" excuse, pick on some small detail that you feel is a weakness and totally ignore all the embarrassing questions you'd rather not address. If you like the contents, then the humour will be seen as a playful, court-jester like addition to what is a seriously analytical book
In conclusion, this book is well written, thought provoking, and above all entertaining (an aspect which seems to be proving almost heretical among some XP advocates). I found this to be a fun read, unlike some books, it was never a chore. It's extremely conversational, like having a cynical, wise-cracking guide. It's a pity more computer books aren't this fun. A spoonful of sugar and all that...
In fact this book is pretty damn wonderful. I know, it may sound a bit gushing, but before you review my review, give the book a read yourself. It's a thing of beauty, a rare mix of positive and negative, sweet and sour, opinion, and XP's favorite emotion, courage, courage enough to say "the emperor has no clothes." I can't see how you could read this book from beginning to end and not see XP in a different light.
In fact, XP programmers would do well to read this book, as it presents the negative path, something other than sunny-day scenarios. Using these warnings and guidelines, they'd have much more successful projects, as this book points out the dangers of lack of XP discipline, fragility and so on.
The truth is that I couldn't do justice to this book in such a short review. There is just so much evidence, so many contradictions pointed out, endless damning words from XPer's own mouths. It was supposed to be a small book to start of with, 150 pages or so, but due to the sheer body of evidence and submitted real life stories from those in the trenches, it bloomed to 400+ pages.
As Doug Rosenberg says "I don't want to be nearby when somebody decides to deploy an air traffic control system or some missile-targeting software that has been developed with no written requirements, and where the programmers made the design up as they went along." At least don't say you weren't warned!
You can purchase Extreme Programming Refactored: The Case Against XP from bn.com. Slashdot welcomes readers' book reviews -- to see your own review here, read the book review guidelines, then visit the submission page.
sPh
> XP like any software process allows you take
> and use only those features that work with
> your project
Sure, you can do whatever you want. But XP without pair programming isn't XP, it's.... something else. "Pretty Adventuresome Programming", maybe.
Some folks don't like XP because it's hard - it's hard to write unit tests, and it's hard to pair program and share knowledge and so forth. That's fine... but if you're rejecting XP because you don't like doing those things, don't blame XP.
The Army reading list
I would have to say that pair programming isn't really worth it. I personally can't stand the lack of productivity that is inherent to having two programmers sit around and watch each other type.
My version of pair programming is to have one developer write a test harness while the other one develops the actual code to be tested. This forces each of them to communicate with each other, generally via a very informal spec or direct communication with the client.
This ties two people closely together with the immediate need, but it doesn't require complete overlap. I'm not convinced that two brains are better than one - typing is an inherently blocking process, and whiteboarding happens either way.
Anway, what good hacker wants to watch someone else type? Or play a game? Or do anything? Coding isn't a spectator sport.
warning: epoll_wait is not implemented and will always fail
having a team of programmers who:
Like to program,
are properly trained and schooled,
who are paid enough,
and are given enough time to do the job right.
Everything else is fluff or a fad.
Your hybrid is not saving the environment. Its purpose is to make you feel good about buying something.
> the best way to deliver software effectively
> is to understand the problem domain
That's part of the process, yup. And there's a lot of other stuff, too.
> You don't need requirements before you
> start coding
What about the planning game? What about user stories? Read this for better information on XP requirements gathering.
XP has the concept of figuring out what customer want... but XP doesn't demand that you and the customer pretend that every detail of the system is in a bunch of white binders filled with stuff like "the system shall...". Instead, you and the customer work together to develop a system that meets the customer's ever-changing requirements.
> if you aren't you need more process to
> make sure you don't FUBAR things
That's the beauty of XP - it assumes that we aren't geniuses and thus we use unit tests, refactoring, pair programming to keep the code under control.
The Army reading list
Q: What do you get when you combine the waterfall model with the spiral approach?
A: The flush model.
On a more serious note, the waterfall model does work for certain problem domains, notably those where the requirements are well understood in advance and unlikely to change significantly over the course of development, and where reliability of the final product is critical. This assumes you have the resources (time and people) to do it properly -- and this is why the waterfall method has gotten a bad reputation -- it's been applied where there is insufficient time/resources or the requirements aren't well understood.
