A thought-triggering article

http://www.techradar.com/news/computing ... ths-644771

The bit about the patriot missile system was a little alarming.

At least the summary isn't misleading.

When writing multi-million pound life endangering mission critical control systems, it would be best if the programmer can maths. Sadly, such people are in short supply. It's amazing how few people understand fully the sometimes dramatic consequences of rounding off, and how they may often be avoided by simply doing a calculation in a different order.

Even when a true professional writes code, there are often time and budget constraints which mean it's released before it's ready. When code runs into millions of lines, it's not possible to guarantee 100% accuracy because it would be obsolete before the tests were complete. Some bugs are inevitable. However, something like that incremental clock bug in the Patriot targeting system was just idiotic by design.

I suspect the fatal overflow in the Ariane V was probably brought about by trying to stretch the performance of the available hardware to the limits. *wikis*



In other words, the computer wasn't fast enough to do it more thoroughly. There are often compromises on the bleeding edge of technology, and the logic sounds acceptable.

However:



LOLWTF.

Given that Ada is a piece is of piss to debug, how the hell was this missed?

Because the code wasn't even looked at. It was a known working subsystem with a proven track record. It was treated as a black box.

People often use library routines which contain redundant code, and it usually doesn't matter.

Reusing code is fine if you're talking about a device driver or a GUI but this wasn't. This was an aerospace subsystem written in a specialist language and installed in a multi-billion pound space vehicle.

If the vehicle was to be that important and expensive, short cuts should not have been acceptable.

Given my dissertation/final year project is on test generation I can completely see why this was not discovered (as far as the google bug of 599 999 999 999 999 - 599 999 999 999 998 = 0 goes)

If you look at testing 2 integers where
a - b = c

Then you have 5 million possibilities for a, 5 million possibilities for b

Testing every possible combination, that gives over a quintillion (1 x10 ^ 18) possible combinations

Assuming each test is automated and takes 1 second to run each test you are looking at 32 billion years to run the tests JUST for a - b = c, that doesn't even include the time taken to verify each answer is correct.

Programmers take short cuts or use unsafe code. Always have and always will until they are made accountable for their errors

Clearly, you're quite correct.

To be honest, without knowing all the details of what the sub system actually did I don't think it's fair to judge the decisions made at the time. Frankly, even presented all the information there are few of us qualified to criticise it. However, on the surface it does appear likely that the decision was based on cost and time rather than best engineering practice. Unfortunately, I suspect these kind of shortcuts happen all too often - even in expensive and critical projects found in aerospace and other industries.

Lets hope the computer systems in our nuclear power stations aren't cobbled together from old bits of code to meet deadlines. Actually, let's hope they were mechanically engineered to be fail-safe regardless.

Coder: "I need a year to complete the project"
Manager: "We launch in 2 weeks"

Grrrr, I really want to comment here, but can't

Because of the official secrets act, or can't you say?

He could tell you but then he'd have to kill you.

That's a risk I'm willing to take, he can tell us then kill JJ

Altered for clarity

Tell everyone! It'll be like putting the ring video on youtube!

I'm guessing you've seen this, so for those who have not: