Slashdot Mirror
Has Software Development Improved?
earnest_deyoung asks: "Twenty-five years ago Frederick Brooks laid out a vision of the future of software engineering in "No Silver Bullet." At the time he thought improvements in the process of software creation were most likely to come from object-oriented programming, of-the-shelf components, rapid prototyping, and cultivation of truly great designers. I've found postings on /. where people tout all sorts of design tools, from languages like Java, Perl, Python, Ruby, and Smalltalk to design aids and processes like UML and eXtreme Programming. I'm in a Computer Science degree program, and I keep wondering what "improvements" over the last quarter century have actually brought progress to the key issue: more quickly and more inexpensively developing software that's more reliable?"
There is no silver bullet, never will be.
Logic requires careful thought, and careful thought requires time.
You are being MICROattacked, from various angles, in a SOFT manner.
Programs that assist programmers in the development process by handling who changes what when, etc - are - IMHO - a huge improvement.
I seriously doubt that a program like Linux could flourish without programs like CVS.
furthermore - many of the programs that do this sort of thing can be used for any programming language... you could even use it for simple documents.
But seriously, a lot of people develop beneath the "enterprise level" and some of the buzzword concepts just don't scale that well for smaller project.
In my opinion, the two things that have really made a difference are databases (as opposed to manually creating file formats) and the object-oriented paradigm.
My best advice is to use everything with a grain of salt because there is always something "better" on the horizon.
If I had a dime for every system I've seen where there was no planning, no logic, just brute force coding I'd be richer than those in the .com era.
This bring me back to the last slash dot post a few weeks ago of the 80 hour a week SW engineer.
True SW architects are hard to find and so is thier logic. A managment that understands the expense and time involved are even rarer entities.
Hardware Architecture keeps building on similar themes, no real inovation going on (I use that term loosely). SW requires a human factor.
more quickly and more inexpensively developing software that's more reliable
Based on the last 20 years either working in or supporting government efforts, I'd say yes. However (there is always a however), it depends on the sophistication of the developing organization. The cowboy shops still suck. Those who have embraced more formal processes have become more reliable. It is a 2-way street though and requires the customer to be more sophisticated as well. It doesn't do a damn thing to have a development shop at CMM-5 if the client doesn't understand the need for process and doesn't understand the software development challenges.
The invention of pencil and paper.
The question of whether a computer can think is no more interesting than the question of whether a submarine can swim.
If you look at how programming languages and sofrware evolve, the errors just get higher level. For example, Java keeps you from having memory leaks and buffer overflows. But a bad coder is a bad coder and will write buggy software where the problems are on a different layer. They will just arrange your off-the-shelf components incorrectly. With wherever there is still really low level C/assembly programming stuff going on, off the shelf just really isn't applicable so you still have the really low level pointer arithmetic problems. As time goes on nothing really changes... it just evolves. So our bugs are just evolving.
The ability to create decoupled software components , combine them into a coherent, functional application, and deploy them into a standards-based container (i.e., an "application server") is a HUGE step in programmer productivity.
Silver bullets still do not exist. New technologies and methodologies are often hyped as such and naturally fail to live up to the hype. However, that does not mean they are useless.
These technologies and methodologies have allowed us to keep pace with Moore's law. The amount of software we develop today would simply be impossible using the state of the art in 1970. We routinely poor out millions of lines of code in mammoth projects that take sever hundreds or thousands of man-years to complete. The only reason we are able to do so is because of improvements in development methodologies and technology.
The (huge) research field that studies these technologies and approaches is called software engineering.
Jilles
High quality, reliable C and C++ compilers have emerged as defacto standards on major platforms.
Now you wouldn't think of developing on UNIX with anything but GCC and the associated build tools.
In 1990 you were stuck with whatever C compiler your vendor shipped, and there were more than a few dodgy compilers out there. Modern compilers with useful error messages have done more than anything else to make debugging and development faster and easier for me.
The focus has definitely shifted away from algorithms and toward abstraction. This was supposed to make things easier, by letting the software do what it does best and keep track of bookkeeping, while we concentrate on building models and governing interations between them.
Some of it actually makes sense - the object oriented paridigm, component models, virtual machines. (VM's, by the way, go back at least 20 years in the literature -- I studied them in college in the late 80's. However, like Pascal, they were originally considered as an instructional tool, and nobody at the time thought that anyone would be damn fool enough to actually try and *implement* such a thing!)
But just like letting grade-school students use calculators for their arithmetic, I'm not certain these things are best for students. Sure, you get usable code out quickly, but without an understanding of the underlying algorithms and logic. I doubt many modern 'c0derz' could properly knock out a simple quick-sort, let alone a fully ACID SQL DBMS.
"Lawyers are for sucks."
- Doug McKenzie
else how would commercial software company's make money ? They need to sell updates and support.
;-)
They code crap. My personal experience with not to big software company's is they like to sell crap if they can and most of the time the other company also sells crap and after wards you dont even own the software you buy. Software bussness sucks atm.
Until the world wakes up and start demanding that software needs to work just like material stuff or else they wont buy it. It wont get better in the near future aslong we except crap from even big company's like Microsoft, which is becoming more reliable and stable but still sells a awful design if you ask me. oke maybe start a flame here but most software is like throw a way dishes instead of a rolce roys. And i think company's like microsoft showed the smaller software houses that its oke to sell crap as long it sells and it bugs me cause software has a bad name, but i know good software exists, bla bla bla, you got me pissed again slashdot....i hate humans
Quazion.
NO.
The project management keeps moving the target. The customer still says "I don't know what I really want, but keep coding and I'll let you know when you get it." Analysts can't analyze, designers can't design, coders can't code, testers don't test (enough), ad nauseam.
Methodologies and philosophies of software development can only make up for so much. Sometimes, they are counterproductive. They "lower the bar", leading management to believe that the methodology can fill in for cheaping-out on hiring good developers. But we /.ers know the truth, don't we: Quality software comes only from quality developers--people, not methods or schools of thought or books by really smart people. Since quality developers are rare (like Leonardo DaVinci rare), quality software is correspondingly rare.
Welcome to the Panopticon. Used to be a prison, now it's your home.
No need to haul out references to books or count buzzwords...just look at the software world and the question answers itself.
Since the early days of computing in the late 70's, we've seen systems grow more larger and more complex. Whereas entire OSes used to be written by a single person (granted, a gifted person like the Woz or Bill Gates), these days we have companies like Sun and Microsoft with literally hundreds of developers and testers for a word processor, let alone the thousands of folks around the world who contribute to Linux, Apache, or KDE.
