The other day a software project that I had spent a year on was “put on the shelf.” A year of effort was unceremoniously flushed down the software sewer. A number of colleagues asked me, “How do you feel about this?” Would you believe relieved?
This is not false bravado or stupid sunny optimism. I am not naïve. I know this failure will hurt me. In modern corporations the parties that launch failures, usually management, seldom take the blame. Blame, like groundwater, sinks to the lowest levels. In software circles it pools on the coding grunts doing the real work. It’s never said, but always implied, that software failures are the exclusive result of programmer flaws. If only we had worked harder and put in those sixty or eighty hour weeks for months at a time; if only we were a higher level of coding wizard that could crank out thousands of error free lines of code per hour, like the entirely fictional software geniuses on TV, things might have been different.
Indulging hypotheticals is like arguing with young Earth creationists about the distance of galaxies.
“Given the absolute constancy of the speed of light how is it possible to see the Andromeda galaxy, something we know is over two million light years away, in a six thousand-year old universe?”
The inevitable reply to this young Earth killing query is always something to the effect that God could have made the universe with light en route. God could have also made farts smell like Chanel #5 but sadly he did not. Similarly, God has also failed to staff corporations with legions of Spock level programmers ready and willing to satisfy whatever au courant notion sets up brain-keeping in management skulls. The Andromeda galaxy is millions of light years away, we don’t live on a six-thousand year old Earth, and software is not created by magic. Projects work or flounder for very real reasons. I was not surprised by the termination of this project. I was surprised that it took so long to give up on something that was never going to work out as expected.
Working with bad legacy code always reminds me of the scene in The Empire Strikes Back when Han Solo rescues Luke on the ice world by cutting open his fallen Tauntaun with a light saber. When the Tauntaun’s guts spill out Solo says, “I thought they smelled bad on the outside.” Legacy code may look pretty bad on the outside, just wait until you cut into it. Only then do you release the full foul forceful stench.
Avoiding Death Marches, or finding ways to shift blame for them, is an essential survival skill for young corporate programmers. It’s not an exaggeration to say that a lot of modern software is built on the naiveté of the young. When you first learn to program you are seduced by its overpowering rationality. Here is a world founded on logic, mathematics and real engineering constraints. Things work, and don’t work, for non-bullshit reasons! There aren’t any pointless arguments with ideological metrosexual pinheads about “privilege” or “gender.” Programs don’t give a crap about the skin color of the programmer or whether their mother was an abusive bull dyke. After a few years of inhaling rarefied logic fumes young programmers often make the gigantic and catastrophic mistake that the greater naked ape society in which they live also works on bullshit free principles. It’s at this delicate stage when you can drive the young programmer hard. There’s an old saying that salesmen like money and engineers like work so in most companies the salesmen get the money and the engineers get the work. As the great sage Homer Simpson would say, “It’s funny because it’s true!”
Death Marches shatter the naïve. Gaining a greater understanding of nasty world we live in always does a naked ape good but there’s a cost; your career will suffer and you will suffer: enlightenment is proportional to pain! The school of hard knocks is real thing and unlike that party school that you boozed through no one graduates alive. I’d strongly recommend a few Death Marches for every working programmer. To paraphrase Tolstoy, “All successful software projects are alike; every unsuccessful project fails in its own way!” You may consider what follows my modest contribution to the vast, comprehensively ignored, literature of software Death Marches.
Code Smell #1: A large mass of poorly written test free legacy code that does something important.
“Refactoring” is a word that I love to mock but the refactoring literature showcases an astonishingly precise term: “code smell.” Code smells are exactly what you expect. Something about the code stinks. The Stinking Tauntaun code base didn’t just stink, it reeked like diuretic dog shit on choleric vomit. This was well-known throughout the company. My general advice to young programmers that catch a whiff of rotting code is simple: flee, run, abort, get out Dodge! Do not spill your precious bodily fluids cleaning up the messes of others.
When recruited a number of experienced hieroglyphic programmers asked me the pointed question, “How do you deal with giant stinking piles of legacy doo-doo?” Maybe it wasn’t quite phrased like that but the intention was clear. They were telling me that they were dealing with a seething mass of half-baked crappy code that somehow, despite its manifold flaws, was useful to the organization and couldn’t be put out of its well deserved misery.
