Table of contents

• Introduction
• Language Tour
• Usage
• Installation-and-build
  • Venv install
  • Global install
  • Build from source
    • Testing
• Benchmark
• TODO

Uza

Uza is a small statically typed programming language. This repo features an uza compiler/interpreter written in Python in the uzac directory. It also features a garbage collected virtual machine, based on the second part of Bob Nystrom's Crafting Interpreters, in the vm directory.

Here is an example of an uza program:

func fib(n : int) => int {
    if n <= 1 then return n
    return fib(n-1) + fib(n-2)
}

const n = 30
println("The 30th fibonacci number is " + fib(30).toString())

Language Tour

Variables

Variables use the var keyword and take an optional type:

var count: int = 42
var str = "hello world" // 'string' type inferred

Constants are declared using const and cannot be reassigned:

const G = 10.0

G = 9.81 // FAILS: UzaTypeError: cannot reassign const variable 'G'

Blocks

Blocks statements allow for creation of a new scope.

var i = 0
{
  var i = 42
  const foo = "unused"
  println(i) // 42
}
println(i) // 0
println(foo) // UzaNameError: variable `foo` not defined in this scope

Functions

Functions are defined using func and must be typed:

func add(a: int, b: int) => int {
    return a + b
}

Functions that have a void return type always return nil:

func printMessage() => void {
    print("Hello, World!")
}

print(printMessage)//Hello, World!nil

Lists

Lists are dynamic arrays with a generic type. A List instead is constructed with List<type>().

var numbers: List<int> = List<int>() // var numbers = List<int>() # to infer type
numbers.append(1)
numbers.append(42)

println(numbers.len()) // 2
println(numbers) // [1, 42]

numbers.set(0, 97)
println(numbers) // [97, 42]

const DESCENDING = false
numbers.sort(DESCENDING)
println(numbers) // [42, 97]

---

Control Flow

Conditionals

const x = 15
if x > 10 {
  println("Large") // Large
}
else {
  println("Small")
}

println(2 > 1 and true) // true

println(not (false or true)) // false

If statements can also take a single expression/statement instead of a block by using the then keyword:

const x = 7

if x > 10 then println("Large")
else if x > 5 then println("Medium") // Medium
else println("Small")

// identical to:
if x > 10 then
  println("Large")
else if x > 5 then
  println("Medium") // Medium
else
  println("Small")

Loops

For loops are take in three optional statements:

• A initializer, that is run before the first iteration.
• A conditional, that is checked every iteration.
• And a statement that is run after each iteration.

for var i = 0; i < 3; i += 1 {
    println(i)
}

While loops only take a conditional that evaluates to a boolean:

var count = 0
while count < 10 {
  count += 1
}

Like if statments, while and for loops can take in a block, or a single expression or statment using the do keyword:

println("pair numbers in [0, 10]: ")
for var i = 0; i <= 10; i += 2 do println(i)

break and continue

A break statement exits the current loop:

for var i = 0; i < 1_000_000_000; i += 1 {
  if i > 0 then break
  println(i)
}
// prints:
// 0

The continue statement skips the current loop iteration:

println("pair numbers in [0, 10]: ")
for var i = 0; i <= 10; i += 1 {
  if i % 2 != 0 then continue
  println(i)
}

Unit Conversions

toInt and toFloat

const i: int = 1
const f: float = i + 0.5 // implicit conversion of i to float
println(f) // 1.5


const str = "1.6"
const bigger = str.toFloat()
println(bigger > f) // true

//truncate int
println(bigger.toInt() < f) // true

toString

const number = 42
println(number.toString() + "24") // 4224

Scientific notation

const str = "1.5E9"
const scientific = str.toFloat()
println(scientific)                          // 1500000000.000
println(scientific.toInt() == 1_500_000_000) // true

String Handling

String Concatenation

var message = "Hello, " + "World!"

Substrings

const str = "aldskfjldjaflj Hello, world!"

var i = 0
while str.get(i) != " " do i += 1

println(str.substring(i + 1, str.len())) // Hello world!

f-strings

String interpolations, with a syntax similar to Python:

const name = "Jane"
const age = 33
println(f"{name} is {age.toString()} years old.")

Utils

Other useful functions:

const foo: float = abs(3.14)

print("hi")
flush() // flush stdout

const t: int = timeNs() // nanoseconds counter

const ms: int = timeMs() // milliseconds counter

const n: int = randInt(1000) // random value between 0 and N (excluded)

sleep(1000) // sleep thread for N ms

---

conway.s.game.of.life.webm

Output of examples/game_of_life.uza, an implementation of Conway's Game of Life. More examples are available in the examples directory.