Given this incredible change in how software is developed, we'd expect to see systems collapse into instability and mayhem, but save for a few exceptions (Windows 9x, anyone?) this has largely not been the case. Windows XP, Mac OS X, and Linux 8.0 have proven, if anything, more stable and reliable than their predecessors. For an even more dramatic example, look at NASA's software or popular video games. It's clear that not only has software development expanded in scope exponentially, but it has become objectively better in the process. Development has never been better, and I see no reason why this trend shouldn't continue.
Karma: Good (despite my invention of the Karma: sig)
Let's say you're trying to improve the work of a painter. Give him a better brush, his strokes look nicer. Give him better paint, his colors look brighter. Give him a CAD system with a robotic painting arm, his paintings look crisper. However, nothing you've done can change the fact that he's still doing velvet Elvis's. Not that there is anything wrong with that, but he isn't churning out a Mona Lisa anytime soon, or ever. You haven't been able to affect the input to the process - his brain. You know the adage, garbage in, garbage out. Well that is the fundamental rule in software.
The big improvement in software productivity will come when we are able to successfully clone human beings. Corporations will create a stable of uber-programmers who they will then clone to be able to mass produce software. In the future, the open source war will be about peoples' genes, the software will be just a downstream product.
There are definite improvements to programming. Tools, concepts, etc. have evolved. There are no true silver bullets, but we've got a good selection of choices out there.
The problem, however, is threefold:
One of the real challenges for IT professionals today is to find the good tools and ideas out there, the ones that really have improved programming, and then actually get them into use. A good IT person is a good researcher and good at justifying their ideas to people.
"The Sage treasures Unity and measures all things by it" - Lao Tzu
what "improvements" over the last quarter century have actually brought progress to the key issue: more quickly and more inexpensively developing software that's more reliable?"
I've only been programming for 19 years (not 25) but I can say that I've seen absolutely no progress in software development givin your contraints of defining "progress" as being able to achieve the three goals (speed, cost, robustness). HOWEVER, I don't really blame the tools, rather the nature of the end result. The software we have to write has become significantly more complex, and at a rate that has surpassed our abilities to effectively create tools to deal with their complexity. "Back in the day" when a "good" business level word processor was 100KLines of code and written by a small group of dedicated and bright people, you could get very good results. Now, something like Word is significantly larger than 1MLines of code being worked on by programmers with 15+ years of experience to those who just got out of school to those who you wouldn't trust to add up your grocery bill let alone write a line of code.
It's like we still have hammers and nails, but the "market" wants us to build steel skyscrappers. So we come up with even better hammers and whiz bang processes to "better" utilize those hammers and nail together steel girders, but the fact is that those tools are woefully inadequate to truely make a dent in improving that which we build.
Think about it. 25 years ago, it was extremely limited. How many people did you know, in 1977 with a net account? I remember coding on a C64 in my cousin's basement for days on end just because we had scrounged enough money to get into town and buy some new books/magazines that helped us overcome some bug. Now, if you're being bitten by the same bug, whaddya do? Hit the net! Some of the responses above, like the sharing of source through GPL wouldn't be as viable an option without the access we have today. The biggest aid to programmers today. Net Access.
Don't park drunk, accidents cause people.
Yeah, as a PhD student you get paid practically nothing (at least when compared to corporations), you have to work silly hours and possibly teach as well.
Yet, if you pick your topic carefully, you'll be working those hours on interesting and intellectually challenging bleeding edge stuff and not on some YAMC (yet another moron client) project that was handed over to you. In a nice group you can also choose whether you want to come in to work early in the morning or at around lunch time. Furthermore, teaching experience can never hurt a nerd. I used to hate teaching and I still don't like it. However, positive teaching experiences and the very act of confronting a group of smart, young people for four years improved my self-confidence, presentation and public speaking skills tremendously.
but in my world, Java is the single largest memory hog and memory-leaking piece of crap I have ever seen.
you're kidding yourself if you think Java keeps you from having memory leaks, and I have enterprise code to prove it
A year spent in artificial intelligence is enough to make one believe in God.
I can't say that Java is a significant programming advantage over C -- it's the Java libraries that beat the snot out of calling the raw Windows/Mac/X APIs.
That's not the only one: The depth of CPAN, the Python library, even MFC/ATL are worlds ahead of hand-coding the guts of an app.
When I started programming, the Mac SDK had to be handled raw, and GDI was brand spankin' new. APIs such as DEC's FMS were considered heaven-sent(talk about dead ends!). Shortly after, class libraries were available, but expensive for an amateur. With Java and Perl, it's part of the (free) package.
I'm sure there's garbage code in those libraries -- there's always going to be bugs. But they're going to be relatively smooth to use, and they're already written, letting me focus on the business logic, instead of the presentation.
Design for Use, not Construction!
If you're coding an app and you are spending time on the GUI you are just creating a maintenance headache for maintenance programmers later on.
Most documentation is horrid if it even exists (learn a human language first and use it to actually write meaningful comments, specifications, test scripts, internal and user documentation.)
Most of this industry doesn't know dick about SLAs or optimization for time (first) or space (last.)
Most of this industry doesn't know dick about configuration management, capacity planning or correctness.
The difference between duffers (most of this industry,) and the pros is that the pros don't "paint little master-pieces" in a a guild-like cottage industry. They generate "art by the yard" in industrial settings for mass dispersal, distribution and "invisible" use.
Good luck and remember, computing is nothing but a problem in N-Dimensional topology. If anybody tell you different, they are part of the problem, not part of the solution.
ALL objects have states and state transitions, code for that first and the rest will follow. Start from a thorough, correct and complete meta-model and you won't get into trouble later.
As for languages, CASE tools, GUIs, IDEs and the rest. Learn to do it the long and hard way first so you'll:
a) know what's being optimized and abstracted out,
b) appreciate it,
c) know what happens when it fails.
MSBPodcast.com The opinions expressed here are my own. If you don't like 'em... Think up your own stuff.
Now I am not saying that you still don't need to have a good understanding of the language, or use good design, implementation, testing, etc. But I have worked at SEI CMM level 3, level 2, and am currently going through the process of evaluation where I work now, for level 2. But being at that level doesn't guarantee that your software will be good. It seems almost like we are attempting to fit the creation of software into the old engineering mold, because it kind of fits.
