Introduction. #

Cyber is a fast, efficient, and concurrent scripting language. The landing page is at cyberscript.dev and contains performance metrics and release notes.

These docs provide a reference manual for the language. You can read it in order or jump around using the navigation. You may come across features that are marked Incomplete or Planned. This is because the docs are written as if all features have been completed already.

Type System. #

Cyber is a statically typed language. However, dynamic typing is also supported. If you're coming from a language such as Python, JavaScript, or Lua, it can be easier to get started with Dynamic Typing.

Hello World. #

use math

var worlds = {'World', '世界', 'दुनिया', 'mundo'}
worlds.append(math.random())
for worlds -> w:
    print "Hello, $(w)!"

Syntax. #

^top

Cyber's syntax is concise and easy to read.

Statements. #

A statement ends with the new line:

var a = 123

To wrap a statement to the next line, the last token before the new line must be an operator or a separator:

var gameover = health <= 0 or
    player.collidesWith(spikes)

if year > 2020 and year <= 2030 and
    month > 0 and month <= 11:
    print 'Valid'

Any token inside a delimited syntax (such as parentheses or braces) can be wrapped to the next line:

var sum = add(1, 2, 3, 4,
    100, 200, 300, 400)

var colors = {'red', 'blue', 'green',
    'purple', 'orange', 'yellow'}

Blocks. #

Some statements can start a new block with a colon. The first statement in a new block must be indented further. Spaces or tabs can be used for indentation but not both.

-- This `if` statement begins a new block.
if true:
    var a = 234

Subsequent statements in the block must follow the same indentation. The block ends when a statement recedes from this indentation:

var items = {10, 20, 30}
for items -> it:
    if it == 20:
        print it
    print it      -- This is the first statement outside of the `if` block.

Compact blocks allow only one statement after the starting block:

-- A single line block.
if true: print 123

if true: print 123
    -- This is an indentation error since the compact block is already consumed.
    print 234

Since blocks require at least one statement, use pass as a placeholder statement:

func foo():
    pass

Variables. #

Variables allow values to be stored into named locations in memory.

Local variables and static variables are supported.

Local variables. #

Local variables exist until the end of their scope. They are declared and initialized using the var keyword:

var a = 123

When declared without a type specifier next to the variable, the type is inferred from the right initializer. In the above example, the variable a is initialized with the type int.

Variables can be assigned afterwards using the = operator:

a = 234

Explicit type constraint. #

When a type specifier is provided, the value assigned to the variable must satisfy the type constraint or a compile error is reported:

var a float = 123.0

var b int = 123.0    --> CompileError. Expected `int`, got `float`.

Any operation afterwards that violates the type constraint of the variable will result in a compile error

a = 'hello'          --> CompileError. Expected `float`, got `String`.

Variable scopes. #

Blocks create a new variable scope. Variables declared in the current scope will take precedence over any parent variables with the same name:

func foo():
    var a = 234

    if true:
        var a = 345     -- New `a` declared.
        print a         --> 345

    print a             --> 234

Static variables. #

Static variables will likely be removed in favor of context variables, constants, and extern variables.

Static variables live until the end of the script. They act as global variables and are visible from anywhere in the script.

Static variables are declared with var like local variables, but a namespace must be provided before the variable name:

var .a = 123

func foo():
    print a     --> 123

The . prefix is used to reference the current module's namespace.

A type specifier can be provided after the variable name, otherwise, it's inferred from the initializer:

var .my_map any = Map{}

Since static variables are initialized outside of a fiber's execution flow, they can not reference any local variables:

var .b = a   --> Compile error, initializer can not reference a local variable.

-- Main execution.
var a = 123

However, they can be reassigned after initialization:

var .b = 0

var a = 123
b = a            -- Reassigning after initializing.

Static variable initializers have a natural order based on when it was encountered by the compiler. In the case of imported variables, the order of the import would affect this order. The following would print '123' before '234':

var .a = print(123)
var .b = print(234)

When the initializers reference other static variables, those child references are initialized first in DFS order and supersede the natural ordering. The following initializes b before a.

var .a = b + 321
var .b = 123

print a        --> 444

Circular references in initializers are not allowed. When initialization encounters a reference that creates a circular dependency an error is reported.

var .a = b
var .b = a     --> CompileError. Referencing `a` creates a circular dependency.

Sometimes, you may want to initialize a static variable by executing multiple statements in order. For this use case, you can use a declaration block. Planned Feature

var .myImage =:
    var img = loadImage('me.png')
    img.resize(100, 100)
    img.filter(.blur, 5)
    break img

The final resulting value that is assigned to the static variable is provided by a break statement. If a break statement is not provided, none is assigned instead.

Context variables. #

Context variables are bound to each virtual thread and are accessible anywhere on the call stack. All fibers created in the same virtual thread reference the same context variables.

Context variables used in the program must be declared in the main source file with a default value: Planned Feature

-- main.cy

context MyInt = 123
context MyString = 'abc'

print MyInt        --> 123
print MyString     --> abc

To reference a context variable in a different source file, redeclare it with the same type:

-- foo.cy

context MyInt int

func foo():
    print MyInt    --> 123

Since this is a redeclaration, an assignment statement is not allowed.

Some context variables such as mem for memory allocations are already declared in every program, so it only needs a redeclaration:

context mem Memory

var a = mem.new(int)
a.* = 123
print a.*          --> 123
mem.free(a)

extern variables. #

Planned Feature

use $global #

When use $global is declared in the module, it allows the use of undeclared variables:

use $global

a = 123
print a    --> 123

Accessing an undeclared variable before it's initialized results in a runtime error:

use $global

print a    --> panic: `a` is not defined in `$global`.

Reserved identifiers. #

Keywords. #

There are 29 general keywords. This list categorizes them:

• Control Flow: if else switch case while for break continue pass
• Operators: or and not
• Variables: var context
• Functions: func return
• Fibers: coinit coyield coresume
• Async: await
• Types: type as
• Type embedding: use
• Error Handling: try catch throw
• Modules: use mod
• Dynamic Typing: dyn

Contextual keywords. #

These keywords only have meaning in a certain context.

• Methods: self Self
• Types: object struct cstruct enum trait
• Catching Errors: caught
• Function Return: void

Literals. #

• Boolean literal: true false
• Symbol literal: symbol
• Error literal: error
• None: none

Operators. #

Cyber supports the following operators. They are ordered from highest to lowest precedence.

Arithmetic Operators. #

The following arithmetic operators are supported for the numeric data types.

1 + 2     -- Addition, evaluates to 3.
100 - 10  -- Subtraction, evaluates to 90.
3 * 4     -- Multiplication, evaluates to 12.
20 / 5    -- Division, evaluates to 4.
2 ^ 4     -- Raise to the power, evaluates to 16.
12 % 5    -- Modulus remainder, evaluates to 2.
-(10)     -- Apply negative, evaluates to -10.

Comparison Operators. #

Cyber supports the following comparison operators. By default, a comparison operator evaluates to a Boolean value.

The equals operator returns true if the two values are equal. For primitive types, the comparison checks the types and the underlying value. For strings, the underlying bytes are compared for equality. For objects, the comparison checks that the two values reference the same object.

1 == 1      -- Evaluates to `true`
1 == 2      -- Evaluates to `false`
1 == true   -- Evaluates to `false`

var a = 'abc'
a == 'abc'  -- Evaluates to `true`

a = {_}
var b = a
a == b      -- Evaluates to `true`
a == {_}    -- Evaluates to `false`

The not equals operator returns true if the two values are not equal.

1 != 1      -- Evaluates to `false`
1 != 2      -- Evaluates to `true`

Number types have additional comparison operators.

a > b    -- `true` if a is greater than b
a >= b   -- `true` if a is greater than or equal to b
a < b    -- `true` if a is less than b
a <= b   -- `true` if a is less than or equal to b

Logic Operators. #

The logical operators and, or, and not are supported.

and evaluates to true if both operands are true. Otherwise, it evaluates to false. If the left operand is false, the evaluation of the right operand is skipped:

true and true    --> true
true and false   --> false
false and true   --> false
false and false  --> false

or evaluates to true if at least one of the operands is true. Otherwise, it evaluates to false. If the left operand is true, the evaluation of the right operand is skipped:

true or true     --> true
true or false    --> true
false or true    --> true
false or false   --> false

The unary operator not performs negation on the boolean value. The unary operator ! can also be used instead of not.

not false     --> true
not true      --> false
!false        --> true
!true         --> false

Bitwise Operators. #

The following bitwise operators are supported for int number values.

-- Bitwise and: any underlying bits that are set in both integers are set in the new integer.
a & b

-- Bitwise or: any underlying bits that are set in either integer a or integer b are set in the new integer.
a | b

-- Bitwise exclusive or: any underlying bits that are set in either integer a or integer b but not both are set in the new integer.
a || b

-- Bitwise right shift: a's bits are shifted b bits to the least significant end. This performs sign-extension on the 32-bit integer.
a >> b

-- Bitwise left shift: a's bits are shifted b bits to the most significant end. This does not perform sign-extension on the 32-bit integer.
a << b

-- Bitwise not: a's integer bits are flipped.
~a

Operator overloading. #

See Operator overloading in Metaprogramming.

Zero values. #

Uninitialized type fields currently default to their zero values. However, this implicit behavior will be removed in the future in favor of a default value clause. Zero values must then be expressed using the reserved zero literal.

The following shows the zero values of builtin or created types:

Type casting. #

The as keyword can be used to cast a value to a specific type. Casting lets the compiler know what the expected type is and does not perform any conversions.

If the compiler knows the cast will always fail at runtime, a compile error is returned instead.

print('123' as int)       --> CompileError. Can not cast `String` to `int`.

If the cast fails at runtime, a panic is returned.

var erased any = 123
add(1, erased as int)     --> Success.
print(erased as String)   --> panic: Can not cast `int` to `String`.

func add(a int, b int):
    return a + b

Comments. #

A single line comment starts with two hyphens and ends at the end of the line.

-- This is a comment.

var a = 123   -- This is a comment on the same line as a statement.

There will be multi-line comments in Cyber but the syntax has not been determined.

CYON. #

CYON or the Cyber object notation is similar to JSON. The format uses the same literal value semantics as Cyber.

{
    name  = 'John Doe',
    'age' = 25,

    -- This is a comment
    cities = {
        'New York',
        'San Francisco',
        'Tokyo',
    },
}

Basic Types. #

^top

In Cyber, there are value types and reference types. This section will describe common builtin types.

Booleans. #

Booleans can be true or false. See type bool.

var a = true
if true:
    print 'a is true'

Numbers. #

Integers. #

int is the default integer type and is an alias for int64. It has 64-bits representing integers in the range -(2⁶³) to 2⁶³-1. Integer types use the two's complement format.

The following integer types are supported: byte, int8, int16, int32, int64. Integer types are named according to how many bits they occupy.

When a numeric literal is used and the type can not be inferred, it will default to the int type:

var a = 123

Integer notations always produce an integer value:

var a = 0xFF     -- hex.
a = 0o17         -- octal.
a = 0b1010       -- binary.
a = `🐶`         -- UTF-8 rune.

Arbitrary values can be converted to a int using the type as a function.

var a = '123'
var b = int(a) 

In addition to arithmetic operations, integers can also perform bitwise operations.

Floats. #

float is the default floating point type. It has a (IEEE 754) 64-bit floating point format. See type float.

Although a float represents a decimal number, it can also represent integers between -(2⁵³-1) and (2⁵³-1). Any integers beyond the safe integer range is not guaranteed to have a unique representation.

A numeric literal can be used to create a float if the inferred type is a float:

var a float = 123

Decimal and scientific notations always produce a float value:

var a = 2.34567
var b = 123.0e4

Arbitrary values can be converted to a float using the type as a function.

var a = '12.3'
var b = float(a) 

Big Numbers. #

Planned Feature

Strings. #

The String type represents a sequence of UTF-8 codepoints. Each code point is stored internally as 1-4 bytes. See type String.

Strings are not validated by default. If the codepoint is invalid at a particular index, the replacement character (0xFFFD) is returned instead.

Strings are immutable, so operations that do string manipulation return a new string. By default, short strings are interned to reduce memory footprint.

Under the hood, there are multiple string implementations to make operations faster by default using SIMD.

Raw string literal. #

A raw string doesn't allow any escape sequences or string interpolation.

Single quotes are used to delimit a single line literal:

var fruit = 'apple'
var str = 'abc🦊xyz🐶'

Since raw strings interprets the sequence of characters as is, a single quote character can not be escaped:

var fruit = 'Miso's apple'    -- ParseError.

Triple single quotes are used to delimit a multi-line literal. It also allows single quotes and double single quotes in the string:

var fruit = '''Miso's apple'''
var greet = '''Hello
World'''

String literal. #

A string literal allows escape sequences and string interpolation.

Double quotes are used to delimit a single line literal:

var fruit = "apple"
var sentence = "The $(fruit) is tasty."
var doc = "A double quote can be escaped: \""

Triple double quotes are used to delimit a multi-line literal:

var title = "last"
var doc = """A double quote " doesn't need to be escaped."""
var str = """line a
line "b"
line $(title)
"""

Escape sequences. #

The following escape sequences are supported in string literals wrapped in double quotes:

Example:

print "\xF0\x9F\x90\xB6"    --> 🐶

String indexing. #

The index operator returns the rune starting at the given byte index:

var a = 'abcxyz'
print a[1]          --> 98
print(a[1] == `b`)  --> true

If an index does not begin a sequence of valid UTF-8 bytes, the replacement character (0xFFFD, 65533) is returned:

var a = '🐶abcxyz'
print a[1]          --> 65533

Since indexing operates at the byte level, it should not be relied upon for iterating runes or rune indexing. However, if the string is known to only contain ASCII runes (each rune occupies one byte), indexing will return the expected rune.

String.seek will always return the correct byte index for a rune index:

var a = '🐶abcxyz'
print a.seek(2)         --> 5
print a[a.seek(2)]      --> 98 (`b`)

Similarily, slicing operates on byte indexes. Using the slice operator will return a view of the string at the given start and end (exclusive) indexes. The start index defaults to 0 and the end index defaults to the string's byte length:

var str = 'abcxyz'
var sub = str[0..3]
print str[0..3]  --> abc
print str[..5]   --> abcxy
print str[1..]   --> bcxyz

String concatenation. #

Concatenate two strings together with the + operator or the method concat.

var res = 'abc' + 'xyz'
res = res.concat('end')

String interpolation. #

Expressions can be embedded into string templates with $():

var name = 'Bob'
var points = 123
var str = "Scoreboard: $(name) $(points)"

String templates can not contain nested string templates.

String formatting. #

Values that can be formatted into a string will have a fmt method:

var file = os.openFile('data.bin', .read)
var bytes = file.readAll()

-- Dump contents in hex.
print "$(bytes.fmt(.x))"

Line-join literal. #

The line-join literal joins string literals with the new line character \n. Planned Feature

This has several properties:

• Ensures the use of a consistent line separator: \n
• Allows lines to have a mix of raw string or string literals.
• Single quotes and double quotes do not need to be escaped.
• Allows each line to be indented along with the surrounding syntax.
• The starting whitespace for each line is made explicit.

var paragraph = {
    \'the line-join literal
    \'hello\nworld
    \"hello $(name)
    \'last line
    \'
}

Mutable strings. #

To mutate an existing string, use type MutString. Planned Feature

Optionals. #

An Optional is a value type that provides null safety by forcing the inner value to be unwrapped before it can be used.

The Option template type is a choice type that either holds a none value or contains some value. The option template is defined as:

type Option[T type] enum:
    case none
    case some T

A type prefixed with ? is the idiomatic way to create an option type. The following String optional types are equivalent:

Option[String]
?String

Wrap value. #

A value is automatically wrapped into the inferred optional's some case:

var a ?String = 'abc'
print a     --> some(abc)'

Wrap none. #

none is automatically initialized to the inferred optional's none case:

var a ?String = none
print a     --> none

Unwrap or panic. #

The .? access operator is used to unwrap an optional. If the expression evaluates to the none case, the runtime panics:

var opt ?int = 123
var v = opt.?
print v     --> 123

Unwrap or default. #

The ?else control flow operator either returns the unwrapped value or a default value when the optional is none:

var opt ?int = none
var v = opt ?else 123
print v     --> 123

?else can be used in an assignment block: Planned Feature

var v = opt ?else:
    break 'empty'

var v = opt ?else:
    throw error.Missing

Optional chaining. #

Given the last member's type T in a chain of ?. access operators, the chain's execution will either return Option[T].none on the first encounter of none or returns the last member as an Option[T].some: Planned Feature

print root?.a?.b?.c?.last

if unwrap. #

The if statement can be amended to unwrap an optional value using the capture -> operator:

var opt ?String = 'abc'
if opt -> v:
    print v     -- Prints 'abc'

while unwrap. #

The while statement can be amended to unwrap an optional value using the capture -> operator. The loop exits when none is encountered:

var iter = dir.walk()
while iter.next() -> entry:
    print entry.name

Arrays. #

An array type is a value type and is denoted as [N]T where N is the size of the array and T is the element type. See type Array.

Arrays are initialized with its type followed by the initializer literal:

var a = [3]int{1, 2, 3}

The number of elements can be inferred using pseudo type [.]T: Planned Feature

var a = [.]int{1, 2, 3}

The array type can be inferred by the dot initializer literal:

var a [3]int = .{1, 2, 3}

Arrays can be indexed:

a[2] = 300
print a[2]    --> 300

Lists. #

Lists are a builtin type that holds an ordered collection of elements. Lists grow or shrink as you insert or remove elements. See type List.

The first element of the list starts at index 0. Lists are initialized with elements in braces:

var list = {1, 2, 3}
print list[0]    --> 1

The empty list is initialized with an underscore as the only element:

var list = {_}
print list.len() --> 0

Lists can be sliced with the range .. clause. The sliced list becomes a new list that you can modify without affecting the original list. The end index is non-inclusive.

var list = {1, 2, 3, 4, 5}
print list[0..0]    --> {_}          
print list[0..3]    --> {1, 2, 3}
print list[3..]     --> {4, 5}
print list[..3]     --> {1, 2, 3}

The +.. invokes the slice operator with an end position that is an increment from the start: Planned Feature

var list = {1, 2, 3, 4, 5}
print list[2+..2]   --> {3, 4}

List operations.

var list = {234}

-- Append a value.
list.append(123)

-- Inserting a value at an index.
list.insert(1, 345)

-- Get the length.
print list.len()  -- Prints '2'

-- Sort the list in place.
list.sort((a, b) => a < b)

-- Iterating a list.
for list -> it:
    print it

-- Remove an element at a specific index.
list.remove(1)

Since List is a generic type, an explicit List type can be attached to the initializer:

var a = List[int]{1, 2, 3}

When the intializer is only prefixed with a dot, it will infer the List type constraint:

var a List[int] = .{1, 2, 3}

Tables. #

A Table is a versatile object that can have an arbitrary set of fields.

By default, the record literal initializes a Table:

var o = {}

o = {a=123}
print o.a          --> 123

A Table can be initialized explicitly using its type name:

var o = Table{a=123}

Any field can be assigned a value. However, accessing a field before it's initialized results in a panic:

o.my_field = 234
print o.my_field   --> 234

print o.foo        --> panic: The field `foo` was not initialized.

Table indexing. #

Indexing can be used to access a dynamic field or an arbitrary key:

var o = {name='Nova'}
var field = 'name'
print o[field]     --> Nova

o[10] = [1, 2, 3]
print o[10]        --> List (3)

If the key is not an identifier string, the value can only be obtained through indexing.

Check field existence. #

Planned Feature

Prototypes. #

Planned Feature

Maps. #

Maps are a builtin type that store key value pairs in dictionaries. See type Map.

Maps are initialized with the Map type and a record literal:

var map = Map{a=123, b=() => 5}

The empty record literal creates an empty map:

var empty = Map{}

Map indexing. #

Get a value from the map using the index operator:

print map['a']

Map operations. #

var map = Map{}

-- Set a key value pair.
map[123] = 234

-- Get the size of the map.
print map.size()

-- Remove an entry by key.
map.remove(123)

-- Iterating a map.
for map -> {val, key}:
    print "$(key) -> $(val)"

Map block. #

Entries can also follow a collection literal block. This gives structure to the entries and has the added benefit of allowing multi-line lambdas. Most likely this will not be implemented in favor of a builder syntax

var colors = {}:
    .red   = 0xFF0000
    .green = 0x00FF00
    .blue  = 0x0000FF
    .dump  = func (c):
        print c.red
        print c.green
        print c.blue

    -- Nested map.
    .darker = {}: 
        .red   = 0xAA0000
        .green = 0x00AA00
        .blue  = 0x0000AA

Symbols. #

Symbol literals begin with symbol., followed by an identifier. Each symbol has a global unique ID.

var currency = symbol.usd
print currency == .usd   --> true
print int(currency)      --> 

any. #

Unlike dyn, any is statically typed and performs type checks at compile-time. any type can hold any value, but copying it to narrowed type destination will result in a compile error:

func square(i int):
    return i * i

var a any = 123
a = {'a', 'list'}         --> Valid assignment to a value with a different type.
a = 10

print square(a)           --> CompileError. Expected `int`, got `any`.

a must be explicitly casted to satisfy the type constraint:

print square(a as int)    --> 100

dyn. #

The dynamic type defers type checking to runtime. However, it also tracks its own recent type in order to surface errors at compile-time. See Dynamic Typing.

Type Declarations. #

^top

Objects. #

An object type contains field and function members. New instances can be created from them similar to a struct or class in other languages. Unlike classes, there is no builtin concept of inheritance but a similar effect can be achieved using composition and embedding.

Object types are declared with type object. The object keyword is optional when using a type declaration. These two declarations are equivalent:

type A object:
    my_field int

type A:
    my_field int

Fields must be declared at the top of the type block with their names and type specifiers:

type Node:
    value int
    next  ?Node

Instantiate. #

New object instances are created using a record literal with a leading type name:

var node = Node{value=123, next=none}
print node.value       -- Prints "123"

A dot record literal can also initialize to the inferred object type:

var node Node = .{value=234, next=none}
print node.value       -- Prints "234"

Field visibility. #

All fields have public visibility. However, when a field is declared with a - prefix, it suggests that it should not be made available to an editor's autocomplete:

type Info:
    a       int
    -b      int
    -secret String

Default field values. #

When a field is omitted in the record literal, it gets initialized to its zero value:

var node Node = .{value=234}
print node.next    --> Option.none

type Student:
    name String
    age  int
    gpa  float

var s = Student{}
print s.name       -->
print s.age        --> 0
print s.gpa        --> 0.0

Circular references. #

Circular type references are allowed if the object can be initialized:

type Node:
    val  any
    next ?Node

var n = Node{}    -- Initializes.

In this example, next has an optional ?Node type so it can be initialized to none when creating a new Node object.

The following example will fail because this version of Node can not be initialized:

type Node:
    val  any
    next Node

var n = Node{}    -- CompileError. Can not zero initialize `next`
                   -- because of circular dependency.

Unnamed object. #

Unnamed object types can be declared and used without an explicit type declaration:

type Node:
    value object:
        a  int
        b  float
    next  ?Node

var node = Node{
    value = .{a=123, b=100.0},
    next  = none,
}

Methods. #

Methods are functions that are invoked with a parent instance using the . operator. When the first parameter of a function contains self, it's declared as a method of the parent type:

type Node:
    value int
    next  ?Node

    func inc(self, n):
        self.value += n

    func incAndPrint(self):
        self.inc(321)
        print value

var n = Node{value=123, next=none}
n.incAndPrint()         -- Prints "444"

self is then used to reference members of the parent instance.

Methods can be declared outside of the type declaration as a flat declaration:

func Node.getNext(self):
    return self.next

Type functions. #

Type functions can be declared within the type block without a self param:

type Node:
    value int
    next  ?Node

    func new():
        return Node{value=123, next=none}

var n = Node.new()

Type functions can also be declared outside of the type block using a flat declaration:

type Node:
    value int
    next  ?Node

func Node.new():
    return Node{value=123, next=none}

var n = Node.new()

Type variables. #

Similarily, type variables are declared outside of the type block:

-- Declare inside the `Node` namespace.
var Node.DefaultValue = 100

print Node.DefaultValue    -- Prints "100"

Type embedding. #

Type embedding facilitates type composition by using the namespace of a child field's type: Planned Feature

type Base:
    a int

    func double(self) int:
        return a * 2

type Container:
    b use Base

var c = Container{b = Base{a=123}}
print c.a         --> 123
print c.double()  --> 246

Note that embedding a type does not declare extra fields or methods in the containing type. It simply augments the type's using namespace by binding the embedding field.

If there is a member name conflict, the containing type's member has a higher precedence:

type Container:
    a int
    b use Base

var c = Container{a=999, b = Base{a=123}}
print c.a         --> 999
print c.double()  --> 246

Since the embedding field is named, it can be used just like any other field:

print c.b.a       --> 123

Structs. #

Struct types can contain field and method members just like object types, but their instances are copied by value rather than by reference. In that sense, they behave like primitive data types.

Unlike objects, structs are value types and do not have a reference count. They can be safely referenced with the ref modifier and their lifetime can be managed with single ownership semantics. Unsafe pointers can not reference structs by default, but there may be an unsafe builtin to allow it anyway.

Declare struct. #

Struct types are created using the type struct declaration:

type Vec2 struct:
    x float
    y float

var v = Vec2{x=30, y=40}

Copy structs. #

Since structs are copied by value, assigning a struct to another variable creates a new struct:

var v = Vec2{x=30, y=40}
var w = v
v.x = 100
print w.x    -- Prints '30'
print v.x    -- Prints '100'

Tuples. #

Tuples can be declared using parentheses to wrap member fields:

type Vec2 struct(x float, y float)

If the fields share the same type, they can be declared in a field group:

type Vec3 struct(x, y, z float)

Function declarations can still be declared under the type:

type Vec2 struct(x float, y float):
    func scale(self, s float):
        self.x *= s
        self.y *= s

Tuples can be initialized with member values corresponding to the order they were declared:

var v = Vec2{3, 4}

The initializer can also be inferred from the target type:

var v Vec2 = .{3, 4}

Tuples can still be initialized with explicit field names:

var v = Vec2{x=3, y=4}

Enums. #

A new enum type can be declared with the type enum declaration. An enum value can only be one of the unique symbols declared in the enum type. By default, the symbols generate unique ids starting from 0.

type Fruit enum:
    case apple
    case orange
    case banana
    case kiwi

var fruit = Fruit.kiwi
print fruit       -- 'Fruit.kiwi'
print int(fruit)  -- '3'

When the type of the value is known to be an enum, it can be assigned using a symbol literal.

var fruit = Fruit.kiwi
fruit = .orange
print(fruit == Fruit.orange)   -- 'true'

Choices. #

Choices are enums with payloads (also known as sum types or tagged unions). An enum declaration becomes a choice type if one of the cases has a payload type specifier:

type Shape enum:
    case rectangle Rectangle
    case circle    object:
        radius float
    case triangle  object:
        base   float
        height float
    case line      float
    case point 

type Rectangle:
    width  float
    height float

Initialize choice. #

The general way to initialize a choice is to invoke the initializer with the payload as the argument:

var rect = Rectangle{width=10, height=20}
var s = Shape.rectangle(rect)

s = Shape.line(20)

If the payload is a record-like type, the choice can be initialized with a record literal:

var s = Shape.rectangle{width=10, height=20}

A choice without a payload is initialized like an enum member:

var s = Shape.point

Choice switch. #

case clauses can match choices and capture the payload:

switch s
case .rectangle -> r:
    print "$(r.width) $(r.height)"
case .circle -> c:
    print "$(c.radius)"
case .triangle -> t:
    print "$(t.base) $(t.height)"
case .line -> len:
    print "$(len)"
case .point:
    print "a point"
else:
    print "Unsupported."

Unwrap choice. #

A choice can be accessed by specifying the unwrap operator .! before the tagged member name. This will either return the payload or panic at runtime:

var s = Shape{line=20}
print s.!line     --> 20

Type aliases. #

A type alias refers to a different target type. Once declared, the alias and the target type can be used interchangeably.

A type alias declaration is denoted as type A -> T where A is the alias type of T:

type Vec2:
    x float
    y float

type Pos2 -> Vec2

var pos = Pos2{x=3, y=4}

Distinct types. #

A distinct type creates a new type by copying a target type.

It's declared with type name declaration followed by the target type specifier:

type Vec2:
    x float
    y float

type Pos2 Vec2

var pos = Pos2{x=3, y=4}

Functions can be declared under the new type's namespace:

use math

type Pos2 Vec2:
    func blockDist(self, o Pos2):
        var dx = math.abs(o.x - x)
        var dy = math.abs(o.y - y)
        return dx + dy

var pos = Pos2{x=3, y=4}
var dst = Pos2{x=4, y=5}
print pos.blockDist(dst)     --> 2

Note that functions declared from the target type do not carry over to the new type.

Unlike a type alias, the new type and the target type can not be used interchangeably since they are different types. However, instances of the new type can be casted to the target type, and vice versa: Planned Feature

type Pos2 Vec2

var a = Pos2{x=3, y=4}

var b Vec2 = a as Vec2

Traits. #

A trait type defines a common interface for implementing types. A trait type is declared with the trait keyword:

type Shape trait:
    func area(self) float

Types can implement a trait using the with keyword:

type Circle:
    with Shape
    radius float

    func area(self) float:
        return 3.14 * self.radius^2

type Rectangle:
    with Shape
    width  float
    height float

    func area(self) float:
        return self.width * self.height

A type that intends to implement a trait but does not satisfy the trait's interface results in a compile error.

Implementing types become assignable to the trait type:

var s Shape = Circle{radius=2}
print s.area()         --> 12.57

s = Rectangle{width=4, height=5}
print s.area()         --> 20

Type templates. #

Type declarations can include template parameters to create a type template:

type MyContainer[T type]:
    id    int
    value T

    func get(self) T:
        return self.value

Expand type template. #

When the type template is invoked with template argument(s), a special version of the type is generated.

In this example, String can be used as an argument since it satisfies the type parameter constraint:

var a MyContainer[String] = .{id=123, value='abc'}
print a.get()      -- Prints 'abc'

Note that invoking the template again with the same argument(s) returns the same generated type. In other words, the generated type is always memoized from the input parameters.

Type specialization. #

Value templates can be used to specialize type templates:

def List[T type] type:
    if T == dyn:
        return DynList
    else:
        return GenList[T]

var a = List[int]{1, 2, 3}

C Types. #

• C primitives.
• C structs.
  • C struct methods.
  • C struct reference.
• Pointers.
  • Dereferencing pointers.
  • Pointer indexing.
  • Pointer arithmetic.
  • Pointer casting.
  • Pointer slicing.
  • Pointer conversions.
• Pointer slices.
• Union types.

^top

C types interop with C-ABI types and are used to represent external or manual memory. Using C types is considered unsafe, but runtime safety checks can be inserted to make it safer at the cost of runtime performance.

C primitives. #

This table shows the size and type compatibility of Cyber and C types:

* Varies depending on C compiler and architecture. Pointers may occupy an extra word size for runtime memory checks.

C structs. #

C structs are unsafe since they can be referenced by unsafe pointers. The underlying memory could be corrupted or invalidated before use. In contrast, structs can be safely used since they can only be instantiated in safe memory.

C structs can be declared with the cstruct keyword:

type Data cstruct:
    x   float
    y   float
    ptr *int
    str [*]byte

A C struct may contain:

• C types.
• Primitive types.
• Container types that contain a compatible element type. This includes enums, choices, and optionals.

It may not contain structs or objects.

C struct methods. #

C structs can be declared with methods:

type Vec2 cstruct:
    x float
    y float

    func add(self, o Vec2):
        return Vec2{
            x = self.x+o.x,
            y = self.y+o.y,
        }

var v = Vec2{x=1, y=2}
print v.add(Vec2{x=2, y=1})   --> Vec2{x=3, y=3}

In this example add passes the receiver by value.

In order to pass the receiver by reference, self must be annotated with *:

type Vec2 cstruct:
    x float
    y float

    func add(*self, o Vec2):
        self.x += o.x
        self.y += o.y

var v = Vec2{x=1, y=2}
v.add(Vec2{x=2, y=1})
print v                 --> Vec2{x=3, y=3}

C struct reference. #

The * operator is used to obtain the pointer to a C struct value:

type Vec2 cstruct:
    x float
    y float

var v = Vec2{x=1, y=2}
var ref = *v
ref.x = 4

print v                 --> Vec2{x=4, y=2}

The reference is a pointer type that points to the C struct:

func scale(a *Vec2, n float):
    a.x *= n
    a.y *= n

add(*v, 10)
print v                 --> Vec2{x=40, y=20}

Pointers. #

A pointer is a reference to a memory location. Its type is denoted as *T where T is the type that the pointer references in memory:

func setName(p *Person, name [*]byte):
    p.name = name

var p = Person{}
setName(*p, 'Spock')

Depending on the target architecture, the alignment of the pointer will either be at least 8 bytes on a 64-bit machine or 4 bytes on a 32-bit machine. Aligned pointers will be supported in a future version.

Pointers are unsafe since they can reference corrupted or invalidated memory. When runtime memory checks are enabled, pointers will occupy an extra word size in order to set traps and prevent unsafe uses of pointers. See Memory / Runtime memory checks.

A pointer can be created with an explicit address using pointer.

var ptr = pointer(void, 0xDEADBEEF)
print ptr.value()     --'3735928559'

Dereferencing pointers. #

Pointers are dereferenced using the accessor operator .*:

var a = 123
var ptr = *a
print ptr.*      --> 123

ptr.* = 10
print a          --> 10

Pointer indexing. #

The index operator is used to read or write to a particular element:

var a = mem.alloc(int, 10)
a[5] = 123
print a[5]       --> 123

Negative indexing will locate the element before the pointer's address.

Pointer arithmetic. #

Planned Feature

Pointer casting. #

Planned Feature

Pointer slicing. #

The slice operator will produce a new slice of type [*]T from a *T pointer type:

var ints = mem.alloc(int, 10)
var slice = ints[0..5]
print typeof(slice)    --> []int

Pointer conversions. #

Planned Feature

Pointer slices. #

Slices are pointers with a length field. They are denoted as [*]T where T is the element type.

A slice can be created by taking the pointer of an array: Planned feature

var arr = [3]int{1, 2, 3}
var s = *arr

Read and write an element with the index operator:

print s[0]        --> 1
s[1] = 123
print s[1]        --> 123

Slices provide bounds checking when runtime safety is enabled: Planned feature

print s[100]      --> Panic. Out of bounds.

Slice operations can be applied: Planned feature

print s[0..2]     --> {1, 2}

Union types. #

Planned Feature

Control Flow. #

^top

Cyber provides the common constructs to branch and enter loops.

Branching. #

if statement. #

Use if and else statements to branch the execution of your code depending on conditions. The else clause can contain a condition which is only evaluated if the previous if or conditional else clause was false.

var a = 10
if a == 10:
    print 'a is 10'
else a == 20:
    print 'a is 20'
else:
    print 'neither 10 nor 20'

if expression. #

An if expression evaluates a condition in parentheses and returns either the true value or false value. Unlike the if statement, the if expression can not contain else conditional cases:

var a = 10
var str = if (a == 10) 'red' else 'blue'

and/or #

Use and and or logical operators to combine conditions:

var a = 10
if a > 5 and a < 15:
    print 'a is between 5 and 15'
if a == 20 or a == 10: 
    print 'a is 10 or 20'

Iterations. #

Infinite while. #

The while keyword starts an infinite loop which continues to run the code in the block until a break or return is reached.

var count = 0
while:
    if count > 100:
        break
    count += 1

Conditional while. #

When the while clause contains a condition, the loop continues to run until the condition is evaluated to false:

var running = true
var count = 0
while running:
    if count > 100:
        running = false
    count += 1

for range. #

for loops can iterate over a range that starts at an int (inclusive) to a target int (exclusive). The capture operator -> is used to capture the loop's counter variable:

for 0..4:
    performAction() 

for 0..100 -> i:
    print i    -- 0, 1, 2, ... , 99

To decrement the counter instead, use -..:

for 100-..0 -> i:
    print i    -- 100, 99, 98, ... , 1

When the range operator .. is replaced with ..=, the target int is inclusive: Planned Feature

for 0..=100 -> i:
    print i    -- 0, 1, 2, ... , 100

for each. #

The for clause can iterate over any type that implements the Iterable trait. An Iterable contains an iterator() method which returns an Iterator object. The for loop continually invokes the Iterator's next() method until none is returned.

Lists can be iterated since they implement the Iterable trait. The capture operator -> is used to capture the value returned from an iterator's next():

var list = {1, 2, 3, 4, 5}

for list -> n:
    print n

Maps can be iterated. next() returns a key and value tuple:

var map = Map{a=123, b=234}

for map -> entry:
    print entry[0]
    print entry[1]

Use the destructure syntax to extract the key and value into two separate variables:

for map -> {key, val}:
    print "key $(key) -> value $(val)"

for each with index. #

A counting index can be declared after the each variable. The count starts at 0 for the first value:

var list = {1, 2, 3, 4, 5}

for list -> val, i:
    print "index $(i), value $(val)"

Exit loop. #

Use break to exit a loop. This loop stops printing once i reaches 4:

for 0..10 -> i:
    if i == 4:
        break
    print i

Next iteration. #

Use continue to skip the rest of the loop and go to the next iteration. This loop prints 0 through 9 but skips 4:

for 0..10 -> i:
    if i == 4:
        continue
    print i

switch matching. #

The switch statement branches to a case block from a matching case condition. The expression that is matched against comes after switch statement. Multiple cases can be grouped together using a comma separator. An optional else fallback case is executed when no other cases were matched. Incomplete: Not all types can be used in the case conditions such as ranges.

var val = 1000
switch val
case 0..100:
    print 'at or between 0 and 99'
case 100:
    print 'val is 100'
case 200:
    print 'val is 200'
case 300, 400:
    print 'combined case'
else:
    print "val is $(val)"

The switch statement requires at least one case block or an else block. When the switch statement begins a new block, the case statements must be indented:

switch val:
    case 0: print 'a'
    case 1: print 'b'
    else: print 'c'

switch assignment. #

Although switch can not be used freely as an expression, it can be assigned to a left variable or destination:

var shu = switch pepper:
    case 'bell'     => 0
    case 'anaheim'  => 500
    case 'jalapeño' => 2000
    case 'serrano'  => 10000

When declaring an assignable switch, the cases must have a return value using the syntax case {cond} => {expr} or else => {expr}.

switch break. #

A break statement exits the current case block and resumes execution after the end of the switch statement: Planned Feature

switch value
case 0..5:
    print value
    if value == 3:
        break case
    print value    -- Skips second print if `value` is 3.

Try/Catch. #

The try catch statement, try else and try expressions provide a way to catch a throwing error and resume execution in a different branch. Learn more about Error Handling.

Deferred Execution. #

Planned Feature

Functions. #

^top

In Cyber, there are first-class functions (or function values) and static functions.

Static functions. #

Functions are declared with the func keyword and must have a name.

use math

func dist(x0 float, y0 float, x1 float, y1 float) float:
    var dx = x0-x1
    var dy = y0-y1
    return math.sqrt(dx^2 + dy^2)

Functions can be invoked with arguments:

print dist(0, 0, 10, 20)

Functions are initially static, but they can be passed around as a lambda or assigned to a variable:

-- Assigning to a local variable.
var bar = dist

func squareDist(dist dyn, size float) float:
    return dist(0.0, 0.0, size, size)
    
-- Passing `dist` as an argument.
print squareDist(dist, 30.0)

Functions can only return one value. However, the value can be destructured: Planned Feature

use {cos, sin} 'math'

func compute(rad float) [2]float:
    return .{cos(rad), sin(rad)}

var {x, y} = compute(pi)

Function overloading. #

Functions can be overloaded by their type signature:

func foo() int:
    return 2 + 2

func foo(n int) int:
    return 10 + n

func foo(n int, m int) int:
    return n * m

print foo()         --> 4
print foo(2)        --> 12
print foo(20, 5)    --> 100

Lambdas. #

Lambdas or function values can be assigned to variables or passed as arguments into other constructs.

When a lambda only returns an expression, it can be declared with a simplified syntax.

-- Passing simple lambda as an argument.
foo(word => toUpper(word))

-- A simple lambda with multiple arguments.
foo((word, prefix) => prefix + toUpper(word))

-- Assigning a simple lambda.
canvas.onUpdate = delta_ms => print delta_ms

Lambdas that need a block of statements can be declared with the func keyword without a name.

-- Assigning lambda block to a variable.
var add = func (a int, b int) int:
    return a + b

-- Passing a lambda block as an argument.
canvas.onUpdate():
    ..func (delta_ms):
        print delta_ms

Passing a lambda block as a call argument is only possible in a call block. Planned Feature See Function calls.

Closures. #

Lambdas can capture local variables from parent blocks. This example shows the lambda f capturing a from the main scope: Incomplete, only variables one parent block away can be captured.

var a = 1
var f = func() int:
    return a + 2
print f()         --> 3

The following lambda expression captures a from the function add:

func add():
    var a = 123
    return b => a + b
var addTo = add()
print addTo(10)   --> 133

Like static variables, static functions can not reference local variables outside of their scope:

var a = 1
func foo():
    print a       --> CompileError: Undeclared variable `a`.

extern functions. #

Planned Feature

Function types. #

A function type is denoted as func(P1, P2, ...) R where Ps are parameter types and R is the return type in addition to any function modifiers:

type AddFn -> func(int, int) int

Function types can include optional parameter names:

type AddFn -> func(a int, b int) int

If one parameter has a name, the other parameters must also have names. Parameter names do not alter the function signature and only serve to document the function type.

Only static functions can be assigned to a function type:

func add(a int, b int) int:
    return a + b

var fn AddFn = add
fn(10, 20)         --> 30

Similarily, a function union type is denoted as Func(P1, P2, ...) R and can store static functions, lambdas, and closures:

var c = 5
func addClosure(a int, b int) int:
    return a + b + c

var fn Func(int, int) int = add
fn(10, 20)         --> 30
fn = addClosure 
fn(10, 20)         --> 35

Function types declared in template parameters represent function symbol types. A function symbol can only be used at compile-time. It can be expanded to a runtime function:

type IntMap[K type, HASH func(K) int]
    ints [*]int

    -- Probing omitted.
    func get(self, key K) int:
        var slot = HASH(key) % self.ints.len
        return self.ints[slot]

func my_str_hash(s String) int:
    return s.len()

var m = IntMap[String, my_str_hash]{}

Parameter groups. #

When multiple parameters share the same type they can be declared together in a sequence:

func sum(a, b, c int) int
    return a + b + c

Named parameters. #

Planned Feature

Variadic parameters. #

Planned Feature

Function calls. #

The straightforward way to call a function is to use parentheses.

var d = dist(100, 100, 200, 200)

You can call functions with named parameters.

