Attendees at last week's Embedded Systems Conference got an earful on the high price to be paid by poorly written or tested code.
"Why did they inspect software only after people died?" asked Ganssle, who said a court case on the crash is still in litigation.
Some pacemakers have stimulated hearts to beat at rates of 190 beats a minute, prompting companies to provide software updates delivered to the implanted devices using capacitive coupling. Unfortunately, other pacemaker patients have had their devices inadvertently reprogrammed when walking through metal detectors. In 2003, the pacemaker of a woman in Japan was accidentally reprogrammed by her rice cooker.
A Thai politician had to have police bust the windows on his BMW 745i after a software glitch caused the electric doors and windows to freeze in a locked state, trapping him inside. Ford recalled some models of its 2000 Explorer because lights and wipers would not work in some circumstances. And the 2004 Pontiac Grand Prix faced a software recall for a leap-year fault.
Part of the problem lies in poor engineering discipline, such as a lack of adequate testing, improper error handling and inherently sloppy languages. Management issues, including a demand for ever more features in compressed schedules, and tight budgets are also to blame.
"We need to test everything up front and integrate testing into the design process. Then we need to believe the data we get when we do test," said Ganssle.
When engineers make a change because of a failed test, they often neglect to go back to the beginning of the test suite to make sure the changes haven't introduced new errors, said Dave Stewart, chief technology officer of Embedded Research Solutions Inc. (Annapolis, Md.) in an ESC session on the top problems in real-time software design.
Engineers need to create error-handling modes in their programs, and the modes must exist as just another state for their systems and treat errors as one of many possible inputs, Stewart added.
Fasanelli of Ericsson gave a detailed prescription for how to find, report and minimize faults in embedded software. Programmers must make it a standard practice to classify all inputs and states of a system and note any illegal inputs or edge states, whether or not they affect a program's ability to run, he said.
In addition, programs should routinely track and report their own performance, idle times and memory integrity. Creating such debug features may affect a system's cost, but that will be offset by reduced maintenance, Fasanelli said.
"Exception handling is particularly hard to test because it's hard to generate the exceptions. These tend to be the most poorly tested parts of code," said Ganssle.
Riding a rough C Ironically, today's most popular programming languages, C and C++, are among the most error prone. That's because C compilers have plenty of latitude to compile and link without providing any diagnostics code that can produce serious run-time errors, especially when ported to a new processor.
"There are a lot of little goodies in C that programmers are not fully aware of," said Dan Saks, an author who has documented nearly 40 "gotchas" he presented in a session at ESC. "The lesson is to understand what you can assume and what you can't."
We welcome your comments on this topic on our social media channels, or [contact us directly] with questions about the site.
2018 State of the CloudCloud adoption is growing, but how are organizations taking advantage of it? Interop ITX and InformationWeek surveyed technology decision-makers to find out, read this report to discover what they had to say!
Infographic: The State of DevOps in 2017Is DevOps helping organizations reduce costs and time-to-market for software releases? What's getting in the way of DevOps adoption? Find out in this InformationWeek and Interop ITX infographic on the state of DevOps in 2017.