An introduction to the Hare programming language

This tutorial will introduce you to the Hare programming language. It should take an hour or two to complete. If you are already familiar with programming in other languages, it might take even less time — feel free to jump around as you see fit.

After you complete this tutorial, you should move on to the standard library introduction.

Table of contents

How to get help

We want you to leverage the Hare community to maximize your odds of success with Hare. Our community strives to help each other build the best Hare programs possible, together. Please join our mailing lists, chat rooms, and so on, and do not hesitate to ask for help, discuss your problems, and offer your knowledge and insights to others.

Here are some quick links to our community resources:

  • hare-users is a great place to ask questions
  • #hare on irc.libera.chat is a good place for IRC users to idle

Please join us!

Getting started

use fmt;

export fn main() void = {
	fmt::println("Hello world!")!;
};

Start by setting up your local development environment according to the installation instructions. To test that it works, copy this code sample into a file named main.ha, and run the following command:

$ hare run main.ha
Hello world!

You can also compile this program into an executable like so:

$ hare build -o example main.ha
$ ./example
Hello world!

Hare programs use 8-column-wide tab characters for indentation. See the style guide for details, and consider installing an editor plugin.

Breaking down "Hello world!"

use fmt;

export fn main() void = {
	fmt::println("Hello world!")!;
};

Let’s briefly break down what’s happening in this program, then go into more detail throughout the remainder of the tutorial.

The first line is an import which pulls in the fmt module from the standard library — click that link to view the documentation for this module. The purpose of this module is formatting text, and one of the functions it provides is println, which formats some text and prints it to the program’s output along with a new line.

The program’s entry point is “main”, which executes after the program’s environment is initialized. The shape of this function is defined by the Hare standard: it accepts no parameters and returns no values (this is what void means), and it must be “exported” so that the Hare runtime can use it. The only thing that this function does is call the “println” function from the fmt module, printing our “Hello world!” message to the program’s standard output.

Error handling in "Hello world!"

use fmt;

export fn main() void = {
	fmt::println("Hello world!")!;
};

Note the ! operator which follows the function call: this is the error assertion operator. It is possible for writing to the standard output to fail, and the programmer is required to address this scenario. Try removing it and compiling this program again — the compiler will complain. The simplest approach to error handling, though usually incorrect, is to use this operator to promise the compiler that the error will never occur. The compiler will hold you to this: if an error does occur, the program will crash. You can see this in action by simulating an error; on Linux you can do this like so:

$ hare run main.ha >/dev/full

Crashing is generally thought to be more desirable than allowing the program to continue under conditions which were thought to be impossible, as this often leads to security vulnerabilities or other undesirable behavior. We will go over better ways to deal with errors later on in the tutorial, but for now it’s okay to use ! to make the compiler happy.

Next, we’ll dive a bit deeper with a slightly more complex example.

Hare basics

Functions & parameters

use bufio;
use fmt;
use os;
use strings;

export fn main() void = {
	const user = askname();
	greet(user);
};

// Asks the user to provide their name.
fn askname() str = {
	fmt::println("Hello! Please enter your name:")!;
	const name = bufio::read_line(os::stdin)! as []u8;
	return strings::fromutf8(name)!;
};

// Greets a user by name.
fn greet(user: str) void = {
	fmt::printfln("Hello, {}!", user)!;
};

The sample program shown here is a bit more complex, and we’re going to examine it in detail over the next few headings. This program prints a greeting, then prompts the user to enter their name, then greets the user by name. There’s a lot to unpack here, but we’ll start by looking at the use of functions.

A function is the basic unit of executable code in a Hare program, which defines a step or series of steps for the program to perform. Functions can “call” each other to deputize various tasks, such as relying on “fmt” to format text for printing.

Functions may accept a list of parameters by placing them between the () tokens when declaring the function. “greet” accepts a str (string) parameter, which it names “user”. Functions can also return values to the function which calls them, such as “askname”, which returns a str. Our main function makes use of the return value by storing it in a variable and passing it to “greet”.

Note: Many code samples in this tutorial make use of concepts which are introduced later on. If you're unsure about how something works, don't worry — it will be explained later.

A closer look at bufio::read_line

use bufio;
use fmt;
use os;
use strings;

export fn main() void = {
	const user = askname();
	greet(user);
};

// Asks the user to provide their name.
fn askname() str = {
	fmt::println("Hello! Please enter your name:")!;
	const name = bufio::read_line(os::stdin)! as []u8;
	return strings::fromutf8(name)!;
};

// Greets a user by name.
fn greet(user: str) void = {
	fmt::printfln("Hello, {}!", user)!;
};

Like “askname”, many standard library functions can return values. bufio::read_line is an interesting example of this. We can read the documentation for this function in our terminal by running the haredoc bufio::read_line command — try this now.

This function’s signature (the combination of its parameters and return type) is a bit more complex in that it can return one of several kinds, or types, of values, forming what’s referred to as a tagged union. The one we’re interested in is []u8, which is a “slice” of bytes containing the line we want to read. It can also return io::EOF, which indicates that the “end of file” was reached, or io::error, which indicates that an I/O error occurred.

Run the haredoc io::error command, and note the ! error flag which prepends the signature, which signifies that io::error is an error type. Now run the haredoc io::EOF command and note the absence of the ! error flag. Encountering the end of the file is not considered an error. The ! in our program is an error assertion operator, which deals with the error case as we described before, leaving []u8 and io::EOF remaining.

To address this, we use the as operator to interpret the value as if it were a []u8 — this is called a type assertion. Like other assertions, the compiler will check our work and, if proven wrong, will cause the program to crash. You can simulate this case by pressing Ctrl+D in your terminal emulator instead of entering your name.

A key take-away from this should be the “haredoc” tool, which you should use liberally as you work with Hare. Try to use it to learn about the other standard library functions we’re using in this sample, such as strings::fromutf8. You can also use it to browse the modules themselves  — try haredoc fmt or haredoc strings.

Using const & let to define variables

use fmt;
use io;
use os;
use strings;

export fn main() void = {
	// Example A
	const source = os::open("main.ha")!;
	const source = io::drain(source)!;
	const source = strings::fromutf8(source)!;
	const source = strings::split(source, "\n");
	first3(source);

	// Example B
	let i: int = 1337, j: int = 42;
	fmt::printfln("{} + {} = {}", i, j, i + j)!;
	j = i;
	fmt::printfln("{} + {} = {}", i, j, i + j)!;
};

fn first3(lines: []str) void = {
	fmt::println("The first three lines of main.ha are:")!;
	fmt::println(lines[0])!;
	fmt::println(lines[1])!;
	fmt::println(lines[2])!;
};

This sample is designed to illustrate a few common ways to use variables in Hare with the const and let keywords. When declaring a variable, you must provide the initial value with an initializer, such as a constant value like 42 or an arbitrary expression like a function call. A variable declared with const cannot be modified after it is initialized, but you can modify let variables with the = operator.

