[throwawaymaths]

I don't think this is the root cause. Declarative languages really paper over details in many cases, and the mapping from the declaration to the imperative execution might not be obvious. When you're on the happy path, it's amazing (if it's a good DSL: because you can be confident you haven't forgotten details, that it's systematically correct, etc). When you encounter a regression, bug, or misunderstanding, in anger, it can be very very difficult to figure out what's going on.