Published in ESP, July 2000
|For novel ideas about building embedded systems (both hardware and firmware), join the 27,000+ engineers who subscribe to The Embedded Muse, a free biweekly newsletter. The Muse has no hype and no vendor PR. It takes just a few seconds (just enter your email, which is shared with absolutely no one) to subscribe.
By Jack Ganssle
The night, illuminated only by a universe of stars, swallows the sounds of our movements as we ease through the brush. Coming over a rocky rise we spy a flickering light, a beckoning distant campfire whose promise of friendly warmth draws us closer. Though it's been a long day and a tiring trek wariness still prevails; we approach these unknown humans downwind and quietly lest the need to protect their turf turns into a violent confrontation.
Closer now, we listen to a language that's been unheard for millennium, but through the miracle of fiction we somehow comprehend. An old man, perhaps 30, gray, bearded, and sickly relates a story in a mixture of song and epic poetry of the history of their people. He heard it from this father, who passed it from other, now dead, generations. The younger folks listen enraptured by the story. In the manner of youngsters everywhere they little realize that this is more than an interesting tale. Years will pass before they, too, pass this oral history on to their genetic legacies.
It's 4000 BC. The printing press lies five thousand years in the future. Writing won't appear for another two millennia. Yet a million years have passed since man became more or less self-aware, capable of learning, a tool-maker who, perhaps in small ways, strives to improve his lot. Though the great explosion of new ideas and revolutionary ways of living lies far off in the future, much has been learned over the generations. Until new technologies come along humans pass this information along just by talk.
Somehow humankind, though, has learned the most effective way to package information passed this way: the story. We remember little of a dry recitation of facts, but are enraptured by stories, as their elements weave permanent patterns in our minds.
We time travelers press the fast forward button on the Way Back machine and zoom ahead. Writing appears; some of the earliest written words are the stories formerly passed around the campfire. Jumping ahead once again we hit the information age. Lo and behold, even now the most reproduced book in the world starts with a recitation of many of those campfire stories. Chapter 1: Genesis. We listen to little children reciting tales of a plague that tortured the world 500 years before - "Ring Around The Rosy".
It appears, despite astonishing new communications methods and channels, new "learning strategies" and developmental fads, that humans learn best from something inalienably simple: stories and storytelling. It's a fact we neglect at our peril.
This longwinded introduction is my tale of how I was struck by Michael Barr's article in the March issue of this magazine about selecting appropriate languages for embedded systems development. I started reading the piece with the expectation of a dry but precise engineering approach to language selection. Instead, he told a story. One that showed both triumph and failure, a high tech version of heroic struggles between the forces of good and evil (if you define "good" as getting a bug-free product out on time!). Beowulf and Barr.
As a vehicle for learning, especially when there probably is no scientific formula to get the most optimized result, the story is king. It's interesting that after 50 years of computer science it seems we still have no definitive way to select a language, even though that's the most fundamental choice we'll make for a project.
Is Ada or Forth the right choice? C or C++? What role should assembly play? I have my own ground rules, though these are honed only through the vicious lessons of bitter experience. Yours may differ. Perhaps, though, I can take a cue from Michael's approach and share some of my own stories with you. Epic they're not. To me, though, they've proven instructional.
A C++ Crash
A couple of years ago an instrumentation company asked me to visit their site and probe into a project. On the phone the manager sounded in-control and confident in the team. The company had a broad range of successful products and, secure in their technology, were working on a new generation device planned to be the basis of future years of products.
Somewhat reluctantly I agreed to the visit, hoping to learn something from how this manager was able to coordinate a decent sized team working on quite complex products. The successes claimed on the phone sounded so good, compared to the normal chaos of development! But why did they want to see me?
At the site I found an army of developers working in spacious quarters. Unusually, private offices were the norm, capital equipment budgets were fat, and the manager was a truly enlightened individual dedicated to bringing the best software processes into the firm. It seemed there was little I could contribute other than admiring comments.