However, though you cannot modify a const variable, you can re-bind it by creating another variable with the same name. When re-binding a variable, you may also change its type: in this example, “source” refers to an io::file, a []u8, a str, and a []str, in that order. This is a useful pattern for building up a variable from a series of intermediate values of various types. This technique is more generally called “shadowing” the variable.

Take note of the syntax as well. Each of the bindings in Example A use type inference, in which the variable automatically assumes the type of the right-hand side of the statement. The let examples demonstrate the use of an explicit type (int). In some cases it may be necessary or helpful to state the type explicitly. This can be done at your discretion or when the compiler asks you to.

More about types

export fn main() void = {
	// Numeric types can be declared explicitly:
	let
		a: int = 10,	// Signed integer
		b: uint = 10,	// Unsigned integer
		c: u8 = 10,	// Unsigned 8-bit integer
		d: f32 = 13.37;	// 32-bit floating point number

	// Or inferred from a suffix:
	let
		a = 10i,	// Signed integer
		b = 10u,	// Unsigned integer
		c = 10u8,	// Unsigned 8-bit integer
		d = 13.37f32;	// 32-bit floating point number

	// Some other types:
	let
		a: str = "hi",			// String
		b: (int, int) = (42, 24),	// Tuple
		c: struct {
			x: int,
			y: int,
		} = struct {
			x: int = 10,
			y: int = 20,
		},				// Struct
		d: [4]int = [1, 2, 3, 4],	// Array
		e: []int = [1, 2, 3, 4];	// Slice
};

Here is a small (non-comprehensive) sample of some other types supported by Hare. Hare supports signed and unsigned integers of various sizes (signed numbers are able to store negative values), as well as 32- and 64-bit floating point types (which can store a fractional component). The exact type can be inferred from context, as in the first set of examples, or specified with an appropriate suffix.

Note: If you are not already familiar with binary floating point arithmetic, you may be surprised when arithmetic using f32 and f64 types gives unexpected results. Programmers unfamiliar with the subject are encouraged to read the Wikipedia page on Floating-point arithmetic to better understand the theory and trade-offs.

Hare also supports a number of composite types, a few examples of which are shown here. We’ll go into more detail on how these work later. There are also a number of more specialized types that are not shown here, such as size and uintptr. Some of these are introduced later in this tutorial, while others are not useful outside of specialized situations and are left for you to discover when you need them.

Struct and tuple types

use fmt;

type coords = struct { x: int, y: int };

export fn main() void = {
	let player1 = struct {
		x: int = 10,
		y: int = 20,
	};
	let player2 = coords {
		y = 10,
		x = 20,
	};
	let player3: (int, int) = (42, 24);
	fmt::printfln("Player 1: ({}, {})", player1.x, player1.y)!;
	fmt::printfln("Player 2: ({}, {})", player2.x, player2.y)!;
	fmt::printfln("Player 3: ({}, {})", player3.0, player3.1)!;
};

Structs are one of the composite types supported by Hare. These define a structured value which is made up of other values in a certain order. In this example, we show two ways of using structs: first manually, and then by using a user-defined type, for “player1” and “player2” respectively. We can access the fields of the struct by using the . operator.

Note the “type coords” declaration above the main function: this defines a type named “coords”. This can be used for any type, not just structs. This type is equivalent to the hand-written struct used for the “player1” variable, but by giving it a name we can skip the types for each field (when creating a variable of that type) and re-order them if we so desire. The other player, “player1”, is using an anonymous, or un-named, struct type.

“player3” is defined with a tuple type, which is very similar to a struct, but does not name its fields. They are accessed by their ordinal position, starting from zero, instead of their names.

Arrays and slices

use fmt;

export fn main() void = {
	let x: [_]int = [1, 3, 3, 7];
	assert(len(x) == 4);	// len() built-in
	assert(x[3] == 7);	// 0-indexed
	x[3] = 8;		// assignment
	assert(x[3] == 8);

	let y: [1024]int = [1, 3, 3, 7, 42...];	// Fill remainder with 42

	printvals(y[..4]);
	printvals(y[2..8]);
};

fn printvals(in: []int) void = {
	fmt::printfln("input: {} integers", len(in))!;
	for (let i = 0z; i < len(in); i += 1) {
		fmt::printfln("in[{}]: {}", i, in[i])!;
	};
	fmt::println()!;
};

We’ll introduce array and slice types next. Arrays store a specific (determined at compile-time) number of ordered values of a uniform subtype; slices store an arbitrary (determined at runtime) number of ordered values of a uniform type.

An array may be declared with a specific length and subtype (e.g. [5]int), or may infer the length from context using an underscore (e.g. [_]int). A slice type leaves the length unwritten: []int. Values of a slice or array type can be indexed to obtain the value of one of their objects, numbered from zero, with the [index] operator as shown in the samples. The length (number of items) of an array or slice may be obtained with the len built-in.

Like all values, arrays must be initialized when they are declared. This can be awkward for large arrays like “y”. In such cases, the ... operator is often useful: it assigns all remaining values to the last value. In this example, most of “y” is initialized to 42.

Arrays and slices, continued

use fmt;

export fn main() void = {
	let x: [_]int = [1, 3, 3, 7];
	assert(len(x) == 4);	// len() built-in
	assert(x[3] == 7);	// 0-indexed
	x[3] = 8;		// assignment
	assert(x[3] == 8);

	let y: [1024]int = [1, 3, 3, 7, 42...];	// Fill remainder with 42

	printvals(y[..4]);
	printvals(y[2..8]);
};

fn printvals(in: []int) void = {
	fmt::printfln("input: {} integers", len(in))!;
	for (let i = 0z; i < len(in); i += 1) {
		fmt::printfln("in[{}]: {}", i, in[i])!;
	};
	fmt::println()!;
};

A slicing expression is used to “slice” arrays and slices with the .. operator. This creates a new slice which references a subset of the source object, such that y[2..5] will produce a slice whose 0th value is the 2nd value of “y”, and whose length is 5 - 2 = 3. Slicing does not copy the underlying data, so modifying the items in a slice will modify the underlying array.

Accesses to arrays and slices are bounds checked, which means that accessing a value beyond the end of their valid objects will cause your program to abort.

let x = [1, 2, 3];
x[4]; // Abort!

It is occasionally useful (but risky!) to skip the bounds check, for interop with code written in other languages, or in carefully reviewed, performance-critical code. To do this, you can use a pointer to an array type with * in place of the array length:

let x: [_]int = [1, 2, 3];
let y: *[*]int = &x;
y[4]; // Undefined behavior!

You have to use *[*]int here — not [*]int directly — because otherwise we would not be able to determine the size of the required stack allocation.

