Introduction #

This article is an introduction to the Odin Programming Language. You’re assumed to have basic knowledge of programming concepts such as variables, statements, and types. You will see how familiar things are done in Odin, while also being introduced to new concepts.

If you haven’t already installed Odin, then you can read how in the Getting Started with Odin guide.

Hellope! #

To begin this tour, let us start with a modified version of the famous “hello world” program:

package main

import "core:fmt"

main :: proc() {
	fmt.println("Hellope!")
}

Save this code to a .odin file, then compile and run it using odin run <dir>. For the current directory:

odin run .

The run command compiles the .odin file to an executable and then runs that executable after compilation. If you do not wish to run the executable after compilation, the build command can be used.

odin build <dir>

Odin thinks in terms of directory-based packages. The odin build <dir> command takes all the files in the directory <dir>, compiles them into a package and then turns that into an executable. You can also tell it to treat a single file as a complete package, by adding -file, like so:

odin run hellope.odin -file

Lexical elements and literals #

Comments #

Comments can be anywhere outside of a string or character literal. Single line comments begin with //:

// A comment

my_integer_variable: int // A comment for documentation

Multi-line comments begin with /* and end with */. Multi-line comments can be also be nested (unlike in C):

/*
	You can have any text or code here and
	have it be commented.
	/*
		NOTE: comments can be nested!
	*/
*/

Comments are parsed as tokens within the compiler. This is to allow for future work on automatic documentation tools.

String and character literals #

String literals are enclosed in double quotes and character literals in single quotes. Special characters are escaped with a backslash \.

"This is a string"
'A'
'\n' // newline character
"C:\\Windows\\notepad.exe"

Raw string literals are enclosed in single back ticks.

`C:\Windows\notepad.exe`

The length of a string can be found using the built-in len proc:

len("Foo")
len(some_string)

If the string passed to len is a compile-time constant, the value from len will be a compile-time constant.

Escape Characters #

• \a - bell (BEL)
• \b - backspace (BS)
• \e - escape (ESC)
• \f - form feed (FF)
• \n - newline
• \r - carriage return
• \t - tab
• \v - vertical tab (VT)
• \\ - backslash
• \" - double quote (if needed)
• \' - single quote (if needed)
• \NNN- octal 6 bit character (3 digits)
• \xNN - hexadecimal 8 bit character (2 digits)
• \uNNNN - hexadecimal 16-bit Unicode character UTF-8 encoded (4 digits)
• \UNNNNNNNN - hexadecimal 32-bit Unicode character UTF-8 encoded (8 digits)

Numbers #

Numerical literals are written similar to most other programming languages. A useful feature in Odin is that underscores are allowed for better readability: 1_000_000_000 (one billion). A number that contains a dot is a floating point literal: 1.0e9 (one billion). If a number literal is suffixed with i, it is an imaginary number literal: 2i (2 multiply the square root of -1).

Binary literals are prefixed with 0b, octal literals with 0o, and hexadecimal literals with 0x. A leading zero does not produce an octal constant (unlike C).

In Odin, if a number constant can be represented by a type without precision loss, it will automatically convert to that type.

x: int = 1.0 // A float literal but it can be represented by an integer without precision loss

Constant literals are “untyped” which means that they can implicitly convert to a type.

x: int // `x` is typed as being of type `int`
x = 1 // `1` is an untyped integer literal which can implicitly convert to `int`

Variable declarations #

A variable declaration declares a new variable for the current scope.

x: int // declares `x` to have type `int`
y, z: int // declares `y` and `z` to have type `int`

Variables are initialized to zero by default unless specified otherwise.

Note: Declarations have to be unique within a scope.

x := 10
x := 20 // Redeclaration of `x` in this scope
y, z := 20, 30
test, z := 20, 30 // not allowed since `z` exists already

Assignment statements #

The assignment statement assigns a new value to a variable/location:

x: int = 123 // declares a new variable `x` with type `int` and assigns a value to it
x = 637 // assigns a new value to `x`

= is the assignment operator.

You can assign multiple variables with it:

x, y := 1, "hello" // declares `x` and `y` and infers the types from the assignments
y, x = "bye", 5

Note: := is two tokens, : and =. The following are all equivalent:

x: int = 123
x:     = 123 // default type for an integer literal is `int`
x := 123

Constant declarations #

Constants are entities (symbols) which have an assigned value. The constant’s value cannot be changed. The constant’s value must be able to be evaluated at compile time:

x :: "what" // constant `x` has the untyped string value "what"

Constants can be explicitly typed like a variable declaration:

y : int : 123
z :: y + 7 // constant computations are possible

For more information regarding value declarations in general, please see the Odin FAQ and Ginger Bill’s article On the Aesthetics of the Syntax of Declarations.

Packages #

Odin programs consist of packages. A package is a directory of Odin code files, all of which have the same package declaration at the top. Execution starts in the package’s main procedure.

import statement #

The following program imports the fmt and os packages from the core library collection.

package main

import "core:fmt"
import "core:os"

main :: proc() {
}

The core: prefix is used to state where the import is meant to look; this is called a library collection. If no prefix is present, the import will look relative to the current file.

Note: By convention, the package name is the same as the last element in the import path. "core:fmt" package comprises of files that begin with the statement package fmt. However, this is not enforced by the compiler, which means the default name for the import name will be determined by the last element in the import path if possible.

A different import name can be used over the default package name:

import "core:fmt"
import foo "core:fmt" // reference a package by a different name

Exported names #

All declarations in a package are public by default.

The private attribute can be applied to an entity to prevent it from being exported from a package.

@(private)
my_variable: int // cannot be accessed outside this package

You may also make an entity private to the file instead of the package.

@(private="file")
my_variable: int // cannot be accessed outside this file

@(private) is equivalent to @(private="package").

Authoring a package #

A package is a directory of Odin code files, all of which have the same package declaration at the top, e.g. package main. Each .odin file must have the same package name. A directory cannot contain more than 1 package.

Organizing packages #

Packages may be thematically organized by placing them in subdirectories of another package. For example: core:image/png and core:image/tga, as subdirectories of core:image. Nesting these packages is a helpful taxonomy. It does not imply a dependency: core:foo/bar does not need to import core:foo and reference anything from it.

Control flow statements #

for statement #

Odin has only one loop statement, the for loop.

Basic for loop #

A basic for loop has three components separated by semicolons:

• The initial statement: executed before the first iteration
• The condition expression: evaluated before every iteration
• The post statement: executed at the end of every iteration

The loop will stop executing when the condition is evaluated to false.

for i := 0; i < 10; i += 1 {
	fmt.println(i)
}

Note: Unlike other languages like C, there are no parentheses ( ) surrounding the three components. Braces { } or a do are always required.

for i := 0; i < 10; i += 1 { }
for i := 0; i < 10; i += 1 do single_statement()

The initial and post statements are optional:

i := 0
for ; i < 10; {
	i += 1
}

These semicolons can be dropped. This for loop is equivalent to C’s while loop:

i := 0
for i < 10 {
	i += 1
}

If the condition is omitted, this produces an infinite loop:

for {
}

Range-based for loop #

The basic for loop

for i := 0; i < 10; i += 1 {
	fmt.println(i)
}

can also be written

for i in 0..<10 {
	fmt.println(i)
}
// or
for i in 0..=9 {
	fmt.println(i)
}

where a..=b denotes a closed interval [a,b], i.e. the upper limit is inclusive, and a..<b denotes a half-open interval [a,b), i.e. the upper limit is exclusive.

Certain built-in types can be iterated over:

some_string := "Hello, 世界"
for character in some_string {
	fmt.println(character)
}

some_array := [3]int{1, 4, 9}
for value in some_array {
	fmt.println(value)
}

some_slice := []int{1, 4, 9}
for value in some_slice {
	fmt.println(value)
}

some_dynamic_array := [dynamic]int{1, 4, 9} // must be enabled with `#+feature dynamic-literals`
defer delete(some_dynamic_array)
for value in some_dynamic_array {
	fmt.println(value)
}

some_map := map[string]int{"A" = 1, "C" = 9, "B" = 4} // must be enabled with `#+feature dynamic-literals`
defer delete(some_map)
for key in some_map {
	fmt.println(key)
}

Alternatively a second index value can be added:

for character, index in some_string {
	fmt.println(index, character)
}
for value, index in some_array {
	fmt.println(index, value)
}
for value, index in some_slice {
	fmt.println(index, value)
}
for value, index in some_dynamic_array {
	fmt.println(index, value)
}
for key, value in some_map {
	fmt.println(key, value)
}

The iterated values are copies and cannot be written to.

When iterating a string, the characters will be runes and not bytes. for in assumes the string is encoded as UTF-8.

str: string = "Some text"
for character in str {
	assert(type_of(character) == rune)
	fmt.println(character)
}

You can iterate arrays and slices by-reference with the address operator:

for &value in some_array {
	value = something
}
for &value in some_slice {
	value = something
}
for &value in some_dynamic_array {
	value = something
}
// does not impact the second index value
for &value, index in some_dynamic_array {
	value = something
}

Map values can be iterated by-reference, but their keys cannot since map keys are immutable:

some_map := map[string]int{"A" = 1, "C" = 9, "B" = 4} // must be enabled with `#+feature dynamic-literals`
defer delete(some_map)

for key, &value in some_map {
	value += 1
}

fmt.println(some_map["A"]) // 2
fmt.println(some_map["C"]) // 10
fmt.println(some_map["B"]) // 5

Note: It is not possible to iterate a string in a by-reference manner as strings are immutable.

for reverse iteration #

The #reverse directive makes a range-based for loop iterate in reverse.

array := [?]int { 10, 20, 30, 40, 50 }

#reverse for x in array {
	fmt.println(x) // 50 40 30 20 10
}

for loop unrolling #

The #unroll directive takes a for loop and expands it at compile-time to the individual statements, repeated for as many times as the loop would normally iterate. This may result in performance improvements or provides the ability to repeat a set of instructions a limited number of times without explicitly writing each out in repetition.