So to answer the question - I don't think that there have been any great improvements to obtain the goals you stated. Software relys on hardware, which seems to contstantly be changing. A bridge is a bridge, but the very idea of what software can do has been changing constantly over the last 25 years.
If you want reliability, look at what NASA has produced. Those are some engineers. Ahh, but you said 'quickly', didn't you? :-) If you want something rock solid, you can't have it tomorrow, if you want it to do anything remotely complex. I think one of the big things we as an industry has to realize is that our product (software) can be simple and fast to market, or more complex and take more time. And all of the variations in between. I haven't been convinced that what we do is engineering, but I haven't been convinced that it isn't. After all, it is a pretty young profession, compared to others. Imagine trying to explain to someone from 100 years ago what a software engineer does.
All that being said, I think that the obvious Slashdot answer is that the GPL and free software have been a huge force, but only in recent years. I think the two biggest forces in software development were the PC and the internet.
My beliefs do not require that you agree with them.
As a professional developer, I've seen complexity for even simple solutions skyrocket with a concomittant loss of performance with the current obsession on over-inheritence and Java-style interpreted/P-code software overall. Add to this GPL/OS that slashes meaningful business value from well engineered software components and I think we have retrograded from the vanilla unix with tough, straightforward C coding.
If it were done when 'tis done, then t'were well it were done quickly... MacBeth
IMHO, one of the keys to writing good software is to not write very much software. Class libraries like the one in Java are almost invariably more feature-complete, more reliable, and better tested than roll-your-own libraries. Using existing libraries means you don't have to worry about off-by-one errors in your linked list implementation, or figuring out a good hash key for your string. While I think that all coders should know how to solve these problems, they shouldn't have to as part of their daily life. It's busywork, and giant class libraries do a wonderful job of making life easier. They let you concentrate on the important stuff - the algorithms, and marketing's feature of the week.
Yes, libraries have been around for more than a decade. The STL and Java both showed up in the mid 90's, though, and I'd argue they've made a huge difference since then. Yes, they have bugs. Yes, there have been other advances. Yes, there are special cases where standard libraries aren't appropriate. There's no silver bullet, but standard libraries are a huge advance.
That's what code libraries are for just rewrite the function in the library, recompile the library, and relink the other programs... or even better, put it in a dynamic library and just recompile the library...
Find some other career. Like be a male stripper or something else that has regular hours.
Is this thing on? Hello?
All the problems with software today aren't a result of poor engineering, but rather of poor software development management.
;) But most business orientated software is bug-ridden, and that's just fine. It's the accepted risk for getting the software out the door as cheaply and quickly as possible.
I can't count the number of times last minute feature requests are required to be in a build. As software developers, we just deal with it. But quality suffers because of it. And the engineers get the bad wrap.
Do you think Intel management requires last minute features to the Pentium core, and tries to push them out the door with no testing? Do you think people building cars for Toyota decide to swap engines at the last minute, becuase GM just put a turbo charger on it's standard Prizm?
It's ludicrous the stuff that gets requested of software engineers.
My brother is an contract electrical engineer. He was complaining one time of a 'last minute feature request'. His project still had a year of development time left! I laughed so hard I almost puked.
Granted, given all that, the current model for software works. When software is required to be bug-free, it can be. Lots of software is bug free. You just don't hear about it, because it works. Look at NASA, the AIN telephone network, or Windows.
I don't have a sig...Do you??
Erlang does have some great applications - but I don't think one can say OOP doesn't scale well(forgive me if I am misreading your post). For example, C++ has been proven to be extremely scalable. Not only can it be scalable performance wise (read speed and size), but it has proven to be scalable project wise (read large collaborative project).
While people seem to be saying there haven't been any advances, and won't be any advances, and such, they're forgetting that the first key step in solving any problem is identifying the problem.
People are complaining about how using stronger employees to lead less-skilled employees doesn't work. People are complaining that the person testing the "bad coder's" code is usually the "bad coder" itself. These complaints begin to define the actual problem of "Why software development isn't improving." Each of these complaints, taken together and addressed, is the first step towards improving software design methodologies.
If I have been able to see further than others, it is because I bought a pair of binoculars.
I think the varying degree of control over the physical machine a developer can now choose when writing a program is the single most important factor in increased productivity. In assembler the programmer must worry about everything. C was the first truly abstract programming language where the programmer could call a routine like printf and not need to worry about the details of printing a string to the screen. Because the language was more abstract, the programmer could do far more complicated things (have you every tried to write a red/black tree in assembler)?
Over the last several years languages have gotten more and more abstract, languages like Java isolated the developer from pretty much everything except the logic they are trying to capture. Developers can now choose the level of abstraction they wish to work with based on the problem domain. (low level library vs. script for renaming files) Higher level (more abstract) programs are usually much easier to write but it is worth noting that there is a price to pay for, generally these programs are not as powerful as their low level cousins. Some languages like VB still try to abstract development out even more, so that it is accessible to everyone. Abstraction has brought the ability to program to a much wider audience and has greatly reduced the time it takes to write basic applications and for that it is the most important change in programming.
Just to nitpick on this particular myth. In construction there is the idea of "fast-path" construction, where building starts before the design is done.
The idea that all requirements and design are done before construction starts is a myth.
People hack buildings too. There were some famous cases of building "bugs" (eg. CityCorp skyscrapper wasn't stiff enough) or famous failures where design or implementation errors caused a building collapse.
Read "To Engineer is Human" and "Design Paradigms" by Henry Petroski for a start.
...richie - It is a good day to code.
I think any programmer who sees the benifts of CVS would understand where im going with this concept. We all have functions we use again and again - and realizing that there is a potential flaw in a given function at one point is always followed by exasperation because one realizes that the function needs to be changes in X number of programs.
You don't need a new version control tool, you need a refactoring tool.
Small-scale development has always been efficient. The challenge facing the industry has been to find ways of doing large-scale development (the type Fred Brooks was talking about) cheaply and effectively.
And in this domain, there has been a revolution, namely the Internet, and the arrival of cheap connectivity between developers anywhere in the world.
Prior to this, the only way for developers to collaborate was to be hired by the same company, work on the same network. Inefficient as hell.
Today any group anywhere in the world can create a project and work on this as efficiently as a small group in the past.
The irony is that the revolution does not care a shit about the technology used, and works as well for COBOL programmers as for companies cracking the human genome. It's about solving a purely human problem: communications.
Sig for sale or rent. One previous user. Inquire within.
