Dust

Dust is a general purpose programming language that emphasises concurrency and correctness.

A basic dust program:

(output  "Hello world!")

Dust can do two (or more) things at the same time with effortless concurrency:

async {
    (output 'will this one finish first?')
    (output 'or will this one?')
}

You can make any block, i.e. {}, run its statements in parallel by changing it to async {}.

if (random_boolean) {
    (output "Do something...")
} else async {
    (output "Do something else instead...")
    (output "And another thing at the same time...")
}

Dust is an interpreted, strictly typed language with first class functions. It emphasises concurrency by allowing any group of statements to be executed in parallel. Dust includes built-in tooling to import and export data in a variety of formats, including JSON, TOML, YAML and CSV.

• Dust
  • Features
  • Usage
  • Installation
  • Benchmarks
  • Implementation
  • The Dust Programming Language
    • Declaring Variables
    • Lists
    • Maps
    • Loops
    • Functions
    • Concurrency
    • Acknowledgements

Features

• Simplicity: Dust is designed to be easy to learn.
• Speed: Dust is built on Tree Sitter and Rust to prioritize performance and correctness. See [Benchmarks] below.
• Concurrency: Safe, effortless parallel code using thread pools.
• Safety: Written in safe, stable Rust.
• Correctness: Type checking makes it easy to write good code.

Usage

Dust is an experimental project under active development. At this stage, features come and go and the API is always changing. It should not be considered for serious use yet.

cargo install dust-lang
dust -c "(output 'Hello world!')"

Installation

You must have the default rust toolchain installed and up-to-date. Install rustup if it is not already installed. Run cargo install dust-lang then run dust to start the interactive shell. Use dust --help to see the full command line options.

To build from source, clone the repository and build the parser. To do so, enter the tree-sitter-dust directory and run tree-sitter-generate. In the project root, run cargo run to start the shell. To see other command line options, use cargo run -- --help.

Benchmarks

Dust is at a very early development stage but performs strongly in preliminary benchmarks. The examples given were tested using Hyperfine on a single-core cloud instance with 1024 MB RAM. Each test was run 1000 times. The test script is shown below. Each test asks the program to read a JSON file and count the objects. Dust is a command line shell, programming language and data manipulation tool so three appropriate targets were chosen for comparison: nushell, NodeJS and jq. The programs produced identical output with the exception that NodeJS printed in color.

For the first test, a file with four entries was used.

Command	Mean [ms]	Min [ms]	Max [ms]
Dust	3.1 ± 0.5	2.4	8.4
jq	33.7 ± 2.2	30.0	61.8
NodeJS	226.4 ± 13.1	197.6	346.2
Nushell	51.6 ± 3.7	45.4	104.3

The second set of data is from the GitHub API, it consists of 100 commits from the jq GitHub repo.

Command	Mean [ms]	Min [ms]	Max [ms]
Dust	6.8 ± 0.6	5.7	12.0
jq	43.3 ± 3.6	37.6	81.6
NodeJS	224.9 ± 12.3	194.8	298.5
Nushell	59.2 ± 5.7	49.7	125.0

This data came from CERN, it is a massive file of 100,000 entries.

Command	Mean [ms]	Min [ms]	Max [ms]
Dust	1080.8 ± 38.7	975.3	1326.6
jq	1305.3 ± 64.3	1159.7	1925.1
NodeJS	1850.5 ± 72.5	1641.9	2395.1
Nushell	1850.5 ± 86.2	1625.5	2400.7

The tests were run after 5 warmup runs and the cache was cleared before each run.

hyperfine \
	--shell none \
	--warmup 5 \
	--prepare "rm -rf /root/.cache" \
	--runs 1000 \
	--parameter-list data_path seaCreatures.json,jq_data.json,dielectron.json \
	--export-markdown test_output.md \
	"dust -c '(length (from_json input))' -p {data_path}" \
	"jq 'length' {data_path}" \
	"node --eval \"require('node:fs').readFile('{data_path}',(err,data)=>{console.log(JSON.parse(data).length)})\"" \
	"nu -c 'open {data_path} | length'"

Implementation

Dust is formally defined as a Tree Sitter grammar in the tree-sitter-dust directory. Tree sitter generates a parser, written in C, from a set of rules defined in JavaScript. Dust itself is a rust binary that calls the C parser using FFI.