Planned Feature

var d = dist(x0=10, x1=20, y0=30, y1=40)

Shorthand syntax. #

Functions can be invoked without parentheses. This only works for calls that take at least one argument. Named arguments are not allowed and nested function calls must use parentheses:

-- Calls the function `dist`.
var d = dist 100, 100, 200, 200

func foo():       
    return 4

var r = foo       -- Returns the function itself as a value.
                     -- Does not call the function `random`.
r = foo()         -- Calls the function `random`.

Call block syntax. #

The call block appends a lambda to a call expression's last argument:

func Button(name String, size int, on_click func() void) ButtonConfig:
    return .{
        name = name,
        size = size,
        on_click = on_click,
    }

Button('Count', 10, ():
    print 'on click'

Piping. #

Planned Feature

Compile-time call. #

Any function can be invoked at compile-time using # before the arguments:

func fib(n int) int:
    if n < 2:
        return n
    return fib(n - 1) + fib(n - 2)

var res = fib#(30)
print res         --> 832040

The result of the function call is used instead of emitting a runtime function call. The call arguments must be compile-time compatible values.

There is no need for # if the caller is already in a compile-time context: Compile-time if is planned.

#if fib(30) == 832040:
    print 'Correct result.'

Function templates. #

Function declarations can include template parameters to create a function template:

func add[T](T type, a T, b T) T:
    return a + b

Only the template parameter names are declared. Their types are then inferred from the function signature.

Explicit template call. #

When the function is invoked with template argument(s), a new runtime function is generated and used:

print add(int, 1, 2)    --> 3
print add(float, 1, 2)  --> 3.0

Note that invoking the function again with the same template argument(s) uses the same generated function. In other words, the generated function is always memoized from the template arguments.

Expand function. #

The function template can be explicitly expanded to a runtime function:

var addInt = add[int]
print addInt(1, 2)      --> 3

Infer param type. #

When a template parameter is first encountered in a function parameter's type specifier, it's inferred from the argument's type:

func add[T](a T, b T) T:
    return a + b

print add(1, 2)         --> 3
print add(1.0, 2.0)     --> 3.0

In the above example, add[int] and add[float] were inferred from the function calls:

Nested template parameters can also be inferred:

func set[K, V](m Map[K, V], key K, val V):
    m.set(key, val)

Modules. #

^top

Modules have their own namespace and contain accessible static symbols. By default, importing another Cyber script returns a module with its declared symbols.

Importing. #

When a use declaration contains only a single identifier, it creates a local alias to a module using the identifier as the module specifier. Cyber's CLI comes with some builtin modules like math and test.

use test
test.eq(123, 123)

use math
print math.cos(0)

The explicit import syntax requires an alias name followed by a module specifier as a raw string:

use m 'math'
print m.random()

When Cyber is embedded into a host application, the module resolver and loader can be overridden using libcyber.

Import file. #

File modules can be imported:

-- Importing a module from the local directory.
use b 'bar.cy'
print b.myFunc()
print b.myVar

Import URL. #

Modules can be imported over the Internet:

-- Importing a module from a CDN.
use rl 'https://mycdn.com/raylib'

When importing using a URL without a file name, Cyber's CLI will look for a mod.cy from the path instead.

Import all. #

If the alias name is the wildcard character, all symbols from the module are imported into the using namespace: This feature is experimental and may be removed in a future version.

use * 'math'
print random()

Main module. #

Only the main module can have top-level statements that aren't static declarations. An imported module containing top-level statements returns an error:

-- main.cy
use a 'foo.cy'
print a.foo

-- foo.cy
use 'bar.cy'
var .foo = 123
print foo         -- Error: Top-level statement not allowed.

Circular imports. #

Circular imports are allowed. In the following example, main.cy and foo.cy import each other without any problems.

-- main.cy
use foo 'foo.cy'

func printB():
    foo.printC()

foo.printA()

-- foo.cy
use main 'main.cy'

func printA():
    main.printB()

func printC():
    print 'done'

Static variable declarations from imports can have circular references. Read more about this in Static variables.

Destructure import. #

Modules can also be destructured using the following syntax:

Planned Feature

use { cos, pi } 'math'
print cos(pi)

Exporting. #

All symbols are exported when the script's module is loaded. However, symbols can be declared with a hidden modifier.

func foo():         -- Exported static function.
    print 123

var .bar = 234      -- Exported static variable.

type Thing:  -- Exported type.
    var a float

Module URI. #

To get the absolute path of the current module, reference the compile-time constant #modUri. This can be used with os.dirName to get the current module directory.

print #modUri              -- Prints '/some/path/foo.cy'

use os
print os.dirName(#modUri)  -- Prints '/some/path'

Symbol visibility. #

All symbols have public visibility. However, when a symbol is declared with a - prefix, it suggests that it should not be made available to an editor's autocomplete:

-type Foo:
    a int
    b int

-func add(a int, b int) int:
    return a + b

Furthermore, the symbol is excluded when its module is included using use *.

Symbol alias. #

use can be used to create an alias to another symbol:

use eng 'lib/engine.cy'
use Vec2 -> eng.Vector2

Builtin modules. #

Builtin modules are the bare minimum that comes with Cyber. The embeddable library contains these modules and nothing more. They include:

• core: Commonly used utilities.
• cy: Cyber related functions.
• math: Math constants and functions.

mod core #

The core module contains functions related to Cyber and common utilities. It is automatically imported into each script's namespace.

Sample usage:

-- `print` and `typeof` are available without imports.
print 'hello'
print typeof('my str').id()

^topic

func allTypes() List[type]

func bitcast(D type, val S) D

func copy(val T) T

Copies a primitive value or creates a shallow copy of an object value.

func choicetag(choice T) T.Tag

Returns the tag of a choice. TODO: This will be moved to T.tag().

func dump(val any) void

Dumps a detailed description of a value.

func eprint(str any) void

Prints a value to the error stream. The host determines how it is printed.

func errorReport() String

func getObjectRc(val any) int

Returns the current reference count of an object.

func is(a any, b any) bool

Returns whether two values refer to the same instance.

func isAlpha(val int) bool

Returns whether a rune is an alphabetic letter.

func isDigit(val int) bool

Returns whether a rune is a digit.

func isNone(val any) bool

Returns whether a boxed value is the none case of a generic Option(T) type. This is temporary and may be removed in a future release.

func must(val any) any

If val is an error, panic(val) is invoked. Otherwise, val is returned.

func panic(err any) dyn

Stop execution in the current fiber and starts unwinding the call stack. See Unexpected Errors.

func performGC() Map

Runs the garbage collector once to detect reference cycles and abandoned objects. Returns the statistics of the run in a map value.

func ptrcast(D type, val S) *D

func print(str any) void

Prints a value. The host determines how it is printed.

func queueTask(fn ) void

Queues a callback function as an async task.

func refcast(T type, ptr *T)

func runestr(val int) String

Converts a rune to a string.

func sizeof(T type) int

func sizeof_(type_id int) int

func typeOf(t ExprType) type

Returns the type of an expression.

func typeInfo(T type) TypeInfo

Returns info about a type.

func typeid(T type) int

type void #

type bool #

func bool.$call(val any) bool

Converts any value to either true or false. Integers and floats equal to 0 return false. Empty strings return false. Otherwise, returns true.

type symbol #

type error #

func sym(self) symbol

Return the underlying symbol.

func error.$call(val any) error

Create an error from symbol.

type byte #

func $prefix~(self) byte

func $infix<(self, o byte) bool

func $infix<=(self, o byte) bool

func $infix>(self, o byte) bool

func $infix>=(self, o byte) bool

func $infix+(self, o byte) byte

func $infix-(self, o byte) byte

func $infix*(self, o byte) byte

func $infix/(self, o byte) byte

func $infix%(self, o byte) byte

func $infix^(self, o byte) byte

func $infix&(self, o byte) byte

func $infix|(self, o byte) byte

func $infix||(self, o byte) byte

func $infix<<(self, o int) byte

func $infix>>(self, o int) byte

func fmt(self, format NumberFormat) String

Formats the byte using a NumberFormat.

func fmt(self, format NumberFormat, config Table) String

opts.pad provides the ASCII rune that is used for padding with a string length of config.width.

func $call(b byte) byte

type int #

func $prefix~(self) int

func $prefix-(self) int

func $infix<(self, o int) bool

func $infix<=(self, o int) bool

func $infix>(self, o int) bool

func $infix>=(self, o int) bool

func $infix+(self, o int) int

func $infix-(self, o int) int

func $infix*(self, o int) int

func $infix/(self, o int) int

func $infix%(self, o int) int

func $infix^(self, o int) int

func $infix&(self, o int) int

func $infix|(self, o int) int

func $infix||(self, o int) int

func $infix<<(self, o int) int

func $infix>>(self, o int) int

func fmt(self, format NumberFormat) String

Formats the integer using a NumberFormat.

func fmt(self, format NumberFormat, config Table) String

opts.pad provides the ASCII rune that is used for padding with a string length of config.width.

func int.$call(val any) int

type float #

func $prefix-(self) float

func $infix<(self, o float) bool

func $infix<=(self, o float) bool

func $infix>(self, o float) bool

func $infix>=(self, o float) bool

func $infix+(self, o float) float

func $infix-(self, o float) float

func $infix*(self, o float) float

func $infix/(self, o float) float

func $infix%(self, o float) float

func $infix^(self, o float) float

func float.$call(val any) float

Converts the value to a float. Panics if type conversion fails.

type placeholder2 #

type placeholder3 #

type taglit #

type dyn #

type any #

type type #

func id(self) int

Returns a unique ID for this type.

type IntInfo #

type FloatInfo #

type HostObjectInfo #

type ObjectInfo #

type ObjectField #

type EnumInfo #

type EnumCase #

type ChoiceInfo #

type ChoiceCase #

type FuncInfo #

type FuncParam #

type ArrayInfo #

type StructInfo #

type StructField #

type TraitInfo #

type PointerInfo #

type OptionInfo #

func type.$call(val any) type

Return the type of a value. See typeof to obtain the type of an expression at compile-time.

type List #

func $index(self, idx int) T

func $index(self, range Range) List[T]

func $setIndex(self, idx int, val T) void

func append(self, val T) void

Appends a value to the end of the list.

func appendAll(self, list List[T]) void

Appends the elements of another list to the end of this list.

func insert(self, idx int, val T) void

Inserts a value at index idx.

func iterator(self) ListIterator[T]

Returns a new iterator over the list elements.

func join(self, sep String) String

Returns a new string that joins the elements with separator.

func len(self) int

Returns the number of elements in the list.

func remove(self, idx int) void

Removes an element at index idx.

func resize(self, size int) void

Resizes the list to len elements. If the new size is bigger, none values are appended to the list. If the new size is smaller, elements at the end of the list are removed.

func resize_(self, elem_t int, size int) void

func sort(self, lessFn ) void

Sorts the list with the given less function. If element a should be ordered before b, the function should return true otherwise false.

func List.fill(val T, n int) List[T]

Creates a list with initial capacity of n and values set to val. If the value is an object, it is shallow copied n times.

type ListIterator #

func next(self) ?T

func next_(self, ret_t int) ?T

type Tuple #

func $index(self, idx int) any

type FuncSig #

type funcptr_t #

type funcunion_t #

type Func #

type funcsym_t #

type Table #

func $initPair(self, key String, value any) void

func $get(self, name String) dyn

func $set(self, name String, value any)

func $index(self, key any) dyn

func $setIndex(self, key any, value any) void

type Map #

func $initPair(self, key any, value any) void

func $index(self, key any) dyn

func $setIndex(self, key any, val any) void

func contains(self, key any) bool

Returns whether there is a value mapped to key.

func get(self, key any) ?any

Returns value mapped to key or returns none.

func remove(self, key any) bool

Removes the element with the given key key.

func size(self) int

Returns the number of key-value pairs in the map.

func iterator(self) MapIterator

Returns a new iterator over the map elements.

type MapIterator #

func next(self) ?any

type String #

func $index(self, idx int) int

Returns the rune at byte index idx. The replacement character (0xFFFD) is returned for an invalid UTF-8 rune.

func $index(self, range Range) String

Returns a slice into this string from a Range with start (inclusive) to end (exclusive) byte indexes.

func $infix+(self, o any) String

Returns a new string that concats this string and str.

func concat(self, o String) String

Returns a new string that concats this string and str.

func count(self) int

Returns the number of runes in the string.

func endsWith(self, suffix String) bool

Returns whether the string ends with suffix.

func find(self, needle String) ?int

Returns the first byte index of substring needle in the string or none if not found. SIMD enabled.

func findAnyByte(self, set List[byte]) ?int

Returns the first index of any byte in set or none if not found.

func findAnyRune(self, runes List[int]) ?int

Returns the first byte index of any rune in runes or none if not found. SIMD enabled.

func findByte(self, b byte) ?int

Returns the first index of byte in the array or none if not found.

func findRune(self, rune int) ?int

Returns the first byte index of a rune needle in the string or none if not found. SIMD enabled.

func fmtBytes(self, format NumberFormat) String

Formats each byte in the string using a NumberFormat. Each byte is zero padded.

func getByte(self, idx int) byte

Returns the byte value (0-255) at the given index idx.

func getInt(self, idx int, endian symbol) int

Returns the int value of the 6 bytes starting from idx with the given endianness (.little or .big).

func getInt32(self, idx int, endian symbol) int

Returns the int value of the 4 bytes starting from idx with the given endianness (.little or .big).

func insert(self, idx int, str String) String

Returns a new string with str inserted at byte index idx.

func insertByte(self, idx int, byte int) String

Returns a new array with byte inserted at index idx.

func isAscii(self) bool

Returns whether the string contains all ASCII runes.

func len(self) int

Returns the byte length of the string. See count() to obtain the number of runes.

func less(self, other String) bool

Returns whether this string is lexicographically before other.

func lower(self) String

Returns this string in lowercase.

func replace(self, needle String, replacement String) String

Returns a new string with all occurrences of needle replaced with replacement.

func repeat(self, n int) String

Returns a new string with this string repeated n times.

func seek(self, idx int) int

Returns the starting byte index for the rune index idx.

func sliceAt(self, idx int) String

Returns the UTF-8 rune starting at byte index idx as a string.

func split(self, sep String) List[String]

Returns a list of UTF-8 strings split at occurrences of sep.

func startsWith(self, prefix String) bool

Returns whether the string starts with prefix.

func trim(self, mode symbol, delims String) String

Returns the string with ends trimmed from runes in delims. mode can be .left, .right, or .ends.

func upper(self) String

Returns this string in uppercase.

func String.$call(val any) String

Converts a value to a string.