First of all, glad to see a lot of positive posts on this topic...I frequently see on this very same site laments about the dismal state of software. I am in agreement with the viewpoint that software developers continue to be more and more productive (through frameworks, code reuse, improved languages and tools, etc.), however the productivity hasn't resulted in improved software quality because we are simply being asked to do more complex tasks with the same schedule and resources.
One thing that has drastically improved my productivity is the Web itself (time on Slashdot notwithstanding), as a way to locate resources for programming. Almost any algorithm, component, or subsystem that is not specific to the problem domain can probably be found on the web, whether as a library, a set of source code, or simply a precise definition of an algorithm. I agree with one post that a lot of young developers would have difficulty writing a correct bubble sort, but anyone that attempts to design and implement a bubble sort on the job is wasting resources, since there is an implementation somewhere on the Web in any language you would want. In addition to software projects, informative articles and API documentation, newsgroup discussions are invaluable as well for pinpointing problems. So the Web is really my most important programming tool.
In conjunction with the Web, the vast supply of open source and free software out there that has drastically improved my productivity. The Apache Jakarta project and CPAN are my favorites, but there are many interesting projects on SourceForge and FreshMeat as well. Often, even if you are determined (or required by corporate standards) to roll your own, free software can give you a good idea about how others have approached the problem and the abstractions and metaphors they've used. In my experience, design reuse is often far more helpful and practical than actual reuse of code.
Software development, at least many types of software development, has changed, in that programmers are much more dependent on large APIs and libraries than they used to be. In theory this is good, because it saves work. In reality, it has turned many types of programming into scavenger hunts for information. Now you have to hang huge posters on your wall showing object heirarchies (you didn't remember that a Button is a subclass of BaseButton which is a sublcass of ButtonControl which is a subclass of WindowControl?). Now you need online forums so you can get a consensus about how a certain API call is supposed to behave in certain circumstances. Quite often I feel I could write a new function faster than locating information about how an existing function is supposed to work.
I can easitly see your point, but at the same time, without knowing much about OneWorld, I have to wonder if it doesn't do a little bit more than you're giving it credit for.
While I no doubt respect ManMan (especially with a kick-ass moniker like that), I could easily see ManMan looking on enviously at OneWorld's application and system interop features. Try RPC on ManMan..
-- jimmycarter
I have over 30 years experience of embedded systems, and I would say that without exception the C ones were leaner, meaner and fitter than the C++ ones.
As for reusability, documentation is the key to reusability. It is conceivable that the syntax of some languages may help with this, but it seems there is more benefit in theory than in practice. All of us use the C standard library all the time (ie anyone using a browser is executing it) - now THAT is reusable. The Fortran library is widely reused, and no one ever accused Fortran of being object oriented. (I have heard Fortran users called "disoriented" :-)
Sent from my ASR33 using ASCII
Software development will never be made reliable and inexpensive until we get rid of the programmers.
Software development is a cottage industry and "art" where artisans produce one of a kind products. No two programmers will produce a program in exactly the same way.
What we need is an easy way to translate analysis directly into code in a reliable way without human intervention.
This will give us the consistency and repeatability that translates into reliability and lower cost.
From the article:
The most a high-level language can do is to furnish all the constructs that the programmer imagines in the abstract program.
A-men. But if it does that well, then it makes the job a lot easier. That's why going from C to Java or Perl was sheer relief. Actual strings? Real associative arrays? Whoohoo! And less memory management grief. Not to mention the component libraries available for things I hadn't even thought of yet. CPAN...
To be sure, the level of our thinking about data structures, data types, and operations is steadily rising, but at an ever decreasing rate. And language development approaches closer and closer to the sophistication of users.
True... but the user sophistication is increasing to. It seems highly apparent to me that with more experience and more shoulders to stand on, language and component developers are able to concieve of more and more useful abstractions. And because of the internet, they're more easily available for sharing, commentary, and change.
To sum up, I am much, much happier with the readily available toolsets I have access to now than the ones I had 15 years ago, or even eight years ago. They make developing much easier and much more fun.
Libertarianism is rich wolves and poor sheep playing gambler's ruin for dinner.
I notice the original post mentions several things that could influence the development time of a software project. I will address a few of these below:
1) Object Oriented Programming :)
This is one of the bigger Silver Bullets to be unleased upon the programming world. I don't think it entirely lived up to the hype. Most OOP is just for local project design, and heaven help you if you have to reuse code somewhere else. It isn't just a case of bad design. Problems like software design are actually ambiguous. The design process is not algorithmic; rather, it's heuristic. You use "templates" and "patterns" to represent your ideas. Trying to shoehorn real-world complexities into these cookie-cutter styles is difficult at best. Trying to further take those styles and integrate them with each other in a very large scale product is a hair-tearing nightmare. I think Tablizer would agree with me on this...
2) Reusable components ;)
The most visible place reusable components come into play is GUI programming. It's very, VERY simple to use a visual-based system (like Visual Basic, C++ Builder, Delphi, etc.) to create a GUI simply by dragging the desired components onto the blank form window. If anything has been sped up significantly in the past several years, it has been the GUI development.
Components are, of course, used in a variety of other places, particularly in run-time libraries of various programming languages. However, learning to use these components effectively takes more time and dedication than one might suspect as the syntax tends to be rather cryptic looking.
3) Java
Don't get me started. I am currently employed as a Java developer. I don't really like it a lot. The file scoping rules bug me. (Similarly, I don't like Python because of the way it enforces indentation.) Also, the Java IDE sucks. Whoever thought the entire GUI needed to actually be written in Java needs to be taken out and beaten with a stick. A large stick.
4) The Internet (and OSS) ;)
One thing I noticed that you hadn't mentioned is the Internet. I have never been exposed to so many programming concepts and new languages. There is an astounding variety of tools, and thanks to Open Source and researchers at various universities, you can try your hand at as many of them as you have disk space for. The 'Net can be a wonderful place, after all.
My advice to any new programmer would be to get online and start reading. Download and try out new languages, especially ones in different paradigms, like functional programming. The tools you need (such as compilers, editors, databases, GUI component libraries, etc.) are ALL there, free for the taking. The only real "silver bullet" is to make yourself the best programmer you can be.
bytesmythe
Hypocrisy is the resin that holds the plywood of society together.