Using default values

TODO
TODO

Memory management

Stack allocation & pass by reference

use crypto::sha256;
use encoding::hex;
use hash;
use io;
use os;
use fmt;

export fn main() void = {
	// Pointer basics
	let i = 10;
	fmt::println(i)!;
	increment(&i);
	fmt::println(i)!;

	// Applied usage
	const hash = sha256::sha256();
	const file = os::open("main.ha")!;
	io::copy(&hash, file)!;

	let sum: [sha256::SZ]u8 = [0...];
	hash::sum(&hash, sum);
	hex::encode(os::stdout, sum)!;
	fmt::println()!;
};

fn increment(ptr: *int) void = {
	*ptr = *ptr + 1;
};

This sample demonstrates the use of pointers and stack allocation, the latter of which is an important design pattern in Hare. First it passes a copy of the “i” variable to the “increment” function by reference, allowing the increment function to modify i from outside of “main” by using the * operator.

When you call a function in Hare (or when the runtime calls “main” for you), a small block of memory called a stack frame is allocated to store all of the function’s variables and parameters. This is automatically cleaned up when you return from the function, which makes it useful for getting rid of resources when you’re done using them. However, it’s important not to allow a reference to any stack-allocated variables to persist after the function ends. Additionally, there is a limited amount of stack space, so it’s often wise to seek alternative strategies when allocating large objects — which we’ll address momentarily.

As a practical demonstration of stack allocation, this sample also computes and prints the SHA-256 hash of its source code. A common pattern in Hare is for a constructor like sha256::sha256 to return an object on the stack, which you can then pass into other functions by reference using &. This is convenient for many objects which can be cleaned up by simply discarding their state on the stack, but other kinds of objects (such as file handles) require additional steps.

Dynamic memory allocation & defer

use fmt;
use io;
use os;

export fn main() void = {
	// Allocation basics
	let x: *int = alloc(42);
	fmt::printfln(" x: {}", x)!;
	fmt::printfln("*x: {}", *x)!;
	free(x);

	// Applied example
	const file = os::open(os::args[1])!;
	defer io::close(file)!;

	let buffer: *[65535]u8 = alloc([0...]);
	defer free(buffer);

	const n = io::read(file, buffer)! as size;
	io::write(os::stdout, buffer[..n])!;
};

Another allocation strategy in Hare is heap or dynamic allocation of resources. A simple sample program is provided here, which opens the file referred to by its first command line argument, reads up to 64 KiB from it, and writes it to the standard output. You can run it on itself like so: hare run main.ha main.ha.

To allocate an object on the heap, use the alloc keyword along with an initializer in parentheses. The runtime will request the necessary memory from the operating system, initialize it to the value you provide here, and return a pointer to this value. The first “fmt” call in this example prints the location (or address) of the allocated memory, and the second call prints the value which was placed there. Even though uses of alloc look like function calls, alloc is not a function and does not work like one. Among the special things it does is automatic size calculation - when allocating a slice of 4 ints, you should write alloc([0...], 4), not alloc([0...], 4*size(int)).

Unlike stack-allocated resources, which clean themselves up when the function exits, heap-allocated resources must be “freed” by the caller using the free keyword. Another concept shown here is the use of defer to defer the execution of an expression to the end of the current scope (that is, the code bound by { and }). This gives you the ability to write the code which cleans up an object right next to the code which creates that object.

There are other kinds of resources that have to be cleaned up when you’re done using them, such as open files. defer is useful for these cases, too, and in this code sample we use it to call io::close to clean up the file opened by os::open.

Note: Failing to free allocated memory causes a memory leak, which is a bug in your program. Failing to close files is another kind of leak. Several of the earlier samples have memory leaks — can you identify and fix them?

Static allocation

use fmt;
use io;
use os;

let items: [4]int = [1, 3, 3, 7];

export fn main() void = {
	// Example A: Static globals
	printitems();
	items[3] = 1;
	items[0] = 7;
	printitems();

	// Example B: Static locals
	fmt::println(increment())!;
	fmt::println(increment())!;
	fmt::println(increment())!;

	// Example C: Applied static allocation
	const file = os::open(os::args[1])!;
	static let buffer: [65535]u8 = [0...];
	const n = io::read(file, buffer)! as size;
	io::write(os::stdout, buffer[..n])!;
};

fn printitems() void = {
	fmt::println(items[0], items[1], items[2], items[3])!;
};

fn increment() int = {
	static let x: int = 41;
	x += 1;
	return x;
};

The third major approach to allocation in Hare is static allocation. This approach involves creating a single object (a singleton) for the whole program to use. This never has to be cleaned up because it never goes away. The most apparent drawback of this approach is that you have to allocate all static resources in advance, when you write your program, and cannot allocate more or fewer static resources at runtime as needs demand them. This can increase the footprint of your program on RAM (and on disk, in some cases), and there are special constraints imposed on the initializers for static variables.

There are two ways to use static allocation in Hare: static locals and static globals. The first kind demonstrated in our example is a static global, which is defined using const or let in a similar manner to local variables (local here meaning local to a function). If we declare them with let, we can modify them.

A static local is also declared with const or let, with the addition of the static keyword. Note that, after being initially set to 41, the value of “x” remains consistent across repeated calls to “increment”. Unlike a stack-allocated local, which is allocated space in that specific function call’s stack frame, static locals are allocated to the function itself.

A practical example of static allocation is shown in the third part of this program, which, like the previous example, reads up to 64 KiB from the file specified by the first command-line argument, and writes it to stdout. In this case, however, instead of dynamically allocating a 64 KiB buffer, we statically allocate the buffer. We do not have to clean it up and the allocation cannot fail because the system is out of memory.

Note: Static variables must be initialized to a value which can be computed at compile-time. You cannot, for example, initialize a static variable based on the user's input at runtime. You can initialize it to a sensible default value and modify it at runtime instead.

Thinking in terms of ownership

use fmt;
use io;
use os;
use strings;

export fn main() void = {
	const file = os::open("main.ha")!;        // Opens file
	defer io::close(file)!;
	const buffer = io::drain(file)!;          // Allocates buffer
	defer free(buffer);
	const string = strings::fromutf8(buffer)!; // Borrows buffer
	fmt::print(string)!;
};

Hare uses manual memory management, which means that you, the programmer, are responsible for planning for and allocating the memory your program uses. Some functions in the standard library, and elsewhere, have consequences for memory management that you should be aware of. These are addressed in the documentation for these functions using standardized language of ownership, transfers and assumption of ownership, and borrowing resources.

Examine the documentation for the functions in this sample code with commands like haredoc io::drain. This function allocates a buffer to store the results into, and returns that buffer to the caller (you), who assumes ownership over the object. You are responsible for destroying this object when you are done with it, freeing resources like memory to be used elsewhere.

Many functions borrow resources to make use of them without taking responsibility for them. strings::fromutf8 is an example of this, as we can learn from its documentation. The return value, “string”, does not need to be (and should not be) freed, and will become invalid when the buffer it’s borrowed from is freed.

It is important to read the documentation for the functions you use to understand the ownership semantics they require, and to plan for this in your program. The standard library is extensively documented and “haredoc” makes it easy to access this information.

Note: We use terms like "borrow" and "ownership" to reason about memory, but this is not enforced at the language level. The compiler does not prevent double-free or use-after-free bugs. Hare does have other safety features, however, which will be addressed later on.

