C3 Specification

THIS SPECIFICATION IS UNDER DEVELOPMENT

Notation

The syntax is specified using Extended Backus-Naur Form (EBNF):

production  ::= PRODUCTION_NAME '::=' expression?
expression  ::= alternative ("|" alternative)*
alternative ::= term term*
term        ::= PRODUCTION_NAME | TOKEN | set | group | option | repetition
set         ::= '[' (range | CHAR) (rang | CHAR)* ']'
range       ::= CHAR '-' CHAR
group       ::= '(' expression ')'
option      ::= expression '?'
repetition  ::= expression '*'

Productions are expressions constructed from terms and the following operators, in increasing precedence:

|   alternation
()  grouping
?  option (0 or 1 times)
*  repetition (0 to n times)

Uppercase production names are used to identify lexical tokens. Non-terminals are in lower case. Lexical tokens are enclosed in single quotes ”.

The form a..b represents the set of characters from a through b as alternatives.

Source code representation

A program consists of one or more translation units stored in files written in the Unicode character set, stored as a sequence of bytes using the UTF-8 encoding. Except for comments and the contents of character and string literals, all input elements are formed only from the ASCII subset (U+0000 to U+007F) of Unicode.

A raw byte stream is translated into a sequence of tokens which white space and comments are discarded. Doc comments may optionally be discarded as well. The resulting input elements form the tokens that are the terminal symbols of the syntactic grammar.

Lexical Translations

The longest possible translation is used at each step, even if the result does not ultimately make a correct program while another lexical translation would.

Example: a--b is translated as a, --, b, which does not form a grammatically correct expression, even though the tokenization a, -, -, b could form a grammatically correct expression.

Line Terminators

The C3 compiler divides the sequence of input bytes into lines by recognizing line terminators

Lines are terminated by the ASCII LF character (U+000A), also known as “newline”. A line termination specifies the termination of the // form of a comment.

Input Elements and Tokens

An input element may be:

White space
Comment
Doc Contract
Token

A token may be:

Identifier
Keyword
Literal
Separator
Operator

A Doc Contract consists of:

A stream of descriptive text
A list of directive Tokens

Those input elements that are not white space or comments are tokens. The tokens are the terminal symbols of the syntactic grammar. Whitespace and comments can serve to separate tokens that might be tokenized in another manner. For example the characters + and = may form the operator token += only if there is no intervening white space or comment.

White Space

White space is defined as the ASCII horizontal tab character (U+0009), form feed character (U+000A), vertical tab ( U+000B), carriage return (U+000D), space character (U+0020) and the line terminator character (U+000D).

WHITESPACE      ::= [ \t\f\v\r\n]

Letters and digits

UC_LETTER       ::= [A-Z]
LC_LETTER       ::= [a-z]
LETTER          ::= UC_LETTER | LC_LETTER
DIGIT           ::= [0-9]
HEX_DIGIT       ::= [0-9a-fA-F]
BINARY_DIGIT    ::= [01]
OCTAL_DIGIT     ::= [0-7]
LC_LETTER_US    ::= LC_LETTER | "_"
UC_LETTER_US    ::= UC_LETTER | "_"
ALPHANUM        ::= LETTER | DIGIT
ALPHANUM_US     ::= ALPHANUM | "_"
UC_ALPHANUM_US  ::= UC_LETTER_US | DIGIT
LC_ALPHANUM_US  ::= LC_LETTER_US | DIGIT

Comments

There are three types of regular comments:

// text a line comment. The text between // and line end is ignored.
/* text */ block comments. The text between /* and */ is ignored. It has nesting behaviour, so for every /* discovered between the first /* and the last */ a corresponding */ must be found.

Doc contract

<* text *> doc block comment. The text between <* and *> is optionally parsed using the doc comment syntactic grammar. A compiler may choose to read <* text *> as a regular comment.

Identifiers

Identifiers name program entities such as variables and types. An identifier is a sequence of one or more letters and digits. The first character in an identifier must be a letter or underscore.

C3 has three types of identifiers: const identifiers - containing only underscore and upper-case letters, type identifiers - starting with an upper case letter followed by at least one underscore letter and regular identifiers, starting with a lower case letter.

IDENTIFIER      ::=  "_"* LC_LETTER ALPHANUM_US*
CONST_IDENT     ::=  "_"* UC_LETTER UC_ALPHANUM_US*
TYPE_IDENT      ::=  "_"* UC_LETTER UC_ALPHANUM_US* LC_LETTER ALPHANUM_US*
CT_IDENT        ::=  "$" IDENTIFIER
CT_CONST_IDENT  ::=  "$" CONST_IDENT
CT_TYPE_IDENT   ::=  "$" TYPE_IDENT
AT_TYPE_IDENT   ::=  "@" TYPE_IDENT
PATH_SEGMENT    ::= "_"* LC_LETTER LC_ALPHANUM_US*

Keywords

The following keywords are reserved and may not be used as identifiers:

asm         any         fault
assert      attribute   break
case        catch       const
continue    default     defer
alias         do          else
enum        extern      false
while       fn          if
import      inline      macro
module      nextcase    null
public      return      struct
switch      true        try
typeid      var         void


bool        int128      double
float       long        ulong
int         uint        byte
short       ushort      char
isz         usz         float16
float128    uint128     bfloat16

$assert     $case       $default
$echo       $else       $error
$endfor     $endforeach $endif
$endswitch  $for        $foreach
$if         $switch     $typef
$vaarg      $vaconst    $vacount
$vaexpr     $vatype

Operators and punctuation

The following character sequences represent operators and punctuation.

&       @       ~       |       ^       :
,       /       $       .       ;       )
>       <       #       {       }       -
(       )       *       [       ]       %
>=      <=      +       +=      -=      !
?       ?:      &&      ??      &=      |=
^=      /=      ..      ==      ({      })
[<      >]      ++      --      %=      !=
||      ::      <<      >>      !!      ...
<<=     >>=

Integer literals

An integer literal is a sequence of digits representing an integer constant. An optional prefix sets a non-decimal base: 0b or 0B for binary, 0o, or 0O for octal, and 0x or 0X for hexadecimal. A single 0 is considered a decimal zero. In hexadecimal literals, letters a through f and A through F represent values 10 through 15.