type array_t #

type Array #

func $index(self, idx int)

func iterator(self) RefSliceIterator[T]

Returns a new iterator over the array.

func len(self) int

Returns the number of elements in the array.

type pointer #

func $index(self, idx int) *T

func $index(self, range Range) [*]T

func $setIndex(self, idx int, val T) void

func addr(self) int

When pointer runtime safety is enabled, this returns the raw pointer address as an int64. Otherwise, the pointer itself is bitcasted to an int64.

func asObject(self) any

Casts the pointer to a Cyber object. The object is retained before it's returned.

func fromCstr(self, offset int) String

Returns a String from a null terminated C string.

func get(self, offset int, ctype symbol) dyn

Dereferences the pointer at a byte offset and returns the C value converted to Cyber.

func getString(self, offset int, len int) String

Returns a String with a copy of the byte data starting from an offset to the specified length.

func set(self, offset int, ctype symbol, val any) void

Converts the value to a compatible C value and writes it to a byte offset from this pointer.

func pointer.$call(T type, addr int) *T

Converts an int to a pointer value.

type ExternFunc #

func addr(self) int

Returns the memory address as an int. The value may be negative since it's bitcasted from an unsigned 48-bit integer but it retains the original pointer bits.

func ptr(self) *void

type Fiber #

func status(self) symbol

type Range #

type TccState #

type Future #

func Future.complete(val T) Future[T]

Returns a Future[T] that has a completed value.

func Future.new(T type) Future[T]

type FutureResolver #

func complete(self, val T) void

func future(self) Future[T]

func FutureResolver.new(T type) FutureResolver[T]

type Ref #

type RefSlice #

func $index(self, idx int)

func $index(self, range Range)

func $setIndex(self, idx int, val T) void

func iterator(self) RefSliceIterator[T]

func len(self) int

type RefSliceIterator #

func next(self) ?T

type PtrSlice #

func $index(self, idx int) *T

func $index(self, range Range) [*]T

func $setIndex(self, idx int, val T) void

func endsWith(self, suffix [*]T) bool

Returns whether the array ends with suffix.

func findScalar(self, needle T) ?int

Returns the first index of needle in the slice or none if not found.

func iterator(self) PtrSliceIterator[T]

func len(self) int

func startsWith(self, prefix [*]T) bool

Returns whether the array starts with prefix.

type PtrSliceIterator #

func next(self) ?T

type IMemory trait #

func alloc(self, len int) [*]byte

func free(self, buf [*]byte) void

type Memory #

func Memory.new(self, T type) *T

func Memory.alloc(self, T type, n int) [*]T

func Memory.free(self, ptr *T)

func Memory.free_(self, slice [*]T)

type DefaultMemory #

func alloc(self, len int) [*]byte

func free(self, buf [*]byte) void

type ExprType #

func getType(self) type

mod cy #

The cy module contains functions related to the Cyber language.

Sample usage:

use cy
print cy.toCyon([1, 2, 3])

^topic

var Success int

var Await int

var ErrorCompile int

var ErrorPanic int

var TypeVoid int

func eval(src String) any

Evaluates source code in an isolated VM. If the last statement is an expression, a primitive or a String can be returned.

func parse(src String) Map

Parses Cyber source string into a structured map object. Currently, only metadata about static declarations is made available but this will be extended to include an AST.

func parseCyon(src String) any

Parses a CYON string into a value.

func repl(read_line any) void

Starts an isolated REPL session. The callback read_line(prefix String) String is responsible for obtaining the input.

func toCyon(val any) String

Encodes a value to CYON string.

type EvalConfig #

type EvalResult #

type Value #

func dump(self) String

func getTypeId(self) int

func toHost(self) any

type VM #

func eval(self, code String) EvalResult

func eval(self, uri String, code String, config EvalConfig) EvalResult

func getErrorSummary(self) String

func getPanicSummary(self) String

func VM.new() VM

Create an isolated VM.

type REPL #

func printIntro(self)

func read(self, read_line dyn) ?String

func evalPrint(self, code String) void

func getPrefix(self) String

func REPL.new() REPL

mod math #

The math module contains commonly used math constants and functions.

Sample usage:

use math

var r = 10.0
print(math.pi * r^2)

^topic

var e float

Euler's number and the base of natural logarithms; approximately 2.718.

var inf float

Infinity.

var log10e float

Base-10 logarithm of E; approximately 0.434.

var log2e float

Base-2 logarithm of E; approximately 1.443.

var ln10 float

Natural logarithm of 10; approximately 2.303.

var ln2 float

Natural logarithm of 2; approximately 0.693.

var maxSafeInt float

The maximum integer value that can be safely represented as a float. 2^53-1 or 9007199254740991.

var minSafeInt float

The minumum integer value that can be safely represented as a float. -(2^53-1) or -9007199254740991.

var nan float

Not a number. Note that nan == nan. However, if a nan came from an arithmetic operation, the comparison is undefined. Use isNaN instead.

var neginf float

Negative infinity.

var pi float

Ratio of a circle's circumference to its diameter; approximately 3.14159.

var sqrt1_2 float

Square root of ½; approximately 0.707.

var sqrt2 float

Square root of 2; approximately 1.414.

func abs(a float) float

Returns the absolute value of x.

func acos(a float) float

Returns the arccosine of x.

func acosh(a float) float

Returns the hyperbolic arccosine of x.

func asin(a float) float

Returns the arcsine of x.

func asinh(a float) float

Returns the hyperbolic arcsine of a number.

func atan(a float) float

Returns the arctangent of x.

func atan2(a float, b float) float

Returns the arctangent of the quotient of its arguments.

func atanh(a float) float

Returns the hyperbolic arctangent of x.

func cbrt(a float) float

Returns the cube root of x.

func ceil(a float) float

Returns the smallest integer greater than or equal to x.

func clz32(a float) float

Returns the number of leading zero bits of the 32-bit integer x.

func cos(a float) float

Returns the cosine of x.

func cosh(a float) float

Returns the hyperbolic cosine of x.

func exp(a float) float

Returns e^x, where x is the argument, and e is Euler's number (2.718…, the base of the natural logarithm).

func expm1(a float) float

Returns subtracting 1 from exp(x).

func floor(a float) float

Returns the largest integer less than or equal to x.

func frac(a float) float

Returns the fractional or decimal part of a float value.

func hypot(a float, b float) float

Returns the square root of the sum of squares of its arguments.

func isInt(a float) bool

Returns true if the float has no fractional part, otherwise false.

func isNaN(a float) bool

Returns whether x is not a number.

func ln(a float) float

Returns the natural logarithm (㏒e; also, ㏑) of x.

func log(a float, b float) float

Returns the logarithm of y with base x.

func log10(a float) float

Returns the base-10 logarithm of x.

func log1p(a float) float

Returns the natural logarithm (㏒e; also ㏑) of 1 + x for the number x.

func log2(a float) float

Returns the base-2 logarithm of x.

func max(a float, b float) float

Returns the largest of two numbers.

func min(a float, b float) float

Returns the smallest of two numbers.

func mul32(a float, b float) float

Returns the result of the 32-bit integer multiplication of x and y. Integer overflow is allowed.

func pow(a float, b float) float

Returns base x to the exponent power y (that is, x^y).

func random() float

Returns a pseudo-random number between 0 and 1.

func round(a float) float

Returns the value of the number x rounded to the nearest integer.

func sign(a float) float

Returns the sign of the x, indicating whether x is positive, negative, or zero.

func sin(a float) float

Returns the sine of x.

func sinh(a float) float

Returns the hyperbolic sine of x.

func sqrt(a float) float

Returns the positive square root of x.

func tan(a float) float

Returns the tangent of x.

func tanh(a float) float

Returns the hyperbolic tangent of x.

func trunc(a float) float

Returns the integer portion of x, removing any fractional digits.

Std modules. #

Std modules come with Cyber's CLI. They include:

• cli: Related to the command line.
• os: System level functions.
• test: Utilities for testing.

mod cli #

The cli module contains functions related to the command line.

Sample usage:

use cli
cli.repl()

^topic

func repl() void

Starts an isolated REPL session. Invokes cy.repl(replReadLine).

func replReadLine(prefix String) String

Default implementation to read a line from the CLI for a REPL.

mod os #

Cyber's os module contains system level functions. It's still undecided as to how much should be included here so it's incomplete. You can still access os and libc functions yourself using Cyber's FFI or embedding API.

Sample usage:

use os

var map = os.getEnvAll()
for map -> {k, v}:
    print "$(k) -> $(v)"

^topic

var cpu String

The current cpu arch's tag name.

var endian symbol

The current arch's endianness: .little, .big

var stderr File

Standard error file descriptor.

var stdin File

Standard input file descriptor.

var stdout File

Standard output file descriptor.

var system String

The current operating system's tag name.

var vecBitSize int

Default SIMD vector bit size.

func access(path String, mode symbol) void

Attempts to access a file at the given path with the .read, .write, or .readWrite mode. Throws an error if unsuccessful.

func args() List[String]

Returns the command line arguments in a List[String].

func cacheUrl(url String) String

Returns the path of a locally cached file of url. If no such file exists locally, it's fetched from url.

func copyFile(srcPath String, dstPath String) void

Copies a file to a destination path.

func createDir(path String) void

Creates the directory at path. Returns true if successful.

func createFile(path String, truncate bool) File

Creates and opens the file at path. If truncate is true, an existing file will be truncated.

func cstr(s String) *void

Returns a null terminated C string.

func cwd() String

Returns the current working directory.

func dirName(path String) ?String

Returns the given path with its last component removed.

func execCmd(args List[String]) Map

Runs a shell command and returns the stdout/stderr.

func exePath() String

Returns the current executable's path.

func exit(status int) void

Exits the program with a status code.

func fetchUrl(url String) String

Fetches the contents at url using the HTTP GET request method.

func free(ptr *void) void

Frees the memory located at ptr.

func getEnv(key String) ?String

Returns an environment variable by key.

func getEnvAll() Map

Returns all environment variables as a Map.

func malloc(size int) *void

Allocates size bytes of memory and returns a pointer.

func milliTime() float

Return the calendar timestamp, in milliseconds, relative to UTC 1970-01-01. For an high resolution timestamp, use now().

func newFFI() FFI

Returns a new FFI context for declaring C mappings and binding a dynamic library.

func now() float

Returns the current time (in high resolution seconds) since an arbitrary point in time.

func openDir(path String) Dir

Invokes openDir(path, false).

func openDir(path String, iterable bool) Dir

Opens a directory at the given path. iterable indicates that the directory's entries can be iterated.

func openFile(path String, mode symbol) File

Opens a file at the given path with the .read, .write, or .readWrite mode.

func parseArgs(options List[dyn]) Table

Given expected ArgOptions, returns a Table of the options and a rest entry which contains the non-option arguments.

func readAll() String

Reads stdin to the EOF as a UTF-8 string. To return the bytes instead, use stdin.readAll().

func readFile(path String) String

Reads the file contents from path as a UTF-8 string. To return the bytes instead, use File.readAll().

func readLine() String

Reads stdin until a new line as a String. This is intended to read user input from the command line. For bulk reads from stdin, use stdin.

func realPath(path String) String

Returns the absolute path of the given path.

func removeDir(path String) void

Removes an empty directory at path. Returns true if successful.

func removeFile(path String) void

Removes the file at path. Returns true if successful.

func setEnv(key String, val String) void

Sets an environment variable by key.

func sleep(ms float) void

Pauses the current thread for given milliseconds.

func unsetEnv(key String) void

Removes an environment variable by key.

func writeFile(path String, contents String) void

Writes contents as a string or bytes to a file.

type File #

func close(self) void

Closes the file handle. File ops invoked afterwards will return error.Closed.

func iterator(self) File

func next(self) String

func read(self, n int) String

Reads at most n bytes as an Array. n must be at least 1. A result with length 0 indicates the end of file was reached.

func readAll(self) String

Reads to the end of the file and returns the content as an Array.

func seek(self, n int) void

Seeks the read/write position to pos bytes from the start. Negative pos is invalid.

func seekFromCur(self, n int) void

Seeks the read/write position by pos bytes from the current position.

func seekFromEnd(self, n int) void

Seeks the read/write position by pos bytes from the end. Positive pos is invalid.

func stat(self) Map

Returns info about the file as a Map.

func streamLines(self) File

Equivalent to streamLines(4096).

func streamLines(self, bufSize int) File

Returns an iterable that streams lines ending in \n, \r, \r\n, or the EOF. The lines returned include the new line character(s). A buffer size of bufSize bytes is allocated for reading. If \r is found at the end of the read buffer, the line is returned instead of waiting to see if the next read has a connecting \n.

func write(self, val String) int

Writes a String at the current file position. The number of bytes written is returned.

type Dir #

func iterator(self) DirIterator

Returns a new iterator over the directory entries. If this directory was not opened with the iterable flag, error.NotAllowed is returned instead.

func stat(self) Map

Returns info about the file as a Map.

func walk(self) DirIterator

Returns a new iterator over the directory recursive entries. If this directory was not opened with the iterable flag, error.NotAllowed is returned instead.

type DirIterator #

func next(self) ?Map

type FFI #

func bindCallback(self, fn any, params List[dyn], ret symbol) ExternFunc

Creates an ExternFunc that contains a C function pointer with the given signature. The extern function is a wrapper that calls the provided user function. Once created, the extern function is retained and managed by the FFI context.

func bindLib(self, path ?String) any

Calls bindLib(path, [:]).

func bindLib(self, path ?String, config Table) any

Creates a handle to a dynamic library and functions declared from cfunc. By default, an anonymous object is returned with the C-functions binded as the object's methods. If config contains gen_table: true, a Table is returned instead with C-functions binded as function values.

func bindObjPtr(self, obj any) *void

Returns a Cyber object's pointer. Operations on the pointer is unsafe, but it can be useful when passing it to C as an opaque pointer. The object is also retained and managed by the FFI context.

func cbind(self, mt type, fields List[dyn]) void

Binds a Cyber type to a C struct.

func cfunc(self, name String, params List[dyn], ret any) void

Declares a C function which will get binded to the library handle created from bindLib.

func new(self, ctype symbol) *void

Allocates memory for a C struct or primitive with the given C type specifier. A pointer to the allocated memory is returned. Eventually this will return a cpointer instead which will be more idiomatic to use.

func unbindObjPtr(self, obj any) void

Releases the object from the FFI context. External code should no longer use the object's pointer since it's not guaranteed to exist or point to the correct object.

type CArray #

type CDimArray #

Map DirEntry #

Map DirWalkEntry #

Table ArgOption #

mod test #

The test module contains utilities for testing.