Tests are written in three places: in the Rust library, in Dust as examples and in the Tree Sitter test format. Generally, features are added by implementing and testing the syntax in the tree-sitter-dust repository, then writing library tests to evaluate the new syntax. Implementation tests run the Dust files in the "examples" directory and should be used to demonstrate and verify that features work together.

Tree Sitter generates a concrete syntax tree, which Dust traverses to create an abstract syntax tree that can run the Dust code. The CST generation is an extra step but it allows easy testing of the parser, defining the language in one file and makes the syntax easy to modify and expand. Because it uses Tree Sitter, developer-friendly features like syntax highlighting and code navigation are already available in any text editor that supports Tree Sitter.

The Dust Programming Language

Dust is easy to learn. Aside from this guide, the best way to learn Dust is to read the examples and tests to get a better idea of what it can do.

Declaring Variables

Variables have two parts: a key and a value. The key is always a string. The value can be any of the following data types:

• string
• integer
• float
• boolean
• list
• map
• option
• function

Here are some examples of variables in dust.

string = "foobar"
integer = 42
float = 42.42
list = [1 2 string integer float] # Commas are optional when writing lists.
map = {
    key = 'value'
}

Note that strings can be wrapped with any kind of quote: single, double or backticks. Numbers are always integers by default. Floats are declared by adding a decimal. If you divide integers or do any kind of math with a float, you will create a float value.

Dust enforces strict type checking, but you don't usually need to write the type, dust can figure it out on its own. The number and any types are special types that allow you to relax the type bounds.

string <string> = "foobar"
integer <int> = 42
float <float> = 42.42

numbers <[number]> = [integer float]

stuff <[any]> = [string integer float]

Lists

Lists are sequential collections. They can be built by grouping values with square brackets. Commas are optional. Values can be indexed by their position using a colon : followed by an integer. Dust lists are zero-indexed.

list = [true 41 "Ok"]

(assert_equal list:0 true)

the_answer = list:1 + 1

(assert_equal the_answer, 42) # You can also use commas when passing values to
                              # a function. 

Maps

Maps are flexible collections with arbitrary key-value pairs, similar to JSON objects. A map is created with a pair of curly braces and its entries are variables declared inside those braces. Map contents can be accessed using a colon :.

reminder = {
    message = "Buy milk"
    tags = ["groceries", "home"]
}

(output reminder:message)

Loops

A while loop continues until a predicate is false.

i = 0
while i < 10 {
    (output i)
    i += 1
}

A for loop operates on a list without mutating it or the items inside. It does not return a value.

list = [ 1, 2, 3 ]

for number in list {
    (output number + 1)
}

Functions

Functions are first-class values in dust, so they are assigned to variables like any other value.

# This simple function has no arguments and no return value.
say_hi = (fn) {
    (output "hi")
}

# This function has one argument and will return a value.
add_one = (fn number <num>) <num> {
    number + 1
}

(say_hi)
(assert_equal (add_one 3), 4)

You don't need commas when listing arguments and you don't need to add whitespace inside the function body but doing so may make your code easier to read.

Option

The option type represents a value that may not be present. It has two variants: some and none. Dust includes built-in functions to work with option values: is_none, is_some and either_or.

say_something = (fn message <option(str)>) <str> {
    (either_or message, "hiya")
}

(say_something some("goodbye"))
# goodbye

(say_something none)
# hiya

Concurrency

Dust features effortless concurrency anywhere in your code. Any block of code can be made to run its contents asynchronously. Dust's concurrency is written in safe Rust and uses a thread pool whose size depends on the number of cores available.

# An async block will run each statement in its own thread.
async {
    (output (random_integer))
    (output (random_float))
    (output (random_boolean))
}

data = async {
    (output "Reading a file...")
    (read "examples/assets/faithful.csv")
}

Acknowledgements

Dust began as a fork of evalexpr. Some of the original code is still in place but the project has dramatically changed and no longer uses any of its parsing or interpreting.