Handling errors

A few words about error handling

use errors;
use fmt;
use fs;
use fs::{flag};
use io;
use os;
use strings;

export fn main() void = {
	const path = os::args[1];
	const oflags = flag::WRONLY | flag::TRUNC;

	const file = match (os::create(path, 0o644, oflags)) {
	case let file: io::file =>
		yield file;
	case errors::noaccess =>
		fmt::fatalf("Error opening {}: Access denied.", path);
	case let err: fs::error =>
		fmt::fatalf("Error opening {}: {}", path, fs::strerror(err));
	};
	defer io::close(file)!;

	const buf = strings::toutf8("Hello world!\n");
	match (io::write(file, buf)) {
	case let err: io::error =>
		fmt::fatalf("Error writing file: {}", io::strerror(err));
	case let z: size =>
		assert(z == len(buf), "Unexpected short write");
	};
};

Many operations can fail. For instance, writing a file could fail if the disk is full, or a network connection could fail if the Ethernet cable is unplugged. In Hare, it is mandatory to consider these cases. In prior examples, we have used the ! operator, which causes the program to crash when an error occurs. Let’s explore some more effective ways of dealing with errors.

This program uses os::create to create a file (named after the first command line argument) and writes “Hello world!\n” to that file with io::write. Both of these functions can fail for a variety of reasons. You can try out some failure cases like so:

$ hare run main.ha example.txt      # should work
$ mkdir test && chmod -w test       # make a directory we cannot write to
$ hare run main.ha test/example.txt # os::create fails
$ hare run main.ha /dev/full        # io::write fails

Error handling is very important in Hare, so we’re going to go over this sample in detail in the next few sections.

Handling errors with match

use errors;
use fmt;
use fs;
use fs::{flag};
use io;
use os;
use strings;

export fn main() void = {
	const path = os::args[1];
	const oflags = flag::WRONLY | flag::TRUNC;

	const file = match (os::create(path, 0o644, oflags)) {
	case let file: io::file =>
		yield file;
	case errors::noaccess =>
		fmt::fatalf("Error opening {}: Access denied.", path);
	case let err: fs::error =>
		fmt::fatalf("Error opening {}: {}", path, fs::strerror(err));
	};
	defer io::close(file)!;

	const buf = strings::toutf8("Hello world!\n");
	match (io::write(file, buf)) {
	case let err: io::error =>
		fmt::fatalf("Error writing file: {}", io::strerror(err));
	case let z: size =>
		assert(z == len(buf), "Unexpected short write");
	};
};

Let’s look at the signature for os::create with haredoc:

fn create(str, fs::mode, fs::flags...) (io::file | fs::error);

This function can return one of two possible values: an io::file if the file was successfully opened, or fs::error if not. You can look up both of these types with haredoc if you like, and your attention is especially drawn to fs::error. This type is a tagged union which represents each of the errors that can be caused during file operations.

One way of handling these errors is with !, which you already know how to use. A more elegant way is to use match. A match expression takes an expression between its parentheses which can return one of several types from a tagged union, and each case handles one of these types. The first case in this example is the successful path, which we’ll talk about more momentarily. The second case handles a specific error: errors::noaccess, and the third case handles any other errors.

The case let syntax is used to bind the value for each case to a variable. In the first branch, this value uses the yield keyword to yield it to the parent expression, which causes it to be “returned”, in a manner of speaking, from the match expression. This assigns the desired io::file value to the file variable.

The third case binds fs::error to an “err” variable, then passes it into fs::strerror to convert it to a string that can be presented to the user in an error message. This is a standard pattern in Hare: most modules will provide a “strerror” function which stringifies all of the errors which can be caused by that module.

Propagating errors

use fmt;
use fs;
use fs::{flag};
use io;
use os;
use strings;

export fn main() void = {
	const path = os::args[1];
	match (writehello(path)) {
	case void =>
		yield;
	case let err: fs::error =>
		fmt::fatalf("Error writing {}: {}", path, fs::strerror(err));
	case let err: io::error =>
		fmt::fatalf("Error writing {}: {}", path, io::strerror(err));
	};
};

fn writehello(path: str) (fs::error | io::error | void) = {
	const oflags = flag::WRONLY | flag::TRUNC;
	const file = os::create(path, 0o644, oflags)?;
	defer io::close(file)!;
	const buf = strings::toutf8("Hello world!\n");
	io::write(file, buf)?;
};

It is cumbersome to use match to enumerate every possible failure for every function that might fail. To make it easier to deal with errors, the ? operator is generally useful. The purpose of this operator is to check for errors and, if found, return them to a higher call frame. If there are no errors, execution will proceed normally.

To use this functionality, it is necessary to establish some error handling code somewhere in the program. In this sample, the “main” function is responsible for all error handling. In more complex programs, you may handle various kinds of errors at different levels throughout the program. An HTTP server, for instance, might have some logic to handle configuration errors by printing a message and stopping the server, but could handle errors related to a specific client by sending them an error response or disconnecting them.

os::create and io::write together can return either an fs::error or an io::error, so the result type for our “writehello” function is a tagged union of either void (nothing, indicating success), fs::error, or io::error. We can then use ? to return these errors immediately and extract the useful types from the return values of these functions, and handle both cases in “main”.

Defining new error types

use bufio;
use fmt;
use io;
use os;
use strconv;
use strings;

export fn main() void = {
	match (prompt()) {
	case void =>
		yield;
	case let err: error =>
		fmt::fatal(strerror(err));
	};
};

// An invalid number was provided.
type invalid = !(strconv::invalid | strconv::overflow);

// An error which indicates that [[io::EOF]] was unexpectedly encountered.
type unexpectedeof = !void;

// Tagged union of all possible errors.
type error = !(io::error | invalid | unexpectedeof);

// Converts an error into a user-friendly string
fn strerror(err: error) str = {
	match (err) {
	case invalid =>
		return "Expected a positive number";
	case unexpectedeof =>
		return "Unexpected end of file";
	case let err: io::error =>
		return io::strerror(err);
	};
};

fn prompt() (void | error) = {
	fmt::println("Please enter a positive number:")!;
	const num = getnumber()?;
	fmt::printfln("{} + 2 is {}", num, num + 2)!;
};

fn getnumber() (uint | error) = {
	const num = match (bufio::read_line(os::stdin)?) {
	case io::EOF =>
		return unexpectedeof;
	case let buf: []u8 =>
		yield strings::fromutf8(buf)!;
	};
	defer free(num);
	return strconv::stou(num)?;
};

Here we have a somewhat more complex sample in which we prompt the user to enter a number and then enumerate all possible error cases, such as entering something other than a number or pressing Ctrl+D to close the input file without entering anything at all. Try doing these things yourself and seeing how the program responds.