Sample usage:

use t 'test'

var a = 123 + 321
t.eq(a, 444)

^topic

func assert(pred bool) void

Panics if pred is false.

func eq(a T, b T) bool

Returns whether two values are equal. Panics with error.AssertError if types or values do not match up.

func eqList(a List[T], b List[T]) bool

Returns true if two lists have the same size and the elements are equal as if eq was called on those corresponding elements.

func eqSlice(a , b ) bool

Returns true if two slices have the same size and the elements are equal as if eq was called on those corresponding elements.

func eqNear(a T, b T) bool

Returns true if two numbers are near each other within epsilon 1e-5.

func fail()

func fail(msg String)

func throws(fn any, err error)

Asserts that an error was thrown when invoking a function.

FFI. #

• FFI context.
• Declare functions.
• Bind library.
  • Search path.
  • Configuration.
  • Finalizer.
• Mappings.
• Bind to Cyber type.
• cbindgen.cy

^top

Cyber supports binding to an existing C ABI compatible library at runtime. This allows you to call into dynamic libraries created in C or other languages. Cyber uses libtcc to JIT compile the bindings so function calls are fast. The example shown below can be found in Examples.

FFI context. #

An FFI context contains declarations that map C to Cyber. Afterwards, it allows you to bind to a dynamic library or create interoperable objects. To create a new FFI context:

use os

var ffi = os.newFFI()

Declare functions. #

Functions from a library are first declared using cfunc which accepts C types in the form of symbols. In a future update they will accept C syntax instead.

ffi.cfunc('add', {.int, .int}, .int)

The first argument refers to the symbol's name in the dynamic library. The second argument contains the function's parameter types and finally the last argument is the function's return type.

The example above maps to this C function:

int add(int a, int b) {
    return a + b;
}

Bind library. #

bindLib accepts the path to the library and returns a object which can be used to invoke the functions declared from cfunc:

dyn lib = ffi.bindLib('./mylib.so')
lib.add(123, 321)

Note that dyn is used to allow lib to be used dynamically since the type is unknown at compile-time.

Search path. #

If the path argument to bindLib is just a filename, the search steps for the library is specific to the operating system. Provide an absolute (eg. '/foo/mylib.so') or relative (eg. './mylib.so') path to load from a direct location instead. When the path argument is none, it loads the currently running executable as a library allowing you to bind exported functions from the Cyber CLI or your own application/runtime.

Configuration. #

By default bindLib returns an anonymous object with the binded C-functions as methods. This is convenient for invoking functions using the method call syntax. If a config is passed into bindLib as the second argument, gen_table=true makes bindLib return a table instead with the binded C-functions as Cyber functions.

Finalizer. #

The resulting object of bindLib holds a reference to an internal TCCState which owns the loaded JIT code. Once the object is released by ARC, the TCCState is also released which removes the JIT code from memory.

Mappings. #

When using cfunc or cbind declarations, symbols are used to represent default type mappings from Cyber to C and back:

Incomplete: This is not the final API for dynamically loading and interfacing with C libraries. The plan is to parse a subset of C headers to bind to Cyber types and functions.

1. Use os.cstr() and pointer.fromCstr() to convert between a Cyber string and a null terminated C string.
2. The mapping from a Cyber object type S and the C-struct can be declared with cbind.

Bind to Cyber type. #

cbind is used to bind a C struct to a Cyber object type. Once declared, the Cyber type can be used as a binding type in function declarations:

use os

type MyObject:
    a float
    b *void
    c bool

ffi.cbind(MyObject, {.float, .voidPtr, .bool})
ffi.cfunc('foo', {MyObject}, MyObject)
dyn lib = ffi.bindLib('./mylib.so')

var res = lib.foo(MyObject{a=123.0, b=os.cstr('foo'), c=true})

The example above maps to these C declarations in mylib.so:

typedef struct MyObject {
    double a;
    char* b;
    bool c;
} MyObject;

MyObject foo(MyObject o) {
    // Do something.
}

cbind also generates ptrTo[Type] as a helper function to dereference an opaque ptr to a new Cyber object:

ffi.cfunc('foo', {MyObject}, .voidPtr)
dyn lib = ffi.bindLib('./mylib.so')

var ptr = lib.foo(MyObject{a=123, b=os.cstr('foo'), c=true})
var res = lib.ptrToMyObject(ptr)

cbindgen.cy #

cbindgen.cy is a Cyber script that automatically generates bindings given a C header file. Some example bindings that were generated include: Raylib and LLVM.

Error Handling. #

^top

Cyber provides a throw/catch mechanism to handle expected errors. For unexpected errors, panics can be used as a fail-fast mechanism to abort the currently running fiber.

Error value. #

error literal. #

An error value contains a symbol. They can be created without a declaration using the error literal:

var err = error.Oops

Use sym() to obtain the underlying symbol:

print err.sym()   -- Prints ".Oops"

Since error is a primitive value, it can be compared using the == operator.

if err == error.Oops:
    handleOops()

-- Alternatively.
if err.sym() == .Oops:
    handleOops()

error payload. #

An payload value can be attached when throwing an error value. Planned Feature

error set type. #

Error set types enumerate error values that belong to the same group of errors: Planned Feature

type MyError error:
    case boom
    case badArgument
    case nameTooLong

var err = MyError.nameTooLong

Throwing errors. #

Use the throw keyword to throw errors. A thrown error continues to bubble up the call stack until it is caught by a try block or try expression.

func fail():
    throw error.Oops      -- Throws an error with the symbol `#Oops`

func fail2():
    throw 123             -- panic: Can only throw an `error` value.

throw can also be used as an expression.

func fail():
    var a = false or throw error.False

Catching errors. #

try block. #

The try block catches thrown errors and resumes execution in a followup catch block:

try:
    funcThatCanFail()
catch err:
    print err      -- 'error.Failed'

caught variable. #

The contextual caught variable is used to reference the caught error: Planned Feature

try:
    funcThatCanFail()
catch:
    print caught   -- 'error.Failed'

catch matching. #

An inner catch block contains a matching clause: Planned Feature

try:
    funcThatCanFail()
catch error.BadDay:
    eatSnack()
catch:
    print caught

Error sets can be matched: Planned Feature

try:
    funcThatCanFail()
catch MyError.Boom:
    print 'Kaboom!'
catch:
    print caught

try expression. #

The try expression either returns a non-error result or the default value from the catch clause:

var res = try funcThatCanFail() catch 123
print res         -- '123'

Since errors bubble up automatically, any errors thrown from sub-expressions are also caught:

var res = try happyFunc(funcThatCanFail()) catch 123
print res         -- '123'

Value or error. #

When the catch clause is omitted, the try expression will return either the value or the error:

var res = try funcThatCanFail()
if res == error.Failed:
    print 'Result is an error.'

Semantic checks. #

Throws modifier. #

The throws modifier ! indicates that a function contains a throwing expression that was not caught with try catch.

The modifier is attached to the function return type as a prefix:

func foo() !void:
    throw error.Failure

This declaration indicates the function can either return an int type or throw an error:

func result(cond bool) !int:
    if cond:
        return 123
    else:
        throw error.Failure

Throws check. #

The compiler requires a throws modifier if the function contains an uncaught throwing expression: Planned Feature

func foo(a int) int:
    if a == 10:
        throw error.Failure   --> CompileError.
    else:
        return a * 2

--> CompileError. Uncaught throwing expression.
--> `foo` requires the `!` throws modifier or
--> the expression must be caught with `try`.

Stack trace. #

When an uncaught error bubbles up to the top, its stack trace from the throw callsite is dumped to the console. The builtin errorTrace() and errorReport() are used to obtain the stack trace info.

try:
    funcThatCanFail()
catch:
    -- Prints the stack trace summary of the caught error.
    print errorReport()

    -- Provides structured info about the stack trace.
    var info = errorTrace()
    print info.frames.len()

Unexpected errors. #

An unexpected error is an error that is not meant to be handled at runtime.

Panics. #

The builtin panic is used as a fail-fast mechanism to quickly exit the current fiber with an error payload:

func kaboom():
    panic(error.danger)

kaboom()     -- Script ends and prints the stack trace.

Panics can not be caught using try catch. Once panic is invoked, the current fiber stops execution and begins to unwind its call stack. Once the error is propagated to the root, the fiber ends and transitions to a panic state. If the main fiber ends this way, the VM begins to shutdown. Otherwise, execution resumes on the next fiber which allows recovery from a panic.

Concurrency. #

^top

Cyber supports asynchronous and cooperative concurrency.

Async. #

Cyber supports asynchronous execution which provides preemptive concurrency. Tasks can be scheduled over a single thread. Multi-threaded execution is planned.

Futures. #

A Future is a promise that some work will either complete or fail at some point in the future. This abstraction allows the current thread to continue execution without waiting for the completion of the future.

The asynchronous work encapsulated by the future has the opportunity to run in parallel. For example, I/O bound work can be delegated to the operating system and CPU bound work can be run across multiple threads. Some work may simply run on the same thread.

If an API function is meant to do work asynchronously, it would return a Future:

use aio

var f = aio.delay(1000)
print f          --> Future(void)

Futures can hold a result value when they are completed:

use aio

var f = aio.readFile('foo.txt')
print f          --> Future(String)

Futures can be created with a completed value:

var f = Future.complete(100)
print f          --> Future(int)
print f.get().?  --> 100

Future are also created when composing them together.

await. #

The await expression asynchronously waits for a Future to resolve. It guarantees that the current execution flow will not resume until the future is completed. Once resumed, the expression evaluates to the completed result:

use aio

var res = await aio.readFile('foo.txt')
print res        --> bar

In the above example, the file contents of foo.txt is "bar".

await suspends the current fiber so the scheduler can execute tasks in its ready queue. When the future resolves, the suspended task is moved to the ready queue waiting to resume execution.

Performing await on other values besides a Future type will evaluate to the value and will not perform any waiting:

var v = await 123
print v          --> 123

Colorless async. #

await can be used in any function. This means that async functions are colorless and don't require a special function modifier.

Future chains. #

Future.then is used to attach a callback that will only be invoked when the future completes, thereby creating an asynchronous chain: Planned Feature

use aio

var f = aio.readFile('foo.txt').then() -> res:
    print res    --> bar

print f          --> Future(void)

By default Future.then evaluates to a new Future(void) and represents the completion of the callback. A callback that returns a result requires a generic parameter or inferred from the lambda: Planned Feature

var f = aio.readFile('foo.txt').then() -> res:
    return res.len()

print f          --> Future(int)
print await f    --> 3

Similarily, Future.catch attaches a callback that will only be invoked when the future fails: Planned Feature

var f = aio.readFile('foo.txt').catch() -> err:
    print err    --> error.FileNotFound

Resolving futures. #

FutureResolver produces a completable Future. In the following example, a future is completed with the FutureResolver after a queued task runs:

var r = FutureResolver[int].new()

queueTask():
    r.complete(234)

var v = await r.future()
print v

Structured concurrency. #

Planned Feature

Asychronous tasks must be created with an Async context which groups together related tasks. This gives tasks well-defined lifetimes and allows policies to be applied to a task group.

Every program begins by pushing a new Async.waitAll context to the async context variable. waitAll is a finish policy that waits for all tasks to complete before it marks its finish task as completed. In the following example, the created tasks are attached to the current Async context:

use aio

aio.delay(1000)
addMoreTasks()

Since the program did not explicitly wait for the completion of the current Async context, execution will end before 1 second has elapsed.

To wait for all the tasks to complete, the program can wait for the Async context to finish:

use aio

context async Async

aio.delay(1000)
addMoreTasks()

await async.finish()

The context variable async is declared to access the current Async context. finish applies the finish policy waitAll which completes when all tasks under the async context have been completed.

After invoking finish, new tasks can no longer be created under the async context. Doing so would result in a runtime error. A future is returned to indicate the completion of the finish task.

Explicitly invoking finish isn't idiomatic and was only demonstrated to show how an async context works. The same feat can be achieved with helpers that can wrap a block of related async logic:

use aio

await Async.block(.waitAll):
    aio.delay(1000)
    addMoreTasks()

Async.block begins by creating a new async context and pushing the context to async. The callback is then invoked which creates a new task under the async context. Once the callback returns, the async context and the result of Async.finish is returned.

Finish policies. #

Planned Feature

Task cancellation. #

Planned Feature

Threads. #

Planned Feature

Fibers. #

A fiber represents a separate execution context as a first-class value. It contains it's own call stack and program counter. Fibers by themselves do not enable parallelism.

Creating fibers. #

The coinit keyword creates and returns a new fiber using a function as the entry point:

var count = 0

var foo = func ():
    count += 1
    coyield
    count += 1

var task = coinit(foo)

print count          -- '0'
coresume task
print count          -- '1'
coresume task
print count          -- '2'

A fiber does not start execution until coresume is invoked on it. coyield pauses the current fiber and execution is returned to the previous fiber that invoked coresume.

Passing arguments. #

Arguments after the callee are passed into the entry function:

var count = 0

var increment = func (inc):
    count += inc

var task = coinit(increment, 5)
coresume task
print count          -- '5'

When the fiber is created, the arguments are saved inside the fiber's stack. Once the first coresume is invoked, the entry function is invoked with the saved arguments.

Reset state. #

To reset a fiber to its initial state, invoke reset(). Planned Feature When reset, the existing stack is unwinded, the program counter returns to the starting point, and the state is set to .init:

func fib(n int) int:
    coyield n
    if n < 2:
        return n
    return fib(n - 1) + fib(n - 2)

var task = coinit(fib, 10)

-- Progress the fiber...
print(coresume task)    -- Prints "10"
print(coresume task)    -- Prints "9"
print(coresume task)    -- Prints "8"

-- Reset back to the start with the `.init` state.
task.reset()
print(coresume task)    -- Prints "10"

Rebinding arguments. #

Arguments attached to the fiber can be rebinded with a different set of values. Planned Feature This allows fiber reuse, instead of creating a new fiber:

var task = coinit(fib, 10)

-- Run task to completion.
var res = 0
while task.status() != .done:
    res = coresume fiber
print res

task.reset()
task.bindArgs(20)

-- Run task again with the new argument...

Fiber block. #

A fiber block is used to construct a fiber without an entry function. Planned Feature The counting example can be rewritten to:

var count = 0

var task = coinit:
    count += 1       -- `count is captured`
    coyield
    count += 1

print count          -- '0'
coresume task
print count          -- '1'
coresume task
print count          -- '2'

Referencing parent variables from the fiber block automatically captures them just like a function closure.