For readability, an underscore character _ may appear after a base prefix or between successive digits; such underscores do not change the literal’s value.

INTEGER         ::= DECIMAL_LIT | BINARY_LIT | OCTAL_LIT | HEX_LIT
DECIMAL_LIT     ::= '0' | [1-9] ('_'* DECIMAL_DIGITS)?
BINARY_LIT      ::= '0' [bB] '_'* BINARY_DIGITS
OCTAL_LIT       ::= '0' [oO] '_'* OCTAL_DIGITS
HEX_LIT         ::= '0' [xX] '_'* HEX_DIGITS

BINARY_DIGIT    ::= [01]
HEX_DIGIT       ::= [0-9a-fA-F]

DECIMAL_DIGITS  ::= DIGIT ('_'* DIGIT)*
BINARY_DIGITS   ::= BINARY_DIGIT ('_'* BINARY_DIGIT)*
OCTAL_DIGITS    ::= OCTAL_DIGIT ('_'* OCTAL_DIGIT)*
HEX_DIGITS      ::= HEX_DIGIT ('_'* HEX_DIGIT)*

42
4_2
0_600
0o600
0O600           // second character is capital letter 'O'
0xBadFace
0xBad_Face
0x_67_7a_2f_cc_40_c6
170141183460469231731687303715884105727
170_141183_460469_231731_687303_715884_105727

0600            // Invalid, non zero decimal number may not start with 0
_42             // an identifier, not an integer literal
42_             // invalid: _ must separate successive digits
0_xBadFace      // invalid: _ must separate successive digits

Floating point literals

A floating-point literal is a decimal or hexadecimal representation of a floating-point constant.

A decimal floating-point literal consists of an integer part (decimal digits), a decimal point, a fractional part (decimal digits), and an exponent part (e or E followed by an optional sign and decimal digits). One of the integer part or the fractional part may be elided; one of the decimal point or the exponent part may be elided. An exponent value exp scales the mantissa (integer and fractional part) by powers of 10.

A hexadecimal floating-point literal consists of a 0x or 0X prefix, an integer part (hexadecimal digits), a radix point, a fractional part (hexadecimal digits), and an exponent part (p or P followed by an optional sign and decimal digits). One of the integer part or the fractional part may be elided; the radix point may be elided as well, but the exponent part is required. An exponent value exp scales the mantissa (integer and fractional part) by powers of 2.

For readability, an underscore character _ may appear after a base prefix or between successive digits; such underscores do not change the literal value.

FLOAT_LIT       ::= DEC_FLOAT_LIT | HEX_FLOAT_LIT
DEC_FLOAT_LIT   ::= DECIMAL_DIGITS '.' DECIMAL_DIGITS? DEC_EXPONENT?
                    | DECIMAL_DIGITS DEC_EXPONENT
                    | '.' DECIMAL_DIGITS DEC_EXPONENT?
DEC_EXPONENT    ::= [eE] [+-]? DECIMAL_DIGITS
HEX_FLOAT_LIT   ::= '0' [xX] HEX_MANTISSA HEX_EXPONENT
HEX_MANTISSA    ::= HEX_DIGITS '.' HEX_DIGITS?
                    | HEX_DIGITS
                    | '.' HEX_DIGITS
HEX_EXPONENT    ::= [pP] [+-] DECIMAL_DIGITS

Characters

Characters are the fundamental components of strings and character literals.

CHAR_ELEMENT    ::= [\x20-\x26] | [\x28-\x5B] | [\x5D-\x7F]
CHAR_LIT_BYTE   ::= CHAR_ELEMENT | \x5C CHAR_ESCAPE
CHAR_ESCAPE     ::= [abefnrtv\'\"\\]
                    | 'x' HEX_DIGIT HEX_DIGIT
UNICODE_CHAR    ::= unicode_char
                    | 'u' HEX_DIGIT HEX_DIGIT HEX_DIGIT HEX_DIGIT
                    | 'U' HEX_DIGIT HEX_DIGIT HEX_DIGIT HEX_DIGIT
                          HEX_DIGIT HEX_DIGIT HEX_DIGIT HEX_DIGIT

Backslash escapes

The following backslash escapes are available for characters and string literals:

\0      0x00 zero value
\a      0x07 alert/bell
\b      0x08 backspace
\e      0x1B escape
\f      0x0C form feed
\n      0x0A newline
\r      0x0D carriage return
\t      0x09 horizontal tab
\v      0x0B vertical tab
\\      0x5C backslash
\'      0x27 single quote '
\"      0x22 double quote "
\x      Escapes a single byte hex value
\u      Escapes a two byte unicode hex value
\U      Escapes a four byte unicode hex value

String literals

A string literal represents a string constant obtained from concatenating a sequence of characters. String literals are character sequences between double quotes, as in “bar”. Within the quotes, any character may appear except newline and unescaped double quote. The text between the quotes forms the value of the literal, with backslash escapes interpreted as they are in rune literals, with the same restrictions. The two-digit hexadecimal (\xnn) escapes represent individual bytes of the resulting string; all other escapes represent the (possibly multibyte) UTF-8 encoding of individual characters. Thus inside a string literal \xFF represent a single byte of value 0xFF = 255, while ÿ, \u00FF, \U000000FF and \xc3\xbf represent the two bytes 0xc3 0xbf of the UTF-8 encoding of character U+00FF.

STRING_LIT      ::= \x22 (CHAR_LIT_BYTE | UNICODE_CHAR)* \x22

Compile time string concatenation

Strings will concatenate if declared in sequence.

Example:

String s = "abc" "def" "ghi";
// This is equivalent to:
String s = "abcdefghi";

Raw string literals

Raw string literals are enclosed between “ and consist of the raw UTF8 in the source code between the ”`”. A sequence of two ”`” will be interpreted as a single escaped ”`” that does not terminate the literal.