-- Scott Meyer
Good, Fast, and Cheap can happen, even in software. If the aircraft industry can pull it off, so can the software industry. Read up on the Lockheed Skunk Works. They did incredible stuff with very few engineers in a very short amount of time. The key is people. You need a top-notch staff and more often than not a world-class leader. Such a team is hard to come by, but when they do get together they can pull off some amazing stuff.
To be honest, to throw my 2 cents in, we are seeing more and more bad programming. 20 years ago, you might have expected a program to have bugs. But know we see some of the biggest software houses and programmers releasing half standard programs, just to get them released on time. With many bugs still with them. And for the most part these bugs are bareable only because, our massive increases in hardware does not notice the odd memory leak here for a long time. To be honest, how many things we have designed over the years, would run on a 286, with 4 meg of ram. Fair enough you can't ask it to hand some of the graphics side, but the bare bones should at least run, even if you would be waiting hours for it to say Yippie. All in all, we can get payed silly money for stuff we write, and the pressure of deadlines makes some awful things. But on the plus side, those that do side projects, do end up being the best things around. As you dont want to release a personal project unless its perfect.
--+> Life, is there any?
I've often thought over the last few years that we've made too little progress in making programmers more productive. I largely blame that on Microsoft, simply because it drives more software development with it's tools than any other entity. One language I've categorically made a decision to avoid is Visual Basic. I have always felt it was basically (sorry) a waste of brain cells. It has certainly done nothing to advance the state of the art.
In my opinion, one of the best things to come along in a long time is Java. The gentle reader may recall earlier posts along those lines. I enjoy C, and have spent the majority of my career doing C and C++. However, I have also spent _way_ too much time tracking down memory-related bugs. Often, they were in third party code. That is no way to run a railroad.
Java addresses almost all of the glaring deficiencies of C++, both in language design and in runtime safety. In my opinion, the best programming tools will be those that enable single programmers to tackle larger and larger projects.
Compared with C++, Java enables me to tackle much more ambitious projects with confidence. A team approach can never attain the efficiency of a single programmer approach. The "sweet spot" of software engineering efficiency is the largest project one person can tackle. Extreme programming is a useful hybrid that attempts to turn two programmers into one programmer. ;-) (Also teams can be nearly as efficient as single programmers if the system is properly decomposed into subsystems separated by simple interfaces. This rarely happens smoothly, in my experience. It takes a top notch group of people.)
One last note on Java - performance is now almost completely on par with C++. On my most recent round of benchmarks, Java (JDK 1.4.1_01) on both Linux and Windows outperformed C++ (gcc 3 and VC 6) on most tests. Dynamic compilation and aggressive inlining are that effective. The VM also soundly spanked the gcj ahead of time compiler in gcc 3. It thoroughly rocks to have truly cross-platform code that runs faster than native! Think how many religous wars would be avoided if 99%+ of software was available on all OS platforms...and think how much it would help Linux! :-)
If you want to see what's out there for Java, download either the NetBeans IDE project, or the Eclipse IDE. Both are free and each has its strong points. NetBeans is a Swing app and includes a Swing GUI designer. Eclipse uses a new open source "native widget wrapper" library from IBM called SWT which has it's interesting points. You'll also need a Java VM (there are also others available from IBM etc.).
One last thought - wouldn't it be cool if web browsers had support for something like Java? I mean, you could deploy apps just by putting them on a web page! It wouldn't matter what the target platform was! What a great idea! (This paragraph was sarcasm in case you were wondering.)
Galileo: "The Earth revolves around the Sun!"
Score: -1 100% Flamebait
I maintaint that there is nothing in software that can be called an Engineering Discipline. It's just not that important. People die if a bridge isn't designed right. People have to do things by hand if the software doesn't work. Which of these is more real?
.mdl that you ftp to some cut-rate sweat shop in Bangalore.
Coding is not an engineering discipline. Coding is typing. Coding can be done from a Rose
I'm going to get flamebait for this but I don't care. I'm a programmer and I know my line of work is a crock of shit, but I'll take all the money my employer will give me.
You are probably right, I am sure there are a large number of successful C+ projects. On the other hand there are probably a number of successful large VB/C/Perl projects out there, which probably indicates the language is not the overiding factor on development.
The main problem I see with OO systems is that they have been put forward as a panacea to all software woes, when in fact they also bring some of there own problems to the party(see MFC for details). The OO paradigm has been oversold over the years in the detriment of other ways of working
Choose your allies carefully, it is highly unlikely you will be held accountable for the actions of your enemies
Not knowing the industry or the apps, I'll just resort to ad hoc speculation. But, hey, this is /. afterall :)
What are the training costs for the two? Is OneWorld significantly easier to learn and use? Does it interoperate with more 3rd party programs? Is it more friendly with respect to data input and output?
Odds are good that ManMan is actually more efficient for a trained operator - but the cost of getting that trained operator is relatively high. On the other hand, you can plop down half a dozen monkeys in front of OneWorld and get results.
True or not? If not, then I definitely have to wonder what advantages OneWorld actually presents over ManMan, other than support and maintainability (which, without a doubt, are huge requirements in the business world).
People die if a bridge isn't designed right. People have to do things by hand if the software doesn't work. Which of these is more real?
On the other hand, if a sidwalk curb isn't designed right, it wastes some concrete. People die if the software in their pacemaker doesn't work.
It's the project, not the discipline.
I wrote my first computer program in 1974 or 75 and have been a professional programmer (meaning that I got paid to write code) since '79 or so. School was mainframes and an early Wang desktop system (Basic and punched cards, oh yeah, baby!) I later moved into minis, mainframes, and I've been working with desktop systems since CP/M and the S-100 bus, so I guess I've seen a little of the history, anyway.
In my experience, the actual process of coding has greatly improved over time but the process of developing software hasn't improved as much. As other posters have pointed out, object-oriented tools, technologies and techniques (among other factors) have greatly facilitated the generation of code but the management of the process; deciding what gets coded, when, by whom, etc. is little better now in actual practice than it was in the late 70's or early 80's. In fact, in my opinion the situation is in some respects worse.
Management of software development today makes the same mistakes and operates under many of the same misguided assumptions as it did back when I spent my day in front of a dumb terminal. Adding outsiders unfamiliar with the details of a project makes the project later, not earlier, etc.: all the platitudes we know and love are still with us, still the butt of Dilbert jokes.
Technology may change; people aren't quite so amenable to upgrades, I think.
Uhm, like Borland Delphi /
C++ Builder / Kylix / JBuilder have been doing for the past 6-7 years ??
I have both used VS 6.0 and the new VS.Net, and I just find it very familiar to what Borland has been doing for a long time.