Pause and resume. #

coyield can be used anywhere in a fiber's call stack to pause execution and return to the previous fiber.

func foo():
    print 'foo'
    bar()

func bar():
    -- Nested coyield in call stack.
    coyield
    print 'bar'

var task = coinit(foo)
coresume task

coresume also returns the resulting value.

func foo():
    return 123

var task = coinit(foo)
print(coresume task)    -- '123'

coyield can return a value back to coresume. Planned Feature

Fiber state. #

Use Fiber.status() to get the current state of the fiber.

func foo():
    coyield
    print 'done'

var task = coinit(foo)
print task.status()   -- '.paused'
coresume task
print task.status()   -- '.paused'
coresume task
print task.status()   -- '.done'

The main execution context is a fiber as well. Once the main fiber has finished, the VM is done and control is returned to the host.

Gas mileage. #

Planned Feature

Dynamic Typing. #

• Dynamic variables.
• Runtime type checking.
• Dynamic tables.
• Dynamic inference.

^top

Dynamic typing is supported with a less restrictive syntax. This can reduce the amount of friction when writing code, but it can also result in more runtime errors.

In Cyber, the dyn keyword is used exclusively for dynamic declarations.

Dynamic variables. #

Variables declared with dyn are implicitly given the dyn type:

dyn a = 123

Typically a dynamic variable defers type checking to runtime, but if the compiler determines that an operation will always fail at runtime, a compile error is reported instead:

dyn a = '100'

print a / 2
--> CompileError: Can not find the symbol `$infix/` in `String`

When a is assigned a different type of value, its recent type is updated so the compiler can continue to surface errors ahead of time:

a = {1, 2, 3}
print a / 2
--> CompileError: Can not find the symbol `$infix/` in `List`

Runtime type checking. #

If the type of a dynamic variable can not be determined at compile-time, type checking is deferred to runtime.

In this example, the type for a is unknown after assigning the return of a dynamic call to erase. Any operation on a would defer type checking to runtime:

dyn a = erase(123)

print a(1, 2, 3)
--> panic: Expected a function.

If a dynamic variable's recent type differs between two branches of execution, the type is considered unknown after the branches are merged. Any operations on the variable afterwards will defer type checking to runtime:

dyn a = 123
if a > 20:
    a = 'hello'

-- Branches are merged. `a` has an unknown type.

print a(1, 2, 3)
--> panic: Expected a function.

Dynamic tables. #

The builtin Table type is used to create dynamic objects. Tables are initialized with the record literal:

dyn a = {}
a.name = 'Nova'
print a.name     --> Nova

Read more about how to use Tables.

Dynamic inference. #

When the inference tag is used in a dynamic context, it will attempt to resolve its value at runtime. In this example, the dynamic value a resolves to a String at runtime and invokes the typed method trim. .left then infers to the correct value at runtime:

print str.trim(.left, ' ')

Metaprogramming. #

^top

Operator overloading. #

All operators are implemented as object methods.

Incomplete: Not all operators have transitioned to the method paradigm.

To overload an operator for an object type, declare $prefix, $infix, $postfix methods. See the available builtin operators. Since operator names aren't allowed as standard identifiers, they are contained in a string literal.

type Vec2:
    x float
    y float

    func '$infix+'(self, o Vec2) Vec2:
        return Vec2{
            x = x + o.x,
            y = y + o.y,
        }

    func '$prefix-'(self) Vec2:
        return Vec2{x=-x, y=-y}

var a = Vec2{x=1, y=2}
var b = a + Vec2{x=3, y=4}
var c = -a

Some special operators have their own name. This example overloads the index operator and the set index operator:

type MyCollection:
    arr List

    func $index(self, idx):
        return arr[idx * 2]

    func $setIndex(self, idx, val):
        arr[idx * 2] = val 

var a = MyCollection{arr={1, 2, 3, 4}}
print a[1]        -- Prints `3`

Builtin operators. #

A list of all supported operators:

Custom operators. #

Planned Feature

Magic functions. #

Call module. #

Declare a $call function to allow invoking a module as a function.

-- Object types are also modules.
type Vec2:
    x float
    y float

func Vec2.$call(x float, y float) Vec2:
    return Vec2{x=x, y=y}

var v = Vec2(1, 2)

$initRecord method. #

Planned Feature

$initPair method. #

The $initPair method overrides the record initializer. After an instance of the type is created from its default record initializer, this method is invoked for each key-value pair in the record literal:

type MyMap:
    func $initPair(self, key any, value any) void:
        print "$(key) = $(value)"

var m = MyMap{a=123, b=234}
--> a = 123
--> b = 234

$initPair is only allowed if the type has a default record initializer or $initRecord is declared.

$get method. #

The $get method allows overriding field accesses for undeclared fields:

type Foo:
    func $get(self, name String):
        return name.len()

var f = Foo{}
print f.abc      --> 3
print f.hello    --> 5

$set method. #

The $set method allows overriding field assignments for undeclared fields:

type Foo:
    func $set(self, name String, value any):
        print "setting $(name) $(value)"

var f = Foo{}
f.abc = 123      --> setting abc 123

Missing method. #

Declare a $missing method as a fallback when a method was not found in an instance.

Planned Feature

type A:

    func $missing(self, args...):
        return args.len

var a = A{}
print a.foo()      -- Output: '0'
print a.bar(1, 2)  -- Output: '2'

Reflection. #

A type metatype object references an internal type. Use the typeof builtin to get the metatype of a value.

var val = 123
print typeof(val)   -- 'type: float'

-- Referencing a type as a value also returns its `metatype`.
print bool          -- 'type: bool'

Attributes. #

Attributes start with @. They are used as declaration modifiers.

@host

Bind a function, variable, or type to the host. See libcyber.

Templates. #

Templates enables parametric polymorphism for types and functions. Template arguments are passed to templates to generate specialized code. This facilitates developing container types and algorithms that operate on different types.

See Type Declarations / Type templates and Functions / Function templates.

Value templates. #

A value template returns a memoized value after being invoked with template arguments at compile-time. It's declared with def:

def StrType[ID String] type:
    if ID == 'bool':
        return bool
    else ID == 'int':
        return int
    else ID == 'String':
        return String
    else
        throw error.Unsupported

var a StrType['int'] = 123
print a         --> 123

This can be useful to evaluate compile-time logic to create new types or specialize other templates. Any compile-time compatible type can also be returned.

Macros. #

Planned Feature

Compile-time execution. #

Planned Feature

Builtin types. #

Builtin types are used internally by the compiler to define it's own primitive types such as bool, int, and float.

type bool_t

type int64_t

type float64_t

Builtin functions. #

func genLabel(name String)

Emits a label during codegen for debugging.

Builtin constants. #

var modUri String

Evaluates to the module's URI as a string. See Module URI.

Runtime execution. #

cy.eval evaluates source code in an isolated VM. If the last statement is an expression, a primitive or String can be returned to the caller:

use cy

var res = cy.eval('1 + 2')
print res        --> 3

libcyber. #

^top

libcyber allows embedding the Cyber compiler and VM into applications. Cyber's core types and the CLI app were built using the library.

The API is defined in the C header file. The examples shown below can be found in the repository under c-embedded. C is used as the host language, but it can be easily translated to C++ or any C-ABI compatible language.

Types and constants from the C-API begin with CL and functions begin with cl.

Getting started. #

Create VM. #

Most operations are tied to a VM handle. To create a new VM instance, call clCreate:

#include "cyber.h"

int main() {
    CLVM* vm = clCreate();
    // ...
    clDestroy(vm);
    return 0;
}

Override print. #

The builtin print function does nothing by default, so it needs to be overrided to print to stdout for example:

void printer(CLVM* vm, CLStr str) {
    printf("Invoked printer: %.*s\n", (int)str.len, str.buf);
}

int main() {
    // ...
    clSetPrinter(vm, printer);
    // ...
}

Note that print invokes the printer twice, once for the value's string and another for the new line character.

Eval script. #

clEval compiles and evaluates a script:

CLStr src = STR(
    "var a = 1\n"
    "print(a + 2)\n"
);

CLValue val;
CLResultCode res = clEval(vm, src, &val);
if (res == CL_SUCCESS) {
    printf("Success!\n");
    clRelease(vm, val);
} else {
    CLStr report = clNewLastErrorReport(vm);
    printf("%s\n", report);
    clFree(vm, report);
}

If a value is returned from the main block of the script, it's saved to the result value argument. Memory is managed by ARC so a value that points to a heap object requires a clRelease when it's no longer needed.

clEval returns a result code that indicates whether it was successful.

Module Loader. #

A module loader describes how a module is loaded when use import statement is encountered during script execution. Only one module loader can be active and is set using clSetModuleLoader:

bool modLoader(CLVM* vm, CLStr spec, CLModule* res) {
    if (strncmp("my_mod", spec.buf, spec.len) == 0) {
        CLStr src = STR(
            "@host func add(a float, b float) float\n"
            "@host var .MyConstant float\n"
            "@host var .MyList     List[dyn]\n"
            "\n"
            "@host\n"
            "type MyNode _:\n"
            "    @host func asList(self) any"
            "\n"
            "@host func MyNode.new(a any, b any) MyNode\n"
        );
        *res = clCreateModule(vm, spec, src);
        CLModuleConfig config = (CLModuleConfig){
            .funcs = (CLSlice){ .ptr = funcs, .len = 3 },
            .types = (CLSlice){ .ptr = types, .len = 1 },
            .varLoader = varLoader,
        };
        clSetModuleConfig(vm, *res, &config);
        return true;
    } else {
        // Fallback to the default module loader to load `core`.
        return clDefaultModuleLoader(vm, spec, out);
    }
}

int main() {
    //...
    clSetModuleLoader(vm, modLoader);
    //...
}

The above example checks whether "my_mod" was imported and returns it's source code. Additional loaders are returned to load the functions, variables, and types from the source code.

Default module loader. #

Since only one module loader can be set to the VM instance, a custom loader is required to handle the "core" import which contains all of the core types and functions in Cyber. This can simply be delegated to clDefaultModuleLoader.

Bind functions. #

An array of function definitions can be assigned to CLModuleConfig.funcs. When a @host function is encountered by the compiler, it will use this mapping to find the correct function pointer:

CLHostFuncEntry funcs[] = {
    CL_FUNC("add",           add),
    CL_FUNC("MyNode.asList", myNodeAsList),
    CL_FUNC("MyNode.new",    myNodeNew),
};

A fallback function loader can be assigned to CLModuleConfig.func_loader. It's invoked if a function could not be found in CLModuleConfig.funcs.

bool funcLoader(CLVM* vm, CLFuncInfo info, CLFuncResult* out) {
    // Check that the name matches before setting the function pointer.
    if (strncmp("missing_func", info.name.buf, info.name.len) == 0) {
        out->ptr = myMissingFunc;
        return true;
    } else {
        return false;
    }
}

Variable loader. #

A variable loader describes how to load a @host variable when it's encountered by the compiler:

// C has limited static initializers (and objects require a vm instance) so initialize them in `main`.
typedef struct { char* n; CLValue v; } NameValue;
NameValue vars[2];

bool varLoader(CLVM* vm, CLVarInfo info, CLValue* out) {
    // Check that the name matches before setting the value.
    if (strncmp(vars[info.idx].n, info.name.buf, info.name.len) == 0) {
        // Objects are consumed by the module.
        *out = vars[info.idx].v;
        return true;
    } else {
        return false;
    }
}

int main() {
    // ...

    // Initialize var array for loader.
    vars[0] = (NameValue){".MyConstant", clFloat(1.23)};
    CLValue myInt = clInteger(123);
    vars[1] = (NameValue){".MyList", clNewList(vm, &myInt, 1)};

    // ...
}

This example uses the same technique as the function loader, but it can be much simpler. It doesn't matter how the mapping is done as long as the variable loader returns a CLValue.

Bind types. #

An array of type definitions can be assigned to CLModuleConfig.types. When a @host type is encountered by the compiler, it will use this mapping to initialize the type:

CLTypeId myNodeId;

CLHostTypeEntry types[] = {
    CL_CUSTOM_TYPE("MyNode", &myNodeId, myNodeGetChildren, myNodeFinalizer),
};

When binding to the "MyNode" type, it's type id is saved to myNodeId. This id is then used to create new instances of this type. See Host types.

A fallback type loader can be assigned to CLModuleConfig.type_loader. It's invoked if a type could not be found in CLModuleConfig.types.

bool typeLoader(CLVM* vm, CLTypeInfo info, CLTypeResult* out) {
    if (strncmp("MissingType", info.name.buf, info.name.len) == 0) {
        out->type = CS_TYPE_OBJECT;
        out->data.object.outTypeId = &myNodeId;
        out->data.object.getChildren = myNodeGetChildren;
        out->data.object.finalizer = myNodeFinalizer;
        return true;
    } else {
        return false;
    }
}

Host functions. #

A host function requires a specific function signature:

CLValue add(CLVM* vm, const CLValue* args, uint8_t nargs) {
    double res = clAsFloat(args[0]) + clAsFloat(args[1]);
    return clFloat(res);
}

A host function should always return a CLValue. clNone() can be returned if the function does not intend to return any value.

Host types. #

A host type are types that are opaque to Cyber scripts but still behave like an object. They can have type functions and methods.

Only the host application can directly create new instances of them, so usually a function is binded to expose a constructor to the user script:

// Binding a C struct with it's own children and finalizer.
// This struct retains 2 VM values and has 2 arbitrary data values unrelated to the VM.
typedef struct MyNode {
    CLValue val1;
    CLValue val2;
    int a;
    double b;
} MyNode;

// Implement the `new` function in MyNode.
CLValue myNodeNew(CLVM* vm, const CLValue* args, uint8_t nargs) {
    // Instantiate our object.
    CLValue new = clNewHostObject(vm, myNodeId, sizeof(MyNode));
    MyNode* my = (MyNode*)clAsHostObject(new);

    // Assign the constructor args passed in and retain them since the new object now references them.
    clRetain(vm, args[0]);
    my->val1 = args[0];
    clRetain(vm, args[1]);
    my->val2 = args[1];

    // Assign non VM values.
    my->a = 123;
    my->b = 9999.999;
    return new;
}

clNewHostObject takes the type id (returned from the Type loader) and size (in bytes) and returns a new heap object. Note that the size is allowed to vary. Different instances of the same type can occupy different amounts of memory.

getChildren #

Since MyNode contains CLValue children, the Type loader requires a getChildren callback so that memory management can reach them:

CLValueSlice myNodeGetChildren(CLVM* vm, void* obj) {
    MyNode* my = (MyNode*)obj;
    return (CLValueSlice){ .ptr = &my->val1, .len = 2 };
}

finalizer #

A type finalizer is optional since the memory and children of an instance will be freed automatically by ARC. However, it can be useful to perform additional cleanup tasks for instances that contain external resources.

void myNodeFinalizer(CLVM* vm, void* obj) {
    printf("MyNode finalizer was called.\n");
}

Memory. #

^top

Cyber provides memory safety by default with structured and automatic memory. Manual memory is also supported but discouraged.