Compile time concatenation

Raw strings will concatenate with other regular strings and raw strings ( see string literal compile time concatenation).

Source code pre-filtering

The source code will pre-filter \r (0x0D) from the source code. This means that it is also implicitly filtered out of raw strings.

Character literals

A character literal is enclosed in ' and may either consist of 1, 2, 4, 8, 16 bytes.

CHARACTER_LIT   ::= "'" (CHAR_LIT_BYTE+) | UNICODE_CHAR "'"

Types

Types consist of built-in types and user-defined types (enums, structs, unions, bitstructs, fault and typedef).

Boolean types

bool may have the two values true and false. It holds a single bit of information but is stored in a char type.

Integer types

The built-in integer types:

char      unsigned 8-bit
ichar     signed 8-bit
ushort    unsigned 16-bit
short     signed 16-bit
uint      unsigned 32-bit
int       signed 32-bit
ulong     unsigned 64-bit
long      signed 64-bit
uint128   unsigned 128-bit
int128    singed 128-bit

In addition, the following type aliases exist:

uptr      unsigned pointer size
iptr      signed pointer size
usz       unsigned pointer offset / object size
isz       signed pointer offset  / object size

Floating point types

Built-in floating point types:

float16   IEEE 16-bit*
bfloat16  Brainfloat*
float     IEEE 32-bit
double    IEEE 64-bit
float128  IEEE 128-bit*

(* optionally supported)

Vector types

A vector lowers to the platform’s vector types where available. A vector has a base type and a width.

vector_type        ::= type "[<" length ">]"

Vector base type

The base type of a vector must be boolean, an integer or a floating point type.

Min width

The vector width must be at least 1.

Element access

Vector elements are accessed using []. It is possible to take the address of a single element.

Alignment

Alignment of vectors are platform dependent, but is at least the alignment of its element type.

Vector operations

Vectors support the same arithmetics as its underlying type, and will perform the operation element-wise.

Example:

int[<2>] a = { 1, 3 };
int[<2>] b = { 2, 7 };

int[<2>] c = a * b;
// Equivalent to
int[<2>] c = { a[0] * b[0], a[1] * b[1] };

Array types

An array has the alignment of its elements. An array must have at least one element.

Slice types

The slice consist of a pointer, followed by an usz length, having the alignment of pointers.

Pointer types

A pointer is the address to memory.

pointer_type       ::= type "*"

Pointee type

The type of the memory pointed to is the pointee type. It may be any runtime type.

`iptr` and `uptr`

A pointer may be losslessly cast to an iptr or uptr. An iptr or uptr may be cast to a pointer of any type.

The wildcard pointer `void*`

The void* may implicitly cast into any other pointer type. The void* [implicitly casts into any other pointer.

A void* pointer may never be dereferenced.

Pointer arithmetic on `void*`

Performing pointer arithmetics on void* will assume that the element size is 1. This includes pointer arithmetics using subscripting.

Subscripting

Subscripting a pointer is equal to performing pointer arithmetics using the index, followed by a deref. Subscripts on pointers may be negative and will never do bounds checks.

Deref

Dereferencing a pointer will return the value in the memory location interpreted as the pointee type.

Struct types

A struct may not have zero members.

Alignment

A non-packed struct has the alignment of the member that has the highest alignment. A packed struct has alignment 1. See align attribute for details on changing the alignment.

Union types

A union may not have zero members.

Alignment

A union has the alignment of the member that has the highest alignment. See align attribute for details on changing the alignment.

Fault types

A fault is a constant which can be used to create an Excuse for an empty optional.

Alignment

A fault type has the same alignment as a pointer. See align attribute for details on changing the alignment.

Enum types

Function types

Typeid type

The typeid is a pointer sized value which uniquely identifies a type.

Any type

The any is a fat pointer (2 pointers wide) holding a pointer to a value and its corresponding typeid. It cannot be dereferenced.

Fields

.ptr returns a void* pointer to the underlying value .type returns the typeid of the underlying value.

Switching over `any`

Switching over an any value creates an any switch.

Anyfault type

Declarations and scope

Expressions

Assignment expression

assignment_expr    ::= ct_type_assign | unary_expr assignment_op expr
ct_type_assign     ::= ct_type_ident "=" type
assignment_op      ::= "=" | "+=" | "-=" | "*=" | "/=" | "%=" | "<<=" | ">>=" | "&=" | "^=" | "|="

Type assign

This assigns a new type to a compile time type variable. The value of the expression is the type assigned.

Combined assign

All assignment operations except for = are combined assign operation. They first perform the operation indicated by the leftmost character(s) in the operator (e.g + for +=, << for <<= etc) with the lhs and the rhs. The result is then assigned to the left hand side. The result of the operation is the new value of the left hand side.

Implicit conversion

If the left hand side is a pointer and the operation is += or -= an attempt to implicitly convert to isz/usz will be tried.

For all other types and operations, an implicit conversion of rhs to the type of lhs will be tried.

Ternary, elvis and or-else expressions

ternary_group_expr ::= suffix_group_expr | ternary_expr | elvis_expr | orelse_expr
ternary_expr       ::= or_expr "?" expr ":" ternary_group_expr
elvis_expr         ::= suffix_expr "?:" ternary_group_expr
orelse_expr        ::= suffix_expr "??" ternary_group_expr

Ternary evaluation

The most left-hand expression is evaluated to a boolean. If it is true, the value of the middle expression is returned, otherwise the last expression is returned.

Only the most left-hand expression and the returned expressions are evaluated.

The middle and last expression are implicitly converted to their unified type.

The resulting type is the unified type.

Elvis evaluation

Lhs and rhs are implicitly converted to their unified type.

The lhs is evaluated, it is then converted to a boolean, if the result it true, return the lhs value before its boolean conversion. Otherwise return the right hand side.

The right hand side is only evaluated if the lhs evaluates to false.

The resulting type is the unified type.

Orelse evaluation

The lhs must be optional. The non-optional type for lhs and rhs are calculated. The unified type of the result is calculated. Lhs are converted to the unified type preserving their optionality.

