Howdy, everyone, thanks for the meetup organizing last night. The thing that I'd mentioned for using property-based testing to expand code coverage wasn't quite right. The links that I can easily find actually discuss another target of testing, but the technique is the same. I just can't recall offhand where "your utility function is # lines of code coverage" came from, sorry! I'll post a follow-up if I can find it.
Kostis Sagonas had a conference presentation a few months ago that discusses TPBT, targeted property-based testing. When testing radio networks with simulated mesh topologies, using a random generator was ... too random. "Interesting graphs" were almost never generated. Kostis's research group created a way to automagically drive random data generators to explore more interesting graphs. They defined a utility function: maximum distance between two nodes in the radio mesh graph. Then they defined the data generator to maximize that distance ... and the new generator reliably generates high-max-distance graphs.
-Scott