I honestly answered. “Systems that have survived for many years, despite their problems and flaws, meet a need. You have to respect that.” I still believe this! Every day I boot up ancient command shells that look pretty much like they did thirty years ago. Why do I use these old tools? It’s simple; they do key tasks very efficiently. I try new tools all the time. I spent a few months playing with PowerShell. It’s a very slick modern shell but it has one big problem. It takes far longer to load than ancient DOS or Bash shells. When I want to execute a few simple commands I find the few extra seconds it takes to get a PowerShell up and running highly annoying. Why should I wait when I don’t need the wonderful new features of PowerShell? The Stinking Tauntaun System also met user needs and, just like I am horrified by the clunky ill-conceived incoherence of old shells, Stinking Tauntaun users held their noses and used the good bits of what’s overall a bad system.
Code Smell #2: No fully automatic version controlled test suite.
There is a tale, possibly apocryphal, from the dawn of programming that goes something like this: a young programmer of the ENIAC generation was having a bad day. He was trying to get some program running and was getting frustrated “re-wiring.” Our young programmer was experiencing one of the first “debugging” sessions back when real insects came into play. In a moment of life searing insight our young programmer realized that from now on most of his time would be consumed hunting down errors in faulty programs. The tale goes dark here. I’ve often wondered what happened to the young programmer. Did he have a happy life or did he, like Sisyphus, push that damn program up the hill only to have it crash down again, and again, until management cried stop!
Debugging is a necessary evil. If you program — you debug. It’s impossible to eliminate debugging but at least you can approach it intelligently, and to this day, the only successful method for controlling program errors is the fully automatic version controlled ever-expanding test suite. Systems with such suites actually improve with time. Systems without such suites always degenerate and take down a few naïve programmers as they decay. The Stinking Tauntaun System lacked an automatic test suite. The Stinking Tauntaun System lacked a non-automatic test suite. The Stinking Tauntaun had no useful test cases what so ever! Testing The Stinking Tauntaun was more painful than dealing with its code. The primary tester had nightmares about The Stinking Tauntaun.
My Power-Point-less illustration of Quality over Time for a system that lacks a fully automatic ever expanding version controlled test suite. Systems without well designed test rigs are almost without exception complete crap. Unless someone is paying you huge bucks it’s best to start sending out resumes when asked to fix such systems. You can ruin your health stirring your yummy ice cream into the shit but it’s almost certain the final mixture will taste more like shit than ice cream.
It’s the 21st century people! In ENIAC days you could overlook missing automatic test suites. Nowadays missing automatic test suites is a damning indictment of the organization that tolerates such heinous crimes against programming humanity!
Code Smell #3: Only one programmer is able to work with legacy code.
When I was hired there were a few hieroglyphic programmers on staff. No single programmer was saddled with the accumulated mistakes of prior decades. We could divide and deal with the pain. During this happy time our work was support oriented. We we’re not undertaking large projects. Then an inevitable bout of corporate reorganization occurred, followed by terminations and a stream of defectors fleeing the new order. Eventually only one hieroglyphic programmer remained: moi! Most programmers would have joined the exodus and not taken on the sins of other fathers but as I have already pointed out I am a software whore and proud of it. Still even software sluts must avoid ending up like the, “there can be only one,” Highlander. Most of those sword duels ended rather badly as I recall.
Code Smell #4: Massive routines with far-ranging name scope.
With noxious fumes already cutting off our air supply we should have stopped in our tracks but we soldiered on because The Stinking Tauntaun System did something important and alternatives were not readily available. The corporation recognized this and started a number of parallel projects to explore New Stinking Tautaun’s. I wasn’t lucky enough to work on any of the parallel projects. I had to get in the stall and shovel the Tauntaun manure.
We weren’t entirely clueless when we started Stinking Tauntaun work. We recognized the precarious hieroglyphic programmer resource constraint and tried to divide the work into units that alleviated it. We decided to implement part of the new features in another programming language and call the new module from the old system. The hope was this would divide the work and possibly create something that might be used outside The Stinking Tauntaun System. This part of the project went reasonably well. The new module is entirely new code and works very well. Unfortunately, it was a small part of the greater effort of bolting new features into The Stinking Tauntaun System.