Structured memory. #

Structured memory is very much incomplete. It will be centered around single value ownership but the semantics are subject to change. Cyber uses single value ownership and variable scopes to determine the lifetime of values and references. When lifetimes are known at compile-time, the memory occupied by values do not need to be manually managed which prevents memory bugs such as:

• Use after free.
• Use after invalidation.
• Free with wrong allocator.
• Double free.
• Memory leaks.
• Null pointer dereferencing.

At the same time, structured memory allows performant code to be written since it provides safe semantics to directly reference values and child values. These safety features are guaranteed for debug and optimized builds with no additional runtime cost.

Value ownership. #

Every value in safe memory has a single owner. An owner can be a variable that binds to a value. Otherwise, the owner can be a parent value or the value itself. The owner is responsible for deinitializing or dropping the value when it has gone out of scope (no longer reachable). For example, at the end of the block, a variable can no longer be accessed so it drops the value that it owns:

var a = 123
print a    --> 123
-- Deinit `a`.

In this case, there is nothing to deinitialize since the value is an integer.

If the value was a String, the deinit logic would release (-1) on a reference counted byte buffer since strings are just immutable views over byte buffers:

var a = 'hello'
print a    --> hello
-- Deinit `a`.
-- `a.buf` is released.
-- `a.buf` is freed.

Since the string buffer's reference count reaches 0, it's freed as well.

Finally, let's take a look at ListValue which manages a dynamically sized array of elements:

var a = ListValue[int]{1, 2, 3}
print a    --> {1, 2, 3}
-- Deinit `a`.
-- `a.buf` is freed.

When a is deinitialized, the buffer that holds the 3 integer elements is freed. You may have surmised that it's named ListValue because it's a value type (it can only be passed around by copying itself). The object type, List, wraps ListValue and can be passed around by reference.

The concept of a value having a single owner is very simple yet powerful. A value can represent any data structure from primitives to dynamically allocated buffers. A value always knows how to deinitialize itself, and the owner knows when to deinitialize the value. Later, we'll see that this same concept also applies to shared ownership.

Copy semantics. #

By default, values are passed around by copying (shallow copying), but not all values can perform a copy.

A primitive, such as an integer, can always be copied:

var a = 123
var b = a
-- Deinit `b`.
-- Deinit `a`.

After a copy, a new value is created and given the owner of b. At the end of the block, both a and b are deinitialized (which does nothing since they are just primitives).

Strings are also copyable since they are immutable views over byte buffers:

var a = 'hello'
var b = a
-- Deinit `b`.
-- `b.buf` is released.
-- Deinit `a`.
-- `a.buf` is released.
-- `a.buf` is freed.

The copy b also reuses the byte buffer of a by retaining (+1) on the reference counted byte buffer. The byte buffer is finally freed once there are no references pointing to it.

Unlike the integer and string, a ListValue can not be copied since doing so requires duping a heap allocated buffer which is considered expensive:

var a = ListValue[int]{1, 2, 3}
var b = a      --> error: Can not copy `ListValue`. Can only be cloned or moved.

Instead a can only be cloned or moved.

By default, a declared value type is copyable if all of it's members are also copyable:

type Foo struct:
    a int
    b String

var a = Foo{a=123, b='hello'}
var b = a

Since integers and strings are both copyable, Foo is also copyable.

Foo is non-copyable if it contains at least one non-copyable member:

type Foo struct:
    a int
    b String
    c ListValue[int]

var a = Foo{a=123, b='hello'}
var b = a      --> error: Can not copy `Foo`. Can only be moved.

Foo is also non-copyable if it contains unsafe types such as pointers or pointer slices:

type Foo struct:
    a int
    b String
    c *Bar
    d [*]float

Foo can implement Copyable to override the default behavior and define it's own copy logic:

type Foo struct:
    with Copyable
    a int
    b String
    c *Bar
    d [*]float

    func copy(self) Foo:
        return .{
            a = self.a,
            b = self.b,
            c = self.c,
            d = self.d,
        }

Likewise, Foo can implement NonCopyable which indicates that it can never be copied:

type Foo struct:
    with NonCopyable
    a int
    b String

Cloning. #

Some value types are not allowed to be copied by default and must be cloned instead:

var a = ListValue[int]{1, 2, 3}
var b = a.clone()

Any Copyable type is also Cloneable. For example, performing a clone on an integer will simply perform a copy:

var a = 123
var b = a.clone()

A value type can implement Cloneable to override the default behavior and define it's own clone logic:

type Foo struct:
    with Cloneable
    a int
    b String

    func clone(self) Foo:
        return .{
            a = self.a + 1,
            b = self.b,
        }

Likewise, Foo can implement NonCloneable which indicates that it can never be cloned:

type Foo struct:
    with NonCloneable
    a int
    b String

Moving. #

Values can be moved, thereby transfering ownership from one variable to another:

var a = 123
var b = move a
print a     --> error: `a` does not own a value.

Some types such as ListValue can not be passed around by default without moving (or cloning) the value:

var a = ListValue[int]{1, 2, 3}
print computeSum(move a)    --> 6

In this case, the list value is moved into the computeSum function, so the list is deinitialized in the function before the function returns.

References. #

References are safe pointers to values. Unlike unsafe pointers, a reference is never concerned with when to free or deinitialize a value since that responsibility always belongs to the value's owner. They are considered safe pointers because they are guaranteed to point to their values and never outlive the lifetime of their values.

References grant in-place mutability which allows a value to be modified as long as it does not invalidate other references. Multiple references can be alive at once as long as an exclusive reference is not also alive.

The & operator is used to obtain a reference to a value:

var a = 123
var ref = &a
ref.* = 234
print a        --> 234

A reference can not outlive the value it's referencing:

var a = 123
var ref = &a
if true:
    var b = 234
    ref = &b   --> error: `ref` can not outlive `b`.

A reference type is denoted as &T where T is the type that the reference points to:

var a = 123

func inc(a &int):
    a.* = a.* + 1

References allow in-place mutation:

var a = ListValue[int]{1, 2, 3}
var third = &a[2]
third.* = 300
print a        --> {1, 2, 300}

The element that third points to can be mutated because it does not invalidate other references.

References however can not perform an unstable mutation. An unstable mutation requires an exclusive reference:

var a = ListValue[int]{1, 2, 3}
var ref = &a
ref.append(4)  --> error: Expected exclusive reference.

Exclusive reference. #

An exclusive reference grants full mutability which allows a value to be modified even if could potentially invalidate unsafe pointers.

A single exclusive reference can be alive as long as no other references are also alive. Since no other references (safe pointers) are allowed to be alive at the same time, no references can become invalidated.

The &! operator is used to obtain an exclusive reference to a value. An exclusive reference type is denoted as &!T where T is the type that the reference points to.

ListValue is an example of a type that requires an exclusive reference for operations that can resize or reallocate its dynamic buffer:

var a = ListValue[int]{1, 2, 3}
a.append(4)
print a        --> {1, 2, 3, 4}

Note that invoking the method append here automatically obtains an exclusive reference for self without an explicit &! operator.

If another reference is alive before append, the compiler will not allow an exclusive reference to be obtained from a. Doing so would allow append to potentially reallocate its dynamic buffer, thereby invalidating other references:

var a = ListValue[int]{1, 2, 3}
var third = &a[2]
a.append(4)    --> error: Can not obtain exclusive reference, `third` is still alive.
print third

self reference. #

By default self has a type of &T when declared in a value T's method:

type Pair struct:
    a int
    b int

    func sum(self) int:
        return self.a + self.b

If self requires an exclusive reference, then it must be prepended with !:

type Pair struct:
    a int
    b int

    func sum(!self) int:
        return self.a + self.b

Invoking methods automatically obtains the correct reference as specified by the method:

var p = Pair{a=1, b=2}
print p.sum()     --> 3

Lifted values. #

Planned Feature

Deferred references. #

Planned Feature

Implicit lifetimes. #

Planned Feature

Reference lifetimes. #

Planned Feature

Shared ownership. #

Planned Feature

Deinitializer. #

Planned Feature

Pointer interop. #

Planned Feature

Automatic memory. #

Cyber uses an ARC/GC hybrid to automatically manage objects instantiated from object types. Value types typically do not need to be automatically managed unless they were lifted by a closure or a dynamic container.

ARC. #

ARC also known as automatic reference counting is deterministic and has less overhead compared to a tracing garbage collector. Reference counting distributes memory management, which reduces GC pauses and makes ARC suitable for realtime applications. One common issue in ARC implementations is reference cycles which Cyber addresses with a GC supplement when it is required.

Objects are managed by ARC. Each object has its own reference counter. Upon creating a new object, it receives a reference count of 1. When the object is copied, it's retained and the reference count increments by 1. When an object value is removed from it's parent or is no longer reachable in the current stack frame, it is released and the reference count decrements by 1.

Once the reference count reaches 0 the object begins its destruction procedure.

Object destructor. #

An object's destructor invoked from ARC performs the following in order:

1. Release child references thereby decrementing their reference counts by 1. If any child reference counts reach 0, their destructors are invoked.
2. If the object has a finalizer, it's invoked.
3. The object is freed from memory.

If the destructor is invoked by the GC instead of ARC, cyclable child references are not released in step 1. Since objects freed by the GC either belongs to a reference cycle or branched from one, the GC will still end up invoking the destructor of all unreachable objects. This implies that the destructor order is not reliable, but destructors are guaranteed to be invoked for all unreachable objects.

Retain optimizations. #

When the lifetime of an object's reference is known on the stack, a large amount of retain/release ops can be avoided. For example, calling a function with an object doesn't need a retain since it is guaranteed to be alive when the function returns. This leaves only cases where an object must retain to ensure correctness such as escaping the stack.

When using dynamic types, the compiler can omit retain/release ops when it can infer the actual type even though they are dynamically typed to the user.

Closures. #

When primitive variables are captured by a closure, they are boxed and allocated on the heap. This means they are managed by ARC and cleaned up when there are no more references to them.

Fibers. #

Fibers are freed by ARC just like any other object. Once there are no references to the fiber, it begins to release it's child references by unwinding it's call stack.

Heap objects. #

Many object types are small enough to be at or under 40 bytes. To take advantage of this, object pools are reserved to quickly allocate and free these small objects with very little bookkeeping. Bigger objects are allocated and managed by mimalloc which has proven to be a fast and reliable general-purpose heap allocator.

GC. #

The garbage collector is only used if the program may contain objects that form reference cycles. This property is statically determined by the compiler. Since ARC frees most objects, the GC's only responsibility is to free abandoned objects that form reference cycles. This reduces the amount of work for GC marking since only cyclable objects (objects that may contain a reference cycle) are considered.

Weak references are not supported for object types because objects are intended to behave like GC objects (the user should not be concerned with reference cycles). If weak references do get supported in the future, they will be introduced as a Weak[T] type that is used with an explicit reference counted Rc[T] type.

Currently, the GC can be manually invoked. However, the plan is for this to be automatic by either running in a separate thread or per virtual thread by running the GC incrementally.

To invoke the GC, call the builtin function: performGC. Incomplete Feature: Only the main fiber stack is cleaned up at the moment.

func foo():
    -- Create a reference cycle.
    var a = {_}
    var b = {_}
    a.append(b)
    b.append(a)

    -- Cycle still alive in the current stack so no cleanup is done.
    var res = performGC()
    print res['numCycFreed']    -- Output: 0
    print res['numObjFreed']    -- Output: 0

foo()
-- `a` and `b` are no longer reachable, so the GC does work.
var res = performGC()
print res['numCycFreed']      -- Output: 2
print res['numObjFreed']      -- Output: 2

Manual memory. #

Planned Feature

Memory allocations. #

Planned Feature

Runtime memory checks. #

Planned Feature When runtime memory checks are enabled, the compiler will insert traps and runtime logic to prevent the following unsafe uses of pointers and memory allocations:

• Use after free.
• Double free.
• Out of bounds memory access.
• Null pointer access.
• Misaligned pointer access.
• Unfreed memory.

With these checks in place, using manual memory will be much less error prone at the cost of some runtime performance and memory cost. Pointers will occupy an extra word size and the runtime will maintain an allocation table that contains the raw pointer and metadata.

CLI. #

• Basic commands.
• REPL.
• JIT compiler.
• C backend.

^top

Basic commands. #

To compile and run a program with the VM, provide the path to the main Cyber source file:

cyber foo.cy
cyber path/to/main.cy

To see more options and commands, print the help screen:

cyber help

# These are aliases to the help command.
cyber -h
cyber --help

REPL. #

The REPL is started by running the CLI without any arguments:

cyber

The REPL starts new sessions with use $global. This allows undeclared variables to be used similar to other dynamic languages:

> a = 123
> a * 2
`int` 246

When the first input ends with :, the REPL will automatically indent the next line. To recede the indentation, provide an empty input. Once the indent returns to the beginning, the entire code block is submitted for evaluation:

> if true:
    | print 'hello!'
    | 
hello!

Top level declarations such as imports, types, and functions can be referenced in subsequent evals:

> use math
> math.random()
`float` 0.3650744641604983

> type Foo:
    | a int
    |
> f = Foo{a=123}
> f.a
`int` 123

Local variables can not be referenced in subsequent evals, since their scope ends with each eval input:

> var a = 123
> a
panic: Variable is not defined in `$global`.

input:1:1 main:
a
^

JIT compiler. #

Cyber's just-in-time compiler is incomplete and unstable. To run your script with JIT enabled:

cyber -jit <script>

The goal of the JIT compiler is to be fast at compilation while still being significantly faster than the interpreter. The codegen involves stitching together pregenerated machine code that targets the same runtime stack slots used by the VM. This technique is also known as copy-and-patch. As the VM transitions to unboxed data types, the generated code will see more performance gains.

C backend. #

The C backend generates a static binary from Cyber source code by first transpiling to C code and relying on a C compiler to produce the final executable. The user can specify the system's cc compiler or the builtin tinyc compiler that is bundled with the CLI. This is currently in progress.