At runtime, lhs is evaluated. If it evaluates to an optional, rhs is returned instead.

Rhs is only evaluated if lhs evaluates to an optional.

The resulting type of the orelse is the post conversion type of the rhs.

Suffix expression

Suffix expressions convert a fault to an optional.

suffix_group_exp   ::= or_group_expr | suffix_expr
suffix_expr        ::= or_group_expr "?" "!"?

Effect of `?`

The ? will convert the expression into an optional. The left hand side must be a fault type. If an optional ! follows, this optional is immediately returned, as if by a return <expr>? statement.

Type of the expression

The type is a wildcard optional. If ! is added, it is a wildcard type.

Rethrow expression

If the expression is optional, implicitly return with the optional value.

rethrow_expr       ::= expr "!"

The expression to rethrow

The expression must have an optional type, otherwise this is a compile time error.

Type

The type of “rethrow” is the inner expr type without optional.

Relational expression

rel_group_expr     ::= add_group_expr | relational_expr
relational_expr    ::= rel_group_expr relational_op add_group_expr
relational_op      ::= "<" | ">" | "<=" | ">="

TODO

And expression

This binary expression evaluates the lhs, and if the result is true evaluates the rhs. The result is true if both lhs and rhs are true.

and_group_expr     ::= rel_group_expr | and_expr
and_expr           ::= and_group_expr "&&" rel_group_expr

Type

The type of the and-expression is bool.

Or expression

This binary expression evaluates the lhs, and if the result is false evaluates the rhs. The result is true if lhs or rhs is true.

or_group_expr      ::= and_group_expr | or_expr
or_expr            ::= or_group_expr "||" and_group_expr

Constant folded arithmetics

Constant folding will happen for constant integers and floating. Vectors operations will not be constant-folded.

Constant folded operations are: multiplication, division, addition, subtraction, bit shifts, bit negation, bitwise and, or and xor, comparison, logical and/or, and negation.

Type

The type of the or-expression is bool.

Casts

Pointer casts

Integer to pointer cast

Any integer of pointer size or larger may be explicitly cast to a pointer. An integer to pointer cast is considered non-constant, except in the special case where the integer == 0. In that case, the result is constant null.

Example:

byte a = 1;
int* b = (int*)a; // Invalid, pointer type is > 8 bits.
int* c = (int*)1; // Valid, but runtime value.
int* d = (int*)0; // Valid and constant value.

Pointer to integer cast

A pointer may be cast to any integer, truncating the pointer value if the size of the pointer is larger than the pointer size. A pointer to integer cast is considered non-constant, except in the special case of a null pointer, where it is equal to the integer value 0.

Example:

fn void test() { ... }
alias VoidFunc = fn void test();

VoidFunc a = &test;
int b = (int)null;
int c = (int)a; // Invalid, not constant
int d = (int)((int*)1); // Invalid, not constant

Subscript operator

The subscript operator may take as its left side a pointer, array or slice. The index may be of any integer type. TODO NOTE The subscript operator is not symmetrical as in C. For example in C3 array[n] = 33 is allowed, but not n[array] = 33. This is a change from C.

Operands

Compound Literals

Compound literals have the format

compound_literal   ::= (type) initializer_list
initializer_list   ::= '{' (initializer_param (',' initializer_param)* ','?)? '}'
initializer_param  ::= expression | designator '=' expression
designator         ::= array_designator | range_designator | field_designator
array_designator   ::= '[' expression ']'
range_designator   ::= '[' range_expression ']'
field_designator   ::= IDENTIFIER
range_expression   ::= (range_index)? '..' (range_index)?
range_index        ::= expression | '^' expression

Taking the address of a compound literal will yield a pointer to stack allocated temporary.

Function calls

Function argument resolution

Call slots are in order: regular slots, vaarg slot, name-only slots.

No regular slots may appear after the vaarg slot, however there may be named parameters with default values after the vaarg slot if it’s not a raw vaarg.

These “name-only” slots need to have a parameter name and a default value, and may only be called as named arguments.

Named arguments may never be splat expressions.

Step through all the arguments, resolve the named arguments and determine if there are any regular arguments.
If there are regular arguments, then named arguments may only be in name-only slots, otherwise it is an error.
If there are named arguments in the regular slots, all slots not provided arguments must have default values.
Proceed with evaluation of arguments from left to right in call invocation order.
Regular arguments are placed in the regular slots from left to right.
If a regular argument is a splat expression, evaluate it without inference and determine if it is an array, vector, untyped list or slice with a known size, otherwise it is an error.
A regular argument splat will be expanded into as many slots as its length, this may expand into vaarg arguments.
In the vaarg slot, splatting a slice will forward it.
In the vaarg slot, splatting an array, vector or untyped list will expand its elements as if they were provided as arguments.
A named argument may never appear more than once.
The vaarg slot may never be accessed using named arguments.

Vaargs

For vaargs, a bool or any integer smaller than what the C ABI specifies for the c int type is cast to int. Any float smaller than a double is cast to double. Compile time floats will be cast to double. Compile time integers will be cast to c int type.

Statements

stmt               ::= compound_stmt | non_compound_stmt
non_compound_stmt  ::= assert_stmt | if_stmt | while_stmt | do_stmt | foreach_stmt | foreach_r_stmt
                       | for_stmt | return_stmt | break_stmt | continue_stmt | var_stmt
                       | declaration_stmt | defer_stmt | nextcase_stmt | asm_block_stmt
                       | ct_echo_stmt | ct_error_stmt | ct_assert_stmt | ct_if_stmt | ct_switch_stmt
                       | ct_for_stmt | ct_foreach_stmt | expr_stmt

Asm block statement

An asm block is either a string expression or a brace enclosed list of asm statements.

asm_block_stmt      ::= "asm" ("(" constant_expr ")" | "{" asm_stmt* "}")
asm_stmt            ::= asm_instr asm_exprs? ";"
asm_instr           ::= ("int" | IDENTIFIER) ("." IDENTIFIER)
asm_expr            ::= CT_IDENT | CT_CONST_IDENT | "&"? IDENTIFIER | CONST_IDENT | FLOAT_LITERAL
                        | INTEGER | "(" expr ")" | "[" asm_addr "]"
asm_addr            ::= asm_expr (additive_op asm_expr asm_addr_trail?)?
asm_addr_trail      ::= "*" INTEGER (additive_op INTEGER)? | (shift_op | additive_op) INTEGER

TODO

Assert statement

The assert statement will evaluate the expression and call the panic function if it evaluates to false.

assert_stmt        ::= "assert" "(" expr ("," assert_message)? ")" ";"
assert_message     ::= constant_expr ("," expr)*

Conditional inclusion

assert statements are only included in “safe” builds. They may turn into assume directives for the compiler on “fast” builds.

