It depends on your source and target. The two parts are practically different disciplines. If you target llvm or naive assembly as output, most of your work wil

It depends on your source and target. The two parts are practically different disciplines. If you target llvm or naive assembly as output, most of your work will be in translating semantics, and depending on the language complexity, there can be a lot to implement. Or not that much, in simple languages like C. The back end is mostly about transforms that preserve meaning, with a long tail of attempts to prove things about the code to make things faster in certain cases. You can also have languages with complex front ends that need lots of work, whether it's preserving runtime type information (a big serialisation job that's not strictly back end work), interesting type systems that also are trying to prove things about the code, features that require increasingly elaborate transformations before they can be handed off to any common back end (like closures, iterators, async, continuation passing style).