Stinking Tauntaun code is magisterial in its monolithic madness. The hieroglyphic programming language is famous for its concise coding style yet the main routine in the Stinking Tauntaun System is close to 10,000 lines long! I had never seen hieroglyphic language routines of such length. I’ve only seen comparable programs a few times in my battle scared career. As a summer student I marveled at a 20,000 line, single routine, FORTRAN program. At first I thought it was the output of some cross compiler but no, some insane, bat shit crazy government drone, (I was working for a government agency in western Canada), had coded it by hand — on freaking punch cards no less! Such stunning ineptitude is rare: most programmers have a passing acquaintance with modular design and know enough to reconsider things when code gets out of hand. We all have different thresholds for judging out-of-hand-ness, for me it’s any routine that’s longer than sixty lines: including the damn comments.
A 10,000 line routine is a monumental red flag but The Stinking Tautaun’s length was not the biggest problem. There is this thing programmer’s call “scope.” Scope marks out name boundaries. It sets limits on where names have meaning or value. Ideally name scope is limited to the smallest feasible extent. You really don’t want global names! You absolutely don’t want global names like “X,” or “i,” or “T,” cavorting in your code! Most of the cryptic names in the Stinking Tauntaun System had far-ranging scope. Vast scopes make it difficult to safely make local changes. This makes it risky, dangerous and time-consuming to change code. Large scopes also increase the burden on testers. If tiny changes have far-ranging consequences you have to retest large parts of the system every time you tweak it. All of this sucks up time and drains the pure bodily fluids of IT staff.
Code Smell #5: Basic assumptions about what is possible quickly prove wrong.
It was looking dark. Timid souls would have run but we plowed on. Our naïve hope rested on the theory that we could change a few touch points in The Stinking Tauntaun’s code base and then get out of Dodge before the dark kludge forest closed in. My first estimate of how many routines I would have to touch was around twenty. Two months into the project I had already altered over a hundred with no end in sight. The small orthogonal change we hoped to make was not possible because The Stinking Tauntaun System did not sensibly do one thing in one place. It did sort of the same thing in many places. You couldn’t make a small change and have things sensibly flow throughout the system. I worked to isolate and modularize the mess but I should have just waved my hands and spent my days on LinkedIn looking for another job.
Code Smell #6: A development culture steeped in counterproductive ceremony.
On top of all these screeching sirens yelling stop there were other impediments. I work in a SOX saturated environment. Readers of my blog know that I consider SOX to be one of the biggest time-wasting piles of DC idiocies ever pushed on American companies. Like many “government solutions” SOX did not drain the intended swamp but forever saddled us with inane time-wasting ceremony and utterly stupid levels of micromanagement. Is Active Directory management really a concern of freaking government? Somehow it’s become one! One of my favorite consequences of SOX is the religious ordering of development environments and the sacraments for promoting code changes. During the long bruising Stinking Tauntaun project many releases boiled down to me making changes to one file! Pushing a new version to test should have been a simple file copy.
Of course it couldn’t be that simple. Numerous “tickets” had to be approved. Another branch of IT, that was even more stressed and suffered higher levels of turnover than ours, was dragged in to execute a single trivial file copy. In many cases more bytes were generated by all this pointless time-wasting ceremony than I had changed in that single file. Of course all this took time, and as hard as I looked for a useless-time-wasting-bullshit category in our web-based hours tracking system, I couldn’t find one. There’s a management mantra: you can only improve what you measure. Funny how companies never measure the bullshit.
In retrospect we should have junked The Stinking Tauntaun System or opted for a radical rewrite. It’s ironic but something like this is what’s going to happen. If I was young I would be bitter but I cashed in on this project. In my consulting days I was always on the lookout for Stinking Tauntauns: they were freaking gold mines for those of us that have acquired a taste for the bracing putrid fumes of rotting code.
 If you are the type of person that gets your panties in a knot about the gender of hypothetical supreme beings please go away.