Assert message

The assert message is optional. It can be followed by an arbitrary number of expressions, in which case the message is understood to be a format string, and the following arguments are passed as values to the format function.

The assert message must be a compile time constant. There are no restriction on the format argument expressions.

Panic function

If the assert message has no format arguments or no assert message is included, then the regular panic function is called. If it has format arguments then panicf is called instead.

In the case the panicf function does not exist (for example, compiling without the standard library), then the format and the format arguments will be ignored and the assert will be treated as if no assert message was available.

Break statement

A break statement exits a while, for, do, foreach or switch scope. A labelled break may also exit a labelled if.

break_stmt         ::= "break" label? ";"

Break labels

If a break has a label, then it will instead exit an outer scope with the label.

Unreachable code

Any statement following break in the same scope is considered unreachable.

Compile time echo statement

During parsing, the compiler will output the text in the statement when it is semantically checked. The statement will be turned into a NOP statement after checking.

ct_echo_stmt       ::= "$echo" constant_expr ";"

The message

The message must be a compile time constant string.

Compile time assert statement

During parsing, the compiler will check the compile time expression and create a compile time error with the optional message. After evaluation, the $assert becomes a NOP statement.

ct_assert_stmt     ::= "$assert" constant_expr (":" constant_expr) ";"

Evaluated expression

The checked expression must evaluate to a boolean compile time constant.

Error message

The second parameter, which is optional, must evaluate to a constant string.

Compile time error statement

During parsing, when semantically checked this statement will output a compile time error with the message given.

ct_error_stmt      ::= "$error" constant_expr ";"

Error message

The parameter must evaluate to a constant string.

Compile time if statement

If the cond expression is true, the then-branch is processed by the compiler. If it evaluates to false, the else-branch is processed if it exists.

ct_if_stmt         ::= "$if" constant_expr ":" stmt* ("$else" stmt*)? "$endif"

Cond expression

The cond expression must be possible to evaluate to true or false at compile time.

Scopes

The “then” and “else” branches will add a compile time scope that is exited when reaching $endif. It adds no runtime scope.

Evaluation

Statements in the branch not picked will not be semantically checked.

Compile time switch statement

ct_switch_stmt     ::= "$switch" (ct_expr_or_type)? ":"
ct_case_stmt       ::= ("$default" | "$case" ct_expr_or_type) ":" stmt*

No cond expression switch

If the cond expression is missing, evaluation will go through each case until one case expression evaluates to true.

Type expressions

If a cond expression is a type, then all case statement expressions must be types as well.

Ranged cases

Compile time switch does not support ranged cases.

Fallthrough

If a case clause has no statements, then when executing the case, rather than exiting the switch, the next case clause immediately following it will be used. If that one should also be missing statements, the procedure will be repeated until a case clause with statements is encountered, or the end of the switch is reached.

Break and nextcase

Compile time switches do not support break nor nextcase.

Evaluation of statements

Only the case which is first matched has its statements processed by the compiler. All other statements are ignored and will not be semantically checked.

Continue statement

A continue statement jumps to the cond expression of a while, for, do or foreach

continue_stmt      ::= "continue" label? ";"

Continue labels

If a continue has a label, then it will jump to the cond of the while/for/do in the outer scope with the corresponding label.

Unreachable code

Any statement following continue in the same scope is considered unreachable.

Declaration statement

A declaration statement adds a new runtime or compile time variable to the current scope. It is available after the declaration statement.

declaration_stmt   ::= const_declaration | local_decl_storage? optional_type decls_after_type ";"
local_decl_storage ::= "tlocal" | "static"
decls_after_type   ::= local_decl_after_type ("," local_decl_after_type)*
decl_after_type    ::= CT_IDENT ("=" constant_expr)? | IDENTIFIER opt_attributes ("=" expr)?

Thread local storage

Using tlocal allocates the runtime variable as a thread local variable. In effect this is the same as declaring the variable as a global tlocal variable, but the visibility is limited to the function. tlocal may not be combined with static.

The initializer for a tlocal variable must be a valid global init expression.

Static storage

Using static allocates the runtime variable as a function global variable. In effect this is the same as declaring a global, but visibility is limited to the function. static may not be combined with tlocal.

The initializer for a static variable must be a valid global init expression.

Scopes

Runtime variables are added to the runtime scope, compile time variables to the compile time scope. See **var statements **.

Multiple declarations

If more than one variable is declared, no init expressions are allowed for any of the variables.

No init expression

If no init expression is provided, the variable is zero initialized.

Opt-out of zero initialization

Using the @noinit attribute opts out of zero initialization.

Self referencing initialization

An init expression may refer to the address of the same variable that is declared, but not the value of the variable.

Example:

void* a = &a;  // Valid
int a = a + 1; // Invalid

Defer statement

The defer statements are executed at (runtime) scope exit, whether through return, break, continue or rethrow.

defer_stmt         ::= "defer" ("try" | "catch")? stmt

Defer in defer

The defer body (statement) may not be a defer statement. However, if the body is a compound statement then this may have any number of defer statements.

Static and tlocal variables in defer

Static and tlocal variables are allowed in a defer statement. Only a single variable is instantiated regardless of the number of inlining locations.

Defer and return

If the return has an expression, then it is evaluated before the defer statements (due to exit from the current function scope), are executed.