Usage

Compile and execute.

uza source.uza

Compile and execute from stdin.

echo 'println("hello world!")' | uza

Compile to bytecode. Execute bytecode without compilation step.

uza code.uza -c // Wrote X bytes to code.uzb
uza code.uzb

Interpret without VM installed (slow execution)

uza source.uza -i

For more options:

uza --help

Installation and Build

Note

Uza is a personal learning project and not meant for production use. Unless you like a challenge :^)

Warning

There are no official prebuilt binaries for Linux and Windows on ARM64 or ARMv7. You will need to build from source for these architectures.

The main way to install uza is through pip, Python's package manager. Installing in a venv removes the need to edit the PATH but requires the venv to be active to use uza.

venv install

The venv environement has to be active to run uza.

UNIX shell

python3 -m venv venv
source ./venv/bin/activate
pip install uza
uza --help

Powershell

python3 -m venv venv
venv\Scripts\activate
pip install uza
uza --help

global pip install

UNIX

pip install uza
uza --help

Windows

pip install uza --force-reinstall --user
uza --help

You might get one of the following warnings when installing uza globally:

WARNING: The script uza is installed in '/opt/homebrew/Cellar/pypy3.10/7.3.17_1/libexec/bin' which is not on PATH.
Consider adding this directory to PATH or, if you prefer to suppress this warning, use --no-warn-script-location.
NOTE: The current PATH contains path(s) starting with `~`, which may not be expanded by all applications.

------------------

WARNING: The script uza is installed in '/home/smith/.local/bin' which is not on PATH.
Consider adding this directory to PATH or, if you prefer to suppress this warning, use --no-warn-script-location.

------------------

But still not in path so you have to   WARNING: The script uza.exe is installed in 'C:\Users\doe\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.13_qbz5n2kfra8p0\LocalCache\local-packages\Python313\Scripts' which is not on PATH.
  Consider adding this directory to PATH or, if you prefer to suppress this warning, use --no-warn-script-location.

To use without adding to path:

python3 -m uzac --help

Build from source

Note

The python compiler also has a tree-walk interpreter, using the -i flag, for which there is no need to compile the VM. To interpret uza in python, prefer using a python interpreter with a JIT such as pypy3 for much better performance.

git clone git@github.com:msanlop/uza.git
cd uza
mkdir build && cd build
cmake ..
make

You can now run uza by executing python uza from root dir of the repo.

Add local installation to path (UNIX)

printf "#!$(which python3)\n$(cat uza)" > uza
export PATH=$(pwd):$PATH
./uza --help

Export in shell config file, e.g. ~/.bashrc, to persist across sessions.

Testing

pip install -r requirements.txt
pytest

Benchmark

This microbenchmark should be taken with a grain of salt. It's not a good representation of the overall speed of each language and interpreter. The only reason I include this is because I like comparing performance and looking at charts :).

This benchmark runs a recursive fibonacci function to compute the 35th fibonacci number. The benchmark code can be found in examples/fibonacci_bench.uza. The following interpreters were tested on a base Apple M4:

• uza (pypy): A tree-walk uza interpreter running on PyPy, a JIT implementation of Python. (Running uza with the -i flag)
• uza: Uza running the bytecode VM in the vm directory
• clox: A Lox bytecode interpreter. The VM implementation is similar to the VM in this repo
• python3.12: The reference Python implementation
• pypy3.10: JIT implementation of Python

The chart above shows just how slow the tree-walk approach is. Running a tree-walk interpreter of uza, inside pypy is orders of magnitude slower than running a C bytecode interpreter, or running a JIT with pypy. Running the same benchmark with a tree-walk interpreter inside of CPython instead of PyPy would probably take minutes!

Taking a closer look at other values, we notice similar performance for the bytecode interpreters. One might expect that uza would perform better than Lox and Python, since it is staically typed, but the current VM implementation is still very close to the clox one. There are still quite a few runtime type checks, which could be avoided by emiting more specialized opcodes. For example, instead of having a single OP_ADD for additions, the compiler could emit OP_IADD, OP_FADD, and OP_STRCONCAT to separately handle integer, float and string additions respectively. Implicit integer conversions when adding integers to floats would also have to be handled at compile-time instead of runtime by emitting OP_ITOF for example — analogous to the JVM's i2f instruction — as the current OP_TOINT does runtime checks for the base type. The compiler also does zero optimisations on the AST, so that's anothing thing to explore.

PyPy's JIT shows an incredible, almost 10x improvement over CPython in this benchmark. Note that in real-world cases, the average speedup is closer to 2.9x. It'll be interesting to see how the new CPython JIT will fare in comparaison to PyPy in the coming versions.

TODO

• Structs
• Closures, lambda functions
• Generics and overloading for user functions
• Iterators
• Maps
• Modules, stdlib
• JIT, add jitting to the VM or try using RPython
• it never ends...