We can define new error types ourselves by using ! to prefix the type declaration. invalid is an error type which derives from the errors which can be returned from strconv::stou, and unexpectedeof is a custom error based on the void type. The latter does not store any additional state other than the type itself, so it has a size of zero. We also define a tagged union containing each of these error types, plus io::error, also using ! to indicate that it is an error type.

We can return our custom error from “getnumber” upon encountering io::EOF (which is not ordinarily considered an error) from bufio::read_line, which is propagated through prompt to “main”. If an I/O error were to occur here, it would be propagated similarly.

We can use any type as an error type if we wish. Some errors are an int or enum type containing an error code, or a struct with additional information like a client IP address or a line and column number in a file, and so on. We can also provide our own “strerror” functions which provide helpful error strings, possibly incorporating information stored in the error type.

Handling allocation failure

TODO
TODO

Testing your code

fn sort(items: []int) void = {
	for (true) {
		let sorted = true;
		for (let i = 1z; i < len(items); i += 1) {
			if (items[i - 1] > items[i]) {
				const x = items[i - 1];
				items[i - 1] = items[i];
				items[i] = x;
				sorted = false;
			};
		};
		if (sorted) {
			break;
		};
	};
};

@test fn sort() void = {
	let items = [5, 4, 3, 2, 1];
	sort(items);
	for (let i = 1z; i < len(items); i += 1) {
		assert(items[i - 1] <= items[i], "list is unsorted");
	};
};

Hare has first-class support for tests via the @test attribute on functions. You can run the tests for this sample program by running hare test in the directory where this file is present.

Our simple sample here is a bubble sort implementation. You may already know that this is not a very good sort algorithm — you will likely wish to use the standard library’s sort module in real-world code.

Note that this code does not have a “main” function — hare run will not work here. You can test code which is not strictly speaking a “program”, such as libraries written in Hare.

Also take note of the use of the assert built-in: given a condition (a bool), if the condition is false, the program is stopped and the message is printed. You can also skip the message if you don’t need to be specific; the file name and line number will be printed and you can generally figure out what went wrong regardless.

Control flow

if & switch statements

use fmt;

type color = enum {
	RED,
	ORANGE,
	YELLOW,
	GREEN,
	BLUE,
	VIOLET,
};

export fn main() void = {
	let stock = [
		(color::RED, 1),
		(color::BLUE, 6),
		(color::VIOLET, 1),
		(color::ORANGE, 4),
	];
	fmt::println("Inventory:")!;
	printstock(stock[0]);
	printstock(stock[1]);
	printstock(stock[2]);
	printstock(stock[3]);
};

fn printstock(item: (color, int)) void = {
	const color = item.0, amount = item.1;
	fmt::printfln("{} paint\t{} liter{}",
		colorstr(color), amount,
		if (amount != 1) "s" else "")!;
};

fn colorstr(c: color) str = {
	switch (c) {
	case color::RED =>
		return "Red";
	case color::ORANGE =>
		return "Orange";
	case color::YELLOW =>
		return "Yellow";
	case color::GREEN =>
		return "Green";
	case color::BLUE =>
		return "Blue";
	case color::VIOLET =>
		return "Violet";
	};
};

We have used these expressions many times without explanation, but now we’ll dive into control statements in depth. We’ll begin with statements that select from one or more “branches” of execution: if and switch.

An if statement selects from one or more branches based on the truth of a boolean expression, such as x > 5. If this statement is true, the corresponding branch is executed. Optionally, you may follow the “true” branch with the else keyword and another branch, which is executed should the conditional expression turn out to be false. Hare also supports else if expressions, which execute their branch if the previous branch is false and a second condition of their own is met.

if (condition 1) {
    // Executes if condition 1 is true
} else if (condition 2) {
    // Executes if condition 1 is not true and condition 2 is true
} else if (condition N...) {
    // And so on...
} else {
    // Executes if all other conditions were false
};

An important note is that if expressions, like most Hare expressions, can appear in any expression and can compute a value. The “printstock” function from this sample code uses this to determine if “liter” should be written in the plural form by testing if the quantity is not one, then selecting between an "s" suffix and an empty string ("").

Switch expressions provide a more structured approach to branching. Based on a single value, the switch value, one of several cases is selected. The switch value can be any primitive type, such as ints or rune, as well as str and enum values. The “colorstr” function in the sample uses a switch expression to select a different branch based on the value of the color parameter, then returns a string representing that value.

Note that switch expressions are required to be exhaustive, which means that every possible value of the switch value should have a corresponding branch. If you do not want to handle every possibility individually, you can add a “default” branch by omitting the value: case =>. This branch will be executed should no more-specific branch suffice. This is also possible with match cases that omit the match type.

Using yield

use fmt;
use fs;
use io;
use os;

export fn main() void = {
	const file = match (os::open(os::args[1])) {
	case let f: io::file =>
		yield f;
	case let err: fs::error =>
		fmt::fatalf("Unable to open {}: {}",
			os::args[1], fs::strerror(err));
	};

	match (io::copy(os::stdout, file)) {
	case size =>
		yield;
	case let err: io::error =>
		fmt::fatalf("copy: {}", fs::strerror(err));
	};
};

In many of the samples so far, we’ve seen the use of { and } to denote blocks of expressions which are evaluated one after another. In many languages, these are decorative, but in Hare they have a semantic meaning: these introduce new compound expressions, and, like most other expressions, these expressions can compute a result using the yield keyword.

The simple example given here is meant to show some simple usages of yield, similar to ones we’ve seen and left unexplained in previous samples. The only clarification we wish to add here is to draw your attention to the { and } characters, and clarify that the yield keyword causes the nearest compound expression to stop and to have the result of that expression set to the value provided to yield. The sample code shown here uses this to cause the value of “f” to be assigned to the “file” variable for further use outside of the match expression.

This example keeps things simple, but, in the field, you will probably find yield useful less frequently in situations that are less obvious. Getting values out of match expressions is likely to be the most common use of this feature in your code. But, keep yield in your tool belt and you might end up using it to solve other problems every now and then, too.

For loops

use fmt;

export fn main() void = {
	const items = [
		"Hello,",
		"world!",
		"Hare",
		"is",
		"cool!",
	];
	let i = 0z;
	for (i < len(items); i += 1) {
		if (items[i] == "Hare") {
			break;
		};
	};
	fmt::printfln(`"Hare" is at index {}`, i)!;
};

All loops in Hare are written with the for keyword, which allows some logic to repeat so long as a condition is met. The syntax is:

for (binding; condition; afterthought)

The binding allows you to declare and initialize a variable which only exists within the loop. The condition determines when the loop terminates — it’s evaluated at the start of each loop iteration, and will terminate the loop if false. The afterthought runs after each iteration, and is a convenient place to update the variables used for the loop condition.