Example:

int a = 0;
defer a++;
return a;
// This is equivalent to
int a = 0;
int temp = a;
a++;
return temp;

Defer and jump statements

A defer body may not contain a break, continue, return or rethrow that would exit the statement.

Defer execution

Defer statements are executed in the reverse order of their declaration, starting from the last declared defer statement.

`defer try`

A defer try type of defer will only execute if the scope is left through normal fallthrough, break, continue or a return with a result.

It will not execute if the exit is through a rethrow or a return with an optional value.

`defer catch`

A defer catch type of defer will only execute if the scope is left through a rethrow or a return with an optional value

It will not execute if the exit is a normal fallthrough, break, continue or a return with a result.

Non-regular returns - longjmp, panic and other errors

Defers will not execute when doing longjmp terminating through a panic or other error. They are only invoked on regular scope exits.

Expr statement

An expression statement evaluates an expression.

expr_stmt          ::= expr ";"

No discard

If the expression is a function or macro call either returning an optional or annotated @nodiscard, then the expression is a compile time error. A function or macro returning an optional can use the @maydiscard attribute to suppress this error.

If statement

An if statement will evaluate the cond expression, then execute the first statement (the “then clause”) in the if-body if it evaluates to “true”, otherwise execute the else clause. If no else clause exists, then the next statement is executed.

if_stmt            ::= "if" (label ":")? "(" cond_expr ")" if_body
if_body            ::= non_compound_stmt | compound_stmt else_clause? | "{" switch_body "}"
else_clause        ::= "else" (if_stmt | compound_stmt)

Scopes

Both the “then” clause and the else clause open new scopes, even if they are non-compound statements. The cond expression scope is valid until the exit of the entire statement, so any declarations in the cond expression are available both in then and else clauses. Declarations in the “then” clause is not available in the else clause and vice versa.

Special parsing of the “then” clause

If the then-clause isn’t a compound statement, then it must follow on the same row as the cond expression. It may not appear on a consecutive row.

Break

It is possible to use labelled break to break out of an if statement. Note that an unlabelled break may not be used.

If-try

The cond expression may be a try-unwrap chain. In this case, the unwrapped variables are scoped to the “then” clause only.

If-catch

The cond expression may be a catch-unwrap. The unwrap is scoped to the “then” clause only. If one or more variables are in the catch, then the “else” clause have these variables implicitly unwrapped.

Example:

int? a = foo();
int? b = foo();
if (catch a, b)
{
    // Do something
}
else
{
    int x = a + b; // Valid, a and b are implicitly unwrapped.
}

If-catch implicit unwrap

If an if-catch’s “then”-clause will jump out of the outer scope in all code paths and the catch is on one or more variables, then this variable(s) will be implicitly unwrapped in the outer scope after the if-statement.

Example:

int? a = foo();
if (catch a)
{
  return;
}
int x = a; // Valid, a is implicitly unwrapped.

Nextcase statement

Nextcase will jump to another switch case.

nextcase_stmt      ::= "nextcase" ((label ":")? (expr | "default"))? ";"

Labels

When a nextcase has a label, the jump is to the switch in an outer scope with the corresponding label.

No expression jumps

A nextcase without any expression jumps to the next case clause in the current switch. It is not possible to use no expression nextcase with labels.

Jumps to default

Using default jumps to the default clause of a switch.

Missing case

If the switch has constant case values, and the nextcase expression is constant, then the value of the expression must match a case clause. Not matching a case is a compile time error.

If one or more cases are non-constant and/or the nextcase expression is non-constant, then no compile time check is made.

Variable expression

If the nextcase has a non-constant expression, or the cases are not all constant, then first the nextcase expression is evaluated. Next, execution will proceed as if the switch was invoked again, but with the nextcase expression as the switch cond expression. See switch statement.

If the switch does not have a cond expression, nextcase with an expression is not allowed.

Unreachable code

Any statement in the same scope after a nextcase are considered unreachable.

Switch statement

switch_stmt        ::= "switch" (label ":")? ("(" cond_expr ")")? switch body
switch_body        ::= "{" case_clause* "}"
case_clause        ::= default_stmt | case_stmt
default_stmt       ::= "default" ":" stmt*
case_stmt          ::= "case" label? expr (".." expr)? ":" stmt*

Regular switch

If the cond expression exists and all case statements have constant expression, then first the cond expression is evaluated, next the case corresponding to the expression’s value will be jumped to and the statement will be executed. After reaching the end of the statements and a new case clause or the end of the switch body, the execution will jump to the first statement after the switch.

If-switch

If the cond expression is missing or the case statements are non-constant expressions, then each case clause will be evaluated in order after the cond expression has been evaluated (if it exists):

If a cond expression exists, calculate the case expression and execute the case if it is matching the cond expression. A default statement has no expression and will always be considered matching the cond expression reached.
If no con expression exists, calculate the case expression and execute the case if the expression evaluates to “true” when implicitly converted to boolean. A default statement will always be considered having the “true” result.

Any-switch

If the cond expression is an any type, the switch is handled as if switching was done over the type field of the any. This field has the type of typeid, and the cases follows the rules for switching over typeid.

If the cond expression is a variable, then this variable is implicitly converted to a pointer with the pointee type given by the case statement.

Example:

any a = abc();
switch (a)
{
    case int:
        int b = *a;   // a is int*
    case float:
        float z = *a; // a is float*
    case Bar:
        Bar f = *a;   // a is Bar*
    default:
        // a is not unwrapped
}

Ranged cases

Cases may be ranged. The start and end of the range must both be constant integer values. The start must be less or equal to the end value. Using non-integers or non-constant values is a compile time error.

Fallthrough

If a case clause has no statements, then when executing the case, rather than exiting the switch, the next case clause immediately following it will be executed. If that one should also be missing statement, the procedure will be repeated until a case clause with statements is encountered (and executed), or the end of the switch is reached.

Exhaustive switch

If a switch case has a default clause or it is switching over an enum and there exists a case for each enum value then the switch is exhaustive.