If our client's don't force us to do a project in VS (some do, because "it's corporate policy"), we use Borland,mostly C++, but the JBuilder for cross-platform. Now that Kylix supports C++, I think we are going to use that for Win32/Linux apps.
Adriaan.
RogerWilco the Adventurous Janitor
Score 4: funny?
I'd say: Score 5: tragic.
But damn I agree. Perl zealot needs not apply: I've been there, I learned, I understood, I realized nothing good is coming from there.
CGI.pm --> the way to generate HTML documents? wtf, the function names are even more verbose than typing in tags directly. That's why people will always embed html in print statements in Perl. My eyes still hurts just remembering how f***ing painful Perl is sometimes.
A message from the system administrator: 'I've upped my priority. Now up yours.'
I work for a Fourtune 500 company (intern) working with a software & hardware development model. The model that we use constantly goes through changes and is very sucessful in terms of delivering software and hardware on time, and defect free. I have no clue about budget, but I'd guess that it is not over budget either. Working here has given me a favorable view of software engineering. Every process is laid out and explained, and there are many tools at our disposal, including a requirements tool, a UML tool, code generation tools, code testing tools, etc.
Now for the rant. I also go to a large University 20 miles down the road, and I am taking their software engineering course (<- yay for relevance). Among CS majors there is a bad reputation for the SE class, that it is never organized well and leaves students with a bad taste for SE. In my class, we are currently in the 'Design' phase for our semester-long project, we will be done tomorrow. The Implementation phase is a week long. That's right, a week. We spent months on the requirement and ooa phases, but they are still unclear (because the document that calls for them is unclear), and now we have a week to iron everything out. This class leaves me with an unfavorable view of software engineering.
I think that more colleges should teach software engineering not as just one class but as an approach along with code classes. It makes sense when teaching people how to code that one should also teach them the likely environment in which they will code, otherwise students will get the impression that they can opt-out of having to deal with software engineering.
Has software engineering become better? Probably. Would you be able to tell when taking the class that I am? Hell no.
You should duck from the Java purist flames because you and the poster you're replying to are wrong. Java hasn't been interpreted for years. All Java virtual machines use some sort of JIT mechanism to compile the code before its run.
My blog: http://jkratz.dyndns.org/~jason/blog/
The idea of course being that good modular design, and good use of classes can increase reuseability ad infinitum, so that eventually there will be no task and no problem that cannot be assembled lego-like from blocks of premade reuseable code. The marvelous technology in Douglas Coupland's Microserfs ( called Goop wasn't it ? ) was really the epitome of this concept, a totally reuseable code system that was so generalized, so modularized that anyone with no more experience in programming than a child could assemble working programs for just about any purpose in a drag-and-drop virtual lego-building reality.
Any student of the history of science and science forecasting should begin to smirk at this point. Is it any surprise that these visions have not materialized? The hard truth, IMHO, is that logic is inherently not conducive to such high degrees of abstraction and modularity. For given tasks which are fixed and well defined with completely known and understood parameters and requirements than yes, abstraction and modularization can be a great boon in optimizing design and improving the development cycle. We can see the great advantages that this has yielded in some areas where the requirements are fixed and the applications are mostly derivative of each other. GUI design is a great example, and there are a multitude of GUI building toolkits & development environments that take great advantage of this.
However the whole thing breaks down when you move from the known to the unknown. When you try to extend this principle from fixed tasks to hypothetical ones in which the basic requirements are not envisioned. I would argue that there is a basic lack of understanding among some of the proponents of these techniques that at a fundamental level logic cannot be generalized out across all possible tasks, and all possible configurations. This was similar to the revelations of Godel's theorem in mathematics already in the 1930's - that any axiomatic system could be consistent but never complete. In reality adapting one set of code to a purpose other than that for which it was designed, or with parameters other than those originally envisioned usually is more trouble than it is worth, and often you would be better served by building new objects from scratch. You will always need intelligent educated people to design these things; there is no such thing as logical lego.
Unfortunately it seems to me that many have not gotten this message yet. Sun and Microsoft are still actively humping that dream of infinite modularity and drag-and-drop programming design. In my experience with both Java and .Net, I have found that I always run into blocks where the established object model is too constraining and has too many built-in assumptions about what you're going to be using the classes for, and so I have ended by coding new versions from scratch. Of course it may simply be the nature of the applications I'm working on, and your mileage may vary. Ultimately I think that for derivative applications this kind of abstraction and generalization is definitely an improvement, but when you come to applications that move beyond those fixed tasks it actually becomes an impediment not an advantage.
There are a thousand forms of subversion, but few can equal the convenience and immediacy of a cream pie -Noel Godin
Yes, I think things have improved.
.Net Beta 3 to generate controls, will understand what I mean.)
Such things as Structured Design, and OOP have made coded reuse better.
What hasn't improved:
1) Programmers STILL refuse to use tools that could help them in productivity. (i.e. source debuggers, instead of writing printfs around everything and printing out your variables)
Tools tools TOOLS people. Use a source debugger and save yourself a great deal of time.
If you can't use a different language or infrastructure to write the code.
Sadly, many programmers still do not use source debuggers, citing a waste of time. But they will sit there and hack over and over again trying to understand the code they write with printfs!
Tsk tsk.
2) Cost Time Estimation. Wow, talk about almost zero improvement there. Almost zero, but not quite zero. After all, most people are now adopting an open source strategy so that even if your estimates are off, the cost penalties are reduced. Furthermore, most people are beginning to realize that you have to complete a full requirements document, and do some fact finding before you attempt to quote work.
3) Finally, the hardware we use to write software is vastly more powerful, and as a result we can run much nicer environments on our machines when we write code, such as API references, etc. I have far fewer references now days to things like Java for example than I use to have to keep on my desk. Primarily because with the rise of IDE's the development environment can answer alot of questions I might have about the language I am using to write the software with.
---
I would also like to point out things have got a little worse. If you believe like I do that 80% of the work in writing software is debugging it and maintaining it over its lifetime, then you like me have problems with our IDE's.
Primarily when our IDE's produce automated code for drag and drop environments. They produce horrible code, at the expense of saving time now, and end up costing a great deal of time later.
(Anyone use the latest
I primarily write only Java code, but even my SunONE environment produces some pretty cruddy stuff if I am writing a desktop app.
I think automatic code generation is a step backwards in many ways, and ends up costing more money to fix or maintain it.
