I used to own a C++ application that was a morass of abstractions and indirections so it was impossible to reason about. It took a number of hours to compile.

I used to own a C++ application that was a morass of abstractions and indirections so it was impossible to reason about. It took a number of hours to compile. On one infamous occasion we were making a, relatively small, patch release. The debug version worked fine but the release version crashed systematically. Even when we backed out all the changes we had the same behaviour. We were screwed. Until one of the team had a bright idea. She stripped strings from the debug build and tested it. To our surprise it not only worked, it was only slightly bigger than the previous release version and it was also slightly faster! We shipped. This experience was the trigger to make me go all-in on a full re-write that I had been contemplating. One of only a couple of times in my career that I've made that decision on a major piece of software. The re-write was a huge success. It was also about 10% of the original in terms of LoC. The day our testing finished, we held a ceremony where we deleted all the old code from the current version. This caused a slightly different issue. At the time, code metrics were starting to get fashionable but LoC wasn't yet the pariah it became. So, a couple of days later I got a concerned call from the metrics guys. Apparently, we had deleted more code than all the other teams combined had added in the previous measurement period. This caused their metric calculation to barf. Their solution? We should add all the code back in! This led to a somewhat heated argument that ended up with me persuading them that deleting code was good and they should, at least, abs(LoC) it. It didn't make the metrics any more useful but meant that we had an application we could reason about. Happy days.