Break

If an unlabelled break, or a break with the switch’s label is encountered, then the execution will jump out of the switch and proceed directly after the end of the switch body.

Unreachable code

If a switch is exhaustive and all case clauses end with a jump instruction, containing no break statement out of the current switch, then the code directly following the switch will be considered unreachable.

Switching over typeid

If the switch cond expression is a typeid, then case declarations may use only the type name after the case, which will be interpreted as having an implicit .typeid. Example: case int: will be interpreted as if written case int.typeid.

Nextcase without expression

Without a value nextcase will jump to the beginning of the next case clause. It is not allowed to put nextcase without an expression if there are no following case clauses.

Nextcase with expression

Nextcase with an expression will evaluate the expression and then jump as if the switch was entered with the cond expression corresponding to the value of the nextcase expression. Nextcase with an expression cannot be used on a switch without a cond expression.

Do statement

The do statement first evaluates its body (inner statement), then evaluates the cond expression. If the cond expression evaluates to true, jumps back into the body and repeats the process.

do_stmt            ::= "do" label? compound_stmt ("while" "(" cond_expr ")")? ";"

Unreachable code

The statement after a do is considered unreachable if the cond expression cannot ever be false and there is no break out of the do.

Break

break will exit the do with execution continuing on the following statement.

Continue

continue will jump directly to the evaluation of the cond, as if the end of the statement had been reached.

Do block

If no while part exists, it will only execute the block once, as if it ended with while (false), this is called a “do block”

For statement

The for statement will perform the (optional) init expression. The cond expression will then be tested. If it evaluates to true then the body will execute, followed by the incr expression. After execution will jump back to the cond expression and execution will repeat until the cond expression evaluates to false.

for_stmt           ::= "for" label? "(" init_expr ";" cond_expr? ";" incr_expr ")" stmt
init_expr          ::= decl_expr_list?
incr_expr          ::= expr_list?

Init expression

The init expression is only executed once before the rest of the for loop is executed. Any declarations in the init expression will be in scope until the for loop exits.

The init expression may optionally be omitted.

Incr expression

The incr expression is evaluated before evaluating the cond expr every time except for the first one.

The incr expression may optionally be omitted.

Cond expression

The cond expression is evaluated every loop. Any declaration in the cond expression is scoped to the current loop, i.e. it will be reinitialized at the start of every loop.

The cond expression may optionally be omitted. This is equivalent to setting the cond expression to always return true.

Unreachable code

The statement after a for is considered unreachable if the cond expression cannot ever be false, or is omitted and there is no break out of the loop.

Break

break will exit the for with execution continuing on the following statement after the for.

Continue

continue will jump directly to the evaluation of the cond, as if the end of the statement had been reached.

Equivalence of `while` and `for`

A while loop is functionally equivalent to a for loop without init and incr expressions.

`foreach` and `foreach_r` statements

The foreach statement will loop over a sequence of values. The foreach_r is equivalent to foreach but the order of traversal is reversed. foreach starts with element 0 and proceeds step by step to element len - 1. foreach_r starts starts with element len - 1 and proceeds step by step to element 0.

foreach_stmt       ::= "foreach" label? "(" foreach_vars ":" expr ")" stmt
foreach_r_stmt     ::= "foreach_r" label? "(" foreach_vars ":" expr ")" stmt
foreach_vars       ::= (foreach_index ",")? foreach_var
foreach_var        ::= type? "&"? IDENTIFIER

Break

break will exit the foreach statement with execution continuing on the following statement after.

Continue

continue will cause the next iteration to commence, as if the end of the statement had been reached.

Iteration by value or reference

Normally iteration are by value. Each element is copied into the foreach variable. If & is added before the variable name, the elements will be retrieved by reference instead, and consequently the type of the variable will be a pointer to the element type instead.

Foreach variable

The foreach variable may omit the type. In this case the type is inferred. If the type differs from the element type, then an implicit conversion will be attempted. Failing this is a compile time error.

Foreach index

If a variable name is added before the foreach variable, then this variable will receive the index of the element. For foreach_r this mean that the first value of the index will be len - 1.

The index type defaults to usz.

If an optional type is added to the index, the index will be converted to this type. The type must be an integer type. The conversion happens as if the conversion was a direct cast. If the actual index value would exceed the maximum representable value of the type, this does not affect the actual iteration, but may cause the index value to take on an incorrect value due to the cast.

For example, if the optional index type is char and the actual index is 256, then the index value would show 0 as (char)256 evaluates to zero.

Modifying the index variable will not affect the foreach iteration.

Foreach support

Foreach is natively supported for any slice, array, pointer to an array, vector and pointer to a vector. These types support both iteration by value and reference.

In addition, a type with operator overload for len and [] will support iteration by value, and a type with operator overload for len and &[] will support iteration by reference.

Return statement

The return statement evaluates its expression (if present) and returns the result.

return_stmt        ::= "return" expr? ";"

Jumps in return statements

If the expression should in itself cause an implicit return, for example due to the rethrow operator !, then this jump will happen before the return.

An example:

return foo()!;
// is equivalent to:
int temp = foo()!;
return temp;

Empty returns

An empty return is equivalent to a return with a void type. Consequently constructs like foo(); return; and return (void)foo(); are equivalent.

Unreachable code

Any statement directly following a return in the same scope are considered unreachable.

While statement

The while statement evaluates the cond expression and executes the statement if it evaluates to true. After this the cond expression is evaluated again and the process is repeated until cond expression returns false.

while_stmt         ::= "while" label? "(" cond_expr ")" stmt

Unreachable code

The statement after a while is considered unreachable if the cond expression cannot ever be false and there is no break out of the while.

Break

break will exit the while with execution continuing on the following statement.

Continue

continue will jump directly to the evaluation of the cond, as if the end of the statement had been reached.

Var statement

A var statement declares a variable with inferred type, or a compile time type variable. It can be used both for runtime and compile time variables. The use for runtime variables is limited to macros.

var_stmt           ::= "var" IDENTIFIER | CT_IDENT | CT_TYPE_IDENT ("=" expr)? ";"

Inferring type

In the case of a runtime variable, the type is inferred from the expression. Not providing an expression is a compile time error. The expression must resolve to a runtime type.