Please note that #unroll may only be used with ranged for loops that have constant intervals known at compile-time.

x: [4]u8 = 0xFF
y: [4]u8 = 0x88
#unroll for i in 0..<len(x) {
	x[i] ~= y[i]
}

if statement #

Odin’s if statements do not need to be surrounded by parentheses ( ) but braces { } or do are required.

if x >= 0 {
	fmt.println("x is positive")
}

Like for, the if statement can start with an initial statement to execute before the condition. Variables declared by the initial statement are only in the scope of that if statement.

if x := foo(); x < 0 {
	fmt.println("x is negative")
}

Variables declared inside an if initial statement are also available to any of the else blocks:

if x := foo(); x < 0 {
	fmt.println("x is negative")
} else if x == 0 {
	fmt.println("x is zero")
} else {
	fmt.println("x is positive")
}

switch statement #

A switch statement is another way to write a sequence of if-else statements. In Odin, the default case is denoted as a case without any expression.

switch arch := ODIN_ARCH; arch {
case .i386, .wasm32, .arm32:
	fmt.println("32 bit")
case .amd64, .wasm64p32, .arm64, .riscv64:
	fmt.println("64 bit")
case .Unknown:
	fmt.println("Unknown architecture")
}

Odin’s switch is like the one in C or C++, except that Odin only runs the selected case. This means that a break statement is not needed at the end of each case. Another important difference is that the case values need not be integers nor constants.

To achieve a C-like fall through into the next case block, the keyword fallthrough can be used.

Switch cases are evaluated from top to bottom, stopping when a case succeeds. For example:

switch i {
case 0:
case foo():
}

foo() does not get called if i==0. If all the case values are constants, the compiler may optimize the switch statement into a jump table (like C).

A switch statement without a condition is the same as switch true. This can be used to write a clean and long if-else chain and have the ability to break if needed

switch {
case x < 0:
	fmt.println("x is negative")
case x == 0:
	fmt.println("x is zero")
case:
	fmt.println("x is positive")
}

A switch statement can also use ranges like a range-based loop:

switch c {
case 'A'..='Z', 'a'..='z', '0'..='9':
	fmt.println("c is alphanumeric")
}

switch x {
case 0..<10:
	fmt.println("units")
case 10..<13:
	fmt.println("pre-teens")
case 13..<20:
	fmt.println("teens")
case 20..<30:
	fmt.println("twenties")
}

#partial switch #

With enum values:

Foo :: enum {
	A,
	B,
	C,
	D,
}

f := Foo.A
switch f {
case .A: fmt.println("A")
case .B: fmt.println("B")
case .C: fmt.println("C")
case .D: fmt.println("D")
case:    fmt.println("?")
}

#partial switch f {
case .A: fmt.println("A")
case .D: fmt.println("D")
}

With union types (see Type switch statement)

Foo :: union {int, bool}
f: Foo = 123
switch _ in f {
case int:  fmt.println("int")
case bool: fmt.println("bool")
case:
}

#partial switch _ in f {
case bool: fmt.println("bool")
}

defer statement #

A defer statement defers the execution of a statement until the end of the scope it is in.

The following will print 4 then 234:

package main

import "core:fmt"

main :: proc() {
	x := 123
	defer fmt.println(x)
	{
		defer x = 4
		x = 2
	}
	fmt.println(x)

	x = 234
}

You can defer an entire block too:

{
	defer {
		foo()
		bar()
	}
	// This is equivalent to `defer { if cond { bar() } }` because the `if` is
	// a statement in its own right.
	defer if cond {
		bar()
	}
}

Defer statements are executed in the reverse order that they were declared:

defer fmt.println("1")
defer fmt.println("2")
defer fmt.println("3")

Will print 3, 2, and then 1.

A real world use case for defer may be something like the following:

f, err := os.open("my_file.txt")
if err != os.ERROR_NONE {
	// handle error
}
defer os.close(f)
// rest of code

In this case, it acts akin to an explicit C++ destructor however, the error handling is basic control flow.

It’s important to note that defer cannot be used to change a procedure’s named return values, as it runs after exit when the values have already been returned.

foo :: proc() -> (n: int) {
	defer {
		n = 456 // This won't affect `n`
	}
	n = 123
	return
}

Note: The defer construct in Odin differs from Go’s defer, which is function-exit and relies on a closure stack system.

when statement #

The when statement is almost identical to the if statement but with some differences:

• Each condition must be a constant expression as a when statement is evaluated at compile time.
• The statements within a branch do not create a new scope
• The compiler checks the semantics and code only for statements that belong to the first condition that is true
• An initial statement is not allowed in a when statement
• when statements are allowed at file scope

Example:

when ODIN_ARCH == .i386 {
	fmt.println("32 bit")
} else when ODIN_ARCH == .amd64 {
	fmt.println("64 bit")
} else {
	fmt.println("Unsupported architecture")
}

The when statement is very useful for writing platform specific code. This is akin to the #if construct in C’s preprocessor. However, in Odin, it is type checked.

See the Conditional compilation section for examples of built-in constants you can use with when statements.

Branch statements #

break statement #

A for loop, conditional, or a switch statement can be left prematurely with a break statement. It leaves the innermost construct, unless a label of a construct is given:

for cond {
	switch {
	case:
		if cond {
			break // break out of the `switch` statement
		}
	}

	break // break out of the `for` statement
}

loop: for cond1 {
	for cond2 {
		break loop // leaves both loops
	}
}

outer: if cond {
	ok := check_something()
	if !ok {
		break outer // label names are required with conditionals
	}
}

exit: {
    if true {
        break exit // works with labeled blocks too
    }
    fmt.println("This line will never print.")
}

continue statement #

As in many programming languages, a continue statement starts the next iteration of a loop prematurely:

for cond {
	if get_foo() {
		continue
	}
	fmt.println("Hellope")
}

fallthrough statement #

Odin’s switch is like the one in C or C++, except that Odin only runs the selected case. This means that a break statement is not needed at the end of each case. Another important difference is that the case values need not be integers nor constants.

fallthrough can be used to explicitly fall through into the next case block:

switch i {
case 0:
	foo()
	fallthrough
case 1:
	bar()
}

Procedures #

In Odin, a procedure is something that can do work, which some languages call functions or methods. A procedure literal in Odin is defined with the proc keyword:

fibonacci :: proc(n: int) -> int {
	switch {
	case n < 1:
		return 0
	case n == 1:
		return 1
	}
	return fibonacci(n-1) + fibonacci(n-2)
}

fmt.println(fibonacci(3)) // 2

For more information regarding value declarations in general, please see the Odin FAQ.

Parameters #

Procedures can take zero or many parameters. The following example is a basic procedure that multiplies two integers together:

multiply :: proc(x: int, y: int) -> int {
	return x * y
}
fmt.println(multiply(137, 432))

When two or more consecutive parameters share a type, you can omit the other types from previous names, like with variable declarations. In this example: x: int, y: int can be shortened to x, y: int, for example:

multiply :: proc(x, y: int) -> int {
	return x * y
}
fmt.println(multiply(137, 432))

Continuing the C family tradition, everything in Odin is passed by value (rather than by reference, e.g. FORTRAN, Java, etc). However, Odin differs from the C/C++ tradition in that all procedure parameters in Odin are immutable values. This allows for numerous optimizations with the Odin calling conventions ("odin" and "contextless") which would not be possible with the original C tradition of always passing a copy of the thing that has been passed.

Passing a pointer value makes a copy of the pointer, not the data it points to. Slices, dynamic arrays, and maps behave like pointers in this case (Internally they are structures that contain values, which include pointers, and the “structure” is passed by value).

Shadowing Parameters #

To mutate the procedure parameter (like in C), an explicit copy is required. This can be done through shadowing the variable declaration:

foo :: proc(x: int) {
	x := x // explicit mutation
	for x > 0 {
		fmt.println(x)
		x -= 1
	}
}

Variadic Parameters #

Procedures can be variadic, taking a varying number of arguments:

sum :: proc(nums: ..int) -> (result: int) {
	for n in nums {
		result += n
	}
	return
}
fmt.println(sum())              // 0
fmt.println(sum(1, 2))          // 3
fmt.println(sum(1, 2, 3, 4, 5)) // 15

odds := []int{1, 3, 5}
fmt.println(sum(..odds))        // 9, passing a slice as varargs

Multiple results #

A procedure in Odin can return any number of results. For example:

swap :: proc(x, y: int) -> (int, int) {
	return y, x
}
a, b := swap(1, 2)
fmt.println(a, b) // 2 1

Named results #

Return values in Odin may be named. If so, they are treated as variables defined at the top of the procedure, like input parameters. A return statement without arguments returns the named return value (or values). “Naked” return statements should only be used in short procedures as it reduces clarity when reading.

do_math :: proc(input: int) -> (x, y: int) {
	x = 2*input + 1
	y = 3*input / 5
	return x, y
}
do_math_with_naked_return :: proc(input: int) -> (x, y: int) {
	x = 2*input + 1
	y = 3*input / 5
	return // A "naked" return statement, as no values are explicitly specified.
}

Assigning default values to named results is also possible:

conditionally_blue :: proc(red: bool) -> (color := "blue") {
    if red {
        return "red"
    }
    return
}

Named arguments #

When calling a procedure, it is not clear in which order parameters might appear. Therefore, the arguments can be named, like a struct literal, to make it clear which argument a parameter is for:

create_window :: proc(title: string, x, y: int, width, height: int, monitor: ^Monitor) -> (^Window, Window_Error) {...}

window, err := create_window(title="Hellope Title", monitor=nil, width=854, height=480, x=0, y=0)

As of the dev-2023-07 release, mixing named and positional arguments is allowed. This is often useful when a procedure has a lot of arguments or you want to customize default values.