I still think a code "repository" built by humans, and nicely documented, like a cvs tree for example, is the better way. Time spent on the CVS code repository for building customized pieces is time much better spent IMHO.
-Hack
Got Geometrodynamics? Awe, too hard to figure out? Too bad.
Look at the software written twenty five years ago, and look at the software written recently.
Bill Joy wrote vi pretty much by himself. Bram Moolenaar write Vim pretty much by himself; it's a huge superset of Joy's editor.
The original portable C compiler (PCC) was about 5,000 lines of C. No one even blinks if a graduate student writes a far more sophisticated language processor, e.g., a C++ compiler, a complete Java development environment (including a compiler and debugger).
The original SimCity was awesome. No one thinks twice of re-implementing it for a PDA or a Java-enabled web browser.
What's changed? Programmers don't have to worry so much about CPU, disk, or memory limitations. The tools (compilers, libraries, source control) are much improved. Some of the new languages are far more productive. There are also new practices, and the results of lessons learned, on how to do development; some programmers take advantage of these (not enough!)
But our abilities haven't kept up with our aspirations. Compare SimCity to the masively multi-player Sims Online. How do vi or PCC stack up against Eclipse? Look at the text formatters of twenty five years ago, and then look at OpenOffice (or Microsoft's stuff); at Unix v6 vs. Max OS X.
Software hasn't kept up with Moore's Law. We're running a Red Queen's race, going as fast as we can just to stay in one place; and we're losing.
Stupid job ads, weird spam, occasional insight at
Software is at best a cottage industry of craftmen that have widely different abilities. At worst, it is a cottery of alchemists that promote their own secret snake oil recipes and only succeed by sheer luck.
And I am a developer that calls himself a software engineer, so this is not a flame.
Why craftmen? Because we develop with tricks and recipes, often called "processes". But neither of these are scientific processes. They cannot predict the outcome of a software project within a definite set of constraints. They cannot be disproved. And they cannot explain failures. So they aren't science, they are rules of thumbs. The guy who makes a living by applying rules of thumbs and learned tricks is a craftmen.
Why alchemists? Because scientists publish their methods and their results. To the contrary, the software industry hides its customer references and does not publish its source code (with the notable exception of Open Source, with remains the exception in large-scale software projects). This is how alchemists and "savants" worked in the Renaissance. They hid their trade secrets, they had confidential relationships with rich patrons, and they feared full disclosure.
On top of that, each subbranch of computing has its own lingo and redefines as many words as possible. Mathematicians who specialize in field theory topology may not understand number theory, but at least, they use distinct, well-defined jargons. In computing, terms like "record", "field", "server", "link" are so overloaded that they are just noise.
About 60% of all software projects are cancelled or written-off as failures. I don't think civil engineering or, say, aeronautics have such a abysmal track record.
I hope that some day, we'll practice and teach Computer Science as, well, a science, not as a craft.
--
Mad science! Robots! Underwear! Cute girls! Full comic online! http://www.girlgeniusonline.com/
Actually you bring up an excellent point that I'm afraid will get lost, but anyway.
What you described is an "ideal" situation, but then the ugly reality of programming rears it's ugly head. What if you want to sort by something other than the first column, your data abstract should support that (a property perhaps). What if you want to sort an object by a member (do you supply a comparison routine that now makes the abstract not quite as abstract), what if you need to sort unicode, what if you're sorting 500MB of data. All these extra requirements have a tendency to break many abstractions and force the programmer to A) have a lot more knowledge about the tool B) make any particular tool less likely to fit the bill.
After years of debating and haggling with OOP fans (I think OO is way overhyped), I have concluded that an important factor is the personal psychology of the individual. IOW, "mindfit".
People simply perceive things differently and are bothered and helped by different conventions, notations, traditions, etc. Each individual has to find their *own* silver bullet, or at least a brass one. Some of the surveys by Ed Yourdon seem to back this. OOP scored higher when "OO fanatics" worked on projects, but as OO went more mainstream, its score faded into the background noise (average).
Perhaps IT shops can focus more on screening for individuals who think alike rather than simple buzzword matching. It perhaps is time for an inkblot-like test for developers.
I used to complain about how Perl is a "write-only language", but I later realized that it is just write-only to *me*. If other Perl fans can figure out Perl and are productive under it (both in writing and maintenance), then I see little reason to complain. A given shop just has to be willing to accept the fact that they are married to Perl and Perl-loving developers. But if they can get things done, then why should I or anybody else fuss? One man's spehgetti is another's gormet favorite.
(It is still important to explore other viewpoints to expand your horizons, but if something does not seem to "click" for you after a little while, then don't feel bad. I am tired of "you just don't get it, neener neener" from various fans of different paradigms or languages.)
Table-ized A.I.
Please come back when you know what you're talking about. Until then, HAND.
Just as the hardware industry has grown from abstracting commonly used circuits into chips, what occurs is the creation of building blocks on standard designs. In the software sector, that would equate to a component architecture which keeps building on what's been done in the past. The big problem here is that if ANYBODY starts trying to hide anything in Microsoft Windows, Microsoft eliminates them from the market. C++ frameworks were very popular in the early 90's but they all but disappeared as Microsoft provided their C-- (object like) way of doing things at a financial loss. Borland was a leader in the C++ dev tool market but their work hide MS Windows API's so much you could start building applications that recompiled on many different operating systems. MS gutted Borland of it's top level design engineers and paid Borland a tiny fee to settle out of court. CORBA was another framework for building applications across a network of different operating systems and languages. We got 3+ years of intense MS-DCOM press coverage and CORBA eventually faded. I worked on a project which was to use CORBA for a large military hospital system but 1 year into it we were told to stop and start using MS tools and languages with no explainations. Java did/does the same thing( OS/API abstraction ) and it too was fought by Microsoft with incredible gusto.
As long as Microsoft holds the major share of the desktop computing platform, they will not allow anybody else to decide what's to become a standard software component or API. And changing this every 2 years or so keeps the $$ flowing from your pockets into theirs.
Sure you can do some of this within your own organization but as an industry, the largest software company in the world opposes such thinking. And with $30 billion in cash, they have the power to change the minds and directions of whole countries.
IMHO
LoB
"Anyone who stands out in the middle of a road looks like roadkill to me." --Linus
However, take a look at Parnas' classic paper on information hiding. In it, he makes the following statement (emphasis mine):
Is there anyone in the crowd that doesn't think he could write a shell script/perl script/etc. to accomplish this in 30 minutes or less? I'd be willing to bet I could do it in Python in under 15 minutes.This paper was written 30 years ago. It only pre-dated Brooks by 5 years. In that time, if my estimate is accurate, the development time for simple programs may have fallen by a factor of a hundred or more.
