The thing that really got me into this subject was building my own virtual machine and then a compiler to turn a high level language into machine code.

It sounds complicated but the VM, in Python, is basically:

  r = {}
  mem = []
  boot(mem)
  pc = 0
  while True:
    op, args = decode(mem[pc++])
    if op == “+”:
      d, a, b = args
      r[d] = r[a] + b
    elif ...:
      ...

The “source” which gets “compiled” can also just be Python function calls that emit op codes — no need to get bogged down with lexing and parsing!

The thing you focus on is the semantics of representing a function call (and return) in machine code and how to manage the stack. It made C make a lot more sense e.g. defining locals up front so the compiler knows how much stack space to use. I wish I could remember more — this was all from teaching A-Level Computer Science a few years ago.