Positional arguments are not allowed after named arguments.

foo :: proc(value: int, name: string, x: bool, y: f32, z := 0) { }
foo(134, "hellope", x=true, y=4.5)

Default values #

The create_window procedure may be easier to use if default values are provided, which will be used if they are not specified:

create_window :: proc(title: string, x := 0, y := 0, width := 854, height := 480, monitor: ^Monitor = nil) -> (^Window, Window_Error) {...}

window1, err1 := create_window("Title1")
window2, err2 := create_window(title="Title1", width=640, height=360)

Default values are assigned at the start of the procedure call and can be overwritten. They may also be assigned to named results.

Note: These default values must be compile time known values, such as a constant value or nil (if the type supports it).

Explicit procedure overloading #

Unlike other languages, Odin provides the ability to explicitly overload procedures:

bool_to_string :: proc(b: bool) -> string {...}
int_to_string  :: proc(i: int)  -> string {...}

to_string :: proc{bool_to_string, int_to_string}

Rationale behind explicit overloading #

The design goals of Odin were explicitness and simplicity. Implicit procedure overloading complicates the scoping system. In C++, you cannot nest procedures within procedures, so all procedure look-ups are done at the global scope. In Odin, procedures can be nested within procedures and, as a result, determining which procedure should be used, in the case of implicit overloading, is complex.

Explicit overloading has many advantages:

• Explicitness of what is overloaded
• Able to refer to the specific procedure if needed
• Clear which scope the entity name belongs to
• Ability to specialize parametric polymorphic procedures if necessary, which have the same parameter but different bounds (see where clauses)

foo :: proc{
	foo_bar,
	foo_baz,
	foo_baz2,
	another_thing_entirely,
}

Basic types #

Odin’s basic types are:

bool b8 b16 b32 b64 // booleans

// integers
int  i8 i16 i32 i64 i128
uint u8 u16 u32 u64 u128 uintptr

// endian specific integers
i16le i32le i64le i128le u16le u32le u64le u128le // little endian
i16be i32be i64be i128be u16be u32be u64be u128be // big endian

f16 f32 f64 // floating point numbers

// endian specific floating point numbers
f16le f32le f64le // little endian
f16be f32be f64be // big endian

complex32 complex64 complex128 // complex numbers

quaternion64 quaternion128 quaternion256 // quaternion numbers

rune // signed 32 bit integer
	 // represents a Unicode code point
	 // is a distinct type to `i32`

// strings
string cstring

// raw pointer type
rawptr

// runtime type information specific type
typeid
any

The uintptr type is pointer sized, and the int, uint types are the “natural” register size, which is guaranteed to greater than or equal to the size of a pointer (i.e. size_of(uint) >= size_of(uintptr)). When you need an integer value, you should default to using int unless you have a specific reason to use a sized or unsigned integer type

Note: The Odin string type stores the pointer to the data and the length of the string. cstring is used to interface with foreign libraries written in/for C that use zero-terminated strings.

Zero values #

Variables declared without an explicit initial value are given their zero value.

The zero value is:

• 0 for numeric and rune types
• false for boolean types
• "" (the empty string) for strings
• nil for pointer, typeid, and any types.

The expression {} can be used for all types to act as a zero type. This is not recommended as it is not clear and if a type has a specific zero value shown above, please prefer that.

Type conversion #

The expression T(v) converts the value v to the type T.

i: int = 123
f: f64 = f64(i)
u: u32 = u32(f)

or with type inference:

i := 123
f := f64(i)
u := u32(f)

Unlike C, assignments between values of a different type require an explicit conversion.

Cast operator #

The cast operator can also be used to do the same thing:

i := 123
f := cast(f64)i
u := cast(u32)f

This is useful in some contexts but has the same semantic meaning.

Transmute operator #

The transmute operator is a bit cast conversion between two types of the same size:

f := f32(123)
u := transmute(u32)f

This is akin to doing the following pointer cast manipulations:

f := f32(123)
u := (^u32)(&f)^

However, transmute does not require taking the address of the value in question, which may not be possible for many expressions.

Untyped types #

In the Odin type system, certain expressions will have an “untyped” type. An untyped type can implicitly convert to a “typed” type.

I :: 42        // untyped integer, will implicitly convert to any numeric type (int, u32, f64, quaternion128 etc)
F :: 1.37      // untyped float,  will implicitly convert to any numeric type that can support fractional parts (f64, quaternion128 etc)
S :: "Hellope" // untyped string,  will implicitly convert to string and cstring
B :: true      // untyped boolean, will implicitly convert to bool, b8, b16, etc.

Auto cast operation #

The auto_cast operator automatically casts an expression to the destination’s type if possible:

x: f32 = 123
y: int = auto_cast x

Note: This operation is only recommended to be used for prototyping and quick tests. Please do not abuse it.

Built-in constants, values, and procedures #

There are a few built-in constants and values in Odin which have different uses:

false // untyped boolean constant equivalent to the expression 0!=0
true  // untyped boolean constant equivalent to the expression 0==0
nil   // untyped nil value used for certain values
---   // untyped undefined value used to explicitly not initialize a variable

--- is useful if you want to explicitly not initialize a variable with any default value:

x: int // initialized with its zero value
y: int = --- // uses uninitialized memory

This is the default behaviour in C, whilst the default behaviour in Odin is to zero the memory.

Note: --- is not a contract in that all memory is uninitialized, but that it may be. For example, as a side-effect of an implementation detail, padding in a struct must be zeroed to permit trivial bitwise memory comparison.

Built-in procedures #

For the full list of builtin-procedures, see the documentation for package builtin.

There are two kinds of built-in procedures in Odin:

• Compiler defined
• Core library defined

string type #

As previously mentioned the odin string type is just a rawptr + len.

The core:strings library was created to help dealing with string cloning, conversion of string<->cstring and other calls you find in standard libraries.

All procedures are documented and can be easily understood with code examples.

string iteration #

Iterating a string can be done in two ways - by runes or by bytes.

// by runes
x := "ABC"
for codepoint, index in x {
	fmt.println(index, codepoint)
	// 0 A
	// 1 B
	// 2 C
}

// by bytes - string length is in bytes
for index in 0..<len(x) {
	fmt.println(index, x[index])
	// 0 65
	// 1 66
	// 2 67
}

Iteration through runes is preferred since odin strings are UTF8. Most core library procedures will be addressed by *_byte if they do input/output an index in byte.

string format printing #

The core:fmt library supports printing strings from byte arrays in structs, when additional tag information is supplied.

Foo :: struct {
	a: [L]u8 `fmt:"s"`, // whole buffer is a string
	b: [N]u8 `fmt:"s,0"`, // 0 terminated string
	c: [M]u8 `fmt:"q,n"`, // string with length determined by n, and use %q rather than %s
	n: int `fmt:"-"`, // ignore this from formatting
}

cstring type #

The cstring type is a c-style string value, which is zero-terminated. It is equivalent to char const * in C. Its primary purpose is for easy interfacing with C. Please see the foreign system for more information.

A cstring is easily convertible to an Odin string. However, to convert a string to a cstring, it requires allocations if the value is not constant.

str:  string  = "Hellope"
cstr: cstring = "Hellope" // constant literal
cstr2 := string(cstr)     // O(n) conversion as it requires search from the zero-terminator
nstr  := len(str)  // O(1)
ncstr := len(cstr) // O(n)

string type conversions #

Here is a short list of possible type conversions - including whether they copy or alias. This is important to understand since Odin gives you the possibility to keep allocations to a low degree.

If some cases are missing please let us know.

Legend:

• copy = get a freshly allocated copy of the ‘from’ data
• alias = reuse the ‘from’ data, without allocation
• stream = get individual values from the string, without allocation
• st = the input string

From string to X #

From cstring to X #

From a string literal to X #

From []u8 to X #

From []rune to string #

From [^]u8 to cstring #

From [^]u8 and length int to string #

Operators #

Operators combine operands into expressions. For binary operations, operand types must be identical or implicitly convertible unless the operation involves shifts or untyped constants.

Arithmetic operators #

Unary:

+                           is 0 + x
-    negation               is 0 - x
~    bitwise complement     is m ~ x where m = "all bits set to 1" for unsigned x
                                     and m = -1 for signed x

Binary:

+       sum                        integers, enums, floats, complex values, quaternions, arrays of numeric types, matrices, constant strings
-       subtraction                integers, enums, floats, complex values, quaternions, arrays of numeric types, matrices
*       multiplication             integers, floats, complex values, quaternions, arrays of numeric types, matrices
/       division                   integers, floats, complex values, quaternions, arrays of numeric types
%       modulo (truncated)         integers
%%      remainder (floored)        integers

|       bitwise or                 integers, enums
~       bitwise xor                integers, enums
&       bitwise and                integers, enums
&~      bitwise and-not            integers, enums
<<      left shift                 integer << integer >= 0
>>      right shift                integer >> integer >= 0

Except for shift operations, if one operand is an untyped constant and the other operand is not, the constant is implicitly converted to the type of the other operand (if possible).

The right operand in a shift expression must have an unsigned integer type or be an untyped constant representable by a typed unsigned integer. If the left operand of a non-constant shift expression is an untyped constant, it is first implicitly converted to the type it would assume if the shift expression were replaced solely by the left operand alone (with type inference and hinting rules applied).

Comparison operators #

==      equal
!=      not equal
<       less
<=      less or equal
>       greater
>=      greater or equal
&&      short-circuiting logical and
||      short-circuiting logical or

In any comparison, the first operand must be assignable to the type of the second, or vice versa.