After a couple of hours listening and looking, bits of pain started to leak out. Though past generations of products were indeed successful, something seemed awry with the current project. 40 firmware developers were clearly working hard on this new gadget, but time leaked from the schedule like cheap wine from a street person's flask. A year into the project they were already 6 months late, getting later at an ever increasing rate.
Worse, the system had grown beyond anyone's expectation in size despite no scope changes. Bugs crawled from every function. An entire group was dedicated to bug fixes, but like a hydra fixing one created three more.
As the day wore on and the manager slowly revealed ever more problems, his facade of success finally broke and almost tearfully asked me what I thought had gone wrong.
The answer, if not clear to him, was obvious to me. Not at all because of superior intellect or experience; rather, I drew on the consultant's secret weapon: theft.
Throughout the day I had had the chance to talk one-on-one with many of the team members. They knew what the problem was. Though some had trouble separating details from menaing, the engineers knew all too well what had happened. I merely stole their thoughts and reformatted them into one idea the manager could understand. Communications problems are at the root of so many troubles, from home life to the office to global political issues.
"The code is junk because the developers are just hacking away," I told the manager. "No one really understands the language, and so they build things almost randomly in the hopes things will work."
I went to the doctor once after reading a newspaper article about a disease that seemed to match my minor symptoms. After telling her about the article the doc rolled her eyes and admonished me against playing amateur medical person. In fact, nothing was wrong; my utter lack of knowledge in this field meant I was ripe to believe the overly simplified description of complex issues jammed into three paragraphs of USA Today.
Similarly, the president of this company had read a Wall Street Journal description of the benefits of Object Oriented Programming. The promise seemed too good to be true: OOP meant, or so this techno-novice read, that all code would become reuseable. I credit the president with identifying that reusable code could solve many problems. Unhappily, just as with my experience at the doctor's office, the Journal's piece couldn't do justice to the complexities of software development. This layperson learned just enough to be dangerous.
He issued an edict: from now on, we're doing all projects in C++! He blithely went on about his normal management, sales and accounting functions, little realizing the terrible implications of such a far-reaching rule.
Is this a rant against C++? Not at all. A year and 100,000 lines of junk code into the project it became apparent that the rule was bad. The edict meant 40 developers were suddenly plunged into an environment they knew nothing about.
For, it turns out, when this project started only one developer, the only one fresh out of college, had even the slightest exposure to C++ and OOP. 39 others were learning on the job, cranking code and making mistakes on a mission-critical project, as they tried to grapple with the very different philosophies of OOP. Working with objects means thinking out the design of a product in a very different way than most C designs. C++ is an approach, as well as a language. Trying to have the developers build an OOPy design as they learned the concepts was a bit like designing the space shuttle's orbital dynamics as the engineers learned calculus.
The result was a mad mishmash of conventional procedural code mixed with shoehorned objects, a design that exploited the worst of both kinds of programming. And so I recommended they trash the code and start over. In C. All 40 developers knew C, all 40 were experts at building systems using a C framework.
The problem wasn't the language but an arbitrary edict that neglected a critical part of the success of any embedded systems: having expert people. Though the president's wish for reusability was right on, his rule was as disastrous as telling a newspaper reporter to write in Latin. Without people skilled in the development environment you're doomed.
And never expect skills to appear magically overnight.
So to this day I hate having told them to toss out the code - though it was the right choice. I made the same recommendation as did Michael in his story, which is unhappily the universal refrain of consultants around the world. "Toss out the code - it's crap - we'll start over." Firmware is so terribly expensive! But it's impossible to retrofit quality into lousy code. Maybe, with enough effort, we can get rid of most of the bugs in a poorly written system, but maintenance becomes impossible. And crummy code always has nasty, lurking bugs that just seem unconquerable.
So the moral of this unhappy tale is that people's skills are far more important to the success of the project than any technology. No tool, no technology, no "new new thing" will save us from the realities of development woes. Skills first, technology second.
