v2.29.4 · 819 tests passing

Watch a programming language work.

A small, hand-rolled language — tokenize, parse, execute, compile to bytecode — with every step animated in front of you. ~25,000 lines, no exec(), no compile-to-Python.

Try it yourself How does it work?

GitHub·Docs·Internals

Live demostep 1/14

Source


1i = 1
2while (i <= 3) {
3    if (i == 3) {
4        coutln("Fizz")
5    } else {
6        coutln(i)
7    }
8    i = i + 1
9}

Output

(waiting…)

Readtokens

i=1

Parseast node

Assign(i, 1)

Runeffect

i←1

Version

v2.29.4

Lines of code

~25,000

Tests passing

819

Opcodes (M2 VM)

The pipeline

Six stages, from raw characters to printed output. The playground animates every one of them.

Open the playground →

Source

Characters in a file — the program before anything is touched.

coutln("hi")

Tokens

A char-by-char lexer turns the source into typed tokens with line + col.

coutln · ( · "hi" · )

AST

Recursive-descent for statements; Pratt for expressions. Every node carries source coordinates.

Call(coutln, ["hi"])

Snapshots

The tree-walking interpreter executes statements; each one captures an env snapshot.

{ i: 1, n: 3 }

Bytecode

A separate compiler lowers the AST to 38 opcodes that run on a stack VM.

LOAD_CONST 1 · STORE_NAME i

Output

stdout, captured and streamed back to the playground.

Watch each stage

Self-contained loops, one per major pipeline stage. The full playground threads them together.

Read the internals docs →

Lexerlive

characters

coutln("hi")

tokens

A char-by-char tokenizer. Groups characters into typed tokens that carry line + column.

Parserlive

tokens

1+2*3

ast

(waiting…)

Recursive descent for statements, Pratt for expressions. Operator precedence falls out naturally.

VMlive

bytecode

0 LOAD_CONST 42
1 STORE_NAME i
2 LOAD_NAME i
3 RETURN

stack

empty

env

(empty)

Bytecode runs on a stack machine — same shape as CPython, Lua, or the JVM, just smaller.

What's inside

The whole language lives in six files. Each one is small enough to read in a sitting.

Read the design retrospective (PDF)

lexer.py~320 LOC

Hand-rolled, char-by-char tokenizer. Emits Tokens with line + col.

View on GitHub

syntax.py~870 LOC

Recursive-descent statements; Pratt expressions. Builds the AST.

View on GitHub

interpreter.py~1,300 LOC

Tree-walking evaluator. Snapshot-per-statement for the playground's step mode.

View on GitHub

codegen.py~690 LOC

AST → bytecode chunks. The M2 compiler. Per-instruction line attribution.

View on GitHub

vm.py~780 LOC

Stack-based bytecode VM. Frames, handlers, 38 opcodes.

View on GitHub

errors.py~90 LOC

RotError + rustc-style rendering — source line, caret, hints.

View on GitHub

Demos

Open a sample in the playground and step through it statement by statement.

FizzBuzz

Loops, if/elseif/else, and the classic interview prompt.

Factorial

Recursion + return value. Watch the call frame stack grow.

Counter

Classes, init, methods, and `this`. Object-oriented in ROT.

Sum List

Lists + for-in + compound assignment.