The equality operators == and != apply to operands that are comparable. The ordering operators <, <=, >, and >= apply to operands that are ordered. These terms and the result of the comparisons are defined as follows:

• Boolean values are comparable.
• Integers values are comparable and ordered.
• Floating-point values are comparable and ordered, defined by the IEEE-754 standard.
• Complex values are comparable.
• Quaternion values are comparable.
• Rune values are comparable and ordered.
• String values are comparable and ordered, lexically byte-wise.
• Matrix values are comparable.
• Pointer values are comparable and ordered.
• Multi-pointer values are comparable and ordered.
• Soa-pointer values are comparable.
• Enum values are comparable and ordered.
• Bit-set values are comparable.
• Struct values are comparable if all their fields are comparable.
• Union values are comparable if all their variants are comparable.
• Array and enumerated array values are comparable if values of the element type are comparable.
• typeid is comparable.
• Simd vectors are comparable.

Bit-set values use different logic compared to integers when comparison operators are used: please see the section of bit sets

Logical operators #

Logical operators apply to boolean values. The right operand is evaluated conditionally

&&      conditional AND    a && b  is "b if a else false"
||      conditional OR     a || b  is "true if a else b"
!       NOT                !a      is "not a"

Compound binary operator and assign #

Like many other languages, there is a shorthand for performing a binary operation and assigning the result to the first operand e.g. x = x + 5. All arithmetic and logical binary operators have this shorthand

+=       sum and assign                   a += b is a = a + b
-=       subtraction and assign           a -= b is a = a - b
*=       multiplication and assign        a *= b is a = a * b
/=       division and assign              a /= b is a = a / b
%=       modulo (truncated) and assign    a %= b is a = a % b
%%=      remainder (floored) and assign   a %%= b is a = a %% b

|=       bitwise or and assign            a |= b is a = a | b
~=       bitwise xor and assign           a ~= b is a = a ~ b
&=       bitwise and and assign           a &= b is a = a & b
&~=      bitwise and-not and assign       a &~= b is a = a &~ b
<<=      left shift and assign            a <<= b is a = a << b
>>=      right shift and assign           a >>= b is a = a >> b

&&=      conditional AND and assign       a &&= b is a = a && b
||=      conditional OR and assign        a ||= b is a = a || b

Address operator #

For an operand x of type T, the address operation &x generates a pointer of ^T to x. The operand must be addressable, meaning that either a variable, pointer indirection, or slice/dynamic array indexing operator; or a field selector of an addressable struct operand; or an array index operation of an addressable array; or a type assertion of an addressable union or any; or a compound literal value.

For an operand x of pointer type ^T, the pointer indirection x^ denotes the variable of type T pointed to by x. If x is an invalid address, such as nil, an attempt to evaluate x^ will result in platform specific behaviour - on most platforms this will be a segmentation fault.

&x
&a[foo(123)]
&Foo{1, 2}

p^
pproc(a)^

x: ^int = nil
x^      // causes a runtime panic

Ternary Operators #

x if cond else y    // ternary runtime conditional expression
x when cond else y  // ternary compile-time conditional expression
cond ? x : y        // ternary runtime conditional expression, equivalent to "x if cond else y"

Other operators #

• or_else
  • see section on or_else
• or_return
  • see section on or_return
• or_continue
  • see section on or_continue
• or_break
  • see section on or_break
• in - set membership (e in A, A contains element e)
  • Used for bit_set types and map types
• not_in - not set membership (e not_in A, A does not contain e)
  • Used for bit_set types and map types
• ..= - inclusive range
• ..< - half open range

The range operations ..= and ..< are only possible within certain contexts:

for x in a..<b {}
for x in a..=b {}

switch x {
case a..<b:
case c..=d:
}

bit_set[a..<b]
bit_set[a..=b]

foo := [?]int{0..=3 = 1} // initialises as: [1, 1, 1, 1]
bar := [?]int{0 = 0, 1..<3 = 1} // initialises as: [0, 1, 1]

in and not_in are not allowed in within a for loop condition without ambiguity:

for x in y {}    // range loop
for (x in y) {}  // condition-only for-loop (while-loop in some other languages)

Operator precedence #

Unary operators have the highest precedence.

There are seven precedence levels for binary (and ternary) operators.

Precedence    Operator
     7           *   /   %   %%   &   &~  <<   >>
     6           +   -   |   ~    in  not_in
     5           ==  !=  <   >    <=  >=
     4           &&
     3           ||
     2           ..=    ..<
     1           or_else     ?    if  when

Binary operators of the same precedence associate from left to right. For instance x / y * z is the same as (x / y) * z.

Integer operators #

For two integers values x and y, the integer quotient q = x/y and remainder r = x%y satisfies the following relationships:

x = q*y + r   and |r| < |y|

with x/y truncated towards zero (truncated division).

For two integers values x and y, the integer quotient q = x/y and remainder r = x%%y satisfies the following relationships:

r = x - y*floor(x/y)

The exception to these rules are when the dividend x is the most non-negative value for the integer type of x, and the quotient q = x/-1 is equal to x (and r or m = 0) due to two’s complement integer overflow.

If the divisor is a constant, it must not be zero. If the divisor is zero at runtime, a runtime panic occurs.

The shift operators shift the left operand by the shift count specified by the right operand, which must be non-negative. The shift operators implement arithmetic shifts if the left operand is a signed integer and logical shifts if the left operand is an unsigned integer. There is not an upper limit on the shift count. Shifts behave as if the left operand is shifted n times by 1 for a shift count of n. Therefore, x<<1 is the same as x*2 and x>>1 is the same as x/2 but truncated towards negative infinity.

// These are equivalent:
x << y
x << y if y < 8*size_of(x) else 0

x >> y
x >> y if y < 8*size_of(x) else 0

Integer overflow #

For unsigned integers, the operations +, -, *, and << are computed modulo 2ⁿ, where n is the bit width of the unsigned integer’s type. In a sense, these unsigned integer operations discard the high bits upon overflow, and programs may rely on “wrap around”.

For signed integers, the operations +, -, *, /, and << may legally overflow and the resulting value exists and is deterministically defined by the signed integer representation. Overflow does not cause a runtime panic. A compiler may not optimize code under the assumption that overflow does not occur. For instance, x < x+1 may not be assumed to be always true.

Floating-point operators #

For floating-point, complex numbers, quaternions, and other floating-point embedded types:

• +x is the same as x
• -x is the negation of x

The result of a floating-point related division by zero is not specified beyond the IEEE-754 standard; a runtime panic will occur.

An implementation may combine multiple floating-point operations into a single fused operation, and produce a result that differs from the value obtained by executing and rounding the instructions individually.

Advanced types #

Type alias #

You can alias a named type with another name:

My_Int :: int
#assert(My_Int == int)

Distinct types #

A distinct type allows for the creation of a new type with the same underlying semantics.

My_Int :: distinct int
#assert(My_Int != int)

Aggregate types (struct, enum, union) will always be distinct even when named.

Foo :: struct {}
#assert(Foo != struct{})

Fixed arrays #

An array is a simplified fixed length container. Each element in an array has the same type. An array’s index can be any integer, character, or enumeration type.

An array can be constructed like the following:

x := [5]int{1, 2, 3, 4, 5}
for i in 0..=4 {
	fmt.println(x[i])
}

Fixed arrays are equivalent to a struct with a field for each element. They are just a number of values in a row in memory.

The notation x[i] is used to access the i-th element of x; and 0-index based (like C).

The above array can also be constructed with a question mark (?) to automatically infer its length:

x := [?]int{1, 2, 3, 4, 5}

Construct an array with designated initializers:

favorite_animals := [?]string{
	// Assign by index
	0 = "Raven",
	1 = "Zebra",
	2 = "Spider",
	// Assign by range of indices
	3..=5 = "Frog",
	6..<8 = "Cat"
}

The built-in len proc returns the array’s length.

x: [5]int
#assert(len(x) == 5)

Array access is bounds checked by default, both at compile-time (with constant indices) and at runtime. This can be disabled and enabled at a per block level with the #no_bounds_check and #bounds_check directives, respectively:

#no_bounds_check {
	x[n] = 123 // n could be in or out of range of valid indices
}

#no_bounds_check can be used to improve performance when the bounds are known to not exceed.

Array programming #

Odin’s fixed length arrays support array programming.

Example:

Vector3 :: [3]f32
a := Vector3{1, 4, 9}
b := Vector3{2, 4, 8}
c := a + b  // {3, 8, 17}
d := a * b  // {2, 16, 72}
e := c != d // true

n.b. Odin also supports matrix types.

Swizzle Operations #

a := [3]f32{10, 20, 30}
b := swizzle(a, 2, 1, 0)
assert(b == [3]f32{30, 20, 10})

c := swizzle(a, 0, 0)
assert(c == [2]f32{10, 10})
assert(c == 10) // assert all elements == 10

Builtin implicit swizzle fields are available on any array with length <= 4 as xyzw and rgba.

Vector3 :: distinct [3]f32
a := Vector3{1, 2, 3}
b := Vector3{5, 6, 7}
c := (a * b)/2 + 1
d := c.x + c.y + c.z
fmt.printf("%.1f\n", d) // 22.0

cross :: proc(a, b: Vector3) -> Vector3 {
	i := swizzle(a, 1, 2, 0) * swizzle(b, 2, 0, 1)
	j := swizzle(a, 2, 0, 1) * swizzle(b, 1, 2, 0)
	return i - j
}

cross_shorter :: proc(a, b: Vector3) -> Vector3 {
	i := a.yzx * b.zxy
	j := a.zxy * b.yzx
	return i - j
}

blah :: proc(a: Vector3) -> f32 {
	return a.x + a.y + a.z
}

x := cross(a, b)
fmt.println(x)
fmt.println(blah(x))

Slices #

Slices look similar to arrays however, their length is not known at compile time. The type []T is a slice with elements of type T. In practice, slices are much more common than arrays.

A slice is formed by specifying two indices, a low and high bound, separated by a colon:

a[low : high]

This selects a half-open range which includes the lower element, but excludes the higher element.

fibonaccis := [6]int{0, 1, 1, 2, 3, 5}
s: []int = fibonaccis[1:4] // creates a slice which includes elements 1 through 3
fmt.println(s) // 1, 1, 2

Slices are like references to arrays; they do not store any data, rather they describe a section, or slice, of underlying data.

