[yegortimoshenko]

I'd argue that most breaking changes are not about fixing problems. They are about API consistency and taste, like Babel 6 mentioned above.

If problem is a bug that people might rely upon, I don't mind fixing it. If problem is fundamental design-wise, however, the right way is to create a new library.

There are a lot of designs that never break anything (or at least try hard not to): Linux, macOS, Erlang, Java, Clojure, x86 instruction set, life on earth, and they all improve over time. And they obviously didn't get everything right the first time.