Software doesn't appear to have improved in that time simply because we keep trying to do more complex things.
Patrick Doyle
I mod down every jackass who puts his moderation policy in his sig. Oh, wait a sec....
May I point out that programming languages can be used to develop things other that software (although related) nowadays.
Probably, the thing I find most interesting is hardware, which can be described using Verilog or VHDL, both Hardware Description Languages. That, together with technologies like FPGA, enables a programmer to design his own microprocessor if he wishes to do so, I find that revolutionary.
That one sentance, which was Brooks' key insight, sums up why progress will always be limited. You can read commentary on that point in a paper I wrote here.
Other's have already pointed out the obvious corollary: good management practices are most important for successful, reliable software development.
I don't think the Agile people need to be trashed as much as they are here. Sure, they are gurus. But they are emphasizing human-centric instead of "software engineering" which is a Good Thing (TM). Just don't get too religious about XP and you'll be fine.
Other than that, the greatest thing for programmers since sliced bread is Python.
And yes, I also agree open source development has pushed this industry light years ahead. But it works because - it's human-centric programming!
Those operating systems were written in Lisp to about the same extent that, for example, Linux is written in C, that is, about everything except some assembler parts was Lisp. And it all was at the control of the user, at runtime. You could debug and browse the OS sources, modify them on the fly (no rebooting just because you changed something about the filesystem implementation or other trivialities), etc. Pretty impressive environments, even from todays point of view.
You can even still buy Genera for Alphas, but it's not exactly cheap, and finding a Symbolics representative may prove harder than someone selling WinXP copies...
Sad, but true... It is really amazing to see what Lisp or Smalltalk users had years ago, and how only parts of this show up in recent "innovative" languages and tools. Sometimes I think that the history of computing should be made an obligatory part of CS education. Quote Richard Gabriel, "our competitors will be spending all their time trying to figure out that it's really possible we're doing what we're doing, because they will be thinking in terms of customization at compile time or link time, not at runtime." (Although I don't agree with every aspect of his rant.)Programming can be fun again. Film at 11.
I think the tech market is very good right now. However, you have to look at it from the right perspective.
I'm in a tech company that has been growing over the last 3 years. The problem is knowing how to find business.
Every business wants to automate what they do. Every business wants to operate better and faster. The problem is, most tech companies know very little about how to actually do this, especially how to treat and care for customers.
Get to know the people and companies around town. Find out what their problems *actually are*. Find a way to fix them, possibly using technology. Present your solution and your price.
There are always people with problems that are willing to pay for them. What happened in the tech industry are:
a) it grew so fast, "developers" were in such demand that completely unqualified incompetents were being hired. In addition, HR had no way of distinguishing between competent and incompetent people. This is probably the biggest problem.
b) Even fewer people cared about actually serving consumer needs. Upgrading to Windows 2000 became more important than making my processes more efficient.
c) Management finally wised up and stopped doing useless things with technology. However, there is still a demand for useful technology, you just have to be able to justify it to a much smarter crowd.
d) Management refuses to go along with technology people who can't communicate.
Tech people are used to being able to just make sales based on the fact that they knew more about technology than anyone else. Now they have to be able to actually help solve problems. I see this as a good thing.
By the way, solving problems is something that can't be outsourced to third-world countries. It requires personal communication. It requires being able to see technology from the point-of-view of the business owner, and being able to speak intelligently and understandably about them and their problems, and only speak about technology when it's absolutely relevant.
Engineering and the Ultimate
Perl and Python are my two favorite languages. Them are fightin words....
Perl is amazing, because you can get so much done in so little time. People say it's hard to read, but it really only takes one or two readings of the camel book to be able to understand all but the toughest perl.
It has support for all stages of development, and is really useful as your project grows from procedural, to object-oriented, and even if it needs some functional aspects. A scripting language as easy-to-use as perl with lexical closures is simply amazing to build with.
In addition, it has it's own documentation system that goes beyond what Javadoc does, and allows you to really document the what's and why's of your interfaces.
Engineering and the Ultimate
I'm sorry for my late response, a virus penetrated my firewall and set the automatic defense system in motion: increased priority to the 'White cell' service at the expense of the Intelligence app.
;) The problem with avoiding change in your app is that you get code rot. The real world changes while the code stays the same. You add local hacks to support feature X, Y and Z, messing up your code. Changes become harder and harder to make. You start to get massive code duplication and (unnecessary) interdependencies. Your resistence to low-level change catches up with you because you are now forced to make all kinds of wide-spread, buggy hacks just to support the features that your customer really need. Finally you throw out the mess and start again.
Named parameters is to *avoid* find-and-replace.
Yes, I know. But the effect will probably be substantially smaller than using refactoring(s). My point was more that they don't exactly overlap, refactoring can do much more than named parameters and also a bit less ( named parameters can be arbitrarily ordered, may increase the readability of function calls and allow functions with the same name+parameter type(s) ).
I actually consider this "refactoring" push a *failure* of OOP, or at least a bad idea. Rather than trying to avoid the Code Change Gods, it is giving into them big-time. I don't get it.
I guess you don't like Extreme Programming then
XP tries to avoid this as long as possible by making it much, much easier to keep your code up to date. Many of the XP rules support this:
- no ownership of code
- no clever tricks, KISS
- rigorous unit tests
- only create the code that you need now
- integrate often
- refactor regularly and in small steps
Having unit tests for just about everything is a very important rule in this regard. It gives you confidence that you can change the implementation without changing the functionality. This allows you the maximum amount of freedom. You keep the code lean and mean. The design evolves along with the addition of new features. You won't be prevented from making necessary changes just because it's 'too hard to do' (which happens in real life - a lot). I understand why it is hard to grasp, this goes against the engineering principles that you where taught: gather all requirements first, make a detailed design, etc. Of course, we all know that the world doesn't conform to the waterfall model and that software is especially prone to changing requirements. There is only so much you can do to reduce the effects of this threat. Instead of fighting it, XP embraces this law of nature (which makes it very effective for some projects). Refactoring is an important part of this strategy and I don't see why you consider it pushing a failure. Is procedural programming immune to code rot and changing requirements? Is any programming paradigm?
The Drowned and the Saved - Primo Levi