Internally, a slice stores a pointer to the data and an integer to store the length of the slice.

The built-in len proc returns the slice’s length.

x: []int = ...
length_of_x := len(x)

Slice literals #

A slice literal is like an array literal without the length. This is an array literal:

[3]int{1, 6, 3}

This is a slice literal which creates the same array as above, and then creates a slice that references it:

[]int{1, 6, 3}

Slice shorthand #

For the array:

a: [6]int

these slice expressions are equivalent:

a[0:6]
a[:6]
a[0:]
a[:]

When grabbing a chunk of a slice:

a[offset:offset+length]

can also be written:

a[offset:][:length]

Nil slices #

The zero value of a slice is nil. A nil slice has a length of 0 and does not point to any underlying memory. Slices can be compared against nil and nothing else.

s: []int
if s == nil {
	fmt.println("s is nil!")
}

Sort slices #

A slice literal can be sorted in ascending order as follows:

s := []int{1, 6, 3, 5 ,7, 3, 0}
slice.sort(s)

or in descending order

r := []int{1, 6, 3, 5 ,7, 3, 0}
slice.reverse_sort(r)

Dynamic arrays #

Dynamic arrays are similar to slices, but their lengths may change during runtime. Dynamic arrays are resizeable and they are allocated using the current context’s allocator.

x: [dynamic]int

Along with the built-in proc len, dynamic arrays also have cap which can used to determine the dynamic array’s current underlying capacity.

Appending to a dynamic array #

It is common to append new elements to a dynamic array; this can be done using the built-in append proc.

x: [dynamic]int
append(&x, 123)
append(&x, 4, 1, 74, 3) // append multiple values at once

y: [dynamic]int
append(&y, ..x[:]) // append a slice

Inject / Assign to a dynamic array #

Injecting into a specific index can be done with inject. It will move other elements upwards when inserted below other elements.

Assign a type at a specific index can be done with assign. It’s the same as doing x[index] = 10.

Both of these procedures will resize the dynamic array len to the wanted index. This can be seen in the example below.

x := make([dynamic]int, 0, 16)
inject_at(&x, 0, 10)
inject_at(&x, 3, 10) // resizes till length
fmt.eprintln(x[:], len(x), cap(x)) // [10, 0, 0, 10] 4 16
assign_at(&x, 3, 20)
assign_at(&x, 4, 30)
fmt.eprintln(x[:], len(x), cap(x)) // [10, 0, 0, 20, 30] 5, 16
assign_at(&x, 5, 40, 50, 60)
fmt.eprintln(x[:], len(x), cap(x)) // [10, 0, 0, 20, 30, 40, 50, 60] 8 16

Removing from a dynamic array #

Removing from a dynamic array can be done in several ways using the built-in procedures:

• pop pops the last element of the array
• unordered_remove removes an element at a specific index. Unordered means it is O(1), since it swaps the last element to the removed location - making the array be sorted differently.
• ordered_remove removes an element at a specific index. Ordered means it will move all elements after the index downwards with a copy - ensuring elements remain in the same order.

x: [dynamic]int
append(&x, 1, 2, 3, 4, 5) // [1, 2, 3, 4, 5]
pop(&x) // [1, 2, 3, 4]
ordered_remove(&x, 0) // [2, 3, 4]
unordered_remove(&x, 0) // [4, 3]

Other variants can be found in the built-in procedures documentation.

Slice & Sort a dynamic array #

Although dynamic arrays and slices are different concepts, dynamic arrays can be ‘sliced’ and sorted as follows:

s: [dynamic]int
append(&s, 1, 6, 3, 5, 7, 3, 0) // [1, 6, 3, 5, 7, 3, 0]
slice.sort(s[:]) // [0, 1, 3, 3, 5, 6, 7]

Making and deleting slices and dynamic arrays #

Slices and dynamic arrays can be explicitly allocated with the built-in make proc.

a := make([]int, 6)           // len(a) == 6
b := make([dynamic]int, 6)    // len(b) == 6, cap(b) == 6
c := make([dynamic]int, 0, 6) // len(c) == 0, cap(c) == 6
d := []int{1, 2, 3}           // a slice literal, for comparison

// with an explicit allocator:
e := make([]int, 6, context.allocator)
f := make([dynamic]int, 0, 6, context.allocator)

Slices and dynamic arrays can be deleted with the built-in delete proc.

delete(a)
delete(b)
delete(c)
// delete(d)                  // no need to clean up slice literals
delete(e)                     // slices are always deleted from context.allocator
delete(f)                     // dynamic arrays remember their allocator

Note: There is no automatic memory management in Odin.

Clearing a dynamic array #

Instead of deleting the array you often want to simply clear the dynamic array. This will set the length len() to be 0, while the capacity cap remains the same.

x: [dynamic]int
append(&x, 1, 2, 3, 4, 5) // [1, 2, 3, 4, 5]
fmt.println(len(x)) // 5
clear(&x) // []
fmt.println(len(x)) // 0

Resize / Reserve with a dynamic array #

Often enough we also want to resize or reserve a specific amount for a dynamic array. It’s important to understand the difference between the two operations.

• resize will try to resize memory of a passed dynamic array to the requested element count (setting the len, and possibly cap).
• reserve will try to reserve memory of a passed dynamic array to the requested element count (setting the cap).
• shrink will shrink the capacity of a dynamic array down to the current length, or the given capacity.

x: [dynamic]int
fmt.println(len(x), cap(x)) // 0 0
append(&x, 1, 2, 3) // [1, 2, 3]
fmt.println(len(x), cap(x)) // 3 8
resize(&x, 5) 
fmt.println(x[:]) // [1, 2, 3, 0, 0] other values are zero'd memory
fmt.println(len(x), cap(x)) // 5 8
reserve(&x, 32)
fmt.println(len(x), cap(x)) // 5 32
shrink(&x)
fmt.println(len(x), cap(x)) // 5 5

Small_Array(N, T) container dynamic array #

The core library also contains a container dynamic array Small_Array(N, T). It implements most dynamic array procedures while being able to remain on the stack.

Short Example:

import sa "core:container/small_array"

x: sa.Small_Array(8, int)
fmt.println(sa.len(x), sa.cap(x)) // 0 8
sa.append(&x, 1, 2, 3) 
fmt.println(sa.len(x), sa.cap(x)) // 3 8
fmt.println(sa.slice(&x)) // [1, 2, 3]

Enumerations #

Enumeration types define a new type whose values consist of the ones specified. The values are ordered, for example:

Direction :: enum{North, East, South, West}

The following holds:

int(Direction.North) == 0
int(Direction.East)  == 1
int(Direction.South) == 2
int(Direction.West)  == 3

Enum fields can be assigned an explicit value:

Foo :: enum {
	A,
	B = 4, // Holes are valid
	C = 7,
	D = 1337,
}

If an enumeration requires a specific size, a backing integer type can be specified. By default, int is used as the backing type for an enumeration.

Foo :: enum u8 {A, B, C} // Foo will only be 8 bits

Implicit Selector Expression #

An implicit selector expression is an abbreviated way to access a member of an enumeration, in a context where type inference can determine the implied type. It has the following form:

.member_name

For example:

Foo :: enum{A, B, C}
f: Foo
f = Foo.A
f = .A

switch f {
case .A:
	fmt.println("foo")
case .B:
	fmt.println("bar")
case .C:
	fmt.println("baz")
}

using can also be used with an enumeration to bring the fields into the current scope:

main :: proc() {
	Foo :: enum {A, B, C}
	using Foo
	a := A

	
}

Note: Implicit selector expression is preferred to using an enumeration as using does pollute the current scope.

Iterating an Enumeration #

Enums can be trivially for looped in odin. This way we can loop through the entire enum and do things like printing or inserting into an Enumerated Array.

Direction :: enum{North, East, South, West}

for direction, index in Direction {
	fmt.println(index, direction) 
	// 0 North
	// 1 East
	// 2 South
	// 3 West
}

Enumerated Array #

Enumerated Arrays allow the use of an Enum to be used as indices to a fixed array.

We’ll extend the Direction enum used previously to add direction vectors.

Direction :: enum{North, East, South, West}

Direction_Vectors :: [Direction][2]int {
	.North = {  0, -1 },
	.East = { +1,  0 },
	.South = {  0, +1 },
	.West = { -1,  0 },
}

assert(Direction_Vectors[.North] == { 0, -1 })
assert(Direction_Vectors[.East] == { 1, 0 })
assert(Direction_Vectors[cast(Direction) 2] == { 0, 1 })

The #partial directive can be used to initialize an enumerated array partially.

The #sparse directive can be used to initialize an enumerated array with an Enum which does not have contiguous values.

arr: [enum {A, B, C}]int
arr = #partial { // without partial the compiler would complain
	.A = 42,
}
fmt.println(arr) // [.A = 42, .B = 0, .C = 0]

Bit sets #

The bit_set type models the mathematical notion of a set. A bit_set’s element type can be either an enumeration or a range:

Direction :: enum{North, East, South, West}

Direction_Set :: bit_set[Direction]

Char_Set :: bit_set['A'..='Z']

Number_Set :: bit_set[0..<10] // bit_set[0..=9]

Bit sets are implemented as bit vectors internally for high performance. The zero value of a bit set is either nil or {}.

x: Char_Set
x = {'A', 'B', 'Y'}
y: Direction_Set
y = {.North, .West}

Bit sets support the following operations:

• A + B - union of two sets (equivalent to A | B)
• A - B - difference of two sets (A without B’s elements) (equivalent to A &~ B)
• A & B - intersection of two sets
• A | B - union of two sets (equivalent to A + B)
• A &~ B - difference of two sets (A without B’s elements) (equivalent to A - B)
• A ~ B - symmetric difference (Elements that are in A and B but not both)
• A == B - set equality
• A != B - set inequality
• A <= B - subset relation (A is a subset of B or equal to B)
• A < B - strict subset relation (A is a proper subset of B)
• A >= B - superset relation (A is a superset of B or equal to B)
• A > B - strict superset relation (A is a proper superset of B)
• e in A - set membership (A contains element e)
• e not_in A - not set membership (A does not contain element e)

Bit sets are often used to denote flags. This is much cleaner than defining integer constants that need to be bitwise or-ed together.

