Writing a tiny language to see what a compiler actually does

bhailang went around my feed — a toy language with Hindi slang for keywords. Everyone shared the joke; I got stuck on the machinery. I use a compiler or interpreter every day and had never once watched one turn text into behaviour. The fastest way I know to understand something is to build a bad version of it myself.

consider declares a variable, print prints one, and expressions can use + - * /. That’s the whole language. The compiler reads a program from a file:

consider x = 20
consider y = 20

consider sum = x + y
print sum

1. 1
   Lexer. Walks the source one character at a time with a cursor. Whitespace is skipped; runs of letters become a keyword (if they’re consider or print) or an identifier; runs of digits become numbers; + - * / = become operators. Anything else throws “Invalid character” — the language’s only error message.
2. 2
   Parser. Consumes tokens and builds a tree: a Program whose body holds Declaration and Print nodes. A declaration takes the next token as its name and, after =, everything up to the next keyword as its value.
3. 3
   Code generator. Walks the tree and emits JavaScript: a declaration becomes let name = value; and a print becomes console.log(name).
4. 4
   Runner. Reads code.txt with Node’s fs, compiles it, and evaluates the generated JavaScript — with any compile or runtime error caught and reported.

let x = 20;
let y = 20;
let sum = x+y;
console.log(sum)

Building it small meant taking shortcuts, and each one turned out to be a signpost to a real compiler problem:

1. 1
   Expressions are strings, not trees. The parser glues the tokens after = back into text. Operator precedence works only because JavaScript’s does — my language never actually understands x + y * 2.
2. 2
   It’s a transpiler, not an interpreter. Nothing in my code evaluates anything; it hands JavaScript to eval. That’s why it was quick to build — and why it can run anything you smuggle into an expression.
3. 3
   Errors know nothing about position. “Invalid character” with no line or column is technically correct and practically useless, which taught me why good compiler errors are their own discipline.

1. 1
   A real expression parser (precedence climbing or Pratt parsing) so + and * mean what the language says, not what JavaScript says.
2. 2
   A tree-walking evaluator to replace eval — the step that turns a transpiler into an actual interpreter.
3. 3
   Conditionals, loops and error messages with line and column numbers.