They Never Die
When Intel invented the microprocessor in 1971 I was working as an electronics technician at a small instrumentation company here in Maryland, putting in too many hours to support myself while in college. Within a year of their 4004, the wizards at Intel came out with the 8008, a truly useful microprocessor that, in my opinion, started the true microprocessor age.
The engineering manager at my company recognized the value of this part to our products and started designing a new product around the part. The problem: the engineers there didn't know how to program computers. Why should they? In the early 70s computers were still mostly inaccessible beasts used more by the IT folks than poor EEs. Somehow they found that I knew assembly language - for a Univac mainframe - and promoted me to the awesome status of engineer. My job was to write the 4k firmware package for the product. Never has so little been known by so many, never has so much on-the-job training happened in such a short period of time.
We eventually did manage to ship a number of these units, which were successful enough that customers demanded ever more features, more than could fit in the 8008's 16k address space. It's 1975 now, and the 8080 has hit the market. Through some quirk I'm running the greatly expanded micro group, and we redesign the instrument around this new processor. It too ships.
We designed the product using another Intel invention, the EPROM. They guarantee data retention of ten years, which of course seemed varnishingly close to infinity to our very young development team.
A decade passes and, as is always the case, individuals disperse even though the company continues. The product is obsolete, but customers still have them working, measuring oil content on polyester fibers in a factor environment. Ten years to the month after shipping the first of these units they start dropping bits in the field.
The repair crew at the company discover that all of the original object files exist on the only media of the mid-70s: paper tape. No one has a clue how even to boot up the Intellec 8 "development system", let alone load the tapes and reprogram the EPROMs. Self-employed now, I'm surprised by a panicked phone call from them. Plumbing the dim recesses of memory I'm somehow able to remember how to use the tools, reload the now-brittle paper tapes, and reburn the EPROMs.
By the mid-90s I expected another call as another decade sailed by, figuring the memories would once again need refreshing. None came; presumably the instrument had finally been retired. But I learned a critical lesson from this experience: embedded systems never seem to die. They run quietly in the background for years. If we don't plan ahead for maintenance that may literally span generations we're fools. And surely the Y2K fiasco should have taught us the same lesson about software in general.
I won't even try to tell you how to select your next embedded lingo. However, from the lessons above, it's clear to me that we simply must select one that has been standardized. Prior to the C's ANSI standard dozens of variants competed for attention. Portability was impossible. In my opinion, C++ has only recently become a viable embedded choice, since its November 1998 ANSI standard. We know that both choices are safe, that a C or C++ programmer ten years from now will be able to work on our ANSI-compliant code.
At the risk of offending lots of people, I don't see Java as a serious embedded contender just yet, unless your system will go away in a year or two. Java is a language in flux, torn asunder by marketing forces of giants. It's coupled, for better or worse, to the two biggest forces of change in the universe: the PC and the Internet. I suspect that eventually it will indeed be standardized, and in fact may even supplant C++ due to its simpler, less overwhelming, syntax. But for now, be assured that the Java we write today will look very different in a few years.
The exception to standards (because there's always an exception in embedded work) is assembly language. Fact is, this is different for every processor, is inherently non-portable, and will always be a tough way to write code. Though I do love it so! Assembly makes sense only for time or space-constrained apps, or for small parts of a bigger project where speed is truly an issue.
When my son was very young I'd end the evening with a made-up story. Most were silly things meant to elicit a laugh. When traveling I'd often write a short bit of nonsense that his mom would read in my absence. Now, just barely a teenager, he seems to have forgotten all of the rules I've so carefully tried to instill! but does remember these stories even long after my failing middle-aged brain hasn't the slimmest recollection.
Stories seem to fit the nature of the way the human brain works, much more efficiently than dreary facts. Urban legends spread like wildfire as arithmetic skills collapse. And so, another moral for us all is to learn to frame important ideas in the context of stories. For only those will be remembered.