Most business applications, for example. The waterfull method is wonderfully suited to something like spacecraft control software, where the spiral approach (we called it "stepwise refinement" twenty-five years ago when I was in college -- there's nothing new) just wouldn't work. But businesses are both constantly changing and adaptable -- a business app that only implements half the requested features is probably still more useful than not, where as software to control complex hardware that's only half done is nearly useless (except that some testing can be done, perhaps).
As always, it's a matter of picking the right tool for the job, rather than picking up your hammer and treating everything as a nail.
-- Alastair
If you don't do pair programming, you need to replace it with a less extreme version of code reviews, such as simply reviewing each patch before it is checked in. The type of pair programming you describe involves no code reviews, and therefore is no substitute for pair programming.
What a fool believes, he sees, no wise man has the power to reason away.
Who hates the idea of pair programming? Smarties ofcourse. They dont want to lose what they feel makes them special.
;-)
;-) but do you just want a stable of predictable but average programmers, or do you want to hold on to those "smarties" who, in a pinch, can deliver good code faster than any eight other programmers put together?
;-)
More likely they're just driven insane by having to wait on the other person. Depends on how bad an impedence mismatch you have. Programmer productivity can vary by an order of magnitude or more between individuals (so called "superprogrammers", although the term is silly). If you have one of the latter, no amount of "skill transference" is going to bring the former up to the same speed, any more than teaming someone with an IQ of 150 with someone with an IQ of 110 is going to raise the latter's IQ to 130. (It might, through frustration, lower the former's though
On the other hand, if you're just talking about a couple of average programmers, one of whom has a couple of years more experience, then yeah, it makes sense.
They are bad for business anyways.
Depends. Nobody needs prima donnas, of course (well, except ballet companies, I suppose
(Of course, you may not have the choice. Superprogrammers aren't nearly as prevalent as the number of average programmers who think they're "super" would indicate
-- Alastair
It's why Liberals need to listen to Rush now and then. It's why neo-cons should read Franken.
So anyone using XP who says "don't read this book" is probably delusioinal in several respects.
As another poster remarked, though, it's a bit annoying to read a review that does not include snips of key points and some analysis. This is endemic in /. reviews. Reviewers, please go look at the form of articles in the Sunday Book Review and follow it!
-dB
"It if was easy to do, we'd find someone cheaper than you to do it."
How would you achieve this? I'd probably do things like:
Of course, by the time you do all that, you've got the heart of XP.
how to invest, a novice's guide
You know, sometimes we just want to pick our noses or read a news story on the web. Pair programming makes this impossible as there is always some idiot practically sitting on your lap. This is why it will always be perceived as punishment.
Unit testing, especially as a foundation for refactoring; continuous integration, read hourly builds; and the flexibility to revise design decisions when the need becomes obvious -- all of these principles are absolutely indispensible in any modern software project. Failing to do any of these things is an outrageous, inexcusable lapse. But these ideas are better understood and more obvious now than they were before Kent Beck came along with XP.
I see all the other posts in the thread so far claiming that some or all of these concepts were known earlier, before they were emphasized by XP. But that does not discredit the approach. Frankly, there's rarely anything new under the sun, but sometimes someone has to come up with a kind of philosophy just to help us realize how effective certain ideas can be. I think all programmatic software methodologies are like this -- they try to emphasize methods that are known to have worked well in the past, and try to draw useful generalizations about them.
Nowadays, most of the UP people are adopting many ideas emphasized by XP -- I attended a tutorial by Craig Larman last year in which he argued that no conflict between the UP and XP exists at all. But no one was saying these things before XP was articulated.
Incidentally, I think it's clear that the waterfall is an unmitigated disaster. I was once in a project that was managed that way, a big fat loser of millions, the worst nightmare of my career, took down a whole dot-com agency. Those were the days.
I don't agree with all of XP -- a complex system has to start with a fairly sophisticated analysis and design (but it better be designed flexibly enough to change over time if necessary). And I can't imagine pair programming, never tried it and don't want to. But even if the Emperor is not wearing purple and crimson ermine minks with linings of silver and gold, he's not naked either. On the contrary, his suit is fairly presentable.
Always keep a sapphire in your mind