If a bit set requires a specific size, the underlying integer type can be specified:

Char_Set :: bit_set['A'..='Z'; u64]
#assert(size_of(Char_Set) == size_of(u64))

To get the number of elements set, its cardinality, of a bit_set, use the built-in card procedure:

x: Direction_Set
x = {.North, .West}
count := card(x)
assert(count == 2)

Pointers #

Odin has pointers. A pointer is a memory address of a value. The type ^T is a pointer to a T value. Its zero value is nil.

p: ^int

The & operator takes the address of its operand (if possible):

i := 123
p := &i

The ^ operator dereferences the pointer’s underlying value:

fmt.println(p^) // read `i` through the pointer `p`
p^ = 1337       // write `i` through the pointer `p`

Note: C programmers may be used to using * to denote pointers. In Odin, the ^ syntax is borrowed from Pascal. This is to keep the convention of the type on the left and its usage on the right:

p: ^int // ^ on the left
x := p^ // ^ on the right

Note: Unlike C, Odin has no pointer arithmetic. If you need a form of pointer arithmetic, please use the ptr_offset and ptr_sub procedures in the "core:mem" package.

Structs #

A struct is a record type in Odin. It is a collection of fields. Struct fields are accessed by using a dot:

Vector2 :: struct {
	x: f32,
	y: f32,
}
v := Vector2{1, 2}
v.x = 4
fmt.println(v.x)

Struct fields can be accessed through a struct pointer:

v := Vector2{1, 2}
p := &v
p.x = 1335
fmt.println(v)

We could write p^.x, however, it is nice to not have to explicitly dereference the pointer. This is very useful when refactoring code to use a pointer rather than a value, and vice versa.

Struct literals #

A struct literal can be denoted by providing the struct’s type followed by {}. A struct literal must either provide all the arguments or none:

Vector3 :: struct {
	x, y, z: f32,
}
v: Vector3
v = Vector3{} // Zero value
v = Vector3{1, 4, 9}

You can list just a subset of the fields if you specify the field by name (the order of the named fields does not matter):

v := Vector3{z=1, y=2}
assert(v.x == 0)
assert(v.y == 2)
assert(v.z == 1)

Structs can be nested by defining a field as a struct.

Foo :: struct {
	a, b, c: int,
	
	bar_1: struct {
		x, y, z: int,
	},

	bar_2: struct {
		x, y, z: int,
	},

	_: struct {
		x, y, z: int,
	},
}

Struct directives #

Structs can be annotated with different memory layout and alignment requirements:

struct #align(4) {...} // align to 4 bytes
struct #packed {...} // remove padding between fields
struct #raw_union {...} // all fields share the same offset (0). This is the same as C's union

Struct field tags #

Struct fields can be tagged with a string literal to attach meta-information which can be used with runtime-type information. Usually this is used to provide transactional information info on how a struct field is encoded to or decoded from another format, but you can store whatever you want within the string literal

User :: struct {
	flag: bool, // untagged field
	age:  int    "custom whatever information",
	name: string `json:"username" xml:"user-name" fmt:"q"`, // `core:reflect` layout
}

Within Odin’s core library, the standard convention is to store a key that denotes the package and then a subsequence "value". For example, json keys are processed and used by core:encoding/json package, fmt keys are processed by core:fmt.

If multiple information is to be passed in the "value", usually it is specified by separating it with a common (,), e.g.

name: string `json:"username,omitempty",

n.b. Field tags also exist for bit_field record types.

Unions #

A union in Odin is a discriminated union, also known as a tagged union or sum type. The zero value of a union is nil.

Value :: union {
	bool,
	i32,
	f32,
	string,
}
v: Value
v = "Hellope"

// type assert that `v` is a `string` and panic otherwise
s1 := v.(string)

// type assert but with an explicit boolean check. This will not panic
s2, ok := v.(string)

Type switch statement #

A type switch is a construct that allows several type assertions in series. A type switch is like a regular switch statement, but the cases are types (not values). For a union, the only case types allowed are that of the union.

value: Value = ...
switch v in value {
case string:
	#assert(type_of(v) == string)

case bool:
	#assert(type_of(v) == bool)

case i32, f32:
	// This case allows for multiple types, therefore we cannot know which type to use
	// `v` remains the original union value
	#assert(type_of(v) == Value)
case:
	// Default case
	// In this case, it is `nil`
}

Union tags #

Applying the #no_nil tag to a union type states it does not have a nil value. Unions with #no_nil must have at least two variants and the first variant is its default type:

Value :: union #no_nil {bool, string}
v: Value
_, ok := v.(bool)
assert(ok)

The #shared_nil tag normalizes each variant’s nil value into nil on assignment. If you assign nil or zero values to a union with #shared_nil the union will be nil. Unions with #shared_nil require all variants to have a nil value.

Error :: union #shared_nil {
	File_Error,
	Memory_Error,
}

File_Error :: enum {
	None = 0,
	File_Not_Found,
	Cannot_Open_File,
}

Memory_Error :: enum {
	None = 0,
	Allocation_Failed,
	Resize_Failed,
}

shared_nil_example :: proc() {
	an_error: Error
	an_error = File_Error.None

	assert(an_error == nil)
}

Unions also have the #align tag, like structures:

union #align(4) {...} // align to 4 bytes

Maps #

A map maps keys to values. The zero value of a map is nil. A nil map has no keys. The built-in make proc returns an initialized map using the current context, and delete can be used to delete a map.

m := make(map[string]int)
defer delete(m)
m["Bob"] = 2
fmt.println(m["Bob"])

To insert or update an element of a map:

m[key] = elem

To retrieve an element:

elem = m[key]

To remove an element:

delete_key(&m, key)

If an element of a key does not exist, the zero value of the element will be returned. Checking to see if an element exists can be done in two ways:

elem, ok := m[key] // `ok` is true if the element for that key exists

or

ok := key in m // `ok` is true if the element for that key exists

The first approach is called the “comma ok idiom”.

You can also initialize maps with map literals:

m := map[string]int{
	"Bob" = 2,
	"Chloe" = 5,
}

Note: To enable compound literals for maps, #+feature dynamic-literals must be enabled per file. This is because dynamic literals will use the current context.allocator and thus implicitly allocate. The opt-in feature is there in order for Odin to not implicitly allocate by default and not give the user any surprises.

Modifying existing map slots needs to be done in two steps. However assigning to a struct field is prohibited.

Test :: struct {
	x: int,
	y: int,
}

m := map[string]Test{
	"Bob" = { 0, 0 },
	"Chloe" = { 1, 1 },
}

value, ok := &m["Bob"]
if ok {
	value^ = { 2, 2 }
}

fmt.println(m["Bob"]) // { 2, 2 }
m["Bob"] = { 3, 3 }
fmt.println(m["Bob"]) // { 3, 3 }
m["Chloe"].x = 0 // PROHIBITED

Map Container Calls #

The built-in map also supports all the standard container calls that can be found with the dynamic array.

Short:

• len(some_map) returns the amount of slots used up
• cap(some_map) returns the capacity of the map - the map will reallocate when exceeded
• clear(&some_map) clears the entire map - dynamically allocated content needs to be freed manually
• reserve(&some_map, capacity) reserves the requested element count
• shrink(&some_map) shrink the capacity down to the current length

Bit Fields #

A bit_field is a record type in Odin that is akin to a bit-packed struct. Note: bit_field is not equivalent to bit_set as it has different semantics and use cases. bit_field fields are accessed by using a dot:

Foo :: bit_field u16 { // backing type must be an integer or array of integers
    x: i32     | 3, // signed integers will be signed extended on use
    y: u16     | 2 + 3, // general expressions
    z: My_Enum | foo.SOME_CONSTANT, // ability to define the bit-width elsewhere
    w: bool    | 2 when foo.SOME_CONSTANT > 10 else 1,
}

v := Foo{}
v.x = 3 // truncates the value to fit into 3 bits
fmt.println(v.x) // accessing will convert `v.x` to an `i32` and do an appropriate sign extension

A bit_field is different from a struct in that you must specify the backing type. This backing type must be an integer or a fixed-length array of integers. This is useful if there needs to be a specific alignment or access pattern for the record.

Foo :: bit_field u32 {...}
Foo :: bit_field [4]u8 {...}

Notes:

• If all of the fields in a bit_field are 1-bit in size and are a boolean, please consider using a bit_set instead.
• Odin’s bit_field and C’s bit-fields might not be compatible
  • Odin’s bit_fields have a well defined layout (Least-Significant-Bit)
  • C’s bit fields on structs are undefined and are not portable across targets and compilers

Procedure type #

A procedure type is internally a pointer to a procedure in memory. nil is the zero value a procedure type.

Examples:

proc(x: int) -> bool
proc(c: proc(x: int) -> bool) -> (i32, f32)

Or you can assign them to a variable:

Callback :: proc() -> int 
a: Callback // nil 
assert(a == nil)
a = proc() -> int { return 0 }
fmt.println(a()) // 0
a = proc() -> int { return 100 }
fmt.println(a()) // 100

Calling conventions #

Odin supports the following calling conventions:

• odin - default convention used for an Odin proc. It is the same as cdecl but passes an implicit context pointer on each call. (Note: This is subject to change)
• contextless - This is the same as odin but without the implicit context pointer.
• stdcall or std – This is the stdcall convention as specified by Microsoft.
• cdecl or c – This is the default calling convention generated of a procedure in C.
• fastcall or fast - This is a compiler dependent calling convention.
• none - This is a compiler dependent calling convention which will do nothing to parameters.

