Have you ever hurt yourself from your own code?

One example that stands out is a case where we had a heating element to seal plastic bags. It turns out those heating elements get ridiculously hot and only need to touch the bag for a fraction of a second. Naïve me felt that 3-4 seconds was a "reasonable" amount of time for a first attempt...it caught on fire.

I was taking a class in software security and had a homework assignment to demonstrate a buffer overflow. For extra credit, we could have our exploit play music without invoking exec. My approach to this was fairly straightforward. Using ffmpeg I converted the audio to raw 16bit little-endian PCM. Then I appended this data to the end of my exploit. When my exploit runs, it sets some ioctls on /dev/audio then starts copying the data from the end of the file to /dev/audio causing sound to play.

As you might have guessed by reading the article, I had an off-by-one error, causing the music data to be misalligned and loud static to play. Fortunately, my system volume was not particularly high, and I was happy to even get sound to play. I never figured out where I made an off-by-one error, since I had just assumed that the static was caused by an alignment issue that could be "fixed" by adding a 0 byte to the start of the audio data.

One key observation, though, when working with code is that the machine becomes so close to predictable that it's more dangerous than if it were unpredictable. A cordless handheld drill is hardly a tool that most people would call 'hazardous'. You turn the trigger, the thing rotates, grab the chuck and you can likely stall it. But that much energy in something driven by code is extremely dangerous, it's waiting for any of who knows how many conditions to take off, a common one is that putting your hands in/on something that seems to be not moving can 'make' a sensor and the machine will take off.

There are definitely two kinds of safety domains, one applies to manually operated tools where the operator of, say, a knife is responsible for their own safety - if they cut themselves, they're at fault, not the knifemaker. There's another kind of safety engineering where if the machine takes off 'by itself' while an operator is in the way it's the machine designer's fault for not stopping that from happening, the operator can't be expected to know what's going on.

The latter should become more common as computers continue to permeate everything we use - that cordless drill is no longer controlled by the operator with a trigger switch that connects power to a DC motor, it's a brushless motor driver with an analog input sensing the position of the trigger and the EMF of the motor, and the charge state of the batteries, and who knows what else. Bump a sensor and it might take off, you don't know why it does what it does.

I'll continue to make my automated cells as safe as I can, but don't trust code that's not so simple you can be confident of how it works.

Dropped to 10 meters, set the datum and started swimming, going well until I crossed into a negative quadrant. My beautiful track went haywire/dead.

Then I swam around trying to get back "into" the positive quadrant, I couldn't find it and consequently couldn't get back to the datum. When I surfaced I couldn't see the dive boat!

After some time contemplating my life/coding choices for literally jumping in without thorough testing, I heard the Zodiac inflatable buzzing around the rising waves looking for me.

Back on the boat I discovered that I had only coded it to work in the positive axis from the datum. In the other axis the location was undefined.