The binding and the afterthought may be omitted, such as in this code sample which omits the binding, allowing us to access the loop item (“i”) outside of the loop. A loop which never terminates may be written like so:

for (true) {
	// ...
};

For-each loops

use fmt;

const items = [
	"apple",
	"banana",
	"carrot",
];
let items_counter = 0z;

// iterator function
fn next() (str | done) = {
	if (items_counter < len(items)) {
		items_counter += 1;
		return items[items_counter - 1];
	};
	return done;
};


export fn main() void = {
	// for-each value
	for (let item .. items) {
		fmt::printfln("{}", item)!;
	};

	// for-each reference
	for (let item_ptr &.. items) {
		fmt::printfln("{} = {}", item_ptr, *item_ptr)!;
	};

	// for-each iterator
	for (let item => next()) {
		fmt::printfln("{}", item)!;
	};
};

In addition to the for loops described above, there are also for-each loops, which make common patters a lot simpler. These have no condition or afterthought, just a binding. The are three forms of for-each loops in Hare: for-each value, for-each pointer and for-each iterator.

The first for-each loop (for-each value) iterates over every element of the items array by value. Each element is copied to the binding. This is good for small sized values or if you don’t want mutate the slice or array directly.

The second for-each loop iterates over every element of the items array by reference. Here, only a pointer to each element is assigned. This is useful if you want to mutate the array/slice itself or want to avoid copies beacuse the values are too big. The example prints the pointer to the element and then the element itself.

In the third for-each loop, the initializer of the binding is a special “iterator” function whose return type is a tagged union which includes the done type. This function is called each iteration and it’s return value, is checked. If the value’s type is done, the loop is terminated. Otherwise, the value is assigned, and its type is that of the return type minus done. In this case, that would be str.

A lot of standard libary functions make use of the done type or aliases of it. You can use a for-each iterator loop in these cases, or create some useful functions yourself.

Note: It is not possible to `insert`, `delete`, `free` and `append` to the slice beeing iterated over. It is encouraged to use normal for loops for this use case instead.

Flow control

use fmt;

export fn main() void = {
	const items = [
		"Hello,",
		"world!",
		"Hare",
		"is",
		"cool!",
	];
	for (let item .. items) {
		if (item == "Hare") {
			continue;
		};
		fmt::printfln(item);
	};
};

We can use the break and continue keywords to manipulate control flow in Hare loops. In the previous sample, we used break to terminate the loop early when encountering a string called “Hare”; in this sample we stop that loop iteration and continue with the next iteration.

Both break and continue have the result type never. What does that mean?

The `never` type

use fmt;

export fn main() void = {
	const color = color::BLUE;
	const name = switch (color) {
	case color::RED =>
		yield "red";
	case color::ORANGE =>
		yield "orange";
	case color::YELLOW =>
		yield "yellow";
	case color::GREEN =>
		abort("green is not a creative color");
	case color::BLUE =>
		yield "blue";
	case color::VIOLET =>
		yield "violet";
	};
	fmt::println("your color is", name)!;
};

type color = enum {
	RED,
	ORANGE,
	YELLOW,
	GREEN,
	BLUE,
	VIOLET,
};

Some expressions, such as return, break, continue, yield, and abort(), as well as some function calls such as os::exit, can never return any value, either because the program is guaranteed to exit or because control flow continues somewhere else. This is represented within Hare’s type system as the never type - a type that can never hold a value.

This is important when you consider that Hare is an expression-oriented language. In the sample here, each branch of our switch expression yields a value (the name of the color) to serve as the result of the switch expression — except for color::GREEN. Because abort()’s result type is never, the branch doesn’t yield a value at all, and is ignored when determining the switch expression’s result type.

We have taken advantage of this behavior many times throughout the tutorial. For example, in Using yield, fmt::fatalf returns never, allowing us to only provide a value in one case.

Types in depth

Promotion and type inference

TODO
TODO

Casting & type assertions

use fmt;

export fn main() void = {
	// Casting between numeric types allows for lossy conversions:
	fmt::println(13.37f32: int)!;

	// Type assertions let you examine the type of a tagged union:
	let x: (int | uint) = 42i;
	assert(x is int);
	fmt::println(x as int)!;

	let y: nullable *(int | uint) = &x;
	assert(!(y is null)); // And nullable pointers

	// You can also use casts for pointer arithmetic, ill-advised as that
	// may be:
	fmt::printfln("{:x} {:x}", y, y: uintptr + 10)!;

	// Casts can also be used to change pointer types, ill-advised as that
	// may be:
	let z = (y: uintptr + size(int): uintptr): *int;
	assert(*z == 42);
};

This sample introduces various ways of working with types explicitly in Hare. One such example is casting types from one to another. One use-case for this is given in the first example: converting between numeric types. Since the fractional part is lost, this conversion is lossy, and therefore must be done explicitly. Converting between signed and unsigned types must also be explicit.

Hare also supports type tests and type assertions via the is and as keywords, which can be used to work with the selected type of a tagged union at runtime. An is expression returns a bool, true if the tagged union is set to a value of the given member type. The as expression is useful for when you know that a tagged union has a particular type and you want to treat it as that type — this “assertion” is tested at runtime and will cause your program to abort if found to be untrue (try swapping x as int for x as uint to demonstrate this).

Casts are tested at compile time to ensure that the desired conversion is generally possible, but no attempt is made to test that it is advisable. Thus, casting can be used to override Hare’s type system when the programmer believes that they know better, and this comes with risks. In the simplest case, converting an f32 to an int might lead to some bugs because you wanted to round instead of truncate. A more dangerous example is shown here as well: pointer arithmetic and conversions between pointer types. Using casts, you can instruct Hare to treat some memory as if it were a given type, regardless of if it actually is or not. Use with caution.

User-defined types

type index = size;
type offs = size;

export fn main() void = {
	let z: (index | offs) = 1337: offs;
	assert(z is offs);
};

You have already seen user-defined types several times throughout this tutorial, defining types like player coordinates, enums for colors, and so on. You have likely already gathered how this works: use the type keyword to begin a new type declaration.

We want to illustrate one additional point here: each user-defined type creates a distinct type identity in the Hare type system. One reason this is important is for tagged unions: even if two types share the same underlying representation (such as “index” and “offs” in our sample code), they can be considered distinct types in a tagged union.

Pointer types in depth

use fmt;

type coords = struct {
	x: int,
	y: int,
};

export fn main() void = {
	let pos = coords { x = 10, y = 20 };
	printcoords(null);
	printcoords(&pos);
};

fn printcoords(pos: nullable *coords) void = {
	match (pos) {
	case null =>
		fmt::println("(null)")!;
	case let pos: *coords =>
		fmt::printfln("({}, {})", pos.x, pos.y)!;
	};
};

Pointers in Hare, by default, cannot be “null”. In other words, all pointers must refer to a valid address. However, it is often useful to signal the absence of a value, and we can do this with a nullable pointer type.

The “printcoords” function in the sample code accepts an argument of type nullable *coords. We cannot dereference this type with the * or . operators like we ordinarily can: we must first test if it is valid. One way to do this is to match against null, as shown here.

Hare also includes a feature called “auto-dereferencing”, which allows you to do things like using the . operator to access struct members via a pointer. This allows us to use pos.x rather than (*pos).x. Many other features work similarly — indexing arrays, append et al (covered later), and so on. This works for any level of indirection: pointers to pointers to pointers to pointers… can also be dereferenced automatically.