Most calling conventions exist only to interface with foreign Windows code.

The default calling convention is odin, unless it is within a foreign block, where it is then cdecl.

A procedure type with a different calling convention can be declared like the following:

proc "c" (n: i32, data: rawptr)
proc "contextless" (s: []int)

Procedure types are only compatible with the procedures that have the same calling convention and parameter types.

When binding to C libraries you’ll often end up using proc "c" and also set the current context. For this you’ll need to explicitly set the context.

typeid type #

A typeid is a unique identifier for an Odin type. This construct is used by the any type to denote what the underlying data’s type is.

a := typeid_of(bool)
i: int = 123
b := typeid_of(type_of(i))

A typeid can be mapped to relevant type information which can be used in applications such as printing types and editing data:

import "base:runtime"

main :: proc() {
	u := u8(123)
	id := typeid_of(type_of(u))
	info: ^runtime.Type_Info
	info = type_info_of(id)
}

any Type #

An any type can reference any data type. Internally it contains a pointer to the underlying data and its relevant typeid. This is a very useful construct in order to have a runtime type safe printing procedure.

Note: The any value is only valid for as long as the underlying data is still valid. Passing a literal to an any will allocate the literal in the current stack frame.

Note: It is highly recommended that you do not use this unless you know what you are doing. Its primary use is for printing procedures.

Multi Pointers #

Multi-Pointers in Odin are a way to describe foreign (C-like) pointers which act like arrays (pointers that map to multiple items). The type [^]T is a multi-pointer to T value(s). Its zero value is nil.

p: [^]int

What multi-pointers support:

• Index (without any bounds checking)
• Slicing (bounds checking on if both the low and high operands are given)
• Implicit conversions between ^T and [^]T
• Implicit conversion to rawptr (like all pointers)

What multi-pointers DO NOT SUPPORT:

• Dereferencing (which makes it closer to a slim-slice than a pointer)

The main purpose of this type is to aid with foreign code and act as a way to auto-document functionality and allow for easier transition to Odin code, especially converting pointers into slices.

The following are the rules for indexing and slicing for multi-pointers, and what type they produce depending on the operands given:

x: [^]T = ...

x[i]   -> T
x[:]   -> [^]T
x[i:]  -> [^]T
x[:n]  -> []T
x[i:n] -> []T

Interacting with Multi-Pointers is easiest using the builtin raw_data() call which can return a Multi-Pointer.

a: [^]int
fmt.println(a) // <nil>
b := [?]int { 10, 20, 30 }
a = raw_data(b[:]) 
fmt.println(a, a[1], b) // 0x7FFCBE9FE688 20 [10, 20, 30]

Note: The name of multi-pointers may be subject to change.

SOA Data Types #

Array of Structures (AoS), Structure of Arrays (SoA), and Array of Structures of Arrays (AoSoA) refer to differing ways to arrange a sequence of data records in memory, with regard to interleaving. These are of interest in SIMD and SIMT programming.

SOA Struct Arrays #

Vector3 :: struct {x, y, z: f32}

N :: 2
v_aos: [N]Vector3
v_aos[0].x = 1
v_aos[0].y = 4
v_aos[0].z = 9

fmt.println(len(v_aos))
fmt.println(v_aos[0])
fmt.println(v_aos[0].x)
fmt.println(&v_aos[0].x)

v_aos[1] = {0, 3, 4}
v_aos[1].x = 2
fmt.println(v_aos[1])
fmt.println(v_aos)

v_soa: #soa[N]Vector3

v_soa[0].x = 1
v_soa[0].y = 4
v_soa[0].z = 9


// Same syntax as AOS and treat as if it was an array
fmt.println(len(v_soa))
fmt.println(v_soa[0])
fmt.println(v_soa[0].x)
fmt.println(&v_soa[0].x)
v_soa[1] = {0, 3, 4}
v_soa[1].x = 2
fmt.println(v_soa[1])

// Can use SOA syntax if necessary
v_soa.x[0] = 1
v_soa.y[0] = 4
v_soa.z[0] = 9
fmt.println(v_soa.x[0])

// Same pointer addresses with both syntaxes
assert(&v_soa[0].x == &v_soa.x[0])


// Same fmt printing
fmt.println(v_aos)
fmt.println(v_soa)

Works with arrays of length <= 4 which have the implicit fields xyzw/rgba

Vector3 :: distinct [3]f32

N :: 2
v_aos: [N]Vector3
v_aos[0].x = 1
v_aos[0].y = 4
v_aos[0].z = 9

v_soa: #soa[N]Vector3

v_soa[0].x = 1
v_soa[0].y = 4
v_soa[0].z = 9

SOA Struct Slices and Dynamic Arrays #

Fixed-length SOA types can be sliced to produce SOA slices.

Vector3 :: struct {x: i8, y: i16, z: f32}

N :: 3
v: #soa[N]Vector3
v[0].x = 1
v[0].y = 4
v[0].z = 9

s: #soa[]Vector3
s = v[:]
assert(len(s) == N)
fmt.println(s)
fmt.println(s[0].x)

a := s[1:2]
assert(len(a) == 1)
fmt.println(a)

To be complete with SOA slices, Odin also supports SOA dynamic arrays.

d: #soa[dynamic]Vector3

append_soa(&d, Vector3{1, 2, 3}, Vector3{4, 5, 9}, Vector3{-4, -4, 3})
fmt.println(d)
fmt.println(len(d))
fmt.println(cap(d))
fmt.println(d[:])

soa_zip and soa_unzip #

SOA is not just useful for high performance scenarios but also for everyday tasks which are normally only achievable in higher level languages. soa_zip is a built-in procedure which allows the user to treat multiple slices as if they are part of the same data structures, utilizing the power of SOA.

x := []i32{1, 3, 9}
y := []f32{2, 4, 16}
z := []b32{true, false, true}

// produce an #soa slice with the normal slices passed
s := soa_zip(a=x, b=y, c=z)

// iterate over the #soa slice
for v, i in s {
	fmt.println(v, i) // exactly the same as s[i]
	// NOTE: `v` is NOT a temporary value but has a specialized addressing mode
	// which means that when accessing v.a etc, it does the correct transformation
	// internally:
	//         s[i].a === s.a[i]
	fmt.println(v.a, v.b, v.c)
}

soa_unzip is a built-in procedure which allows the user to recover the slices from an #soa slice.

// Recover the slices from the #soa slice
a, b, c := soa_unzip(s)
fmt.println(a, b, c)

matrix type #

A matrix is a mathematical type built into Odin. It is a regular array of numbers, arranged in rows and columns.

The following represents a matrix that has 2 rows and 3 columns:

m: matrix[2, 3]f32

m = matrix[2, 3]f32{
	1, 9, -13,
	20, 5, -6,
}

Element types of integers, float, and complex numbers are supported by matrices. There is no support for booleans, quaternions, or any compound type.

Indexing a matrix can be used with the matrix indexing syntax. This mirrors other type usages: type on the left, usage on the right.

elem := m[1, 2] // row 1, column 2

Scalars act as if they are scaled identity matrices and can be assigned to matrices as them

b := matrix[2, 2]f32{}
f := f32(3)
b = f

fmt.println("b", b)
fmt.println("b == f", b == f)

Matrices support multiplication between matrices:

a := matrix[2, 3]f32{
	2, 3, 1,
	4, 5, 0,
}

b := matrix[3, 2]f32{
	1, 2,
	3, 4,
	5, 6,
}

fmt.println("a", a)
fmt.println("b", b)

c := a * b
#assert(type_of(c) == matrix[2, 2]f32)
fmt.tprintln("c = a * b", c)		

Matrices support multiplication between matrices and arrays:

m := matrix[4, 4]f32{
	1, 2, 3, 4, 
	5, 5, 4, 2, 
	0, 1, 3, 0, 
	0, 1, 4, 1,
}

v := [4]f32{1, 5, 4, 3}

// treating `v` as a column vector
fmt.println("m * v", m * v)

// treating `v` as a row vector
fmt.println("v * m", v * m)

// Support with non-square matrices
s := matrix[2, 4]f32{ // [4][2]f32 default layout
	2, 4, 3, 1, 
	7, 8, 6, 5, 
}

w := [2]f32{1, 2}
r: [4]f32 = w * s
fmt.println("r", r)

Component-wise operations:

// if the element type supports it
// Not support for '/', '%', or '%%' operations

a := matrix[2, 2]i32{
	1, 2,
	3, 4,
}

b := matrix[2, 2]i32{
	-5,  1,
	 9, -7,
}

c0 := a + b
c1 := a - b
c2 := a & b
c3 := a | b
c4 := a ~ b
c5 := a &~ b

// component-wise multiplication
// since a * b would be a standard matrix multiplication
c6 := hadamard_product(a, b) 

fmt.println("a + b",  c0)
fmt.println("a - b",  c1)
fmt.println("a & b",  c2)
fmt.println("a | b",  c3)
fmt.println("a ~ b",  c4)
fmt.println("a &~ b", c5)
fmt.println("hadamard_product(a, b)", c6)

Submatrix Casting #

Submatrix casting square matrices #

Casting a square matrix to another square matrix with same element type is supported.

