Imagine this. I'm using Cool-Lang version 2. Now Cool-Lang 3 comes out. It has lots of features that I want to use. If the new version is compatible with the old one, I download it and start using it. If not, I have to convert my whole codebase first. In practice, that means it may be many years before I can use version 3. Look at Python. In 3.0, they decided to clean up some stuff from 2.x. They even wrote a tool 2to3 which will automatically fix most of the incompatible stuff for you, so converting the whole codebase is almost automatic. The problem is "almost" automatic is not enough. Many people and groups have yet to convert their old code.

Py2 to py3 is too far a jump which is made obvious from the fork-like nature of the project. Minor changes like API deprecation is of course already taking place in most projects. Just need to strike a balance.

I rewrote 100klines at least from C#1 to C#2 when generics came in 2.0. It was compatible in that the non generic code still worked, but I don't really see what difference that made. If we didn't want to upgrade the cod me then we wouldn't have updated the compiler!

Obviously if you don't want the benefits of Py3 then you don't have to pay the upgrade cost either.