I second this based on my experience. My masters thesis was in code generation for Haskell where in I spent fair bit of time learning tree pattern matching, AST

I second this based on my experience. My masters thesis was in code generation for Haskell where in I spent fair bit of time learning tree pattern matching, AST manipulation, register allocation etc. One can achieve a non trivial speed up by understanding these different phases of code generation. Another reason, in the context of compilers, is that there’s not much to be gained by micro-optimizing lexing and parsing. After all, a code is compiled just once and executed multiple times. So you know where much of the fun lies. The undergrads unfortunately usually miss out on the meatier part of the compiler course.