Skip to main content

The Tawla programming language and its tawlac compiler

Project description

Tawla

Tawla is a small programming language with its own compiler, tawlac, built from scratch. It looks a lot like C#: you write classes with fields and methods, you get inheritance and interfaces, and everything is statically typed. Under the hood tawlac turns your code into real machine code using LLVM and runs it on the spot, so there's no separate "compile then run" dance.

It started as a learning project to understand how compilers actually work, and it grew into a genuinely usable little language. Source files end in .twl.

Documentation: https://haddajidev.github.io/tawla-lang-docs/

Here's the whole "hello world":

class Main {
    void main() {
        print("Hello, Tawla!");
    }
}
$ tawlac run hello.twl
Hello, Tawla!

Getting it running

You need Python 3.11 or newer. Install it from PyPI with pip:

pip install tawla

That pulls in everything it needs (including LLVM, via llvmlite) and gives you the tawlac command. Works the same on Windows, macOS, and Linux.

No Python? Download a standalone binary

Each release also ships a self-contained tawlac that bundles everything it needs — no Python install required. Grab the one for your OS from the Releases page and put it on your PATH:

  • Windows: download tawlac-windows.exe, rename it to tawlac.exe, move it into a folder such as C:\tools\tawla, then add that folder to your Path environment variable (Settings → "Edit the system environment variables" → Environment Variables → edit Path). Open a new terminal and run tawlac.
  • macOS: download tawlac-macos, then:
    mv tawlac-macos tawlac
    chmod +x tawlac
    sudo mv tawlac /usr/local/bin/
    
    First launch may be blocked because the binary is unsigned — allow it under System Settings → Privacy & Security, or run xattr -d com.apple.quarantine /usr/local/bin/tawlac.
  • Linux: download tawlac-linux, then:
    mv tawlac-linux tawlac
    chmod +x tawlac
    sudo mv tawlac /usr/local/bin/
    

Then tawlac run app.twl works from anywhere, exactly like the pip version.

Hacking on it from a clone of this repo instead? That works too:

pip install llvmlite
python -m tawla run examples/hello.twl

Either way you get the tawlac command (or python -m tawla) with a few subcommands:

tawlac run app.twl     # compile and run a file
tawlac new myapp       # scaffold a new project (like cargo new)
tawlac init            # scaffold one in the current folder
tawlac run             # run the current project (reads Tawla.toml)
tawlac version         # what version is this
tawlac help            # or: tawlac help run

What the language can do

  • The basics: int, float/double, bool, and string; arithmetic and comparisons; if/else, while, for, and functions (recursion works fine).
  • Numbers: integer math with int, and 64-bit floating point with float (or double — same thing). Ints widen to floats automatically when you mix them, so 7.0 / 2 is 3.5 while 7 / 2 stays 3.
  • for loops: the C-style for (int i = 0; i < n; i++) { ... }, with the loop variable scoped to the loop.
  • Increment/decrement: i++, ++i, i--, --i as shorthand for i = i + 1 / i = i - 1 (statement form — works on variables, fields, and array elements).
  • Logical operators: &&, ||, and ! on bools, with short-circuit evaluation (so u != null && u.alive() is safe).
  • Ternary: cond ? a : b picks a value inline (lazy — only the taken branch runs).
  • var: skip the type and let it be inferred — var x = 5;.
  • null & defaults: reference types (objects, strings, arrays) can be null, and a declaration can skip the initializer — int x; is 0, bool b; is false, User u; is null. Using a null (calling a method, reading a field, indexing) gives a clean "null reference" error instead of a crash. Value types (int, float, bool) can't be null.
  • Classes: fields, methods, constructors, this, and new. Objects live on the heap.
  • Inheritance: class Dog : Animal { ... }, with method overriding and super(...) to call the base constructor.
  • Encapsulation: members are private by default; mark them public to expose them or protected to share with subclasses. Constructors are public by default. Access is checked at compile time. (Heads up: code written for Tawla 0.x needs public added to anything used across class boundaries.)
  • Polymorphism: methods are virtual, so the right one gets picked at runtime based on the actual object type (this is done with vtables).
  • Interfaces: interface Shape { int area(); }, and any class can implement it, even classes that share no common parent.
  • Abstract classes: mark a class abstract and leave some methods without a body for subclasses to fill in.
  • Generics: class Box<T> { ... }, used as Box<int> or Box<string>.
  • Strings: literals with escapes, s.length, ==, and + to join them, plus charAt(s, i) (character code), substring(s, a, b), toInt(s) / toFloat(s), and toString(n) (number to string).
  • Arrays: new int[n], indexing with a[i] (bounds-checked, so you get a clear error instead of a crash), and a.length.
  • Comments: // like this.
  • Imports: split code across files with import "other.twl"; — the path is relative to the file importing it, and the .twl is optional.
  • IO library: import "IO.twl"; (a small module that ships with the compiler) gives you readLine(), readInt(), readFloat(), and write(s) (print with no trailing newline) for reading input and prompting.
  • Collections: import "Collections.twl"; gives you a growable List<T> (add, get, set, size) and a Map<K,V> (put, get, has, size). Map.get returns null for a missing object value. (No nested generics yet, e.g. Map<string, List<int>>.)
  • JSON: import "Json.twl"; then parseJson(text) returns a Json value you navigate with get/at/size and read with asInt/asFloat/asBool/asString (plus isNull/isArray/... checks). Build with jsonObject()/jsonArray() + setString/setInt/set/push… and toString(); in a handler, req.respondJson(200, out.toString()) sends application/json. The whole parser/serializer is written in Tawla itself.
  • panic(s): print a message and abort, for unrecoverable errors.
  • Exceptions: fuck_around { ... } find_out (e) { ... } is try/catch — e is the error message string. throw "msg"; raises one, and built-in errors (panic, null dereference, array-out-of-bounds) are catchable too. Use bare find_out { ... } to ignore the message.
  • HTTP server: import "Http.twl"; gives you a Server, a Request, and an Express-style Router with Handler classes. Routes take path params — router.get("/users/:id", new GetUser()) — and inside a handler req.param("id"), req.query("page"), and req.header("Authorization") read the path param, query string, and request header (each null when absent). req.method()/ path()/body()/respond()/respondJson() round it out. Single-threaded, minimal HTTP/1.1.
  • HTTP client (fetch): fetch(url) (GET) or httpRequest(method, url, body) returns a Response with status() and body() — call other services. Network failures come back as status() == 0.
  • SQLite: import "Sql.twl"; gives you Db, prepared Stmts, and a Rows cursor — Db db = new Db("app.db"); Stmt q = db.prepare("SELECT name FROM users WHERE age > ?"); q.bindInt(0, 18); Rows r = q.query(); then r.next() / r.getString("name"). Parameters bind by index (injection-safe); SQL errors throw (catch with fuck_around/find_out).
  • Built-in functions: a handful of predefined functions you can call without declaring anything — sqrt, pow, abs, min, max, floor, ceil for math, plus collect()/__live() for the GC.
  • Garbage collection: you don't free memory by hand. Call collect() when you want a cleanup pass; __live() tells you how many objects are still around.

