Too true.
I still remember when java5 came out, many new features, great potential for a massive refactoring of my interactive climate model. Within that, I had an idea called "parallel worlds" for comparing scenarios, whereby for efficiency data was shared for parts of the system, and split across parts that varied as user adjusted parameters. So I pulled apart the whole codebase, and joined it back together again... - about two years later, by which time colleagues had given up interest.
[ story simplified to relate to point of OP - not only task in two years! ].
Now I develop a derivative climate system model in scala, but evidently it's more interesting to develop some new complex part of the science code, than fix a graphical interface for beginners. But moods vary - some days lacking energy for refactoring, could be satisfied ticking off a few small tasks in a todo list. Yet after some time, brain craving for another big new complex idea...