• If the cast is to a smaller matrix type, the top-left submatrix is taken.
• If the cast is to a larger matrix type, the matrix is extended with zeros everywhere and ones in the diagonal for the unfilled elements of the extended matrix.

mat2 :: distinct matrix[2, 2]f32
mat4 :: distinct matrix[4, 4]f32

m2 := mat2{
	1, 3,
	2, 4,
}

m4 := mat4(m2)
assert(m4[2, 2] == 1)
assert(m4[3, 3] == 1)
fmt.printf("m2 %#v\n", m2)
fmt.println("m4", m4)
fmt.println("mat2(m4)", mat2(m4))
assert(mat2(m4) == m2)

b4 := mat4{
	1, 2, 0, 0,
	3, 4, 0, 0,
	5, 0, 6, 0,
	0, 7, 0, 8,
}
fmt.println("b4", matrix_flatten(b4))

Casting non-square matrices #

Casting a matrix to another matrix is allowed as long as they share the same element type and the number of elements (rows*columns). Matrices in Odin are stored in column-major order, which means the casts will preserve this element order.

mat2x4 :: distinct matrix[2, 4]f32
mat4x2 :: distinct matrix[4, 2]f32

x := mat2x4{
	1, 3, 5, 7, 
	2, 4, 6, 8,
}

y := mat4x2(x)
fmt.println("x", x)
fmt.println("y", y)

Technical Information of matrix Types #

The default internal representation of a matrix in Odin is stored in column-major format e.g. matrix[2, 3]f32 is internally [3][2]f32 (with a different alignment requirement).

Column-major is used in order to utilize (SIMD) vector instructions effectively on modern hardware, if possible.

If a row-major storage format is required, the #row_major directive can be used. e.g. #row_major matrix[2, 3]f32 is internally [2][3]f32 (with a different alignment requirement).

Unlike normal arrays, matrices try to maximize alignment to allow for the (SIMD) vectorization properties whilst keeping zero padding (either between columns (assuming default layout) or at the end of the type).

Zero padding is a compromise for use with third-party libraries, instead of optimizing for performance. Padding between columns (assuming default layout) was not taken even if that would have allowed each column to be loaded individually into a SIMD register with the correct alignment properties.

Currently, matrices are limited to a maximum of 16 elements (rows*columns), and a minimum of 1 element. This is because matrices are stored as values (not a reference type), and thus operations on them will be stored on the stack. Restricting the maximum element count minimizes the possibility of stack overflows.

Built-in Procedures (Compiler Level) in base:intrinsics and core:math/linalg:

• transpose(m) transposes a matrix
• outer_product(a, b) takes two array-like data types and returns the outer product of the values in a matrix
• hadamard_product(a, b) component-wise multiplication of two matrices of the same type
• matrix_flatten(m)
  • converts the matrix into a flatten array of elements in column-major order.
• conj(x)
  • conjugates the elements of a matrix for complex element types only

Procedures (Runtime Level) (all square matrix procedures) in core:math/linalg:

• determinant(m)
• adjugate(m)
• inverse(m)
• inverse_transpose(m)
• hermitian_adjoint(m)
• matrix_trace(m)
• matrix_minor(m)

raw_data procedure #

raw_data is a built-in procedure which returns the underlying data of a built-in data type as a Multi-Pointer.

raw_data([]$E)         -> [^]E    // slices
raw_data([dynamic]$E)  -> [^]E    // dynamic arrays
raw_data(^[$N]$E)      -> [^]E    // fixed array and enumerated arrays 
raw_data(^#simd[$N]$E) -> [^]E    // simd vectors 
raw_data(string)       -> [^]byte // 	

using statement #

using can be used to bring entities declared in a scope/namespace into the current scope. This can be applied to import names, struct fields, procedure fields, and struct values.

import "foo"
bar :: proc() {
	// imports all the exported entities from the `foo` package into this scope
	using foo
}

Using statement with structs #

Let’s take a very simple entity struct:

Vector3 :: struct{x, y, z: f32}
Entity :: struct {
	position: Vector3,
	orientation: quaternion128,
}

It can be used like this:

foo :: proc(entity: ^Entity) {
	fmt.println(entity.position.x, entity.position.y, entity.position.z)
}

The entity members can be brought into the procedure scope by using it:

foo :: proc(entity: ^Entity) {
	using entity
	fmt.println(position.x, position.y, position.z)
}

The using can be applied to the parameter directly:

foo :: proc(using entity: ^Entity) {
	fmt.println(position.x, position.y, position.z)
}

It can also be applied to sub-fields:

foo :: proc(entity: ^Entity) {
	using entity.position
	fmt.println(x, y, z)
}

We can also apply the using statement to the struct fields directly, making all the fields of position appear as if they are on Entity itself:

Entity :: struct {
	using position: Vector3,
	orientation: quaternion128,
}
foo :: proc(entity: ^Entity) {
	fmt.println(entity.x, entity.y, entity.z)
}

Subtype polymorphism #

It is possible to get subtype polymorphism, similar to inheritance-like functionality in C++, but without the requirement of vtables or unknown struct layout:

foo :: proc(entity: Entity) {
	fmt.println(entity.x, entity.y, entity.z)
}

Frog :: struct {
	ribbit_volume: f32,
	using entity: Entity,
}

frog: Frog
// Both work
frog.x = 123
foo(frog)

Note: using can be applied to arbitrarily many things, which allows the ability to have multiple subtype polymorphism (but also its issues).

Note: using’d fields can still be referred by name.

or_else expression #

or_else is an infix binary operator that allows the user to define default values for certain expressions with optional-ok semantics.

m: map[string]int
i: int
ok: bool

if i, ok = m["hellope"]; !ok {
	i = 123
}
// The above can be mapped to `or_else`
i = m["hellope"] or_else 123

assert(i == 123)

or_else can be used with type assertions too, as they have optional-ok semantics.

v: union{int, f64}
i: int
i = v.(int) or_else 123
i = v.? or_else 123 // Type inference magic
assert(i == 123)

m: Maybe(int)
i = m.? or_else 456
assert(i == 456)

or_return operator #

The concept of or_return will work by popping off the end value in a multiple valued expression and checking whether it was not nil or was false, and if so, set the end return value to value if possible. If the procedure only has one return value, it will do a simple return. If the procedure had multiple return values, or_return will require that all parameters be named so that the end value could be assigned to by name and then an empty return could be called.

Error :: enum {
	None,
	Something_Bad,
	Something_Worse,
	The_Worst,
	Your_Mum,
}

caller_1 :: proc() -> Error {
	return .None
}

caller_2 :: proc() -> (int, Error) {
	return 123, .None
}
caller_3 :: proc() -> (int, int, Error) {
	return 123, 345, .None
}

foo_1 :: proc() -> Error {
	// This can be a common idiom in many code bases
	n0, err := caller_2()
	if err != nil {
		return err
	}

	// The above idiom can be transformed into the following
	n1 := caller_2() or_return


	// And if the expression is 1-valued, it can be used like this
	caller_1() or_return
	// which is functionally equivalent to
	if err1 := caller_1(); err1 != nil {
		return err1
	}

	// Multiple return values still work with `or_return` as it only
	// pops off the end value in the multi-valued expression
	n0, n1 = caller_3() or_return

	return .None
}
foo_2 :: proc() -> (n: int, err: Error) {
	// It is more common that your procedure returns multiple values
	// If `or_return` is used within a procedure that returns multiple 
	// values (2+), then all the returned values must be named 
	// so that a bare `return` statement can be used

	// This can be a common idiom in many code bases
	x: int
	x, err = caller_2()
	if err != nil {
		return
	}

	// The above idiom can be transformed into the following
	y := caller_2() or_return
	_ = y

	// And if the expression is 1-valued, it can be used like this
	caller_1() or_return

	// which is functionally equivalent to
	if err1 := caller_1(); err1 != nil {
		err = err1
		return
	}

	// If using a non-bare `return` statement is required, setting the return values
	// using the normal idiom is a better choice and clearer to read
	if z, zerr := caller_2(); zerr != nil {
		return -345 * z, zerr
	}

	n = 123
	return
}

caller_4 :: proc() -> (n: int, ok: bool) {
    return 3, true
}

foo_3 :: proc() -> (ok: bool) {
    // `or_return` also supports the ok semantics common in Odin code.
    // Note an error is indicated by `ok` being `false`
    x := caller_4() or_return

    if (x < 5) {
        ok = true
    }

    return
}

or_continue operator #

or_continue is an operator that allows for simplified control flow in a loop. It functions much like a combination of or_return and continue in that it checks if the last value in a multiple-valued expression is false or not nil, and if so, behaves like a continue statement.

import "core:fmt"

Job :: struct {
	id:         int,
	tasks_left: int,
	result:     int,
	error_has_occurred: bool,
}

// Returns true if it completed its last task.
do_work_until_done :: proc(job: ^Job) -> bool {
	if job.tasks_left == 0 {
		return false
	}

	job.result *= 3
	job.tasks_left -= 1
	return job.tasks_left == 0
}

// Returns a result and true if everything worked out.
// Otherwise, returns zero and false.
tally :: proc(job: ^Job) -> (int, bool) {
	if job.error_has_occurred {
		return 0, false
	}

	return job.result, true
}

main :: proc() {
	jobs: [dynamic]Job
	append(&jobs, Job{ id = 1, })
	append(&jobs, Job{ id = 3, tasks_left = 2, result = 7 })
	append(&jobs, Job{ id = 5, tasks_left = 1, result = 5 })
	append(&jobs, Job{ id = 7, tasks_left = 1, result = 5, error_has_occurred = true })

	job_loop: for &job in jobs {
		do_work_until_done(&job) or_continue job_loop
		result := tally(&job)    or_continue job_loop

		fmt.printfln("Job #%i has completed its work. Result is: %i", job.id, result)
	}
}

The example above will print the following:

Job #5 has completed its work. Result is: 15

Only job #5 completes, because it has only 1 task left and its error_has_occurred is the default value, which is false in this case, as every variable in Odin is implicitly zeroed out unless made explicit.

Without or_continue, the above loop would look more like this:

// [...]
	job_loop: for &job in jobs {
		if !do_work_until_done(&job) {
			continue job_loop
		}
		result, ok := tally(&job)
		if !ok {
			continue job_loop
		}

		fmt.printfln("Job #%i has completed its work. Result is: %i", job.id, result)
	}
// [...]

The underlying operation is the same, but the code is more verbose where it doesn’t need to be. As shown in the example, or_continue also works with labels, just like continue.