There's a runnable example for pretty much every feature in examples/ — poke around in there to see how things look in practice.

How it works, roughly

tawlac runs your code through the usual compiler stages:

source.twl -> lexer -> parser -> sema -> codegen -> LLVM -> JIT -> runs
  • the lexer chops the text into tokens
  • the parser builds a tree out of those tokens
  • sema checks the types and catches mistakes before any code is generated
  • codegen turns the checked tree into LLVM instructions
  • LLVM compiles that to machine code and the JIT runs it immediately

Each piece lives in its own file under tawla/, so it's not hard to follow if you want to read along.

Running the tests

./venv/Scripts/python -m pytest

There are around 200 tests. Programs that print output are checked by running tawlac as a separate process and looking at what it prints (this sidesteps a Windows quirk where output from JIT-compiled code is hard to capture in-process).

What's not done yet

It's a real language, but it's still a young one. A few honest gaps:

  • tawlac build (making a standalone .exe from your Tawla program) isn't done — for now everything runs through tawlac run.
  • Generics only cover classes, not standalone functions, and you can't nest them like Box<Box<int>>.
  • Garbage collection has to be triggered with collect(); it doesn't kick in on its own yet.

The full design and the step-by-step history of how it was built are in docs/superpowers/specs/2026-05-29-tawla-language-design.md if you're curious.

License

MIT — see LICENSE.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

tawla-1.7.0.tar.gz (82.9 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

tawla-1.7.0-py3-none-any.whl (59.6 kB view details)

Uploaded Python 3

File details

Details for the file tawla-1.7.0.tar.gz.

File metadata

  • Download URL: tawla-1.7.0.tar.gz
  • Upload date:
  • Size: 82.9 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.13.5

File hashes

Hashes for tawla-1.7.0.tar.gz
Algorithm Hash digest
SHA256 4258ba127700b47845ed29c89564c29f8916d02eede1a78a87941affef47d2ec
MD5 0f3eb08c237196c392ba7bc3df9e0bc4
BLAKE2b-256 1bef8189b38bdb1b103bc9ee81dac5a7daef522090c6307200fdaa71d57153cc

See more details on using hashes here.

File details

Details for the file tawla-1.7.0-py3-none-any.whl.

File metadata

  • Download URL: tawla-1.7.0-py3-none-any.whl
  • Upload date:
  • Size: 59.6 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.13.5

File hashes

Hashes for tawla-1.7.0-py3-none-any.whl
Algorithm Hash digest
SHA256 69474869f029b5bd4daabc3c9e960e2a6a0a2258f224e8ec716f8ff23abfaf15
MD5 712c7336f48a67cc140cb0f98c43b782
BLAKE2b-256 0e3b1ebcd889f91268e72019bbd955814b9d1f4b36be36da53d98d47c3295456

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page