Struct sub-typing

use fmt;
use io;
use os;

type limitstream = struct {
	stream: io::stream,
	sink: io::handle,
	limit: size,
};

const limit_vtable: io::vtable = io::vtable {
	writer = &limit_write,
	...
};

fn limitwriter(sink: io::handle, limit: size) limitstream = {
	return limitstream {
		stream = &limit_vtable,
		sink = sink,
		limit = limit,
	};
};

fn limit_write(st: *io::stream, buf: const []u8) (size | io::error) = {
	const st = st: *limitstream;
	const buf = if (len(buf) > st.limit) {
		yield buf[..st.limit];
	} else {
		yield buf;
	};
	st.limit -= len(buf);
	return io::write(st.sink, buf);
};

export fn main() void = {
	const limit = limitwriter(os::stdout, 5);
	fmt::fprintln(&limit, "Hello world!")!;
};

Pointers to structs enjoy a special feature in Hare: automatic sub-typing. Essentially, a pointer to one type can be automatically cast to a pointer of another type so long as the second type appears at the beginning of the first type. This is useful for implementing abstract interfaces in Hare programs.

This sample program takes advantage of this by implementing a custom io::stream that limits the total amount of data which can be written, essentially demonstrating a slimmed down version of the standard library’s built-in limitwriter. Running this program only writes the first five bytes to stdout: “Hello”.

The pointer to &limit passed to the fmt::fprintln call is of type *limitstream, which embeds the io::stream type as the first field. Thus, we can safely pass it to a function expecting an *io::stream pointer, such as fmt::fprintln. In the “limit_write” function, we cast this pointer back to *limitstream so that we can access additional data we need to store to implement the limit writer functionality.

Tagged unions in depth

use bufio;
use hare::ast;
use hare::lex;
use hare::parse;
use hare::types;
use io;
use strings;

type signed = (int | i8 | i16 | i32 | i64);
// equivalent to:
// type signed = (i64 | (i32 | (i16 | (i8 | int | int | int))));
type unsigned = (uint | u8 | u16 | u32 | u64);
type integer = (...unsigned | ...signed);
type floating = (f32 | f64);
type numeric = (...integer | ...floating);

type numeric_repr = struct {
	id: u32,
	union {
		_int: int,
		_i8: i8,
		_i16: i16,
		_i32: i32,
		_i64: i64,
		// ...
	},
};

export fn main() void = {
	const input = bufio::fixed(strings::toutf8("int"), io::mode::READ);
	const lexer = lex::init(&input, "<string>");
	const _type = parse::_type(&lexer)!;
	defer ast::type_finish(&_type);
	const store = types::store(types::x86_64, null, null);
	defer types::store_free(store);
	const itype = types::lookup(store, &_type) as const *types::_type;

	const obj: numeric = 1337;
	const ptr = &obj: *numeric_repr;
	assert(ptr.id == itype.id);
	assert(ptr._int == 1337);
};

Let’s take a closer look at tagged unions in Hare. Tagged union types have three important properties: they are commutative, associative, and reductive. This is illustrated by the commented-out version of the “signed” type in the sample code — the order of types, inclusion of the same type more than once, or use of nested tagged unions has no effect on the final type.

An exception to this occurs when using type aliases: a tagged union which contains a type alias referring to another tagged union does not reduce, unless you use the ... operator. The “integer” and “numeric” types in our sample make use of this behavior. This reduces the size of the final type (by not nesting tagged unions), but is undesirable when the underlying representations should be distinct — two error types which are type aliases for !void would not be distinguishable if you used ... to incorporate them into the same tagged union.

The rest of the sample is not important to understand, but does illuminate some of the internal implementation details of tagged unions. Each type in Hare is assigned a unique ID, which is stored as the first field of a tagged union type to indicate which type is stored there. Following the tag, a union of all of the possible types is stored.

The Hare standard library provides access to tools for parsing and introspecting Hare programs, which the sample makes use of to find the type ID of int and compare it against the one stored in the tagged union. We’re not going to go into detail here on how any of this works, but feel free to browse the relevant modules with haredoc if you’re curious.

Working with slices

Growable slices

use bufio;
use fmt;
use io;
use os;
use sort;
use strings;

export fn main() void = {
	fmt::println("Enter a list of strings, then press <Ctrl+D>:")!;

	let lines: []str = [];
	defer strings::freeall(lines);

	for (true) {
		const line = match (bufio::read_line(os::stdin)!) {
		case let line: []u8 =>
			yield line;
		case io::EOF =>
			break;
		};
		append(lines, strings::fromutf8(line)!);
	};

	insert(lines[0], strings::dup("test line"));

	for (let i = 0z; i < len(lines); i += 1) {
		if (lines[i] == "foobar") {
			free(lines[i]);
			delete(lines[i]);
			i -= 1;
		};
	};

	sort::strings(lines);

	fmt::println("Your strings, sorted:")!;
	for (let line .. lines) {
		fmt::printfln(line)!;
	};
};

Slices in Hare can grow and shrink dynamically via the use of the append, insert, and delete keywords. The code sample shown here makes use of all of these features to sort a user-provided list of strings.

You can create an empty slice by simply using [] in the initializer as shown here. You can also create a slice pre-filled with data with the alloc keyword, such as let x: []int = alloc([1, 2, 3]). In either case, once you have a dynamically allocated slice, you must free it. You can pass a slice directly to free, but in this sample we also need to free the strings returned by bufio — so we use strings::freeall, which frees both the slice and the strings inside of it. To meet the criteria of strings being allocated, we had to copy a literal string to the heap using strings::dup.

There’s another form of delete which is not shown here: you can delete more than one element of a slice at a time by specifying a slicing expression rather than an indexing expression, such as delete(x[1..4]).

Static slice operations

use bufio;
use fmt;
use io;
use os;
use sort;
use strings;

export fn main() void = {
	fmt::println("Enter no more than 10 strings, then press <Ctrl+D>:")!;

	let buf: [10]str = [""...];
	let lines = buf[..0];
	defer for (let line .. lines) {
		free(line);
	};

	for (true) {
		const line = match (bufio::read_line(os::stdin)!) {
		case let line: []u8 =>
			yield line;
		case io::EOF =>
			break;
		};
		static append(lines, strings::fromutf8(line)!);
	};

	static insert(lines[0], strings::dup("test line"));

	for (let i = 0z; i < len(lines); i += 1) {
		if (lines[i] == "foobar") {
			free(lines[i]);
			static delete(lines[i]);
			i -= 1;
		};
	};

	sort::strings(lines);

	fmt::println("Your strings, sorted:")!;
	for (let line .. lines) {
		fmt::println(line)!;
	};
};

The slice operations introduced in the prior sample cover dynamic allocations, where the memory required to store the slice is allocated at runtime. Hare also supports static slice operations, where the underlying storage is provided for by the user. This is useful for many situations where memory allocation is undesirable, such as in kernels or embedded programming.