For compile time variables, the expression is optional. The expression may resolve to a runtime or compile time type.

Scope

Runtime variables will follow the runtime scopes, identical to behaviour in a declaration statement. The compile time variables will follow the compile time scopes which are delimited by scoping compile time statements ($if, $switch, $foreach and $for).

Attributes

Attributes are modifiers attached to modules, variables, type declarations etc.

name	used with
`@align`	fn, const, variables, user-defined types, struct member
`@benchmark`	module, fn
`@bigendian`	bitstruct only
`@builtin`	macro, fn, global, constant
`@callconv`	fn, call
`@deprecated`	fn, macro, interface, variables, constants, user-defined types, struct member
`@dynamic`	fn
`@export`	fn, globals, constants, struct, union, enum, faultdef
`@extern`	fn, globals, constants, user-defined types, faultdef
`@if`	all except local variables and calls
`@inline`	fn, call
`@interface`	fn
`@littleendian`	bitstruct only
`@local`	module, fn, macro, globals, constants, user-defined types, attributes and aliases
`@maydiscard`	fn, macro
`@naked`	fn
`@nodiscard`	fn, macro
`@noinit`	variables
`@noinline`	fn, call
`@noreturn`	fn, macro
`@nostrip`	fn, globals, constants, struct, union, enum, faultdef
`@obfuscate`	enum, faultdef
`@operator`	fn, macro
`@optional`	interface methods
`@overlap`	bitstruct only
`@packed`	struct, union
`@priority`	initializer/finalizer
`@private`	module, fn, macro, globals, constants, user-defined types, attributes and aliases
`@public`	module, fn, macro, globals, constants, user-defined types, attributes and aliases
`@pure`	call
`@reflect`	fn, globals, constants, user-defined types
`@section`	fn, globals, constants
`@test`	module, fn
`@unused`	all except call and initializer/finalizers
`@used`	all except call and initializer/finalizers
`@weak`	fn, globals, constants
`@winmain`	fn

`@deprecated`

Takes an optional constant string. If the node is in use, print the deprecation and add the optional string if present.

`@optional`

Marks an interface method as optional, and so does not need to be implemented by a conforming type.

`@winmain`

Marks a main function as a win32 winmain function, which is the entrypoint for a windowed application on Windows. This allows the main function to take a different set of arguments than usual.

`@callconv`

@callconv can be used with a function or a call. It takes a constant string which is either “veccall”, “stdcall” or “cdecl”. If more than one @callconv is applied to a function or call, the last one takes precedence.

User defined attributes

User defined attributes group a list of attributes.

attribute_decl     ::= "attrdef" AT_TYPE_IDENT ("(" parameters ")")? attribute* "=" "{" attribute* "}" ";"

Empty list of attributes

The list of attributes may be empty.

Parameter arguments

Arguments given to user defined attributes will be passed on to the attributes in the list.

Expansion

When a user defined attribute is encountered, its list of attributes is copied and appended instead of the user defined attribute. Any argument passed to the attribute is evaluated and passed as a constant by the name of the parameter to the evaluation of the attribute parameters in the list.

Nesting

A user defined attribute can contain other user defined attributes. The definition may not be cyclic.

Methods

Operator overloading

@operator overloads may only be added to user defined types (typedef, unions, struct, enum and fault).

Indexing operator (`[]`)

This requires a return type and a method parameter, which is the index.

Reference indexing operator (`&[]`)

This requires a return type and a method parameter, which is the index. If [] is implemented, it should return a pointer to [].

Assigning index operator (`=[]`)

This has a void return type, and index should match that of [] and &[]. Value should match that of [] and be the pointee of the result of &[].

Len operator (`len`)

This must have an integer return type.

Dynamic methods

@dynamic may be used on methods for any type except any and interfaces.

Modules

Module paths are hierarchal, with each sub-path appended with ’::’ + the name:

path               ::= PATH_SEGMENT ("::" PATH_SEGMENT)

Each module declaration starts its own module section. All imports and all @local declarations are only visible in the current module section.

module_section     ::= "module" path opt_generic_params? attributes? ";"
generic_param      ::= TYPE_IDENT | CONST_IDENT
opt_generic_params ::= "{" generic_param ("," generic_param)* "}"

Any visibility attribute defined in a module section will be the default visibility in all declarations in the section.

If the @benchmark attribute is applied to the module section then all function declarations will implicitly have the @benchmark attribute.

If the @test attribute is applied to the module section then all function declarations will implicitly have the @test attribute.

C3 Specification

Notation

Source code representation

Lexical Translations

Line Terminators

Input Elements and Tokens

White Space

Letters and digits

Comments

Doc contract

Identifiers

Keywords

Operators and punctuation

Integer literals

Floating point literals

Characters

Backslash escapes

String literals

Compile time string concatenation

Raw string literals

Compile time concatenation

Source code pre-filtering

Character literals

Types

Boolean types

Integer types

Floating point types

Vector types

Vector base type

Min width

Element access

Alignment

Vector operations

Array types

Slice types

Pointer types

Pointee type

iptr and uptr

The wildcard pointer void*

Pointer arithmetic on void*

Subscripting

Deref

Struct types

Alignment

Union types

Alignment

Fault types

Alignment

Enum types

Function types

Typeid type

Any type

Fields

Switching over any

Anyfault type

Declarations and scope

Expressions

Assignment expression

Type assign

Combined assign

Implicit conversion

Ternary, elvis and or-else expressions

Ternary evaluation

Elvis evaluation

Orelse evaluation

Suffix expression

Effect of ?

Type of the expression

Rethrow expression

The expression to rethrow

Type

Relational expression

And expression

Type

Or expression

Constant folded arithmetics

Type

Casts

Pointer casts

Integer to pointer cast

`iptr` and `uptr`

The wildcard pointer `void*`

Pointer arithmetic on `void*`

Switching over `any`

Effect of `?`

`defer try`

`defer catch`