Build a stack machine

The loop every chip runs: read the next instruction, work out what it wants, do it, go back for the next. This lesson writes that same loop as a few dozen lines of Go, and then makes it run real programs.

Before you start. A chip runs one loop forever: fetch the next instruction, decode which operation it is, execute it, with one number (the program counter) saying where the next instruction sits, and a jump being nothing more than writing a new number there (The CPU (fetch-decode-execute, the ISA, assembly)). A compiler turns the words you write into those instructions, some languages stop halfway at a portable in-between code that a small program reads, and a function call leaves a note saying where to come back to (How Code Becomes Execution (compilers, stack & heap)). You need Go installed to run the code; reading it needs nothing.

The idea in 30 seconds

A virtual machine is an ordinary program that pretends to be a computer: a loop that reads one instruction, picks what it means, does it, and moves on.
A stack machine keeps its working numbers on one pile: each instruction takes its inputs from the top and puts its answer back on top, so no instruction needs to say where its numbers live.
Bytecode is a program stored as a list of small numbers for a virtual machine to read. Python, Java and WebAssembly all run bytecode.
An assembler turns readable instruction words into those numbers, and fills in each named line’s address so that inserting a line never breaks a jump.
One instruction that jumps only when the top number is zero is enough to build every loop and every if. A second stack of come-back addresses lets the program jump into a piece of code and return to where it was, even when that code calls itself.
Running a program through a virtual machine costs speed: measured on one laptop, a loop ran about 90 times slower that way than when the same loop was translated ahead of time into the chip’s own instructions.

Read the story · Skip to the numbers