In this example, instead of letting the runtime allocate an arbitrary number of strings, we start with a fixed buffer of ten empty strings, and create a zero-length slice from it: buf[..0]. Each of our append, delete, and insert operations is prefixed with the static keyword here, which prevents Hare from re-allocating a slice with insufficient storage space to perform the desired operation.

When using static slices, the burden lies with you to ensure that the program will not exceed the allocated space. Try entering more than 10 items: the program will terminate with an error.

Slice assignment

use fmt;

export fn main() void = {
	let x: []int = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10];
	printints(x);
	x[2..8] = [0, 0, 0, 0, 0, 0];
	printints(x);
};

fn printints(x: []int) void = {
	for (let i = 0z; i < len(x); i += 1) {
		fmt::printf("{}{}", x[i], if (i + 1 < len(x)) ", " else "\n")!;
	};
};

It is often useful in Hare programs to copy data from one slice to another. We can accomplish this using the slice assignment syntax by placing a slicing expression on the left side of an assignment (=). Like other operations involving slices, this is bounds-checked: the right value must have a length equal to the length of the slice on the left.

This behaves similarly to C’s memmove function: it’s safe to copy data from a slice to itself using this technique.

Functions in depth

Variadic functions

use fmt;

export fn main() void = {
	printints("args ", 1, 2, 3, 4, 5);
	let x: []int = [5, 4, 3, 2, 1];
	printints("slice ", x...);
};

fn printints(prefix: str, x: int...) void = {
	fmt::print(prefix)!;
	for (let i = 0z; i < len(x); i += 1) {
		fmt::printf("{}{}", x[i], if (i + 1 < len(x)) ", " else "\n")!;
	};
};

The functions we have written so far have a fixed arity. Functions taking one argument are unary, two arguments binary, three arguments trinary, and so on. Hare also supports functions with a variable arity; such functions are called variadic.

To declare a function as variadic, place a ... after the type of its final parameter. This will cause the type of the paramter to be a slice of this type; in the example, “x” becomes type []int. Each additional parameter passed to the function is appended to this slice.

We can also call variadic functions in one of two ways: either by listing all of the parameters we want to pass directly, as in the first call to “printints”, or by passing a slice with the ... operator again, as in the second example.

The “fmt” family of functions we’ve been using throughout this tutorial is also variadic, which is why we have been able to pass it an arbitrary number of arguments throughout the tutorial.

Function pointers

use fmt;
use os;
use strconv;

fn usage() never = fmt::fatalf("usage: {} <add|sub> <x> <y>", os::args[0]);

export fn main() void = {
	if (len(os::args) != 4) usage();

	const func: *fn(int, int) int = switch (os::args[1]) {
	case "add" =>
		yield &add;
	case "sub" =>
		yield &sub;
	case => usage();
	};

	const x = match (strconv::stoi(os::args[2])) {
	case let i: int =>
		yield i;
	case => usage();
	};

	const y = match (strconv::stoi(os::args[3])) {
	case let i: int =>
		yield i;
	case => usage();
	};

	fmt::println(func(x, y))!;
};

fn add(x: int, y: int) int = x + y;
fn sub(x: int, y: int) int = x - y;

Much like other objects in Hare, it is possible to take the address of a function with the & operator. You may then call this variable like you would any ordinary function. This allows you to choose between different behaviors at runtime.

In this sample, we take the address of the add and sub functions and store it in a variable which has a function pointer type. The type is optional here — it could be inferred — but we chose to spell it out here to illustrate how these types are written. You can include these types in struct fields, function parameters, and so on, like you could with any other value.

Note: This code sample violates a few of the guidelines set forth by the style guide — can you spot them? The guidelines are not set in stone: if your code is better when you ignore some of them, do it.

We also make use of never here so that switch and match branches which call “usage” are excluded from result type selection, as explained earlier.

@init, @fini, @test

use io;
use os;
use strings;

export fn main() void = {
	io::write(os::stdout, strings::toutf8(src))!;
};

let src: str = "";

@init fn init() void = {
	const file = os::open("main.ha")!;
	defer io::close(file)!;
	const data = io::drain(file)!;
	src = strings::fromutf8(data)!;
};

@fini fn fini() void = {
	free(src);
};

We have already seen some function attributes throughout this tutorial. We’d like to explain the ones you’ve already seen, and introduce two more: @init and @fini.

Decorating a function with @init causes it to be run when your program starts, and is useful for initializing your application state. @fini is run when your program is terminated, which is useful for the opposite. In this sample, we use it to prepare the global “src” variable with the program’s own source code.

You’ve already seen @test: this marks a function as a test function, which is only used when you run “hare test”. There are a few other attributes supported by Hare, but they’re generally only useful in specific, uncommon situations — you’ll know if you need it.

Modules

Organizing your code in many files

Sample project layout:

  • ./
    • main.ha
    • hello.ha
    • example/
      • print.ha
  • /usr/src/hare/stdlib
    • fmt/
      • README
      • fmt.ha
      ...
// main.ha
use example;

export fn main() void = {
	example::print(hello);
};
// hello.ha
const hello: str = "Hello world!";
// example/print.ha
use fmt;

export fn print(what: str) size = fmt::println(what)!;

So far we’ve been putting all of our code into the same file, but as you step out into the world to write real Hare code you would be well advised to organize your code somewhat more carefully.

Each Hare module exists in a directory on your filesystem and contains some or all of the .ha files in that directory. Each file in this directory can use each other’s private declarations (types, functions, and so on), and has its own separate namespace of imports (use statements). Thus, to expand your program into multiple files, you can simply write your code into several files in a directory and pass that path to hare build or hare run. You can also omit the path entirely and these commands will assume you want to use the current directory.

When you use a module, Hare will search for that module in your HAREPATH, which is a colon-delimited list of directories to look through. The first directory is always the current directory, so if you want to add a private module called “example” you can place its files at ./example/*.ha. The standard library is usually installed (on Unix systems) at /usr/src/hare/stdlib, and third-party modules are installed at /usr/src/hare/third-party, both of which are configured in your HAREPATH by default. We recommend that you make liberal use of these resources in your work — don’t hesitate to read the source code for your dependencies.

These are the basics — enough to get started with real-world Hare programs. If you want to learn more, about how to organize large projects, how to write Hare libraries, how to compile different code for each platform, and so on, check out the modules tutorial next.

And that’s the Hare programming language! Nice work getting through all of that. We didn’t cover everything here — check out the Hare specification if you want the comprehensive reference. If you encounter any unusual code out there in the wild that you don’t understand, don’t hesitate to reach out for help.

Speaking of which, this is a good time to remind you about our community resources:

If you want to read some more real-world Hare code samples, also check out the hautils project, which offers several small and straightforward implementations of common shell commands like “tee”. This is also a great project to get started with writing some real-world Hare code — maybe you’ll send us a patch?

The next tutorial is the standard library introduction.